xref: /freebsd/sys/kern/vfs_default.c (revision dc60165b73e4c4d829a2cb9fed5cce585e93d9a9)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed
6  * to Berkeley by John Heidemann of the UCLA Ficus project.
7  *
8  * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/bio.h>
41 #include <sys/buf.h>
42 #include <sys/conf.h>
43 #include <sys/event.h>
44 #include <sys/kernel.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/lockf.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/fcntl.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <sys/dirent.h>
56 #include <sys/poll.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_extern.h>
61 #include <vm/pmap.h>
62 #include <vm/vm_map.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_pager.h>
65 #include <vm/vnode_pager.h>
66 
67 static int	vop_nolookup(struct vop_lookup_args *);
68 static int	vop_nostrategy(struct vop_strategy_args *);
69 static int	get_next_dirent(struct vnode *vp, struct dirent **dpp,
70 				char *dirbuf, int dirbuflen, off_t *off,
71 				char **cpos, int *len, int *eofflag,
72 				struct thread *td);
73 static int	dirent_exists(struct vnode *vp, const char *dirname,
74 			      struct thread *td);
75 
76 #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4)
77 
78 /*
79  * This vnode table stores what we want to do if the filesystem doesn't
80  * implement a particular VOP.
81  *
82  * If there is no specific entry here, we will return EOPNOTSUPP.
83  *
84  */
85 
86 struct vop_vector default_vnodeops = {
87 	.vop_default =		NULL,
88 	.vop_bypass =		VOP_EOPNOTSUPP,
89 
90 	.vop_advlock =		vop_stdadvlock,
91 	.vop_advlockasync =	vop_stdadvlockasync,
92 	.vop_bmap =		vop_stdbmap,
93 	.vop_close =		VOP_NULL,
94 	.vop_fsync =		VOP_NULL,
95 	.vop_getpages =		vop_stdgetpages,
96 	.vop_getwritemount = 	vop_stdgetwritemount,
97 	.vop_inactive =		VOP_NULL,
98 	.vop_ioctl =		VOP_ENOTTY,
99 	.vop_kqfilter =		vop_stdkqfilter,
100 	.vop_islocked =		vop_stdislocked,
101 	.vop_lease =		VOP_NULL,
102 	.vop_lock1 =		vop_stdlock,
103 	.vop_lookup =		vop_nolookup,
104 	.vop_open =		VOP_NULL,
105 	.vop_pathconf =		VOP_EINVAL,
106 	.vop_poll =		vop_nopoll,
107 	.vop_putpages =		vop_stdputpages,
108 	.vop_readlink =		VOP_EINVAL,
109 	.vop_revoke =		VOP_PANIC,
110 	.vop_strategy =		vop_nostrategy,
111 	.vop_unlock =		vop_stdunlock,
112 	.vop_vptocnp =		vop_stdvptocnp,
113 	.vop_vptofh =		vop_stdvptofh,
114 };
115 
116 /*
117  * Series of placeholder functions for various error returns for
118  * VOPs.
119  */
120 
121 int
122 vop_eopnotsupp(struct vop_generic_args *ap)
123 {
124 	/*
125 	printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name);
126 	*/
127 
128 	return (EOPNOTSUPP);
129 }
130 
131 int
132 vop_ebadf(struct vop_generic_args *ap)
133 {
134 
135 	return (EBADF);
136 }
137 
138 int
139 vop_enotty(struct vop_generic_args *ap)
140 {
141 
142 	return (ENOTTY);
143 }
144 
145 int
146 vop_einval(struct vop_generic_args *ap)
147 {
148 
149 	return (EINVAL);
150 }
151 
152 int
153 vop_enoent(struct vop_generic_args *ap)
154 {
155 
156 	return (ENOENT);
157 }
158 
159 int
160 vop_null(struct vop_generic_args *ap)
161 {
162 
163 	return (0);
164 }
165 
166 /*
167  * Helper function to panic on some bad VOPs in some filesystems.
168  */
169 int
170 vop_panic(struct vop_generic_args *ap)
171 {
172 
173 	panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name);
174 }
175 
176 /*
177  * vop_std<something> and vop_no<something> are default functions for use by
178  * filesystems that need the "default reasonable" implementation for a
179  * particular operation.
180  *
181  * The documentation for the operations they implement exists (if it exists)
182  * in the VOP_<SOMETHING>(9) manpage (all uppercase).
183  */
184 
185 /*
186  * Default vop for filesystems that do not support name lookup
187  */
188 static int
189 vop_nolookup(ap)
190 	struct vop_lookup_args /* {
191 		struct vnode *a_dvp;
192 		struct vnode **a_vpp;
193 		struct componentname *a_cnp;
194 	} */ *ap;
195 {
196 
197 	*ap->a_vpp = NULL;
198 	return (ENOTDIR);
199 }
200 
201 /*
202  *	vop_nostrategy:
203  *
204  *	Strategy routine for VFS devices that have none.
205  *
206  *	BIO_ERROR and B_INVAL must be cleared prior to calling any strategy
207  *	routine.  Typically this is done for a BIO_READ strategy call.
208  *	Typically B_INVAL is assumed to already be clear prior to a write
209  *	and should not be cleared manually unless you just made the buffer
210  *	invalid.  BIO_ERROR should be cleared either way.
211  */
212 
213 static int
214 vop_nostrategy (struct vop_strategy_args *ap)
215 {
216 	printf("No strategy for buffer at %p\n", ap->a_bp);
217 	vprint("vnode", ap->a_vp);
218 	ap->a_bp->b_ioflags |= BIO_ERROR;
219 	ap->a_bp->b_error = EOPNOTSUPP;
220 	bufdone(ap->a_bp);
221 	return (EOPNOTSUPP);
222 }
223 
224 static int
225 get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf,
226 		int dirbuflen, off_t *off, char **cpos, int *len,
227 		int *eofflag, struct thread *td)
228 {
229 	int error, reclen;
230 	struct uio uio;
231 	struct iovec iov;
232 	struct dirent *dp;
233 
234 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
235 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
236 
237 	if (*len == 0) {
238 		iov.iov_base = dirbuf;
239 		iov.iov_len = dirbuflen;
240 
241 		uio.uio_iov = &iov;
242 		uio.uio_iovcnt = 1;
243 		uio.uio_offset = *off;
244 		uio.uio_resid = dirbuflen;
245 		uio.uio_segflg = UIO_SYSSPACE;
246 		uio.uio_rw = UIO_READ;
247 		uio.uio_td = td;
248 
249 		*eofflag = 0;
250 
251 #ifdef MAC
252 		error = mac_vnode_check_readdir(td->td_ucred, vp);
253 		if (error == 0)
254 #endif
255 			error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag,
256 		    		NULL, NULL);
257 		if (error)
258 			return (error);
259 
260 		*off = uio.uio_offset;
261 
262 		*cpos = dirbuf;
263 		*len = (dirbuflen - uio.uio_resid);
264 	}
265 
266 	dp = (struct dirent *)(*cpos);
267 	reclen = dp->d_reclen;
268 	*dpp = dp;
269 
270 	/* check for malformed directory.. */
271 	if (reclen < DIRENT_MINSIZE)
272 		return (EINVAL);
273 
274 	*cpos += reclen;
275 	*len -= reclen;
276 
277 	return (0);
278 }
279 
280 /*
281  * Check if a named file exists in a given directory vnode.
282  */
283 static int
284 dirent_exists(struct vnode *vp, const char *dirname, struct thread *td)
285 {
286 	char *dirbuf, *cpos;
287 	int error, eofflag, dirbuflen, len, found;
288 	off_t off;
289 	struct dirent *dp;
290 	struct vattr va;
291 
292 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
293 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
294 
295 	found = 0;
296 
297 	error = VOP_GETATTR(vp, &va, td->td_ucred);
298 	if (error)
299 		return (found);
300 
301 	dirbuflen = DEV_BSIZE;
302 	if (dirbuflen < va.va_blocksize)
303 		dirbuflen = va.va_blocksize;
304 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
305 
306 	off = 0;
307 	len = 0;
308 	do {
309 		error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off,
310 					&cpos, &len, &eofflag, td);
311 		if (error)
312 			goto out;
313 
314 		if ((dp->d_type != DT_WHT) &&
315 		    !strcmp(dp->d_name, dirname)) {
316 			found = 1;
317 			goto out;
318 		}
319 	} while (len > 0 || !eofflag);
320 
321 out:
322 	free(dirbuf, M_TEMP);
323 	return (found);
324 }
325 
326 /*
327  * Advisory record locking support
328  */
329 int
330 vop_stdadvlock(struct vop_advlock_args *ap)
331 {
332 	struct vnode *vp;
333 	struct ucred *cred;
334 	struct vattr vattr;
335 	int error;
336 
337 	vp = ap->a_vp;
338 	cred = curthread->td_ucred;
339 	vn_lock(vp, LK_SHARED | LK_RETRY);
340 	error = VOP_GETATTR(vp, &vattr, cred);
341 	VOP_UNLOCK(vp, 0);
342 	if (error)
343 		return (error);
344 
345 	return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size));
346 }
347 
348 int
349 vop_stdadvlockasync(struct vop_advlockasync_args *ap)
350 {
351 	struct vnode *vp;
352 	struct ucred *cred;
353 	struct vattr vattr;
354 	int error;
355 
356 	vp = ap->a_vp;
357 	cred = curthread->td_ucred;
358 	vn_lock(vp, LK_SHARED | LK_RETRY);
359 	error = VOP_GETATTR(vp, &vattr, cred);
360 	VOP_UNLOCK(vp, 0);
361 	if (error)
362 		return (error);
363 
364 	return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size));
365 }
366 
367 /*
368  * vop_stdpathconf:
369  *
370  * Standard implementation of POSIX pathconf, to get information about limits
371  * for a filesystem.
372  * Override per filesystem for the case where the filesystem has smaller
373  * limits.
374  */
375 int
376 vop_stdpathconf(ap)
377 	struct vop_pathconf_args /* {
378 	struct vnode *a_vp;
379 	int a_name;
380 	int *a_retval;
381 	} */ *ap;
382 {
383 
384 	switch (ap->a_name) {
385 		case _PC_NAME_MAX:
386 			*ap->a_retval = NAME_MAX;
387 			return (0);
388 		case _PC_PATH_MAX:
389 			*ap->a_retval = PATH_MAX;
390 			return (0);
391 		case _PC_LINK_MAX:
392 			*ap->a_retval = LINK_MAX;
393 			return (0);
394 		case _PC_MAX_CANON:
395 			*ap->a_retval = MAX_CANON;
396 			return (0);
397 		case _PC_MAX_INPUT:
398 			*ap->a_retval = MAX_INPUT;
399 			return (0);
400 		case _PC_PIPE_BUF:
401 			*ap->a_retval = PIPE_BUF;
402 			return (0);
403 		case _PC_CHOWN_RESTRICTED:
404 			*ap->a_retval = 1;
405 			return (0);
406 		case _PC_VDISABLE:
407 			*ap->a_retval = _POSIX_VDISABLE;
408 			return (0);
409 		default:
410 			return (EINVAL);
411 	}
412 	/* NOTREACHED */
413 }
414 
415 /*
416  * Standard lock, unlock and islocked functions.
417  */
418 int
419 vop_stdlock(ap)
420 	struct vop_lock1_args /* {
421 		struct vnode *a_vp;
422 		int a_flags;
423 		char *file;
424 		int line;
425 	} */ *ap;
426 {
427 	struct vnode *vp = ap->a_vp;
428 
429 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
430 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
431 	    ap->a_line));
432 }
433 
434 /* See above. */
435 int
436 vop_stdunlock(ap)
437 	struct vop_unlock_args /* {
438 		struct vnode *a_vp;
439 		int a_flags;
440 	} */ *ap;
441 {
442 	struct vnode *vp = ap->a_vp;
443 
444 	return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp)));
445 }
446 
447 /* See above. */
448 int
449 vop_stdislocked(ap)
450 	struct vop_islocked_args /* {
451 		struct vnode *a_vp;
452 	} */ *ap;
453 {
454 
455 	return (lockstatus(ap->a_vp->v_vnlock));
456 }
457 
458 /*
459  * Return true for select/poll.
460  */
461 int
462 vop_nopoll(ap)
463 	struct vop_poll_args /* {
464 		struct vnode *a_vp;
465 		int  a_events;
466 		struct ucred *a_cred;
467 		struct thread *a_td;
468 	} */ *ap;
469 {
470 
471 	return (poll_no_poll(ap->a_events));
472 }
473 
474 /*
475  * Implement poll for local filesystems that support it.
476  */
477 int
478 vop_stdpoll(ap)
479 	struct vop_poll_args /* {
480 		struct vnode *a_vp;
481 		int  a_events;
482 		struct ucred *a_cred;
483 		struct thread *a_td;
484 	} */ *ap;
485 {
486 	if (ap->a_events & ~POLLSTANDARD)
487 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
488 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
489 }
490 
491 /*
492  * Return our mount point, as we will take charge of the writes.
493  */
494 int
495 vop_stdgetwritemount(ap)
496 	struct vop_getwritemount_args /* {
497 		struct vnode *a_vp;
498 		struct mount **a_mpp;
499 	} */ *ap;
500 {
501 	struct mount *mp;
502 
503 	/*
504 	 * XXX Since this is called unlocked we may be recycled while
505 	 * attempting to ref the mount.  If this is the case or mountpoint
506 	 * will be set to NULL.  We only have to prevent this call from
507 	 * returning with a ref to an incorrect mountpoint.  It is not
508 	 * harmful to return with a ref to our previous mountpoint.
509 	 */
510 	mp = ap->a_vp->v_mount;
511 	if (mp != NULL) {
512 		vfs_ref(mp);
513 		if (mp != ap->a_vp->v_mount) {
514 			vfs_rel(mp);
515 			mp = NULL;
516 		}
517 	}
518 	*(ap->a_mpp) = mp;
519 	return (0);
520 }
521 
522 /* XXX Needs good comment and VOP_BMAP(9) manpage */
523 int
524 vop_stdbmap(ap)
525 	struct vop_bmap_args /* {
526 		struct vnode *a_vp;
527 		daddr_t  a_bn;
528 		struct bufobj **a_bop;
529 		daddr_t *a_bnp;
530 		int *a_runp;
531 		int *a_runb;
532 	} */ *ap;
533 {
534 
535 	if (ap->a_bop != NULL)
536 		*ap->a_bop = &ap->a_vp->v_bufobj;
537 	if (ap->a_bnp != NULL)
538 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
539 	if (ap->a_runp != NULL)
540 		*ap->a_runp = 0;
541 	if (ap->a_runb != NULL)
542 		*ap->a_runb = 0;
543 	return (0);
544 }
545 
546 int
547 vop_stdfsync(ap)
548 	struct vop_fsync_args /* {
549 		struct vnode *a_vp;
550 		struct ucred *a_cred;
551 		int a_waitfor;
552 		struct thread *a_td;
553 	} */ *ap;
554 {
555 	struct vnode *vp = ap->a_vp;
556 	struct buf *bp;
557 	struct bufobj *bo;
558 	struct buf *nbp;
559 	int error = 0;
560 	int maxretry = 1000;     /* large, arbitrarily chosen */
561 
562 	bo = &vp->v_bufobj;
563 	BO_LOCK(bo);
564 loop1:
565 	/*
566 	 * MARK/SCAN initialization to avoid infinite loops.
567 	 */
568         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
569                 bp->b_vflags &= ~BV_SCANNED;
570 		bp->b_error = 0;
571 	}
572 
573 	/*
574 	 * Flush all dirty buffers associated with a vnode.
575 	 */
576 loop2:
577 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
578 		if ((bp->b_vflags & BV_SCANNED) != 0)
579 			continue;
580 		bp->b_vflags |= BV_SCANNED;
581 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL))
582 			continue;
583 		BO_UNLOCK(bo);
584 		KASSERT(bp->b_bufobj == bo,
585 		    ("bp %p wrong b_bufobj %p should be %p",
586 		    bp, bp->b_bufobj, bo));
587 		if ((bp->b_flags & B_DELWRI) == 0)
588 			panic("fsync: not dirty");
589 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
590 			vfs_bio_awrite(bp);
591 		} else {
592 			bremfree(bp);
593 			bawrite(bp);
594 		}
595 		BO_LOCK(bo);
596 		goto loop2;
597 	}
598 
599 	/*
600 	 * If synchronous the caller expects us to completely resolve all
601 	 * dirty buffers in the system.  Wait for in-progress I/O to
602 	 * complete (which could include background bitmap writes), then
603 	 * retry if dirty blocks still exist.
604 	 */
605 	if (ap->a_waitfor == MNT_WAIT) {
606 		bufobj_wwait(bo, 0, 0);
607 		if (bo->bo_dirty.bv_cnt > 0) {
608 			/*
609 			 * If we are unable to write any of these buffers
610 			 * then we fail now rather than trying endlessly
611 			 * to write them out.
612 			 */
613 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
614 				if ((error = bp->b_error) == 0)
615 					continue;
616 			if (error == 0 && --maxretry >= 0)
617 				goto loop1;
618 			error = EAGAIN;
619 		}
620 	}
621 	BO_UNLOCK(bo);
622 	if (error == EAGAIN)
623 		vprint("fsync: giving up on dirty", vp);
624 
625 	return (error);
626 }
627 
628 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
629 int
630 vop_stdgetpages(ap)
631 	struct vop_getpages_args /* {
632 		struct vnode *a_vp;
633 		vm_page_t *a_m;
634 		int a_count;
635 		int a_reqpage;
636 		vm_ooffset_t a_offset;
637 	} */ *ap;
638 {
639 
640 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
641 	    ap->a_count, ap->a_reqpage);
642 }
643 
644 int
645 vop_stdkqfilter(struct vop_kqfilter_args *ap)
646 {
647 	return vfs_kqfilter(ap);
648 }
649 
650 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
651 int
652 vop_stdputpages(ap)
653 	struct vop_putpages_args /* {
654 		struct vnode *a_vp;
655 		vm_page_t *a_m;
656 		int a_count;
657 		int a_sync;
658 		int *a_rtvals;
659 		vm_ooffset_t a_offset;
660 	} */ *ap;
661 {
662 
663 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
664 	     ap->a_sync, ap->a_rtvals);
665 }
666 
667 int
668 vop_stdvptofh(struct vop_vptofh_args *ap)
669 {
670 	return (EOPNOTSUPP);
671 }
672 
673 int
674 vop_stdvptocnp(struct vop_vptocnp_args *ap)
675 {
676 	struct vnode *vp = ap->a_vp;
677 	struct vnode **dvp = ap->a_vpp;
678 	char *buf = ap->a_buf;
679 	int *buflen = ap->a_buflen;
680 	char *dirbuf, *cpos;
681 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
682 	off_t off;
683 	ino_t fileno;
684 	struct vattr va;
685 	struct nameidata nd;
686 	struct thread *td;
687 	struct dirent *dp;
688 	struct vnode *mvp;
689 
690 	i = *buflen;
691 	error = 0;
692 	covered = 0;
693 	td = curthread;
694 
695 	if (vp->v_type != VDIR)
696 		return (ENOENT);
697 
698 	error = VOP_GETATTR(vp, &va, td->td_ucred);
699 	if (error)
700 		return (error);
701 
702 	VREF(vp);
703 	locked = VOP_ISLOCKED(vp);
704 	VOP_UNLOCK(vp, 0);
705 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE,
706 	    "..", vp, td);
707 	flags = FREAD;
708 	error = vn_open(&nd, &flags, 0, NULL);
709 	if (error) {
710 		vn_lock(vp, locked | LK_RETRY);
711 		return (error);
712 	}
713 	NDFREE(&nd, NDF_ONLY_PNBUF);
714 
715 	mvp = *dvp = nd.ni_vp;
716 
717 	if (vp->v_mount != (*dvp)->v_mount &&
718 	    ((*dvp)->v_vflag & VV_ROOT) &&
719 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
720 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
721 		VREF(mvp);
722 		VOP_UNLOCK(mvp, 0);
723 		vn_close(mvp, FREAD, td->td_ucred, td);
724 		VREF(*dvp);
725 		vn_lock(*dvp, LK_EXCLUSIVE | LK_RETRY);
726 		covered = 1;
727 	}
728 
729 	fileno = va.va_fileid;
730 
731 	dirbuflen = DEV_BSIZE;
732 	if (dirbuflen < va.va_blocksize)
733 		dirbuflen = va.va_blocksize;
734 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
735 
736 	if ((*dvp)->v_type != VDIR) {
737 		error = ENOENT;
738 		goto out;
739 	}
740 
741 	off = 0;
742 	len = 0;
743 	do {
744 		/* call VOP_READDIR of parent */
745 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
746 					&cpos, &len, &eofflag, td);
747 		if (error)
748 			goto out;
749 
750 		if ((dp->d_type != DT_WHT) &&
751 		    (dp->d_fileno == fileno)) {
752 			if (covered) {
753 				VOP_UNLOCK(*dvp, 0);
754 				vn_lock(mvp, LK_EXCLUSIVE | LK_RETRY);
755 				if (dirent_exists(mvp, dp->d_name, td)) {
756 					error = ENOENT;
757 					VOP_UNLOCK(mvp, 0);
758 					vn_lock(*dvp, LK_EXCLUSIVE | LK_RETRY);
759 					goto out;
760 				}
761 				VOP_UNLOCK(mvp, 0);
762 				vn_lock(*dvp, LK_EXCLUSIVE | LK_RETRY);
763 			}
764 			i -= dp->d_namlen;
765 
766 			if (i < 0) {
767 				error = ENOMEM;
768 				goto out;
769 			}
770 			bcopy(dp->d_name, buf + i, dp->d_namlen);
771 			error = 0;
772 			goto out;
773 		}
774 	} while (len > 0 || !eofflag);
775 	error = ENOENT;
776 
777 out:
778 	free(dirbuf, M_TEMP);
779 	if (!error) {
780 		*buflen = i;
781 		vhold(*dvp);
782 	}
783 	if (covered) {
784 		vput(*dvp);
785 		vrele(mvp);
786 	} else {
787 		VOP_UNLOCK(mvp, 0);
788 		vn_close(mvp, FREAD, td->td_ucred, td);
789 	}
790 	vn_lock(vp, locked | LK_RETRY);
791 	return (error);
792 }
793 
794 /*
795  * vfs default ops
796  * used to fill the vfs function table to get reasonable default return values.
797  */
798 int
799 vfs_stdroot (mp, flags, vpp, td)
800 	struct mount *mp;
801 	int flags;
802 	struct vnode **vpp;
803 	struct thread *td;
804 {
805 
806 	return (EOPNOTSUPP);
807 }
808 
809 int
810 vfs_stdstatfs (mp, sbp, td)
811 	struct mount *mp;
812 	struct statfs *sbp;
813 	struct thread *td;
814 {
815 
816 	return (EOPNOTSUPP);
817 }
818 
819 int
820 vfs_stdquotactl (mp, cmds, uid, arg, td)
821 	struct mount *mp;
822 	int cmds;
823 	uid_t uid;
824 	void *arg;
825 	struct thread *td;
826 {
827 
828 	return (EOPNOTSUPP);
829 }
830 
831 int
832 vfs_stdsync(mp, waitfor, td)
833 	struct mount *mp;
834 	int waitfor;
835 	struct thread *td;
836 {
837 	struct vnode *vp, *mvp;
838 	int error, lockreq, allerror = 0;
839 
840 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
841 	if (waitfor != MNT_WAIT)
842 		lockreq |= LK_NOWAIT;
843 	/*
844 	 * Force stale buffer cache information to be flushed.
845 	 */
846 	MNT_ILOCK(mp);
847 loop:
848 	MNT_VNODE_FOREACH(vp, mp, mvp) {
849 		/* bv_cnt is an acceptable race here. */
850 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0)
851 			continue;
852 		VI_LOCK(vp);
853 		MNT_IUNLOCK(mp);
854 		if ((error = vget(vp, lockreq, td)) != 0) {
855 			MNT_ILOCK(mp);
856 			if (error == ENOENT) {
857 				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
858 				goto loop;
859 			}
860 			continue;
861 		}
862 		error = VOP_FSYNC(vp, waitfor, td);
863 		if (error)
864 			allerror = error;
865 
866 		/* Do not turn this into vput.  td is not always curthread. */
867 		VOP_UNLOCK(vp, 0);
868 		vrele(vp);
869 		MNT_ILOCK(mp);
870 	}
871 	MNT_IUNLOCK(mp);
872 	return (allerror);
873 }
874 
875 int
876 vfs_stdnosync (mp, waitfor, td)
877 	struct mount *mp;
878 	int waitfor;
879 	struct thread *td;
880 {
881 
882 	return (0);
883 }
884 
885 int
886 vfs_stdvget (mp, ino, flags, vpp)
887 	struct mount *mp;
888 	ino_t ino;
889 	int flags;
890 	struct vnode **vpp;
891 {
892 
893 	return (EOPNOTSUPP);
894 }
895 
896 int
897 vfs_stdfhtovp (mp, fhp, vpp)
898 	struct mount *mp;
899 	struct fid *fhp;
900 	struct vnode **vpp;
901 {
902 
903 	return (EOPNOTSUPP);
904 }
905 
906 int
907 vfs_stdinit (vfsp)
908 	struct vfsconf *vfsp;
909 {
910 
911 	return (0);
912 }
913 
914 int
915 vfs_stduninit (vfsp)
916 	struct vfsconf *vfsp;
917 {
918 
919 	return(0);
920 }
921 
922 int
923 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname, td)
924 	struct mount *mp;
925 	int cmd;
926 	struct vnode *filename_vp;
927 	int attrnamespace;
928 	const char *attrname;
929 	struct thread *td;
930 {
931 
932 	if (filename_vp != NULL)
933 		VOP_UNLOCK(filename_vp, 0);
934 	return (EOPNOTSUPP);
935 }
936 
937 int
938 vfs_stdsysctl(mp, op, req)
939 	struct mount *mp;
940 	fsctlop_t op;
941 	struct sysctl_req *req;
942 {
943 
944 	return (EOPNOTSUPP);
945 }
946 
947 /* end of vfs default ops */
948