xref: /freebsd/sys/kern/vfs_default.c (revision b2d48be1bc7df45ddd13b143a160d0acb5a383c5)
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/namei.h>
51 #include <sys/rwlock.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 <security/mac/mac_framework.h>
59 
60 #include <vm/vm.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_extern.h>
63 #include <vm/pmap.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_page.h>
66 #include <vm/vm_pager.h>
67 #include <vm/vnode_pager.h>
68 
69 static int	vop_nolookup(struct vop_lookup_args *);
70 static int	vop_norename(struct vop_rename_args *);
71 static int	vop_nostrategy(struct vop_strategy_args *);
72 static int	get_next_dirent(struct vnode *vp, struct dirent **dpp,
73 				char *dirbuf, int dirbuflen, off_t *off,
74 				char **cpos, int *len, int *eofflag,
75 				struct thread *td);
76 static int	dirent_exists(struct vnode *vp, const char *dirname,
77 			      struct thread *td);
78 
79 #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4)
80 
81 static int vop_stdis_text(struct vop_is_text_args *ap);
82 static int vop_stdset_text(struct vop_set_text_args *ap);
83 static int vop_stdunset_text(struct vop_unset_text_args *ap);
84 static int vop_stdget_writecount(struct vop_get_writecount_args *ap);
85 static int vop_stdadd_writecount(struct vop_add_writecount_args *ap);
86 static int vop_stdgetpages_async(struct vop_getpages_async_args *ap);
87 
88 /*
89  * This vnode table stores what we want to do if the filesystem doesn't
90  * implement a particular VOP.
91  *
92  * If there is no specific entry here, we will return EOPNOTSUPP.
93  *
94  * Note that every filesystem has to implement either vop_access
95  * or vop_accessx; failing to do so will result in immediate crash
96  * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(),
97  * which calls vop_stdaccess() etc.
98  */
99 
100 struct vop_vector default_vnodeops = {
101 	.vop_default =		NULL,
102 	.vop_bypass =		VOP_EOPNOTSUPP,
103 
104 	.vop_access =		vop_stdaccess,
105 	.vop_accessx =		vop_stdaccessx,
106 	.vop_advise =		vop_stdadvise,
107 	.vop_advlock =		vop_stdadvlock,
108 	.vop_advlockasync =	vop_stdadvlockasync,
109 	.vop_advlockpurge =	vop_stdadvlockpurge,
110 	.vop_allocate =		vop_stdallocate,
111 	.vop_bmap =		vop_stdbmap,
112 	.vop_close =		VOP_NULL,
113 	.vop_fsync =		VOP_NULL,
114 	.vop_getpages =		vop_stdgetpages,
115 	.vop_getpages_async =	vop_stdgetpages_async,
116 	.vop_getwritemount = 	vop_stdgetwritemount,
117 	.vop_inactive =		VOP_NULL,
118 	.vop_ioctl =		VOP_ENOTTY,
119 	.vop_kqfilter =		vop_stdkqfilter,
120 	.vop_islocked =		vop_stdislocked,
121 	.vop_lock1 =		vop_stdlock,
122 	.vop_lookup =		vop_nolookup,
123 	.vop_open =		VOP_NULL,
124 	.vop_pathconf =		VOP_EINVAL,
125 	.vop_poll =		vop_nopoll,
126 	.vop_putpages =		vop_stdputpages,
127 	.vop_readlink =		VOP_EINVAL,
128 	.vop_rename =		vop_norename,
129 	.vop_revoke =		VOP_PANIC,
130 	.vop_strategy =		vop_nostrategy,
131 	.vop_unlock =		vop_stdunlock,
132 	.vop_vptocnp =		vop_stdvptocnp,
133 	.vop_vptofh =		vop_stdvptofh,
134 	.vop_unp_bind =		vop_stdunp_bind,
135 	.vop_unp_connect =	vop_stdunp_connect,
136 	.vop_unp_detach =	vop_stdunp_detach,
137 	.vop_is_text =		vop_stdis_text,
138 	.vop_set_text =		vop_stdset_text,
139 	.vop_unset_text =	vop_stdunset_text,
140 	.vop_get_writecount =	vop_stdget_writecount,
141 	.vop_add_writecount =	vop_stdadd_writecount,
142 };
143 
144 /*
145  * Series of placeholder functions for various error returns for
146  * VOPs.
147  */
148 
149 int
150 vop_eopnotsupp(struct vop_generic_args *ap)
151 {
152 	/*
153 	printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name);
154 	*/
155 
156 	return (EOPNOTSUPP);
157 }
158 
159 int
160 vop_ebadf(struct vop_generic_args *ap)
161 {
162 
163 	return (EBADF);
164 }
165 
166 int
167 vop_enotty(struct vop_generic_args *ap)
168 {
169 
170 	return (ENOTTY);
171 }
172 
173 int
174 vop_einval(struct vop_generic_args *ap)
175 {
176 
177 	return (EINVAL);
178 }
179 
180 int
181 vop_enoent(struct vop_generic_args *ap)
182 {
183 
184 	return (ENOENT);
185 }
186 
187 int
188 vop_null(struct vop_generic_args *ap)
189 {
190 
191 	return (0);
192 }
193 
194 /*
195  * Helper function to panic on some bad VOPs in some filesystems.
196  */
197 int
198 vop_panic(struct vop_generic_args *ap)
199 {
200 
201 	panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name);
202 }
203 
204 /*
205  * vop_std<something> and vop_no<something> are default functions for use by
206  * filesystems that need the "default reasonable" implementation for a
207  * particular operation.
208  *
209  * The documentation for the operations they implement exists (if it exists)
210  * in the VOP_<SOMETHING>(9) manpage (all uppercase).
211  */
212 
213 /*
214  * Default vop for filesystems that do not support name lookup
215  */
216 static int
217 vop_nolookup(ap)
218 	struct vop_lookup_args /* {
219 		struct vnode *a_dvp;
220 		struct vnode **a_vpp;
221 		struct componentname *a_cnp;
222 	} */ *ap;
223 {
224 
225 	*ap->a_vpp = NULL;
226 	return (ENOTDIR);
227 }
228 
229 /*
230  * vop_norename:
231  *
232  * Handle unlock and reference counting for arguments of vop_rename
233  * for filesystems that do not implement rename operation.
234  */
235 static int
236 vop_norename(struct vop_rename_args *ap)
237 {
238 
239 	vop_rename_fail(ap);
240 	return (EOPNOTSUPP);
241 }
242 
243 /*
244  *	vop_nostrategy:
245  *
246  *	Strategy routine for VFS devices that have none.
247  *
248  *	BIO_ERROR and B_INVAL must be cleared prior to calling any strategy
249  *	routine.  Typically this is done for a BIO_READ strategy call.
250  *	Typically B_INVAL is assumed to already be clear prior to a write
251  *	and should not be cleared manually unless you just made the buffer
252  *	invalid.  BIO_ERROR should be cleared either way.
253  */
254 
255 static int
256 vop_nostrategy (struct vop_strategy_args *ap)
257 {
258 	printf("No strategy for buffer at %p\n", ap->a_bp);
259 	vprint("vnode", ap->a_vp);
260 	ap->a_bp->b_ioflags |= BIO_ERROR;
261 	ap->a_bp->b_error = EOPNOTSUPP;
262 	bufdone(ap->a_bp);
263 	return (EOPNOTSUPP);
264 }
265 
266 static int
267 get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf,
268 		int dirbuflen, off_t *off, char **cpos, int *len,
269 		int *eofflag, struct thread *td)
270 {
271 	int error, reclen;
272 	struct uio uio;
273 	struct iovec iov;
274 	struct dirent *dp;
275 
276 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
277 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
278 
279 	if (*len == 0) {
280 		iov.iov_base = dirbuf;
281 		iov.iov_len = dirbuflen;
282 
283 		uio.uio_iov = &iov;
284 		uio.uio_iovcnt = 1;
285 		uio.uio_offset = *off;
286 		uio.uio_resid = dirbuflen;
287 		uio.uio_segflg = UIO_SYSSPACE;
288 		uio.uio_rw = UIO_READ;
289 		uio.uio_td = td;
290 
291 		*eofflag = 0;
292 
293 #ifdef MAC
294 		error = mac_vnode_check_readdir(td->td_ucred, vp);
295 		if (error == 0)
296 #endif
297 			error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag,
298 		    		NULL, NULL);
299 		if (error)
300 			return (error);
301 
302 		*off = uio.uio_offset;
303 
304 		*cpos = dirbuf;
305 		*len = (dirbuflen - uio.uio_resid);
306 
307 		if (*len == 0)
308 			return (ENOENT);
309 	}
310 
311 	dp = (struct dirent *)(*cpos);
312 	reclen = dp->d_reclen;
313 	*dpp = dp;
314 
315 	/* check for malformed directory.. */
316 	if (reclen < DIRENT_MINSIZE)
317 		return (EINVAL);
318 
319 	*cpos += reclen;
320 	*len -= reclen;
321 
322 	return (0);
323 }
324 
325 /*
326  * Check if a named file exists in a given directory vnode.
327  */
328 static int
329 dirent_exists(struct vnode *vp, const char *dirname, struct thread *td)
330 {
331 	char *dirbuf, *cpos;
332 	int error, eofflag, dirbuflen, len, found;
333 	off_t off;
334 	struct dirent *dp;
335 	struct vattr va;
336 
337 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
338 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
339 
340 	found = 0;
341 
342 	error = VOP_GETATTR(vp, &va, td->td_ucred);
343 	if (error)
344 		return (found);
345 
346 	dirbuflen = DEV_BSIZE;
347 	if (dirbuflen < va.va_blocksize)
348 		dirbuflen = va.va_blocksize;
349 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
350 
351 	off = 0;
352 	len = 0;
353 	do {
354 		error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off,
355 					&cpos, &len, &eofflag, td);
356 		if (error)
357 			goto out;
358 
359 		if (dp->d_type != DT_WHT && dp->d_fileno != 0 &&
360 		    strcmp(dp->d_name, dirname) == 0) {
361 			found = 1;
362 			goto out;
363 		}
364 	} while (len > 0 || !eofflag);
365 
366 out:
367 	free(dirbuf, M_TEMP);
368 	return (found);
369 }
370 
371 int
372 vop_stdaccess(struct vop_access_args *ap)
373 {
374 
375 	KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN |
376 	    VAPPEND)) == 0, ("invalid bit in accmode"));
377 
378 	return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td));
379 }
380 
381 int
382 vop_stdaccessx(struct vop_accessx_args *ap)
383 {
384 	int error;
385 	accmode_t accmode = ap->a_accmode;
386 
387 	error = vfs_unixify_accmode(&accmode);
388 	if (error != 0)
389 		return (error);
390 
391 	if (accmode == 0)
392 		return (0);
393 
394 	return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td));
395 }
396 
397 /*
398  * Advisory record locking support
399  */
400 int
401 vop_stdadvlock(struct vop_advlock_args *ap)
402 {
403 	struct vnode *vp;
404 	struct vattr vattr;
405 	int error;
406 
407 	vp = ap->a_vp;
408 	if (ap->a_fl->l_whence == SEEK_END) {
409 		/*
410 		 * The NFSv4 server must avoid doing a vn_lock() here, since it
411 		 * can deadlock the nfsd threads, due to a LOR.  Fortunately
412 		 * the NFSv4 server always uses SEEK_SET and this code is
413 		 * only required for the SEEK_END case.
414 		 */
415 		vn_lock(vp, LK_SHARED | LK_RETRY);
416 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
417 		VOP_UNLOCK(vp, 0);
418 		if (error)
419 			return (error);
420 	} else
421 		vattr.va_size = 0;
422 
423 	return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size));
424 }
425 
426 int
427 vop_stdadvlockasync(struct vop_advlockasync_args *ap)
428 {
429 	struct vnode *vp;
430 	struct vattr vattr;
431 	int error;
432 
433 	vp = ap->a_vp;
434 	if (ap->a_fl->l_whence == SEEK_END) {
435 		/* The size argument is only needed for SEEK_END. */
436 		vn_lock(vp, LK_SHARED | LK_RETRY);
437 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
438 		VOP_UNLOCK(vp, 0);
439 		if (error)
440 			return (error);
441 	} else
442 		vattr.va_size = 0;
443 
444 	return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size));
445 }
446 
447 int
448 vop_stdadvlockpurge(struct vop_advlockpurge_args *ap)
449 {
450 	struct vnode *vp;
451 
452 	vp = ap->a_vp;
453 	lf_purgelocks(vp, &vp->v_lockf);
454 	return (0);
455 }
456 
457 /*
458  * vop_stdpathconf:
459  *
460  * Standard implementation of POSIX pathconf, to get information about limits
461  * for a filesystem.
462  * Override per filesystem for the case where the filesystem has smaller
463  * limits.
464  */
465 int
466 vop_stdpathconf(ap)
467 	struct vop_pathconf_args /* {
468 	struct vnode *a_vp;
469 	int a_name;
470 	int *a_retval;
471 	} */ *ap;
472 {
473 
474 	switch (ap->a_name) {
475 		case _PC_NAME_MAX:
476 			*ap->a_retval = NAME_MAX;
477 			return (0);
478 		case _PC_PATH_MAX:
479 			*ap->a_retval = PATH_MAX;
480 			return (0);
481 		case _PC_LINK_MAX:
482 			*ap->a_retval = LINK_MAX;
483 			return (0);
484 		case _PC_MAX_CANON:
485 			*ap->a_retval = MAX_CANON;
486 			return (0);
487 		case _PC_MAX_INPUT:
488 			*ap->a_retval = MAX_INPUT;
489 			return (0);
490 		case _PC_PIPE_BUF:
491 			*ap->a_retval = PIPE_BUF;
492 			return (0);
493 		case _PC_CHOWN_RESTRICTED:
494 			*ap->a_retval = 1;
495 			return (0);
496 		case _PC_VDISABLE:
497 			*ap->a_retval = _POSIX_VDISABLE;
498 			return (0);
499 		default:
500 			return (EINVAL);
501 	}
502 	/* NOTREACHED */
503 }
504 
505 /*
506  * Standard lock, unlock and islocked functions.
507  */
508 int
509 vop_stdlock(ap)
510 	struct vop_lock1_args /* {
511 		struct vnode *a_vp;
512 		int a_flags;
513 		char *file;
514 		int line;
515 	} */ *ap;
516 {
517 	struct vnode *vp = ap->a_vp;
518 
519 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
520 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
521 	    ap->a_line));
522 }
523 
524 /* See above. */
525 int
526 vop_stdunlock(ap)
527 	struct vop_unlock_args /* {
528 		struct vnode *a_vp;
529 		int a_flags;
530 	} */ *ap;
531 {
532 	struct vnode *vp = ap->a_vp;
533 
534 	return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp)));
535 }
536 
537 /* See above. */
538 int
539 vop_stdislocked(ap)
540 	struct vop_islocked_args /* {
541 		struct vnode *a_vp;
542 	} */ *ap;
543 {
544 
545 	return (lockstatus(ap->a_vp->v_vnlock));
546 }
547 
548 /*
549  * Return true for select/poll.
550  */
551 int
552 vop_nopoll(ap)
553 	struct vop_poll_args /* {
554 		struct vnode *a_vp;
555 		int  a_events;
556 		struct ucred *a_cred;
557 		struct thread *a_td;
558 	} */ *ap;
559 {
560 
561 	return (poll_no_poll(ap->a_events));
562 }
563 
564 /*
565  * Implement poll for local filesystems that support it.
566  */
567 int
568 vop_stdpoll(ap)
569 	struct vop_poll_args /* {
570 		struct vnode *a_vp;
571 		int  a_events;
572 		struct ucred *a_cred;
573 		struct thread *a_td;
574 	} */ *ap;
575 {
576 	if (ap->a_events & ~POLLSTANDARD)
577 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
578 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
579 }
580 
581 /*
582  * Return our mount point, as we will take charge of the writes.
583  */
584 int
585 vop_stdgetwritemount(ap)
586 	struct vop_getwritemount_args /* {
587 		struct vnode *a_vp;
588 		struct mount **a_mpp;
589 	} */ *ap;
590 {
591 	struct mount *mp;
592 
593 	/*
594 	 * XXX Since this is called unlocked we may be recycled while
595 	 * attempting to ref the mount.  If this is the case or mountpoint
596 	 * will be set to NULL.  We only have to prevent this call from
597 	 * returning with a ref to an incorrect mountpoint.  It is not
598 	 * harmful to return with a ref to our previous mountpoint.
599 	 */
600 	mp = ap->a_vp->v_mount;
601 	if (mp != NULL) {
602 		vfs_ref(mp);
603 		if (mp != ap->a_vp->v_mount) {
604 			vfs_rel(mp);
605 			mp = NULL;
606 		}
607 	}
608 	*(ap->a_mpp) = mp;
609 	return (0);
610 }
611 
612 /* XXX Needs good comment and VOP_BMAP(9) manpage */
613 int
614 vop_stdbmap(ap)
615 	struct vop_bmap_args /* {
616 		struct vnode *a_vp;
617 		daddr_t  a_bn;
618 		struct bufobj **a_bop;
619 		daddr_t *a_bnp;
620 		int *a_runp;
621 		int *a_runb;
622 	} */ *ap;
623 {
624 
625 	if (ap->a_bop != NULL)
626 		*ap->a_bop = &ap->a_vp->v_bufobj;
627 	if (ap->a_bnp != NULL)
628 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
629 	if (ap->a_runp != NULL)
630 		*ap->a_runp = 0;
631 	if (ap->a_runb != NULL)
632 		*ap->a_runb = 0;
633 	return (0);
634 }
635 
636 int
637 vop_stdfsync(ap)
638 	struct vop_fsync_args /* {
639 		struct vnode *a_vp;
640 		struct ucred *a_cred;
641 		int a_waitfor;
642 		struct thread *a_td;
643 	} */ *ap;
644 {
645 	struct vnode *vp = ap->a_vp;
646 	struct buf *bp;
647 	struct bufobj *bo;
648 	struct buf *nbp;
649 	int error = 0;
650 	int maxretry = 1000;     /* large, arbitrarily chosen */
651 
652 	bo = &vp->v_bufobj;
653 	BO_LOCK(bo);
654 loop1:
655 	/*
656 	 * MARK/SCAN initialization to avoid infinite loops.
657 	 */
658         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
659                 bp->b_vflags &= ~BV_SCANNED;
660 		bp->b_error = 0;
661 	}
662 
663 	/*
664 	 * Flush all dirty buffers associated with a vnode.
665 	 */
666 loop2:
667 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
668 		if ((bp->b_vflags & BV_SCANNED) != 0)
669 			continue;
670 		bp->b_vflags |= BV_SCANNED;
671 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
672 			if (ap->a_waitfor != MNT_WAIT)
673 				continue;
674 			if (BUF_LOCK(bp,
675 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
676 			    BO_LOCKPTR(bo)) != 0) {
677 				BO_LOCK(bo);
678 				goto loop1;
679 			}
680 			BO_LOCK(bo);
681 		}
682 		BO_UNLOCK(bo);
683 		KASSERT(bp->b_bufobj == bo,
684 		    ("bp %p wrong b_bufobj %p should be %p",
685 		    bp, bp->b_bufobj, bo));
686 		if ((bp->b_flags & B_DELWRI) == 0)
687 			panic("fsync: not dirty");
688 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
689 			vfs_bio_awrite(bp);
690 		} else {
691 			bremfree(bp);
692 			bawrite(bp);
693 		}
694 		BO_LOCK(bo);
695 		goto loop2;
696 	}
697 
698 	/*
699 	 * If synchronous the caller expects us to completely resolve all
700 	 * dirty buffers in the system.  Wait for in-progress I/O to
701 	 * complete (which could include background bitmap writes), then
702 	 * retry if dirty blocks still exist.
703 	 */
704 	if (ap->a_waitfor == MNT_WAIT) {
705 		bufobj_wwait(bo, 0, 0);
706 		if (bo->bo_dirty.bv_cnt > 0) {
707 			/*
708 			 * If we are unable to write any of these buffers
709 			 * then we fail now rather than trying endlessly
710 			 * to write them out.
711 			 */
712 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
713 				if ((error = bp->b_error) == 0)
714 					continue;
715 			if (error == 0 && --maxretry >= 0)
716 				goto loop1;
717 			error = EAGAIN;
718 		}
719 	}
720 	BO_UNLOCK(bo);
721 	if (error == EAGAIN)
722 		vprint("fsync: giving up on dirty", vp);
723 
724 	return (error);
725 }
726 
727 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
728 int
729 vop_stdgetpages(ap)
730 	struct vop_getpages_args /* {
731 		struct vnode *a_vp;
732 		vm_page_t *a_m;
733 		int a_count;
734 		int a_reqpage;
735 	} */ *ap;
736 {
737 
738 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
739 	    ap->a_count, ap->a_reqpage, NULL, NULL);
740 }
741 
742 static int
743 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
744 {
745 	int error;
746 
747 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_reqpage);
748 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_reqpage, error);
749 	return (error);
750 }
751 
752 int
753 vop_stdkqfilter(struct vop_kqfilter_args *ap)
754 {
755 	return vfs_kqfilter(ap);
756 }
757 
758 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
759 int
760 vop_stdputpages(ap)
761 	struct vop_putpages_args /* {
762 		struct vnode *a_vp;
763 		vm_page_t *a_m;
764 		int a_count;
765 		int a_sync;
766 		int *a_rtvals;
767 	} */ *ap;
768 {
769 
770 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
771 	     ap->a_sync, ap->a_rtvals);
772 }
773 
774 int
775 vop_stdvptofh(struct vop_vptofh_args *ap)
776 {
777 	return (EOPNOTSUPP);
778 }
779 
780 int
781 vop_stdvptocnp(struct vop_vptocnp_args *ap)
782 {
783 	struct vnode *vp = ap->a_vp;
784 	struct vnode **dvp = ap->a_vpp;
785 	struct ucred *cred = ap->a_cred;
786 	char *buf = ap->a_buf;
787 	int *buflen = ap->a_buflen;
788 	char *dirbuf, *cpos;
789 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
790 	off_t off;
791 	ino_t fileno;
792 	struct vattr va;
793 	struct nameidata nd;
794 	struct thread *td;
795 	struct dirent *dp;
796 	struct vnode *mvp;
797 
798 	i = *buflen;
799 	error = 0;
800 	covered = 0;
801 	td = curthread;
802 
803 	if (vp->v_type != VDIR)
804 		return (ENOENT);
805 
806 	error = VOP_GETATTR(vp, &va, cred);
807 	if (error)
808 		return (error);
809 
810 	VREF(vp);
811 	locked = VOP_ISLOCKED(vp);
812 	VOP_UNLOCK(vp, 0);
813 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
814 	    "..", vp, td);
815 	flags = FREAD;
816 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
817 	if (error) {
818 		vn_lock(vp, locked | LK_RETRY);
819 		return (error);
820 	}
821 	NDFREE(&nd, NDF_ONLY_PNBUF);
822 
823 	mvp = *dvp = nd.ni_vp;
824 
825 	if (vp->v_mount != (*dvp)->v_mount &&
826 	    ((*dvp)->v_vflag & VV_ROOT) &&
827 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
828 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
829 		VREF(mvp);
830 		VOP_UNLOCK(mvp, 0);
831 		vn_close(mvp, FREAD, cred, td);
832 		VREF(*dvp);
833 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
834 		covered = 1;
835 	}
836 
837 	fileno = va.va_fileid;
838 
839 	dirbuflen = DEV_BSIZE;
840 	if (dirbuflen < va.va_blocksize)
841 		dirbuflen = va.va_blocksize;
842 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
843 
844 	if ((*dvp)->v_type != VDIR) {
845 		error = ENOENT;
846 		goto out;
847 	}
848 
849 	off = 0;
850 	len = 0;
851 	do {
852 		/* call VOP_READDIR of parent */
853 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
854 					&cpos, &len, &eofflag, td);
855 		if (error)
856 			goto out;
857 
858 		if ((dp->d_type != DT_WHT) &&
859 		    (dp->d_fileno == fileno)) {
860 			if (covered) {
861 				VOP_UNLOCK(*dvp, 0);
862 				vn_lock(mvp, LK_SHARED | LK_RETRY);
863 				if (dirent_exists(mvp, dp->d_name, td)) {
864 					error = ENOENT;
865 					VOP_UNLOCK(mvp, 0);
866 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
867 					goto out;
868 				}
869 				VOP_UNLOCK(mvp, 0);
870 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
871 			}
872 			i -= dp->d_namlen;
873 
874 			if (i < 0) {
875 				error = ENOMEM;
876 				goto out;
877 			}
878 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
879 				error = ENOENT;
880 			} else {
881 				bcopy(dp->d_name, buf + i, dp->d_namlen);
882 				error = 0;
883 			}
884 			goto out;
885 		}
886 	} while (len > 0 || !eofflag);
887 	error = ENOENT;
888 
889 out:
890 	free(dirbuf, M_TEMP);
891 	if (!error) {
892 		*buflen = i;
893 		vref(*dvp);
894 	}
895 	if (covered) {
896 		vput(*dvp);
897 		vrele(mvp);
898 	} else {
899 		VOP_UNLOCK(mvp, 0);
900 		vn_close(mvp, FREAD, cred, td);
901 	}
902 	vn_lock(vp, locked | LK_RETRY);
903 	return (error);
904 }
905 
906 int
907 vop_stdallocate(struct vop_allocate_args *ap)
908 {
909 #ifdef __notyet__
910 	struct statfs sfs;
911 #endif
912 	struct iovec aiov;
913 	struct vattr vattr, *vap;
914 	struct uio auio;
915 	off_t fsize, len, cur, offset;
916 	uint8_t *buf;
917 	struct thread *td;
918 	struct vnode *vp;
919 	size_t iosize;
920 	int error;
921 
922 	buf = NULL;
923 	error = 0;
924 	td = curthread;
925 	vap = &vattr;
926 	vp = ap->a_vp;
927 	len = *ap->a_len;
928 	offset = *ap->a_offset;
929 
930 	error = VOP_GETATTR(vp, vap, td->td_ucred);
931 	if (error != 0)
932 		goto out;
933 	fsize = vap->va_size;
934 	iosize = vap->va_blocksize;
935 	if (iosize == 0)
936 		iosize = BLKDEV_IOSIZE;
937 	if (iosize > MAXPHYS)
938 		iosize = MAXPHYS;
939 	buf = malloc(iosize, M_TEMP, M_WAITOK);
940 
941 #ifdef __notyet__
942 	/*
943 	 * Check if the filesystem sets f_maxfilesize; if not use
944 	 * VOP_SETATTR to perform the check.
945 	 */
946 	error = VFS_STATFS(vp->v_mount, &sfs, td);
947 	if (error != 0)
948 		goto out;
949 	if (sfs.f_maxfilesize) {
950 		if (offset > sfs.f_maxfilesize || len > sfs.f_maxfilesize ||
951 		    offset + len > sfs.f_maxfilesize) {
952 			error = EFBIG;
953 			goto out;
954 		}
955 	} else
956 #endif
957 	if (offset + len > vap->va_size) {
958 		/*
959 		 * Test offset + len against the filesystem's maxfilesize.
960 		 */
961 		VATTR_NULL(vap);
962 		vap->va_size = offset + len;
963 		error = VOP_SETATTR(vp, vap, td->td_ucred);
964 		if (error != 0)
965 			goto out;
966 		VATTR_NULL(vap);
967 		vap->va_size = fsize;
968 		error = VOP_SETATTR(vp, vap, td->td_ucred);
969 		if (error != 0)
970 			goto out;
971 	}
972 
973 	for (;;) {
974 		/*
975 		 * Read and write back anything below the nominal file
976 		 * size.  There's currently no way outside the filesystem
977 		 * to know whether this area is sparse or not.
978 		 */
979 		cur = iosize;
980 		if ((offset % iosize) != 0)
981 			cur -= (offset % iosize);
982 		if (cur > len)
983 			cur = len;
984 		if (offset < fsize) {
985 			aiov.iov_base = buf;
986 			aiov.iov_len = cur;
987 			auio.uio_iov = &aiov;
988 			auio.uio_iovcnt = 1;
989 			auio.uio_offset = offset;
990 			auio.uio_resid = cur;
991 			auio.uio_segflg = UIO_SYSSPACE;
992 			auio.uio_rw = UIO_READ;
993 			auio.uio_td = td;
994 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
995 			if (error != 0)
996 				break;
997 			if (auio.uio_resid > 0) {
998 				bzero(buf + cur - auio.uio_resid,
999 				    auio.uio_resid);
1000 			}
1001 		} else {
1002 			bzero(buf, cur);
1003 		}
1004 
1005 		aiov.iov_base = buf;
1006 		aiov.iov_len = cur;
1007 		auio.uio_iov = &aiov;
1008 		auio.uio_iovcnt = 1;
1009 		auio.uio_offset = offset;
1010 		auio.uio_resid = cur;
1011 		auio.uio_segflg = UIO_SYSSPACE;
1012 		auio.uio_rw = UIO_WRITE;
1013 		auio.uio_td = td;
1014 
1015 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
1016 		if (error != 0)
1017 			break;
1018 
1019 		len -= cur;
1020 		offset += cur;
1021 		if (len == 0)
1022 			break;
1023 		if (should_yield())
1024 			break;
1025 	}
1026 
1027  out:
1028 	*ap->a_len = len;
1029 	*ap->a_offset = offset;
1030 	free(buf, M_TEMP);
1031 	return (error);
1032 }
1033 
1034 int
1035 vop_stdadvise(struct vop_advise_args *ap)
1036 {
1037 	struct buf *bp;
1038 	struct buflists *bl;
1039 	struct vnode *vp;
1040 	daddr_t bn, startn, endn;
1041 	off_t start, end;
1042 	int bsize, error;
1043 
1044 	vp = ap->a_vp;
1045 	switch (ap->a_advice) {
1046 	case POSIX_FADV_WILLNEED:
1047 		/*
1048 		 * Do nothing for now.  Filesystems should provide a
1049 		 * custom method which starts an asynchronous read of
1050 		 * the requested region.
1051 		 */
1052 		error = 0;
1053 		break;
1054 	case POSIX_FADV_DONTNEED:
1055 		error = 0;
1056 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1057 		if (vp->v_iflag & VI_DOOMED) {
1058 			VOP_UNLOCK(vp, 0);
1059 			break;
1060 		}
1061 
1062 		/*
1063 		 * Deactivate pages in the specified range from the backing VM
1064 		 * object.  Pages that are resident in the buffer cache will
1065 		 * remain wired until their corresponding buffers are released
1066 		 * below.
1067 		 */
1068 		if (vp->v_object != NULL) {
1069 			start = trunc_page(ap->a_start);
1070 			end = round_page(ap->a_end);
1071 			VM_OBJECT_WLOCK(vp->v_object);
1072 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1073 			    OFF_TO_IDX(end));
1074 			VM_OBJECT_WUNLOCK(vp->v_object);
1075 		}
1076 
1077 		BO_RLOCK(&vp->v_bufobj);
1078 		bsize = vp->v_bufobj.bo_bsize;
1079 		startn = ap->a_start / bsize;
1080 		endn = -1;
1081 		bl = &vp->v_bufobj.bo_clean.bv_hd;
1082 		if (!TAILQ_EMPTY(bl))
1083 			endn = TAILQ_LAST(bl, buflists)->b_lblkno;
1084 		bl = &vp->v_bufobj.bo_dirty.bv_hd;
1085 		if (!TAILQ_EMPTY(bl) &&
1086 		    endn < TAILQ_LAST(bl, buflists)->b_lblkno)
1087 			endn = TAILQ_LAST(bl, buflists)->b_lblkno;
1088 		if (ap->a_end != OFF_MAX && endn != -1)
1089 			endn = ap->a_end / bsize;
1090 		BO_RUNLOCK(&vp->v_bufobj);
1091 		/*
1092 		 * In the VMIO case, use the B_NOREUSE flag to hint that the
1093 		 * pages backing each buffer in the range are unlikely to be
1094 		 * reused.  Dirty buffers will have the hint applied once
1095 		 * they've been written.
1096 		 */
1097 		for (bn = startn; bn <= endn; bn++) {
1098 			bp = getblk(vp, bn, bsize, 0, 0, GB_NOCREAT |
1099 			    GB_UNMAPPED);
1100 			if (bp == NULL)
1101 				continue;
1102 			bp->b_flags |= B_RELBUF;
1103 			if (vp->v_object != NULL)
1104 				bp->b_flags |= B_NOREUSE;
1105 			brelse(bp);
1106 		}
1107 		VOP_UNLOCK(vp, 0);
1108 		break;
1109 	default:
1110 		error = EINVAL;
1111 		break;
1112 	}
1113 	return (error);
1114 }
1115 
1116 int
1117 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1118 {
1119 
1120 	ap->a_vp->v_socket = ap->a_socket;
1121 	return (0);
1122 }
1123 
1124 int
1125 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1126 {
1127 
1128 	*ap->a_socket = ap->a_vp->v_socket;
1129 	return (0);
1130 }
1131 
1132 int
1133 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1134 {
1135 
1136 	ap->a_vp->v_socket = NULL;
1137 	return (0);
1138 }
1139 
1140 static int
1141 vop_stdis_text(struct vop_is_text_args *ap)
1142 {
1143 
1144 	return ((ap->a_vp->v_vflag & VV_TEXT) != 0);
1145 }
1146 
1147 static int
1148 vop_stdset_text(struct vop_set_text_args *ap)
1149 {
1150 
1151 	ap->a_vp->v_vflag |= VV_TEXT;
1152 	return (0);
1153 }
1154 
1155 static int
1156 vop_stdunset_text(struct vop_unset_text_args *ap)
1157 {
1158 
1159 	ap->a_vp->v_vflag &= ~VV_TEXT;
1160 	return (0);
1161 }
1162 
1163 static int
1164 vop_stdget_writecount(struct vop_get_writecount_args *ap)
1165 {
1166 
1167 	*ap->a_writecount = ap->a_vp->v_writecount;
1168 	return (0);
1169 }
1170 
1171 static int
1172 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1173 {
1174 
1175 	ap->a_vp->v_writecount += ap->a_inc;
1176 	return (0);
1177 }
1178 
1179 /*
1180  * vfs default ops
1181  * used to fill the vfs function table to get reasonable default return values.
1182  */
1183 int
1184 vfs_stdroot (mp, flags, vpp)
1185 	struct mount *mp;
1186 	int flags;
1187 	struct vnode **vpp;
1188 {
1189 
1190 	return (EOPNOTSUPP);
1191 }
1192 
1193 int
1194 vfs_stdstatfs (mp, sbp)
1195 	struct mount *mp;
1196 	struct statfs *sbp;
1197 {
1198 
1199 	return (EOPNOTSUPP);
1200 }
1201 
1202 int
1203 vfs_stdquotactl (mp, cmds, uid, arg)
1204 	struct mount *mp;
1205 	int cmds;
1206 	uid_t uid;
1207 	void *arg;
1208 {
1209 
1210 	return (EOPNOTSUPP);
1211 }
1212 
1213 int
1214 vfs_stdsync(mp, waitfor)
1215 	struct mount *mp;
1216 	int waitfor;
1217 {
1218 	struct vnode *vp, *mvp;
1219 	struct thread *td;
1220 	int error, lockreq, allerror = 0;
1221 
1222 	td = curthread;
1223 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1224 	if (waitfor != MNT_WAIT)
1225 		lockreq |= LK_NOWAIT;
1226 	/*
1227 	 * Force stale buffer cache information to be flushed.
1228 	 */
1229 loop:
1230 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1231 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1232 			VI_UNLOCK(vp);
1233 			continue;
1234 		}
1235 		if ((error = vget(vp, lockreq, td)) != 0) {
1236 			if (error == ENOENT) {
1237 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1238 				goto loop;
1239 			}
1240 			continue;
1241 		}
1242 		error = VOP_FSYNC(vp, waitfor, td);
1243 		if (error)
1244 			allerror = error;
1245 		vput(vp);
1246 	}
1247 	return (allerror);
1248 }
1249 
1250 int
1251 vfs_stdnosync (mp, waitfor)
1252 	struct mount *mp;
1253 	int waitfor;
1254 {
1255 
1256 	return (0);
1257 }
1258 
1259 int
1260 vfs_stdvget (mp, ino, flags, vpp)
1261 	struct mount *mp;
1262 	ino_t ino;
1263 	int flags;
1264 	struct vnode **vpp;
1265 {
1266 
1267 	return (EOPNOTSUPP);
1268 }
1269 
1270 int
1271 vfs_stdfhtovp (mp, fhp, flags, vpp)
1272 	struct mount *mp;
1273 	struct fid *fhp;
1274 	int flags;
1275 	struct vnode **vpp;
1276 {
1277 
1278 	return (EOPNOTSUPP);
1279 }
1280 
1281 int
1282 vfs_stdinit (vfsp)
1283 	struct vfsconf *vfsp;
1284 {
1285 
1286 	return (0);
1287 }
1288 
1289 int
1290 vfs_stduninit (vfsp)
1291 	struct vfsconf *vfsp;
1292 {
1293 
1294 	return(0);
1295 }
1296 
1297 int
1298 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1299 	struct mount *mp;
1300 	int cmd;
1301 	struct vnode *filename_vp;
1302 	int attrnamespace;
1303 	const char *attrname;
1304 {
1305 
1306 	if (filename_vp != NULL)
1307 		VOP_UNLOCK(filename_vp, 0);
1308 	return (EOPNOTSUPP);
1309 }
1310 
1311 int
1312 vfs_stdsysctl(mp, op, req)
1313 	struct mount *mp;
1314 	fsctlop_t op;
1315 	struct sysctl_req *req;
1316 {
1317 
1318 	return (EOPNOTSUPP);
1319 }
1320 
1321 /* end of vfs default ops */
1322