xref: /dflybsd-src/sys/kern/vfs_subr.c (revision cd29885abfb8f68adb0c082e313b891156d66964)
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/file.h>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/mount.h>
62 #include <sys/proc.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <machine/limits.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_kern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vnode_pager.h>
83 #include <vm/vm_zone.h>
84 
85 #include <sys/buf2.h>
86 #include <sys/thread2.h>
87 #include <sys/sysref2.h>
88 
89 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
90 
91 int numvnodes;
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93 int vfs_fastdev = 1;
94 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
95 
96 enum vtype iftovt_tab[16] = {
97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 };
100 int vttoif_tab[9] = {
101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 	S_IFSOCK, S_IFIFO, S_IFMT,
103 };
104 
105 static int reassignbufcalls;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
107 		&reassignbufcalls, 0, "");
108 static int reassignbufloops;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
110 		&reassignbufloops, 0, "");
111 static int reassignbufsortgood;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
113 		&reassignbufsortgood, 0, "");
114 static int reassignbufsortbad;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
116 		&reassignbufsortbad, 0, "");
117 static int reassignbufmethod = 1;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
119 		&reassignbufmethod, 0, "");
120 
121 int	nfs_mount_type = -1;
122 static struct lwkt_token spechash_token;
123 struct nfs_public nfs_pub;	/* publicly exported FS */
124 
125 int desiredvnodes;
126 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
127 		&desiredvnodes, 0, "Maximum number of vnodes");
128 
129 static void	vfs_free_addrlist (struct netexport *nep);
130 static int	vfs_free_netcred (struct radix_node *rn, void *w);
131 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
132 				       const struct export_args *argp);
133 
134 extern int dev_ref_debug;
135 
136 /*
137  * Red black tree functions
138  */
139 static int rb_buf_compare(struct buf *b1, struct buf *b2);
140 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
141 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
142 
143 static int
144 rb_buf_compare(struct buf *b1, struct buf *b2)
145 {
146 	if (b1->b_loffset < b2->b_loffset)
147 		return(-1);
148 	if (b1->b_loffset > b2->b_loffset)
149 		return(1);
150 	return(0);
151 }
152 
153 /*
154  * Returns non-zero if the vnode is a candidate for lazy msyncing.
155  */
156 static __inline int
157 vshouldmsync(struct vnode *vp)
158 {
159 	if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
160 		return (0);		/* other holders */
161 	if (vp->v_object &&
162 	    (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
163 		return (0);
164 	}
165 	return (1);
166 }
167 
168 /*
169  * Initialize the vnode management data structures.
170  *
171  * Called from vfsinit()
172  */
173 void
174 vfs_subr_init(void)
175 {
176 	/*
177 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
178 	 * according to available system memory but we may also have
179 	 * to limit it based on available KVM, which is capped on 32 bit
180 	 * systems.
181 	 */
182 	desiredvnodes = min(maxproc + vmstats.v_page_count / 4,
183 			    KvaSize / (20 *
184 			    (sizeof(struct vm_object) + sizeof(struct vnode))));
185 
186 	lwkt_token_init(&spechash_token);
187 }
188 
189 /*
190  * Knob to control the precision of file timestamps:
191  *
192  *   0 = seconds only; nanoseconds zeroed.
193  *   1 = seconds and nanoseconds, accurate within 1/HZ.
194  *   2 = seconds and nanoseconds, truncated to microseconds.
195  * >=3 = seconds and nanoseconds, maximum precision.
196  */
197 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
198 
199 static int timestamp_precision = TSP_SEC;
200 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
201 		&timestamp_precision, 0, "");
202 
203 /*
204  * Get a current timestamp.
205  *
206  * MPSAFE
207  */
208 void
209 vfs_timestamp(struct timespec *tsp)
210 {
211 	struct timeval tv;
212 
213 	switch (timestamp_precision) {
214 	case TSP_SEC:
215 		tsp->tv_sec = time_second;
216 		tsp->tv_nsec = 0;
217 		break;
218 	case TSP_HZ:
219 		getnanotime(tsp);
220 		break;
221 	case TSP_USEC:
222 		microtime(&tv);
223 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
224 		break;
225 	case TSP_NSEC:
226 	default:
227 		nanotime(tsp);
228 		break;
229 	}
230 }
231 
232 /*
233  * Set vnode attributes to VNOVAL
234  */
235 void
236 vattr_null(struct vattr *vap)
237 {
238 	vap->va_type = VNON;
239 	vap->va_size = VNOVAL;
240 	vap->va_bytes = VNOVAL;
241 	vap->va_mode = VNOVAL;
242 	vap->va_nlink = VNOVAL;
243 	vap->va_uid = VNOVAL;
244 	vap->va_gid = VNOVAL;
245 	vap->va_fsid = VNOVAL;
246 	vap->va_fileid = VNOVAL;
247 	vap->va_blocksize = VNOVAL;
248 	vap->va_rmajor = VNOVAL;
249 	vap->va_rminor = VNOVAL;
250 	vap->va_atime.tv_sec = VNOVAL;
251 	vap->va_atime.tv_nsec = VNOVAL;
252 	vap->va_mtime.tv_sec = VNOVAL;
253 	vap->va_mtime.tv_nsec = VNOVAL;
254 	vap->va_ctime.tv_sec = VNOVAL;
255 	vap->va_ctime.tv_nsec = VNOVAL;
256 	vap->va_flags = VNOVAL;
257 	vap->va_gen = VNOVAL;
258 	vap->va_vaflags = 0;
259 	vap->va_fsmid = VNOVAL;
260 	/* va_*_uuid fields are only valid if related flags are set */
261 }
262 
263 /*
264  * Flush out and invalidate all buffers associated with a vnode.
265  *
266  * vp must be locked.
267  */
268 static int vinvalbuf_bp(struct buf *bp, void *data);
269 
270 struct vinvalbuf_bp_info {
271 	struct vnode *vp;
272 	int slptimeo;
273 	int lkflags;
274 	int flags;
275 };
276 
277 void
278 vupdatefsmid(struct vnode *vp)
279 {
280 	atomic_set_int(&vp->v_flag, VFSMID);
281 }
282 
283 int
284 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
285 {
286 	struct vinvalbuf_bp_info info;
287 	vm_object_t object;
288 	lwkt_tokref vlock;
289 	int error;
290 
291 	lwkt_gettoken(&vlock, &vp->v_token);
292 
293 	/*
294 	 * If we are being asked to save, call fsync to ensure that the inode
295 	 * is updated.
296 	 */
297 	if (flags & V_SAVE) {
298 		error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
299 		if (error)
300 			goto done;
301 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
302 			if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0)
303 				goto done;
304 
305 			/*
306 			 * Dirty bufs may be left or generated via races
307 			 * in circumstances where vinvalbuf() is called on
308 			 * a vnode not undergoing reclamation.   Only
309 			 * panic if we are trying to reclaim the vnode.
310 			 */
311 			if ((vp->v_flag & VRECLAIMED) &&
312 			    (bio_track_active(&vp->v_track_write) ||
313 			    !RB_EMPTY(&vp->v_rbdirty_tree))) {
314 				panic("vinvalbuf: dirty bufs");
315 			}
316 		}
317   	}
318 	info.slptimeo = slptimeo;
319 	info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
320 	if (slpflag & PCATCH)
321 		info.lkflags |= LK_PCATCH;
322 	info.flags = flags;
323 	info.vp = vp;
324 
325 	/*
326 	 * Flush the buffer cache until nothing is left.
327 	 */
328 	while (!RB_EMPTY(&vp->v_rbclean_tree) ||
329 	       !RB_EMPTY(&vp->v_rbdirty_tree)) {
330 		error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
331 				vinvalbuf_bp, &info);
332 		if (error == 0) {
333 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
334 					vinvalbuf_bp, &info);
335 		}
336 	}
337 
338 	/*
339 	 * Wait for I/O completion.  We may block in the pip code so we have
340 	 * to re-check.
341 	 */
342 	do {
343 		bio_track_wait(&vp->v_track_write, 0, 0);
344 		if ((object = vp->v_object) != NULL) {
345 			while (object->paging_in_progress)
346 				vm_object_pip_sleep(object, "vnvlbx");
347 		}
348 	} while (bio_track_active(&vp->v_track_write));
349 
350 	/*
351 	 * Destroy the copy in the VM cache, too.
352 	 */
353 	if ((object = vp->v_object) != NULL) {
354 		vm_object_page_remove(object, 0, 0,
355 			(flags & V_SAVE) ? TRUE : FALSE);
356 	}
357 
358 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
359 		panic("vinvalbuf: flush failed");
360 	if (!RB_EMPTY(&vp->v_rbhash_tree))
361 		panic("vinvalbuf: flush failed, buffers still present");
362 	error = 0;
363 done:
364 	lwkt_reltoken(&vlock);
365 	return (error);
366 }
367 
368 static int
369 vinvalbuf_bp(struct buf *bp, void *data)
370 {
371 	struct vinvalbuf_bp_info *info = data;
372 	int error;
373 
374 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
375 		error = BUF_TIMELOCK(bp, info->lkflags,
376 				     "vinvalbuf", info->slptimeo);
377 		if (error == 0) {
378 			BUF_UNLOCK(bp);
379 			error = ENOLCK;
380 		}
381 		if (error == ENOLCK)
382 			return(0);
383 		return (-error);
384 	}
385 
386 	KKASSERT(bp->b_vp == info->vp);
387 
388 	/*
389 	 * XXX Since there are no node locks for NFS, I
390 	 * believe there is a slight chance that a delayed
391 	 * write will occur while sleeping just above, so
392 	 * check for it.  Note that vfs_bio_awrite expects
393 	 * buffers to reside on a queue, while bwrite() and
394 	 * brelse() do not.
395 	 *
396 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
397 	 * check.  This code will write out the buffer, period.
398 	 */
399 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
400 	    (info->flags & V_SAVE)) {
401 		if (bp->b_vp == info->vp) {
402 			if (bp->b_flags & B_CLUSTEROK) {
403 				vfs_bio_awrite(bp);
404 			} else {
405 				bremfree(bp);
406 				bawrite(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 		brelse(bp);
421 	} else {
422 		bremfree(bp);
423 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
424 		brelse(bp);
425 	}
426 	return(0);
427 }
428 
429 /*
430  * Truncate a file's buffer and pages to a specified length.  This
431  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
432  * sync activity.
433  *
434  * The vnode must be locked.
435  */
436 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
437 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
438 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
439 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
440 
441 int
442 vtruncbuf(struct vnode *vp, off_t length, int blksize)
443 {
444 	off_t truncloffset;
445 	const char *filename;
446 	lwkt_tokref vlock;
447 	int count;
448 
449 	/*
450 	 * Round up to the *next* block, then destroy the buffers in question.
451 	 * Since we are only removing some of the buffers we must rely on the
452 	 * scan count to determine whether a loop is necessary.
453 	 */
454 	if ((count = (int)(length % blksize)) != 0)
455 		truncloffset = length + (blksize - count);
456 	else
457 		truncloffset = length;
458 
459 	lwkt_gettoken(&vlock, &vp->v_token);
460 	do {
461 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
462 				vtruncbuf_bp_trunc_cmp,
463 				vtruncbuf_bp_trunc, &truncloffset);
464 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
465 				vtruncbuf_bp_trunc_cmp,
466 				vtruncbuf_bp_trunc, &truncloffset);
467 	} while(count);
468 
469 	/*
470 	 * For safety, fsync any remaining metadata if the file is not being
471 	 * truncated to 0.  Since the metadata does not represent the entire
472 	 * dirty list we have to rely on the hit count to ensure that we get
473 	 * all of it.
474 	 */
475 	if (length > 0) {
476 		do {
477 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
478 					vtruncbuf_bp_metasync_cmp,
479 					vtruncbuf_bp_metasync, vp);
480 		} while (count);
481 	}
482 
483 	/*
484 	 * Clean out any left over VM backing store.
485 	 *
486 	 * It is possible to have in-progress I/O from buffers that were
487 	 * not part of the truncation.  This should not happen if we
488 	 * are truncating to 0-length.
489 	 */
490 	vnode_pager_setsize(vp, length);
491 	bio_track_wait(&vp->v_track_write, 0, 0);
492 
493 	filename = TAILQ_FIRST(&vp->v_namecache) ?
494 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
495 
496 	/*
497 	 * Make sure no buffers were instantiated while we were trying
498 	 * to clean out the remaining VM pages.  This could occur due
499 	 * to busy dirty VM pages being flushed out to disk.
500 	 */
501 	do {
502 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
503 				vtruncbuf_bp_trunc_cmp,
504 				vtruncbuf_bp_trunc, &truncloffset);
505 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
506 				vtruncbuf_bp_trunc_cmp,
507 				vtruncbuf_bp_trunc, &truncloffset);
508 		if (count) {
509 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
510 			       "left over buffers in %s\n", count, filename);
511 		}
512 	} while(count);
513 
514 	lwkt_reltoken(&vlock);
515 
516 	return (0);
517 }
518 
519 /*
520  * The callback buffer is beyond the new file EOF and must be destroyed.
521  * Note that the compare function must conform to the RB_SCAN's requirements.
522  */
523 static
524 int
525 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
526 {
527 	if (bp->b_loffset >= *(off_t *)data)
528 		return(0);
529 	return(-1);
530 }
531 
532 static
533 int
534 vtruncbuf_bp_trunc(struct buf *bp, void *data)
535 {
536 	/*
537 	 * Do not try to use a buffer we cannot immediately lock, but sleep
538 	 * anyway to prevent a livelock.  The code will loop until all buffers
539 	 * can be acted upon.
540 	 */
541 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
542 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
543 			BUF_UNLOCK(bp);
544 	} else {
545 		bremfree(bp);
546 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
547 		brelse(bp);
548 	}
549 	return(1);
550 }
551 
552 /*
553  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
554  * blocks (with a negative loffset) are scanned.
555  * Note that the compare function must conform to the RB_SCAN's requirements.
556  */
557 static int
558 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
559 {
560 	if (bp->b_loffset < 0)
561 		return(0);
562 	return(1);
563 }
564 
565 static int
566 vtruncbuf_bp_metasync(struct buf *bp, void *data)
567 {
568 	struct vnode *vp = data;
569 
570 	if (bp->b_flags & B_DELWRI) {
571 		/*
572 		 * Do not try to use a buffer we cannot immediately lock,
573 		 * but sleep anyway to prevent a livelock.  The code will
574 		 * loop until all buffers can be acted upon.
575 		 */
576 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
577 			if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
578 				BUF_UNLOCK(bp);
579 		} else {
580 			bremfree(bp);
581 			if (bp->b_vp == vp)
582 				bawrite(bp);
583 			else
584 				bwrite(bp);
585 		}
586 		return(1);
587 	} else {
588 		return(0);
589 	}
590 }
591 
592 /*
593  * vfsync - implements a multipass fsync on a file which understands
594  * dependancies and meta-data.  The passed vnode must be locked.  The
595  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
596  *
597  * When fsyncing data asynchronously just do one consolidated pass starting
598  * with the most negative block number.  This may not get all the data due
599  * to dependancies.
600  *
601  * When fsyncing data synchronously do a data pass, then a metadata pass,
602  * then do additional data+metadata passes to try to get all the data out.
603  */
604 static int vfsync_wait_output(struct vnode *vp,
605 			    int (*waitoutput)(struct vnode *, struct thread *));
606 static int vfsync_data_only_cmp(struct buf *bp, void *data);
607 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
608 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
609 static int vfsync_bp(struct buf *bp, void *data);
610 
611 struct vfsync_info {
612 	struct vnode *vp;
613 	int synchronous;
614 	int syncdeps;
615 	int lazycount;
616 	int lazylimit;
617 	int skippedbufs;
618 	int (*checkdef)(struct buf *);
619 };
620 
621 int
622 vfsync(struct vnode *vp, int waitfor, int passes,
623 	int (*checkdef)(struct buf *),
624 	int (*waitoutput)(struct vnode *, struct thread *))
625 {
626 	struct vfsync_info info;
627 	lwkt_tokref vlock;
628 	int error;
629 
630 	bzero(&info, sizeof(info));
631 	info.vp = vp;
632 	if ((info.checkdef = checkdef) == NULL)
633 		info.syncdeps = 1;
634 
635 	lwkt_gettoken(&vlock, &vp->v_token);
636 
637 	switch(waitfor) {
638 	case MNT_LAZY:
639 		/*
640 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
641 		 * number of data (not meta) pages we try to flush to 1MB.
642 		 * A non-zero return means that lazy limit was reached.
643 		 */
644 		info.lazylimit = 1024 * 1024;
645 		info.syncdeps = 1;
646 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
647 				vfsync_lazy_range_cmp, vfsync_bp, &info);
648 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
649 				vfsync_meta_only_cmp, vfsync_bp, &info);
650 		if (error == 0)
651 			vp->v_lazyw = 0;
652 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
653 			vn_syncer_add_to_worklist(vp, 1);
654 		error = 0;
655 		break;
656 	case MNT_NOWAIT:
657 		/*
658 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
659 		 */
660 		info.syncdeps = 1;
661 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
662 			vfsync_bp, &info);
663 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
664 			vfsync_bp, &info);
665 		error = 0;
666 		break;
667 	default:
668 		/*
669 		 * Synchronous.  Do a data-only pass, then a meta-data+data
670 		 * pass, then additional integrated passes to try to get
671 		 * all the dependancies flushed.
672 		 */
673 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
674 			vfsync_bp, &info);
675 		error = vfsync_wait_output(vp, waitoutput);
676 		if (error == 0) {
677 			info.skippedbufs = 0;
678 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
679 				vfsync_bp, &info);
680 			error = vfsync_wait_output(vp, waitoutput);
681 			if (info.skippedbufs)
682 				kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
683 		}
684 		while (error == 0 && passes > 0 &&
685 		       !RB_EMPTY(&vp->v_rbdirty_tree)
686 		) {
687 			if (--passes == 0) {
688 				info.synchronous = 1;
689 				info.syncdeps = 1;
690 			}
691 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
692 				vfsync_bp, &info);
693 			if (error < 0)
694 				error = -error;
695 			info.syncdeps = 1;
696 			if (error == 0)
697 				error = vfsync_wait_output(vp, waitoutput);
698 		}
699 		break;
700 	}
701 	lwkt_reltoken(&vlock);
702 	return(error);
703 }
704 
705 static int
706 vfsync_wait_output(struct vnode *vp,
707 		   int (*waitoutput)(struct vnode *, struct thread *))
708 {
709 	int error;
710 
711 	error = bio_track_wait(&vp->v_track_write, 0, 0);
712 	if (waitoutput)
713 		error = waitoutput(vp, curthread);
714 	return(error);
715 }
716 
717 static int
718 vfsync_data_only_cmp(struct buf *bp, void *data)
719 {
720 	if (bp->b_loffset < 0)
721 		return(-1);
722 	return(0);
723 }
724 
725 static int
726 vfsync_meta_only_cmp(struct buf *bp, void *data)
727 {
728 	if (bp->b_loffset < 0)
729 		return(0);
730 	return(1);
731 }
732 
733 static int
734 vfsync_lazy_range_cmp(struct buf *bp, void *data)
735 {
736 	struct vfsync_info *info = data;
737 	if (bp->b_loffset < info->vp->v_lazyw)
738 		return(-1);
739 	return(0);
740 }
741 
742 static int
743 vfsync_bp(struct buf *bp, void *data)
744 {
745 	struct vfsync_info *info = data;
746 	struct vnode *vp = info->vp;
747 	int error;
748 
749 	/*
750 	 * if syncdeps is not set we do not try to write buffers which have
751 	 * dependancies.
752 	 */
753 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
754 		return(0);
755 
756 	/*
757 	 * Ignore buffers that we cannot immediately lock.  XXX
758 	 */
759 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
760 		kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
761 		++info->skippedbufs;
762 		return(0);
763 	}
764 	if ((bp->b_flags & B_DELWRI) == 0)
765 		panic("vfsync_bp: buffer not dirty");
766 	if (vp != bp->b_vp)
767 		panic("vfsync_bp: buffer vp mismatch");
768 
769 	/*
770 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
771 	 * has been written but an additional handshake with the device
772 	 * is required before we can dispose of the buffer.  We have no idea
773 	 * how to do this so we have to skip these buffers.
774 	 */
775 	if (bp->b_flags & B_NEEDCOMMIT) {
776 		BUF_UNLOCK(bp);
777 		return(0);
778 	}
779 
780 	/*
781 	 * Ask bioops if it is ok to sync
782 	 */
783 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
784 		bremfree(bp);
785 		brelse(bp);
786 		return(0);
787 	}
788 
789 	if (info->synchronous) {
790 		/*
791 		 * Synchronous flushing.  An error may be returned.
792 		 */
793 		bremfree(bp);
794 		error = bwrite(bp);
795 	} else {
796 		/*
797 		 * Asynchronous flushing.  A negative return value simply
798 		 * stops the scan and is not considered an error.  We use
799 		 * this to support limited MNT_LAZY flushes.
800 		 */
801 		vp->v_lazyw = bp->b_loffset;
802 		if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
803 			info->lazycount += vfs_bio_awrite(bp);
804 		} else {
805 			info->lazycount += bp->b_bufsize;
806 			bremfree(bp);
807 			bawrite(bp);
808 		}
809 		if (info->lazylimit && info->lazycount >= info->lazylimit)
810 			error = 1;
811 		else
812 			error = 0;
813 	}
814 	return(-error);
815 }
816 
817 /*
818  * Associate a buffer with a vnode.
819  *
820  * MPSAFE
821  */
822 int
823 bgetvp(struct vnode *vp, struct buf *bp)
824 {
825 	lwkt_tokref vlock;
826 
827 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
828 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
829 
830 	/*
831 	 * Insert onto list for new vnode.
832 	 */
833 	lwkt_gettoken(&vlock, &vp->v_token);
834 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
835 		lwkt_reltoken(&vlock);
836 		return (EEXIST);
837 	}
838 	bp->b_vp = vp;
839 	bp->b_flags |= B_HASHED;
840 	bp->b_flags |= B_VNCLEAN;
841 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
842 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
843 	vhold(vp);
844 	lwkt_reltoken(&vlock);
845 	return(0);
846 }
847 
848 /*
849  * Disassociate a buffer from a vnode.
850  */
851 void
852 brelvp(struct buf *bp)
853 {
854 	struct vnode *vp;
855 	lwkt_tokref vlock;
856 
857 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
858 
859 	/*
860 	 * Delete from old vnode list, if on one.
861 	 */
862 	vp = bp->b_vp;
863 	lwkt_gettoken(&vlock, &vp->v_token);
864 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
865 		if (bp->b_flags & B_VNDIRTY)
866 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
867 		else
868 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
869 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
870 	}
871 	if (bp->b_flags & B_HASHED) {
872 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
873 		bp->b_flags &= ~B_HASHED;
874 	}
875 	if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
876 		vp->v_flag &= ~VONWORKLST;
877 		LIST_REMOVE(vp, v_synclist);
878 	}
879 	bp->b_vp = NULL;
880 	lwkt_reltoken(&vlock);
881 
882 	vdrop(vp);
883 }
884 
885 /*
886  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
887  * This routine is called when the state of the B_DELWRI bit is changed.
888  *
889  * MPSAFE
890  */
891 void
892 reassignbuf(struct buf *bp)
893 {
894 	struct vnode *vp = bp->b_vp;
895 	lwkt_tokref vlock;
896 	int delay;
897 
898 	KKASSERT(vp != NULL);
899 	++reassignbufcalls;
900 
901 	/*
902 	 * B_PAGING flagged buffers cannot be reassigned because their vp
903 	 * is not fully linked in.
904 	 */
905 	if (bp->b_flags & B_PAGING)
906 		panic("cannot reassign paging buffer");
907 
908 	lwkt_gettoken(&vlock, &vp->v_token);
909 	if (bp->b_flags & B_DELWRI) {
910 		/*
911 		 * Move to the dirty list, add the vnode to the worklist
912 		 */
913 		if (bp->b_flags & B_VNCLEAN) {
914 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
915 			bp->b_flags &= ~B_VNCLEAN;
916 		}
917 		if ((bp->b_flags & B_VNDIRTY) == 0) {
918 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
919 				panic("reassignbuf: dup lblk vp %p bp %p",
920 				      vp, bp);
921 			}
922 			bp->b_flags |= B_VNDIRTY;
923 		}
924 		if ((vp->v_flag & VONWORKLST) == 0) {
925 			switch (vp->v_type) {
926 			case VDIR:
927 				delay = dirdelay;
928 				break;
929 			case VCHR:
930 			case VBLK:
931 				if (vp->v_rdev &&
932 				    vp->v_rdev->si_mountpoint != NULL) {
933 					delay = metadelay;
934 					break;
935 				}
936 				/* fall through */
937 			default:
938 				delay = filedelay;
939 			}
940 			vn_syncer_add_to_worklist(vp, delay);
941 		}
942 	} else {
943 		/*
944 		 * Move to the clean list, remove the vnode from the worklist
945 		 * if no dirty blocks remain.
946 		 */
947 		if (bp->b_flags & B_VNDIRTY) {
948 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
949 			bp->b_flags &= ~B_VNDIRTY;
950 		}
951 		if ((bp->b_flags & B_VNCLEAN) == 0) {
952 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
953 				panic("reassignbuf: dup lblk vp %p bp %p",
954 				      vp, bp);
955 			}
956 			bp->b_flags |= B_VNCLEAN;
957 		}
958 		if ((vp->v_flag & VONWORKLST) &&
959 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
960 			vp->v_flag &= ~VONWORKLST;
961 			LIST_REMOVE(vp, v_synclist);
962 		}
963 	}
964 	lwkt_reltoken(&vlock);
965 }
966 
967 /*
968  * Create a vnode for a block device.
969  * Used for mounting the root file system.
970  */
971 extern struct vop_ops *devfs_vnode_dev_vops_p;
972 int
973 bdevvp(cdev_t dev, struct vnode **vpp)
974 {
975 	struct vnode *vp;
976 	struct vnode *nvp;
977 	int error;
978 
979 	if (dev == NULL) {
980 		*vpp = NULLVP;
981 		return (ENXIO);
982 	}
983 	error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p/*&spec_vnode_vops_p*/, &nvp, 0, 0);
984 	if (error) {
985 		*vpp = NULLVP;
986 		return (error);
987 	}
988 	vp = nvp;
989 	vp->v_type = VCHR;
990 	vp->v_rdev = dev;
991 	vp->v_umajor = dev->si_umajor;
992 	vp->v_uminor = dev->si_uminor;
993 	vx_unlock(vp);
994 	*vpp = vp;
995 	return (0);
996 }
997 
998 int
999 v_associate_rdev(struct vnode *vp, cdev_t dev)
1000 {
1001 	lwkt_tokref ilock;
1002 
1003 	if (dev == NULL)
1004 		return(ENXIO);
1005 	if (dev_is_good(dev) == 0)
1006 		return(ENXIO);
1007 	KKASSERT(vp->v_rdev == NULL);
1008 	if (dev_ref_debug)
1009 		kprintf("Z1");
1010 	vp->v_rdev = reference_dev(dev);
1011 	lwkt_gettoken(&ilock, &spechash_token);
1012 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1013 	lwkt_reltoken(&ilock);
1014 	return(0);
1015 }
1016 
1017 void
1018 v_release_rdev(struct vnode *vp)
1019 {
1020 	lwkt_tokref ilock;
1021 	cdev_t dev;
1022 
1023 	if ((dev = vp->v_rdev) != NULL) {
1024 		lwkt_gettoken(&ilock, &spechash_token);
1025 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1026 		vp->v_rdev = NULL;
1027 		release_dev(dev);
1028 		lwkt_reltoken(&ilock);
1029 	}
1030 }
1031 
1032 /*
1033  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1034  * the device number with the vnode.  The actual device is not associated
1035  * until the vnode is opened (usually in spec_open()), and will be
1036  * disassociated on last close.
1037  */
1038 void
1039 addaliasu(struct vnode *nvp, int x, int y)
1040 {
1041 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1042 		panic("addaliasu on non-special vnode");
1043 	nvp->v_umajor = x;
1044 	nvp->v_uminor = y;
1045 }
1046 
1047 /*
1048  * Simple call that a filesystem can make to try to get rid of a
1049  * vnode.  It will fail if anyone is referencing the vnode (including
1050  * the caller).
1051  *
1052  * The filesystem can check whether its in-memory inode structure still
1053  * references the vp on return.
1054  */
1055 void
1056 vclean_unlocked(struct vnode *vp)
1057 {
1058 	vx_get(vp);
1059 	if (sysref_isactive(&vp->v_sysref) == 0)
1060 		vgone_vxlocked(vp);
1061 	vx_put(vp);
1062 }
1063 
1064 /*
1065  * Disassociate a vnode from its underlying filesystem.
1066  *
1067  * The vnode must be VX locked and referenced.  In all normal situations
1068  * there are no active references.  If vclean_vxlocked() is called while
1069  * there are active references, the vnode is being ripped out and we have
1070  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1071  */
1072 void
1073 vclean_vxlocked(struct vnode *vp, int flags)
1074 {
1075 	int active;
1076 	int n;
1077 	vm_object_t object;
1078 
1079 	/*
1080 	 * If the vnode has already been reclaimed we have nothing to do.
1081 	 */
1082 	if (vp->v_flag & VRECLAIMED)
1083 		return;
1084 	vp->v_flag |= VRECLAIMED;
1085 
1086 	/*
1087 	 * Scrap the vfs cache
1088 	 */
1089 	while (cache_inval_vp(vp, 0) != 0) {
1090 		kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1091 		tsleep(vp, 0, "vclninv", 2);
1092 	}
1093 
1094 	/*
1095 	 * Check to see if the vnode is in use. If so we have to reference it
1096 	 * before we clean it out so that its count cannot fall to zero and
1097 	 * generate a race against ourselves to recycle it.
1098 	 */
1099 	active = sysref_isactive(&vp->v_sysref);
1100 
1101 	/*
1102 	 * Clean out any buffers associated with the vnode and destroy its
1103 	 * object, if it has one.
1104 	 */
1105 	vinvalbuf(vp, V_SAVE, 0, 0);
1106 
1107 	/*
1108 	 * If purging an active vnode (typically during a forced unmount
1109 	 * or reboot), it must be closed and deactivated before being
1110 	 * reclaimed.  This isn't really all that safe, but what can
1111 	 * we do? XXX.
1112 	 *
1113 	 * Note that neither of these routines unlocks the vnode.
1114 	 */
1115 	if (active && (flags & DOCLOSE)) {
1116 		while ((n = vp->v_opencount) != 0) {
1117 			if (vp->v_writecount)
1118 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1119 			else
1120 				VOP_CLOSE(vp, FNONBLOCK);
1121 			if (vp->v_opencount == n) {
1122 				kprintf("Warning: unable to force-close"
1123 				       " vnode %p\n", vp);
1124 				break;
1125 			}
1126 		}
1127 	}
1128 
1129 	/*
1130 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1131 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1132 	 * again to make sure they all get flushed.
1133 	 *
1134 	 * This can occur if a file with a link count of 0 needs to be
1135 	 * truncated.
1136 	 */
1137 	if ((vp->v_flag & VINACTIVE) == 0) {
1138 		vp->v_flag |= VINACTIVE;
1139 		VOP_INACTIVE(vp);
1140 		vinvalbuf(vp, V_SAVE, 0, 0);
1141 	}
1142 
1143 	/*
1144 	 * If the vnode has an object, destroy it.
1145 	 */
1146 	if ((object = vp->v_object) != NULL) {
1147 		if (object->ref_count == 0) {
1148 			if ((object->flags & OBJ_DEAD) == 0)
1149 				vm_object_terminate(object);
1150 		} else {
1151 			vm_pager_deallocate(object);
1152 		}
1153 		vp->v_flag &= ~VOBJBUF;
1154 	}
1155 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1156 
1157 	/*
1158 	 * Reclaim the vnode.
1159 	 */
1160 	if (VOP_RECLAIM(vp))
1161 		panic("vclean: cannot reclaim");
1162 
1163 	/*
1164 	 * Done with purge, notify sleepers of the grim news.
1165 	 */
1166 	vp->v_ops = &dead_vnode_vops_p;
1167 	vn_pollgone(vp);
1168 	vp->v_tag = VT_NON;
1169 
1170 	/*
1171 	 * If we are destroying an active vnode, reactivate it now that
1172 	 * we have reassociated it with deadfs.  This prevents the system
1173 	 * from crashing on the vnode due to it being unexpectedly marked
1174 	 * as inactive or reclaimed.
1175 	 */
1176 	if (active && (flags & DOCLOSE)) {
1177 		vp->v_flag &= ~(VINACTIVE|VRECLAIMED);
1178 	}
1179 }
1180 
1181 /*
1182  * Eliminate all activity associated with the requested vnode
1183  * and with all vnodes aliased to the requested vnode.
1184  *
1185  * The vnode must be referenced but should not be locked.
1186  */
1187 int
1188 vrevoke(struct vnode *vp, struct ucred *cred)
1189 {
1190 	struct vnode *vq;
1191 	struct vnode *vqn;
1192 	lwkt_tokref ilock;
1193 	cdev_t dev;
1194 	int error;
1195 
1196 	/*
1197 	 * If the vnode has a device association, scrap all vnodes associated
1198 	 * with the device.  Don't let the device disappear on us while we
1199 	 * are scrapping the vnodes.
1200 	 *
1201 	 * The passed vp will probably show up in the list, do not VX lock
1202 	 * it twice!
1203 	 *
1204 	 * Releasing the vnode's rdev here can mess up specfs's call to
1205 	 * device close, so don't do it.  The vnode has been disassociated
1206 	 * and the device will be closed after the last ref on the related
1207 	 * fp goes away (if not still open by e.g. the kernel).
1208 	 */
1209 	if (vp->v_type != VCHR) {
1210 		error = fdrevoke(vp, DTYPE_VNODE, cred);
1211 		return (error);
1212 	}
1213 	if ((dev = vp->v_rdev) == NULL) {
1214 		return(0);
1215 	}
1216 	reference_dev(dev);
1217 	lwkt_gettoken(&ilock, &spechash_token);
1218 
1219 	vqn = SLIST_FIRST(&dev->si_hlist);
1220 	if (vqn)
1221 		vref(vqn);
1222 	while ((vq = vqn) != NULL) {
1223 		vqn = SLIST_NEXT(vqn, v_cdevnext);
1224 		if (vqn)
1225 			vref(vqn);
1226 		fdrevoke(vq, DTYPE_VNODE, cred);
1227 		/*v_release_rdev(vq);*/
1228 		vrele(vq);
1229 	}
1230 	lwkt_reltoken(&ilock);
1231 	dev_drevoke(dev);
1232 	//release_dev(dev);
1233 	return (0);
1234 }
1235 
1236 /*
1237  * This is called when the object underlying a vnode is being destroyed,
1238  * such as in a remove().  Try to recycle the vnode immediately if the
1239  * only active reference is our reference.
1240  *
1241  * Directory vnodes in the namecache with children cannot be immediately
1242  * recycled because numerous VOP_N*() ops require them to be stable.
1243  */
1244 int
1245 vrecycle(struct vnode *vp)
1246 {
1247 	if (vp->v_sysref.refcnt <= 1) {
1248 		if (cache_inval_vp_nonblock(vp))
1249 			return(0);
1250 		vgone_vxlocked(vp);
1251 		return (1);
1252 	}
1253 	return (0);
1254 }
1255 
1256 /*
1257  * Return the maximum I/O size allowed for strategy calls on VP.
1258  *
1259  * If vp is VCHR or VBLK we dive the device, otherwise we use
1260  * the vp's mount info.
1261  */
1262 int
1263 vmaxiosize(struct vnode *vp)
1264 {
1265 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1266 		return(vp->v_rdev->si_iosize_max);
1267 	} else {
1268 		return(vp->v_mount->mnt_iosize_max);
1269 	}
1270 }
1271 
1272 /*
1273  * Eliminate all activity associated with a vnode in preparation for reuse.
1274  *
1275  * The vnode must be VX locked and refd and will remain VX locked and refd
1276  * on return.  This routine may be called with the vnode in any state, as
1277  * long as it is VX locked.  The vnode will be cleaned out and marked
1278  * VRECLAIMED but will not actually be reused until all existing refs and
1279  * holds go away.
1280  *
1281  * NOTE: This routine may be called on a vnode which has not yet been
1282  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1283  * already been reclaimed.
1284  *
1285  * This routine is not responsible for placing us back on the freelist.
1286  * Instead, it happens automatically when the caller releases the VX lock
1287  * (assuming there aren't any other references).
1288  */
1289 
1290 void
1291 vgone_vxlocked(struct vnode *vp)
1292 {
1293 	/*
1294 	 * assert that the VX lock is held.  This is an absolute requirement
1295 	 * now for vgone_vxlocked() to be called.
1296 	 */
1297 	KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1298 
1299 	/*
1300 	 * Clean out the filesystem specific data and set the VRECLAIMED
1301 	 * bit.  Also deactivate the vnode if necessary.
1302 	 */
1303 	vclean_vxlocked(vp, DOCLOSE);
1304 
1305 	/*
1306 	 * Delete from old mount point vnode list, if on one.
1307 	 */
1308 	if (vp->v_mount != NULL)
1309 		insmntque(vp, NULL);
1310 
1311 	/*
1312 	 * If special device, remove it from special device alias list
1313 	 * if it is on one.  This should normally only occur if a vnode is
1314 	 * being revoked as the device should otherwise have been released
1315 	 * naturally.
1316 	 */
1317 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1318 		v_release_rdev(vp);
1319 	}
1320 
1321 	/*
1322 	 * Set us to VBAD
1323 	 */
1324 	vp->v_type = VBAD;
1325 }
1326 
1327 /*
1328  * Lookup a vnode by device number.
1329  *
1330  * Returns non-zero and *vpp set to a vref'd vnode on success.
1331  * Returns zero on failure.
1332  */
1333 int
1334 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1335 {
1336 	lwkt_tokref ilock;
1337 	struct vnode *vp;
1338 
1339 	lwkt_gettoken(&ilock, &spechash_token);
1340 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1341 		if (type == vp->v_type) {
1342 			*vpp = vp;
1343 			vref(vp);
1344 			lwkt_reltoken(&ilock);
1345 			return (1);
1346 		}
1347 	}
1348 	lwkt_reltoken(&ilock);
1349 	return (0);
1350 }
1351 
1352 /*
1353  * Calculate the total number of references to a special device.  This
1354  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1355  * an overloaded field.  Since udev2dev can now return NULL, we have
1356  * to check for a NULL v_rdev.
1357  */
1358 int
1359 count_dev(cdev_t dev)
1360 {
1361 	lwkt_tokref ilock;
1362 	struct vnode *vp;
1363 	int count = 0;
1364 
1365 	if (SLIST_FIRST(&dev->si_hlist)) {
1366 		lwkt_gettoken(&ilock, &spechash_token);
1367 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1368 			if (vp->v_sysref.refcnt > 0)
1369 				count += vp->v_sysref.refcnt;
1370 		}
1371 		lwkt_reltoken(&ilock);
1372 	}
1373 	return(count);
1374 }
1375 
1376 int
1377 count_udev(int x, int y)
1378 {
1379 	cdev_t dev;
1380 
1381 	if ((dev = get_dev(x, y)) == NULL)
1382 		return(0);
1383 	return(count_dev(dev));
1384 }
1385 
1386 int
1387 vcount(struct vnode *vp)
1388 {
1389 	if (vp->v_rdev == NULL)
1390 		return(0);
1391 	return(count_dev(vp->v_rdev));
1392 }
1393 
1394 /*
1395  * Initialize VMIO for a vnode.  This routine MUST be called before a
1396  * VFS can issue buffer cache ops on a vnode.  It is typically called
1397  * when a vnode is initialized from its inode.
1398  */
1399 int
1400 vinitvmio(struct vnode *vp, off_t filesize)
1401 {
1402 	vm_object_t object;
1403 	int error = 0;
1404 
1405 retry:
1406 	if ((object = vp->v_object) == NULL) {
1407 		object = vnode_pager_alloc(vp, filesize, 0, 0);
1408 		/*
1409 		 * Dereference the reference we just created.  This assumes
1410 		 * that the object is associated with the vp.
1411 		 */
1412 		object->ref_count--;
1413 		vrele(vp);
1414 	} else {
1415 		if (object->flags & OBJ_DEAD) {
1416 			vn_unlock(vp);
1417 			vm_object_dead_sleep(object, "vodead");
1418 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1419 			goto retry;
1420 		}
1421 	}
1422 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1423 	vp->v_flag |= VOBJBUF;
1424 	return (error);
1425 }
1426 
1427 
1428 /*
1429  * Print out a description of a vnode.
1430  */
1431 static char *typename[] =
1432 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1433 
1434 void
1435 vprint(char *label, struct vnode *vp)
1436 {
1437 	char buf[96];
1438 
1439 	if (label != NULL)
1440 		kprintf("%s: %p: ", label, (void *)vp);
1441 	else
1442 		kprintf("%p: ", (void *)vp);
1443 	kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1444 		typename[vp->v_type],
1445 		vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1446 	buf[0] = '\0';
1447 	if (vp->v_flag & VROOT)
1448 		strcat(buf, "|VROOT");
1449 	if (vp->v_flag & VPFSROOT)
1450 		strcat(buf, "|VPFSROOT");
1451 	if (vp->v_flag & VTEXT)
1452 		strcat(buf, "|VTEXT");
1453 	if (vp->v_flag & VSYSTEM)
1454 		strcat(buf, "|VSYSTEM");
1455 	if (vp->v_flag & VFREE)
1456 		strcat(buf, "|VFREE");
1457 	if (vp->v_flag & VOBJBUF)
1458 		strcat(buf, "|VOBJBUF");
1459 	if (buf[0] != '\0')
1460 		kprintf(" flags (%s)", &buf[1]);
1461 	if (vp->v_data == NULL) {
1462 		kprintf("\n");
1463 	} else {
1464 		kprintf("\n\t");
1465 		VOP_PRINT(vp);
1466 	}
1467 }
1468 
1469 #ifdef DDB
1470 #include <ddb/ddb.h>
1471 
1472 static int db_show_locked_vnodes(struct mount *mp, void *data);
1473 
1474 /*
1475  * List all of the locked vnodes in the system.
1476  * Called when debugging the kernel.
1477  */
1478 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1479 {
1480 	kprintf("Locked vnodes\n");
1481 	mountlist_scan(db_show_locked_vnodes, NULL,
1482 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1483 }
1484 
1485 static int
1486 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1487 {
1488 	struct vnode *vp;
1489 
1490 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1491 		if (vn_islocked(vp))
1492 			vprint(NULL, vp);
1493 	}
1494 	return(0);
1495 }
1496 #endif
1497 
1498 /*
1499  * Top level filesystem related information gathering.
1500  */
1501 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1502 
1503 static int
1504 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1505 {
1506 	int *name = (int *)arg1 - 1;	/* XXX */
1507 	u_int namelen = arg2 + 1;	/* XXX */
1508 	struct vfsconf *vfsp;
1509 	int maxtypenum;
1510 
1511 #if 1 || defined(COMPAT_PRELITE2)
1512 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1513 	if (namelen == 1)
1514 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1515 #endif
1516 
1517 #ifdef notyet
1518 	/* all sysctl names at this level are at least name and field */
1519 	if (namelen < 2)
1520 		return (ENOTDIR);		/* overloaded */
1521 	if (name[0] != VFS_GENERIC) {
1522 		vfsp = vfsconf_find_by_typenum(name[0]);
1523 		if (vfsp == NULL)
1524 			return (EOPNOTSUPP);
1525 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1526 		    oldp, oldlenp, newp, newlen, p));
1527 	}
1528 #endif
1529 	switch (name[1]) {
1530 	case VFS_MAXTYPENUM:
1531 		if (namelen != 2)
1532 			return (ENOTDIR);
1533 		maxtypenum = vfsconf_get_maxtypenum();
1534 		return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1535 	case VFS_CONF:
1536 		if (namelen != 3)
1537 			return (ENOTDIR);	/* overloaded */
1538 		vfsp = vfsconf_find_by_typenum(name[2]);
1539 		if (vfsp == NULL)
1540 			return (EOPNOTSUPP);
1541 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1542 	}
1543 	return (EOPNOTSUPP);
1544 }
1545 
1546 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1547 	"Generic filesystem");
1548 
1549 #if 1 || defined(COMPAT_PRELITE2)
1550 
1551 static int
1552 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1553 {
1554 	int error;
1555 	struct ovfsconf ovfs;
1556 	struct sysctl_req *req = (struct sysctl_req*) data;
1557 
1558 	bzero(&ovfs, sizeof(ovfs));
1559 	ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1560 	strcpy(ovfs.vfc_name, vfsp->vfc_name);
1561 	ovfs.vfc_index = vfsp->vfc_typenum;
1562 	ovfs.vfc_refcount = vfsp->vfc_refcount;
1563 	ovfs.vfc_flags = vfsp->vfc_flags;
1564 	error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1565 	if (error)
1566 		return error; /* abort iteration with error code */
1567 	else
1568 		return 0; /* continue iterating with next element */
1569 }
1570 
1571 static int
1572 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1573 {
1574 	return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1575 }
1576 
1577 #endif /* 1 || COMPAT_PRELITE2 */
1578 
1579 /*
1580  * Check to see if a filesystem is mounted on a block device.
1581  */
1582 int
1583 vfs_mountedon(struct vnode *vp)
1584 {
1585 	cdev_t dev;
1586 
1587 	if ((dev = vp->v_rdev) == NULL) {
1588 /*		if (vp->v_type != VBLK)
1589 			dev = get_dev(vp->v_uminor, vp->v_umajor); */
1590 	}
1591 	if (dev != NULL && dev->si_mountpoint)
1592 		return (EBUSY);
1593 	return (0);
1594 }
1595 
1596 /*
1597  * Unmount all filesystems. The list is traversed in reverse order
1598  * of mounting to avoid dependencies.
1599  */
1600 
1601 static int vfs_umountall_callback(struct mount *mp, void *data);
1602 
1603 void
1604 vfs_unmountall(void)
1605 {
1606 	int count;
1607 
1608 	do {
1609 		count = mountlist_scan(vfs_umountall_callback,
1610 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1611 	} while (count);
1612 }
1613 
1614 static
1615 int
1616 vfs_umountall_callback(struct mount *mp, void *data)
1617 {
1618 	int error;
1619 
1620 	error = dounmount(mp, MNT_FORCE);
1621 	if (error) {
1622 		mountlist_remove(mp);
1623 		kprintf("unmount of filesystem mounted from %s failed (",
1624 			mp->mnt_stat.f_mntfromname);
1625 		if (error == EBUSY)
1626 			kprintf("BUSY)\n");
1627 		else
1628 			kprintf("%d)\n", error);
1629 	}
1630 	return(1);
1631 }
1632 
1633 /*
1634  * Build hash lists of net addresses and hang them off the mount point.
1635  * Called by ufs_mount() to set up the lists of export addresses.
1636  */
1637 static int
1638 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1639 		const struct export_args *argp)
1640 {
1641 	struct netcred *np;
1642 	struct radix_node_head *rnh;
1643 	int i;
1644 	struct radix_node *rn;
1645 	struct sockaddr *saddr, *smask = 0;
1646 	struct domain *dom;
1647 	int error;
1648 
1649 	if (argp->ex_addrlen == 0) {
1650 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1651 			return (EPERM);
1652 		np = &nep->ne_defexported;
1653 		np->netc_exflags = argp->ex_flags;
1654 		np->netc_anon = argp->ex_anon;
1655 		np->netc_anon.cr_ref = 1;
1656 		mp->mnt_flag |= MNT_DEFEXPORTED;
1657 		return (0);
1658 	}
1659 
1660 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1661 		return (EINVAL);
1662 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1663 		return (EINVAL);
1664 
1665 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1666 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1667 	saddr = (struct sockaddr *) (np + 1);
1668 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1669 		goto out;
1670 	if (saddr->sa_len > argp->ex_addrlen)
1671 		saddr->sa_len = argp->ex_addrlen;
1672 	if (argp->ex_masklen) {
1673 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1674 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1675 		if (error)
1676 			goto out;
1677 		if (smask->sa_len > argp->ex_masklen)
1678 			smask->sa_len = argp->ex_masklen;
1679 	}
1680 	i = saddr->sa_family;
1681 	if ((rnh = nep->ne_rtable[i]) == 0) {
1682 		/*
1683 		 * Seems silly to initialize every AF when most are not used,
1684 		 * do so on demand here
1685 		 */
1686 		SLIST_FOREACH(dom, &domains, dom_next)
1687 			if (dom->dom_family == i && dom->dom_rtattach) {
1688 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1689 				    dom->dom_rtoffset);
1690 				break;
1691 			}
1692 		if ((rnh = nep->ne_rtable[i]) == 0) {
1693 			error = ENOBUFS;
1694 			goto out;
1695 		}
1696 	}
1697 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1698 	    np->netc_rnodes);
1699 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1700 		error = EPERM;
1701 		goto out;
1702 	}
1703 	np->netc_exflags = argp->ex_flags;
1704 	np->netc_anon = argp->ex_anon;
1705 	np->netc_anon.cr_ref = 1;
1706 	return (0);
1707 out:
1708 	kfree(np, M_NETADDR);
1709 	return (error);
1710 }
1711 
1712 /* ARGSUSED */
1713 static int
1714 vfs_free_netcred(struct radix_node *rn, void *w)
1715 {
1716 	struct radix_node_head *rnh = (struct radix_node_head *) w;
1717 
1718 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1719 	kfree((caddr_t) rn, M_NETADDR);
1720 	return (0);
1721 }
1722 
1723 /*
1724  * Free the net address hash lists that are hanging off the mount points.
1725  */
1726 static void
1727 vfs_free_addrlist(struct netexport *nep)
1728 {
1729 	int i;
1730 	struct radix_node_head *rnh;
1731 
1732 	for (i = 0; i <= AF_MAX; i++)
1733 		if ((rnh = nep->ne_rtable[i])) {
1734 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1735 			    (caddr_t) rnh);
1736 			kfree((caddr_t) rnh, M_RTABLE);
1737 			nep->ne_rtable[i] = 0;
1738 		}
1739 }
1740 
1741 int
1742 vfs_export(struct mount *mp, struct netexport *nep,
1743 	   const struct export_args *argp)
1744 {
1745 	int error;
1746 
1747 	if (argp->ex_flags & MNT_DELEXPORT) {
1748 		if (mp->mnt_flag & MNT_EXPUBLIC) {
1749 			vfs_setpublicfs(NULL, NULL, NULL);
1750 			mp->mnt_flag &= ~MNT_EXPUBLIC;
1751 		}
1752 		vfs_free_addrlist(nep);
1753 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1754 	}
1755 	if (argp->ex_flags & MNT_EXPORTED) {
1756 		if (argp->ex_flags & MNT_EXPUBLIC) {
1757 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1758 				return (error);
1759 			mp->mnt_flag |= MNT_EXPUBLIC;
1760 		}
1761 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1762 			return (error);
1763 		mp->mnt_flag |= MNT_EXPORTED;
1764 	}
1765 	return (0);
1766 }
1767 
1768 
1769 /*
1770  * Set the publicly exported filesystem (WebNFS). Currently, only
1771  * one public filesystem is possible in the spec (RFC 2054 and 2055)
1772  */
1773 int
1774 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1775 		const struct export_args *argp)
1776 {
1777 	int error;
1778 	struct vnode *rvp;
1779 	char *cp;
1780 
1781 	/*
1782 	 * mp == NULL -> invalidate the current info, the FS is
1783 	 * no longer exported. May be called from either vfs_export
1784 	 * or unmount, so check if it hasn't already been done.
1785 	 */
1786 	if (mp == NULL) {
1787 		if (nfs_pub.np_valid) {
1788 			nfs_pub.np_valid = 0;
1789 			if (nfs_pub.np_index != NULL) {
1790 				FREE(nfs_pub.np_index, M_TEMP);
1791 				nfs_pub.np_index = NULL;
1792 			}
1793 		}
1794 		return (0);
1795 	}
1796 
1797 	/*
1798 	 * Only one allowed at a time.
1799 	 */
1800 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1801 		return (EBUSY);
1802 
1803 	/*
1804 	 * Get real filehandle for root of exported FS.
1805 	 */
1806 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1807 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1808 
1809 	if ((error = VFS_ROOT(mp, &rvp)))
1810 		return (error);
1811 
1812 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1813 		return (error);
1814 
1815 	vput(rvp);
1816 
1817 	/*
1818 	 * If an indexfile was specified, pull it in.
1819 	 */
1820 	if (argp->ex_indexfile != NULL) {
1821 		int namelen;
1822 
1823 		error = vn_get_namelen(rvp, &namelen);
1824 		if (error)
1825 			return (error);
1826 		MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1827 		    M_WAITOK);
1828 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1829 		    namelen, NULL);
1830 		if (!error) {
1831 			/*
1832 			 * Check for illegal filenames.
1833 			 */
1834 			for (cp = nfs_pub.np_index; *cp; cp++) {
1835 				if (*cp == '/') {
1836 					error = EINVAL;
1837 					break;
1838 				}
1839 			}
1840 		}
1841 		if (error) {
1842 			FREE(nfs_pub.np_index, M_TEMP);
1843 			return (error);
1844 		}
1845 	}
1846 
1847 	nfs_pub.np_mount = mp;
1848 	nfs_pub.np_valid = 1;
1849 	return (0);
1850 }
1851 
1852 struct netcred *
1853 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1854 		struct sockaddr *nam)
1855 {
1856 	struct netcred *np;
1857 	struct radix_node_head *rnh;
1858 	struct sockaddr *saddr;
1859 
1860 	np = NULL;
1861 	if (mp->mnt_flag & MNT_EXPORTED) {
1862 		/*
1863 		 * Lookup in the export list first.
1864 		 */
1865 		if (nam != NULL) {
1866 			saddr = nam;
1867 			rnh = nep->ne_rtable[saddr->sa_family];
1868 			if (rnh != NULL) {
1869 				np = (struct netcred *)
1870 					(*rnh->rnh_matchaddr)((char *)saddr,
1871 							      rnh);
1872 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1873 					np = NULL;
1874 			}
1875 		}
1876 		/*
1877 		 * If no address match, use the default if it exists.
1878 		 */
1879 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1880 			np = &nep->ne_defexported;
1881 	}
1882 	return (np);
1883 }
1884 
1885 /*
1886  * perform msync on all vnodes under a mount point.  The mount point must
1887  * be locked.  This code is also responsible for lazy-freeing unreferenced
1888  * vnodes whos VM objects no longer contain pages.
1889  *
1890  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1891  *
1892  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1893  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
1894  * way up in this high level function.
1895  */
1896 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1897 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1898 
1899 void
1900 vfs_msync(struct mount *mp, int flags)
1901 {
1902 	int vmsc_flags;
1903 
1904 	vmsc_flags = VMSC_GETVP;
1905 	if (flags != MNT_WAIT)
1906 		vmsc_flags |= VMSC_NOWAIT;
1907 	vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1908 			(void *)(intptr_t)flags);
1909 }
1910 
1911 /*
1912  * scan1 is a fast pre-check.  There could be hundreds of thousands of
1913  * vnodes, we cannot afford to do anything heavy weight until we have a
1914  * fairly good indication that there is work to do.
1915  */
1916 static
1917 int
1918 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1919 {
1920 	int flags = (int)(intptr_t)data;
1921 
1922 	if ((vp->v_flag & VRECLAIMED) == 0) {
1923 		if (vshouldmsync(vp))
1924 			return(0);	/* call scan2 */
1925 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1926 		    (vp->v_flag & VOBJDIRTY) &&
1927 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
1928 			return(0);	/* call scan2 */
1929 		}
1930 	}
1931 
1932 	/*
1933 	 * do not call scan2, continue the loop
1934 	 */
1935 	return(-1);
1936 }
1937 
1938 /*
1939  * This callback is handed a locked vnode.
1940  */
1941 static
1942 int
1943 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1944 {
1945 	vm_object_t obj;
1946 	int flags = (int)(intptr_t)data;
1947 
1948 	if (vp->v_flag & VRECLAIMED)
1949 		return(0);
1950 
1951 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1952 		if ((obj = vp->v_object) != NULL) {
1953 			vm_object_page_clean(obj, 0, 0,
1954 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1955 		}
1956 	}
1957 	return(0);
1958 }
1959 
1960 /*
1961  * Record a process's interest in events which might happen to
1962  * a vnode.  Because poll uses the historic select-style interface
1963  * internally, this routine serves as both the ``check for any
1964  * pending events'' and the ``record my interest in future events''
1965  * functions.  (These are done together, while the lock is held,
1966  * to avoid race conditions.)
1967  */
1968 int
1969 vn_pollrecord(struct vnode *vp, int events)
1970 {
1971 	lwkt_tokref vlock;
1972 
1973 	KKASSERT(curthread->td_proc != NULL);
1974 
1975 	lwkt_gettoken(&vlock, &vp->v_token);
1976 	if (vp->v_pollinfo.vpi_revents & events) {
1977 		/*
1978 		 * This leaves events we are not interested
1979 		 * in available for the other process which
1980 		 * which presumably had requested them
1981 		 * (otherwise they would never have been
1982 		 * recorded).
1983 		 */
1984 		events &= vp->v_pollinfo.vpi_revents;
1985 		vp->v_pollinfo.vpi_revents &= ~events;
1986 
1987 		lwkt_reltoken(&vlock);
1988 		return events;
1989 	}
1990 	vp->v_pollinfo.vpi_events |= events;
1991 	selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
1992 	lwkt_reltoken(&vlock);
1993 	return 0;
1994 }
1995 
1996 /*
1997  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
1998  * it is possible for us to miss an event due to race conditions, but
1999  * that condition is expected to be rare, so for the moment it is the
2000  * preferred interface.
2001  */
2002 void
2003 vn_pollevent(struct vnode *vp, int events)
2004 {
2005 	lwkt_tokref vlock;
2006 
2007 	lwkt_gettoken(&vlock, &vp->v_token);
2008 	if (vp->v_pollinfo.vpi_events & events) {
2009 		/*
2010 		 * We clear vpi_events so that we don't
2011 		 * call selwakeup() twice if two events are
2012 		 * posted before the polling process(es) is
2013 		 * awakened.  This also ensures that we take at
2014 		 * most one selwakeup() if the polling process
2015 		 * is no longer interested.  However, it does
2016 		 * mean that only one event can be noticed at
2017 		 * a time.  (Perhaps we should only clear those
2018 		 * event bits which we note?) XXX
2019 		 */
2020 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2021 		vp->v_pollinfo.vpi_revents |= events;
2022 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2023 	}
2024 	lwkt_reltoken(&vlock);
2025 }
2026 
2027 /*
2028  * Wake up anyone polling on vp because it is being revoked.
2029  * This depends on dead_poll() returning POLLHUP for correct
2030  * behavior.
2031  */
2032 void
2033 vn_pollgone(struct vnode *vp)
2034 {
2035 	lwkt_tokref vlock;
2036 
2037 	lwkt_gettoken(&vlock, &vp->v_token);
2038 	if (vp->v_pollinfo.vpi_events) {
2039 		vp->v_pollinfo.vpi_events = 0;
2040 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2041 	}
2042 	lwkt_reltoken(&vlock);
2043 }
2044 
2045 /*
2046  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2047  * (or v_rdev might be NULL).
2048  */
2049 cdev_t
2050 vn_todev(struct vnode *vp)
2051 {
2052 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2053 		return (NULL);
2054 	KKASSERT(vp->v_rdev != NULL);
2055 	return (vp->v_rdev);
2056 }
2057 
2058 /*
2059  * Check if vnode represents a disk device.  The vnode does not need to be
2060  * opened.
2061  *
2062  * MPALMOSTSAFE
2063  */
2064 int
2065 vn_isdisk(struct vnode *vp, int *errp)
2066 {
2067 	cdev_t dev;
2068 
2069 	if (vp->v_type != VCHR) {
2070 		if (errp != NULL)
2071 			*errp = ENOTBLK;
2072 		return (0);
2073 	}
2074 
2075 	dev = vp->v_rdev;
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;
2101 	register_t retval[2];
2102 
2103 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2104 	if (error)
2105 		return (error);
2106 	*namelen = (int)retval[0];
2107 	return (0);
2108 }
2109 
2110 int
2111 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2112 		uint16_t d_namlen, const char *d_name)
2113 {
2114 	struct dirent *dp;
2115 	size_t len;
2116 
2117 	len = _DIRENT_RECLEN(d_namlen);
2118 	if (len > uio->uio_resid)
2119 		return(1);
2120 
2121 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2122 
2123 	dp->d_ino = d_ino;
2124 	dp->d_namlen = d_namlen;
2125 	dp->d_type = d_type;
2126 	bcopy(d_name, dp->d_name, d_namlen);
2127 
2128 	*error = uiomove((caddr_t)dp, len, uio);
2129 
2130 	kfree(dp, M_TEMP);
2131 
2132 	return(0);
2133 }
2134 
2135 void
2136 vn_mark_atime(struct vnode *vp, struct thread *td)
2137 {
2138 	struct proc *p = td->td_proc;
2139 	struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2140 
2141 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2142 		VOP_MARKATIME(vp, cred);
2143 	}
2144 }
2145