xref: /dflybsd-src/sys/kern/vfs_subr.c (revision 5cccfb7b21444e0e73a738d924f82daf27b4854d)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_subr.c,v 1.118 2008/09/17 21:44:18 dillon Exp $
41  */
42 
43 /*
44  * External virtual filesystem routines
45  */
46 #include "opt_ddb.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/file.h>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/malloc.h>
60 #include <sys/mbuf.h>
61 #include <sys/mount.h>
62 #include <sys/proc.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <machine/limits.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_kern.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_pager.h>
82 #include <vm/vnode_pager.h>
83 #include <vm/vm_zone.h>
84 
85 #include <sys/buf2.h>
86 #include <sys/thread2.h>
87 #include <sys/sysref2.h>
88 
89 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
90 
91 int numvnodes;
92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93 int vfs_fastdev = 1;
94 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
95 
96 enum vtype iftovt_tab[16] = {
97 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
98 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
99 };
100 int vttoif_tab[9] = {
101 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
102 	S_IFSOCK, S_IFIFO, S_IFMT,
103 };
104 
105 static int reassignbufcalls;
106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
107 		&reassignbufcalls, 0, "");
108 static int reassignbufloops;
109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
110 		&reassignbufloops, 0, "");
111 static int reassignbufsortgood;
112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
113 		&reassignbufsortgood, 0, "");
114 static int reassignbufsortbad;
115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
116 		&reassignbufsortbad, 0, "");
117 static int reassignbufmethod = 1;
118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
119 		&reassignbufmethod, 0, "");
120 
121 int	nfs_mount_type = -1;
122 static struct lwkt_token spechash_token;
123 struct nfs_public nfs_pub;	/* publicly exported FS */
124 
125 int desiredvnodes;
126 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
127 		&desiredvnodes, 0, "Maximum number of vnodes");
128 
129 static void	vfs_free_addrlist (struct netexport *nep);
130 static int	vfs_free_netcred (struct radix_node *rn, void *w);
131 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
132 				       const struct export_args *argp);
133 
134 extern int dev_ref_debug;
135 
136 /*
137  * Red black tree functions
138  */
139 static int rb_buf_compare(struct buf *b1, struct buf *b2);
140 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
141 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
142 
143 static int
144 rb_buf_compare(struct buf *b1, struct buf *b2)
145 {
146 	if (b1->b_loffset < b2->b_loffset)
147 		return(-1);
148 	if (b1->b_loffset > b2->b_loffset)
149 		return(1);
150 	return(0);
151 }
152 
153 /*
154  * Returns non-zero if the vnode is a candidate for lazy msyncing.
155  */
156 static __inline int
157 vshouldmsync(struct vnode *vp)
158 {
159 	if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
160 		return (0);		/* other holders */
161 	if (vp->v_object &&
162 	    (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
163 		return (0);
164 	}
165 	return (1);
166 }
167 
168 /*
169  * Initialize the vnode management data structures.
170  *
171  * Called from vfsinit()
172  */
173 void
174 vfs_subr_init(void)
175 {
176 	/*
177 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
178 	 * according to available system memory but we may also have
179 	 * to limit it based on available KVM, which is capped on 32 bit
180 	 * systems.
181 	 */
182 	desiredvnodes = min(maxproc + vmstats.v_page_count / 4,
183 			    KvaSize / (20 *
184 			    (sizeof(struct vm_object) + sizeof(struct vnode))));
185 
186 	lwkt_token_init(&spechash_token);
187 }
188 
189 /*
190  * Knob to control the precision of file timestamps:
191  *
192  *   0 = seconds only; nanoseconds zeroed.
193  *   1 = seconds and nanoseconds, accurate within 1/HZ.
194  *   2 = seconds and nanoseconds, truncated to microseconds.
195  * >=3 = seconds and nanoseconds, maximum precision.
196  */
197 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
198 
199 static int timestamp_precision = TSP_SEC;
200 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
201 		&timestamp_precision, 0, "");
202 
203 /*
204  * Get a current timestamp.
205  *
206  * MPSAFE
207  */
208 void
209 vfs_timestamp(struct timespec *tsp)
210 {
211 	struct timeval tv;
212 
213 	switch (timestamp_precision) {
214 	case TSP_SEC:
215 		tsp->tv_sec = time_second;
216 		tsp->tv_nsec = 0;
217 		break;
218 	case TSP_HZ:
219 		getnanotime(tsp);
220 		break;
221 	case TSP_USEC:
222 		microtime(&tv);
223 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
224 		break;
225 	case TSP_NSEC:
226 	default:
227 		nanotime(tsp);
228 		break;
229 	}
230 }
231 
232 /*
233  * Set vnode attributes to VNOVAL
234  */
235 void
236 vattr_null(struct vattr *vap)
237 {
238 	vap->va_type = VNON;
239 	vap->va_size = VNOVAL;
240 	vap->va_bytes = VNOVAL;
241 	vap->va_mode = VNOVAL;
242 	vap->va_nlink = VNOVAL;
243 	vap->va_uid = VNOVAL;
244 	vap->va_gid = VNOVAL;
245 	vap->va_fsid = VNOVAL;
246 	vap->va_fileid = VNOVAL;
247 	vap->va_blocksize = VNOVAL;
248 	vap->va_rmajor = VNOVAL;
249 	vap->va_rminor = VNOVAL;
250 	vap->va_atime.tv_sec = VNOVAL;
251 	vap->va_atime.tv_nsec = VNOVAL;
252 	vap->va_mtime.tv_sec = VNOVAL;
253 	vap->va_mtime.tv_nsec = VNOVAL;
254 	vap->va_ctime.tv_sec = VNOVAL;
255 	vap->va_ctime.tv_nsec = VNOVAL;
256 	vap->va_flags = VNOVAL;
257 	vap->va_gen = VNOVAL;
258 	vap->va_vaflags = 0;
259 	vap->va_fsmid = VNOVAL;
260 	/* va_*_uuid fields are only valid if related flags are set */
261 }
262 
263 /*
264  * Flush out and invalidate all buffers associated with a vnode.
265  *
266  * vp must be locked.
267  */
268 static int vinvalbuf_bp(struct buf *bp, void *data);
269 
270 struct vinvalbuf_bp_info {
271 	struct vnode *vp;
272 	int slptimeo;
273 	int lkflags;
274 	int flags;
275 };
276 
277 void
278 vupdatefsmid(struct vnode *vp)
279 {
280 	atomic_set_int(&vp->v_flag, VFSMID);
281 }
282 
283 int
284 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
285 {
286 	struct vinvalbuf_bp_info info;
287 	vm_object_t object;
288 	lwkt_tokref vlock;
289 	int error;
290 
291 	lwkt_gettoken(&vlock, &vp->v_token);
292 
293 	/*
294 	 * If we are being asked to save, call fsync to ensure that the inode
295 	 * is updated.
296 	 */
297 	if (flags & V_SAVE) {
298 		error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
299 		if (error)
300 			goto done;
301 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
302 			if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0)
303 				goto done;
304 
305 			/*
306 			 * Dirty bufs may be left or generated via races
307 			 * in circumstances where vinvalbuf() is called on
308 			 * a vnode not undergoing reclamation.   Only
309 			 * panic if we are trying to reclaim the vnode.
310 			 */
311 			if ((vp->v_flag & VRECLAIMED) &&
312 			    (bio_track_active(&vp->v_track_write) ||
313 			    !RB_EMPTY(&vp->v_rbdirty_tree))) {
314 				panic("vinvalbuf: dirty bufs");
315 			}
316 		}
317   	}
318 	info.slptimeo = slptimeo;
319 	info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
320 	if (slpflag & PCATCH)
321 		info.lkflags |= LK_PCATCH;
322 	info.flags = flags;
323 	info.vp = vp;
324 
325 	/*
326 	 * Flush the buffer cache until nothing is left.
327 	 */
328 	while (!RB_EMPTY(&vp->v_rbclean_tree) ||
329 	       !RB_EMPTY(&vp->v_rbdirty_tree)) {
330 		error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
331 				vinvalbuf_bp, &info);
332 		if (error == 0) {
333 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
334 					vinvalbuf_bp, &info);
335 		}
336 	}
337 
338 	/*
339 	 * Wait for I/O completion.  We may block in the pip code so we have
340 	 * to re-check.
341 	 */
342 	do {
343 		bio_track_wait(&vp->v_track_write, 0, 0);
344 		if ((object = vp->v_object) != NULL) {
345 			while (object->paging_in_progress)
346 				vm_object_pip_sleep(object, "vnvlbx");
347 		}
348 	} while (bio_track_active(&vp->v_track_write));
349 
350 	/*
351 	 * Destroy the copy in the VM cache, too.
352 	 */
353 	if ((object = vp->v_object) != NULL) {
354 		vm_object_page_remove(object, 0, 0,
355 			(flags & V_SAVE) ? TRUE : FALSE);
356 	}
357 
358 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
359 		panic("vinvalbuf: flush failed");
360 	if (!RB_EMPTY(&vp->v_rbhash_tree))
361 		panic("vinvalbuf: flush failed, buffers still present");
362 	error = 0;
363 done:
364 	lwkt_reltoken(&vlock);
365 	return (error);
366 }
367 
368 static int
369 vinvalbuf_bp(struct buf *bp, void *data)
370 {
371 	struct vinvalbuf_bp_info *info = data;
372 	int error;
373 
374 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
375 		error = BUF_TIMELOCK(bp, info->lkflags,
376 				     "vinvalbuf", info->slptimeo);
377 		if (error == 0) {
378 			BUF_UNLOCK(bp);
379 			error = ENOLCK;
380 		}
381 		if (error == ENOLCK)
382 			return(0);
383 		return (-error);
384 	}
385 
386 	KKASSERT(bp->b_vp == info->vp);
387 
388 	/*
389 	 * XXX Since there are no node locks for NFS, I
390 	 * believe there is a slight chance that a delayed
391 	 * write will occur while sleeping just above, so
392 	 * check for it.  Note that vfs_bio_awrite expects
393 	 * buffers to reside on a queue, while bwrite() and
394 	 * brelse() do not.
395 	 *
396 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
397 	 * check.  This code will write out the buffer, period.
398 	 */
399 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
400 	    (info->flags & V_SAVE)) {
401 		if (bp->b_vp == info->vp) {
402 			if (bp->b_flags & B_CLUSTEROK) {
403 				vfs_bio_awrite(bp);
404 			} else {
405 				bremfree(bp);
406 				bp->b_flags |= B_ASYNC;
407 				bwrite(bp);
408 			}
409 		} else {
410 			bremfree(bp);
411 			bwrite(bp);
412 		}
413 	} else if (info->flags & V_SAVE) {
414 		/*
415 		 * Cannot set B_NOCACHE on a clean buffer as this will
416 		 * destroy the VM backing store which might actually
417 		 * be dirty (and unsynchronized).
418 		 */
419 		bremfree(bp);
420 		bp->b_flags |= (B_INVAL | B_RELBUF);
421 		bp->b_flags &= ~B_ASYNC;
422 		brelse(bp);
423 	} else {
424 		bremfree(bp);
425 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
426 		bp->b_flags &= ~B_ASYNC;
427 		brelse(bp);
428 	}
429 	return(0);
430 }
431 
432 /*
433  * Truncate a file's buffer and pages to a specified length.  This
434  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
435  * sync activity.
436  *
437  * The vnode must be locked.
438  */
439 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
440 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
441 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
442 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
443 
444 int
445 vtruncbuf(struct vnode *vp, off_t length, int blksize)
446 {
447 	off_t truncloffset;
448 	const char *filename;
449 	lwkt_tokref vlock;
450 	int count;
451 
452 	/*
453 	 * Round up to the *next* block, then destroy the buffers in question.
454 	 * Since we are only removing some of the buffers we must rely on the
455 	 * scan count to determine whether a loop is necessary.
456 	 */
457 	if ((count = (int)(length % blksize)) != 0)
458 		truncloffset = length + (blksize - count);
459 	else
460 		truncloffset = length;
461 
462 	lwkt_gettoken(&vlock, &vp->v_token);
463 	do {
464 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
465 				vtruncbuf_bp_trunc_cmp,
466 				vtruncbuf_bp_trunc, &truncloffset);
467 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
468 				vtruncbuf_bp_trunc_cmp,
469 				vtruncbuf_bp_trunc, &truncloffset);
470 	} while(count);
471 
472 	/*
473 	 * For safety, fsync any remaining metadata if the file is not being
474 	 * truncated to 0.  Since the metadata does not represent the entire
475 	 * dirty list we have to rely on the hit count to ensure that we get
476 	 * all of it.
477 	 */
478 	if (length > 0) {
479 		do {
480 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
481 					vtruncbuf_bp_metasync_cmp,
482 					vtruncbuf_bp_metasync, vp);
483 		} while (count);
484 	}
485 
486 	/*
487 	 * Clean out any left over VM backing store.
488 	 *
489 	 * It is possible to have in-progress I/O from buffers that were
490 	 * not part of the truncation.  This should not happen if we
491 	 * are truncating to 0-length.
492 	 */
493 	vnode_pager_setsize(vp, length);
494 	bio_track_wait(&vp->v_track_write, 0, 0);
495 
496 	filename = TAILQ_FIRST(&vp->v_namecache) ?
497 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
498 
499 	/*
500 	 * Make sure no buffers were instantiated while we were trying
501 	 * to clean out the remaining VM pages.  This could occur due
502 	 * to busy dirty VM pages being flushed out to disk.
503 	 */
504 	do {
505 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
506 				vtruncbuf_bp_trunc_cmp,
507 				vtruncbuf_bp_trunc, &truncloffset);
508 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
509 				vtruncbuf_bp_trunc_cmp,
510 				vtruncbuf_bp_trunc, &truncloffset);
511 		if (count) {
512 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
513 			       "left over buffers in %s\n", count, filename);
514 		}
515 	} while(count);
516 
517 	lwkt_reltoken(&vlock);
518 
519 	return (0);
520 }
521 
522 /*
523  * The callback buffer is beyond the new file EOF and must be destroyed.
524  * Note that the compare function must conform to the RB_SCAN's requirements.
525  */
526 static
527 int
528 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
529 {
530 	if (bp->b_loffset >= *(off_t *)data)
531 		return(0);
532 	return(-1);
533 }
534 
535 static
536 int
537 vtruncbuf_bp_trunc(struct buf *bp, void *data)
538 {
539 	/*
540 	 * Do not try to use a buffer we cannot immediately lock, but sleep
541 	 * anyway to prevent a livelock.  The code will loop until all buffers
542 	 * can be acted upon.
543 	 */
544 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
545 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
546 			BUF_UNLOCK(bp);
547 	} else {
548 		bremfree(bp);
549 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
550 		bp->b_flags &= ~B_ASYNC;
551 		brelse(bp);
552 	}
553 	return(1);
554 }
555 
556 /*
557  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
558  * blocks (with a negative loffset) are scanned.
559  * Note that the compare function must conform to the RB_SCAN's requirements.
560  */
561 static int
562 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
563 {
564 	if (bp->b_loffset < 0)
565 		return(0);
566 	return(1);
567 }
568 
569 static int
570 vtruncbuf_bp_metasync(struct buf *bp, void *data)
571 {
572 	struct vnode *vp = data;
573 
574 	if (bp->b_flags & B_DELWRI) {
575 		/*
576 		 * Do not try to use a buffer we cannot immediately lock,
577 		 * but sleep anyway to prevent a livelock.  The code will
578 		 * loop until all buffers can be acted upon.
579 		 */
580 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
581 			if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
582 				BUF_UNLOCK(bp);
583 		} else {
584 			bremfree(bp);
585 			if (bp->b_vp == vp) {
586 				bp->b_flags |= B_ASYNC;
587 			} else {
588 				bp->b_flags &= ~B_ASYNC;
589 			}
590 			bwrite(bp);
591 		}
592 		return(1);
593 	} else {
594 		return(0);
595 	}
596 }
597 
598 /*
599  * vfsync - implements a multipass fsync on a file which understands
600  * dependancies and meta-data.  The passed vnode must be locked.  The
601  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
602  *
603  * When fsyncing data asynchronously just do one consolidated pass starting
604  * with the most negative block number.  This may not get all the data due
605  * to dependancies.
606  *
607  * When fsyncing data synchronously do a data pass, then a metadata pass,
608  * then do additional data+metadata passes to try to get all the data out.
609  */
610 static int vfsync_wait_output(struct vnode *vp,
611 			    int (*waitoutput)(struct vnode *, struct thread *));
612 static int vfsync_data_only_cmp(struct buf *bp, void *data);
613 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
614 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
615 static int vfsync_bp(struct buf *bp, void *data);
616 
617 struct vfsync_info {
618 	struct vnode *vp;
619 	int synchronous;
620 	int syncdeps;
621 	int lazycount;
622 	int lazylimit;
623 	int skippedbufs;
624 	int (*checkdef)(struct buf *);
625 };
626 
627 int
628 vfsync(struct vnode *vp, int waitfor, int passes,
629 	int (*checkdef)(struct buf *),
630 	int (*waitoutput)(struct vnode *, struct thread *))
631 {
632 	struct vfsync_info info;
633 	lwkt_tokref vlock;
634 	int error;
635 
636 	bzero(&info, sizeof(info));
637 	info.vp = vp;
638 	if ((info.checkdef = checkdef) == NULL)
639 		info.syncdeps = 1;
640 
641 	lwkt_gettoken(&vlock, &vp->v_token);
642 
643 	switch(waitfor) {
644 	case MNT_LAZY:
645 		/*
646 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
647 		 * number of data (not meta) pages we try to flush to 1MB.
648 		 * A non-zero return means that lazy limit was reached.
649 		 */
650 		info.lazylimit = 1024 * 1024;
651 		info.syncdeps = 1;
652 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
653 				vfsync_lazy_range_cmp, vfsync_bp, &info);
654 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
655 				vfsync_meta_only_cmp, vfsync_bp, &info);
656 		if (error == 0)
657 			vp->v_lazyw = 0;
658 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
659 			vn_syncer_add_to_worklist(vp, 1);
660 		error = 0;
661 		break;
662 	case MNT_NOWAIT:
663 		/*
664 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
665 		 */
666 		info.syncdeps = 1;
667 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
668 			vfsync_bp, &info);
669 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
670 			vfsync_bp, &info);
671 		error = 0;
672 		break;
673 	default:
674 		/*
675 		 * Synchronous.  Do a data-only pass, then a meta-data+data
676 		 * pass, then additional integrated passes to try to get
677 		 * all the dependancies flushed.
678 		 */
679 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
680 			vfsync_bp, &info);
681 		error = vfsync_wait_output(vp, waitoutput);
682 		if (error == 0) {
683 			info.skippedbufs = 0;
684 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
685 				vfsync_bp, &info);
686 			error = vfsync_wait_output(vp, waitoutput);
687 			if (info.skippedbufs)
688 				kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
689 		}
690 		while (error == 0 && passes > 0 &&
691 		       !RB_EMPTY(&vp->v_rbdirty_tree)
692 		) {
693 			if (--passes == 0) {
694 				info.synchronous = 1;
695 				info.syncdeps = 1;
696 			}
697 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
698 				vfsync_bp, &info);
699 			if (error < 0)
700 				error = -error;
701 			info.syncdeps = 1;
702 			if (error == 0)
703 				error = vfsync_wait_output(vp, waitoutput);
704 		}
705 		break;
706 	}
707 	lwkt_reltoken(&vlock);
708 	return(error);
709 }
710 
711 static int
712 vfsync_wait_output(struct vnode *vp,
713 		   int (*waitoutput)(struct vnode *, struct thread *))
714 {
715 	int error;
716 
717 	error = bio_track_wait(&vp->v_track_write, 0, 0);
718 	if (waitoutput)
719 		error = waitoutput(vp, curthread);
720 	return(error);
721 }
722 
723 static int
724 vfsync_data_only_cmp(struct buf *bp, void *data)
725 {
726 	if (bp->b_loffset < 0)
727 		return(-1);
728 	return(0);
729 }
730 
731 static int
732 vfsync_meta_only_cmp(struct buf *bp, void *data)
733 {
734 	if (bp->b_loffset < 0)
735 		return(0);
736 	return(1);
737 }
738 
739 static int
740 vfsync_lazy_range_cmp(struct buf *bp, void *data)
741 {
742 	struct vfsync_info *info = data;
743 	if (bp->b_loffset < info->vp->v_lazyw)
744 		return(-1);
745 	return(0);
746 }
747 
748 static int
749 vfsync_bp(struct buf *bp, void *data)
750 {
751 	struct vfsync_info *info = data;
752 	struct vnode *vp = info->vp;
753 	int error;
754 
755 	/*
756 	 * if syncdeps is not set we do not try to write buffers which have
757 	 * dependancies.
758 	 */
759 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
760 		return(0);
761 
762 	/*
763 	 * Ignore buffers that we cannot immediately lock.  XXX
764 	 */
765 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
766 		kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
767 		++info->skippedbufs;
768 		return(0);
769 	}
770 	if ((bp->b_flags & B_DELWRI) == 0)
771 		panic("vfsync_bp: buffer not dirty");
772 	if (vp != bp->b_vp)
773 		panic("vfsync_bp: buffer vp mismatch");
774 
775 	/*
776 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
777 	 * has been written but an additional handshake with the device
778 	 * is required before we can dispose of the buffer.  We have no idea
779 	 * how to do this so we have to skip these buffers.
780 	 */
781 	if (bp->b_flags & B_NEEDCOMMIT) {
782 		BUF_UNLOCK(bp);
783 		return(0);
784 	}
785 
786 	/*
787 	 * Ask bioops if it is ok to sync
788 	 */
789 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
790 		bremfree(bp);
791 		brelse(bp);
792 		return(0);
793 	}
794 
795 	if (info->synchronous) {
796 		/*
797 		 * Synchronous flushing.  An error may be returned.
798 		 */
799 		bremfree(bp);
800 		error = bwrite(bp);
801 	} else {
802 		/*
803 		 * Asynchronous flushing.  A negative return value simply
804 		 * stops the scan and is not considered an error.  We use
805 		 * this to support limited MNT_LAZY flushes.
806 		 */
807 		vp->v_lazyw = bp->b_loffset;
808 		if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
809 			info->lazycount += vfs_bio_awrite(bp);
810 		} else {
811 			info->lazycount += bp->b_bufsize;
812 			bremfree(bp);
813 			bawrite(bp);
814 		}
815 		if (info->lazylimit && info->lazycount >= info->lazylimit)
816 			error = 1;
817 		else
818 			error = 0;
819 	}
820 	return(-error);
821 }
822 
823 /*
824  * Associate a buffer with a vnode.
825  *
826  * MPSAFE
827  */
828 int
829 bgetvp(struct vnode *vp, struct buf *bp)
830 {
831 	lwkt_tokref vlock;
832 
833 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
834 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
835 
836 	/*
837 	 * Insert onto list for new vnode.
838 	 */
839 	lwkt_gettoken(&vlock, &vp->v_token);
840 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
841 		lwkt_reltoken(&vlock);
842 		return (EEXIST);
843 	}
844 	bp->b_vp = vp;
845 	bp->b_flags |= B_HASHED;
846 	bp->b_flags |= B_VNCLEAN;
847 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
848 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
849 	vhold(vp);
850 	lwkt_reltoken(&vlock);
851 	return(0);
852 }
853 
854 /*
855  * Disassociate a buffer from a vnode.
856  */
857 void
858 brelvp(struct buf *bp)
859 {
860 	struct vnode *vp;
861 	lwkt_tokref vlock;
862 
863 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
864 
865 	/*
866 	 * Delete from old vnode list, if on one.
867 	 */
868 	vp = bp->b_vp;
869 	lwkt_gettoken(&vlock, &vp->v_token);
870 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
871 		if (bp->b_flags & B_VNDIRTY)
872 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
873 		else
874 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
875 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
876 	}
877 	if (bp->b_flags & B_HASHED) {
878 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
879 		bp->b_flags &= ~B_HASHED;
880 	}
881 	if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
882 		vp->v_flag &= ~VONWORKLST;
883 		LIST_REMOVE(vp, v_synclist);
884 	}
885 	bp->b_vp = NULL;
886 	lwkt_reltoken(&vlock);
887 
888 	vdrop(vp);
889 }
890 
891 /*
892  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
893  * This routine is called when the state of the B_DELWRI bit is changed.
894  *
895  * MPSAFE
896  */
897 void
898 reassignbuf(struct buf *bp)
899 {
900 	struct vnode *vp = bp->b_vp;
901 	lwkt_tokref vlock;
902 	int delay;
903 
904 	KKASSERT(vp != NULL);
905 	++reassignbufcalls;
906 
907 	/*
908 	 * B_PAGING flagged buffers cannot be reassigned because their vp
909 	 * is not fully linked in.
910 	 */
911 	if (bp->b_flags & B_PAGING)
912 		panic("cannot reassign paging buffer");
913 
914 	lwkt_gettoken(&vlock, &vp->v_token);
915 	if (bp->b_flags & B_DELWRI) {
916 		/*
917 		 * Move to the dirty list, add the vnode to the worklist
918 		 */
919 		if (bp->b_flags & B_VNCLEAN) {
920 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
921 			bp->b_flags &= ~B_VNCLEAN;
922 		}
923 		if ((bp->b_flags & B_VNDIRTY) == 0) {
924 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
925 				panic("reassignbuf: dup lblk vp %p bp %p",
926 				      vp, bp);
927 			}
928 			bp->b_flags |= B_VNDIRTY;
929 		}
930 		if ((vp->v_flag & VONWORKLST) == 0) {
931 			switch (vp->v_type) {
932 			case VDIR:
933 				delay = dirdelay;
934 				break;
935 			case VCHR:
936 			case VBLK:
937 				if (vp->v_rdev &&
938 				    vp->v_rdev->si_mountpoint != NULL) {
939 					delay = metadelay;
940 					break;
941 				}
942 				/* fall through */
943 			default:
944 				delay = filedelay;
945 			}
946 			vn_syncer_add_to_worklist(vp, delay);
947 		}
948 	} else {
949 		/*
950 		 * Move to the clean list, remove the vnode from the worklist
951 		 * if no dirty blocks remain.
952 		 */
953 		if (bp->b_flags & B_VNDIRTY) {
954 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
955 			bp->b_flags &= ~B_VNDIRTY;
956 		}
957 		if ((bp->b_flags & B_VNCLEAN) == 0) {
958 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
959 				panic("reassignbuf: dup lblk vp %p bp %p",
960 				      vp, bp);
961 			}
962 			bp->b_flags |= B_VNCLEAN;
963 		}
964 		if ((vp->v_flag & VONWORKLST) &&
965 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
966 			vp->v_flag &= ~VONWORKLST;
967 			LIST_REMOVE(vp, v_synclist);
968 		}
969 	}
970 	lwkt_reltoken(&vlock);
971 }
972 
973 /*
974  * Create a vnode for a block device.
975  * Used for mounting the root file system.
976  */
977 int
978 bdevvp(cdev_t dev, struct vnode **vpp)
979 {
980 	struct vnode *vp;
981 	struct vnode *nvp;
982 	int error;
983 
984 	if (dev == NULL) {
985 		*vpp = NULLVP;
986 		return (ENXIO);
987 	}
988 	error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0);
989 	if (error) {
990 		*vpp = NULLVP;
991 		return (error);
992 	}
993 	vp = nvp;
994 	vp->v_type = VCHR;
995 	vp->v_umajor = dev->si_umajor;
996 	vp->v_uminor = dev->si_uminor;
997 	vx_unlock(vp);
998 	*vpp = vp;
999 	return (0);
1000 }
1001 
1002 int
1003 v_associate_rdev(struct vnode *vp, cdev_t dev)
1004 {
1005 	lwkt_tokref ilock;
1006 
1007 	if (dev == NULL)
1008 		return(ENXIO);
1009 	if (dev_is_good(dev) == 0)
1010 		return(ENXIO);
1011 	KKASSERT(vp->v_rdev == NULL);
1012 	if (dev_ref_debug)
1013 		kprintf("Z1");
1014 	vp->v_rdev = reference_dev(dev);
1015 	lwkt_gettoken(&ilock, &spechash_token);
1016 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1017 	lwkt_reltoken(&ilock);
1018 	return(0);
1019 }
1020 
1021 void
1022 v_release_rdev(struct vnode *vp)
1023 {
1024 	lwkt_tokref ilock;
1025 	cdev_t dev;
1026 
1027 	if ((dev = vp->v_rdev) != NULL) {
1028 		lwkt_gettoken(&ilock, &spechash_token);
1029 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1030 		vp->v_rdev = NULL;
1031 		release_dev(dev);
1032 		lwkt_reltoken(&ilock);
1033 	}
1034 }
1035 
1036 /*
1037  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1038  * the device number with the vnode.  The actual device is not associated
1039  * until the vnode is opened (usually in spec_open()), and will be
1040  * disassociated on last close.
1041  */
1042 void
1043 addaliasu(struct vnode *nvp, int x, int y)
1044 {
1045 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1046 		panic("addaliasu on non-special vnode");
1047 	nvp->v_umajor = x;
1048 	nvp->v_uminor = y;
1049 }
1050 
1051 /*
1052  * Simple call that a filesystem can make to try to get rid of a
1053  * vnode.  It will fail if anyone is referencing the vnode (including
1054  * the caller).
1055  *
1056  * The filesystem can check whether its in-memory inode structure still
1057  * references the vp on return.
1058  */
1059 void
1060 vclean_unlocked(struct vnode *vp)
1061 {
1062 	vx_get(vp);
1063 	if (sysref_isactive(&vp->v_sysref) == 0)
1064 		vgone_vxlocked(vp);
1065 	vx_put(vp);
1066 }
1067 
1068 /*
1069  * Disassociate a vnode from its underlying filesystem.
1070  *
1071  * The vnode must be VX locked and referenced.  In all normal situations
1072  * there are no active references.  If vclean_vxlocked() is called while
1073  * there are active references, the vnode is being ripped out and we have
1074  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1075  */
1076 void
1077 vclean_vxlocked(struct vnode *vp, int flags)
1078 {
1079 	int active;
1080 	int n;
1081 	vm_object_t object;
1082 
1083 	/*
1084 	 * If the vnode has already been reclaimed we have nothing to do.
1085 	 */
1086 	if (vp->v_flag & VRECLAIMED)
1087 		return;
1088 	vp->v_flag |= VRECLAIMED;
1089 
1090 	/*
1091 	 * Scrap the vfs cache
1092 	 */
1093 	while (cache_inval_vp(vp, 0) != 0) {
1094 		kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1095 		tsleep(vp, 0, "vclninv", 2);
1096 	}
1097 
1098 	/*
1099 	 * Check to see if the vnode is in use. If so we have to reference it
1100 	 * before we clean it out so that its count cannot fall to zero and
1101 	 * generate a race against ourselves to recycle it.
1102 	 */
1103 	active = sysref_isactive(&vp->v_sysref);
1104 
1105 	/*
1106 	 * Clean out any buffers associated with the vnode and destroy its
1107 	 * object, if it has one.
1108 	 */
1109 	vinvalbuf(vp, V_SAVE, 0, 0);
1110 
1111 	/*
1112 	 * If purging an active vnode (typically during a forced unmount
1113 	 * or reboot), it must be closed and deactivated before being
1114 	 * reclaimed.  This isn't really all that safe, but what can
1115 	 * we do? XXX.
1116 	 *
1117 	 * Note that neither of these routines unlocks the vnode.
1118 	 */
1119 	if (active && (flags & DOCLOSE)) {
1120 		while ((n = vp->v_opencount) != 0) {
1121 			if (vp->v_writecount)
1122 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1123 			else
1124 				VOP_CLOSE(vp, FNONBLOCK);
1125 			if (vp->v_opencount == n) {
1126 				kprintf("Warning: unable to force-close"
1127 				       " vnode %p\n", vp);
1128 				break;
1129 			}
1130 		}
1131 	}
1132 
1133 	/*
1134 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1135 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1136 	 * again to make sure they all get flushed.
1137 	 *
1138 	 * This can occur if a file with a link count of 0 needs to be
1139 	 * truncated.
1140 	 */
1141 	if ((vp->v_flag & VINACTIVE) == 0) {
1142 		vp->v_flag |= VINACTIVE;
1143 		VOP_INACTIVE(vp);
1144 		vinvalbuf(vp, V_SAVE, 0, 0);
1145 	}
1146 
1147 	/*
1148 	 * If the vnode has an object, destroy it.
1149 	 */
1150 	if ((object = vp->v_object) != NULL) {
1151 		if (object->ref_count == 0) {
1152 			if ((object->flags & OBJ_DEAD) == 0)
1153 				vm_object_terminate(object);
1154 		} else {
1155 			vm_pager_deallocate(object);
1156 		}
1157 		vp->v_flag &= ~VOBJBUF;
1158 	}
1159 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1160 
1161 	/*
1162 	 * Reclaim the vnode.
1163 	 */
1164 	if (VOP_RECLAIM(vp))
1165 		panic("vclean: cannot reclaim");
1166 
1167 	/*
1168 	 * Done with purge, notify sleepers of the grim news.
1169 	 */
1170 	vp->v_ops = &dead_vnode_vops_p;
1171 	vn_pollgone(vp);
1172 	vp->v_tag = VT_NON;
1173 
1174 	/*
1175 	 * If we are destroying an active vnode, reactivate it now that
1176 	 * we have reassociated it with deadfs.  This prevents the system
1177 	 * from crashing on the vnode due to it being unexpectedly marked
1178 	 * as inactive or reclaimed.
1179 	 */
1180 	if (active && (flags & DOCLOSE)) {
1181 		vp->v_flag &= ~(VINACTIVE|VRECLAIMED);
1182 	}
1183 }
1184 
1185 /*
1186  * Eliminate all activity associated with the requested vnode
1187  * and with all vnodes aliased to the requested vnode.
1188  *
1189  * The vnode must be referenced but should not be locked.
1190  */
1191 int
1192 vrevoke(struct vnode *vp, struct ucred *cred)
1193 {
1194 	struct vnode *vq;
1195 	struct vnode *vqn;
1196 	lwkt_tokref ilock;
1197 	cdev_t dev;
1198 	int error;
1199 
1200 	/*
1201 	 * If the vnode has a device association, scrap all vnodes associated
1202 	 * with the device.  Don't let the device disappear on us while we
1203 	 * are scrapping the vnodes.
1204 	 *
1205 	 * The passed vp will probably show up in the list, do not VX lock
1206 	 * it twice!
1207 	 *
1208 	 * Releasing the vnode's rdev here can mess up specfs's call to
1209 	 * device close, so don't do it.  The vnode has been disassociated
1210 	 * and the device will be closed after the last ref on the related
1211 	 * fp goes away (if not still open by e.g. the kernel).
1212 	 */
1213 	if (vp->v_type != VCHR) {
1214 		error = fdrevoke(vp, DTYPE_VNODE, cred);
1215 		return (error);
1216 	}
1217 	if ((dev = vp->v_rdev) == NULL) {
1218 		if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL)
1219 			return(0);
1220 	}
1221 	reference_dev(dev);
1222 	lwkt_gettoken(&ilock, &spechash_token);
1223 
1224 	vqn = SLIST_FIRST(&dev->si_hlist);
1225 	if (vqn)
1226 		vref(vqn);
1227 	while ((vq = vqn) != NULL) {
1228 		vqn = SLIST_NEXT(vqn, v_cdevnext);
1229 		if (vqn)
1230 			vref(vqn);
1231 		fdrevoke(vq, DTYPE_VNODE, cred);
1232 		/*v_release_rdev(vq);*/
1233 		vrele(vq);
1234 	}
1235 	lwkt_reltoken(&ilock);
1236 	dev_drevoke(dev);
1237 	release_dev(dev);
1238 	return (0);
1239 }
1240 
1241 /*
1242  * This is called when the object underlying a vnode is being destroyed,
1243  * such as in a remove().  Try to recycle the vnode immediately if the
1244  * only active reference is our reference.
1245  *
1246  * Directory vnodes in the namecache with children cannot be immediately
1247  * recycled because numerous VOP_N*() ops require them to be stable.
1248  */
1249 int
1250 vrecycle(struct vnode *vp)
1251 {
1252 	if (vp->v_sysref.refcnt <= 1) {
1253 		if (cache_inval_vp_nonblock(vp))
1254 			return(0);
1255 		vgone_vxlocked(vp);
1256 		return (1);
1257 	}
1258 	return (0);
1259 }
1260 
1261 /*
1262  * Return the maximum I/O size allowed for strategy calls on VP.
1263  *
1264  * If vp is VCHR or VBLK we dive the device, otherwise we use
1265  * the vp's mount info.
1266  */
1267 int
1268 vmaxiosize(struct vnode *vp)
1269 {
1270 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1271 		return(vp->v_rdev->si_iosize_max);
1272 	} else {
1273 		return(vp->v_mount->mnt_iosize_max);
1274 	}
1275 }
1276 
1277 /*
1278  * Eliminate all activity associated with a vnode in preparation for reuse.
1279  *
1280  * The vnode must be VX locked and refd and will remain VX locked and refd
1281  * on return.  This routine may be called with the vnode in any state, as
1282  * long as it is VX locked.  The vnode will be cleaned out and marked
1283  * VRECLAIMED but will not actually be reused until all existing refs and
1284  * holds go away.
1285  *
1286  * NOTE: This routine may be called on a vnode which has not yet been
1287  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1288  * already been reclaimed.
1289  *
1290  * This routine is not responsible for placing us back on the freelist.
1291  * Instead, it happens automatically when the caller releases the VX lock
1292  * (assuming there aren't any other references).
1293  */
1294 
1295 void
1296 vgone_vxlocked(struct vnode *vp)
1297 {
1298 	/*
1299 	 * assert that the VX lock is held.  This is an absolute requirement
1300 	 * now for vgone_vxlocked() to be called.
1301 	 */
1302 	KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1303 
1304 	/*
1305 	 * Clean out the filesystem specific data and set the VRECLAIMED
1306 	 * bit.  Also deactivate the vnode if necessary.
1307 	 */
1308 	vclean_vxlocked(vp, DOCLOSE);
1309 
1310 	/*
1311 	 * Delete from old mount point vnode list, if on one.
1312 	 */
1313 	if (vp->v_mount != NULL)
1314 		insmntque(vp, NULL);
1315 
1316 	/*
1317 	 * If special device, remove it from special device alias list
1318 	 * if it is on one.  This should normally only occur if a vnode is
1319 	 * being revoked as the device should otherwise have been released
1320 	 * naturally.
1321 	 */
1322 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1323 		v_release_rdev(vp);
1324 	}
1325 
1326 	/*
1327 	 * Set us to VBAD
1328 	 */
1329 	vp->v_type = VBAD;
1330 }
1331 
1332 /*
1333  * Lookup a vnode by device number.
1334  *
1335  * Returns non-zero and *vpp set to a vref'd vnode on success.
1336  * Returns zero on failure.
1337  */
1338 int
1339 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1340 {
1341 	lwkt_tokref ilock;
1342 	struct vnode *vp;
1343 
1344 	lwkt_gettoken(&ilock, &spechash_token);
1345 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1346 		if (type == vp->v_type) {
1347 			*vpp = vp;
1348 			vref(vp);
1349 			lwkt_reltoken(&ilock);
1350 			return (1);
1351 		}
1352 	}
1353 	lwkt_reltoken(&ilock);
1354 	return (0);
1355 }
1356 
1357 /*
1358  * Calculate the total number of references to a special device.  This
1359  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1360  * an overloaded field.  Since udev2dev can now return NULL, we have
1361  * to check for a NULL v_rdev.
1362  */
1363 int
1364 count_dev(cdev_t dev)
1365 {
1366 	lwkt_tokref ilock;
1367 	struct vnode *vp;
1368 	int count = 0;
1369 
1370 	if (SLIST_FIRST(&dev->si_hlist)) {
1371 		lwkt_gettoken(&ilock, &spechash_token);
1372 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1373 			if (vp->v_sysref.refcnt > 0)
1374 				count += vp->v_sysref.refcnt;
1375 		}
1376 		lwkt_reltoken(&ilock);
1377 	}
1378 	return(count);
1379 }
1380 
1381 int
1382 count_udev(int x, int y)
1383 {
1384 	cdev_t dev;
1385 
1386 	if ((dev = get_dev(x, y)) == NULL)
1387 		return(0);
1388 	return(count_dev(dev));
1389 }
1390 
1391 int
1392 vcount(struct vnode *vp)
1393 {
1394 	if (vp->v_rdev == NULL)
1395 		return(0);
1396 	return(count_dev(vp->v_rdev));
1397 }
1398 
1399 /*
1400  * Initialize VMIO for a vnode.  This routine MUST be called before a
1401  * VFS can issue buffer cache ops on a vnode.  It is typically called
1402  * when a vnode is initialized from its inode.
1403  */
1404 int
1405 vinitvmio(struct vnode *vp, off_t filesize)
1406 {
1407 	vm_object_t object;
1408 	int error = 0;
1409 
1410 retry:
1411 	if ((object = vp->v_object) == NULL) {
1412 		object = vnode_pager_alloc(vp, filesize, 0, 0);
1413 		/*
1414 		 * Dereference the reference we just created.  This assumes
1415 		 * that the object is associated with the vp.
1416 		 */
1417 		object->ref_count--;
1418 		vrele(vp);
1419 	} else {
1420 		if (object->flags & OBJ_DEAD) {
1421 			vn_unlock(vp);
1422 			vm_object_dead_sleep(object, "vodead");
1423 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1424 			goto retry;
1425 		}
1426 	}
1427 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1428 	vp->v_flag |= VOBJBUF;
1429 	return (error);
1430 }
1431 
1432 
1433 /*
1434  * Print out a description of a vnode.
1435  */
1436 static char *typename[] =
1437 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1438 
1439 void
1440 vprint(char *label, struct vnode *vp)
1441 {
1442 	char buf[96];
1443 
1444 	if (label != NULL)
1445 		kprintf("%s: %p: ", label, (void *)vp);
1446 	else
1447 		kprintf("%p: ", (void *)vp);
1448 	kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1449 		typename[vp->v_type],
1450 		vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1451 	buf[0] = '\0';
1452 	if (vp->v_flag & VROOT)
1453 		strcat(buf, "|VROOT");
1454 	if (vp->v_flag & VPFSROOT)
1455 		strcat(buf, "|VPFSROOT");
1456 	if (vp->v_flag & VTEXT)
1457 		strcat(buf, "|VTEXT");
1458 	if (vp->v_flag & VSYSTEM)
1459 		strcat(buf, "|VSYSTEM");
1460 	if (vp->v_flag & VFREE)
1461 		strcat(buf, "|VFREE");
1462 	if (vp->v_flag & VOBJBUF)
1463 		strcat(buf, "|VOBJBUF");
1464 	if (buf[0] != '\0')
1465 		kprintf(" flags (%s)", &buf[1]);
1466 	if (vp->v_data == NULL) {
1467 		kprintf("\n");
1468 	} else {
1469 		kprintf("\n\t");
1470 		VOP_PRINT(vp);
1471 	}
1472 }
1473 
1474 #ifdef DDB
1475 #include <ddb/ddb.h>
1476 
1477 static int db_show_locked_vnodes(struct mount *mp, void *data);
1478 
1479 /*
1480  * List all of the locked vnodes in the system.
1481  * Called when debugging the kernel.
1482  */
1483 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1484 {
1485 	kprintf("Locked vnodes\n");
1486 	mountlist_scan(db_show_locked_vnodes, NULL,
1487 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1488 }
1489 
1490 static int
1491 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1492 {
1493 	struct vnode *vp;
1494 
1495 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1496 		if (vn_islocked(vp))
1497 			vprint(NULL, vp);
1498 	}
1499 	return(0);
1500 }
1501 #endif
1502 
1503 /*
1504  * Top level filesystem related information gathering.
1505  */
1506 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1507 
1508 static int
1509 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1510 {
1511 	int *name = (int *)arg1 - 1;	/* XXX */
1512 	u_int namelen = arg2 + 1;	/* XXX */
1513 	struct vfsconf *vfsp;
1514 	int maxtypenum;
1515 
1516 #if 1 || defined(COMPAT_PRELITE2)
1517 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1518 	if (namelen == 1)
1519 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1520 #endif
1521 
1522 #ifdef notyet
1523 	/* all sysctl names at this level are at least name and field */
1524 	if (namelen < 2)
1525 		return (ENOTDIR);		/* overloaded */
1526 	if (name[0] != VFS_GENERIC) {
1527 		vfsp = vfsconf_find_by_typenum(name[0]);
1528 		if (vfsp == NULL)
1529 			return (EOPNOTSUPP);
1530 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1531 		    oldp, oldlenp, newp, newlen, p));
1532 	}
1533 #endif
1534 	switch (name[1]) {
1535 	case VFS_MAXTYPENUM:
1536 		if (namelen != 2)
1537 			return (ENOTDIR);
1538 		maxtypenum = vfsconf_get_maxtypenum();
1539 		return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
1540 	case VFS_CONF:
1541 		if (namelen != 3)
1542 			return (ENOTDIR);	/* overloaded */
1543 		vfsp = vfsconf_find_by_typenum(name[2]);
1544 		if (vfsp == NULL)
1545 			return (EOPNOTSUPP);
1546 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1547 	}
1548 	return (EOPNOTSUPP);
1549 }
1550 
1551 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1552 	"Generic filesystem");
1553 
1554 #if 1 || defined(COMPAT_PRELITE2)
1555 
1556 static int
1557 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
1558 {
1559 	int error;
1560 	struct ovfsconf ovfs;
1561 	struct sysctl_req *req = (struct sysctl_req*) data;
1562 
1563 	bzero(&ovfs, sizeof(ovfs));
1564 	ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1565 	strcpy(ovfs.vfc_name, vfsp->vfc_name);
1566 	ovfs.vfc_index = vfsp->vfc_typenum;
1567 	ovfs.vfc_refcount = vfsp->vfc_refcount;
1568 	ovfs.vfc_flags = vfsp->vfc_flags;
1569 	error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1570 	if (error)
1571 		return error; /* abort iteration with error code */
1572 	else
1573 		return 0; /* continue iterating with next element */
1574 }
1575 
1576 static int
1577 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1578 {
1579 	return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
1580 }
1581 
1582 #endif /* 1 || COMPAT_PRELITE2 */
1583 
1584 /*
1585  * Check to see if a filesystem is mounted on a block device.
1586  */
1587 int
1588 vfs_mountedon(struct vnode *vp)
1589 {
1590 	cdev_t dev;
1591 
1592 	if ((dev = vp->v_rdev) == NULL) {
1593 		if (vp->v_type != VBLK)
1594 			dev = get_dev(vp->v_uminor, vp->v_umajor);
1595 	}
1596 	if (dev != NULL && dev->si_mountpoint)
1597 		return (EBUSY);
1598 	return (0);
1599 }
1600 
1601 /*
1602  * Unmount all filesystems. The list is traversed in reverse order
1603  * of mounting to avoid dependencies.
1604  */
1605 
1606 static int vfs_umountall_callback(struct mount *mp, void *data);
1607 
1608 void
1609 vfs_unmountall(void)
1610 {
1611 	int count;
1612 
1613 	do {
1614 		count = mountlist_scan(vfs_umountall_callback,
1615 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1616 	} while (count);
1617 }
1618 
1619 static
1620 int
1621 vfs_umountall_callback(struct mount *mp, void *data)
1622 {
1623 	int error;
1624 
1625 	error = dounmount(mp, MNT_FORCE);
1626 	if (error) {
1627 		mountlist_remove(mp);
1628 		kprintf("unmount of filesystem mounted from %s failed (",
1629 			mp->mnt_stat.f_mntfromname);
1630 		if (error == EBUSY)
1631 			kprintf("BUSY)\n");
1632 		else
1633 			kprintf("%d)\n", error);
1634 	}
1635 	return(1);
1636 }
1637 
1638 /*
1639  * Build hash lists of net addresses and hang them off the mount point.
1640  * Called by ufs_mount() to set up the lists of export addresses.
1641  */
1642 static int
1643 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1644 		const struct export_args *argp)
1645 {
1646 	struct netcred *np;
1647 	struct radix_node_head *rnh;
1648 	int i;
1649 	struct radix_node *rn;
1650 	struct sockaddr *saddr, *smask = 0;
1651 	struct domain *dom;
1652 	int error;
1653 
1654 	if (argp->ex_addrlen == 0) {
1655 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1656 			return (EPERM);
1657 		np = &nep->ne_defexported;
1658 		np->netc_exflags = argp->ex_flags;
1659 		np->netc_anon = argp->ex_anon;
1660 		np->netc_anon.cr_ref = 1;
1661 		mp->mnt_flag |= MNT_DEFEXPORTED;
1662 		return (0);
1663 	}
1664 
1665 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1666 		return (EINVAL);
1667 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1668 		return (EINVAL);
1669 
1670 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1671 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1672 	saddr = (struct sockaddr *) (np + 1);
1673 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1674 		goto out;
1675 	if (saddr->sa_len > argp->ex_addrlen)
1676 		saddr->sa_len = argp->ex_addrlen;
1677 	if (argp->ex_masklen) {
1678 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1679 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1680 		if (error)
1681 			goto out;
1682 		if (smask->sa_len > argp->ex_masklen)
1683 			smask->sa_len = argp->ex_masklen;
1684 	}
1685 	i = saddr->sa_family;
1686 	if ((rnh = nep->ne_rtable[i]) == 0) {
1687 		/*
1688 		 * Seems silly to initialize every AF when most are not used,
1689 		 * do so on demand here
1690 		 */
1691 		SLIST_FOREACH(dom, &domains, dom_next)
1692 			if (dom->dom_family == i && dom->dom_rtattach) {
1693 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1694 				    dom->dom_rtoffset);
1695 				break;
1696 			}
1697 		if ((rnh = nep->ne_rtable[i]) == 0) {
1698 			error = ENOBUFS;
1699 			goto out;
1700 		}
1701 	}
1702 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1703 	    np->netc_rnodes);
1704 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1705 		error = EPERM;
1706 		goto out;
1707 	}
1708 	np->netc_exflags = argp->ex_flags;
1709 	np->netc_anon = argp->ex_anon;
1710 	np->netc_anon.cr_ref = 1;
1711 	return (0);
1712 out:
1713 	kfree(np, M_NETADDR);
1714 	return (error);
1715 }
1716 
1717 /* ARGSUSED */
1718 static int
1719 vfs_free_netcred(struct radix_node *rn, void *w)
1720 {
1721 	struct radix_node_head *rnh = (struct radix_node_head *) w;
1722 
1723 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1724 	kfree((caddr_t) rn, M_NETADDR);
1725 	return (0);
1726 }
1727 
1728 /*
1729  * Free the net address hash lists that are hanging off the mount points.
1730  */
1731 static void
1732 vfs_free_addrlist(struct netexport *nep)
1733 {
1734 	int i;
1735 	struct radix_node_head *rnh;
1736 
1737 	for (i = 0; i <= AF_MAX; i++)
1738 		if ((rnh = nep->ne_rtable[i])) {
1739 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1740 			    (caddr_t) rnh);
1741 			kfree((caddr_t) rnh, M_RTABLE);
1742 			nep->ne_rtable[i] = 0;
1743 		}
1744 }
1745 
1746 int
1747 vfs_export(struct mount *mp, struct netexport *nep,
1748 	   const struct export_args *argp)
1749 {
1750 	int error;
1751 
1752 	if (argp->ex_flags & MNT_DELEXPORT) {
1753 		if (mp->mnt_flag & MNT_EXPUBLIC) {
1754 			vfs_setpublicfs(NULL, NULL, NULL);
1755 			mp->mnt_flag &= ~MNT_EXPUBLIC;
1756 		}
1757 		vfs_free_addrlist(nep);
1758 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1759 	}
1760 	if (argp->ex_flags & MNT_EXPORTED) {
1761 		if (argp->ex_flags & MNT_EXPUBLIC) {
1762 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1763 				return (error);
1764 			mp->mnt_flag |= MNT_EXPUBLIC;
1765 		}
1766 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1767 			return (error);
1768 		mp->mnt_flag |= MNT_EXPORTED;
1769 	}
1770 	return (0);
1771 }
1772 
1773 
1774 /*
1775  * Set the publicly exported filesystem (WebNFS). Currently, only
1776  * one public filesystem is possible in the spec (RFC 2054 and 2055)
1777  */
1778 int
1779 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1780 		const struct export_args *argp)
1781 {
1782 	int error;
1783 	struct vnode *rvp;
1784 	char *cp;
1785 
1786 	/*
1787 	 * mp == NULL -> invalidate the current info, the FS is
1788 	 * no longer exported. May be called from either vfs_export
1789 	 * or unmount, so check if it hasn't already been done.
1790 	 */
1791 	if (mp == NULL) {
1792 		if (nfs_pub.np_valid) {
1793 			nfs_pub.np_valid = 0;
1794 			if (nfs_pub.np_index != NULL) {
1795 				FREE(nfs_pub.np_index, M_TEMP);
1796 				nfs_pub.np_index = NULL;
1797 			}
1798 		}
1799 		return (0);
1800 	}
1801 
1802 	/*
1803 	 * Only one allowed at a time.
1804 	 */
1805 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1806 		return (EBUSY);
1807 
1808 	/*
1809 	 * Get real filehandle for root of exported FS.
1810 	 */
1811 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1812 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1813 
1814 	if ((error = VFS_ROOT(mp, &rvp)))
1815 		return (error);
1816 
1817 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1818 		return (error);
1819 
1820 	vput(rvp);
1821 
1822 	/*
1823 	 * If an indexfile was specified, pull it in.
1824 	 */
1825 	if (argp->ex_indexfile != NULL) {
1826 		int namelen;
1827 
1828 		error = vn_get_namelen(rvp, &namelen);
1829 		if (error)
1830 			return (error);
1831 		MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1832 		    M_WAITOK);
1833 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1834 		    namelen, NULL);
1835 		if (!error) {
1836 			/*
1837 			 * Check for illegal filenames.
1838 			 */
1839 			for (cp = nfs_pub.np_index; *cp; cp++) {
1840 				if (*cp == '/') {
1841 					error = EINVAL;
1842 					break;
1843 				}
1844 			}
1845 		}
1846 		if (error) {
1847 			FREE(nfs_pub.np_index, M_TEMP);
1848 			return (error);
1849 		}
1850 	}
1851 
1852 	nfs_pub.np_mount = mp;
1853 	nfs_pub.np_valid = 1;
1854 	return (0);
1855 }
1856 
1857 struct netcred *
1858 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1859 		struct sockaddr *nam)
1860 {
1861 	struct netcred *np;
1862 	struct radix_node_head *rnh;
1863 	struct sockaddr *saddr;
1864 
1865 	np = NULL;
1866 	if (mp->mnt_flag & MNT_EXPORTED) {
1867 		/*
1868 		 * Lookup in the export list first.
1869 		 */
1870 		if (nam != NULL) {
1871 			saddr = nam;
1872 			rnh = nep->ne_rtable[saddr->sa_family];
1873 			if (rnh != NULL) {
1874 				np = (struct netcred *)
1875 					(*rnh->rnh_matchaddr)((char *)saddr,
1876 							      rnh);
1877 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1878 					np = NULL;
1879 			}
1880 		}
1881 		/*
1882 		 * If no address match, use the default if it exists.
1883 		 */
1884 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1885 			np = &nep->ne_defexported;
1886 	}
1887 	return (np);
1888 }
1889 
1890 /*
1891  * perform msync on all vnodes under a mount point.  The mount point must
1892  * be locked.  This code is also responsible for lazy-freeing unreferenced
1893  * vnodes whos VM objects no longer contain pages.
1894  *
1895  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1896  *
1897  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1898  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
1899  * way up in this high level function.
1900  */
1901 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1902 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1903 
1904 void
1905 vfs_msync(struct mount *mp, int flags)
1906 {
1907 	int vmsc_flags;
1908 
1909 	vmsc_flags = VMSC_GETVP;
1910 	if (flags != MNT_WAIT)
1911 		vmsc_flags |= VMSC_NOWAIT;
1912 	vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1913 			(void *)(intptr_t)flags);
1914 }
1915 
1916 /*
1917  * scan1 is a fast pre-check.  There could be hundreds of thousands of
1918  * vnodes, we cannot afford to do anything heavy weight until we have a
1919  * fairly good indication that there is work to do.
1920  */
1921 static
1922 int
1923 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1924 {
1925 	int flags = (int)(intptr_t)data;
1926 
1927 	if ((vp->v_flag & VRECLAIMED) == 0) {
1928 		if (vshouldmsync(vp))
1929 			return(0);	/* call scan2 */
1930 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1931 		    (vp->v_flag & VOBJDIRTY) &&
1932 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
1933 			return(0);	/* call scan2 */
1934 		}
1935 	}
1936 
1937 	/*
1938 	 * do not call scan2, continue the loop
1939 	 */
1940 	return(-1);
1941 }
1942 
1943 /*
1944  * This callback is handed a locked vnode.
1945  */
1946 static
1947 int
1948 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1949 {
1950 	vm_object_t obj;
1951 	int flags = (int)(intptr_t)data;
1952 
1953 	if (vp->v_flag & VRECLAIMED)
1954 		return(0);
1955 
1956 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1957 		if ((obj = vp->v_object) != NULL) {
1958 			vm_object_page_clean(obj, 0, 0,
1959 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1960 		}
1961 	}
1962 	return(0);
1963 }
1964 
1965 /*
1966  * Record a process's interest in events which might happen to
1967  * a vnode.  Because poll uses the historic select-style interface
1968  * internally, this routine serves as both the ``check for any
1969  * pending events'' and the ``record my interest in future events''
1970  * functions.  (These are done together, while the lock is held,
1971  * to avoid race conditions.)
1972  */
1973 int
1974 vn_pollrecord(struct vnode *vp, int events)
1975 {
1976 	lwkt_tokref vlock;
1977 
1978 	KKASSERT(curthread->td_proc != NULL);
1979 
1980 	lwkt_gettoken(&vlock, &vp->v_token);
1981 	if (vp->v_pollinfo.vpi_revents & events) {
1982 		/*
1983 		 * This leaves events we are not interested
1984 		 * in available for the other process which
1985 		 * which presumably had requested them
1986 		 * (otherwise they would never have been
1987 		 * recorded).
1988 		 */
1989 		events &= vp->v_pollinfo.vpi_revents;
1990 		vp->v_pollinfo.vpi_revents &= ~events;
1991 
1992 		lwkt_reltoken(&vlock);
1993 		return events;
1994 	}
1995 	vp->v_pollinfo.vpi_events |= events;
1996 	selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
1997 	lwkt_reltoken(&vlock);
1998 	return 0;
1999 }
2000 
2001 /*
2002  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2003  * it is possible for us to miss an event due to race conditions, but
2004  * that condition is expected to be rare, so for the moment it is the
2005  * preferred interface.
2006  */
2007 void
2008 vn_pollevent(struct vnode *vp, int events)
2009 {
2010 	lwkt_tokref vlock;
2011 
2012 	lwkt_gettoken(&vlock, &vp->v_token);
2013 	if (vp->v_pollinfo.vpi_events & events) {
2014 		/*
2015 		 * We clear vpi_events so that we don't
2016 		 * call selwakeup() twice if two events are
2017 		 * posted before the polling process(es) is
2018 		 * awakened.  This also ensures that we take at
2019 		 * most one selwakeup() if the polling process
2020 		 * is no longer interested.  However, it does
2021 		 * mean that only one event can be noticed at
2022 		 * a time.  (Perhaps we should only clear those
2023 		 * event bits which we note?) XXX
2024 		 */
2025 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2026 		vp->v_pollinfo.vpi_revents |= events;
2027 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2028 	}
2029 	lwkt_reltoken(&vlock);
2030 }
2031 
2032 /*
2033  * Wake up anyone polling on vp because it is being revoked.
2034  * This depends on dead_poll() returning POLLHUP for correct
2035  * behavior.
2036  */
2037 void
2038 vn_pollgone(struct vnode *vp)
2039 {
2040 	lwkt_tokref vlock;
2041 
2042 	lwkt_gettoken(&vlock, &vp->v_token);
2043 	if (vp->v_pollinfo.vpi_events) {
2044 		vp->v_pollinfo.vpi_events = 0;
2045 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2046 	}
2047 	lwkt_reltoken(&vlock);
2048 }
2049 
2050 /*
2051  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2052  * (or v_rdev might be NULL).
2053  */
2054 cdev_t
2055 vn_todev(struct vnode *vp)
2056 {
2057 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2058 		return (NULL);
2059 	KKASSERT(vp->v_rdev != NULL);
2060 	return (vp->v_rdev);
2061 }
2062 
2063 /*
2064  * Check if vnode represents a disk device.  The vnode does not need to be
2065  * opened.
2066  *
2067  * MPALMOSTSAFE
2068  */
2069 int
2070 vn_isdisk(struct vnode *vp, int *errp)
2071 {
2072 	cdev_t dev;
2073 
2074 	if (vp->v_type != VCHR) {
2075 		if (errp != NULL)
2076 			*errp = ENOTBLK;
2077 		return (0);
2078 	}
2079 
2080 	if ((dev = vp->v_rdev) == NULL) {
2081 		get_mplock();
2082 		dev = get_dev(vp->v_umajor, vp->v_uminor);
2083 		rel_mplock();
2084 	}
2085 
2086 	if (dev == NULL) {
2087 		if (errp != NULL)
2088 			*errp = ENXIO;
2089 		return (0);
2090 	}
2091 	if (dev_is_good(dev) == 0) {
2092 		if (errp != NULL)
2093 			*errp = ENXIO;
2094 		return (0);
2095 	}
2096 	if ((dev_dflags(dev) & D_DISK) == 0) {
2097 		if (errp != NULL)
2098 			*errp = ENOTBLK;
2099 		return (0);
2100 	}
2101 	if (errp != NULL)
2102 		*errp = 0;
2103 	return (1);
2104 }
2105 
2106 int
2107 vn_get_namelen(struct vnode *vp, int *namelen)
2108 {
2109 	int error;
2110 	register_t retval[2];
2111 
2112 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2113 	if (error)
2114 		return (error);
2115 	*namelen = (int)retval[0];
2116 	return (0);
2117 }
2118 
2119 int
2120 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2121 		uint16_t d_namlen, const char *d_name)
2122 {
2123 	struct dirent *dp;
2124 	size_t len;
2125 
2126 	len = _DIRENT_RECLEN(d_namlen);
2127 	if (len > uio->uio_resid)
2128 		return(1);
2129 
2130 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2131 
2132 	dp->d_ino = d_ino;
2133 	dp->d_namlen = d_namlen;
2134 	dp->d_type = d_type;
2135 	bcopy(d_name, dp->d_name, d_namlen);
2136 
2137 	*error = uiomove((caddr_t)dp, len, uio);
2138 
2139 	kfree(dp, M_TEMP);
2140 
2141 	return(0);
2142 }
2143 
2144 void
2145 vn_mark_atime(struct vnode *vp, struct thread *td)
2146 {
2147 	struct proc *p = td->td_proc;
2148 	struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
2149 
2150 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
2151 		VOP_MARKATIME(vp, cred);
2152 	}
2153 }
2154