xref: /dflybsd-src/sys/kern/vfs_subr.c (revision ff837cd5d97df9cfc8d66fc9ebc73557657d023e)
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
36  */
37 
38 /*
39  * External virtual filesystem routines
40  */
41 #include "opt_ddb.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/dirent.h>
48 #include <sys/domain.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
51 #include <sys/file.h>
52 #include <sys/kernel.h>
53 #include <sys/kthread.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/mount.h>
57 #include <sys/priv.h>
58 #include <sys/proc.h>
59 #include <sys/reboot.h>
60 #include <sys/socket.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/syslog.h>
64 #include <sys/unistd.h>
65 #include <sys/vmmeter.h>
66 #include <sys/vnode.h>
67 
68 #include <machine/limits.h>
69 
70 #include <vm/vm.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_kern.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_pager.h>
78 #include <vm/vnode_pager.h>
79 #include <vm/vm_zone.h>
80 
81 #include <sys/buf2.h>
82 #include <sys/thread2.h>
83 #include <sys/sysref2.h>
84 #include <sys/mplock2.h>
85 
86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
87 
88 int numvnodes;
89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
90     "Number of vnodes allocated");
91 int verbose_reclaims;
92 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0,
93     "Output filename of reclaimed vnode(s)");
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, &reassignbufcalls,
106     0, "Number of times buffers have been reassigned to the proper list");
107 
108 static int check_buf_overlap = 2;	/* invasive check */
109 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap,
110     0, "Enable overlapping buffer checks");
111 
112 int	nfs_mount_type = -1;
113 static struct lwkt_token spechash_token;
114 struct nfs_public nfs_pub;	/* publicly exported FS */
115 
116 int desiredvnodes;
117 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
118 		&desiredvnodes, 0, "Maximum number of vnodes");
119 
120 static void	vfs_free_addrlist (struct netexport *nep);
121 static int	vfs_free_netcred (struct radix_node *rn, void *w);
122 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
123 				       const struct export_args *argp);
124 
125 /*
126  * Red black tree functions
127  */
128 static int rb_buf_compare(struct buf *b1, struct buf *b2);
129 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
130 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
131 
132 static int
133 rb_buf_compare(struct buf *b1, struct buf *b2)
134 {
135 	if (b1->b_loffset < b2->b_loffset)
136 		return(-1);
137 	if (b1->b_loffset > b2->b_loffset)
138 		return(1);
139 	return(0);
140 }
141 
142 /*
143  * Returns non-zero if the vnode is a candidate for lazy msyncing.
144  *
145  * NOTE: v_object is not stable (this scan can race), however the
146  *	 mntvnodescan code holds vmobj_token so any VM object we
147  *	 do find will remain stable storage.
148  */
149 static __inline int
150 vshouldmsync(struct vnode *vp)
151 {
152 	vm_object_t object;
153 
154 	if (vp->v_auxrefs != 0 || VREFCNT(vp) > 0)
155 		return (0);		/* other holders */
156 	object = vp->v_object;
157 	cpu_ccfence();
158 	if (object && (object->ref_count || object->resident_page_count))
159 		return(0);
160 	return (1);
161 }
162 
163 /*
164  * Initialize the vnode management data structures.
165  *
166  * Called from vfsinit()
167  */
168 void
169 vfs_subr_init(void)
170 {
171 	int factor1;
172 	int factor2;
173 
174 	/*
175 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
176 	 * according to available system memory but we may also have
177 	 * to limit it based on available KVM, which is capped on 32 bit
178 	 * systems, to ~80K vnodes or so.
179 	 *
180 	 * WARNING!  For machines with 64-256M of ram we have to be sure
181 	 *	     that the default limit scales down well due to HAMMER
182 	 *	     taking up significantly more memory per-vnode vs UFS.
183 	 *	     We want around ~5800 on a 128M machine.
184 	 */
185 	factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
186 	factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode));
187 	desiredvnodes =
188 		imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
189 		     KvaSize / factor2);
190 	desiredvnodes = imax(desiredvnodes, maxproc * 8);
191 
192 	lwkt_token_init(&spechash_token, "spechash");
193 }
194 
195 /*
196  * Knob to control the precision of file timestamps:
197  *
198  *   0 = seconds only; nanoseconds zeroed.
199  *   1 = seconds and nanoseconds, accurate within 1/HZ.
200  *   2 = seconds and nanoseconds, truncated to microseconds.
201  * >=3 = seconds and nanoseconds, maximum precision.
202  */
203 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
204 
205 static int timestamp_precision = TSP_SEC;
206 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
207 		&timestamp_precision, 0, "Precision of file timestamps");
208 
209 /*
210  * Get a current timestamp.
211  *
212  * MPSAFE
213  */
214 void
215 vfs_timestamp(struct timespec *tsp)
216 {
217 	struct timeval tv;
218 
219 	switch (timestamp_precision) {
220 	case TSP_SEC:
221 		tsp->tv_sec = time_second;
222 		tsp->tv_nsec = 0;
223 		break;
224 	case TSP_HZ:
225 		getnanotime(tsp);
226 		break;
227 	case TSP_USEC:
228 		microtime(&tv);
229 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
230 		break;
231 	case TSP_NSEC:
232 	default:
233 		nanotime(tsp);
234 		break;
235 	}
236 }
237 
238 /*
239  * Set vnode attributes to VNOVAL
240  */
241 void
242 vattr_null(struct vattr *vap)
243 {
244 	vap->va_type = VNON;
245 	vap->va_size = VNOVAL;
246 	vap->va_bytes = VNOVAL;
247 	vap->va_mode = VNOVAL;
248 	vap->va_nlink = VNOVAL;
249 	vap->va_uid = VNOVAL;
250 	vap->va_gid = VNOVAL;
251 	vap->va_fsid = VNOVAL;
252 	vap->va_fileid = VNOVAL;
253 	vap->va_blocksize = VNOVAL;
254 	vap->va_rmajor = VNOVAL;
255 	vap->va_rminor = VNOVAL;
256 	vap->va_atime.tv_sec = VNOVAL;
257 	vap->va_atime.tv_nsec = VNOVAL;
258 	vap->va_mtime.tv_sec = VNOVAL;
259 	vap->va_mtime.tv_nsec = VNOVAL;
260 	vap->va_ctime.tv_sec = VNOVAL;
261 	vap->va_ctime.tv_nsec = VNOVAL;
262 	vap->va_flags = VNOVAL;
263 	vap->va_gen = VNOVAL;
264 	vap->va_vaflags = 0;
265 	/* va_*_uuid fields are only valid if related flags are set */
266 }
267 
268 /*
269  * Flush out and invalidate all buffers associated with a vnode.
270  *
271  * vp must be locked.
272  */
273 static int vinvalbuf_bp(struct buf *bp, void *data);
274 
275 struct vinvalbuf_bp_info {
276 	struct vnode *vp;
277 	int slptimeo;
278 	int lkflags;
279 	int flags;
280 	int clean;
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 	int error;
289 
290 	lwkt_gettoken(&vp->v_token);
291 
292 	/*
293 	 * If we are being asked to save, call fsync to ensure that the inode
294 	 * is updated.
295 	 */
296 	if (flags & V_SAVE) {
297 		error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
298 		if (error)
299 			goto done;
300 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
301 			if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
302 				goto done;
303 #if 0
304 			/*
305 			 * Dirty bufs may be left or generated via races
306 			 * in circumstances where vinvalbuf() is called on
307 			 * a vnode not undergoing reclamation.   Only
308 			 * panic if we are trying to reclaim the vnode.
309 			 */
310 			if ((vp->v_flag & VRECLAIMED) &&
311 			    (bio_track_active(&vp->v_track_write) ||
312 			    !RB_EMPTY(&vp->v_rbdirty_tree))) {
313 				panic("vinvalbuf: dirty bufs");
314 			}
315 #endif
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, wait for all I/O
327 	 * to complete.  At least one pass is required.  We might block
328 	 * in the pip code so we have to re-check.  Order is important.
329 	 */
330 	do {
331 		/*
332 		 * Flush buffer cache
333 		 */
334 		if (!RB_EMPTY(&vp->v_rbclean_tree)) {
335 			info.clean = 1;
336 			error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
337 					NULL, vinvalbuf_bp, &info);
338 		}
339 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
340 			info.clean = 0;
341 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
342 					NULL, vinvalbuf_bp, &info);
343 		}
344 
345 		/*
346 		 * Wait for I/O completion.
347 		 */
348 		bio_track_wait(&vp->v_track_write, 0, 0);
349 		if ((object = vp->v_object) != NULL)
350 			refcount_wait(&object->paging_in_progress, "vnvlbx");
351 	} while (bio_track_active(&vp->v_track_write) ||
352 		 !RB_EMPTY(&vp->v_rbclean_tree) ||
353 		 !RB_EMPTY(&vp->v_rbdirty_tree));
354 
355 	/*
356 	 * Destroy the copy in the VM cache, too.
357 	 */
358 	if ((object = vp->v_object) != NULL) {
359 		vm_object_page_remove(object, 0, 0,
360 			(flags & V_SAVE) ? TRUE : FALSE);
361 	}
362 
363 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
364 		panic("vinvalbuf: flush failed");
365 	if (!RB_EMPTY(&vp->v_rbhash_tree))
366 		panic("vinvalbuf: flush failed, buffers still present");
367 	error = 0;
368 done:
369 	lwkt_reltoken(&vp->v_token);
370 	return (error);
371 }
372 
373 static int
374 vinvalbuf_bp(struct buf *bp, void *data)
375 {
376 	struct vinvalbuf_bp_info *info = data;
377 	int error;
378 
379 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
380 		atomic_add_int(&bp->b_refs, 1);
381 		error = BUF_TIMELOCK(bp, info->lkflags,
382 				     "vinvalbuf", info->slptimeo);
383 		atomic_subtract_int(&bp->b_refs, 1);
384 		if (error == 0) {
385 			BUF_UNLOCK(bp);
386 			error = ENOLCK;
387 		}
388 		if (error == ENOLCK)
389 			return(0);
390 		return (-error);
391 	}
392 	KKASSERT(bp->b_vp == info->vp);
393 
394 	/*
395 	 * Must check clean/dirty status after successfully locking as
396 	 * it may race.
397 	 */
398 	if ((info->clean && (bp->b_flags & B_DELWRI)) ||
399 	    (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) {
400 		BUF_UNLOCK(bp);
401 		return(0);
402 	}
403 
404 	/*
405 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
406 	 * check.  This code will write out the buffer, period.
407 	 */
408 	bremfree(bp);
409 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
410 	    (info->flags & V_SAVE)) {
411 		cluster_awrite(bp);
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 		bp->b_flags |= (B_INVAL | B_RELBUF);
419 		brelse(bp);
420 	} else {
421 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
422 		brelse(bp);
423 	}
424 	return(0);
425 }
426 
427 /*
428  * Truncate a file's buffer and pages to a specified length.  This
429  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
430  * sync activity.
431  *
432  * The vnode must be locked.
433  */
434 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
435 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
436 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
437 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
438 
439 struct vtruncbuf_info {
440 	struct vnode *vp;
441 	off_t	truncloffset;
442 	int	clean;
443 };
444 
445 int
446 vtruncbuf(struct vnode *vp, off_t length, int blksize)
447 {
448 	struct vtruncbuf_info info;
449 	const char *filename;
450 	int count;
451 
452 	/*
453 	 * Round up to the *next* block, then destroy the buffers in question.
454 	 * Since we are only removing some of the buffers we must rely on the
455 	 * scan count to determine whether a loop is necessary.
456 	 */
457 	if ((count = (int)(length % blksize)) != 0)
458 		info.truncloffset = length + (blksize - count);
459 	else
460 		info.truncloffset = length;
461 	info.vp = vp;
462 
463 	lwkt_gettoken(&vp->v_token);
464 	do {
465 		info.clean = 1;
466 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
467 				vtruncbuf_bp_trunc_cmp,
468 				vtruncbuf_bp_trunc, &info);
469 		info.clean = 0;
470 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
471 				vtruncbuf_bp_trunc_cmp,
472 				vtruncbuf_bp_trunc, &info);
473 	} while(count);
474 
475 	/*
476 	 * For safety, fsync any remaining metadata if the file is not being
477 	 * truncated to 0.  Since the metadata does not represent the entire
478 	 * dirty list we have to rely on the hit count to ensure that we get
479 	 * all of it.
480 	 */
481 	if (length > 0) {
482 		do {
483 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
484 					vtruncbuf_bp_metasync_cmp,
485 					vtruncbuf_bp_metasync, &info);
486 		} while (count);
487 	}
488 
489 	/*
490 	 * Clean out any left over VM backing store.
491 	 *
492 	 * It is possible to have in-progress I/O from buffers that were
493 	 * not part of the truncation.  This should not happen if we
494 	 * are truncating to 0-length.
495 	 */
496 	vnode_pager_setsize(vp, length);
497 	bio_track_wait(&vp->v_track_write, 0, 0);
498 
499 	/*
500 	 * Debugging only
501 	 */
502 	spin_lock(&vp->v_spin);
503 	filename = TAILQ_FIRST(&vp->v_namecache) ?
504 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
505 	spin_unlock(&vp->v_spin);
506 
507 	/*
508 	 * Make sure no buffers were instantiated while we were trying
509 	 * to clean out the remaining VM pages.  This could occur due
510 	 * to busy dirty VM pages being flushed out to disk.
511 	 */
512 	do {
513 		info.clean = 1;
514 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
515 				vtruncbuf_bp_trunc_cmp,
516 				vtruncbuf_bp_trunc, &info);
517 		info.clean = 0;
518 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
519 				vtruncbuf_bp_trunc_cmp,
520 				vtruncbuf_bp_trunc, &info);
521 		if (count) {
522 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
523 			       "left over buffers in %s\n", count, filename);
524 		}
525 	} while(count);
526 
527 	lwkt_reltoken(&vp->v_token);
528 
529 	return (0);
530 }
531 
532 /*
533  * The callback buffer is beyond the new file EOF and must be destroyed.
534  * Note that the compare function must conform to the RB_SCAN's requirements.
535  */
536 static
537 int
538 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
539 {
540 	struct vtruncbuf_info *info = data;
541 
542 	if (bp->b_loffset >= info->truncloffset)
543 		return(0);
544 	return(-1);
545 }
546 
547 static
548 int
549 vtruncbuf_bp_trunc(struct buf *bp, void *data)
550 {
551 	struct vtruncbuf_info *info = data;
552 
553 	/*
554 	 * Do not try to use a buffer we cannot immediately lock, but sleep
555 	 * anyway to prevent a livelock.  The code will loop until all buffers
556 	 * can be acted upon.
557 	 *
558 	 * We must always revalidate the buffer after locking it to deal
559 	 * with MP races.
560 	 */
561 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
562 		atomic_add_int(&bp->b_refs, 1);
563 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
564 			BUF_UNLOCK(bp);
565 		atomic_subtract_int(&bp->b_refs, 1);
566 	} else if ((info->clean && (bp->b_flags & B_DELWRI)) ||
567 		   (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) ||
568 		   bp->b_vp != info->vp ||
569 		   vtruncbuf_bp_trunc_cmp(bp, data)) {
570 		BUF_UNLOCK(bp);
571 	} else {
572 		bremfree(bp);
573 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
574 		brelse(bp);
575 	}
576 	return(1);
577 }
578 
579 /*
580  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
581  * blocks (with a negative loffset) are scanned.
582  * Note that the compare function must conform to the RB_SCAN's requirements.
583  */
584 static int
585 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused)
586 {
587 	if (bp->b_loffset < 0)
588 		return(0);
589 	return(1);
590 }
591 
592 static int
593 vtruncbuf_bp_metasync(struct buf *bp, void *data)
594 {
595 	struct vtruncbuf_info *info = data;
596 
597 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
598 		atomic_add_int(&bp->b_refs, 1);
599 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
600 			BUF_UNLOCK(bp);
601 		atomic_subtract_int(&bp->b_refs, 1);
602 	} else if ((bp->b_flags & B_DELWRI) == 0 ||
603 		   bp->b_vp != info->vp ||
604 		   vtruncbuf_bp_metasync_cmp(bp, data)) {
605 		BUF_UNLOCK(bp);
606 	} else {
607 		bremfree(bp);
608 		if (bp->b_vp == info->vp)
609 			bawrite(bp);
610 		else
611 			bwrite(bp);
612 	}
613 	return(1);
614 }
615 
616 /*
617  * vfsync - implements a multipass fsync on a file which understands
618  * dependancies and meta-data.  The passed vnode must be locked.  The
619  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
620  *
621  * When fsyncing data asynchronously just do one consolidated pass starting
622  * with the most negative block number.  This may not get all the data due
623  * to dependancies.
624  *
625  * When fsyncing data synchronously do a data pass, then a metadata pass,
626  * then do additional data+metadata passes to try to get all the data out.
627  */
628 static int vfsync_wait_output(struct vnode *vp,
629 			    int (*waitoutput)(struct vnode *, struct thread *));
630 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused);
631 static int vfsync_data_only_cmp(struct buf *bp, void *data);
632 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
633 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
634 static int vfsync_bp(struct buf *bp, void *data);
635 
636 struct vfsync_info {
637 	struct vnode *vp;
638 	int synchronous;
639 	int syncdeps;
640 	int lazycount;
641 	int lazylimit;
642 	int skippedbufs;
643 	int (*checkdef)(struct buf *);
644 	int (*cmpfunc)(struct buf *, void *);
645 };
646 
647 int
648 vfsync(struct vnode *vp, int waitfor, int passes,
649 	int (*checkdef)(struct buf *),
650 	int (*waitoutput)(struct vnode *, struct thread *))
651 {
652 	struct vfsync_info info;
653 	int error;
654 
655 	bzero(&info, sizeof(info));
656 	info.vp = vp;
657 	if ((info.checkdef = checkdef) == NULL)
658 		info.syncdeps = 1;
659 
660 	lwkt_gettoken(&vp->v_token);
661 
662 	switch(waitfor) {
663 	case MNT_LAZY | MNT_NOWAIT:
664 	case MNT_LAZY:
665 		/*
666 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
667 		 * number of data (not meta) pages we try to flush to 1MB.
668 		 * A non-zero return means that lazy limit was reached.
669 		 */
670 		info.lazylimit = 1024 * 1024;
671 		info.syncdeps = 1;
672 		info.cmpfunc = vfsync_lazy_range_cmp;
673 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
674 				vfsync_lazy_range_cmp, vfsync_bp, &info);
675 		info.cmpfunc = vfsync_meta_only_cmp;
676 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
677 			vfsync_meta_only_cmp, vfsync_bp, &info);
678 		if (error == 0)
679 			vp->v_lazyw = 0;
680 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
681 			vn_syncer_add(vp, 1);
682 		error = 0;
683 		break;
684 	case MNT_NOWAIT:
685 		/*
686 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
687 		 */
688 		info.syncdeps = 1;
689 		info.cmpfunc = vfsync_data_only_cmp;
690 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
691 			vfsync_bp, &info);
692 		info.cmpfunc = vfsync_meta_only_cmp;
693 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
694 			vfsync_bp, &info);
695 		error = 0;
696 		break;
697 	default:
698 		/*
699 		 * Synchronous.  Do a data-only pass, then a meta-data+data
700 		 * pass, then additional integrated passes to try to get
701 		 * all the dependancies flushed.
702 		 */
703 		info.cmpfunc = vfsync_data_only_cmp;
704 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
705 			vfsync_bp, &info);
706 		error = vfsync_wait_output(vp, waitoutput);
707 		if (error == 0) {
708 			info.skippedbufs = 0;
709 			info.cmpfunc = vfsync_dummy_cmp;
710 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
711 				vfsync_bp, &info);
712 			error = vfsync_wait_output(vp, waitoutput);
713 			if (info.skippedbufs) {
714 				kprintf("Warning: vfsync skipped %d dirty "
715 					"bufs in pass2!\n", info.skippedbufs);
716 			}
717 		}
718 		while (error == 0 && passes > 0 &&
719 		       !RB_EMPTY(&vp->v_rbdirty_tree)
720 		) {
721 			if (--passes == 0) {
722 				info.synchronous = 1;
723 				info.syncdeps = 1;
724 			}
725 			info.cmpfunc = vfsync_dummy_cmp;
726 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
727 					vfsync_bp, &info);
728 			if (error < 0)
729 				error = -error;
730 			info.syncdeps = 1;
731 			if (error == 0)
732 				error = vfsync_wait_output(vp, waitoutput);
733 		}
734 		break;
735 	}
736 	lwkt_reltoken(&vp->v_token);
737 	return(error);
738 }
739 
740 static int
741 vfsync_wait_output(struct vnode *vp,
742 		   int (*waitoutput)(struct vnode *, struct thread *))
743 {
744 	int error;
745 
746 	error = bio_track_wait(&vp->v_track_write, 0, 0);
747 	if (waitoutput)
748 		error = waitoutput(vp, curthread);
749 	return(error);
750 }
751 
752 static int
753 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
754 {
755 	return(0);
756 }
757 
758 static int
759 vfsync_data_only_cmp(struct buf *bp, void *data)
760 {
761 	if (bp->b_loffset < 0)
762 		return(-1);
763 	return(0);
764 }
765 
766 static int
767 vfsync_meta_only_cmp(struct buf *bp, void *data)
768 {
769 	if (bp->b_loffset < 0)
770 		return(0);
771 	return(1);
772 }
773 
774 static int
775 vfsync_lazy_range_cmp(struct buf *bp, void *data)
776 {
777 	struct vfsync_info *info = data;
778 
779 	if (bp->b_loffset < info->vp->v_lazyw)
780 		return(-1);
781 	return(0);
782 }
783 
784 static int
785 vfsync_bp(struct buf *bp, void *data)
786 {
787 	struct vfsync_info *info = data;
788 	struct vnode *vp = info->vp;
789 	int error;
790 
791 	/*
792 	 * Ignore buffers that we cannot immediately lock.
793 	 */
794 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
795 		++info->skippedbufs;
796 		return(0);
797 	}
798 
799 	/*
800 	 * We must revalidate the buffer after locking.
801 	 */
802 	if ((bp->b_flags & B_DELWRI) == 0 ||
803 	    bp->b_vp != info->vp ||
804 	    info->cmpfunc(bp, data)) {
805 		BUF_UNLOCK(bp);
806 		return(0);
807 	}
808 
809 	/*
810 	 * If syncdeps is not set we do not try to write buffers which have
811 	 * dependancies.
812 	 */
813 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
814 		BUF_UNLOCK(bp);
815 		return(0);
816 	}
817 
818 	/*
819 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
820 	 * has been written but an additional handshake with the device
821 	 * is required before we can dispose of the buffer.  We have no idea
822 	 * how to do this so we have to skip these buffers.
823 	 */
824 	if (bp->b_flags & B_NEEDCOMMIT) {
825 		BUF_UNLOCK(bp);
826 		return(0);
827 	}
828 
829 	/*
830 	 * Ask bioops if it is ok to sync.  If not the VFS may have
831 	 * set B_LOCKED so we have to cycle the buffer.
832 	 */
833 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
834 		bremfree(bp);
835 		brelse(bp);
836 		return(0);
837 	}
838 
839 	if (info->synchronous) {
840 		/*
841 		 * Synchronous flushing.  An error may be returned.
842 		 */
843 		bremfree(bp);
844 		error = bwrite(bp);
845 	} else {
846 		/*
847 		 * Asynchronous flushing.  A negative return value simply
848 		 * stops the scan and is not considered an error.  We use
849 		 * this to support limited MNT_LAZY flushes.
850 		 */
851 		vp->v_lazyw = bp->b_loffset;
852 		bremfree(bp);
853 		info->lazycount += cluster_awrite(bp);
854 		waitrunningbufspace();
855 		vm_wait_nominal();
856 		if (info->lazylimit && info->lazycount >= info->lazylimit)
857 			error = 1;
858 		else
859 			error = 0;
860 	}
861 	return(-error);
862 }
863 
864 /*
865  * Associate a buffer with a vnode.
866  *
867  * MPSAFE
868  */
869 int
870 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
871 {
872 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
873 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
874 
875 	/*
876 	 * Insert onto list for new vnode.
877 	 */
878 	lwkt_gettoken(&vp->v_token);
879 
880 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
881 		lwkt_reltoken(&vp->v_token);
882 		return (EEXIST);
883 	}
884 
885 	/*
886 	 * Diagnostics (mainly for HAMMER debugging).  Check for
887 	 * overlapping buffers.
888 	 */
889 	if (check_buf_overlap) {
890 		struct buf *bx;
891 		bx = buf_rb_hash_RB_PREV(bp);
892 		if (bx) {
893 			if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
894 				kprintf("bgetvp: overlapl %016jx/%d %016jx "
895 					"bx %p bp %p\n",
896 					(intmax_t)bx->b_loffset,
897 					bx->b_bufsize,
898 					(intmax_t)bp->b_loffset,
899 					bx, bp);
900 				if (check_buf_overlap > 1)
901 					panic("bgetvp - overlapping buffer");
902 			}
903 		}
904 		bx = buf_rb_hash_RB_NEXT(bp);
905 		if (bx) {
906 			if (bp->b_loffset + testsize > bx->b_loffset) {
907 				kprintf("bgetvp: overlapr %016jx/%d %016jx "
908 					"bp %p bx %p\n",
909 					(intmax_t)bp->b_loffset,
910 					testsize,
911 					(intmax_t)bx->b_loffset,
912 					bp, bx);
913 				if (check_buf_overlap > 1)
914 					panic("bgetvp - overlapping buffer");
915 			}
916 		}
917 	}
918 	bp->b_vp = vp;
919 	bp->b_flags |= B_HASHED;
920 	bp->b_flags |= B_VNCLEAN;
921 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
922 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
923 	/*vhold(vp);*/
924 	lwkt_reltoken(&vp->v_token);
925 	return(0);
926 }
927 
928 /*
929  * Disassociate a buffer from a vnode.
930  *
931  * MPSAFE
932  */
933 void
934 brelvp(struct buf *bp)
935 {
936 	struct vnode *vp;
937 
938 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
939 
940 	/*
941 	 * Delete from old vnode list, if on one.
942 	 */
943 	vp = bp->b_vp;
944 	lwkt_gettoken(&vp->v_token);
945 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
946 		if (bp->b_flags & B_VNDIRTY)
947 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
948 		else
949 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
950 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
951 	}
952 	if (bp->b_flags & B_HASHED) {
953 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
954 		bp->b_flags &= ~B_HASHED;
955 	}
956 
957 	/*
958 	 * Only remove from synclist when no dirty buffers are left AND
959 	 * the VFS has not flagged the vnode's inode as being dirty.
960 	 */
961 	if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST &&
962 	    RB_EMPTY(&vp->v_rbdirty_tree)) {
963 		vn_syncer_remove(vp);
964 	}
965 	bp->b_vp = NULL;
966 
967 	lwkt_reltoken(&vp->v_token);
968 
969 	/*vdrop(vp);*/
970 }
971 
972 /*
973  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
974  * This routine is called when the state of the B_DELWRI bit is changed.
975  *
976  * Must be called with vp->v_token held.
977  * MPSAFE
978  */
979 void
980 reassignbuf(struct buf *bp)
981 {
982 	struct vnode *vp = bp->b_vp;
983 	int delay;
984 
985 	ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
986 	++reassignbufcalls;
987 
988 	/*
989 	 * B_PAGING flagged buffers cannot be reassigned because their vp
990 	 * is not fully linked in.
991 	 */
992 	if (bp->b_flags & B_PAGING)
993 		panic("cannot reassign paging buffer");
994 
995 	if (bp->b_flags & B_DELWRI) {
996 		/*
997 		 * Move to the dirty list, add the vnode to the worklist
998 		 */
999 		if (bp->b_flags & B_VNCLEAN) {
1000 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
1001 			bp->b_flags &= ~B_VNCLEAN;
1002 		}
1003 		if ((bp->b_flags & B_VNDIRTY) == 0) {
1004 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
1005 				panic("reassignbuf: dup lblk vp %p bp %p",
1006 				      vp, bp);
1007 			}
1008 			bp->b_flags |= B_VNDIRTY;
1009 		}
1010 		if ((vp->v_flag & VONWORKLST) == 0) {
1011 			switch (vp->v_type) {
1012 			case VDIR:
1013 				delay = dirdelay;
1014 				break;
1015 			case VCHR:
1016 			case VBLK:
1017 				if (vp->v_rdev &&
1018 				    vp->v_rdev->si_mountpoint != NULL) {
1019 					delay = metadelay;
1020 					break;
1021 				}
1022 				/* fall through */
1023 			default:
1024 				delay = filedelay;
1025 			}
1026 			vn_syncer_add(vp, delay);
1027 		}
1028 	} else {
1029 		/*
1030 		 * Move to the clean list, remove the vnode from the worklist
1031 		 * if no dirty blocks remain.
1032 		 */
1033 		if (bp->b_flags & B_VNDIRTY) {
1034 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
1035 			bp->b_flags &= ~B_VNDIRTY;
1036 		}
1037 		if ((bp->b_flags & B_VNCLEAN) == 0) {
1038 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
1039 				panic("reassignbuf: dup lblk vp %p bp %p",
1040 				      vp, bp);
1041 			}
1042 			bp->b_flags |= B_VNCLEAN;
1043 		}
1044 
1045 		/*
1046 		 * Only remove from synclist when no dirty buffers are left
1047 		 * AND the VFS has not flagged the vnode's inode as being
1048 		 * dirty.
1049 		 */
1050 		if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) ==
1051 		     VONWORKLST &&
1052 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
1053 			vn_syncer_remove(vp);
1054 		}
1055 	}
1056 }
1057 
1058 /*
1059  * Create a vnode for a block device.  Used for mounting the root file
1060  * system.
1061  *
1062  * A vref()'d vnode is returned.
1063  */
1064 extern struct vop_ops *devfs_vnode_dev_vops_p;
1065 int
1066 bdevvp(cdev_t dev, struct vnode **vpp)
1067 {
1068 	struct vnode *vp;
1069 	struct vnode *nvp;
1070 	int error;
1071 
1072 	if (dev == NULL) {
1073 		*vpp = NULLVP;
1074 		return (ENXIO);
1075 	}
1076 	error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
1077 				&nvp, 0, 0);
1078 	if (error) {
1079 		*vpp = NULLVP;
1080 		return (error);
1081 	}
1082 	vp = nvp;
1083 	vp->v_type = VCHR;
1084 #if 0
1085 	vp->v_rdev = dev;
1086 #endif
1087 	v_associate_rdev(vp, dev);
1088 	vp->v_umajor = dev->si_umajor;
1089 	vp->v_uminor = dev->si_uminor;
1090 	vx_unlock(vp);
1091 	*vpp = vp;
1092 	return (0);
1093 }
1094 
1095 int
1096 v_associate_rdev(struct vnode *vp, cdev_t dev)
1097 {
1098 	if (dev == NULL)
1099 		return(ENXIO);
1100 	if (dev_is_good(dev) == 0)
1101 		return(ENXIO);
1102 	KKASSERT(vp->v_rdev == NULL);
1103 	vp->v_rdev = reference_dev(dev);
1104 	lwkt_gettoken(&spechash_token);
1105 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1106 	lwkt_reltoken(&spechash_token);
1107 	return(0);
1108 }
1109 
1110 void
1111 v_release_rdev(struct vnode *vp)
1112 {
1113 	cdev_t dev;
1114 
1115 	if ((dev = vp->v_rdev) != NULL) {
1116 		lwkt_gettoken(&spechash_token);
1117 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1118 		vp->v_rdev = NULL;
1119 		release_dev(dev);
1120 		lwkt_reltoken(&spechash_token);
1121 	}
1122 }
1123 
1124 /*
1125  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1126  * the device number with the vnode.  The actual device is not associated
1127  * until the vnode is opened (usually in spec_open()), and will be
1128  * disassociated on last close.
1129  */
1130 void
1131 addaliasu(struct vnode *nvp, int x, int y)
1132 {
1133 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1134 		panic("addaliasu on non-special vnode");
1135 	nvp->v_umajor = x;
1136 	nvp->v_uminor = y;
1137 }
1138 
1139 /*
1140  * Simple call that a filesystem can make to try to get rid of a
1141  * vnode.  It will fail if anyone is referencing the vnode (including
1142  * the caller).
1143  *
1144  * The filesystem can check whether its in-memory inode structure still
1145  * references the vp on return.
1146  *
1147  * May only be called if the vnode is in a known state (i.e. being prevented
1148  * from being deallocated by some other condition such as a vfs inode hold).
1149  */
1150 void
1151 vclean_unlocked(struct vnode *vp)
1152 {
1153 	vx_get(vp);
1154 	if (VREFCNT(vp) <= 0)
1155 		vgone_vxlocked(vp);
1156 	vx_put(vp);
1157 }
1158 
1159 /*
1160  * Disassociate a vnode from its underlying filesystem.
1161  *
1162  * The vnode must be VX locked and referenced.  In all normal situations
1163  * there are no active references.  If vclean_vxlocked() is called while
1164  * there are active references, the vnode is being ripped out and we have
1165  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1166  */
1167 void
1168 vclean_vxlocked(struct vnode *vp, int flags)
1169 {
1170 	int active;
1171 	int n;
1172 	vm_object_t object;
1173 	struct namecache *ncp;
1174 
1175 	/*
1176 	 * If the vnode has already been reclaimed we have nothing to do.
1177 	 */
1178 	if (vp->v_flag & VRECLAIMED)
1179 		return;
1180 
1181 	/*
1182 	 * Set flag to interlock operation, flag finalization to ensure
1183 	 * that the vnode winds up on the inactive list, and set v_act to 0.
1184 	 */
1185 	vsetflags(vp, VRECLAIMED);
1186 	atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1187 	vp->v_act = 0;
1188 
1189 	if (verbose_reclaims) {
1190 		if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
1191 			kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
1192 	}
1193 
1194 	/*
1195 	 * Scrap the vfs cache
1196 	 */
1197 	while (cache_inval_vp(vp, 0) != 0) {
1198 		kprintf("Warning: vnode %p clean/cache_resolution "
1199 			"race detected\n", vp);
1200 		tsleep(vp, 0, "vclninv", 2);
1201 	}
1202 
1203 	/*
1204 	 * Check to see if the vnode is in use. If so we have to reference it
1205 	 * before we clean it out so that its count cannot fall to zero and
1206 	 * generate a race against ourselves to recycle it.
1207 	 */
1208 	active = (VREFCNT(vp) > 0);
1209 
1210 	/*
1211 	 * Clean out any buffers associated with the vnode and destroy its
1212 	 * object, if it has one.
1213 	 */
1214 	vinvalbuf(vp, V_SAVE, 0, 0);
1215 	KKASSERT(lockcountnb(&vp->v_lock) == 1);
1216 
1217 	/*
1218 	 * If purging an active vnode (typically during a forced unmount
1219 	 * or reboot), it must be closed and deactivated before being
1220 	 * reclaimed.  This isn't really all that safe, but what can
1221 	 * we do? XXX.
1222 	 *
1223 	 * Note that neither of these routines unlocks the vnode.
1224 	 */
1225 	if (active && (flags & DOCLOSE)) {
1226 		while ((n = vp->v_opencount) != 0) {
1227 			if (vp->v_writecount)
1228 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1229 			else
1230 				VOP_CLOSE(vp, FNONBLOCK);
1231 			if (vp->v_opencount == n) {
1232 				kprintf("Warning: unable to force-close"
1233 				       " vnode %p\n", vp);
1234 				break;
1235 			}
1236 		}
1237 	}
1238 
1239 	/*
1240 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1241 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1242 	 * again to make sure they all get flushed.
1243 	 *
1244 	 * This can occur if a file with a link count of 0 needs to be
1245 	 * truncated.
1246 	 *
1247 	 * If the vnode is already dead don't try to deactivate it.
1248 	 */
1249 	if ((vp->v_flag & VINACTIVE) == 0) {
1250 		vsetflags(vp, VINACTIVE);
1251 		if (vp->v_mount)
1252 			VOP_INACTIVE(vp);
1253 		vinvalbuf(vp, V_SAVE, 0, 0);
1254 	}
1255 	KKASSERT(lockcountnb(&vp->v_lock) == 1);
1256 
1257 	/*
1258 	 * If the vnode has an object, destroy it.
1259 	 */
1260 	while ((object = vp->v_object) != NULL) {
1261 		vm_object_hold(object);
1262 		if (object == vp->v_object)
1263 			break;
1264 		vm_object_drop(object);
1265 	}
1266 
1267 	if (object != NULL) {
1268 		if (object->ref_count == 0) {
1269 			if ((object->flags & OBJ_DEAD) == 0)
1270 				vm_object_terminate(object);
1271 			vm_object_drop(object);
1272 			vclrflags(vp, VOBJBUF);
1273 		} else {
1274 			vm_pager_deallocate(object);
1275 			vclrflags(vp, VOBJBUF);
1276 			vm_object_drop(object);
1277 		}
1278 	}
1279 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1280 
1281 	/*
1282 	 * Reclaim the vnode if not already dead.
1283 	 */
1284 	if (vp->v_mount && VOP_RECLAIM(vp))
1285 		panic("vclean: cannot reclaim");
1286 
1287 	/*
1288 	 * Done with purge, notify sleepers of the grim news.
1289 	 */
1290 	vp->v_ops = &dead_vnode_vops_p;
1291 	vn_gone(vp);
1292 	vp->v_tag = VT_NON;
1293 
1294 	/*
1295 	 * If we are destroying an active vnode, reactivate it now that
1296 	 * we have reassociated it with deadfs.  This prevents the system
1297 	 * from crashing on the vnode due to it being unexpectedly marked
1298 	 * as inactive or reclaimed.
1299 	 */
1300 	if (active && (flags & DOCLOSE)) {
1301 		vclrflags(vp, VINACTIVE | VRECLAIMED);
1302 	}
1303 }
1304 
1305 /*
1306  * Eliminate all activity associated with the requested vnode
1307  * and with all vnodes aliased to the requested vnode.
1308  *
1309  * The vnode must be referenced but should not be locked.
1310  */
1311 int
1312 vrevoke(struct vnode *vp, struct ucred *cred)
1313 {
1314 	struct vnode *vq;
1315 	struct vnode *vqn;
1316 	cdev_t dev;
1317 	int error;
1318 
1319 	/*
1320 	 * If the vnode has a device association, scrap all vnodes associated
1321 	 * with the device.  Don't let the device disappear on us while we
1322 	 * are scrapping the vnodes.
1323 	 *
1324 	 * The passed vp will probably show up in the list, do not VX lock
1325 	 * it twice!
1326 	 *
1327 	 * Releasing the vnode's rdev here can mess up specfs's call to
1328 	 * device close, so don't do it.  The vnode has been disassociated
1329 	 * and the device will be closed after the last ref on the related
1330 	 * fp goes away (if not still open by e.g. the kernel).
1331 	 */
1332 	if (vp->v_type != VCHR) {
1333 		error = fdrevoke(vp, DTYPE_VNODE, cred);
1334 		return (error);
1335 	}
1336 	if ((dev = vp->v_rdev) == NULL) {
1337 		return(0);
1338 	}
1339 	reference_dev(dev);
1340 	lwkt_gettoken(&spechash_token);
1341 
1342 restart:
1343 	vqn = SLIST_FIRST(&dev->si_hlist);
1344 	if (vqn)
1345 		vhold(vqn);
1346 	while ((vq = vqn) != NULL) {
1347 		if (VREFCNT(vq) > 0) {
1348 			vref(vq);
1349 			fdrevoke(vq, DTYPE_VNODE, cred);
1350 			/*v_release_rdev(vq);*/
1351 			vrele(vq);
1352 			if (vq->v_rdev != dev) {
1353 				vdrop(vq);
1354 				goto restart;
1355 			}
1356 		}
1357 		vqn = SLIST_NEXT(vq, v_cdevnext);
1358 		if (vqn)
1359 			vhold(vqn);
1360 		vdrop(vq);
1361 	}
1362 	lwkt_reltoken(&spechash_token);
1363 	dev_drevoke(dev);
1364 	release_dev(dev);
1365 	return (0);
1366 }
1367 
1368 /*
1369  * This is called when the object underlying a vnode is being destroyed,
1370  * such as in a remove().  Try to recycle the vnode immediately if the
1371  * only active reference is our reference.
1372  *
1373  * Directory vnodes in the namecache with children cannot be immediately
1374  * recycled because numerous VOP_N*() ops require them to be stable.
1375  *
1376  * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1377  * function is a NOP if VRECLAIMED is already set.
1378  */
1379 int
1380 vrecycle(struct vnode *vp)
1381 {
1382 	if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1383 		if (cache_inval_vp_nonblock(vp))
1384 			return(0);
1385 		vgone_vxlocked(vp);
1386 		return (1);
1387 	}
1388 	return (0);
1389 }
1390 
1391 /*
1392  * Return the maximum I/O size allowed for strategy calls on VP.
1393  *
1394  * If vp is VCHR or VBLK we dive the device, otherwise we use
1395  * the vp's mount info.
1396  *
1397  * The returned value is clamped at MAXPHYS as most callers cannot use
1398  * buffers larger than that size.
1399  */
1400 int
1401 vmaxiosize(struct vnode *vp)
1402 {
1403 	int maxiosize;
1404 
1405 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1406 		maxiosize = vp->v_rdev->si_iosize_max;
1407 	else
1408 		maxiosize = vp->v_mount->mnt_iosize_max;
1409 
1410 	if (maxiosize > MAXPHYS)
1411 		maxiosize = MAXPHYS;
1412 	return (maxiosize);
1413 }
1414 
1415 /*
1416  * Eliminate all activity associated with a vnode in preparation for
1417  * destruction.
1418  *
1419  * The vnode must be VX locked and refd and will remain VX locked and refd
1420  * on return.  This routine may be called with the vnode in any state, as
1421  * long as it is VX locked.  The vnode will be cleaned out and marked
1422  * VRECLAIMED but will not actually be reused until all existing refs and
1423  * holds go away.
1424  *
1425  * NOTE: This routine may be called on a vnode which has not yet been
1426  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1427  * already been reclaimed.
1428  *
1429  * This routine is not responsible for placing us back on the freelist.
1430  * Instead, it happens automatically when the caller releases the VX lock
1431  * (assuming there aren't any other references).
1432  */
1433 void
1434 vgone_vxlocked(struct vnode *vp)
1435 {
1436 	/*
1437 	 * assert that the VX lock is held.  This is an absolute requirement
1438 	 * now for vgone_vxlocked() to be called.
1439 	 */
1440 	KKASSERT(lockcountnb(&vp->v_lock) == 1);
1441 
1442 	/*
1443 	 * Clean out the filesystem specific data and set the VRECLAIMED
1444 	 * bit.  Also deactivate the vnode if necessary.
1445 	 */
1446 	vclean_vxlocked(vp, DOCLOSE);
1447 
1448 	/*
1449 	 * Delete from old mount point vnode list, if on one.
1450 	 */
1451 	if (vp->v_mount != NULL) {
1452 		KKASSERT(vp->v_data == NULL);
1453 		insmntque(vp, NULL);
1454 	}
1455 
1456 	/*
1457 	 * If special device, remove it from special device alias list
1458 	 * if it is on one.  This should normally only occur if a vnode is
1459 	 * being revoked as the device should otherwise have been released
1460 	 * naturally.
1461 	 */
1462 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1463 		v_release_rdev(vp);
1464 	}
1465 
1466 	/*
1467 	 * Set us to VBAD
1468 	 */
1469 	vp->v_type = VBAD;
1470 }
1471 
1472 /*
1473  * Lookup a vnode by device number.
1474  *
1475  * Returns non-zero and *vpp set to a vref'd vnode on success.
1476  * Returns zero on failure.
1477  */
1478 int
1479 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1480 {
1481 	struct vnode *vp;
1482 
1483 	lwkt_gettoken(&spechash_token);
1484 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1485 		if (type == vp->v_type) {
1486 			*vpp = vp;
1487 			vref(vp);
1488 			lwkt_reltoken(&spechash_token);
1489 			return (1);
1490 		}
1491 	}
1492 	lwkt_reltoken(&spechash_token);
1493 	return (0);
1494 }
1495 
1496 /*
1497  * Calculate the total number of references to a special device.  This
1498  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1499  * an overloaded field.  Since udev2dev can now return NULL, we have
1500  * to check for a NULL v_rdev.
1501  */
1502 int
1503 count_dev(cdev_t dev)
1504 {
1505 	struct vnode *vp;
1506 	int count = 0;
1507 
1508 	if (SLIST_FIRST(&dev->si_hlist)) {
1509 		lwkt_gettoken(&spechash_token);
1510 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1511 			count += vp->v_opencount;
1512 		}
1513 		lwkt_reltoken(&spechash_token);
1514 	}
1515 	return(count);
1516 }
1517 
1518 int
1519 vcount(struct vnode *vp)
1520 {
1521 	if (vp->v_rdev == NULL)
1522 		return(0);
1523 	return(count_dev(vp->v_rdev));
1524 }
1525 
1526 /*
1527  * Initialize VMIO for a vnode.  This routine MUST be called before a
1528  * VFS can issue buffer cache ops on a vnode.  It is typically called
1529  * when a vnode is initialized from its inode.
1530  */
1531 int
1532 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1533 {
1534 	vm_object_t object;
1535 	int error = 0;
1536 
1537 retry:
1538 	while ((object = vp->v_object) != NULL) {
1539 		vm_object_hold(object);
1540 		if (object == vp->v_object)
1541 			break;
1542 		vm_object_drop(object);
1543 	}
1544 
1545 	if (object == NULL) {
1546 		object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1547 
1548 		/*
1549 		 * Dereference the reference we just created.  This assumes
1550 		 * that the object is associated with the vp.
1551 		 */
1552 		vm_object_hold(object);
1553 		atomic_add_int(&object->ref_count, -1);
1554 		vrele(vp);
1555 	} else {
1556 		if (object->flags & OBJ_DEAD) {
1557 			vn_unlock(vp);
1558 			if (vp->v_object == object)
1559 				vm_object_dead_sleep(object, "vodead");
1560 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1561 			vm_object_drop(object);
1562 			goto retry;
1563 		}
1564 	}
1565 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1566 	vsetflags(vp, VOBJBUF);
1567 	vm_object_drop(object);
1568 
1569 	return (error);
1570 }
1571 
1572 
1573 /*
1574  * Print out a description of a vnode.
1575  */
1576 static char *typename[] =
1577 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1578 
1579 void
1580 vprint(char *label, struct vnode *vp)
1581 {
1582 	char buf[96];
1583 
1584 	if (label != NULL)
1585 		kprintf("%s: %p: ", label, (void *)vp);
1586 	else
1587 		kprintf("%p: ", (void *)vp);
1588 	kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
1589 		typename[vp->v_type],
1590 		vp->v_refcnt, vp->v_writecount, vp->v_auxrefs);
1591 	buf[0] = '\0';
1592 	if (vp->v_flag & VROOT)
1593 		strcat(buf, "|VROOT");
1594 	if (vp->v_flag & VPFSROOT)
1595 		strcat(buf, "|VPFSROOT");
1596 	if (vp->v_flag & VTEXT)
1597 		strcat(buf, "|VTEXT");
1598 	if (vp->v_flag & VSYSTEM)
1599 		strcat(buf, "|VSYSTEM");
1600 	if (vp->v_flag & VOBJBUF)
1601 		strcat(buf, "|VOBJBUF");
1602 	if (buf[0] != '\0')
1603 		kprintf(" flags (%s)", &buf[1]);
1604 	if (vp->v_data == NULL) {
1605 		kprintf("\n");
1606 	} else {
1607 		kprintf("\n\t");
1608 		VOP_PRINT(vp);
1609 	}
1610 }
1611 
1612 /*
1613  * Do the usual access checking.
1614  * file_mode, uid and gid are from the vnode in question,
1615  * while acc_mode and cred are from the VOP_ACCESS parameter list
1616  */
1617 int
1618 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1619     mode_t acc_mode, struct ucred *cred)
1620 {
1621 	mode_t mask;
1622 	int ismember;
1623 
1624 	/*
1625 	 * Super-user always gets read/write access, but execute access depends
1626 	 * on at least one execute bit being set.
1627 	 */
1628 	if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1629 		if ((acc_mode & VEXEC) && type != VDIR &&
1630 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1631 			return (EACCES);
1632 		return (0);
1633 	}
1634 
1635 	mask = 0;
1636 
1637 	/* Otherwise, check the owner. */
1638 	if (cred->cr_uid == uid) {
1639 		if (acc_mode & VEXEC)
1640 			mask |= S_IXUSR;
1641 		if (acc_mode & VREAD)
1642 			mask |= S_IRUSR;
1643 		if (acc_mode & VWRITE)
1644 			mask |= S_IWUSR;
1645 		return ((file_mode & mask) == mask ? 0 : EACCES);
1646 	}
1647 
1648 	/* Otherwise, check the groups. */
1649 	ismember = groupmember(gid, cred);
1650 	if (cred->cr_svgid == gid || ismember) {
1651 		if (acc_mode & VEXEC)
1652 			mask |= S_IXGRP;
1653 		if (acc_mode & VREAD)
1654 			mask |= S_IRGRP;
1655 		if (acc_mode & VWRITE)
1656 			mask |= S_IWGRP;
1657 		return ((file_mode & mask) == mask ? 0 : EACCES);
1658 	}
1659 
1660 	/* Otherwise, check everyone else. */
1661 	if (acc_mode & VEXEC)
1662 		mask |= S_IXOTH;
1663 	if (acc_mode & VREAD)
1664 		mask |= S_IROTH;
1665 	if (acc_mode & VWRITE)
1666 		mask |= S_IWOTH;
1667 	return ((file_mode & mask) == mask ? 0 : EACCES);
1668 }
1669 
1670 #ifdef DDB
1671 #include <ddb/ddb.h>
1672 
1673 static int db_show_locked_vnodes(struct mount *mp, void *data);
1674 
1675 /*
1676  * List all of the locked vnodes in the system.
1677  * Called when debugging the kernel.
1678  */
1679 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1680 {
1681 	kprintf("Locked vnodes\n");
1682 	mountlist_scan(db_show_locked_vnodes, NULL,
1683 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1684 }
1685 
1686 static int
1687 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1688 {
1689 	struct vnode *vp;
1690 
1691 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1692 		if (vn_islocked(vp))
1693 			vprint(NULL, vp);
1694 	}
1695 	return(0);
1696 }
1697 #endif
1698 
1699 /*
1700  * Top level filesystem related information gathering.
1701  */
1702 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1703 
1704 static int
1705 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1706 {
1707 	int *name = (int *)arg1 - 1;	/* XXX */
1708 	u_int namelen = arg2 + 1;	/* XXX */
1709 	struct vfsconf *vfsp;
1710 	int maxtypenum;
1711 
1712 #if 1 || defined(COMPAT_PRELITE2)
1713 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1714 	if (namelen == 1)
1715 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1716 #endif
1717 
1718 #ifdef notyet
1719 	/* all sysctl names at this level are at least name and field */
1720 	if (namelen < 2)
1721 		return (ENOTDIR);		/* overloaded */
1722 	if (name[0] != VFS_GENERIC) {
1723 		vfsp = vfsconf_find_by_typenum(name[0]);
1724 		if (vfsp == NULL)
1725 			return (EOPNOTSUPP);
1726 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1727 		    oldp, oldlenp, newp, newlen, p));
1728 	}
1729 #endif
1730 	switch (name[1]) {
1731 	case VFS_MAXTYPENUM:
1732 		if (namelen != 2)
1733 			return (ENOTDIR);
1734 		maxtypenum = vfsconf_get_maxtypenum();
1735 		return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1736 	case VFS_CONF:
1737 		if (namelen != 3)
1738 			return (ENOTDIR);	/* overloaded */
1739 		vfsp = vfsconf_find_by_typenum(name[2]);
1740 		if (vfsp == NULL)
1741 			return (EOPNOTSUPP);
1742 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1743 	}
1744 	return (EOPNOTSUPP);
1745 }
1746 
1747 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1748 	"Generic filesystem");
1749 
1750 #if 1 || defined(COMPAT_PRELITE2)
1751 
1752 static int
1753 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1754 {
1755 	int error;
1756 	struct ovfsconf ovfs;
1757 	struct sysctl_req *req = (struct sysctl_req*) data;
1758 
1759 	bzero(&ovfs, sizeof(ovfs));
1760 	ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1761 	strcpy(ovfs.vfc_name, vfsp->vfc_name);
1762 	ovfs.vfc_index = vfsp->vfc_typenum;
1763 	ovfs.vfc_refcount = vfsp->vfc_refcount;
1764 	ovfs.vfc_flags = vfsp->vfc_flags;
1765 	error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1766 	if (error)
1767 		return error; /* abort iteration with error code */
1768 	else
1769 		return 0; /* continue iterating with next element */
1770 }
1771 
1772 static int
1773 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1774 {
1775 	return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1776 }
1777 
1778 #endif /* 1 || COMPAT_PRELITE2 */
1779 
1780 /*
1781  * Check to see if a filesystem is mounted on a block device.
1782  */
1783 int
1784 vfs_mountedon(struct vnode *vp)
1785 {
1786 	cdev_t dev;
1787 
1788 	if ((dev = vp->v_rdev) == NULL) {
1789 /*		if (vp->v_type != VBLK)
1790 			dev = get_dev(vp->v_uminor, vp->v_umajor); */
1791 	}
1792 	if (dev != NULL && dev->si_mountpoint)
1793 		return (EBUSY);
1794 	return (0);
1795 }
1796 
1797 /*
1798  * Unmount all filesystems. The list is traversed in reverse order
1799  * of mounting to avoid dependencies.
1800  */
1801 
1802 static int vfs_umountall_callback(struct mount *mp, void *data);
1803 
1804 void
1805 vfs_unmountall(void)
1806 {
1807 	int count;
1808 
1809 	do {
1810 		count = mountlist_scan(vfs_umountall_callback,
1811 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1812 	} while (count);
1813 }
1814 
1815 static
1816 int
1817 vfs_umountall_callback(struct mount *mp, void *data)
1818 {
1819 	int error;
1820 
1821 	error = dounmount(mp, MNT_FORCE);
1822 	if (error) {
1823 		mountlist_remove(mp);
1824 		kprintf("unmount of filesystem mounted from %s failed (",
1825 			mp->mnt_stat.f_mntfromname);
1826 		if (error == EBUSY)
1827 			kprintf("BUSY)\n");
1828 		else
1829 			kprintf("%d)\n", error);
1830 	}
1831 	return(1);
1832 }
1833 
1834 /*
1835  * Checks the mount flags for parameter mp and put the names comma-separated
1836  * into a string buffer buf with a size limit specified by len.
1837  *
1838  * It returns the number of bytes written into buf, and (*errorp) will be
1839  * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1840  * not large enough).  The buffer will be 0-terminated if len was not 0.
1841  */
1842 size_t
1843 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1844 	       char *buf, size_t len, int *errorp)
1845 {
1846 	static const struct mountctl_opt optnames[] = {
1847 		{ MNT_ASYNC,            "asynchronous" },
1848 		{ MNT_EXPORTED,         "NFS exported" },
1849 		{ MNT_LOCAL,            "local" },
1850 		{ MNT_NOATIME,          "noatime" },
1851 		{ MNT_NODEV,            "nodev" },
1852 		{ MNT_NOEXEC,           "noexec" },
1853 		{ MNT_NOSUID,           "nosuid" },
1854 		{ MNT_NOSYMFOLLOW,      "nosymfollow" },
1855 		{ MNT_QUOTA,            "with-quotas" },
1856 		{ MNT_RDONLY,           "read-only" },
1857 		{ MNT_SYNCHRONOUS,      "synchronous" },
1858 		{ MNT_UNION,            "union" },
1859 		{ MNT_NOCLUSTERR,       "noclusterr" },
1860 		{ MNT_NOCLUSTERW,       "noclusterw" },
1861 		{ MNT_SUIDDIR,          "suiddir" },
1862 		{ MNT_SOFTDEP,          "soft-updates" },
1863 		{ MNT_IGNORE,           "ignore" },
1864 		{ 0,			NULL}
1865 	};
1866 	int bwritten;
1867 	int bleft;
1868 	int optlen;
1869 	int actsize;
1870 
1871 	*errorp = 0;
1872 	bwritten = 0;
1873 	bleft = len - 1;	/* leave room for trailing \0 */
1874 
1875 	/*
1876 	 * Checks the size of the string. If it contains
1877 	 * any data, then we will append the new flags to
1878 	 * it.
1879 	 */
1880 	actsize = strlen(buf);
1881 	if (actsize > 0)
1882 		buf += actsize;
1883 
1884 	/* Default flags if no flags passed */
1885 	if (optp == NULL)
1886 		optp = optnames;
1887 
1888 	if (bleft < 0) {	/* degenerate case, 0-length buffer */
1889 		*errorp = EINVAL;
1890 		return(0);
1891 	}
1892 
1893 	for (; flags && optp->o_opt; ++optp) {
1894 		if ((flags & optp->o_opt) == 0)
1895 			continue;
1896 		optlen = strlen(optp->o_name);
1897 		if (bwritten || actsize > 0) {
1898 			if (bleft < 2) {
1899 				*errorp = ENOSPC;
1900 				break;
1901 			}
1902 			buf[bwritten++] = ',';
1903 			buf[bwritten++] = ' ';
1904 			bleft -= 2;
1905 		}
1906 		if (bleft < optlen) {
1907 			*errorp = ENOSPC;
1908 			break;
1909 		}
1910 		bcopy(optp->o_name, buf + bwritten, optlen);
1911 		bwritten += optlen;
1912 		bleft -= optlen;
1913 		flags &= ~optp->o_opt;
1914 	}
1915 
1916 	/*
1917 	 * Space already reserved for trailing \0
1918 	 */
1919 	buf[bwritten] = 0;
1920 	return (bwritten);
1921 }
1922 
1923 /*
1924  * Build hash lists of net addresses and hang them off the mount point.
1925  * Called by ufs_mount() to set up the lists of export addresses.
1926  */
1927 static int
1928 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1929 		const struct export_args *argp)
1930 {
1931 	struct netcred *np;
1932 	struct radix_node_head *rnh;
1933 	int i;
1934 	struct radix_node *rn;
1935 	struct sockaddr *saddr, *smask = NULL;
1936 	struct domain *dom;
1937 	int error;
1938 
1939 	if (argp->ex_addrlen == 0) {
1940 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1941 			return (EPERM);
1942 		np = &nep->ne_defexported;
1943 		np->netc_exflags = argp->ex_flags;
1944 		np->netc_anon = argp->ex_anon;
1945 		np->netc_anon.cr_ref = 1;
1946 		mp->mnt_flag |= MNT_DEFEXPORTED;
1947 		return (0);
1948 	}
1949 
1950 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1951 		return (EINVAL);
1952 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1953 		return (EINVAL);
1954 
1955 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1956 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1957 	saddr = (struct sockaddr *) (np + 1);
1958 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1959 		goto out;
1960 	if (saddr->sa_len > argp->ex_addrlen)
1961 		saddr->sa_len = argp->ex_addrlen;
1962 	if (argp->ex_masklen) {
1963 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1964 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1965 		if (error)
1966 			goto out;
1967 		if (smask->sa_len > argp->ex_masklen)
1968 			smask->sa_len = argp->ex_masklen;
1969 	}
1970 	i = saddr->sa_family;
1971 	if ((rnh = nep->ne_rtable[i]) == NULL) {
1972 		/*
1973 		 * Seems silly to initialize every AF when most are not used,
1974 		 * do so on demand here
1975 		 */
1976 		SLIST_FOREACH(dom, &domains, dom_next)
1977 			if (dom->dom_family == i && dom->dom_rtattach) {
1978 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1979 				    dom->dom_rtoffset);
1980 				break;
1981 			}
1982 		if ((rnh = nep->ne_rtable[i]) == NULL) {
1983 			error = ENOBUFS;
1984 			goto out;
1985 		}
1986 	}
1987 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1988 	    np->netc_rnodes);
1989 	if (rn == NULL || np != (struct netcred *) rn) {	/* already exists */
1990 		error = EPERM;
1991 		goto out;
1992 	}
1993 	np->netc_exflags = argp->ex_flags;
1994 	np->netc_anon = argp->ex_anon;
1995 	np->netc_anon.cr_ref = 1;
1996 	return (0);
1997 out:
1998 	kfree(np, M_NETADDR);
1999 	return (error);
2000 }
2001 
2002 /* ARGSUSED */
2003 static int
2004 vfs_free_netcred(struct radix_node *rn, void *w)
2005 {
2006 	struct radix_node_head *rnh = (struct radix_node_head *) w;
2007 
2008 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2009 	kfree((caddr_t) rn, M_NETADDR);
2010 	return (0);
2011 }
2012 
2013 /*
2014  * Free the net address hash lists that are hanging off the mount points.
2015  */
2016 static void
2017 vfs_free_addrlist(struct netexport *nep)
2018 {
2019 	int i;
2020 	struct radix_node_head *rnh;
2021 
2022 	for (i = 0; i <= AF_MAX; i++)
2023 		if ((rnh = nep->ne_rtable[i])) {
2024 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2025 			    (caddr_t) rnh);
2026 			kfree((caddr_t) rnh, M_RTABLE);
2027 			nep->ne_rtable[i] = 0;
2028 		}
2029 }
2030 
2031 int
2032 vfs_export(struct mount *mp, struct netexport *nep,
2033 	   const struct export_args *argp)
2034 {
2035 	int error;
2036 
2037 	if (argp->ex_flags & MNT_DELEXPORT) {
2038 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2039 			vfs_setpublicfs(NULL, NULL, NULL);
2040 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2041 		}
2042 		vfs_free_addrlist(nep);
2043 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2044 	}
2045 	if (argp->ex_flags & MNT_EXPORTED) {
2046 		if (argp->ex_flags & MNT_EXPUBLIC) {
2047 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2048 				return (error);
2049 			mp->mnt_flag |= MNT_EXPUBLIC;
2050 		}
2051 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2052 			return (error);
2053 		mp->mnt_flag |= MNT_EXPORTED;
2054 	}
2055 	return (0);
2056 }
2057 
2058 
2059 /*
2060  * Set the publicly exported filesystem (WebNFS). Currently, only
2061  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2062  */
2063 int
2064 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
2065 		const struct export_args *argp)
2066 {
2067 	int error;
2068 	struct vnode *rvp;
2069 	char *cp;
2070 
2071 	/*
2072 	 * mp == NULL -> invalidate the current info, the FS is
2073 	 * no longer exported. May be called from either vfs_export
2074 	 * or unmount, so check if it hasn't already been done.
2075 	 */
2076 	if (mp == NULL) {
2077 		if (nfs_pub.np_valid) {
2078 			nfs_pub.np_valid = 0;
2079 			if (nfs_pub.np_index != NULL) {
2080 				kfree(nfs_pub.np_index, M_TEMP);
2081 				nfs_pub.np_index = NULL;
2082 			}
2083 		}
2084 		return (0);
2085 	}
2086 
2087 	/*
2088 	 * Only one allowed at a time.
2089 	 */
2090 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2091 		return (EBUSY);
2092 
2093 	/*
2094 	 * Get real filehandle for root of exported FS.
2095 	 */
2096 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2097 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2098 
2099 	if ((error = VFS_ROOT(mp, &rvp)))
2100 		return (error);
2101 
2102 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2103 		return (error);
2104 
2105 	vput(rvp);
2106 
2107 	/*
2108 	 * If an indexfile was specified, pull it in.
2109 	 */
2110 	if (argp->ex_indexfile != NULL) {
2111 		int namelen;
2112 
2113 		error = vn_get_namelen(rvp, &namelen);
2114 		if (error)
2115 			return (error);
2116 		nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
2117 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2118 		    namelen, NULL);
2119 		if (!error) {
2120 			/*
2121 			 * Check for illegal filenames.
2122 			 */
2123 			for (cp = nfs_pub.np_index; *cp; cp++) {
2124 				if (*cp == '/') {
2125 					error = EINVAL;
2126 					break;
2127 				}
2128 			}
2129 		}
2130 		if (error) {
2131 			kfree(nfs_pub.np_index, M_TEMP);
2132 			return (error);
2133 		}
2134 	}
2135 
2136 	nfs_pub.np_mount = mp;
2137 	nfs_pub.np_valid = 1;
2138 	return (0);
2139 }
2140 
2141 struct netcred *
2142 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2143 		struct sockaddr *nam)
2144 {
2145 	struct netcred *np;
2146 	struct radix_node_head *rnh;
2147 	struct sockaddr *saddr;
2148 
2149 	np = NULL;
2150 	if (mp->mnt_flag & MNT_EXPORTED) {
2151 		/*
2152 		 * Lookup in the export list first.
2153 		 */
2154 		if (nam != NULL) {
2155 			saddr = nam;
2156 			rnh = nep->ne_rtable[saddr->sa_family];
2157 			if (rnh != NULL) {
2158 				np = (struct netcred *)
2159 					(*rnh->rnh_matchaddr)((char *)saddr,
2160 							      rnh);
2161 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2162 					np = NULL;
2163 			}
2164 		}
2165 		/*
2166 		 * If no address match, use the default if it exists.
2167 		 */
2168 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2169 			np = &nep->ne_defexported;
2170 	}
2171 	return (np);
2172 }
2173 
2174 /*
2175  * perform msync on all vnodes under a mount point.  The mount point must
2176  * be locked.  This code is also responsible for lazy-freeing unreferenced
2177  * vnodes whos VM objects no longer contain pages.
2178  *
2179  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2180  *
2181  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2182  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
2183  * way up in this high level function.
2184  */
2185 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2186 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2187 
2188 void
2189 vfs_msync(struct mount *mp, int flags)
2190 {
2191 	int vmsc_flags;
2192 
2193 	/*
2194 	 * tmpfs sets this flag to prevent msync(), sync, and the
2195 	 * filesystem periodic syncer from trying to flush VM pages
2196 	 * to swap.  Only pure memory pressure flushes tmpfs VM pages
2197 	 * to swap.
2198 	 */
2199 	if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2200 		return;
2201 
2202 	/*
2203 	 * Ok, scan the vnodes for work.  If the filesystem is using the
2204 	 * syncer thread feature we can use vsyncscan() instead of
2205 	 * vmntvnodescan(), which is much faster.
2206 	 */
2207 	vmsc_flags = VMSC_GETVP;
2208 	if (flags != MNT_WAIT)
2209 		vmsc_flags |= VMSC_NOWAIT;
2210 
2211 	if (mp->mnt_kern_flag & MNTK_THR_SYNC) {
2212 		vsyncscan(mp, vmsc_flags, vfs_msync_scan2,
2213 			  (void *)(intptr_t)flags);
2214 	} else {
2215 		vmntvnodescan(mp, vmsc_flags,
2216 			      vfs_msync_scan1, vfs_msync_scan2,
2217 			      (void *)(intptr_t)flags);
2218 	}
2219 }
2220 
2221 /*
2222  * scan1 is a fast pre-check.  There could be hundreds of thousands of
2223  * vnodes, we cannot afford to do anything heavy weight until we have a
2224  * fairly good indication that there is work to do.
2225  */
2226 static
2227 int
2228 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2229 {
2230 	int flags = (int)(intptr_t)data;
2231 
2232 	if ((vp->v_flag & VRECLAIMED) == 0) {
2233 		if (vshouldmsync(vp))
2234 			return(0);	/* call scan2 */
2235 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2236 		    (vp->v_flag & VOBJDIRTY) &&
2237 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2238 			return(0);	/* call scan2 */
2239 		}
2240 	}
2241 
2242 	/*
2243 	 * do not call scan2, continue the loop
2244 	 */
2245 	return(-1);
2246 }
2247 
2248 /*
2249  * This callback is handed a locked vnode.
2250  */
2251 static
2252 int
2253 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2254 {
2255 	vm_object_t obj;
2256 	int flags = (int)(intptr_t)data;
2257 
2258 	if (vp->v_flag & VRECLAIMED)
2259 		return(0);
2260 
2261 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2262 		if ((obj = vp->v_object) != NULL) {
2263 			vm_object_page_clean(obj, 0, 0,
2264 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2265 		}
2266 	}
2267 	return(0);
2268 }
2269 
2270 /*
2271  * Wake up anyone interested in vp because it is being revoked.
2272  */
2273 void
2274 vn_gone(struct vnode *vp)
2275 {
2276 	lwkt_gettoken(&vp->v_token);
2277 	KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
2278 	lwkt_reltoken(&vp->v_token);
2279 }
2280 
2281 /*
2282  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2283  * (or v_rdev might be NULL).
2284  */
2285 cdev_t
2286 vn_todev(struct vnode *vp)
2287 {
2288 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2289 		return (NULL);
2290 	KKASSERT(vp->v_rdev != NULL);
2291 	return (vp->v_rdev);
2292 }
2293 
2294 /*
2295  * Check if vnode represents a disk device.  The vnode does not need to be
2296  * opened.
2297  *
2298  * MPALMOSTSAFE
2299  */
2300 int
2301 vn_isdisk(struct vnode *vp, int *errp)
2302 {
2303 	cdev_t dev;
2304 
2305 	if (vp->v_type != VCHR) {
2306 		if (errp != NULL)
2307 			*errp = ENOTBLK;
2308 		return (0);
2309 	}
2310 
2311 	dev = vp->v_rdev;
2312 
2313 	if (dev == NULL) {
2314 		if (errp != NULL)
2315 			*errp = ENXIO;
2316 		return (0);
2317 	}
2318 	if (dev_is_good(dev) == 0) {
2319 		if (errp != NULL)
2320 			*errp = ENXIO;
2321 		return (0);
2322 	}
2323 	if ((dev_dflags(dev) & D_DISK) == 0) {
2324 		if (errp != NULL)
2325 			*errp = ENOTBLK;
2326 		return (0);
2327 	}
2328 	if (errp != NULL)
2329 		*errp = 0;
2330 	return (1);
2331 }
2332 
2333 int
2334 vn_get_namelen(struct vnode *vp, int *namelen)
2335 {
2336 	int error;
2337 	register_t retval[2];
2338 
2339 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2340 	if (error)
2341 		return (error);
2342 	*namelen = (int)retval[0];
2343 	return (0);
2344 }
2345 
2346 int
2347 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2348 		uint16_t d_namlen, const char *d_name)
2349 {
2350 	struct dirent *dp;
2351 	size_t len;
2352 
2353 	len = _DIRENT_RECLEN(d_namlen);
2354 	if (len > uio->uio_resid)
2355 		return(1);
2356 
2357 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2358 
2359 	dp->d_ino = d_ino;
2360 	dp->d_namlen = d_namlen;
2361 	dp->d_type = d_type;
2362 	bcopy(d_name, dp->d_name, d_namlen);
2363 
2364 	*error = uiomove((caddr_t)dp, len, uio);
2365 
2366 	kfree(dp, M_TEMP);
2367 
2368 	return(0);
2369 }
2370 
2371 void
2372 vn_mark_atime(struct vnode *vp, struct thread *td)
2373 {
2374 	struct proc *p = td->td_proc;
2375 	struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2376 
2377 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2378 		VOP_MARKATIME(vp, cred);
2379 	}
2380 }
2381