xref: /freebsd/sys/kern/vfs_default.c (revision 3fc36ee018bb836bd1796067cf4ef8683f166ebc)
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_stdfdatasync(struct vop_fdatasync_args *ap);
87 static int vop_stdgetpages_async(struct vop_getpages_async_args *ap);
88 
89 /*
90  * This vnode table stores what we want to do if the filesystem doesn't
91  * implement a particular VOP.
92  *
93  * If there is no specific entry here, we will return EOPNOTSUPP.
94  *
95  * Note that every filesystem has to implement either vop_access
96  * or vop_accessx; failing to do so will result in immediate crash
97  * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(),
98  * which calls vop_stdaccess() etc.
99  */
100 
101 struct vop_vector default_vnodeops = {
102 	.vop_default =		NULL,
103 	.vop_bypass =		VOP_EOPNOTSUPP,
104 
105 	.vop_access =		vop_stdaccess,
106 	.vop_accessx =		vop_stdaccessx,
107 	.vop_advise =		vop_stdadvise,
108 	.vop_advlock =		vop_stdadvlock,
109 	.vop_advlockasync =	vop_stdadvlockasync,
110 	.vop_advlockpurge =	vop_stdadvlockpurge,
111 	.vop_allocate =		vop_stdallocate,
112 	.vop_bmap =		vop_stdbmap,
113 	.vop_close =		VOP_NULL,
114 	.vop_fsync =		VOP_NULL,
115 	.vop_fdatasync =	vop_stdfdatasync,
116 	.vop_getpages =		vop_stdgetpages,
117 	.vop_getpages_async =	vop_stdgetpages_async,
118 	.vop_getwritemount = 	vop_stdgetwritemount,
119 	.vop_inactive =		VOP_NULL,
120 	.vop_ioctl =		VOP_ENOTTY,
121 	.vop_kqfilter =		vop_stdkqfilter,
122 	.vop_islocked =		vop_stdislocked,
123 	.vop_lock1 =		vop_stdlock,
124 	.vop_lookup =		vop_nolookup,
125 	.vop_open =		VOP_NULL,
126 	.vop_pathconf =		VOP_EINVAL,
127 	.vop_poll =		vop_nopoll,
128 	.vop_putpages =		vop_stdputpages,
129 	.vop_readlink =		VOP_EINVAL,
130 	.vop_rename =		vop_norename,
131 	.vop_revoke =		VOP_PANIC,
132 	.vop_strategy =		vop_nostrategy,
133 	.vop_unlock =		vop_stdunlock,
134 	.vop_vptocnp =		vop_stdvptocnp,
135 	.vop_vptofh =		vop_stdvptofh,
136 	.vop_unp_bind =		vop_stdunp_bind,
137 	.vop_unp_connect =	vop_stdunp_connect,
138 	.vop_unp_detach =	vop_stdunp_detach,
139 	.vop_is_text =		vop_stdis_text,
140 	.vop_set_text =		vop_stdset_text,
141 	.vop_unset_text =	vop_stdunset_text,
142 	.vop_get_writecount =	vop_stdget_writecount,
143 	.vop_add_writecount =	vop_stdadd_writecount,
144 };
145 
146 /*
147  * Series of placeholder functions for various error returns for
148  * VOPs.
149  */
150 
151 int
152 vop_eopnotsupp(struct vop_generic_args *ap)
153 {
154 	/*
155 	printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name);
156 	*/
157 
158 	return (EOPNOTSUPP);
159 }
160 
161 int
162 vop_ebadf(struct vop_generic_args *ap)
163 {
164 
165 	return (EBADF);
166 }
167 
168 int
169 vop_enotty(struct vop_generic_args *ap)
170 {
171 
172 	return (ENOTTY);
173 }
174 
175 int
176 vop_einval(struct vop_generic_args *ap)
177 {
178 
179 	return (EINVAL);
180 }
181 
182 int
183 vop_enoent(struct vop_generic_args *ap)
184 {
185 
186 	return (ENOENT);
187 }
188 
189 int
190 vop_null(struct vop_generic_args *ap)
191 {
192 
193 	return (0);
194 }
195 
196 /*
197  * Helper function to panic on some bad VOPs in some filesystems.
198  */
199 int
200 vop_panic(struct vop_generic_args *ap)
201 {
202 
203 	panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name);
204 }
205 
206 /*
207  * vop_std<something> and vop_no<something> are default functions for use by
208  * filesystems that need the "default reasonable" implementation for a
209  * particular operation.
210  *
211  * The documentation for the operations they implement exists (if it exists)
212  * in the VOP_<SOMETHING>(9) manpage (all uppercase).
213  */
214 
215 /*
216  * Default vop for filesystems that do not support name lookup
217  */
218 static int
219 vop_nolookup(ap)
220 	struct vop_lookup_args /* {
221 		struct vnode *a_dvp;
222 		struct vnode **a_vpp;
223 		struct componentname *a_cnp;
224 	} */ *ap;
225 {
226 
227 	*ap->a_vpp = NULL;
228 	return (ENOTDIR);
229 }
230 
231 /*
232  * vop_norename:
233  *
234  * Handle unlock and reference counting for arguments of vop_rename
235  * for filesystems that do not implement rename operation.
236  */
237 static int
238 vop_norename(struct vop_rename_args *ap)
239 {
240 
241 	vop_rename_fail(ap);
242 	return (EOPNOTSUPP);
243 }
244 
245 /*
246  *	vop_nostrategy:
247  *
248  *	Strategy routine for VFS devices that have none.
249  *
250  *	BIO_ERROR and B_INVAL must be cleared prior to calling any strategy
251  *	routine.  Typically this is done for a BIO_READ strategy call.
252  *	Typically B_INVAL is assumed to already be clear prior to a write
253  *	and should not be cleared manually unless you just made the buffer
254  *	invalid.  BIO_ERROR should be cleared either way.
255  */
256 
257 static int
258 vop_nostrategy (struct vop_strategy_args *ap)
259 {
260 	printf("No strategy for buffer at %p\n", ap->a_bp);
261 	vn_printf(ap->a_vp, "vnode ");
262 	ap->a_bp->b_ioflags |= BIO_ERROR;
263 	ap->a_bp->b_error = EOPNOTSUPP;
264 	bufdone(ap->a_bp);
265 	return (EOPNOTSUPP);
266 }
267 
268 static int
269 get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf,
270 		int dirbuflen, off_t *off, char **cpos, int *len,
271 		int *eofflag, struct thread *td)
272 {
273 	int error, reclen;
274 	struct uio uio;
275 	struct iovec iov;
276 	struct dirent *dp;
277 
278 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
279 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
280 
281 	if (*len == 0) {
282 		iov.iov_base = dirbuf;
283 		iov.iov_len = dirbuflen;
284 
285 		uio.uio_iov = &iov;
286 		uio.uio_iovcnt = 1;
287 		uio.uio_offset = *off;
288 		uio.uio_resid = dirbuflen;
289 		uio.uio_segflg = UIO_SYSSPACE;
290 		uio.uio_rw = UIO_READ;
291 		uio.uio_td = td;
292 
293 		*eofflag = 0;
294 
295 #ifdef MAC
296 		error = mac_vnode_check_readdir(td->td_ucred, vp);
297 		if (error == 0)
298 #endif
299 			error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag,
300 		    		NULL, NULL);
301 		if (error)
302 			return (error);
303 
304 		*off = uio.uio_offset;
305 
306 		*cpos = dirbuf;
307 		*len = (dirbuflen - uio.uio_resid);
308 
309 		if (*len == 0)
310 			return (ENOENT);
311 	}
312 
313 	dp = (struct dirent *)(*cpos);
314 	reclen = dp->d_reclen;
315 	*dpp = dp;
316 
317 	/* check for malformed directory.. */
318 	if (reclen < DIRENT_MINSIZE)
319 		return (EINVAL);
320 
321 	*cpos += reclen;
322 	*len -= reclen;
323 
324 	return (0);
325 }
326 
327 /*
328  * Check if a named file exists in a given directory vnode.
329  */
330 static int
331 dirent_exists(struct vnode *vp, const char *dirname, struct thread *td)
332 {
333 	char *dirbuf, *cpos;
334 	int error, eofflag, dirbuflen, len, found;
335 	off_t off;
336 	struct dirent *dp;
337 	struct vattr va;
338 
339 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
340 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
341 
342 	found = 0;
343 
344 	error = VOP_GETATTR(vp, &va, td->td_ucred);
345 	if (error)
346 		return (found);
347 
348 	dirbuflen = DEV_BSIZE;
349 	if (dirbuflen < va.va_blocksize)
350 		dirbuflen = va.va_blocksize;
351 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
352 
353 	off = 0;
354 	len = 0;
355 	do {
356 		error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off,
357 					&cpos, &len, &eofflag, td);
358 		if (error)
359 			goto out;
360 
361 		if (dp->d_type != DT_WHT && dp->d_fileno != 0 &&
362 		    strcmp(dp->d_name, dirname) == 0) {
363 			found = 1;
364 			goto out;
365 		}
366 	} while (len > 0 || !eofflag);
367 
368 out:
369 	free(dirbuf, M_TEMP);
370 	return (found);
371 }
372 
373 int
374 vop_stdaccess(struct vop_access_args *ap)
375 {
376 
377 	KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN |
378 	    VAPPEND)) == 0, ("invalid bit in accmode"));
379 
380 	return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td));
381 }
382 
383 int
384 vop_stdaccessx(struct vop_accessx_args *ap)
385 {
386 	int error;
387 	accmode_t accmode = ap->a_accmode;
388 
389 	error = vfs_unixify_accmode(&accmode);
390 	if (error != 0)
391 		return (error);
392 
393 	if (accmode == 0)
394 		return (0);
395 
396 	return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td));
397 }
398 
399 /*
400  * Advisory record locking support
401  */
402 int
403 vop_stdadvlock(struct vop_advlock_args *ap)
404 {
405 	struct vnode *vp;
406 	struct vattr vattr;
407 	int error;
408 
409 	vp = ap->a_vp;
410 	if (ap->a_fl->l_whence == SEEK_END) {
411 		/*
412 		 * The NFSv4 server must avoid doing a vn_lock() here, since it
413 		 * can deadlock the nfsd threads, due to a LOR.  Fortunately
414 		 * the NFSv4 server always uses SEEK_SET and this code is
415 		 * only required for the SEEK_END case.
416 		 */
417 		vn_lock(vp, LK_SHARED | LK_RETRY);
418 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
419 		VOP_UNLOCK(vp, 0);
420 		if (error)
421 			return (error);
422 	} else
423 		vattr.va_size = 0;
424 
425 	return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size));
426 }
427 
428 int
429 vop_stdadvlockasync(struct vop_advlockasync_args *ap)
430 {
431 	struct vnode *vp;
432 	struct vattr vattr;
433 	int error;
434 
435 	vp = ap->a_vp;
436 	if (ap->a_fl->l_whence == SEEK_END) {
437 		/* The size argument is only needed for SEEK_END. */
438 		vn_lock(vp, LK_SHARED | LK_RETRY);
439 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
440 		VOP_UNLOCK(vp, 0);
441 		if (error)
442 			return (error);
443 	} else
444 		vattr.va_size = 0;
445 
446 	return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size));
447 }
448 
449 int
450 vop_stdadvlockpurge(struct vop_advlockpurge_args *ap)
451 {
452 	struct vnode *vp;
453 
454 	vp = ap->a_vp;
455 	lf_purgelocks(vp, &vp->v_lockf);
456 	return (0);
457 }
458 
459 /*
460  * vop_stdpathconf:
461  *
462  * Standard implementation of POSIX pathconf, to get information about limits
463  * for a filesystem.
464  * Override per filesystem for the case where the filesystem has smaller
465  * limits.
466  */
467 int
468 vop_stdpathconf(ap)
469 	struct vop_pathconf_args /* {
470 	struct vnode *a_vp;
471 	int a_name;
472 	int *a_retval;
473 	} */ *ap;
474 {
475 
476 	switch (ap->a_name) {
477 		case _PC_ASYNC_IO:
478 			*ap->a_retval = _POSIX_ASYNCHRONOUS_IO;
479 			return (0);
480 		case _PC_NAME_MAX:
481 			*ap->a_retval = NAME_MAX;
482 			return (0);
483 		case _PC_PATH_MAX:
484 			*ap->a_retval = PATH_MAX;
485 			return (0);
486 		case _PC_LINK_MAX:
487 			*ap->a_retval = LINK_MAX;
488 			return (0);
489 		case _PC_MAX_CANON:
490 			*ap->a_retval = MAX_CANON;
491 			return (0);
492 		case _PC_MAX_INPUT:
493 			*ap->a_retval = MAX_INPUT;
494 			return (0);
495 		case _PC_PIPE_BUF:
496 			*ap->a_retval = PIPE_BUF;
497 			return (0);
498 		case _PC_CHOWN_RESTRICTED:
499 			*ap->a_retval = 1;
500 			return (0);
501 		case _PC_VDISABLE:
502 			*ap->a_retval = _POSIX_VDISABLE;
503 			return (0);
504 		default:
505 			return (EINVAL);
506 	}
507 	/* NOTREACHED */
508 }
509 
510 /*
511  * Standard lock, unlock and islocked functions.
512  */
513 int
514 vop_stdlock(ap)
515 	struct vop_lock1_args /* {
516 		struct vnode *a_vp;
517 		int a_flags;
518 		char *file;
519 		int line;
520 	} */ *ap;
521 {
522 	struct vnode *vp = ap->a_vp;
523 
524 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
525 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
526 	    ap->a_line));
527 }
528 
529 /* See above. */
530 int
531 vop_stdunlock(ap)
532 	struct vop_unlock_args /* {
533 		struct vnode *a_vp;
534 		int a_flags;
535 	} */ *ap;
536 {
537 	struct vnode *vp = ap->a_vp;
538 
539 	return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp)));
540 }
541 
542 /* See above. */
543 int
544 vop_stdislocked(ap)
545 	struct vop_islocked_args /* {
546 		struct vnode *a_vp;
547 	} */ *ap;
548 {
549 
550 	return (lockstatus(ap->a_vp->v_vnlock));
551 }
552 
553 /*
554  * Return true for select/poll.
555  */
556 int
557 vop_nopoll(ap)
558 	struct vop_poll_args /* {
559 		struct vnode *a_vp;
560 		int  a_events;
561 		struct ucred *a_cred;
562 		struct thread *a_td;
563 	} */ *ap;
564 {
565 
566 	return (poll_no_poll(ap->a_events));
567 }
568 
569 /*
570  * Implement poll for local filesystems that support it.
571  */
572 int
573 vop_stdpoll(ap)
574 	struct vop_poll_args /* {
575 		struct vnode *a_vp;
576 		int  a_events;
577 		struct ucred *a_cred;
578 		struct thread *a_td;
579 	} */ *ap;
580 {
581 	if (ap->a_events & ~POLLSTANDARD)
582 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
583 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
584 }
585 
586 /*
587  * Return our mount point, as we will take charge of the writes.
588  */
589 int
590 vop_stdgetwritemount(ap)
591 	struct vop_getwritemount_args /* {
592 		struct vnode *a_vp;
593 		struct mount **a_mpp;
594 	} */ *ap;
595 {
596 	struct mount *mp;
597 
598 	/*
599 	 * XXX Since this is called unlocked we may be recycled while
600 	 * attempting to ref the mount.  If this is the case or mountpoint
601 	 * will be set to NULL.  We only have to prevent this call from
602 	 * returning with a ref to an incorrect mountpoint.  It is not
603 	 * harmful to return with a ref to our previous mountpoint.
604 	 */
605 	mp = ap->a_vp->v_mount;
606 	if (mp != NULL) {
607 		vfs_ref(mp);
608 		if (mp != ap->a_vp->v_mount) {
609 			vfs_rel(mp);
610 			mp = NULL;
611 		}
612 	}
613 	*(ap->a_mpp) = mp;
614 	return (0);
615 }
616 
617 /* XXX Needs good comment and VOP_BMAP(9) manpage */
618 int
619 vop_stdbmap(ap)
620 	struct vop_bmap_args /* {
621 		struct vnode *a_vp;
622 		daddr_t  a_bn;
623 		struct bufobj **a_bop;
624 		daddr_t *a_bnp;
625 		int *a_runp;
626 		int *a_runb;
627 	} */ *ap;
628 {
629 
630 	if (ap->a_bop != NULL)
631 		*ap->a_bop = &ap->a_vp->v_bufobj;
632 	if (ap->a_bnp != NULL)
633 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
634 	if (ap->a_runp != NULL)
635 		*ap->a_runp = 0;
636 	if (ap->a_runb != NULL)
637 		*ap->a_runb = 0;
638 	return (0);
639 }
640 
641 int
642 vop_stdfsync(ap)
643 	struct vop_fsync_args /* {
644 		struct vnode *a_vp;
645 		int a_waitfor;
646 		struct thread *a_td;
647 	} */ *ap;
648 {
649 	struct vnode *vp = ap->a_vp;
650 	struct buf *bp;
651 	struct bufobj *bo;
652 	struct buf *nbp;
653 	int error = 0;
654 	int maxretry = 1000;     /* large, arbitrarily chosen */
655 
656 	bo = &vp->v_bufobj;
657 	BO_LOCK(bo);
658 loop1:
659 	/*
660 	 * MARK/SCAN initialization to avoid infinite loops.
661 	 */
662         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
663                 bp->b_vflags &= ~BV_SCANNED;
664 		bp->b_error = 0;
665 	}
666 
667 	/*
668 	 * Flush all dirty buffers associated with a vnode.
669 	 */
670 loop2:
671 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
672 		if ((bp->b_vflags & BV_SCANNED) != 0)
673 			continue;
674 		bp->b_vflags |= BV_SCANNED;
675 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
676 			if (ap->a_waitfor != MNT_WAIT)
677 				continue;
678 			if (BUF_LOCK(bp,
679 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
680 			    BO_LOCKPTR(bo)) != 0) {
681 				BO_LOCK(bo);
682 				goto loop1;
683 			}
684 			BO_LOCK(bo);
685 		}
686 		BO_UNLOCK(bo);
687 		KASSERT(bp->b_bufobj == bo,
688 		    ("bp %p wrong b_bufobj %p should be %p",
689 		    bp, bp->b_bufobj, bo));
690 		if ((bp->b_flags & B_DELWRI) == 0)
691 			panic("fsync: not dirty");
692 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
693 			vfs_bio_awrite(bp);
694 		} else {
695 			bremfree(bp);
696 			bawrite(bp);
697 		}
698 		BO_LOCK(bo);
699 		goto loop2;
700 	}
701 
702 	/*
703 	 * If synchronous the caller expects us to completely resolve all
704 	 * dirty buffers in the system.  Wait for in-progress I/O to
705 	 * complete (which could include background bitmap writes), then
706 	 * retry if dirty blocks still exist.
707 	 */
708 	if (ap->a_waitfor == MNT_WAIT) {
709 		bufobj_wwait(bo, 0, 0);
710 		if (bo->bo_dirty.bv_cnt > 0) {
711 			/*
712 			 * If we are unable to write any of these buffers
713 			 * then we fail now rather than trying endlessly
714 			 * to write them out.
715 			 */
716 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
717 				if ((error = bp->b_error) == 0)
718 					continue;
719 			if (error == 0 && --maxretry >= 0)
720 				goto loop1;
721 			error = EAGAIN;
722 		}
723 	}
724 	BO_UNLOCK(bo);
725 	if (error == EAGAIN)
726 		vn_printf(vp, "fsync: giving up on dirty ");
727 
728 	return (error);
729 }
730 
731 static int
732 vop_stdfdatasync(struct vop_fdatasync_args *ap)
733 {
734 
735 	return (VOP_FSYNC(ap->a_vp, MNT_WAIT, ap->a_td));
736 }
737 
738 int
739 vop_stdfdatasync_buf(struct vop_fdatasync_args *ap)
740 {
741 	struct vop_fsync_args apf;
742 
743 	apf.a_vp = ap->a_vp;
744 	apf.a_waitfor = MNT_WAIT;
745 	apf.a_td = ap->a_td;
746 	return (vop_stdfsync(&apf));
747 }
748 
749 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
750 int
751 vop_stdgetpages(ap)
752 	struct vop_getpages_args /* {
753 		struct vnode *a_vp;
754 		vm_page_t *a_m;
755 		int a_count;
756 		int *a_rbehind;
757 		int *a_rahead;
758 	} */ *ap;
759 {
760 
761 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
762 	    ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL);
763 }
764 
765 static int
766 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
767 {
768 	int error;
769 
770 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
771 	    ap->a_rahead);
772 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
773 	return (error);
774 }
775 
776 int
777 vop_stdkqfilter(struct vop_kqfilter_args *ap)
778 {
779 	return vfs_kqfilter(ap);
780 }
781 
782 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
783 int
784 vop_stdputpages(ap)
785 	struct vop_putpages_args /* {
786 		struct vnode *a_vp;
787 		vm_page_t *a_m;
788 		int a_count;
789 		int a_sync;
790 		int *a_rtvals;
791 	} */ *ap;
792 {
793 
794 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
795 	     ap->a_sync, ap->a_rtvals);
796 }
797 
798 int
799 vop_stdvptofh(struct vop_vptofh_args *ap)
800 {
801 	return (EOPNOTSUPP);
802 }
803 
804 int
805 vop_stdvptocnp(struct vop_vptocnp_args *ap)
806 {
807 	struct vnode *vp = ap->a_vp;
808 	struct vnode **dvp = ap->a_vpp;
809 	struct ucred *cred = ap->a_cred;
810 	char *buf = ap->a_buf;
811 	int *buflen = ap->a_buflen;
812 	char *dirbuf, *cpos;
813 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
814 	off_t off;
815 	ino_t fileno;
816 	struct vattr va;
817 	struct nameidata nd;
818 	struct thread *td;
819 	struct dirent *dp;
820 	struct vnode *mvp;
821 
822 	i = *buflen;
823 	error = 0;
824 	covered = 0;
825 	td = curthread;
826 
827 	if (vp->v_type != VDIR)
828 		return (ENOENT);
829 
830 	error = VOP_GETATTR(vp, &va, cred);
831 	if (error)
832 		return (error);
833 
834 	VREF(vp);
835 	locked = VOP_ISLOCKED(vp);
836 	VOP_UNLOCK(vp, 0);
837 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
838 	    "..", vp, td);
839 	flags = FREAD;
840 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
841 	if (error) {
842 		vn_lock(vp, locked | LK_RETRY);
843 		return (error);
844 	}
845 	NDFREE(&nd, NDF_ONLY_PNBUF);
846 
847 	mvp = *dvp = nd.ni_vp;
848 
849 	if (vp->v_mount != (*dvp)->v_mount &&
850 	    ((*dvp)->v_vflag & VV_ROOT) &&
851 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
852 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
853 		VREF(mvp);
854 		VOP_UNLOCK(mvp, 0);
855 		vn_close(mvp, FREAD, cred, td);
856 		VREF(*dvp);
857 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
858 		covered = 1;
859 	}
860 
861 	fileno = va.va_fileid;
862 
863 	dirbuflen = DEV_BSIZE;
864 	if (dirbuflen < va.va_blocksize)
865 		dirbuflen = va.va_blocksize;
866 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
867 
868 	if ((*dvp)->v_type != VDIR) {
869 		error = ENOENT;
870 		goto out;
871 	}
872 
873 	off = 0;
874 	len = 0;
875 	do {
876 		/* call VOP_READDIR of parent */
877 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
878 					&cpos, &len, &eofflag, td);
879 		if (error)
880 			goto out;
881 
882 		if ((dp->d_type != DT_WHT) &&
883 		    (dp->d_fileno == fileno)) {
884 			if (covered) {
885 				VOP_UNLOCK(*dvp, 0);
886 				vn_lock(mvp, LK_SHARED | LK_RETRY);
887 				if (dirent_exists(mvp, dp->d_name, td)) {
888 					error = ENOENT;
889 					VOP_UNLOCK(mvp, 0);
890 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
891 					goto out;
892 				}
893 				VOP_UNLOCK(mvp, 0);
894 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
895 			}
896 			i -= dp->d_namlen;
897 
898 			if (i < 0) {
899 				error = ENOMEM;
900 				goto out;
901 			}
902 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
903 				error = ENOENT;
904 			} else {
905 				bcopy(dp->d_name, buf + i, dp->d_namlen);
906 				error = 0;
907 			}
908 			goto out;
909 		}
910 	} while (len > 0 || !eofflag);
911 	error = ENOENT;
912 
913 out:
914 	free(dirbuf, M_TEMP);
915 	if (!error) {
916 		*buflen = i;
917 		vref(*dvp);
918 	}
919 	if (covered) {
920 		vput(*dvp);
921 		vrele(mvp);
922 	} else {
923 		VOP_UNLOCK(mvp, 0);
924 		vn_close(mvp, FREAD, cred, td);
925 	}
926 	vn_lock(vp, locked | LK_RETRY);
927 	return (error);
928 }
929 
930 int
931 vop_stdallocate(struct vop_allocate_args *ap)
932 {
933 #ifdef __notyet__
934 	struct statfs sfs;
935 #endif
936 	struct iovec aiov;
937 	struct vattr vattr, *vap;
938 	struct uio auio;
939 	off_t fsize, len, cur, offset;
940 	uint8_t *buf;
941 	struct thread *td;
942 	struct vnode *vp;
943 	size_t iosize;
944 	int error;
945 
946 	buf = NULL;
947 	error = 0;
948 	td = curthread;
949 	vap = &vattr;
950 	vp = ap->a_vp;
951 	len = *ap->a_len;
952 	offset = *ap->a_offset;
953 
954 	error = VOP_GETATTR(vp, vap, td->td_ucred);
955 	if (error != 0)
956 		goto out;
957 	fsize = vap->va_size;
958 	iosize = vap->va_blocksize;
959 	if (iosize == 0)
960 		iosize = BLKDEV_IOSIZE;
961 	if (iosize > MAXPHYS)
962 		iosize = MAXPHYS;
963 	buf = malloc(iosize, M_TEMP, M_WAITOK);
964 
965 #ifdef __notyet__
966 	/*
967 	 * Check if the filesystem sets f_maxfilesize; if not use
968 	 * VOP_SETATTR to perform the check.
969 	 */
970 	error = VFS_STATFS(vp->v_mount, &sfs, td);
971 	if (error != 0)
972 		goto out;
973 	if (sfs.f_maxfilesize) {
974 		if (offset > sfs.f_maxfilesize || len > sfs.f_maxfilesize ||
975 		    offset + len > sfs.f_maxfilesize) {
976 			error = EFBIG;
977 			goto out;
978 		}
979 	} else
980 #endif
981 	if (offset + len > vap->va_size) {
982 		/*
983 		 * Test offset + len against the filesystem's maxfilesize.
984 		 */
985 		VATTR_NULL(vap);
986 		vap->va_size = offset + len;
987 		error = VOP_SETATTR(vp, vap, td->td_ucred);
988 		if (error != 0)
989 			goto out;
990 		VATTR_NULL(vap);
991 		vap->va_size = fsize;
992 		error = VOP_SETATTR(vp, vap, td->td_ucred);
993 		if (error != 0)
994 			goto out;
995 	}
996 
997 	for (;;) {
998 		/*
999 		 * Read and write back anything below the nominal file
1000 		 * size.  There's currently no way outside the filesystem
1001 		 * to know whether this area is sparse or not.
1002 		 */
1003 		cur = iosize;
1004 		if ((offset % iosize) != 0)
1005 			cur -= (offset % iosize);
1006 		if (cur > len)
1007 			cur = len;
1008 		if (offset < fsize) {
1009 			aiov.iov_base = buf;
1010 			aiov.iov_len = cur;
1011 			auio.uio_iov = &aiov;
1012 			auio.uio_iovcnt = 1;
1013 			auio.uio_offset = offset;
1014 			auio.uio_resid = cur;
1015 			auio.uio_segflg = UIO_SYSSPACE;
1016 			auio.uio_rw = UIO_READ;
1017 			auio.uio_td = td;
1018 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
1019 			if (error != 0)
1020 				break;
1021 			if (auio.uio_resid > 0) {
1022 				bzero(buf + cur - auio.uio_resid,
1023 				    auio.uio_resid);
1024 			}
1025 		} else {
1026 			bzero(buf, cur);
1027 		}
1028 
1029 		aiov.iov_base = buf;
1030 		aiov.iov_len = cur;
1031 		auio.uio_iov = &aiov;
1032 		auio.uio_iovcnt = 1;
1033 		auio.uio_offset = offset;
1034 		auio.uio_resid = cur;
1035 		auio.uio_segflg = UIO_SYSSPACE;
1036 		auio.uio_rw = UIO_WRITE;
1037 		auio.uio_td = td;
1038 
1039 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
1040 		if (error != 0)
1041 			break;
1042 
1043 		len -= cur;
1044 		offset += cur;
1045 		if (len == 0)
1046 			break;
1047 		if (should_yield())
1048 			break;
1049 	}
1050 
1051  out:
1052 	*ap->a_len = len;
1053 	*ap->a_offset = offset;
1054 	free(buf, M_TEMP);
1055 	return (error);
1056 }
1057 
1058 int
1059 vop_stdadvise(struct vop_advise_args *ap)
1060 {
1061 	struct vnode *vp;
1062 	struct bufobj *bo;
1063 	daddr_t startn, endn;
1064 	off_t start, end;
1065 	int bsize, error;
1066 
1067 	vp = ap->a_vp;
1068 	switch (ap->a_advice) {
1069 	case POSIX_FADV_WILLNEED:
1070 		/*
1071 		 * Do nothing for now.  Filesystems should provide a
1072 		 * custom method which starts an asynchronous read of
1073 		 * the requested region.
1074 		 */
1075 		error = 0;
1076 		break;
1077 	case POSIX_FADV_DONTNEED:
1078 		error = 0;
1079 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1080 		if (vp->v_iflag & VI_DOOMED) {
1081 			VOP_UNLOCK(vp, 0);
1082 			break;
1083 		}
1084 
1085 		/*
1086 		 * Deactivate pages in the specified range from the backing VM
1087 		 * object.  Pages that are resident in the buffer cache will
1088 		 * remain wired until their corresponding buffers are released
1089 		 * below.
1090 		 */
1091 		if (vp->v_object != NULL) {
1092 			start = trunc_page(ap->a_start);
1093 			end = round_page(ap->a_end);
1094 			VM_OBJECT_WLOCK(vp->v_object);
1095 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1096 			    OFF_TO_IDX(end));
1097 			VM_OBJECT_WUNLOCK(vp->v_object);
1098 		}
1099 
1100 		bo = &vp->v_bufobj;
1101 		BO_RLOCK(bo);
1102 		bsize = vp->v_bufobj.bo_bsize;
1103 		startn = ap->a_start / bsize;
1104 		endn = ap->a_end / bsize;
1105 		error = bnoreuselist(&bo->bo_clean, bo, startn, endn);
1106 		if (error == 0)
1107 			error = bnoreuselist(&bo->bo_dirty, bo, startn, endn);
1108 		BO_RUNLOCK(bo);
1109 		VOP_UNLOCK(vp, 0);
1110 		break;
1111 	default:
1112 		error = EINVAL;
1113 		break;
1114 	}
1115 	return (error);
1116 }
1117 
1118 int
1119 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1120 {
1121 
1122 	ap->a_vp->v_socket = ap->a_socket;
1123 	return (0);
1124 }
1125 
1126 int
1127 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1128 {
1129 
1130 	*ap->a_socket = ap->a_vp->v_socket;
1131 	return (0);
1132 }
1133 
1134 int
1135 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1136 {
1137 
1138 	ap->a_vp->v_socket = NULL;
1139 	return (0);
1140 }
1141 
1142 static int
1143 vop_stdis_text(struct vop_is_text_args *ap)
1144 {
1145 
1146 	return ((ap->a_vp->v_vflag & VV_TEXT) != 0);
1147 }
1148 
1149 static int
1150 vop_stdset_text(struct vop_set_text_args *ap)
1151 {
1152 
1153 	ap->a_vp->v_vflag |= VV_TEXT;
1154 	return (0);
1155 }
1156 
1157 static int
1158 vop_stdunset_text(struct vop_unset_text_args *ap)
1159 {
1160 
1161 	ap->a_vp->v_vflag &= ~VV_TEXT;
1162 	return (0);
1163 }
1164 
1165 static int
1166 vop_stdget_writecount(struct vop_get_writecount_args *ap)
1167 {
1168 
1169 	*ap->a_writecount = ap->a_vp->v_writecount;
1170 	return (0);
1171 }
1172 
1173 static int
1174 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1175 {
1176 
1177 	ap->a_vp->v_writecount += ap->a_inc;
1178 	return (0);
1179 }
1180 
1181 /*
1182  * vfs default ops
1183  * used to fill the vfs function table to get reasonable default return values.
1184  */
1185 int
1186 vfs_stdroot (mp, flags, vpp)
1187 	struct mount *mp;
1188 	int flags;
1189 	struct vnode **vpp;
1190 {
1191 
1192 	return (EOPNOTSUPP);
1193 }
1194 
1195 int
1196 vfs_stdstatfs (mp, sbp)
1197 	struct mount *mp;
1198 	struct statfs *sbp;
1199 {
1200 
1201 	return (EOPNOTSUPP);
1202 }
1203 
1204 int
1205 vfs_stdquotactl (mp, cmds, uid, arg)
1206 	struct mount *mp;
1207 	int cmds;
1208 	uid_t uid;
1209 	void *arg;
1210 {
1211 
1212 	return (EOPNOTSUPP);
1213 }
1214 
1215 int
1216 vfs_stdsync(mp, waitfor)
1217 	struct mount *mp;
1218 	int waitfor;
1219 {
1220 	struct vnode *vp, *mvp;
1221 	struct thread *td;
1222 	int error, lockreq, allerror = 0;
1223 
1224 	td = curthread;
1225 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1226 	if (waitfor != MNT_WAIT)
1227 		lockreq |= LK_NOWAIT;
1228 	/*
1229 	 * Force stale buffer cache information to be flushed.
1230 	 */
1231 loop:
1232 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1233 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1234 			VI_UNLOCK(vp);
1235 			continue;
1236 		}
1237 		if ((error = vget(vp, lockreq, td)) != 0) {
1238 			if (error == ENOENT) {
1239 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1240 				goto loop;
1241 			}
1242 			continue;
1243 		}
1244 		error = VOP_FSYNC(vp, waitfor, td);
1245 		if (error)
1246 			allerror = error;
1247 		vput(vp);
1248 	}
1249 	return (allerror);
1250 }
1251 
1252 int
1253 vfs_stdnosync (mp, waitfor)
1254 	struct mount *mp;
1255 	int waitfor;
1256 {
1257 
1258 	return (0);
1259 }
1260 
1261 int
1262 vfs_stdvget (mp, ino, flags, vpp)
1263 	struct mount *mp;
1264 	ino_t ino;
1265 	int flags;
1266 	struct vnode **vpp;
1267 {
1268 
1269 	return (EOPNOTSUPP);
1270 }
1271 
1272 int
1273 vfs_stdfhtovp (mp, fhp, flags, vpp)
1274 	struct mount *mp;
1275 	struct fid *fhp;
1276 	int flags;
1277 	struct vnode **vpp;
1278 {
1279 
1280 	return (EOPNOTSUPP);
1281 }
1282 
1283 int
1284 vfs_stdinit (vfsp)
1285 	struct vfsconf *vfsp;
1286 {
1287 
1288 	return (0);
1289 }
1290 
1291 int
1292 vfs_stduninit (vfsp)
1293 	struct vfsconf *vfsp;
1294 {
1295 
1296 	return(0);
1297 }
1298 
1299 int
1300 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1301 	struct mount *mp;
1302 	int cmd;
1303 	struct vnode *filename_vp;
1304 	int attrnamespace;
1305 	const char *attrname;
1306 {
1307 
1308 	if (filename_vp != NULL)
1309 		VOP_UNLOCK(filename_vp, 0);
1310 	return (EOPNOTSUPP);
1311 }
1312 
1313 int
1314 vfs_stdsysctl(mp, op, req)
1315 	struct mount *mp;
1316 	fsctlop_t op;
1317 	struct sysctl_req *req;
1318 {
1319 
1320 	return (EOPNOTSUPP);
1321 }
1322 
1323 /* end of vfs default ops */
1324