xref: /freebsd/sys/kern/vfs_default.c (revision 55620f43deef5c0eb5b4b0f675de18b30c8d1c2d)
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_ASYNC_IO:
476 			*ap->a_retval = _POSIX_ASYNCHRONOUS_IO;
477 			return (0);
478 		case _PC_NAME_MAX:
479 			*ap->a_retval = NAME_MAX;
480 			return (0);
481 		case _PC_PATH_MAX:
482 			*ap->a_retval = PATH_MAX;
483 			return (0);
484 		case _PC_LINK_MAX:
485 			*ap->a_retval = LINK_MAX;
486 			return (0);
487 		case _PC_MAX_CANON:
488 			*ap->a_retval = MAX_CANON;
489 			return (0);
490 		case _PC_MAX_INPUT:
491 			*ap->a_retval = MAX_INPUT;
492 			return (0);
493 		case _PC_PIPE_BUF:
494 			*ap->a_retval = PIPE_BUF;
495 			return (0);
496 		case _PC_CHOWN_RESTRICTED:
497 			*ap->a_retval = 1;
498 			return (0);
499 		case _PC_VDISABLE:
500 			*ap->a_retval = _POSIX_VDISABLE;
501 			return (0);
502 		default:
503 			return (EINVAL);
504 	}
505 	/* NOTREACHED */
506 }
507 
508 /*
509  * Standard lock, unlock and islocked functions.
510  */
511 int
512 vop_stdlock(ap)
513 	struct vop_lock1_args /* {
514 		struct vnode *a_vp;
515 		int a_flags;
516 		char *file;
517 		int line;
518 	} */ *ap;
519 {
520 	struct vnode *vp = ap->a_vp;
521 
522 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
523 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
524 	    ap->a_line));
525 }
526 
527 /* See above. */
528 int
529 vop_stdunlock(ap)
530 	struct vop_unlock_args /* {
531 		struct vnode *a_vp;
532 		int a_flags;
533 	} */ *ap;
534 {
535 	struct vnode *vp = ap->a_vp;
536 
537 	return (lockmgr(vp->v_vnlock, ap->a_flags | LK_RELEASE, VI_MTX(vp)));
538 }
539 
540 /* See above. */
541 int
542 vop_stdislocked(ap)
543 	struct vop_islocked_args /* {
544 		struct vnode *a_vp;
545 	} */ *ap;
546 {
547 
548 	return (lockstatus(ap->a_vp->v_vnlock));
549 }
550 
551 /*
552  * Return true for select/poll.
553  */
554 int
555 vop_nopoll(ap)
556 	struct vop_poll_args /* {
557 		struct vnode *a_vp;
558 		int  a_events;
559 		struct ucred *a_cred;
560 		struct thread *a_td;
561 	} */ *ap;
562 {
563 
564 	return (poll_no_poll(ap->a_events));
565 }
566 
567 /*
568  * Implement poll for local filesystems that support it.
569  */
570 int
571 vop_stdpoll(ap)
572 	struct vop_poll_args /* {
573 		struct vnode *a_vp;
574 		int  a_events;
575 		struct ucred *a_cred;
576 		struct thread *a_td;
577 	} */ *ap;
578 {
579 	if (ap->a_events & ~POLLSTANDARD)
580 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
581 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
582 }
583 
584 /*
585  * Return our mount point, as we will take charge of the writes.
586  */
587 int
588 vop_stdgetwritemount(ap)
589 	struct vop_getwritemount_args /* {
590 		struct vnode *a_vp;
591 		struct mount **a_mpp;
592 	} */ *ap;
593 {
594 	struct mount *mp;
595 
596 	/*
597 	 * XXX Since this is called unlocked we may be recycled while
598 	 * attempting to ref the mount.  If this is the case or mountpoint
599 	 * will be set to NULL.  We only have to prevent this call from
600 	 * returning with a ref to an incorrect mountpoint.  It is not
601 	 * harmful to return with a ref to our previous mountpoint.
602 	 */
603 	mp = ap->a_vp->v_mount;
604 	if (mp != NULL) {
605 		vfs_ref(mp);
606 		if (mp != ap->a_vp->v_mount) {
607 			vfs_rel(mp);
608 			mp = NULL;
609 		}
610 	}
611 	*(ap->a_mpp) = mp;
612 	return (0);
613 }
614 
615 /* XXX Needs good comment and VOP_BMAP(9) manpage */
616 int
617 vop_stdbmap(ap)
618 	struct vop_bmap_args /* {
619 		struct vnode *a_vp;
620 		daddr_t  a_bn;
621 		struct bufobj **a_bop;
622 		daddr_t *a_bnp;
623 		int *a_runp;
624 		int *a_runb;
625 	} */ *ap;
626 {
627 
628 	if (ap->a_bop != NULL)
629 		*ap->a_bop = &ap->a_vp->v_bufobj;
630 	if (ap->a_bnp != NULL)
631 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
632 	if (ap->a_runp != NULL)
633 		*ap->a_runp = 0;
634 	if (ap->a_runb != NULL)
635 		*ap->a_runb = 0;
636 	return (0);
637 }
638 
639 int
640 vop_stdfsync(ap)
641 	struct vop_fsync_args /* {
642 		struct vnode *a_vp;
643 		struct ucred *a_cred;
644 		int a_waitfor;
645 		struct thread *a_td;
646 	} */ *ap;
647 {
648 	struct vnode *vp = ap->a_vp;
649 	struct buf *bp;
650 	struct bufobj *bo;
651 	struct buf *nbp;
652 	int error = 0;
653 	int maxretry = 1000;     /* large, arbitrarily chosen */
654 
655 	bo = &vp->v_bufobj;
656 	BO_LOCK(bo);
657 loop1:
658 	/*
659 	 * MARK/SCAN initialization to avoid infinite loops.
660 	 */
661         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
662                 bp->b_vflags &= ~BV_SCANNED;
663 		bp->b_error = 0;
664 	}
665 
666 	/*
667 	 * Flush all dirty buffers associated with a vnode.
668 	 */
669 loop2:
670 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
671 		if ((bp->b_vflags & BV_SCANNED) != 0)
672 			continue;
673 		bp->b_vflags |= BV_SCANNED;
674 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
675 			if (ap->a_waitfor != MNT_WAIT)
676 				continue;
677 			if (BUF_LOCK(bp,
678 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
679 			    BO_LOCKPTR(bo)) != 0) {
680 				BO_LOCK(bo);
681 				goto loop1;
682 			}
683 			BO_LOCK(bo);
684 		}
685 		BO_UNLOCK(bo);
686 		KASSERT(bp->b_bufobj == bo,
687 		    ("bp %p wrong b_bufobj %p should be %p",
688 		    bp, bp->b_bufobj, bo));
689 		if ((bp->b_flags & B_DELWRI) == 0)
690 			panic("fsync: not dirty");
691 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
692 			vfs_bio_awrite(bp);
693 		} else {
694 			bremfree(bp);
695 			bawrite(bp);
696 		}
697 		BO_LOCK(bo);
698 		goto loop2;
699 	}
700 
701 	/*
702 	 * If synchronous the caller expects us to completely resolve all
703 	 * dirty buffers in the system.  Wait for in-progress I/O to
704 	 * complete (which could include background bitmap writes), then
705 	 * retry if dirty blocks still exist.
706 	 */
707 	if (ap->a_waitfor == MNT_WAIT) {
708 		bufobj_wwait(bo, 0, 0);
709 		if (bo->bo_dirty.bv_cnt > 0) {
710 			/*
711 			 * If we are unable to write any of these buffers
712 			 * then we fail now rather than trying endlessly
713 			 * to write them out.
714 			 */
715 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
716 				if ((error = bp->b_error) == 0)
717 					continue;
718 			if (error == 0 && --maxretry >= 0)
719 				goto loop1;
720 			error = EAGAIN;
721 		}
722 	}
723 	BO_UNLOCK(bo);
724 	if (error == EAGAIN)
725 		vprint("fsync: giving up on dirty", vp);
726 
727 	return (error);
728 }
729 
730 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
731 int
732 vop_stdgetpages(ap)
733 	struct vop_getpages_args /* {
734 		struct vnode *a_vp;
735 		vm_page_t *a_m;
736 		int a_count;
737 		int *a_rbehind;
738 		int *a_rahead;
739 	} */ *ap;
740 {
741 
742 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
743 	    ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL);
744 }
745 
746 static int
747 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
748 {
749 	int error;
750 
751 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
752 	    ap->a_rahead);
753 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
754 	return (error);
755 }
756 
757 int
758 vop_stdkqfilter(struct vop_kqfilter_args *ap)
759 {
760 	return vfs_kqfilter(ap);
761 }
762 
763 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
764 int
765 vop_stdputpages(ap)
766 	struct vop_putpages_args /* {
767 		struct vnode *a_vp;
768 		vm_page_t *a_m;
769 		int a_count;
770 		int a_sync;
771 		int *a_rtvals;
772 	} */ *ap;
773 {
774 
775 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
776 	     ap->a_sync, ap->a_rtvals);
777 }
778 
779 int
780 vop_stdvptofh(struct vop_vptofh_args *ap)
781 {
782 	return (EOPNOTSUPP);
783 }
784 
785 int
786 vop_stdvptocnp(struct vop_vptocnp_args *ap)
787 {
788 	struct vnode *vp = ap->a_vp;
789 	struct vnode **dvp = ap->a_vpp;
790 	struct ucred *cred = ap->a_cred;
791 	char *buf = ap->a_buf;
792 	int *buflen = ap->a_buflen;
793 	char *dirbuf, *cpos;
794 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
795 	off_t off;
796 	ino_t fileno;
797 	struct vattr va;
798 	struct nameidata nd;
799 	struct thread *td;
800 	struct dirent *dp;
801 	struct vnode *mvp;
802 
803 	i = *buflen;
804 	error = 0;
805 	covered = 0;
806 	td = curthread;
807 
808 	if (vp->v_type != VDIR)
809 		return (ENOENT);
810 
811 	error = VOP_GETATTR(vp, &va, cred);
812 	if (error)
813 		return (error);
814 
815 	VREF(vp);
816 	locked = VOP_ISLOCKED(vp);
817 	VOP_UNLOCK(vp, 0);
818 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
819 	    "..", vp, td);
820 	flags = FREAD;
821 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
822 	if (error) {
823 		vn_lock(vp, locked | LK_RETRY);
824 		return (error);
825 	}
826 	NDFREE(&nd, NDF_ONLY_PNBUF);
827 
828 	mvp = *dvp = nd.ni_vp;
829 
830 	if (vp->v_mount != (*dvp)->v_mount &&
831 	    ((*dvp)->v_vflag & VV_ROOT) &&
832 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
833 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
834 		VREF(mvp);
835 		VOP_UNLOCK(mvp, 0);
836 		vn_close(mvp, FREAD, cred, td);
837 		VREF(*dvp);
838 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
839 		covered = 1;
840 	}
841 
842 	fileno = va.va_fileid;
843 
844 	dirbuflen = DEV_BSIZE;
845 	if (dirbuflen < va.va_blocksize)
846 		dirbuflen = va.va_blocksize;
847 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
848 
849 	if ((*dvp)->v_type != VDIR) {
850 		error = ENOENT;
851 		goto out;
852 	}
853 
854 	off = 0;
855 	len = 0;
856 	do {
857 		/* call VOP_READDIR of parent */
858 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
859 					&cpos, &len, &eofflag, td);
860 		if (error)
861 			goto out;
862 
863 		if ((dp->d_type != DT_WHT) &&
864 		    (dp->d_fileno == fileno)) {
865 			if (covered) {
866 				VOP_UNLOCK(*dvp, 0);
867 				vn_lock(mvp, LK_SHARED | LK_RETRY);
868 				if (dirent_exists(mvp, dp->d_name, td)) {
869 					error = ENOENT;
870 					VOP_UNLOCK(mvp, 0);
871 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
872 					goto out;
873 				}
874 				VOP_UNLOCK(mvp, 0);
875 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
876 			}
877 			i -= dp->d_namlen;
878 
879 			if (i < 0) {
880 				error = ENOMEM;
881 				goto out;
882 			}
883 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
884 				error = ENOENT;
885 			} else {
886 				bcopy(dp->d_name, buf + i, dp->d_namlen);
887 				error = 0;
888 			}
889 			goto out;
890 		}
891 	} while (len > 0 || !eofflag);
892 	error = ENOENT;
893 
894 out:
895 	free(dirbuf, M_TEMP);
896 	if (!error) {
897 		*buflen = i;
898 		vref(*dvp);
899 	}
900 	if (covered) {
901 		vput(*dvp);
902 		vrele(mvp);
903 	} else {
904 		VOP_UNLOCK(mvp, 0);
905 		vn_close(mvp, FREAD, cred, td);
906 	}
907 	vn_lock(vp, locked | LK_RETRY);
908 	return (error);
909 }
910 
911 int
912 vop_stdallocate(struct vop_allocate_args *ap)
913 {
914 #ifdef __notyet__
915 	struct statfs sfs;
916 #endif
917 	struct iovec aiov;
918 	struct vattr vattr, *vap;
919 	struct uio auio;
920 	off_t fsize, len, cur, offset;
921 	uint8_t *buf;
922 	struct thread *td;
923 	struct vnode *vp;
924 	size_t iosize;
925 	int error;
926 
927 	buf = NULL;
928 	error = 0;
929 	td = curthread;
930 	vap = &vattr;
931 	vp = ap->a_vp;
932 	len = *ap->a_len;
933 	offset = *ap->a_offset;
934 
935 	error = VOP_GETATTR(vp, vap, td->td_ucred);
936 	if (error != 0)
937 		goto out;
938 	fsize = vap->va_size;
939 	iosize = vap->va_blocksize;
940 	if (iosize == 0)
941 		iosize = BLKDEV_IOSIZE;
942 	if (iosize > MAXPHYS)
943 		iosize = MAXPHYS;
944 	buf = malloc(iosize, M_TEMP, M_WAITOK);
945 
946 #ifdef __notyet__
947 	/*
948 	 * Check if the filesystem sets f_maxfilesize; if not use
949 	 * VOP_SETATTR to perform the check.
950 	 */
951 	error = VFS_STATFS(vp->v_mount, &sfs, td);
952 	if (error != 0)
953 		goto out;
954 	if (sfs.f_maxfilesize) {
955 		if (offset > sfs.f_maxfilesize || len > sfs.f_maxfilesize ||
956 		    offset + len > sfs.f_maxfilesize) {
957 			error = EFBIG;
958 			goto out;
959 		}
960 	} else
961 #endif
962 	if (offset + len > vap->va_size) {
963 		/*
964 		 * Test offset + len against the filesystem's maxfilesize.
965 		 */
966 		VATTR_NULL(vap);
967 		vap->va_size = offset + len;
968 		error = VOP_SETATTR(vp, vap, td->td_ucred);
969 		if (error != 0)
970 			goto out;
971 		VATTR_NULL(vap);
972 		vap->va_size = fsize;
973 		error = VOP_SETATTR(vp, vap, td->td_ucred);
974 		if (error != 0)
975 			goto out;
976 	}
977 
978 	for (;;) {
979 		/*
980 		 * Read and write back anything below the nominal file
981 		 * size.  There's currently no way outside the filesystem
982 		 * to know whether this area is sparse or not.
983 		 */
984 		cur = iosize;
985 		if ((offset % iosize) != 0)
986 			cur -= (offset % iosize);
987 		if (cur > len)
988 			cur = len;
989 		if (offset < fsize) {
990 			aiov.iov_base = buf;
991 			aiov.iov_len = cur;
992 			auio.uio_iov = &aiov;
993 			auio.uio_iovcnt = 1;
994 			auio.uio_offset = offset;
995 			auio.uio_resid = cur;
996 			auio.uio_segflg = UIO_SYSSPACE;
997 			auio.uio_rw = UIO_READ;
998 			auio.uio_td = td;
999 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
1000 			if (error != 0)
1001 				break;
1002 			if (auio.uio_resid > 0) {
1003 				bzero(buf + cur - auio.uio_resid,
1004 				    auio.uio_resid);
1005 			}
1006 		} else {
1007 			bzero(buf, cur);
1008 		}
1009 
1010 		aiov.iov_base = buf;
1011 		aiov.iov_len = cur;
1012 		auio.uio_iov = &aiov;
1013 		auio.uio_iovcnt = 1;
1014 		auio.uio_offset = offset;
1015 		auio.uio_resid = cur;
1016 		auio.uio_segflg = UIO_SYSSPACE;
1017 		auio.uio_rw = UIO_WRITE;
1018 		auio.uio_td = td;
1019 
1020 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
1021 		if (error != 0)
1022 			break;
1023 
1024 		len -= cur;
1025 		offset += cur;
1026 		if (len == 0)
1027 			break;
1028 		if (should_yield())
1029 			break;
1030 	}
1031 
1032  out:
1033 	*ap->a_len = len;
1034 	*ap->a_offset = offset;
1035 	free(buf, M_TEMP);
1036 	return (error);
1037 }
1038 
1039 int
1040 vop_stdadvise(struct vop_advise_args *ap)
1041 {
1042 	struct vnode *vp;
1043 	struct bufobj *bo;
1044 	daddr_t startn, endn;
1045 	off_t start, end;
1046 	int bsize, error;
1047 
1048 	vp = ap->a_vp;
1049 	switch (ap->a_advice) {
1050 	case POSIX_FADV_WILLNEED:
1051 		/*
1052 		 * Do nothing for now.  Filesystems should provide a
1053 		 * custom method which starts an asynchronous read of
1054 		 * the requested region.
1055 		 */
1056 		error = 0;
1057 		break;
1058 	case POSIX_FADV_DONTNEED:
1059 		error = 0;
1060 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1061 		if (vp->v_iflag & VI_DOOMED) {
1062 			VOP_UNLOCK(vp, 0);
1063 			break;
1064 		}
1065 
1066 		/*
1067 		 * Deactivate pages in the specified range from the backing VM
1068 		 * object.  Pages that are resident in the buffer cache will
1069 		 * remain wired until their corresponding buffers are released
1070 		 * below.
1071 		 */
1072 		if (vp->v_object != NULL) {
1073 			start = trunc_page(ap->a_start);
1074 			end = round_page(ap->a_end);
1075 			VM_OBJECT_WLOCK(vp->v_object);
1076 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1077 			    OFF_TO_IDX(end));
1078 			VM_OBJECT_WUNLOCK(vp->v_object);
1079 		}
1080 
1081 		bo = &vp->v_bufobj;
1082 		BO_RLOCK(bo);
1083 		bsize = vp->v_bufobj.bo_bsize;
1084 		startn = ap->a_start / bsize;
1085 		endn = ap->a_end / bsize;
1086 		error = bnoreuselist(&bo->bo_clean, bo, startn, endn);
1087 		if (error == 0)
1088 			error = bnoreuselist(&bo->bo_dirty, bo, startn, endn);
1089 		BO_RUNLOCK(bo);
1090 		VOP_UNLOCK(vp, 0);
1091 		break;
1092 	default:
1093 		error = EINVAL;
1094 		break;
1095 	}
1096 	return (error);
1097 }
1098 
1099 int
1100 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1101 {
1102 
1103 	ap->a_vp->v_socket = ap->a_socket;
1104 	return (0);
1105 }
1106 
1107 int
1108 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1109 {
1110 
1111 	*ap->a_socket = ap->a_vp->v_socket;
1112 	return (0);
1113 }
1114 
1115 int
1116 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1117 {
1118 
1119 	ap->a_vp->v_socket = NULL;
1120 	return (0);
1121 }
1122 
1123 static int
1124 vop_stdis_text(struct vop_is_text_args *ap)
1125 {
1126 
1127 	return ((ap->a_vp->v_vflag & VV_TEXT) != 0);
1128 }
1129 
1130 static int
1131 vop_stdset_text(struct vop_set_text_args *ap)
1132 {
1133 
1134 	ap->a_vp->v_vflag |= VV_TEXT;
1135 	return (0);
1136 }
1137 
1138 static int
1139 vop_stdunset_text(struct vop_unset_text_args *ap)
1140 {
1141 
1142 	ap->a_vp->v_vflag &= ~VV_TEXT;
1143 	return (0);
1144 }
1145 
1146 static int
1147 vop_stdget_writecount(struct vop_get_writecount_args *ap)
1148 {
1149 
1150 	*ap->a_writecount = ap->a_vp->v_writecount;
1151 	return (0);
1152 }
1153 
1154 static int
1155 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1156 {
1157 
1158 	ap->a_vp->v_writecount += ap->a_inc;
1159 	return (0);
1160 }
1161 
1162 /*
1163  * vfs default ops
1164  * used to fill the vfs function table to get reasonable default return values.
1165  */
1166 int
1167 vfs_stdroot (mp, flags, vpp)
1168 	struct mount *mp;
1169 	int flags;
1170 	struct vnode **vpp;
1171 {
1172 
1173 	return (EOPNOTSUPP);
1174 }
1175 
1176 int
1177 vfs_stdstatfs (mp, sbp)
1178 	struct mount *mp;
1179 	struct statfs *sbp;
1180 {
1181 
1182 	return (EOPNOTSUPP);
1183 }
1184 
1185 int
1186 vfs_stdquotactl (mp, cmds, uid, arg)
1187 	struct mount *mp;
1188 	int cmds;
1189 	uid_t uid;
1190 	void *arg;
1191 {
1192 
1193 	return (EOPNOTSUPP);
1194 }
1195 
1196 int
1197 vfs_stdsync(mp, waitfor)
1198 	struct mount *mp;
1199 	int waitfor;
1200 {
1201 	struct vnode *vp, *mvp;
1202 	struct thread *td;
1203 	int error, lockreq, allerror = 0;
1204 
1205 	td = curthread;
1206 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1207 	if (waitfor != MNT_WAIT)
1208 		lockreq |= LK_NOWAIT;
1209 	/*
1210 	 * Force stale buffer cache information to be flushed.
1211 	 */
1212 loop:
1213 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1214 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1215 			VI_UNLOCK(vp);
1216 			continue;
1217 		}
1218 		if ((error = vget(vp, lockreq, td)) != 0) {
1219 			if (error == ENOENT) {
1220 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1221 				goto loop;
1222 			}
1223 			continue;
1224 		}
1225 		error = VOP_FSYNC(vp, waitfor, td);
1226 		if (error)
1227 			allerror = error;
1228 		vput(vp);
1229 	}
1230 	return (allerror);
1231 }
1232 
1233 int
1234 vfs_stdnosync (mp, waitfor)
1235 	struct mount *mp;
1236 	int waitfor;
1237 {
1238 
1239 	return (0);
1240 }
1241 
1242 int
1243 vfs_stdvget (mp, ino, flags, vpp)
1244 	struct mount *mp;
1245 	ino_t ino;
1246 	int flags;
1247 	struct vnode **vpp;
1248 {
1249 
1250 	return (EOPNOTSUPP);
1251 }
1252 
1253 int
1254 vfs_stdfhtovp (mp, fhp, flags, vpp)
1255 	struct mount *mp;
1256 	struct fid *fhp;
1257 	int flags;
1258 	struct vnode **vpp;
1259 {
1260 
1261 	return (EOPNOTSUPP);
1262 }
1263 
1264 int
1265 vfs_stdinit (vfsp)
1266 	struct vfsconf *vfsp;
1267 {
1268 
1269 	return (0);
1270 }
1271 
1272 int
1273 vfs_stduninit (vfsp)
1274 	struct vfsconf *vfsp;
1275 {
1276 
1277 	return(0);
1278 }
1279 
1280 int
1281 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1282 	struct mount *mp;
1283 	int cmd;
1284 	struct vnode *filename_vp;
1285 	int attrnamespace;
1286 	const char *attrname;
1287 {
1288 
1289 	if (filename_vp != NULL)
1290 		VOP_UNLOCK(filename_vp, 0);
1291 	return (EOPNOTSUPP);
1292 }
1293 
1294 int
1295 vfs_stdsysctl(mp, op, req)
1296 	struct mount *mp;
1297 	fsctlop_t op;
1298 	struct sysctl_req *req;
1299 {
1300 
1301 	return (EOPNOTSUPP);
1302 }
1303 
1304 /* end of vfs default ops */
1305