xref: /dflybsd-src/sys/kern/vfs_subr.c (revision 1c4f2fa48567f20b32ef2963ca51c6d456c4b58b)
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/priv.h>
63 #include <sys/proc.h>
64 #include <sys/reboot.h>
65 #include <sys/socket.h>
66 #include <sys/stat.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/unistd.h>
70 #include <sys/vmmeter.h>
71 #include <sys/vnode.h>
72 
73 #include <machine/limits.h>
74 
75 #include <vm/vm.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_kern.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_pager.h>
83 #include <vm/vnode_pager.h>
84 #include <vm/vm_zone.h>
85 
86 #include <sys/buf2.h>
87 #include <sys/thread2.h>
88 #include <sys/sysref2.h>
89 #include <sys/mplock2.h>
90 
91 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
92 
93 int numvnodes;
94 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
95 int vfs_fastdev = 1;
96 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
97 
98 enum vtype iftovt_tab[16] = {
99 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
100 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
101 };
102 int vttoif_tab[9] = {
103 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
104 	S_IFSOCK, S_IFIFO, S_IFMT,
105 };
106 
107 static int reassignbufcalls;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
109 		&reassignbufcalls, 0, "");
110 static int reassignbufloops;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
112 		&reassignbufloops, 0, "");
113 static int reassignbufsortgood;
114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
115 		&reassignbufsortgood, 0, "");
116 static int reassignbufsortbad;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
118 		&reassignbufsortbad, 0, "");
119 static int reassignbufmethod = 1;
120 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
121 		&reassignbufmethod, 0, "");
122 
123 int	nfs_mount_type = -1;
124 static struct lwkt_token spechash_token;
125 struct nfs_public nfs_pub;	/* publicly exported FS */
126 
127 int desiredvnodes;
128 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
129 		&desiredvnodes, 0, "Maximum number of vnodes");
130 
131 static void	vfs_free_addrlist (struct netexport *nep);
132 static int	vfs_free_netcred (struct radix_node *rn, void *w);
133 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
134 				       const struct export_args *argp);
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 	int factor1;
177 	int factor2;
178 
179 	/*
180 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
181 	 * according to available system memory but we may also have
182 	 * to limit it based on available KVM, which is capped on 32 bit
183 	 * systems.
184 	 *
185 	 * WARNING!  For machines with 64-256M of ram we have to be sure
186 	 *	     that the default limit scales down well due to HAMMER
187 	 *	     taking up significantly more memory per-vnode vs UFS.
188 	 *	     We want around ~5800 on a 128M machine.
189 	 */
190 	factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
191 	factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode));
192 	desiredvnodes =
193 		imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
194 		     KvaSize / factor2);
195 	desiredvnodes = imax(desiredvnodes, maxproc * 8);
196 
197 	lwkt_token_init(&spechash_token);
198 }
199 
200 /*
201  * Knob to control the precision of file timestamps:
202  *
203  *   0 = seconds only; nanoseconds zeroed.
204  *   1 = seconds and nanoseconds, accurate within 1/HZ.
205  *   2 = seconds and nanoseconds, truncated to microseconds.
206  * >=3 = seconds and nanoseconds, maximum precision.
207  */
208 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
209 
210 static int timestamp_precision = TSP_SEC;
211 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
212 		&timestamp_precision, 0, "");
213 
214 /*
215  * Get a current timestamp.
216  *
217  * MPSAFE
218  */
219 void
220 vfs_timestamp(struct timespec *tsp)
221 {
222 	struct timeval tv;
223 
224 	switch (timestamp_precision) {
225 	case TSP_SEC:
226 		tsp->tv_sec = time_second;
227 		tsp->tv_nsec = 0;
228 		break;
229 	case TSP_HZ:
230 		getnanotime(tsp);
231 		break;
232 	case TSP_USEC:
233 		microtime(&tv);
234 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
235 		break;
236 	case TSP_NSEC:
237 	default:
238 		nanotime(tsp);
239 		break;
240 	}
241 }
242 
243 /*
244  * Set vnode attributes to VNOVAL
245  */
246 void
247 vattr_null(struct vattr *vap)
248 {
249 	vap->va_type = VNON;
250 	vap->va_size = VNOVAL;
251 	vap->va_bytes = VNOVAL;
252 	vap->va_mode = VNOVAL;
253 	vap->va_nlink = VNOVAL;
254 	vap->va_uid = VNOVAL;
255 	vap->va_gid = VNOVAL;
256 	vap->va_fsid = VNOVAL;
257 	vap->va_fileid = VNOVAL;
258 	vap->va_blocksize = VNOVAL;
259 	vap->va_rmajor = VNOVAL;
260 	vap->va_rminor = VNOVAL;
261 	vap->va_atime.tv_sec = VNOVAL;
262 	vap->va_atime.tv_nsec = VNOVAL;
263 	vap->va_mtime.tv_sec = VNOVAL;
264 	vap->va_mtime.tv_nsec = VNOVAL;
265 	vap->va_ctime.tv_sec = VNOVAL;
266 	vap->va_ctime.tv_nsec = VNOVAL;
267 	vap->va_flags = VNOVAL;
268 	vap->va_gen = VNOVAL;
269 	vap->va_vaflags = 0;
270 	/* va_*_uuid fields are only valid if related flags are set */
271 }
272 
273 /*
274  * Flush out and invalidate all buffers associated with a vnode.
275  *
276  * vp must be locked.
277  */
278 static int vinvalbuf_bp(struct buf *bp, void *data);
279 
280 struct vinvalbuf_bp_info {
281 	struct vnode *vp;
282 	int slptimeo;
283 	int lkflags;
284 	int flags;
285 };
286 
287 int
288 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
289 {
290 	struct vinvalbuf_bp_info info;
291 	vm_object_t object;
292 	lwkt_tokref vlock;
293 	int error;
294 
295 	lwkt_gettoken(&vlock, &vp->v_token);
296 
297 	/*
298 	 * If we are being asked to save, call fsync to ensure that the inode
299 	 * is updated.
300 	 */
301 	if (flags & V_SAVE) {
302 		error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
303 		if (error)
304 			goto done;
305 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
306 			if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
307 				goto done;
308 
309 			/*
310 			 * Dirty bufs may be left or generated via races
311 			 * in circumstances where vinvalbuf() is called on
312 			 * a vnode not undergoing reclamation.   Only
313 			 * panic if we are trying to reclaim the vnode.
314 			 */
315 			if ((vp->v_flag & VRECLAIMED) &&
316 			    (bio_track_active(&vp->v_track_write) ||
317 			    !RB_EMPTY(&vp->v_rbdirty_tree))) {
318 				panic("vinvalbuf: dirty bufs");
319 			}
320 		}
321   	}
322 	info.slptimeo = slptimeo;
323 	info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
324 	if (slpflag & PCATCH)
325 		info.lkflags |= LK_PCATCH;
326 	info.flags = flags;
327 	info.vp = vp;
328 
329 	/*
330 	 * Flush the buffer cache until nothing is left.
331 	 */
332 	while (!RB_EMPTY(&vp->v_rbclean_tree) ||
333 	       !RB_EMPTY(&vp->v_rbdirty_tree)) {
334 		error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
335 				vinvalbuf_bp, &info);
336 		if (error == 0) {
337 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
338 					vinvalbuf_bp, &info);
339 		}
340 	}
341 
342 	/*
343 	 * Wait for I/O completion.  We may block in the pip code so we have
344 	 * to re-check.
345 	 */
346 	do {
347 		bio_track_wait(&vp->v_track_write, 0, 0);
348 		if ((object = vp->v_object) != NULL) {
349 			while (object->paging_in_progress)
350 				vm_object_pip_sleep(object, "vnvlbx");
351 		}
352 	} while (bio_track_active(&vp->v_track_write));
353 
354 	/*
355 	 * Destroy the copy in the VM cache, too.
356 	 */
357 	if ((object = vp->v_object) != NULL) {
358 		vm_object_page_remove(object, 0, 0,
359 			(flags & V_SAVE) ? TRUE : FALSE);
360 	}
361 
362 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
363 		panic("vinvalbuf: flush failed");
364 	if (!RB_EMPTY(&vp->v_rbhash_tree))
365 		panic("vinvalbuf: flush failed, buffers still present");
366 	error = 0;
367 done:
368 	lwkt_reltoken(&vlock);
369 	return (error);
370 }
371 
372 static int
373 vinvalbuf_bp(struct buf *bp, void *data)
374 {
375 	struct vinvalbuf_bp_info *info = data;
376 	int error;
377 
378 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
379 		error = BUF_TIMELOCK(bp, info->lkflags,
380 				     "vinvalbuf", info->slptimeo);
381 		if (error == 0) {
382 			BUF_UNLOCK(bp);
383 			error = ENOLCK;
384 		}
385 		if (error == ENOLCK)
386 			return(0);
387 		return (-error);
388 	}
389 
390 	KKASSERT(bp->b_vp == info->vp);
391 
392 	/*
393 	 * XXX Since there are no node locks for NFS, I
394 	 * believe there is a slight chance that a delayed
395 	 * write will occur while sleeping just above, so
396 	 * check for it.  Note that vfs_bio_awrite expects
397 	 * buffers to reside on a queue, while bwrite() and
398 	 * brelse() do not.
399 	 *
400 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
401 	 * check.  This code will write out the buffer, period.
402 	 */
403 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
404 	    (info->flags & V_SAVE)) {
405 		if (bp->b_vp == info->vp) {
406 			if (bp->b_flags & B_CLUSTEROK) {
407 				vfs_bio_awrite(bp);
408 			} else {
409 				bremfree(bp);
410 				bawrite(bp);
411 			}
412 		} else {
413 			bremfree(bp);
414 			bwrite(bp);
415 		}
416 	} else if (info->flags & V_SAVE) {
417 		/*
418 		 * Cannot set B_NOCACHE on a clean buffer as this will
419 		 * destroy the VM backing store which might actually
420 		 * be dirty (and unsynchronized).
421 		 */
422 		bremfree(bp);
423 		bp->b_flags |= (B_INVAL | B_RELBUF);
424 		brelse(bp);
425 	} else {
426 		bremfree(bp);
427 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
428 		brelse(bp);
429 	}
430 	return(0);
431 }
432 
433 /*
434  * Truncate a file's buffer and pages to a specified length.  This
435  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
436  * sync activity.
437  *
438  * The vnode must be locked.
439  */
440 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
441 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
442 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
443 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
444 
445 int
446 vtruncbuf(struct vnode *vp, off_t length, int blksize)
447 {
448 	off_t truncloffset;
449 	const char *filename;
450 	lwkt_tokref vlock;
451 	int count;
452 
453 	/*
454 	 * Round up to the *next* block, then destroy the buffers in question.
455 	 * Since we are only removing some of the buffers we must rely on the
456 	 * scan count to determine whether a loop is necessary.
457 	 */
458 	if ((count = (int)(length % blksize)) != 0)
459 		truncloffset = length + (blksize - count);
460 	else
461 		truncloffset = length;
462 
463 	lwkt_gettoken(&vlock, &vp->v_token);
464 	do {
465 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
466 				vtruncbuf_bp_trunc_cmp,
467 				vtruncbuf_bp_trunc, &truncloffset);
468 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
469 				vtruncbuf_bp_trunc_cmp,
470 				vtruncbuf_bp_trunc, &truncloffset);
471 	} while(count);
472 
473 	/*
474 	 * For safety, fsync any remaining metadata if the file is not being
475 	 * truncated to 0.  Since the metadata does not represent the entire
476 	 * dirty list we have to rely on the hit count to ensure that we get
477 	 * all of it.
478 	 */
479 	if (length > 0) {
480 		do {
481 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
482 					vtruncbuf_bp_metasync_cmp,
483 					vtruncbuf_bp_metasync, vp);
484 		} while (count);
485 	}
486 
487 	/*
488 	 * Clean out any left over VM backing store.
489 	 *
490 	 * It is possible to have in-progress I/O from buffers that were
491 	 * not part of the truncation.  This should not happen if we
492 	 * are truncating to 0-length.
493 	 */
494 	vnode_pager_setsize(vp, length);
495 	bio_track_wait(&vp->v_track_write, 0, 0);
496 
497 	/*
498 	 * Debugging only
499 	 */
500 	spin_lock_wr(&vp->v_spinlock);
501 	filename = TAILQ_FIRST(&vp->v_namecache) ?
502 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
503 	spin_unlock_wr(&vp->v_spinlock);
504 
505 	/*
506 	 * Make sure no buffers were instantiated while we were trying
507 	 * to clean out the remaining VM pages.  This could occur due
508 	 * to busy dirty VM pages being flushed out to disk.
509 	 */
510 	do {
511 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
512 				vtruncbuf_bp_trunc_cmp,
513 				vtruncbuf_bp_trunc, &truncloffset);
514 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
515 				vtruncbuf_bp_trunc_cmp,
516 				vtruncbuf_bp_trunc, &truncloffset);
517 		if (count) {
518 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
519 			       "left over buffers in %s\n", count, filename);
520 		}
521 	} while(count);
522 
523 	lwkt_reltoken(&vlock);
524 
525 	return (0);
526 }
527 
528 /*
529  * The callback buffer is beyond the new file EOF and must be destroyed.
530  * Note that the compare function must conform to the RB_SCAN's requirements.
531  */
532 static
533 int
534 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
535 {
536 	if (bp->b_loffset >= *(off_t *)data)
537 		return(0);
538 	return(-1);
539 }
540 
541 static
542 int
543 vtruncbuf_bp_trunc(struct buf *bp, void *data)
544 {
545 	/*
546 	 * Do not try to use a buffer we cannot immediately lock, but sleep
547 	 * anyway to prevent a livelock.  The code will loop until all buffers
548 	 * can be acted upon.
549 	 */
550 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
551 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
552 			BUF_UNLOCK(bp);
553 	} else {
554 		bremfree(bp);
555 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
556 		brelse(bp);
557 	}
558 	return(1);
559 }
560 
561 /*
562  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
563  * blocks (with a negative loffset) are scanned.
564  * Note that the compare function must conform to the RB_SCAN's requirements.
565  */
566 static int
567 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
568 {
569 	if (bp->b_loffset < 0)
570 		return(0);
571 	return(1);
572 }
573 
574 static int
575 vtruncbuf_bp_metasync(struct buf *bp, void *data)
576 {
577 	struct vnode *vp = data;
578 
579 	if (bp->b_flags & B_DELWRI) {
580 		/*
581 		 * Do not try to use a buffer we cannot immediately lock,
582 		 * but sleep anyway to prevent a livelock.  The code will
583 		 * loop until all buffers can be acted upon.
584 		 */
585 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
586 			if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
587 				BUF_UNLOCK(bp);
588 		} else {
589 			bremfree(bp);
590 			if (bp->b_vp == vp)
591 				bawrite(bp);
592 			else
593 				bwrite(bp);
594 		}
595 		return(1);
596 	} else {
597 		return(0);
598 	}
599 }
600 
601 /*
602  * vfsync - implements a multipass fsync on a file which understands
603  * dependancies and meta-data.  The passed vnode must be locked.  The
604  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
605  *
606  * When fsyncing data asynchronously just do one consolidated pass starting
607  * with the most negative block number.  This may not get all the data due
608  * to dependancies.
609  *
610  * When fsyncing data synchronously do a data pass, then a metadata pass,
611  * then do additional data+metadata passes to try to get all the data out.
612  */
613 static int vfsync_wait_output(struct vnode *vp,
614 			    int (*waitoutput)(struct vnode *, struct thread *));
615 static int vfsync_data_only_cmp(struct buf *bp, void *data);
616 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
617 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
618 static int vfsync_bp(struct buf *bp, void *data);
619 
620 struct vfsync_info {
621 	struct vnode *vp;
622 	int synchronous;
623 	int syncdeps;
624 	int lazycount;
625 	int lazylimit;
626 	int skippedbufs;
627 	int (*checkdef)(struct buf *);
628 };
629 
630 int
631 vfsync(struct vnode *vp, int waitfor, int passes,
632 	int (*checkdef)(struct buf *),
633 	int (*waitoutput)(struct vnode *, struct thread *))
634 {
635 	struct vfsync_info info;
636 	lwkt_tokref vlock;
637 	int error;
638 
639 	bzero(&info, sizeof(info));
640 	info.vp = vp;
641 	if ((info.checkdef = checkdef) == NULL)
642 		info.syncdeps = 1;
643 
644 	lwkt_gettoken(&vlock, &vp->v_token);
645 
646 	switch(waitfor) {
647 	case MNT_LAZY:
648 		/*
649 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
650 		 * number of data (not meta) pages we try to flush to 1MB.
651 		 * A non-zero return means that lazy limit was reached.
652 		 */
653 		info.lazylimit = 1024 * 1024;
654 		info.syncdeps = 1;
655 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
656 				vfsync_lazy_range_cmp, vfsync_bp, &info);
657 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
658 				vfsync_meta_only_cmp, vfsync_bp, &info);
659 		if (error == 0)
660 			vp->v_lazyw = 0;
661 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
662 			vn_syncer_add_to_worklist(vp, 1);
663 		error = 0;
664 		break;
665 	case MNT_NOWAIT:
666 		/*
667 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
668 		 */
669 		info.syncdeps = 1;
670 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
671 			vfsync_bp, &info);
672 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
673 			vfsync_bp, &info);
674 		error = 0;
675 		break;
676 	default:
677 		/*
678 		 * Synchronous.  Do a data-only pass, then a meta-data+data
679 		 * pass, then additional integrated passes to try to get
680 		 * all the dependancies flushed.
681 		 */
682 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
683 			vfsync_bp, &info);
684 		error = vfsync_wait_output(vp, waitoutput);
685 		if (error == 0) {
686 			info.skippedbufs = 0;
687 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
688 				vfsync_bp, &info);
689 			error = vfsync_wait_output(vp, waitoutput);
690 			if (info.skippedbufs)
691 				kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
692 		}
693 		while (error == 0 && passes > 0 &&
694 		       !RB_EMPTY(&vp->v_rbdirty_tree)
695 		) {
696 			if (--passes == 0) {
697 				info.synchronous = 1;
698 				info.syncdeps = 1;
699 			}
700 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
701 				vfsync_bp, &info);
702 			if (error < 0)
703 				error = -error;
704 			info.syncdeps = 1;
705 			if (error == 0)
706 				error = vfsync_wait_output(vp, waitoutput);
707 		}
708 		break;
709 	}
710 	lwkt_reltoken(&vlock);
711 	return(error);
712 }
713 
714 static int
715 vfsync_wait_output(struct vnode *vp,
716 		   int (*waitoutput)(struct vnode *, struct thread *))
717 {
718 	int error;
719 
720 	error = bio_track_wait(&vp->v_track_write, 0, 0);
721 	if (waitoutput)
722 		error = waitoutput(vp, curthread);
723 	return(error);
724 }
725 
726 static int
727 vfsync_data_only_cmp(struct buf *bp, void *data)
728 {
729 	if (bp->b_loffset < 0)
730 		return(-1);
731 	return(0);
732 }
733 
734 static int
735 vfsync_meta_only_cmp(struct buf *bp, void *data)
736 {
737 	if (bp->b_loffset < 0)
738 		return(0);
739 	return(1);
740 }
741 
742 static int
743 vfsync_lazy_range_cmp(struct buf *bp, void *data)
744 {
745 	struct vfsync_info *info = data;
746 	if (bp->b_loffset < info->vp->v_lazyw)
747 		return(-1);
748 	return(0);
749 }
750 
751 static int
752 vfsync_bp(struct buf *bp, void *data)
753 {
754 	struct vfsync_info *info = data;
755 	struct vnode *vp = info->vp;
756 	int error;
757 
758 	/*
759 	 * if syncdeps is not set we do not try to write buffers which have
760 	 * dependancies.
761 	 */
762 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
763 		return(0);
764 
765 	/*
766 	 * Ignore buffers that we cannot immediately lock.  XXX
767 	 */
768 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
769 		kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
770 		++info->skippedbufs;
771 		return(0);
772 	}
773 	if ((bp->b_flags & B_DELWRI) == 0)
774 		panic("vfsync_bp: buffer not dirty");
775 	if (vp != bp->b_vp)
776 		panic("vfsync_bp: buffer vp mismatch");
777 
778 	/*
779 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
780 	 * has been written but an additional handshake with the device
781 	 * is required before we can dispose of the buffer.  We have no idea
782 	 * how to do this so we have to skip these buffers.
783 	 */
784 	if (bp->b_flags & B_NEEDCOMMIT) {
785 		BUF_UNLOCK(bp);
786 		return(0);
787 	}
788 
789 	/*
790 	 * Ask bioops if it is ok to sync
791 	 */
792 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
793 		bremfree(bp);
794 		brelse(bp);
795 		return(0);
796 	}
797 
798 	if (info->synchronous) {
799 		/*
800 		 * Synchronous flushing.  An error may be returned.
801 		 */
802 		bremfree(bp);
803 		error = bwrite(bp);
804 	} else {
805 		/*
806 		 * Asynchronous flushing.  A negative return value simply
807 		 * stops the scan and is not considered an error.  We use
808 		 * this to support limited MNT_LAZY flushes.
809 		 */
810 		vp->v_lazyw = bp->b_loffset;
811 		if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
812 			info->lazycount += vfs_bio_awrite(bp);
813 		} else {
814 			info->lazycount += bp->b_bufsize;
815 			bremfree(bp);
816 			bawrite(bp);
817 		}
818 		if (info->lazylimit && info->lazycount >= info->lazylimit)
819 			error = 1;
820 		else
821 			error = 0;
822 	}
823 	return(-error);
824 }
825 
826 /*
827  * Associate a buffer with a vnode.
828  *
829  * MPSAFE
830  */
831 int
832 bgetvp(struct vnode *vp, struct buf *bp)
833 {
834 	lwkt_tokref vlock;
835 
836 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
837 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
838 
839 	/*
840 	 * Insert onto list for new vnode.
841 	 */
842 	lwkt_gettoken(&vlock, &vp->v_token);
843 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
844 		lwkt_reltoken(&vlock);
845 		return (EEXIST);
846 	}
847 	bp->b_vp = vp;
848 	bp->b_flags |= B_HASHED;
849 	bp->b_flags |= B_VNCLEAN;
850 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
851 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
852 	vhold(vp);
853 	lwkt_reltoken(&vlock);
854 	return(0);
855 }
856 
857 /*
858  * Disassociate a buffer from a vnode.
859  */
860 void
861 brelvp(struct buf *bp)
862 {
863 	struct vnode *vp;
864 	lwkt_tokref vlock;
865 
866 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
867 
868 	/*
869 	 * Delete from old vnode list, if on one.
870 	 */
871 	vp = bp->b_vp;
872 	lwkt_gettoken(&vlock, &vp->v_token);
873 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
874 		if (bp->b_flags & B_VNDIRTY)
875 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
876 		else
877 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
878 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
879 	}
880 	if (bp->b_flags & B_HASHED) {
881 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
882 		bp->b_flags &= ~B_HASHED;
883 	}
884 	if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
885 		vclrflags(vp, VONWORKLST);
886 		LIST_REMOVE(vp, v_synclist);
887 	}
888 	bp->b_vp = NULL;
889 	lwkt_reltoken(&vlock);
890 
891 	vdrop(vp);
892 }
893 
894 /*
895  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
896  * This routine is called when the state of the B_DELWRI bit is changed.
897  *
898  * MPSAFE
899  */
900 void
901 reassignbuf(struct buf *bp)
902 {
903 	struct vnode *vp = bp->b_vp;
904 	lwkt_tokref vlock;
905 	int delay;
906 
907 	KKASSERT(vp != NULL);
908 	++reassignbufcalls;
909 
910 	/*
911 	 * B_PAGING flagged buffers cannot be reassigned because their vp
912 	 * is not fully linked in.
913 	 */
914 	if (bp->b_flags & B_PAGING)
915 		panic("cannot reassign paging buffer");
916 
917 	lwkt_gettoken(&vlock, &vp->v_token);
918 	if (bp->b_flags & B_DELWRI) {
919 		/*
920 		 * Move to the dirty list, add the vnode to the worklist
921 		 */
922 		if (bp->b_flags & B_VNCLEAN) {
923 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
924 			bp->b_flags &= ~B_VNCLEAN;
925 		}
926 		if ((bp->b_flags & B_VNDIRTY) == 0) {
927 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
928 				panic("reassignbuf: dup lblk vp %p bp %p",
929 				      vp, bp);
930 			}
931 			bp->b_flags |= B_VNDIRTY;
932 		}
933 		if ((vp->v_flag & VONWORKLST) == 0) {
934 			switch (vp->v_type) {
935 			case VDIR:
936 				delay = dirdelay;
937 				break;
938 			case VCHR:
939 			case VBLK:
940 				if (vp->v_rdev &&
941 				    vp->v_rdev->si_mountpoint != NULL) {
942 					delay = metadelay;
943 					break;
944 				}
945 				/* fall through */
946 			default:
947 				delay = filedelay;
948 			}
949 			vn_syncer_add_to_worklist(vp, delay);
950 		}
951 	} else {
952 		/*
953 		 * Move to the clean list, remove the vnode from the worklist
954 		 * if no dirty blocks remain.
955 		 */
956 		if (bp->b_flags & B_VNDIRTY) {
957 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
958 			bp->b_flags &= ~B_VNDIRTY;
959 		}
960 		if ((bp->b_flags & B_VNCLEAN) == 0) {
961 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
962 				panic("reassignbuf: dup lblk vp %p bp %p",
963 				      vp, bp);
964 			}
965 			bp->b_flags |= B_VNCLEAN;
966 		}
967 		if ((vp->v_flag & VONWORKLST) &&
968 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
969 			vclrflags(vp, VONWORKLST);
970 			LIST_REMOVE(vp, v_synclist);
971 		}
972 	}
973 	lwkt_reltoken(&vlock);
974 }
975 
976 /*
977  * Create a vnode for a block device.
978  * Used for mounting the root file system.
979  */
980 extern struct vop_ops *devfs_vnode_dev_vops_p;
981 int
982 bdevvp(cdev_t dev, struct vnode **vpp)
983 {
984 	struct vnode *vp;
985 	struct vnode *nvp;
986 	int error;
987 
988 	if (dev == NULL) {
989 		*vpp = NULLVP;
990 		return (ENXIO);
991 	}
992 	error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
993 				&nvp, 0, 0);
994 	if (error) {
995 		*vpp = NULLVP;
996 		return (error);
997 	}
998 	vp = nvp;
999 	vp->v_type = VCHR;
1000 #if 0
1001 	vp->v_rdev = dev;
1002 #endif
1003 	v_associate_rdev(vp, dev);
1004 	vp->v_umajor = dev->si_umajor;
1005 	vp->v_uminor = dev->si_uminor;
1006 	vx_unlock(vp);
1007 	*vpp = vp;
1008 	return (0);
1009 }
1010 
1011 int
1012 v_associate_rdev(struct vnode *vp, cdev_t dev)
1013 {
1014 	lwkt_tokref ilock;
1015 
1016 	if (dev == NULL)
1017 		return(ENXIO);
1018 	if (dev_is_good(dev) == 0)
1019 		return(ENXIO);
1020 	KKASSERT(vp->v_rdev == NULL);
1021 	vp->v_rdev = reference_dev(dev);
1022 	lwkt_gettoken(&ilock, &spechash_token);
1023 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1024 	lwkt_reltoken(&ilock);
1025 	return(0);
1026 }
1027 
1028 void
1029 v_release_rdev(struct vnode *vp)
1030 {
1031 	lwkt_tokref ilock;
1032 	cdev_t dev;
1033 
1034 	if ((dev = vp->v_rdev) != NULL) {
1035 		lwkt_gettoken(&ilock, &spechash_token);
1036 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1037 		vp->v_rdev = NULL;
1038 		release_dev(dev);
1039 		lwkt_reltoken(&ilock);
1040 	}
1041 }
1042 
1043 /*
1044  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1045  * the device number with the vnode.  The actual device is not associated
1046  * until the vnode is opened (usually in spec_open()), and will be
1047  * disassociated on last close.
1048  */
1049 void
1050 addaliasu(struct vnode *nvp, int x, int y)
1051 {
1052 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1053 		panic("addaliasu on non-special vnode");
1054 	nvp->v_umajor = x;
1055 	nvp->v_uminor = y;
1056 }
1057 
1058 /*
1059  * Simple call that a filesystem can make to try to get rid of a
1060  * vnode.  It will fail if anyone is referencing the vnode (including
1061  * the caller).
1062  *
1063  * The filesystem can check whether its in-memory inode structure still
1064  * references the vp on return.
1065  */
1066 void
1067 vclean_unlocked(struct vnode *vp)
1068 {
1069 	vx_get(vp);
1070 	if (sysref_isactive(&vp->v_sysref) == 0)
1071 		vgone_vxlocked(vp);
1072 	vx_put(vp);
1073 }
1074 
1075 /*
1076  * Disassociate a vnode from its underlying filesystem.
1077  *
1078  * The vnode must be VX locked and referenced.  In all normal situations
1079  * there are no active references.  If vclean_vxlocked() is called while
1080  * there are active references, the vnode is being ripped out and we have
1081  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1082  */
1083 void
1084 vclean_vxlocked(struct vnode *vp, int flags)
1085 {
1086 	int active;
1087 	int n;
1088 	vm_object_t object;
1089 
1090 	/*
1091 	 * If the vnode has already been reclaimed we have nothing to do.
1092 	 */
1093 	if (vp->v_flag & VRECLAIMED)
1094 		return;
1095 	vsetflags(vp, VRECLAIMED);
1096 
1097 	/*
1098 	 * Scrap the vfs cache
1099 	 */
1100 	while (cache_inval_vp(vp, 0) != 0) {
1101 		kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1102 		tsleep(vp, 0, "vclninv", 2);
1103 	}
1104 
1105 	/*
1106 	 * Check to see if the vnode is in use. If so we have to reference it
1107 	 * before we clean it out so that its count cannot fall to zero and
1108 	 * generate a race against ourselves to recycle it.
1109 	 */
1110 	active = sysref_isactive(&vp->v_sysref);
1111 
1112 	/*
1113 	 * Clean out any buffers associated with the vnode and destroy its
1114 	 * object, if it has one.
1115 	 */
1116 	vinvalbuf(vp, V_SAVE, 0, 0);
1117 
1118 	/*
1119 	 * If purging an active vnode (typically during a forced unmount
1120 	 * or reboot), it must be closed and deactivated before being
1121 	 * reclaimed.  This isn't really all that safe, but what can
1122 	 * we do? XXX.
1123 	 *
1124 	 * Note that neither of these routines unlocks the vnode.
1125 	 */
1126 	if (active && (flags & DOCLOSE)) {
1127 		while ((n = vp->v_opencount) != 0) {
1128 			if (vp->v_writecount)
1129 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1130 			else
1131 				VOP_CLOSE(vp, FNONBLOCK);
1132 			if (vp->v_opencount == n) {
1133 				kprintf("Warning: unable to force-close"
1134 				       " vnode %p\n", vp);
1135 				break;
1136 			}
1137 		}
1138 	}
1139 
1140 	/*
1141 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1142 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1143 	 * again to make sure they all get flushed.
1144 	 *
1145 	 * This can occur if a file with a link count of 0 needs to be
1146 	 * truncated.
1147 	 *
1148 	 * If the vnode is already dead don't try to deactivate it.
1149 	 */
1150 	if ((vp->v_flag & VINACTIVE) == 0) {
1151 		vsetflags(vp, VINACTIVE);
1152 		if (vp->v_mount)
1153 			VOP_INACTIVE(vp);
1154 		vinvalbuf(vp, V_SAVE, 0, 0);
1155 	}
1156 
1157 	/*
1158 	 * If the vnode has an object, destroy it.
1159 	 */
1160 	if ((object = vp->v_object) != NULL) {
1161 		if (object->ref_count == 0) {
1162 			if ((object->flags & OBJ_DEAD) == 0)
1163 				vm_object_terminate(object);
1164 		} else {
1165 			vm_pager_deallocate(object);
1166 		}
1167 		vclrflags(vp, VOBJBUF);
1168 	}
1169 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1170 
1171 	/*
1172 	 * Reclaim the vnode if not already dead.
1173 	 */
1174 	if (vp->v_mount && VOP_RECLAIM(vp))
1175 		panic("vclean: cannot reclaim");
1176 
1177 	/*
1178 	 * Done with purge, notify sleepers of the grim news.
1179 	 */
1180 	vp->v_ops = &dead_vnode_vops_p;
1181 	vn_pollgone(vp);
1182 	vp->v_tag = VT_NON;
1183 
1184 	/*
1185 	 * If we are destroying an active vnode, reactivate it now that
1186 	 * we have reassociated it with deadfs.  This prevents the system
1187 	 * from crashing on the vnode due to it being unexpectedly marked
1188 	 * as inactive or reclaimed.
1189 	 */
1190 	if (active && (flags & DOCLOSE)) {
1191 		vclrflags(vp, VINACTIVE | VRECLAIMED);
1192 	}
1193 }
1194 
1195 /*
1196  * Eliminate all activity associated with the requested vnode
1197  * and with all vnodes aliased to the requested vnode.
1198  *
1199  * The vnode must be referenced but should not be locked.
1200  */
1201 int
1202 vrevoke(struct vnode *vp, struct ucred *cred)
1203 {
1204 	struct vnode *vq;
1205 	struct vnode *vqn;
1206 	lwkt_tokref ilock;
1207 	cdev_t dev;
1208 	int error;
1209 
1210 	/*
1211 	 * If the vnode has a device association, scrap all vnodes associated
1212 	 * with the device.  Don't let the device disappear on us while we
1213 	 * are scrapping the vnodes.
1214 	 *
1215 	 * The passed vp will probably show up in the list, do not VX lock
1216 	 * it twice!
1217 	 *
1218 	 * Releasing the vnode's rdev here can mess up specfs's call to
1219 	 * device close, so don't do it.  The vnode has been disassociated
1220 	 * and the device will be closed after the last ref on the related
1221 	 * fp goes away (if not still open by e.g. the kernel).
1222 	 */
1223 	if (vp->v_type != VCHR) {
1224 		error = fdrevoke(vp, DTYPE_VNODE, cred);
1225 		return (error);
1226 	}
1227 	if ((dev = vp->v_rdev) == NULL) {
1228 		return(0);
1229 	}
1230 	reference_dev(dev);
1231 	lwkt_gettoken(&ilock, &spechash_token);
1232 
1233 	vqn = SLIST_FIRST(&dev->si_hlist);
1234 	if (vqn)
1235 		vref(vqn);
1236 	while ((vq = vqn) != NULL) {
1237 		vqn = SLIST_NEXT(vqn, v_cdevnext);
1238 		if (vqn)
1239 			vref(vqn);
1240 		fdrevoke(vq, DTYPE_VNODE, cred);
1241 		/*v_release_rdev(vq);*/
1242 		vrele(vq);
1243 	}
1244 	lwkt_reltoken(&ilock);
1245 	dev_drevoke(dev);
1246 	release_dev(dev);
1247 	return (0);
1248 }
1249 
1250 /*
1251  * This is called when the object underlying a vnode is being destroyed,
1252  * such as in a remove().  Try to recycle the vnode immediately if the
1253  * only active reference is our reference.
1254  *
1255  * Directory vnodes in the namecache with children cannot be immediately
1256  * recycled because numerous VOP_N*() ops require them to be stable.
1257  *
1258  * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1259  * function is a NOP if VRECLAIMED is already set.
1260  */
1261 int
1262 vrecycle(struct vnode *vp)
1263 {
1264 	if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1265 		if (cache_inval_vp_nonblock(vp))
1266 			return(0);
1267 		vgone_vxlocked(vp);
1268 		return (1);
1269 	}
1270 	return (0);
1271 }
1272 
1273 /*
1274  * Return the maximum I/O size allowed for strategy calls on VP.
1275  *
1276  * If vp is VCHR or VBLK we dive the device, otherwise we use
1277  * the vp's mount info.
1278  */
1279 int
1280 vmaxiosize(struct vnode *vp)
1281 {
1282 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1283 		return(vp->v_rdev->si_iosize_max);
1284 	} else {
1285 		return(vp->v_mount->mnt_iosize_max);
1286 	}
1287 }
1288 
1289 /*
1290  * Eliminate all activity associated with a vnode in preparation for reuse.
1291  *
1292  * The vnode must be VX locked and refd and will remain VX locked and refd
1293  * on return.  This routine may be called with the vnode in any state, as
1294  * long as it is VX locked.  The vnode will be cleaned out and marked
1295  * VRECLAIMED but will not actually be reused until all existing refs and
1296  * holds go away.
1297  *
1298  * NOTE: This routine may be called on a vnode which has not yet been
1299  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1300  * already been reclaimed.
1301  *
1302  * This routine is not responsible for placing us back on the freelist.
1303  * Instead, it happens automatically when the caller releases the VX lock
1304  * (assuming there aren't any other references).
1305  */
1306 void
1307 vgone_vxlocked(struct vnode *vp)
1308 {
1309 	/*
1310 	 * assert that the VX lock is held.  This is an absolute requirement
1311 	 * now for vgone_vxlocked() to be called.
1312 	 */
1313 	KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1314 
1315 	get_mplock();
1316 
1317 	/*
1318 	 * Clean out the filesystem specific data and set the VRECLAIMED
1319 	 * bit.  Also deactivate the vnode if necessary.
1320 	 */
1321 	vclean_vxlocked(vp, DOCLOSE);
1322 
1323 	/*
1324 	 * Delete from old mount point vnode list, if on one.
1325 	 */
1326 	if (vp->v_mount != NULL) {
1327 		KKASSERT(vp->v_data == NULL);
1328 		insmntque(vp, NULL);
1329 	}
1330 
1331 	/*
1332 	 * If special device, remove it from special device alias list
1333 	 * if it is on one.  This should normally only occur if a vnode is
1334 	 * being revoked as the device should otherwise have been released
1335 	 * naturally.
1336 	 */
1337 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1338 		v_release_rdev(vp);
1339 	}
1340 
1341 	/*
1342 	 * Set us to VBAD
1343 	 */
1344 	vp->v_type = VBAD;
1345 	rel_mplock();
1346 }
1347 
1348 /*
1349  * Lookup a vnode by device number.
1350  *
1351  * Returns non-zero and *vpp set to a vref'd vnode on success.
1352  * Returns zero on failure.
1353  */
1354 int
1355 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1356 {
1357 	lwkt_tokref ilock;
1358 	struct vnode *vp;
1359 
1360 	lwkt_gettoken(&ilock, &spechash_token);
1361 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1362 		if (type == vp->v_type) {
1363 			*vpp = vp;
1364 			vref(vp);
1365 			lwkt_reltoken(&ilock);
1366 			return (1);
1367 		}
1368 	}
1369 	lwkt_reltoken(&ilock);
1370 	return (0);
1371 }
1372 
1373 /*
1374  * Calculate the total number of references to a special device.  This
1375  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1376  * an overloaded field.  Since udev2dev can now return NULL, we have
1377  * to check for a NULL v_rdev.
1378  */
1379 int
1380 count_dev(cdev_t dev)
1381 {
1382 	lwkt_tokref ilock;
1383 	struct vnode *vp;
1384 	int count = 0;
1385 
1386 	if (SLIST_FIRST(&dev->si_hlist)) {
1387 		lwkt_gettoken(&ilock, &spechash_token);
1388 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1389 			count += vp->v_opencount;
1390 		}
1391 		lwkt_reltoken(&ilock);
1392 	}
1393 	return(count);
1394 }
1395 
1396 int
1397 vcount(struct vnode *vp)
1398 {
1399 	if (vp->v_rdev == NULL)
1400 		return(0);
1401 	return(count_dev(vp->v_rdev));
1402 }
1403 
1404 /*
1405  * Initialize VMIO for a vnode.  This routine MUST be called before a
1406  * VFS can issue buffer cache ops on a vnode.  It is typically called
1407  * when a vnode is initialized from its inode.
1408  */
1409 int
1410 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1411 {
1412 	vm_object_t object;
1413 	int error = 0;
1414 
1415 retry:
1416 	if ((object = vp->v_object) == NULL) {
1417 		object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1418 		/*
1419 		 * Dereference the reference we just created.  This assumes
1420 		 * that the object is associated with the vp.
1421 		 */
1422 		object->ref_count--;
1423 		vrele(vp);
1424 	} else {
1425 		if (object->flags & OBJ_DEAD) {
1426 			vn_unlock(vp);
1427 			vm_object_dead_sleep(object, "vodead");
1428 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1429 			goto retry;
1430 		}
1431 	}
1432 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1433 	vsetflags(vp, VOBJBUF);
1434 	return (error);
1435 }
1436 
1437 
1438 /*
1439  * Print out a description of a vnode.
1440  */
1441 static char *typename[] =
1442 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1443 
1444 void
1445 vprint(char *label, struct vnode *vp)
1446 {
1447 	char buf[96];
1448 
1449 	if (label != NULL)
1450 		kprintf("%s: %p: ", label, (void *)vp);
1451 	else
1452 		kprintf("%p: ", (void *)vp);
1453 	kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1454 		typename[vp->v_type],
1455 		vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1456 	buf[0] = '\0';
1457 	if (vp->v_flag & VROOT)
1458 		strcat(buf, "|VROOT");
1459 	if (vp->v_flag & VPFSROOT)
1460 		strcat(buf, "|VPFSROOT");
1461 	if (vp->v_flag & VTEXT)
1462 		strcat(buf, "|VTEXT");
1463 	if (vp->v_flag & VSYSTEM)
1464 		strcat(buf, "|VSYSTEM");
1465 	if (vp->v_flag & VFREE)
1466 		strcat(buf, "|VFREE");
1467 	if (vp->v_flag & VOBJBUF)
1468 		strcat(buf, "|VOBJBUF");
1469 	if (buf[0] != '\0')
1470 		kprintf(" flags (%s)", &buf[1]);
1471 	if (vp->v_data == NULL) {
1472 		kprintf("\n");
1473 	} else {
1474 		kprintf("\n\t");
1475 		VOP_PRINT(vp);
1476 	}
1477 }
1478 
1479 /*
1480  * Do the usual access checking.
1481  * file_mode, uid and gid are from the vnode in question,
1482  * while acc_mode and cred are from the VOP_ACCESS parameter list
1483  */
1484 int
1485 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1486     mode_t acc_mode, struct ucred *cred)
1487 {
1488 	mode_t mask;
1489 	int ismember;
1490 
1491 	/*
1492 	 * Super-user always gets read/write access, but execute access depends
1493 	 * on at least one execute bit being set.
1494 	 */
1495 	if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1496 		if ((acc_mode & VEXEC) && type != VDIR &&
1497 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1498 			return (EACCES);
1499 		return (0);
1500 	}
1501 
1502 	mask = 0;
1503 
1504 	/* Otherwise, check the owner. */
1505 	if (cred->cr_uid == uid) {
1506 		if (acc_mode & VEXEC)
1507 			mask |= S_IXUSR;
1508 		if (acc_mode & VREAD)
1509 			mask |= S_IRUSR;
1510 		if (acc_mode & VWRITE)
1511 			mask |= S_IWUSR;
1512 		return ((file_mode & mask) == mask ? 0 : EACCES);
1513 	}
1514 
1515 	/* Otherwise, check the groups. */
1516 	ismember = groupmember(gid, cred);
1517 	if (cred->cr_svgid == gid || ismember) {
1518 		if (acc_mode & VEXEC)
1519 			mask |= S_IXGRP;
1520 		if (acc_mode & VREAD)
1521 			mask |= S_IRGRP;
1522 		if (acc_mode & VWRITE)
1523 			mask |= S_IWGRP;
1524 		return ((file_mode & mask) == mask ? 0 : EACCES);
1525 	}
1526 
1527 	/* Otherwise, check everyone else. */
1528 	if (acc_mode & VEXEC)
1529 		mask |= S_IXOTH;
1530 	if (acc_mode & VREAD)
1531 		mask |= S_IROTH;
1532 	if (acc_mode & VWRITE)
1533 		mask |= S_IWOTH;
1534 	return ((file_mode & mask) == mask ? 0 : EACCES);
1535 }
1536 
1537 #ifdef DDB
1538 #include <ddb/ddb.h>
1539 
1540 static int db_show_locked_vnodes(struct mount *mp, void *data);
1541 
1542 /*
1543  * List all of the locked vnodes in the system.
1544  * Called when debugging the kernel.
1545  */
1546 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1547 {
1548 	kprintf("Locked vnodes\n");
1549 	mountlist_scan(db_show_locked_vnodes, NULL,
1550 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1551 }
1552 
1553 static int
1554 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1555 {
1556 	struct vnode *vp;
1557 
1558 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1559 		if (vn_islocked(vp))
1560 			vprint(NULL, vp);
1561 	}
1562 	return(0);
1563 }
1564 #endif
1565 
1566 /*
1567  * Top level filesystem related information gathering.
1568  */
1569 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1570 
1571 static int
1572 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1573 {
1574 	int *name = (int *)arg1 - 1;	/* XXX */
1575 	u_int namelen = arg2 + 1;	/* XXX */
1576 	struct vfsconf *vfsp;
1577 	int maxtypenum;
1578 
1579 #if 1 || defined(COMPAT_PRELITE2)
1580 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1581 	if (namelen == 1)
1582 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1583 #endif
1584 
1585 #ifdef notyet
1586 	/* all sysctl names at this level are at least name and field */
1587 	if (namelen < 2)
1588 		return (ENOTDIR);		/* overloaded */
1589 	if (name[0] != VFS_GENERIC) {
1590 		vfsp = vfsconf_find_by_typenum(name[0]);
1591 		if (vfsp == NULL)
1592 			return (EOPNOTSUPP);
1593 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1594 		    oldp, oldlenp, newp, newlen, p));
1595 	}
1596 #endif
1597 	switch (name[1]) {
1598 	case VFS_MAXTYPENUM:
1599 		if (namelen != 2)
1600 			return (ENOTDIR);
1601 		maxtypenum = vfsconf_get_maxtypenum();
1602 		return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1603 	case VFS_CONF:
1604 		if (namelen != 3)
1605 			return (ENOTDIR);	/* overloaded */
1606 		vfsp = vfsconf_find_by_typenum(name[2]);
1607 		if (vfsp == NULL)
1608 			return (EOPNOTSUPP);
1609 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1610 	}
1611 	return (EOPNOTSUPP);
1612 }
1613 
1614 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1615 	"Generic filesystem");
1616 
1617 #if 1 || defined(COMPAT_PRELITE2)
1618 
1619 static int
1620 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1621 {
1622 	int error;
1623 	struct ovfsconf ovfs;
1624 	struct sysctl_req *req = (struct sysctl_req*) data;
1625 
1626 	bzero(&ovfs, sizeof(ovfs));
1627 	ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1628 	strcpy(ovfs.vfc_name, vfsp->vfc_name);
1629 	ovfs.vfc_index = vfsp->vfc_typenum;
1630 	ovfs.vfc_refcount = vfsp->vfc_refcount;
1631 	ovfs.vfc_flags = vfsp->vfc_flags;
1632 	error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1633 	if (error)
1634 		return error; /* abort iteration with error code */
1635 	else
1636 		return 0; /* continue iterating with next element */
1637 }
1638 
1639 static int
1640 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1641 {
1642 	return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1643 }
1644 
1645 #endif /* 1 || COMPAT_PRELITE2 */
1646 
1647 /*
1648  * Check to see if a filesystem is mounted on a block device.
1649  */
1650 int
1651 vfs_mountedon(struct vnode *vp)
1652 {
1653 	cdev_t dev;
1654 
1655 	if ((dev = vp->v_rdev) == NULL) {
1656 /*		if (vp->v_type != VBLK)
1657 			dev = get_dev(vp->v_uminor, vp->v_umajor); */
1658 	}
1659 	if (dev != NULL && dev->si_mountpoint)
1660 		return (EBUSY);
1661 	return (0);
1662 }
1663 
1664 /*
1665  * Unmount all filesystems. The list is traversed in reverse order
1666  * of mounting to avoid dependencies.
1667  */
1668 
1669 static int vfs_umountall_callback(struct mount *mp, void *data);
1670 
1671 void
1672 vfs_unmountall(void)
1673 {
1674 	int count;
1675 
1676 	do {
1677 		count = mountlist_scan(vfs_umountall_callback,
1678 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1679 	} while (count);
1680 }
1681 
1682 static
1683 int
1684 vfs_umountall_callback(struct mount *mp, void *data)
1685 {
1686 	int error;
1687 
1688 	error = dounmount(mp, MNT_FORCE);
1689 	if (error) {
1690 		mountlist_remove(mp);
1691 		kprintf("unmount of filesystem mounted from %s failed (",
1692 			mp->mnt_stat.f_mntfromname);
1693 		if (error == EBUSY)
1694 			kprintf("BUSY)\n");
1695 		else
1696 			kprintf("%d)\n", error);
1697 	}
1698 	return(1);
1699 }
1700 
1701 /*
1702  * Checks the mount flags for parameter mp and put the names comma-separated
1703  * into a string buffer buf with a size limit specified by len.
1704  *
1705  * It returns the number of bytes written into buf, and (*errorp) will be
1706  * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1707  * not large enough).  The buffer will be 0-terminated if len was not 0.
1708  */
1709 size_t
1710 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1711 	       char *buf, size_t len, int *errorp)
1712 {
1713 	static const struct mountctl_opt optnames[] = {
1714 		{ MNT_ASYNC,            "asynchronous" },
1715 		{ MNT_EXPORTED,         "NFS exported" },
1716 		{ MNT_LOCAL,            "local" },
1717 		{ MNT_NOATIME,          "noatime" },
1718 		{ MNT_NODEV,            "nodev" },
1719 		{ MNT_NOEXEC,           "noexec" },
1720 		{ MNT_NOSUID,           "nosuid" },
1721 		{ MNT_NOSYMFOLLOW,      "nosymfollow" },
1722 		{ MNT_QUOTA,            "with-quotas" },
1723 		{ MNT_RDONLY,           "read-only" },
1724 		{ MNT_SYNCHRONOUS,      "synchronous" },
1725 		{ MNT_UNION,            "union" },
1726 		{ MNT_NOCLUSTERR,       "noclusterr" },
1727 		{ MNT_NOCLUSTERW,       "noclusterw" },
1728 		{ MNT_SUIDDIR,          "suiddir" },
1729 		{ MNT_SOFTDEP,          "soft-updates" },
1730 		{ MNT_IGNORE,           "ignore" },
1731 		{ 0,			NULL}
1732 	};
1733 	int bwritten;
1734 	int bleft;
1735 	int optlen;
1736 	int actsize;
1737 
1738 	*errorp = 0;
1739 	bwritten = 0;
1740 	bleft = len - 1;	/* leave room for trailing \0 */
1741 
1742 	/*
1743 	 * Checks the size of the string. If it contains
1744 	 * any data, then we will append the new flags to
1745 	 * it.
1746 	 */
1747 	actsize = strlen(buf);
1748 	if (actsize > 0)
1749 		buf += actsize;
1750 
1751 	/* Default flags if no flags passed */
1752 	if (optp == NULL)
1753 		optp = optnames;
1754 
1755 	if (bleft < 0) {	/* degenerate case, 0-length buffer */
1756 		*errorp = EINVAL;
1757 		return(0);
1758 	}
1759 
1760 	for (; flags && optp->o_opt; ++optp) {
1761 		if ((flags & optp->o_opt) == 0)
1762 			continue;
1763 		optlen = strlen(optp->o_name);
1764 		if (bwritten || actsize > 0) {
1765 			if (bleft < 2) {
1766 				*errorp = ENOSPC;
1767 				break;
1768 			}
1769 			buf[bwritten++] = ',';
1770 			buf[bwritten++] = ' ';
1771 			bleft -= 2;
1772 		}
1773 		if (bleft < optlen) {
1774 			*errorp = ENOSPC;
1775 			break;
1776 		}
1777 		bcopy(optp->o_name, buf + bwritten, optlen);
1778 		bwritten += optlen;
1779 		bleft -= optlen;
1780 		flags &= ~optp->o_opt;
1781 	}
1782 
1783 	/*
1784 	 * Space already reserved for trailing \0
1785 	 */
1786 	buf[bwritten] = 0;
1787 	return (bwritten);
1788 }
1789 
1790 /*
1791  * Build hash lists of net addresses and hang them off the mount point.
1792  * Called by ufs_mount() to set up the lists of export addresses.
1793  */
1794 static int
1795 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1796 		const struct export_args *argp)
1797 {
1798 	struct netcred *np;
1799 	struct radix_node_head *rnh;
1800 	int i;
1801 	struct radix_node *rn;
1802 	struct sockaddr *saddr, *smask = 0;
1803 	struct domain *dom;
1804 	int error;
1805 
1806 	if (argp->ex_addrlen == 0) {
1807 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1808 			return (EPERM);
1809 		np = &nep->ne_defexported;
1810 		np->netc_exflags = argp->ex_flags;
1811 		np->netc_anon = argp->ex_anon;
1812 		np->netc_anon.cr_ref = 1;
1813 		mp->mnt_flag |= MNT_DEFEXPORTED;
1814 		return (0);
1815 	}
1816 
1817 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1818 		return (EINVAL);
1819 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1820 		return (EINVAL);
1821 
1822 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1823 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1824 	saddr = (struct sockaddr *) (np + 1);
1825 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1826 		goto out;
1827 	if (saddr->sa_len > argp->ex_addrlen)
1828 		saddr->sa_len = argp->ex_addrlen;
1829 	if (argp->ex_masklen) {
1830 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1831 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1832 		if (error)
1833 			goto out;
1834 		if (smask->sa_len > argp->ex_masklen)
1835 			smask->sa_len = argp->ex_masklen;
1836 	}
1837 	i = saddr->sa_family;
1838 	if ((rnh = nep->ne_rtable[i]) == 0) {
1839 		/*
1840 		 * Seems silly to initialize every AF when most are not used,
1841 		 * do so on demand here
1842 		 */
1843 		SLIST_FOREACH(dom, &domains, dom_next)
1844 			if (dom->dom_family == i && dom->dom_rtattach) {
1845 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1846 				    dom->dom_rtoffset);
1847 				break;
1848 			}
1849 		if ((rnh = nep->ne_rtable[i]) == 0) {
1850 			error = ENOBUFS;
1851 			goto out;
1852 		}
1853 	}
1854 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1855 	    np->netc_rnodes);
1856 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1857 		error = EPERM;
1858 		goto out;
1859 	}
1860 	np->netc_exflags = argp->ex_flags;
1861 	np->netc_anon = argp->ex_anon;
1862 	np->netc_anon.cr_ref = 1;
1863 	return (0);
1864 out:
1865 	kfree(np, M_NETADDR);
1866 	return (error);
1867 }
1868 
1869 /* ARGSUSED */
1870 static int
1871 vfs_free_netcred(struct radix_node *rn, void *w)
1872 {
1873 	struct radix_node_head *rnh = (struct radix_node_head *) w;
1874 
1875 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1876 	kfree((caddr_t) rn, M_NETADDR);
1877 	return (0);
1878 }
1879 
1880 /*
1881  * Free the net address hash lists that are hanging off the mount points.
1882  */
1883 static void
1884 vfs_free_addrlist(struct netexport *nep)
1885 {
1886 	int i;
1887 	struct radix_node_head *rnh;
1888 
1889 	for (i = 0; i <= AF_MAX; i++)
1890 		if ((rnh = nep->ne_rtable[i])) {
1891 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1892 			    (caddr_t) rnh);
1893 			kfree((caddr_t) rnh, M_RTABLE);
1894 			nep->ne_rtable[i] = 0;
1895 		}
1896 }
1897 
1898 int
1899 vfs_export(struct mount *mp, struct netexport *nep,
1900 	   const struct export_args *argp)
1901 {
1902 	int error;
1903 
1904 	if (argp->ex_flags & MNT_DELEXPORT) {
1905 		if (mp->mnt_flag & MNT_EXPUBLIC) {
1906 			vfs_setpublicfs(NULL, NULL, NULL);
1907 			mp->mnt_flag &= ~MNT_EXPUBLIC;
1908 		}
1909 		vfs_free_addrlist(nep);
1910 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1911 	}
1912 	if (argp->ex_flags & MNT_EXPORTED) {
1913 		if (argp->ex_flags & MNT_EXPUBLIC) {
1914 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1915 				return (error);
1916 			mp->mnt_flag |= MNT_EXPUBLIC;
1917 		}
1918 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1919 			return (error);
1920 		mp->mnt_flag |= MNT_EXPORTED;
1921 	}
1922 	return (0);
1923 }
1924 
1925 
1926 /*
1927  * Set the publicly exported filesystem (WebNFS). Currently, only
1928  * one public filesystem is possible in the spec (RFC 2054 and 2055)
1929  */
1930 int
1931 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1932 		const struct export_args *argp)
1933 {
1934 	int error;
1935 	struct vnode *rvp;
1936 	char *cp;
1937 
1938 	/*
1939 	 * mp == NULL -> invalidate the current info, the FS is
1940 	 * no longer exported. May be called from either vfs_export
1941 	 * or unmount, so check if it hasn't already been done.
1942 	 */
1943 	if (mp == NULL) {
1944 		if (nfs_pub.np_valid) {
1945 			nfs_pub.np_valid = 0;
1946 			if (nfs_pub.np_index != NULL) {
1947 				FREE(nfs_pub.np_index, M_TEMP);
1948 				nfs_pub.np_index = NULL;
1949 			}
1950 		}
1951 		return (0);
1952 	}
1953 
1954 	/*
1955 	 * Only one allowed at a time.
1956 	 */
1957 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1958 		return (EBUSY);
1959 
1960 	/*
1961 	 * Get real filehandle for root of exported FS.
1962 	 */
1963 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1964 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1965 
1966 	if ((error = VFS_ROOT(mp, &rvp)))
1967 		return (error);
1968 
1969 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1970 		return (error);
1971 
1972 	vput(rvp);
1973 
1974 	/*
1975 	 * If an indexfile was specified, pull it in.
1976 	 */
1977 	if (argp->ex_indexfile != NULL) {
1978 		int namelen;
1979 
1980 		error = vn_get_namelen(rvp, &namelen);
1981 		if (error)
1982 			return (error);
1983 		MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1984 		    M_WAITOK);
1985 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1986 		    namelen, NULL);
1987 		if (!error) {
1988 			/*
1989 			 * Check for illegal filenames.
1990 			 */
1991 			for (cp = nfs_pub.np_index; *cp; cp++) {
1992 				if (*cp == '/') {
1993 					error = EINVAL;
1994 					break;
1995 				}
1996 			}
1997 		}
1998 		if (error) {
1999 			FREE(nfs_pub.np_index, M_TEMP);
2000 			return (error);
2001 		}
2002 	}
2003 
2004 	nfs_pub.np_mount = mp;
2005 	nfs_pub.np_valid = 1;
2006 	return (0);
2007 }
2008 
2009 struct netcred *
2010 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2011 		struct sockaddr *nam)
2012 {
2013 	struct netcred *np;
2014 	struct radix_node_head *rnh;
2015 	struct sockaddr *saddr;
2016 
2017 	np = NULL;
2018 	if (mp->mnt_flag & MNT_EXPORTED) {
2019 		/*
2020 		 * Lookup in the export list first.
2021 		 */
2022 		if (nam != NULL) {
2023 			saddr = nam;
2024 			rnh = nep->ne_rtable[saddr->sa_family];
2025 			if (rnh != NULL) {
2026 				np = (struct netcred *)
2027 					(*rnh->rnh_matchaddr)((char *)saddr,
2028 							      rnh);
2029 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2030 					np = NULL;
2031 			}
2032 		}
2033 		/*
2034 		 * If no address match, use the default if it exists.
2035 		 */
2036 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2037 			np = &nep->ne_defexported;
2038 	}
2039 	return (np);
2040 }
2041 
2042 /*
2043  * perform msync on all vnodes under a mount point.  The mount point must
2044  * be locked.  This code is also responsible for lazy-freeing unreferenced
2045  * vnodes whos VM objects no longer contain pages.
2046  *
2047  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2048  *
2049  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2050  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
2051  * way up in this high level function.
2052  */
2053 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2054 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2055 
2056 void
2057 vfs_msync(struct mount *mp, int flags)
2058 {
2059 	int vmsc_flags;
2060 
2061 	/*
2062 	 * tmpfs sets this flag to prevent msync(), sync, and the
2063 	 * filesystem periodic syncer from trying to flush VM pages
2064 	 * to swap.  Only pure memory pressure flushes tmpfs VM pages
2065 	 * to swap.
2066 	 */
2067 	if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2068 		return;
2069 
2070 	/*
2071 	 * Ok, scan the vnodes for work.
2072 	 */
2073 	vmsc_flags = VMSC_GETVP;
2074 	if (flags != MNT_WAIT)
2075 		vmsc_flags |= VMSC_NOWAIT;
2076 	vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
2077 			(void *)(intptr_t)flags);
2078 }
2079 
2080 /*
2081  * scan1 is a fast pre-check.  There could be hundreds of thousands of
2082  * vnodes, we cannot afford to do anything heavy weight until we have a
2083  * fairly good indication that there is work to do.
2084  */
2085 static
2086 int
2087 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2088 {
2089 	int flags = (int)(intptr_t)data;
2090 
2091 	if ((vp->v_flag & VRECLAIMED) == 0) {
2092 		if (vshouldmsync(vp))
2093 			return(0);	/* call scan2 */
2094 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2095 		    (vp->v_flag & VOBJDIRTY) &&
2096 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2097 			return(0);	/* call scan2 */
2098 		}
2099 	}
2100 
2101 	/*
2102 	 * do not call scan2, continue the loop
2103 	 */
2104 	return(-1);
2105 }
2106 
2107 /*
2108  * This callback is handed a locked vnode.
2109  */
2110 static
2111 int
2112 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2113 {
2114 	vm_object_t obj;
2115 	int flags = (int)(intptr_t)data;
2116 
2117 	if (vp->v_flag & VRECLAIMED)
2118 		return(0);
2119 
2120 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2121 		if ((obj = vp->v_object) != NULL) {
2122 			vm_object_page_clean(obj, 0, 0,
2123 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2124 		}
2125 	}
2126 	return(0);
2127 }
2128 
2129 /*
2130  * Record a process's interest in events which might happen to
2131  * a vnode.  Because poll uses the historic select-style interface
2132  * internally, this routine serves as both the ``check for any
2133  * pending events'' and the ``record my interest in future events''
2134  * functions.  (These are done together, while the lock is held,
2135  * to avoid race conditions.)
2136  */
2137 int
2138 vn_pollrecord(struct vnode *vp, int events)
2139 {
2140 	lwkt_tokref vlock;
2141 
2142 	KKASSERT(curthread->td_proc != NULL);
2143 
2144 	lwkt_gettoken(&vlock, &vp->v_token);
2145 	if (vp->v_pollinfo.vpi_revents & events) {
2146 		/*
2147 		 * This leaves events we are not interested
2148 		 * in available for the other process which
2149 		 * which presumably had requested them
2150 		 * (otherwise they would never have been
2151 		 * recorded).
2152 		 */
2153 		events &= vp->v_pollinfo.vpi_revents;
2154 		vp->v_pollinfo.vpi_revents &= ~events;
2155 
2156 		lwkt_reltoken(&vlock);
2157 		return events;
2158 	}
2159 	vp->v_pollinfo.vpi_events |= events;
2160 	selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
2161 	lwkt_reltoken(&vlock);
2162 	return 0;
2163 }
2164 
2165 /*
2166  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2167  * it is possible for us to miss an event due to race conditions, but
2168  * that condition is expected to be rare, so for the moment it is the
2169  * preferred interface.
2170  */
2171 void
2172 vn_pollevent(struct vnode *vp, int events)
2173 {
2174 	lwkt_tokref vlock;
2175 
2176 	lwkt_gettoken(&vlock, &vp->v_token);
2177 	if (vp->v_pollinfo.vpi_events & events) {
2178 		/*
2179 		 * We clear vpi_events so that we don't
2180 		 * call selwakeup() twice if two events are
2181 		 * posted before the polling process(es) is
2182 		 * awakened.  This also ensures that we take at
2183 		 * most one selwakeup() if the polling process
2184 		 * is no longer interested.  However, it does
2185 		 * mean that only one event can be noticed at
2186 		 * a time.  (Perhaps we should only clear those
2187 		 * event bits which we note?) XXX
2188 		 */
2189 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2190 		vp->v_pollinfo.vpi_revents |= events;
2191 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2192 	}
2193 	lwkt_reltoken(&vlock);
2194 }
2195 
2196 /*
2197  * Wake up anyone polling on vp because it is being revoked.
2198  * This depends on dead_poll() returning POLLHUP for correct
2199  * behavior.
2200  */
2201 void
2202 vn_pollgone(struct vnode *vp)
2203 {
2204 	lwkt_tokref vlock;
2205 
2206 	lwkt_gettoken(&vlock, &vp->v_token);
2207 	if (vp->v_pollinfo.vpi_events) {
2208 		vp->v_pollinfo.vpi_events = 0;
2209 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2210 	}
2211 	lwkt_reltoken(&vlock);
2212 }
2213 
2214 /*
2215  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2216  * (or v_rdev might be NULL).
2217  */
2218 cdev_t
2219 vn_todev(struct vnode *vp)
2220 {
2221 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2222 		return (NULL);
2223 	KKASSERT(vp->v_rdev != NULL);
2224 	return (vp->v_rdev);
2225 }
2226 
2227 /*
2228  * Check if vnode represents a disk device.  The vnode does not need to be
2229  * opened.
2230  *
2231  * MPALMOSTSAFE
2232  */
2233 int
2234 vn_isdisk(struct vnode *vp, int *errp)
2235 {
2236 	cdev_t dev;
2237 
2238 	if (vp->v_type != VCHR) {
2239 		if (errp != NULL)
2240 			*errp = ENOTBLK;
2241 		return (0);
2242 	}
2243 
2244 	dev = vp->v_rdev;
2245 
2246 	if (dev == NULL) {
2247 		if (errp != NULL)
2248 			*errp = ENXIO;
2249 		return (0);
2250 	}
2251 	if (dev_is_good(dev) == 0) {
2252 		if (errp != NULL)
2253 			*errp = ENXIO;
2254 		return (0);
2255 	}
2256 	if ((dev_dflags(dev) & D_DISK) == 0) {
2257 		if (errp != NULL)
2258 			*errp = ENOTBLK;
2259 		return (0);
2260 	}
2261 	if (errp != NULL)
2262 		*errp = 0;
2263 	return (1);
2264 }
2265 
2266 int
2267 vn_get_namelen(struct vnode *vp, int *namelen)
2268 {
2269 	int error;
2270 	register_t retval[2];
2271 
2272 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2273 	if (error)
2274 		return (error);
2275 	*namelen = (int)retval[0];
2276 	return (0);
2277 }
2278 
2279 int
2280 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2281 		uint16_t d_namlen, const char *d_name)
2282 {
2283 	struct dirent *dp;
2284 	size_t len;
2285 
2286 	len = _DIRENT_RECLEN(d_namlen);
2287 	if (len > uio->uio_resid)
2288 		return(1);
2289 
2290 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2291 
2292 	dp->d_ino = d_ino;
2293 	dp->d_namlen = d_namlen;
2294 	dp->d_type = d_type;
2295 	bcopy(d_name, dp->d_name, d_namlen);
2296 
2297 	*error = uiomove((caddr_t)dp, len, uio);
2298 
2299 	kfree(dp, M_TEMP);
2300 
2301 	return(0);
2302 }
2303 
2304 void
2305 vn_mark_atime(struct vnode *vp, struct thread *td)
2306 {
2307 	struct proc *p = td->td_proc;
2308 	struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2309 
2310 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2311 		VOP_MARKATIME(vp, cred);
2312 	}
2313 }
2314