xref: /dflybsd-src/sys/kern/vfs_subr.c (revision e7302aa08274de307cd2c3345fc64c56dbe56e21)
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 
96 enum vtype iftovt_tab[16] = {
97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 };
100 int vttoif_tab[9] = {
101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 	S_IFSOCK, S_IFIFO, S_IFMT,
103 };
104 
105 static int reassignbufcalls;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
107 		&reassignbufcalls, 0, "");
108 static int reassignbufloops;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
110 		&reassignbufloops, 0, "");
111 static int reassignbufsortgood;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
113 		&reassignbufsortgood, 0, "");
114 static int reassignbufsortbad;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
116 		&reassignbufsortbad, 0, "");
117 static int reassignbufmethod = 1;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
119 		&reassignbufmethod, 0, "");
120 
121 int	nfs_mount_type = -1;
122 static struct lwkt_token spechash_token;
123 struct nfs_public nfs_pub;	/* publicly exported FS */
124 
125 int desiredvnodes;
126 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
127 		&desiredvnodes, 0, "Maximum number of vnodes");
128 
129 static void	vfs_free_addrlist (struct netexport *nep);
130 static int	vfs_free_netcred (struct radix_node *rn, void *w);
131 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
132 				       const struct export_args *argp);
133 
134 /*
135  * Red black tree functions
136  */
137 static int rb_buf_compare(struct buf *b1, struct buf *b2);
138 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
139 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
140 
141 static int
142 rb_buf_compare(struct buf *b1, struct buf *b2)
143 {
144 	if (b1->b_loffset < b2->b_loffset)
145 		return(-1);
146 	if (b1->b_loffset > b2->b_loffset)
147 		return(1);
148 	return(0);
149 }
150 
151 /*
152  * Returns non-zero if the vnode is a candidate for lazy msyncing.
153  */
154 static __inline int
155 vshouldmsync(struct vnode *vp)
156 {
157 	if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
158 		return (0);		/* other holders */
159 	if (vp->v_object &&
160 	    (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
161 		return (0);
162 	}
163 	return (1);
164 }
165 
166 /*
167  * Initialize the vnode management data structures.
168  *
169  * Called from vfsinit()
170  */
171 void
172 vfs_subr_init(void)
173 {
174 	int factor1;
175 	int factor2;
176 
177 	/*
178 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
179 	 * according to available system memory but we may also have
180 	 * to limit it based on available KVM, which is capped on 32 bit
181 	 * systems.
182 	 *
183 	 * WARNING!  For machines with 64-256M of ram we have to be sure
184 	 *	     that the default limit scales down well due to HAMMER
185 	 *	     taking up significantly more memory per-vnode vs UFS.
186 	 *	     We want around ~5800 on a 128M machine.
187 	 */
188 	factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
189 	factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode));
190 	desiredvnodes =
191 		imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
192 		     KvaSize / factor2);
193 	desiredvnodes = imax(desiredvnodes, maxproc * 8);
194 
195 	lwkt_token_init(&spechash_token);
196 }
197 
198 /*
199  * Knob to control the precision of file timestamps:
200  *
201  *   0 = seconds only; nanoseconds zeroed.
202  *   1 = seconds and nanoseconds, accurate within 1/HZ.
203  *   2 = seconds and nanoseconds, truncated to microseconds.
204  * >=3 = seconds and nanoseconds, maximum precision.
205  */
206 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
207 
208 static int timestamp_precision = TSP_SEC;
209 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
210 		&timestamp_precision, 0, "");
211 
212 /*
213  * Get a current timestamp.
214  *
215  * MPSAFE
216  */
217 void
218 vfs_timestamp(struct timespec *tsp)
219 {
220 	struct timeval tv;
221 
222 	switch (timestamp_precision) {
223 	case TSP_SEC:
224 		tsp->tv_sec = time_second;
225 		tsp->tv_nsec = 0;
226 		break;
227 	case TSP_HZ:
228 		getnanotime(tsp);
229 		break;
230 	case TSP_USEC:
231 		microtime(&tv);
232 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
233 		break;
234 	case TSP_NSEC:
235 	default:
236 		nanotime(tsp);
237 		break;
238 	}
239 }
240 
241 /*
242  * Set vnode attributes to VNOVAL
243  */
244 void
245 vattr_null(struct vattr *vap)
246 {
247 	vap->va_type = VNON;
248 	vap->va_size = VNOVAL;
249 	vap->va_bytes = VNOVAL;
250 	vap->va_mode = VNOVAL;
251 	vap->va_nlink = VNOVAL;
252 	vap->va_uid = VNOVAL;
253 	vap->va_gid = VNOVAL;
254 	vap->va_fsid = VNOVAL;
255 	vap->va_fileid = VNOVAL;
256 	vap->va_blocksize = VNOVAL;
257 	vap->va_rmajor = VNOVAL;
258 	vap->va_rminor = VNOVAL;
259 	vap->va_atime.tv_sec = VNOVAL;
260 	vap->va_atime.tv_nsec = VNOVAL;
261 	vap->va_mtime.tv_sec = VNOVAL;
262 	vap->va_mtime.tv_nsec = VNOVAL;
263 	vap->va_ctime.tv_sec = VNOVAL;
264 	vap->va_ctime.tv_nsec = VNOVAL;
265 	vap->va_flags = VNOVAL;
266 	vap->va_gen = VNOVAL;
267 	vap->va_vaflags = 0;
268 	/* va_*_uuid fields are only valid if related flags are set */
269 }
270 
271 /*
272  * Flush out and invalidate all buffers associated with a vnode.
273  *
274  * vp must be locked.
275  */
276 static int vinvalbuf_bp(struct buf *bp, void *data);
277 
278 struct vinvalbuf_bp_info {
279 	struct vnode *vp;
280 	int slptimeo;
281 	int lkflags;
282 	int flags;
283 };
284 
285 int
286 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
287 {
288 	struct vinvalbuf_bp_info info;
289 	vm_object_t object;
290 	lwkt_tokref vlock;
291 	int error;
292 
293 	lwkt_gettoken(&vlock, &vp->v_token);
294 
295 	/*
296 	 * If we are being asked to save, call fsync to ensure that the inode
297 	 * is updated.
298 	 */
299 	if (flags & V_SAVE) {
300 		error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
301 		if (error)
302 			goto done;
303 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
304 			if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
305 				goto done;
306 
307 			/*
308 			 * Dirty bufs may be left or generated via races
309 			 * in circumstances where vinvalbuf() is called on
310 			 * a vnode not undergoing reclamation.   Only
311 			 * panic if we are trying to reclaim the vnode.
312 			 */
313 			if ((vp->v_flag & VRECLAIMED) &&
314 			    (bio_track_active(&vp->v_track_write) ||
315 			    !RB_EMPTY(&vp->v_rbdirty_tree))) {
316 				panic("vinvalbuf: dirty bufs");
317 			}
318 		}
319   	}
320 	info.slptimeo = slptimeo;
321 	info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
322 	if (slpflag & PCATCH)
323 		info.lkflags |= LK_PCATCH;
324 	info.flags = flags;
325 	info.vp = vp;
326 
327 	/*
328 	 * Flush the buffer cache until nothing is left.
329 	 */
330 	while (!RB_EMPTY(&vp->v_rbclean_tree) ||
331 	       !RB_EMPTY(&vp->v_rbdirty_tree)) {
332 		error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
333 				vinvalbuf_bp, &info);
334 		if (error == 0) {
335 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
336 					vinvalbuf_bp, &info);
337 		}
338 	}
339 
340 	/*
341 	 * Wait for I/O completion.  We may block in the pip code so we have
342 	 * to re-check.
343 	 */
344 	do {
345 		bio_track_wait(&vp->v_track_write, 0, 0);
346 		if ((object = vp->v_object) != NULL) {
347 			while (object->paging_in_progress)
348 				vm_object_pip_sleep(object, "vnvlbx");
349 		}
350 	} while (bio_track_active(&vp->v_track_write));
351 
352 	/*
353 	 * Destroy the copy in the VM cache, too.
354 	 */
355 	if ((object = vp->v_object) != NULL) {
356 		vm_object_page_remove(object, 0, 0,
357 			(flags & V_SAVE) ? TRUE : FALSE);
358 	}
359 
360 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
361 		panic("vinvalbuf: flush failed");
362 	if (!RB_EMPTY(&vp->v_rbhash_tree))
363 		panic("vinvalbuf: flush failed, buffers still present");
364 	error = 0;
365 done:
366 	lwkt_reltoken(&vlock);
367 	return (error);
368 }
369 
370 static int
371 vinvalbuf_bp(struct buf *bp, void *data)
372 {
373 	struct vinvalbuf_bp_info *info = data;
374 	int error;
375 
376 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
377 		error = BUF_TIMELOCK(bp, info->lkflags,
378 				     "vinvalbuf", info->slptimeo);
379 		if (error == 0) {
380 			BUF_UNLOCK(bp);
381 			error = ENOLCK;
382 		}
383 		if (error == ENOLCK)
384 			return(0);
385 		return (-error);
386 	}
387 
388 	KKASSERT(bp->b_vp == info->vp);
389 
390 	/*
391 	 * XXX Since there are no node locks for NFS, I
392 	 * believe there is a slight chance that a delayed
393 	 * write will occur while sleeping just above, so
394 	 * check for it.  Note that vfs_bio_awrite expects
395 	 * buffers to reside on a queue, while bwrite() and
396 	 * brelse() do not.
397 	 *
398 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
399 	 * check.  This code will write out the buffer, period.
400 	 */
401 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
402 	    (info->flags & V_SAVE)) {
403 		if (bp->b_vp == info->vp) {
404 			if (bp->b_flags & B_CLUSTEROK) {
405 				vfs_bio_awrite(bp);
406 			} else {
407 				bremfree(bp);
408 				bawrite(bp);
409 			}
410 		} else {
411 			bremfree(bp);
412 			bwrite(bp);
413 		}
414 	} else if (info->flags & V_SAVE) {
415 		/*
416 		 * Cannot set B_NOCACHE on a clean buffer as this will
417 		 * destroy the VM backing store which might actually
418 		 * be dirty (and unsynchronized).
419 		 */
420 		bremfree(bp);
421 		bp->b_flags |= (B_INVAL | B_RELBUF);
422 		brelse(bp);
423 	} else {
424 		bremfree(bp);
425 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
426 		brelse(bp);
427 	}
428 	return(0);
429 }
430 
431 /*
432  * Truncate a file's buffer and pages to a specified length.  This
433  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
434  * sync activity.
435  *
436  * The vnode must be locked.
437  */
438 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
439 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
440 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
441 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
442 
443 int
444 vtruncbuf(struct vnode *vp, off_t length, int blksize)
445 {
446 	off_t truncloffset;
447 	const char *filename;
448 	lwkt_tokref vlock;
449 	int count;
450 
451 	/*
452 	 * Round up to the *next* block, then destroy the buffers in question.
453 	 * Since we are only removing some of the buffers we must rely on the
454 	 * scan count to determine whether a loop is necessary.
455 	 */
456 	if ((count = (int)(length % blksize)) != 0)
457 		truncloffset = length + (blksize - count);
458 	else
459 		truncloffset = length;
460 
461 	lwkt_gettoken(&vlock, &vp->v_token);
462 	do {
463 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
464 				vtruncbuf_bp_trunc_cmp,
465 				vtruncbuf_bp_trunc, &truncloffset);
466 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
467 				vtruncbuf_bp_trunc_cmp,
468 				vtruncbuf_bp_trunc, &truncloffset);
469 	} while(count);
470 
471 	/*
472 	 * For safety, fsync any remaining metadata if the file is not being
473 	 * truncated to 0.  Since the metadata does not represent the entire
474 	 * dirty list we have to rely on the hit count to ensure that we get
475 	 * all of it.
476 	 */
477 	if (length > 0) {
478 		do {
479 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
480 					vtruncbuf_bp_metasync_cmp,
481 					vtruncbuf_bp_metasync, vp);
482 		} while (count);
483 	}
484 
485 	/*
486 	 * Clean out any left over VM backing store.
487 	 *
488 	 * It is possible to have in-progress I/O from buffers that were
489 	 * not part of the truncation.  This should not happen if we
490 	 * are truncating to 0-length.
491 	 */
492 	vnode_pager_setsize(vp, length);
493 	bio_track_wait(&vp->v_track_write, 0, 0);
494 
495 	/*
496 	 * Debugging only
497 	 */
498 	spin_lock_wr(&vp->v_spinlock);
499 	filename = TAILQ_FIRST(&vp->v_namecache) ?
500 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
501 	spin_unlock_wr(&vp->v_spinlock);
502 
503 	/*
504 	 * Make sure no buffers were instantiated while we were trying
505 	 * to clean out the remaining VM pages.  This could occur due
506 	 * to busy dirty VM pages being flushed out to disk.
507 	 */
508 	do {
509 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
510 				vtruncbuf_bp_trunc_cmp,
511 				vtruncbuf_bp_trunc, &truncloffset);
512 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
513 				vtruncbuf_bp_trunc_cmp,
514 				vtruncbuf_bp_trunc, &truncloffset);
515 		if (count) {
516 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
517 			       "left over buffers in %s\n", count, filename);
518 		}
519 	} while(count);
520 
521 	lwkt_reltoken(&vlock);
522 
523 	return (0);
524 }
525 
526 /*
527  * The callback buffer is beyond the new file EOF and must be destroyed.
528  * Note that the compare function must conform to the RB_SCAN's requirements.
529  */
530 static
531 int
532 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
533 {
534 	if (bp->b_loffset >= *(off_t *)data)
535 		return(0);
536 	return(-1);
537 }
538 
539 static
540 int
541 vtruncbuf_bp_trunc(struct buf *bp, void *data)
542 {
543 	/*
544 	 * Do not try to use a buffer we cannot immediately lock, but sleep
545 	 * anyway to prevent a livelock.  The code will loop until all buffers
546 	 * can be acted upon.
547 	 */
548 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
549 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
550 			BUF_UNLOCK(bp);
551 	} else {
552 		bremfree(bp);
553 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
554 		brelse(bp);
555 	}
556 	return(1);
557 }
558 
559 /*
560  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
561  * blocks (with a negative loffset) are scanned.
562  * Note that the compare function must conform to the RB_SCAN's requirements.
563  */
564 static int
565 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
566 {
567 	if (bp->b_loffset < 0)
568 		return(0);
569 	return(1);
570 }
571 
572 static int
573 vtruncbuf_bp_metasync(struct buf *bp, void *data)
574 {
575 	struct vnode *vp = data;
576 
577 	if (bp->b_flags & B_DELWRI) {
578 		/*
579 		 * Do not try to use a buffer we cannot immediately lock,
580 		 * but sleep anyway to prevent a livelock.  The code will
581 		 * loop until all buffers can be acted upon.
582 		 */
583 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
584 			if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
585 				BUF_UNLOCK(bp);
586 		} else {
587 			bremfree(bp);
588 			if (bp->b_vp == vp)
589 				bawrite(bp);
590 			else
591 				bwrite(bp);
592 		}
593 		return(1);
594 	} else {
595 		return(0);
596 	}
597 }
598 
599 /*
600  * vfsync - implements a multipass fsync on a file which understands
601  * dependancies and meta-data.  The passed vnode must be locked.  The
602  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
603  *
604  * When fsyncing data asynchronously just do one consolidated pass starting
605  * with the most negative block number.  This may not get all the data due
606  * to dependancies.
607  *
608  * When fsyncing data synchronously do a data pass, then a metadata pass,
609  * then do additional data+metadata passes to try to get all the data out.
610  */
611 static int vfsync_wait_output(struct vnode *vp,
612 			    int (*waitoutput)(struct vnode *, struct thread *));
613 static int vfsync_data_only_cmp(struct buf *bp, void *data);
614 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
615 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
616 static int vfsync_bp(struct buf *bp, void *data);
617 
618 struct vfsync_info {
619 	struct vnode *vp;
620 	int synchronous;
621 	int syncdeps;
622 	int lazycount;
623 	int lazylimit;
624 	int skippedbufs;
625 	int (*checkdef)(struct buf *);
626 };
627 
628 int
629 vfsync(struct vnode *vp, int waitfor, int passes,
630 	int (*checkdef)(struct buf *),
631 	int (*waitoutput)(struct vnode *, struct thread *))
632 {
633 	struct vfsync_info info;
634 	lwkt_tokref vlock;
635 	int error;
636 
637 	bzero(&info, sizeof(info));
638 	info.vp = vp;
639 	if ((info.checkdef = checkdef) == NULL)
640 		info.syncdeps = 1;
641 
642 	lwkt_gettoken(&vlock, &vp->v_token);
643 
644 	switch(waitfor) {
645 	case MNT_LAZY:
646 		/*
647 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
648 		 * number of data (not meta) pages we try to flush to 1MB.
649 		 * A non-zero return means that lazy limit was reached.
650 		 */
651 		info.lazylimit = 1024 * 1024;
652 		info.syncdeps = 1;
653 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
654 				vfsync_lazy_range_cmp, vfsync_bp, &info);
655 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
656 				vfsync_meta_only_cmp, vfsync_bp, &info);
657 		if (error == 0)
658 			vp->v_lazyw = 0;
659 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
660 			vn_syncer_add_to_worklist(vp, 1);
661 		error = 0;
662 		break;
663 	case MNT_NOWAIT:
664 		/*
665 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
666 		 */
667 		info.syncdeps = 1;
668 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
669 			vfsync_bp, &info);
670 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
671 			vfsync_bp, &info);
672 		error = 0;
673 		break;
674 	default:
675 		/*
676 		 * Synchronous.  Do a data-only pass, then a meta-data+data
677 		 * pass, then additional integrated passes to try to get
678 		 * all the dependancies flushed.
679 		 */
680 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
681 			vfsync_bp, &info);
682 		error = vfsync_wait_output(vp, waitoutput);
683 		if (error == 0) {
684 			info.skippedbufs = 0;
685 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
686 				vfsync_bp, &info);
687 			error = vfsync_wait_output(vp, waitoutput);
688 			if (info.skippedbufs)
689 				kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
690 		}
691 		while (error == 0 && passes > 0 &&
692 		       !RB_EMPTY(&vp->v_rbdirty_tree)
693 		) {
694 			if (--passes == 0) {
695 				info.synchronous = 1;
696 				info.syncdeps = 1;
697 			}
698 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
699 				vfsync_bp, &info);
700 			if (error < 0)
701 				error = -error;
702 			info.syncdeps = 1;
703 			if (error == 0)
704 				error = vfsync_wait_output(vp, waitoutput);
705 		}
706 		break;
707 	}
708 	lwkt_reltoken(&vlock);
709 	return(error);
710 }
711 
712 static int
713 vfsync_wait_output(struct vnode *vp,
714 		   int (*waitoutput)(struct vnode *, struct thread *))
715 {
716 	int error;
717 
718 	error = bio_track_wait(&vp->v_track_write, 0, 0);
719 	if (waitoutput)
720 		error = waitoutput(vp, curthread);
721 	return(error);
722 }
723 
724 static int
725 vfsync_data_only_cmp(struct buf *bp, void *data)
726 {
727 	if (bp->b_loffset < 0)
728 		return(-1);
729 	return(0);
730 }
731 
732 static int
733 vfsync_meta_only_cmp(struct buf *bp, void *data)
734 {
735 	if (bp->b_loffset < 0)
736 		return(0);
737 	return(1);
738 }
739 
740 static int
741 vfsync_lazy_range_cmp(struct buf *bp, void *data)
742 {
743 	struct vfsync_info *info = data;
744 	if (bp->b_loffset < info->vp->v_lazyw)
745 		return(-1);
746 	return(0);
747 }
748 
749 static int
750 vfsync_bp(struct buf *bp, void *data)
751 {
752 	struct vfsync_info *info = data;
753 	struct vnode *vp = info->vp;
754 	int error;
755 
756 	/*
757 	 * if syncdeps is not set we do not try to write buffers which have
758 	 * dependancies.
759 	 */
760 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
761 		return(0);
762 
763 	/*
764 	 * Ignore buffers that we cannot immediately lock.  XXX
765 	 */
766 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
767 		kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
768 		++info->skippedbufs;
769 		return(0);
770 	}
771 	if ((bp->b_flags & B_DELWRI) == 0)
772 		panic("vfsync_bp: buffer not dirty");
773 	if (vp != bp->b_vp)
774 		panic("vfsync_bp: buffer vp mismatch");
775 
776 	/*
777 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
778 	 * has been written but an additional handshake with the device
779 	 * is required before we can dispose of the buffer.  We have no idea
780 	 * how to do this so we have to skip these buffers.
781 	 */
782 	if (bp->b_flags & B_NEEDCOMMIT) {
783 		BUF_UNLOCK(bp);
784 		return(0);
785 	}
786 
787 	/*
788 	 * Ask bioops if it is ok to sync
789 	 */
790 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
791 		bremfree(bp);
792 		brelse(bp);
793 		return(0);
794 	}
795 
796 	if (info->synchronous) {
797 		/*
798 		 * Synchronous flushing.  An error may be returned.
799 		 */
800 		bremfree(bp);
801 		error = bwrite(bp);
802 	} else {
803 		/*
804 		 * Asynchronous flushing.  A negative return value simply
805 		 * stops the scan and is not considered an error.  We use
806 		 * this to support limited MNT_LAZY flushes.
807 		 */
808 		vp->v_lazyw = bp->b_loffset;
809 		if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
810 			info->lazycount += vfs_bio_awrite(bp);
811 		} else {
812 			info->lazycount += bp->b_bufsize;
813 			bremfree(bp);
814 			bawrite(bp);
815 		}
816 		if (info->lazylimit && info->lazycount >= info->lazylimit)
817 			error = 1;
818 		else
819 			error = 0;
820 	}
821 	return(-error);
822 }
823 
824 /*
825  * Associate a buffer with a vnode.
826  *
827  * MPSAFE
828  */
829 int
830 bgetvp(struct vnode *vp, struct buf *bp)
831 {
832 	lwkt_tokref vlock;
833 
834 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
835 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
836 
837 	/*
838 	 * Insert onto list for new vnode.
839 	 */
840 	lwkt_gettoken(&vlock, &vp->v_token);
841 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
842 		lwkt_reltoken(&vlock);
843 		return (EEXIST);
844 	}
845 	bp->b_vp = vp;
846 	bp->b_flags |= B_HASHED;
847 	bp->b_flags |= B_VNCLEAN;
848 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
849 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
850 	vhold(vp);
851 	lwkt_reltoken(&vlock);
852 	return(0);
853 }
854 
855 /*
856  * Disassociate a buffer from a vnode.
857  */
858 void
859 brelvp(struct buf *bp)
860 {
861 	struct vnode *vp;
862 	lwkt_tokref vlock;
863 
864 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
865 
866 	/*
867 	 * Delete from old vnode list, if on one.
868 	 */
869 	vp = bp->b_vp;
870 	lwkt_gettoken(&vlock, &vp->v_token);
871 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
872 		if (bp->b_flags & B_VNDIRTY)
873 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
874 		else
875 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
876 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
877 	}
878 	if (bp->b_flags & B_HASHED) {
879 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
880 		bp->b_flags &= ~B_HASHED;
881 	}
882 	if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
883 		vclrflags(vp, VONWORKLST);
884 		LIST_REMOVE(vp, v_synclist);
885 	}
886 	bp->b_vp = NULL;
887 	lwkt_reltoken(&vlock);
888 
889 	vdrop(vp);
890 }
891 
892 /*
893  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
894  * This routine is called when the state of the B_DELWRI bit is changed.
895  *
896  * MPSAFE
897  */
898 void
899 reassignbuf(struct buf *bp)
900 {
901 	struct vnode *vp = bp->b_vp;
902 	lwkt_tokref vlock;
903 	int delay;
904 
905 	KKASSERT(vp != NULL);
906 	++reassignbufcalls;
907 
908 	/*
909 	 * B_PAGING flagged buffers cannot be reassigned because their vp
910 	 * is not fully linked in.
911 	 */
912 	if (bp->b_flags & B_PAGING)
913 		panic("cannot reassign paging buffer");
914 
915 	lwkt_gettoken(&vlock, &vp->v_token);
916 	if (bp->b_flags & B_DELWRI) {
917 		/*
918 		 * Move to the dirty list, add the vnode to the worklist
919 		 */
920 		if (bp->b_flags & B_VNCLEAN) {
921 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
922 			bp->b_flags &= ~B_VNCLEAN;
923 		}
924 		if ((bp->b_flags & B_VNDIRTY) == 0) {
925 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
926 				panic("reassignbuf: dup lblk vp %p bp %p",
927 				      vp, bp);
928 			}
929 			bp->b_flags |= B_VNDIRTY;
930 		}
931 		if ((vp->v_flag & VONWORKLST) == 0) {
932 			switch (vp->v_type) {
933 			case VDIR:
934 				delay = dirdelay;
935 				break;
936 			case VCHR:
937 			case VBLK:
938 				if (vp->v_rdev &&
939 				    vp->v_rdev->si_mountpoint != NULL) {
940 					delay = metadelay;
941 					break;
942 				}
943 				/* fall through */
944 			default:
945 				delay = filedelay;
946 			}
947 			vn_syncer_add_to_worklist(vp, delay);
948 		}
949 	} else {
950 		/*
951 		 * Move to the clean list, remove the vnode from the worklist
952 		 * if no dirty blocks remain.
953 		 */
954 		if (bp->b_flags & B_VNDIRTY) {
955 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
956 			bp->b_flags &= ~B_VNDIRTY;
957 		}
958 		if ((bp->b_flags & B_VNCLEAN) == 0) {
959 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
960 				panic("reassignbuf: dup lblk vp %p bp %p",
961 				      vp, bp);
962 			}
963 			bp->b_flags |= B_VNCLEAN;
964 		}
965 		if ((vp->v_flag & VONWORKLST) &&
966 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
967 			vclrflags(vp, VONWORKLST);
968 			LIST_REMOVE(vp, v_synclist);
969 		}
970 	}
971 	lwkt_reltoken(&vlock);
972 }
973 
974 /*
975  * Create a vnode for a block device.
976  * Used for mounting the root file system.
977  */
978 extern struct vop_ops *devfs_vnode_dev_vops_p;
979 int
980 bdevvp(cdev_t dev, struct vnode **vpp)
981 {
982 	struct vnode *vp;
983 	struct vnode *nvp;
984 	int error;
985 
986 	if (dev == NULL) {
987 		*vpp = NULLVP;
988 		return (ENXIO);
989 	}
990 	error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
991 				&nvp, 0, 0);
992 	if (error) {
993 		*vpp = NULLVP;
994 		return (error);
995 	}
996 	vp = nvp;
997 	vp->v_type = VCHR;
998 #if 0
999 	vp->v_rdev = dev;
1000 #endif
1001 	v_associate_rdev(vp, dev);
1002 	vp->v_umajor = dev->si_umajor;
1003 	vp->v_uminor = dev->si_uminor;
1004 	vx_unlock(vp);
1005 	*vpp = vp;
1006 	return (0);
1007 }
1008 
1009 int
1010 v_associate_rdev(struct vnode *vp, cdev_t dev)
1011 {
1012 	lwkt_tokref ilock;
1013 
1014 	if (dev == NULL)
1015 		return(ENXIO);
1016 	if (dev_is_good(dev) == 0)
1017 		return(ENXIO);
1018 	KKASSERT(vp->v_rdev == NULL);
1019 	vp->v_rdev = reference_dev(dev);
1020 	lwkt_gettoken(&ilock, &spechash_token);
1021 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1022 	lwkt_reltoken(&ilock);
1023 	return(0);
1024 }
1025 
1026 void
1027 v_release_rdev(struct vnode *vp)
1028 {
1029 	lwkt_tokref ilock;
1030 	cdev_t dev;
1031 
1032 	if ((dev = vp->v_rdev) != NULL) {
1033 		lwkt_gettoken(&ilock, &spechash_token);
1034 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1035 		vp->v_rdev = NULL;
1036 		release_dev(dev);
1037 		lwkt_reltoken(&ilock);
1038 	}
1039 }
1040 
1041 /*
1042  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1043  * the device number with the vnode.  The actual device is not associated
1044  * until the vnode is opened (usually in spec_open()), and will be
1045  * disassociated on last close.
1046  */
1047 void
1048 addaliasu(struct vnode *nvp, int x, int y)
1049 {
1050 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1051 		panic("addaliasu on non-special vnode");
1052 	nvp->v_umajor = x;
1053 	nvp->v_uminor = y;
1054 }
1055 
1056 /*
1057  * Simple call that a filesystem can make to try to get rid of a
1058  * vnode.  It will fail if anyone is referencing the vnode (including
1059  * the caller).
1060  *
1061  * The filesystem can check whether its in-memory inode structure still
1062  * references the vp on return.
1063  */
1064 void
1065 vclean_unlocked(struct vnode *vp)
1066 {
1067 	vx_get(vp);
1068 	if (sysref_isactive(&vp->v_sysref) == 0)
1069 		vgone_vxlocked(vp);
1070 	vx_put(vp);
1071 }
1072 
1073 /*
1074  * Disassociate a vnode from its underlying filesystem.
1075  *
1076  * The vnode must be VX locked and referenced.  In all normal situations
1077  * there are no active references.  If vclean_vxlocked() is called while
1078  * there are active references, the vnode is being ripped out and we have
1079  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1080  */
1081 void
1082 vclean_vxlocked(struct vnode *vp, int flags)
1083 {
1084 	int active;
1085 	int n;
1086 	vm_object_t object;
1087 
1088 	/*
1089 	 * If the vnode has already been reclaimed we have nothing to do.
1090 	 */
1091 	if (vp->v_flag & VRECLAIMED)
1092 		return;
1093 	vsetflags(vp, VRECLAIMED);
1094 
1095 	/*
1096 	 * Scrap the vfs cache
1097 	 */
1098 	while (cache_inval_vp(vp, 0) != 0) {
1099 		kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1100 		tsleep(vp, 0, "vclninv", 2);
1101 	}
1102 
1103 	/*
1104 	 * Check to see if the vnode is in use. If so we have to reference it
1105 	 * before we clean it out so that its count cannot fall to zero and
1106 	 * generate a race against ourselves to recycle it.
1107 	 */
1108 	active = sysref_isactive(&vp->v_sysref);
1109 
1110 	/*
1111 	 * Clean out any buffers associated with the vnode and destroy its
1112 	 * object, if it has one.
1113 	 */
1114 	vinvalbuf(vp, V_SAVE, 0, 0);
1115 
1116 	/*
1117 	 * If purging an active vnode (typically during a forced unmount
1118 	 * or reboot), it must be closed and deactivated before being
1119 	 * reclaimed.  This isn't really all that safe, but what can
1120 	 * we do? XXX.
1121 	 *
1122 	 * Note that neither of these routines unlocks the vnode.
1123 	 */
1124 	if (active && (flags & DOCLOSE)) {
1125 		while ((n = vp->v_opencount) != 0) {
1126 			if (vp->v_writecount)
1127 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1128 			else
1129 				VOP_CLOSE(vp, FNONBLOCK);
1130 			if (vp->v_opencount == n) {
1131 				kprintf("Warning: unable to force-close"
1132 				       " vnode %p\n", vp);
1133 				break;
1134 			}
1135 		}
1136 	}
1137 
1138 	/*
1139 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1140 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1141 	 * again to make sure they all get flushed.
1142 	 *
1143 	 * This can occur if a file with a link count of 0 needs to be
1144 	 * truncated.
1145 	 *
1146 	 * If the vnode is already dead don't try to deactivate it.
1147 	 */
1148 	if ((vp->v_flag & VINACTIVE) == 0) {
1149 		vsetflags(vp, VINACTIVE);
1150 		if (vp->v_mount)
1151 			VOP_INACTIVE(vp);
1152 		vinvalbuf(vp, V_SAVE, 0, 0);
1153 	}
1154 
1155 	/*
1156 	 * If the vnode has an object, destroy it.
1157 	 */
1158 	if ((object = vp->v_object) != NULL) {
1159 		if (object->ref_count == 0) {
1160 			if ((object->flags & OBJ_DEAD) == 0)
1161 				vm_object_terminate(object);
1162 		} else {
1163 			vm_pager_deallocate(object);
1164 		}
1165 		vclrflags(vp, VOBJBUF);
1166 	}
1167 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1168 
1169 	/*
1170 	 * Reclaim the vnode if not already dead.
1171 	 */
1172 	if (vp->v_mount && VOP_RECLAIM(vp))
1173 		panic("vclean: cannot reclaim");
1174 
1175 	/*
1176 	 * Done with purge, notify sleepers of the grim news.
1177 	 */
1178 	vp->v_ops = &dead_vnode_vops_p;
1179 	vn_pollgone(vp);
1180 	vp->v_tag = VT_NON;
1181 
1182 	/*
1183 	 * If we are destroying an active vnode, reactivate it now that
1184 	 * we have reassociated it with deadfs.  This prevents the system
1185 	 * from crashing on the vnode due to it being unexpectedly marked
1186 	 * as inactive or reclaimed.
1187 	 */
1188 	if (active && (flags & DOCLOSE)) {
1189 		vclrflags(vp, VINACTIVE | VRECLAIMED);
1190 	}
1191 }
1192 
1193 /*
1194  * Eliminate all activity associated with the requested vnode
1195  * and with all vnodes aliased to the requested vnode.
1196  *
1197  * The vnode must be referenced but should not be locked.
1198  */
1199 int
1200 vrevoke(struct vnode *vp, struct ucred *cred)
1201 {
1202 	struct vnode *vq;
1203 	struct vnode *vqn;
1204 	lwkt_tokref ilock;
1205 	cdev_t dev;
1206 	int error;
1207 
1208 	/*
1209 	 * If the vnode has a device association, scrap all vnodes associated
1210 	 * with the device.  Don't let the device disappear on us while we
1211 	 * are scrapping the vnodes.
1212 	 *
1213 	 * The passed vp will probably show up in the list, do not VX lock
1214 	 * it twice!
1215 	 *
1216 	 * Releasing the vnode's rdev here can mess up specfs's call to
1217 	 * device close, so don't do it.  The vnode has been disassociated
1218 	 * and the device will be closed after the last ref on the related
1219 	 * fp goes away (if not still open by e.g. the kernel).
1220 	 */
1221 	if (vp->v_type != VCHR) {
1222 		error = fdrevoke(vp, DTYPE_VNODE, cred);
1223 		return (error);
1224 	}
1225 	if ((dev = vp->v_rdev) == NULL) {
1226 		return(0);
1227 	}
1228 	reference_dev(dev);
1229 	lwkt_gettoken(&ilock, &spechash_token);
1230 
1231 	vqn = SLIST_FIRST(&dev->si_hlist);
1232 	if (vqn)
1233 		vref(vqn);
1234 	while ((vq = vqn) != NULL) {
1235 		vqn = SLIST_NEXT(vqn, v_cdevnext);
1236 		if (vqn)
1237 			vref(vqn);
1238 		fdrevoke(vq, DTYPE_VNODE, cred);
1239 		/*v_release_rdev(vq);*/
1240 		vrele(vq);
1241 	}
1242 	lwkt_reltoken(&ilock);
1243 	dev_drevoke(dev);
1244 	release_dev(dev);
1245 	return (0);
1246 }
1247 
1248 /*
1249  * This is called when the object underlying a vnode is being destroyed,
1250  * such as in a remove().  Try to recycle the vnode immediately if the
1251  * only active reference is our reference.
1252  *
1253  * Directory vnodes in the namecache with children cannot be immediately
1254  * recycled because numerous VOP_N*() ops require them to be stable.
1255  *
1256  * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1257  * function is a NOP if VRECLAIMED is already set.
1258  */
1259 int
1260 vrecycle(struct vnode *vp)
1261 {
1262 	if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
1263 		if (cache_inval_vp_nonblock(vp))
1264 			return(0);
1265 		vgone_vxlocked(vp);
1266 		return (1);
1267 	}
1268 	return (0);
1269 }
1270 
1271 /*
1272  * Return the maximum I/O size allowed for strategy calls on VP.
1273  *
1274  * If vp is VCHR or VBLK we dive the device, otherwise we use
1275  * the vp's mount info.
1276  */
1277 int
1278 vmaxiosize(struct vnode *vp)
1279 {
1280 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1281 		return(vp->v_rdev->si_iosize_max);
1282 	} else {
1283 		return(vp->v_mount->mnt_iosize_max);
1284 	}
1285 }
1286 
1287 /*
1288  * Eliminate all activity associated with a vnode in preparation for reuse.
1289  *
1290  * The vnode must be VX locked and refd and will remain VX locked and refd
1291  * on return.  This routine may be called with the vnode in any state, as
1292  * long as it is VX locked.  The vnode will be cleaned out and marked
1293  * VRECLAIMED but will not actually be reused until all existing refs and
1294  * holds go away.
1295  *
1296  * NOTE: This routine may be called on a vnode which has not yet been
1297  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1298  * already been reclaimed.
1299  *
1300  * This routine is not responsible for placing us back on the freelist.
1301  * Instead, it happens automatically when the caller releases the VX lock
1302  * (assuming there aren't any other references).
1303  */
1304 void
1305 vgone_vxlocked(struct vnode *vp)
1306 {
1307 	/*
1308 	 * assert that the VX lock is held.  This is an absolute requirement
1309 	 * now for vgone_vxlocked() to be called.
1310 	 */
1311 	KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1312 
1313 	get_mplock();
1314 
1315 	/*
1316 	 * Clean out the filesystem specific data and set the VRECLAIMED
1317 	 * bit.  Also deactivate the vnode if necessary.
1318 	 */
1319 	vclean_vxlocked(vp, DOCLOSE);
1320 
1321 	/*
1322 	 * Delete from old mount point vnode list, if on one.
1323 	 */
1324 	if (vp->v_mount != NULL) {
1325 		KKASSERT(vp->v_data == NULL);
1326 		insmntque(vp, NULL);
1327 	}
1328 
1329 	/*
1330 	 * If special device, remove it from special device alias list
1331 	 * if it is on one.  This should normally only occur if a vnode is
1332 	 * being revoked as the device should otherwise have been released
1333 	 * naturally.
1334 	 */
1335 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1336 		v_release_rdev(vp);
1337 	}
1338 
1339 	/*
1340 	 * Set us to VBAD
1341 	 */
1342 	vp->v_type = VBAD;
1343 	rel_mplock();
1344 }
1345 
1346 /*
1347  * Lookup a vnode by device number.
1348  *
1349  * Returns non-zero and *vpp set to a vref'd vnode on success.
1350  * Returns zero on failure.
1351  */
1352 int
1353 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1354 {
1355 	lwkt_tokref ilock;
1356 	struct vnode *vp;
1357 
1358 	lwkt_gettoken(&ilock, &spechash_token);
1359 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1360 		if (type == vp->v_type) {
1361 			*vpp = vp;
1362 			vref(vp);
1363 			lwkt_reltoken(&ilock);
1364 			return (1);
1365 		}
1366 	}
1367 	lwkt_reltoken(&ilock);
1368 	return (0);
1369 }
1370 
1371 /*
1372  * Calculate the total number of references to a special device.  This
1373  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1374  * an overloaded field.  Since udev2dev can now return NULL, we have
1375  * to check for a NULL v_rdev.
1376  */
1377 int
1378 count_dev(cdev_t dev)
1379 {
1380 	lwkt_tokref ilock;
1381 	struct vnode *vp;
1382 	int count = 0;
1383 
1384 	if (SLIST_FIRST(&dev->si_hlist)) {
1385 		lwkt_gettoken(&ilock, &spechash_token);
1386 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1387 			count += vp->v_opencount;
1388 		}
1389 		lwkt_reltoken(&ilock);
1390 	}
1391 	return(count);
1392 }
1393 
1394 int
1395 vcount(struct vnode *vp)
1396 {
1397 	if (vp->v_rdev == NULL)
1398 		return(0);
1399 	return(count_dev(vp->v_rdev));
1400 }
1401 
1402 /*
1403  * Initialize VMIO for a vnode.  This routine MUST be called before a
1404  * VFS can issue buffer cache ops on a vnode.  It is typically called
1405  * when a vnode is initialized from its inode.
1406  */
1407 int
1408 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
1409 {
1410 	vm_object_t object;
1411 	int error = 0;
1412 
1413 retry:
1414 	if ((object = vp->v_object) == NULL) {
1415 		object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
1416 		/*
1417 		 * Dereference the reference we just created.  This assumes
1418 		 * that the object is associated with the vp.
1419 		 */
1420 		object->ref_count--;
1421 		vrele(vp);
1422 	} else {
1423 		if (object->flags & OBJ_DEAD) {
1424 			vn_unlock(vp);
1425 			vm_object_dead_sleep(object, "vodead");
1426 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1427 			goto retry;
1428 		}
1429 	}
1430 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1431 	vsetflags(vp, VOBJBUF);
1432 	return (error);
1433 }
1434 
1435 
1436 /*
1437  * Print out a description of a vnode.
1438  */
1439 static char *typename[] =
1440 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1441 
1442 void
1443 vprint(char *label, struct vnode *vp)
1444 {
1445 	char buf[96];
1446 
1447 	if (label != NULL)
1448 		kprintf("%s: %p: ", label, (void *)vp);
1449 	else
1450 		kprintf("%p: ", (void *)vp);
1451 	kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1452 		typename[vp->v_type],
1453 		vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1454 	buf[0] = '\0';
1455 	if (vp->v_flag & VROOT)
1456 		strcat(buf, "|VROOT");
1457 	if (vp->v_flag & VPFSROOT)
1458 		strcat(buf, "|VPFSROOT");
1459 	if (vp->v_flag & VTEXT)
1460 		strcat(buf, "|VTEXT");
1461 	if (vp->v_flag & VSYSTEM)
1462 		strcat(buf, "|VSYSTEM");
1463 	if (vp->v_flag & VFREE)
1464 		strcat(buf, "|VFREE");
1465 	if (vp->v_flag & VOBJBUF)
1466 		strcat(buf, "|VOBJBUF");
1467 	if (buf[0] != '\0')
1468 		kprintf(" flags (%s)", &buf[1]);
1469 	if (vp->v_data == NULL) {
1470 		kprintf("\n");
1471 	} else {
1472 		kprintf("\n\t");
1473 		VOP_PRINT(vp);
1474 	}
1475 }
1476 
1477 /*
1478  * Do the usual access checking.
1479  * file_mode, uid and gid are from the vnode in question,
1480  * while acc_mode and cred are from the VOP_ACCESS parameter list
1481  */
1482 int
1483 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1484     mode_t acc_mode, struct ucred *cred)
1485 {
1486 	mode_t mask;
1487 	int ismember;
1488 
1489 	/*
1490 	 * Super-user always gets read/write access, but execute access depends
1491 	 * on at least one execute bit being set.
1492 	 */
1493 	if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
1494 		if ((acc_mode & VEXEC) && type != VDIR &&
1495 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1496 			return (EACCES);
1497 		return (0);
1498 	}
1499 
1500 	mask = 0;
1501 
1502 	/* Otherwise, check the owner. */
1503 	if (cred->cr_uid == uid) {
1504 		if (acc_mode & VEXEC)
1505 			mask |= S_IXUSR;
1506 		if (acc_mode & VREAD)
1507 			mask |= S_IRUSR;
1508 		if (acc_mode & VWRITE)
1509 			mask |= S_IWUSR;
1510 		return ((file_mode & mask) == mask ? 0 : EACCES);
1511 	}
1512 
1513 	/* Otherwise, check the groups. */
1514 	ismember = groupmember(gid, cred);
1515 	if (cred->cr_svgid == gid || ismember) {
1516 		if (acc_mode & VEXEC)
1517 			mask |= S_IXGRP;
1518 		if (acc_mode & VREAD)
1519 			mask |= S_IRGRP;
1520 		if (acc_mode & VWRITE)
1521 			mask |= S_IWGRP;
1522 		return ((file_mode & mask) == mask ? 0 : EACCES);
1523 	}
1524 
1525 	/* Otherwise, check everyone else. */
1526 	if (acc_mode & VEXEC)
1527 		mask |= S_IXOTH;
1528 	if (acc_mode & VREAD)
1529 		mask |= S_IROTH;
1530 	if (acc_mode & VWRITE)
1531 		mask |= S_IWOTH;
1532 	return ((file_mode & mask) == mask ? 0 : EACCES);
1533 }
1534 
1535 #ifdef DDB
1536 #include <ddb/ddb.h>
1537 
1538 static int db_show_locked_vnodes(struct mount *mp, void *data);
1539 
1540 /*
1541  * List all of the locked vnodes in the system.
1542  * Called when debugging the kernel.
1543  */
1544 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1545 {
1546 	kprintf("Locked vnodes\n");
1547 	mountlist_scan(db_show_locked_vnodes, NULL,
1548 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1549 }
1550 
1551 static int
1552 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1553 {
1554 	struct vnode *vp;
1555 
1556 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1557 		if (vn_islocked(vp))
1558 			vprint(NULL, vp);
1559 	}
1560 	return(0);
1561 }
1562 #endif
1563 
1564 /*
1565  * Top level filesystem related information gathering.
1566  */
1567 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1568 
1569 static int
1570 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1571 {
1572 	int *name = (int *)arg1 - 1;	/* XXX */
1573 	u_int namelen = arg2 + 1;	/* XXX */
1574 	struct vfsconf *vfsp;
1575 	int maxtypenum;
1576 
1577 #if 1 || defined(COMPAT_PRELITE2)
1578 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1579 	if (namelen == 1)
1580 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1581 #endif
1582 
1583 #ifdef notyet
1584 	/* all sysctl names at this level are at least name and field */
1585 	if (namelen < 2)
1586 		return (ENOTDIR);		/* overloaded */
1587 	if (name[0] != VFS_GENERIC) {
1588 		vfsp = vfsconf_find_by_typenum(name[0]);
1589 		if (vfsp == NULL)
1590 			return (EOPNOTSUPP);
1591 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1592 		    oldp, oldlenp, newp, newlen, p));
1593 	}
1594 #endif
1595 	switch (name[1]) {
1596 	case VFS_MAXTYPENUM:
1597 		if (namelen != 2)
1598 			return (ENOTDIR);
1599 		maxtypenum = vfsconf_get_maxtypenum();
1600 		return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1601 	case VFS_CONF:
1602 		if (namelen != 3)
1603 			return (ENOTDIR);	/* overloaded */
1604 		vfsp = vfsconf_find_by_typenum(name[2]);
1605 		if (vfsp == NULL)
1606 			return (EOPNOTSUPP);
1607 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1608 	}
1609 	return (EOPNOTSUPP);
1610 }
1611 
1612 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1613 	"Generic filesystem");
1614 
1615 #if 1 || defined(COMPAT_PRELITE2)
1616 
1617 static int
1618 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1619 {
1620 	int error;
1621 	struct ovfsconf ovfs;
1622 	struct sysctl_req *req = (struct sysctl_req*) data;
1623 
1624 	bzero(&ovfs, sizeof(ovfs));
1625 	ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1626 	strcpy(ovfs.vfc_name, vfsp->vfc_name);
1627 	ovfs.vfc_index = vfsp->vfc_typenum;
1628 	ovfs.vfc_refcount = vfsp->vfc_refcount;
1629 	ovfs.vfc_flags = vfsp->vfc_flags;
1630 	error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1631 	if (error)
1632 		return error; /* abort iteration with error code */
1633 	else
1634 		return 0; /* continue iterating with next element */
1635 }
1636 
1637 static int
1638 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1639 {
1640 	return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1641 }
1642 
1643 #endif /* 1 || COMPAT_PRELITE2 */
1644 
1645 /*
1646  * Check to see if a filesystem is mounted on a block device.
1647  */
1648 int
1649 vfs_mountedon(struct vnode *vp)
1650 {
1651 	cdev_t dev;
1652 
1653 	if ((dev = vp->v_rdev) == NULL) {
1654 /*		if (vp->v_type != VBLK)
1655 			dev = get_dev(vp->v_uminor, vp->v_umajor); */
1656 	}
1657 	if (dev != NULL && dev->si_mountpoint)
1658 		return (EBUSY);
1659 	return (0);
1660 }
1661 
1662 /*
1663  * Unmount all filesystems. The list is traversed in reverse order
1664  * of mounting to avoid dependencies.
1665  */
1666 
1667 static int vfs_umountall_callback(struct mount *mp, void *data);
1668 
1669 void
1670 vfs_unmountall(void)
1671 {
1672 	int count;
1673 
1674 	do {
1675 		count = mountlist_scan(vfs_umountall_callback,
1676 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1677 	} while (count);
1678 }
1679 
1680 static
1681 int
1682 vfs_umountall_callback(struct mount *mp, void *data)
1683 {
1684 	int error;
1685 
1686 	error = dounmount(mp, MNT_FORCE);
1687 	if (error) {
1688 		mountlist_remove(mp);
1689 		kprintf("unmount of filesystem mounted from %s failed (",
1690 			mp->mnt_stat.f_mntfromname);
1691 		if (error == EBUSY)
1692 			kprintf("BUSY)\n");
1693 		else
1694 			kprintf("%d)\n", error);
1695 	}
1696 	return(1);
1697 }
1698 
1699 /*
1700  * Checks the mount flags for parameter mp and put the names comma-separated
1701  * into a string buffer buf with a size limit specified by len.
1702  *
1703  * It returns the number of bytes written into buf, and (*errorp) will be
1704  * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1705  * not large enough).  The buffer will be 0-terminated if len was not 0.
1706  */
1707 size_t
1708 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
1709 	       char *buf, size_t len, int *errorp)
1710 {
1711 	static const struct mountctl_opt optnames[] = {
1712 		{ MNT_ASYNC,            "asynchronous" },
1713 		{ MNT_EXPORTED,         "NFS exported" },
1714 		{ MNT_LOCAL,            "local" },
1715 		{ MNT_NOATIME,          "noatime" },
1716 		{ MNT_NODEV,            "nodev" },
1717 		{ MNT_NOEXEC,           "noexec" },
1718 		{ MNT_NOSUID,           "nosuid" },
1719 		{ MNT_NOSYMFOLLOW,      "nosymfollow" },
1720 		{ MNT_QUOTA,            "with-quotas" },
1721 		{ MNT_RDONLY,           "read-only" },
1722 		{ MNT_SYNCHRONOUS,      "synchronous" },
1723 		{ MNT_UNION,            "union" },
1724 		{ MNT_NOCLUSTERR,       "noclusterr" },
1725 		{ MNT_NOCLUSTERW,       "noclusterw" },
1726 		{ MNT_SUIDDIR,          "suiddir" },
1727 		{ MNT_SOFTDEP,          "soft-updates" },
1728 		{ MNT_IGNORE,           "ignore" },
1729 		{ 0,			NULL}
1730 	};
1731 	int bwritten;
1732 	int bleft;
1733 	int optlen;
1734 	int actsize;
1735 
1736 	*errorp = 0;
1737 	bwritten = 0;
1738 	bleft = len - 1;	/* leave room for trailing \0 */
1739 
1740 	/*
1741 	 * Checks the size of the string. If it contains
1742 	 * any data, then we will append the new flags to
1743 	 * it.
1744 	 */
1745 	actsize = strlen(buf);
1746 	if (actsize > 0)
1747 		buf += actsize;
1748 
1749 	/* Default flags if no flags passed */
1750 	if (optp == NULL)
1751 		optp = optnames;
1752 
1753 	if (bleft < 0) {	/* degenerate case, 0-length buffer */
1754 		*errorp = EINVAL;
1755 		return(0);
1756 	}
1757 
1758 	for (; flags && optp->o_opt; ++optp) {
1759 		if ((flags & optp->o_opt) == 0)
1760 			continue;
1761 		optlen = strlen(optp->o_name);
1762 		if (bwritten || actsize > 0) {
1763 			if (bleft < 2) {
1764 				*errorp = ENOSPC;
1765 				break;
1766 			}
1767 			buf[bwritten++] = ',';
1768 			buf[bwritten++] = ' ';
1769 			bleft -= 2;
1770 		}
1771 		if (bleft < optlen) {
1772 			*errorp = ENOSPC;
1773 			break;
1774 		}
1775 		bcopy(optp->o_name, buf + bwritten, optlen);
1776 		bwritten += optlen;
1777 		bleft -= optlen;
1778 		flags &= ~optp->o_opt;
1779 	}
1780 
1781 	/*
1782 	 * Space already reserved for trailing \0
1783 	 */
1784 	buf[bwritten] = 0;
1785 	return (bwritten);
1786 }
1787 
1788 /*
1789  * Build hash lists of net addresses and hang them off the mount point.
1790  * Called by ufs_mount() to set up the lists of export addresses.
1791  */
1792 static int
1793 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1794 		const struct export_args *argp)
1795 {
1796 	struct netcred *np;
1797 	struct radix_node_head *rnh;
1798 	int i;
1799 	struct radix_node *rn;
1800 	struct sockaddr *saddr, *smask = 0;
1801 	struct domain *dom;
1802 	int error;
1803 
1804 	if (argp->ex_addrlen == 0) {
1805 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1806 			return (EPERM);
1807 		np = &nep->ne_defexported;
1808 		np->netc_exflags = argp->ex_flags;
1809 		np->netc_anon = argp->ex_anon;
1810 		np->netc_anon.cr_ref = 1;
1811 		mp->mnt_flag |= MNT_DEFEXPORTED;
1812 		return (0);
1813 	}
1814 
1815 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1816 		return (EINVAL);
1817 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1818 		return (EINVAL);
1819 
1820 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1821 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1822 	saddr = (struct sockaddr *) (np + 1);
1823 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1824 		goto out;
1825 	if (saddr->sa_len > argp->ex_addrlen)
1826 		saddr->sa_len = argp->ex_addrlen;
1827 	if (argp->ex_masklen) {
1828 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1829 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1830 		if (error)
1831 			goto out;
1832 		if (smask->sa_len > argp->ex_masklen)
1833 			smask->sa_len = argp->ex_masklen;
1834 	}
1835 	i = saddr->sa_family;
1836 	if ((rnh = nep->ne_rtable[i]) == 0) {
1837 		/*
1838 		 * Seems silly to initialize every AF when most are not used,
1839 		 * do so on demand here
1840 		 */
1841 		SLIST_FOREACH(dom, &domains, dom_next)
1842 			if (dom->dom_family == i && dom->dom_rtattach) {
1843 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1844 				    dom->dom_rtoffset);
1845 				break;
1846 			}
1847 		if ((rnh = nep->ne_rtable[i]) == 0) {
1848 			error = ENOBUFS;
1849 			goto out;
1850 		}
1851 	}
1852 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1853 	    np->netc_rnodes);
1854 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1855 		error = EPERM;
1856 		goto out;
1857 	}
1858 	np->netc_exflags = argp->ex_flags;
1859 	np->netc_anon = argp->ex_anon;
1860 	np->netc_anon.cr_ref = 1;
1861 	return (0);
1862 out:
1863 	kfree(np, M_NETADDR);
1864 	return (error);
1865 }
1866 
1867 /* ARGSUSED */
1868 static int
1869 vfs_free_netcred(struct radix_node *rn, void *w)
1870 {
1871 	struct radix_node_head *rnh = (struct radix_node_head *) w;
1872 
1873 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1874 	kfree((caddr_t) rn, M_NETADDR);
1875 	return (0);
1876 }
1877 
1878 /*
1879  * Free the net address hash lists that are hanging off the mount points.
1880  */
1881 static void
1882 vfs_free_addrlist(struct netexport *nep)
1883 {
1884 	int i;
1885 	struct radix_node_head *rnh;
1886 
1887 	for (i = 0; i <= AF_MAX; i++)
1888 		if ((rnh = nep->ne_rtable[i])) {
1889 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1890 			    (caddr_t) rnh);
1891 			kfree((caddr_t) rnh, M_RTABLE);
1892 			nep->ne_rtable[i] = 0;
1893 		}
1894 }
1895 
1896 int
1897 vfs_export(struct mount *mp, struct netexport *nep,
1898 	   const struct export_args *argp)
1899 {
1900 	int error;
1901 
1902 	if (argp->ex_flags & MNT_DELEXPORT) {
1903 		if (mp->mnt_flag & MNT_EXPUBLIC) {
1904 			vfs_setpublicfs(NULL, NULL, NULL);
1905 			mp->mnt_flag &= ~MNT_EXPUBLIC;
1906 		}
1907 		vfs_free_addrlist(nep);
1908 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1909 	}
1910 	if (argp->ex_flags & MNT_EXPORTED) {
1911 		if (argp->ex_flags & MNT_EXPUBLIC) {
1912 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1913 				return (error);
1914 			mp->mnt_flag |= MNT_EXPUBLIC;
1915 		}
1916 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1917 			return (error);
1918 		mp->mnt_flag |= MNT_EXPORTED;
1919 	}
1920 	return (0);
1921 }
1922 
1923 
1924 /*
1925  * Set the publicly exported filesystem (WebNFS). Currently, only
1926  * one public filesystem is possible in the spec (RFC 2054 and 2055)
1927  */
1928 int
1929 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1930 		const struct export_args *argp)
1931 {
1932 	int error;
1933 	struct vnode *rvp;
1934 	char *cp;
1935 
1936 	/*
1937 	 * mp == NULL -> invalidate the current info, the FS is
1938 	 * no longer exported. May be called from either vfs_export
1939 	 * or unmount, so check if it hasn't already been done.
1940 	 */
1941 	if (mp == NULL) {
1942 		if (nfs_pub.np_valid) {
1943 			nfs_pub.np_valid = 0;
1944 			if (nfs_pub.np_index != NULL) {
1945 				FREE(nfs_pub.np_index, M_TEMP);
1946 				nfs_pub.np_index = NULL;
1947 			}
1948 		}
1949 		return (0);
1950 	}
1951 
1952 	/*
1953 	 * Only one allowed at a time.
1954 	 */
1955 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1956 		return (EBUSY);
1957 
1958 	/*
1959 	 * Get real filehandle for root of exported FS.
1960 	 */
1961 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1962 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1963 
1964 	if ((error = VFS_ROOT(mp, &rvp)))
1965 		return (error);
1966 
1967 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1968 		return (error);
1969 
1970 	vput(rvp);
1971 
1972 	/*
1973 	 * If an indexfile was specified, pull it in.
1974 	 */
1975 	if (argp->ex_indexfile != NULL) {
1976 		int namelen;
1977 
1978 		error = vn_get_namelen(rvp, &namelen);
1979 		if (error)
1980 			return (error);
1981 		MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1982 		    M_WAITOK);
1983 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1984 		    namelen, NULL);
1985 		if (!error) {
1986 			/*
1987 			 * Check for illegal filenames.
1988 			 */
1989 			for (cp = nfs_pub.np_index; *cp; cp++) {
1990 				if (*cp == '/') {
1991 					error = EINVAL;
1992 					break;
1993 				}
1994 			}
1995 		}
1996 		if (error) {
1997 			FREE(nfs_pub.np_index, M_TEMP);
1998 			return (error);
1999 		}
2000 	}
2001 
2002 	nfs_pub.np_mount = mp;
2003 	nfs_pub.np_valid = 1;
2004 	return (0);
2005 }
2006 
2007 struct netcred *
2008 vfs_export_lookup(struct mount *mp, struct netexport *nep,
2009 		struct sockaddr *nam)
2010 {
2011 	struct netcred *np;
2012 	struct radix_node_head *rnh;
2013 	struct sockaddr *saddr;
2014 
2015 	np = NULL;
2016 	if (mp->mnt_flag & MNT_EXPORTED) {
2017 		/*
2018 		 * Lookup in the export list first.
2019 		 */
2020 		if (nam != NULL) {
2021 			saddr = nam;
2022 			rnh = nep->ne_rtable[saddr->sa_family];
2023 			if (rnh != NULL) {
2024 				np = (struct netcred *)
2025 					(*rnh->rnh_matchaddr)((char *)saddr,
2026 							      rnh);
2027 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2028 					np = NULL;
2029 			}
2030 		}
2031 		/*
2032 		 * If no address match, use the default if it exists.
2033 		 */
2034 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2035 			np = &nep->ne_defexported;
2036 	}
2037 	return (np);
2038 }
2039 
2040 /*
2041  * perform msync on all vnodes under a mount point.  The mount point must
2042  * be locked.  This code is also responsible for lazy-freeing unreferenced
2043  * vnodes whos VM objects no longer contain pages.
2044  *
2045  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2046  *
2047  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2048  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
2049  * way up in this high level function.
2050  */
2051 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
2052 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
2053 
2054 void
2055 vfs_msync(struct mount *mp, int flags)
2056 {
2057 	int vmsc_flags;
2058 
2059 	/*
2060 	 * tmpfs sets this flag to prevent msync(), sync, and the
2061 	 * filesystem periodic syncer from trying to flush VM pages
2062 	 * to swap.  Only pure memory pressure flushes tmpfs VM pages
2063 	 * to swap.
2064 	 */
2065 	if (mp->mnt_kern_flag & MNTK_NOMSYNC)
2066 		return;
2067 
2068 	/*
2069 	 * Ok, scan the vnodes for work.
2070 	 */
2071 	vmsc_flags = VMSC_GETVP;
2072 	if (flags != MNT_WAIT)
2073 		vmsc_flags |= VMSC_NOWAIT;
2074 	vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
2075 			(void *)(intptr_t)flags);
2076 }
2077 
2078 /*
2079  * scan1 is a fast pre-check.  There could be hundreds of thousands of
2080  * vnodes, we cannot afford to do anything heavy weight until we have a
2081  * fairly good indication that there is work to do.
2082  */
2083 static
2084 int
2085 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
2086 {
2087 	int flags = (int)(intptr_t)data;
2088 
2089 	if ((vp->v_flag & VRECLAIMED) == 0) {
2090 		if (vshouldmsync(vp))
2091 			return(0);	/* call scan2 */
2092 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
2093 		    (vp->v_flag & VOBJDIRTY) &&
2094 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
2095 			return(0);	/* call scan2 */
2096 		}
2097 	}
2098 
2099 	/*
2100 	 * do not call scan2, continue the loop
2101 	 */
2102 	return(-1);
2103 }
2104 
2105 /*
2106  * This callback is handed a locked vnode.
2107  */
2108 static
2109 int
2110 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
2111 {
2112 	vm_object_t obj;
2113 	int flags = (int)(intptr_t)data;
2114 
2115 	if (vp->v_flag & VRECLAIMED)
2116 		return(0);
2117 
2118 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
2119 		if ((obj = vp->v_object) != NULL) {
2120 			vm_object_page_clean(obj, 0, 0,
2121 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2122 		}
2123 	}
2124 	return(0);
2125 }
2126 
2127 /*
2128  * Record a process's interest in events which might happen to
2129  * a vnode.  Because poll uses the historic select-style interface
2130  * internally, this routine serves as both the ``check for any
2131  * pending events'' and the ``record my interest in future events''
2132  * functions.  (These are done together, while the lock is held,
2133  * to avoid race conditions.)
2134  */
2135 int
2136 vn_pollrecord(struct vnode *vp, int events)
2137 {
2138 	lwkt_tokref vlock;
2139 
2140 	KKASSERT(curthread->td_proc != NULL);
2141 
2142 	lwkt_gettoken(&vlock, &vp->v_token);
2143 	if (vp->v_pollinfo.vpi_revents & events) {
2144 		/*
2145 		 * This leaves events we are not interested
2146 		 * in available for the other process which
2147 		 * which presumably had requested them
2148 		 * (otherwise they would never have been
2149 		 * recorded).
2150 		 */
2151 		events &= vp->v_pollinfo.vpi_revents;
2152 		vp->v_pollinfo.vpi_revents &= ~events;
2153 
2154 		lwkt_reltoken(&vlock);
2155 		return events;
2156 	}
2157 	vp->v_pollinfo.vpi_events |= events;
2158 	selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
2159 	lwkt_reltoken(&vlock);
2160 	return 0;
2161 }
2162 
2163 /*
2164  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2165  * it is possible for us to miss an event due to race conditions, but
2166  * that condition is expected to be rare, so for the moment it is the
2167  * preferred interface.
2168  */
2169 void
2170 vn_pollevent(struct vnode *vp, int events)
2171 {
2172 	lwkt_tokref vlock;
2173 
2174 	lwkt_gettoken(&vlock, &vp->v_token);
2175 	if (vp->v_pollinfo.vpi_events & events) {
2176 		/*
2177 		 * We clear vpi_events so that we don't
2178 		 * call selwakeup() twice if two events are
2179 		 * posted before the polling process(es) is
2180 		 * awakened.  This also ensures that we take at
2181 		 * most one selwakeup() if the polling process
2182 		 * is no longer interested.  However, it does
2183 		 * mean that only one event can be noticed at
2184 		 * a time.  (Perhaps we should only clear those
2185 		 * event bits which we note?) XXX
2186 		 */
2187 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2188 		vp->v_pollinfo.vpi_revents |= events;
2189 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2190 	}
2191 	lwkt_reltoken(&vlock);
2192 }
2193 
2194 /*
2195  * Wake up anyone polling on vp because it is being revoked.
2196  * This depends on dead_poll() returning POLLHUP for correct
2197  * behavior.
2198  */
2199 void
2200 vn_pollgone(struct vnode *vp)
2201 {
2202 	lwkt_tokref vlock;
2203 
2204 	lwkt_gettoken(&vlock, &vp->v_token);
2205 	if (vp->v_pollinfo.vpi_events) {
2206 		vp->v_pollinfo.vpi_events = 0;
2207 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2208 	}
2209 	lwkt_reltoken(&vlock);
2210 }
2211 
2212 /*
2213  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2214  * (or v_rdev might be NULL).
2215  */
2216 cdev_t
2217 vn_todev(struct vnode *vp)
2218 {
2219 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2220 		return (NULL);
2221 	KKASSERT(vp->v_rdev != NULL);
2222 	return (vp->v_rdev);
2223 }
2224 
2225 /*
2226  * Check if vnode represents a disk device.  The vnode does not need to be
2227  * opened.
2228  *
2229  * MPALMOSTSAFE
2230  */
2231 int
2232 vn_isdisk(struct vnode *vp, int *errp)
2233 {
2234 	cdev_t dev;
2235 
2236 	if (vp->v_type != VCHR) {
2237 		if (errp != NULL)
2238 			*errp = ENOTBLK;
2239 		return (0);
2240 	}
2241 
2242 	dev = vp->v_rdev;
2243 
2244 	if (dev == NULL) {
2245 		if (errp != NULL)
2246 			*errp = ENXIO;
2247 		return (0);
2248 	}
2249 	if (dev_is_good(dev) == 0) {
2250 		if (errp != NULL)
2251 			*errp = ENXIO;
2252 		return (0);
2253 	}
2254 	if ((dev_dflags(dev) & D_DISK) == 0) {
2255 		if (errp != NULL)
2256 			*errp = ENOTBLK;
2257 		return (0);
2258 	}
2259 	if (errp != NULL)
2260 		*errp = 0;
2261 	return (1);
2262 }
2263 
2264 int
2265 vn_get_namelen(struct vnode *vp, int *namelen)
2266 {
2267 	int error;
2268 	register_t retval[2];
2269 
2270 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2271 	if (error)
2272 		return (error);
2273 	*namelen = (int)retval[0];
2274 	return (0);
2275 }
2276 
2277 int
2278 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2279 		uint16_t d_namlen, const char *d_name)
2280 {
2281 	struct dirent *dp;
2282 	size_t len;
2283 
2284 	len = _DIRENT_RECLEN(d_namlen);
2285 	if (len > uio->uio_resid)
2286 		return(1);
2287 
2288 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2289 
2290 	dp->d_ino = d_ino;
2291 	dp->d_namlen = d_namlen;
2292 	dp->d_type = d_type;
2293 	bcopy(d_name, dp->d_name, d_namlen);
2294 
2295 	*error = uiomove((caddr_t)dp, len, uio);
2296 
2297 	kfree(dp, M_TEMP);
2298 
2299 	return(0);
2300 }
2301 
2302 void
2303 vn_mark_atime(struct vnode *vp, struct thread *td)
2304 {
2305 	struct proc *p = td->td_proc;
2306 	struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2307 
2308 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2309 		VOP_MARKATIME(vp, cred);
2310 	}
2311 }
2312