xref: /freebsd/sys/kern/vfs_default.c (revision 137a344c6341d1469432e9deb3a25593f96672ad)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed
8  * to Berkeley by John Heidemann of the UCLA Ficus project.
9  *
10  * Source: * @(#)i405_init.c 2.10 92/04/27 UCLA Ficus project
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/bio.h>
43 #include <sys/buf.h>
44 #include <sys/conf.h>
45 #include <sys/event.h>
46 #include <sys/kernel.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/lockf.h>
50 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/namei.h>
53 #include <sys/rwlock.h>
54 #include <sys/fcntl.h>
55 #include <sys/unistd.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/poll.h>
59 
60 #include <security/mac/mac_framework.h>
61 
62 #include <vm/vm.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_extern.h>
65 #include <vm/pmap.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_pager.h>
69 #include <vm/vnode_pager.h>
70 
71 static int	vop_nolookup(struct vop_lookup_args *);
72 static int	vop_norename(struct vop_rename_args *);
73 static int	vop_nostrategy(struct vop_strategy_args *);
74 static int	get_next_dirent(struct vnode *vp, struct dirent **dpp,
75 				char *dirbuf, int dirbuflen, off_t *off,
76 				char **cpos, int *len, int *eofflag,
77 				struct thread *td);
78 static int	dirent_exists(struct vnode *vp, const char *dirname,
79 			      struct thread *td);
80 
81 #define DIRENT_MINSIZE (sizeof(struct dirent) - (MAXNAMLEN+1) + 4)
82 
83 static int vop_stdis_text(struct vop_is_text_args *ap);
84 static int vop_stdset_text(struct vop_set_text_args *ap);
85 static int vop_stdunset_text(struct vop_unset_text_args *ap);
86 static int vop_stdget_writecount(struct vop_get_writecount_args *ap);
87 static int vop_stdadd_writecount(struct vop_add_writecount_args *ap);
88 static int vop_stdfdatasync(struct vop_fdatasync_args *ap);
89 static int vop_stdgetpages_async(struct vop_getpages_async_args *ap);
90 
91 /*
92  * This vnode table stores what we want to do if the filesystem doesn't
93  * implement a particular VOP.
94  *
95  * If there is no specific entry here, we will return EOPNOTSUPP.
96  *
97  * Note that every filesystem has to implement either vop_access
98  * or vop_accessx; failing to do so will result in immediate crash
99  * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(),
100  * which calls vop_stdaccess() etc.
101  */
102 
103 struct vop_vector default_vnodeops = {
104 	.vop_default =		NULL,
105 	.vop_bypass =		VOP_EOPNOTSUPP,
106 
107 	.vop_access =		vop_stdaccess,
108 	.vop_accessx =		vop_stdaccessx,
109 	.vop_advise =		vop_stdadvise,
110 	.vop_advlock =		vop_stdadvlock,
111 	.vop_advlockasync =	vop_stdadvlockasync,
112 	.vop_advlockpurge =	vop_stdadvlockpurge,
113 	.vop_allocate =		vop_stdallocate,
114 	.vop_bmap =		vop_stdbmap,
115 	.vop_close =		VOP_NULL,
116 	.vop_fsync =		VOP_NULL,
117 	.vop_fdatasync =	vop_stdfdatasync,
118 	.vop_getpages =		vop_stdgetpages,
119 	.vop_getpages_async =	vop_stdgetpages_async,
120 	.vop_getwritemount = 	vop_stdgetwritemount,
121 	.vop_inactive =		VOP_NULL,
122 	.vop_ioctl =		VOP_ENOTTY,
123 	.vop_kqfilter =		vop_stdkqfilter,
124 	.vop_islocked =		vop_stdislocked,
125 	.vop_lock1 =		vop_stdlock,
126 	.vop_lookup =		vop_nolookup,
127 	.vop_open =		VOP_NULL,
128 	.vop_pathconf =		VOP_EINVAL,
129 	.vop_poll =		vop_nopoll,
130 	.vop_putpages =		vop_stdputpages,
131 	.vop_readlink =		VOP_EINVAL,
132 	.vop_rename =		vop_norename,
133 	.vop_revoke =		VOP_PANIC,
134 	.vop_strategy =		vop_nostrategy,
135 	.vop_unlock =		vop_stdunlock,
136 	.vop_vptocnp =		vop_stdvptocnp,
137 	.vop_vptofh =		vop_stdvptofh,
138 	.vop_unp_bind =		vop_stdunp_bind,
139 	.vop_unp_connect =	vop_stdunp_connect,
140 	.vop_unp_detach =	vop_stdunp_detach,
141 	.vop_is_text =		vop_stdis_text,
142 	.vop_set_text =		vop_stdset_text,
143 	.vop_unset_text =	vop_stdunset_text,
144 	.vop_get_writecount =	vop_stdget_writecount,
145 	.vop_add_writecount =	vop_stdadd_writecount,
146 };
147 
148 /*
149  * Series of placeholder functions for various error returns for
150  * VOPs.
151  */
152 
153 int
154 vop_eopnotsupp(struct vop_generic_args *ap)
155 {
156 	/*
157 	printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name);
158 	*/
159 
160 	return (EOPNOTSUPP);
161 }
162 
163 int
164 vop_ebadf(struct vop_generic_args *ap)
165 {
166 
167 	return (EBADF);
168 }
169 
170 int
171 vop_enotty(struct vop_generic_args *ap)
172 {
173 
174 	return (ENOTTY);
175 }
176 
177 int
178 vop_einval(struct vop_generic_args *ap)
179 {
180 
181 	return (EINVAL);
182 }
183 
184 int
185 vop_enoent(struct vop_generic_args *ap)
186 {
187 
188 	return (ENOENT);
189 }
190 
191 int
192 vop_null(struct vop_generic_args *ap)
193 {
194 
195 	return (0);
196 }
197 
198 /*
199  * Helper function to panic on some bad VOPs in some filesystems.
200  */
201 int
202 vop_panic(struct vop_generic_args *ap)
203 {
204 
205 	panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name);
206 }
207 
208 /*
209  * vop_std<something> and vop_no<something> are default functions for use by
210  * filesystems that need the "default reasonable" implementation for a
211  * particular operation.
212  *
213  * The documentation for the operations they implement exists (if it exists)
214  * in the VOP_<SOMETHING>(9) manpage (all uppercase).
215  */
216 
217 /*
218  * Default vop for filesystems that do not support name lookup
219  */
220 static int
221 vop_nolookup(ap)
222 	struct vop_lookup_args /* {
223 		struct vnode *a_dvp;
224 		struct vnode **a_vpp;
225 		struct componentname *a_cnp;
226 	} */ *ap;
227 {
228 
229 	*ap->a_vpp = NULL;
230 	return (ENOTDIR);
231 }
232 
233 /*
234  * vop_norename:
235  *
236  * Handle unlock and reference counting for arguments of vop_rename
237  * for filesystems that do not implement rename operation.
238  */
239 static int
240 vop_norename(struct vop_rename_args *ap)
241 {
242 
243 	vop_rename_fail(ap);
244 	return (EOPNOTSUPP);
245 }
246 
247 /*
248  *	vop_nostrategy:
249  *
250  *	Strategy routine for VFS devices that have none.
251  *
252  *	BIO_ERROR and B_INVAL must be cleared prior to calling any strategy
253  *	routine.  Typically this is done for a BIO_READ strategy call.
254  *	Typically B_INVAL is assumed to already be clear prior to a write
255  *	and should not be cleared manually unless you just made the buffer
256  *	invalid.  BIO_ERROR should be cleared either way.
257  */
258 
259 static int
260 vop_nostrategy (struct vop_strategy_args *ap)
261 {
262 	printf("No strategy for buffer at %p\n", ap->a_bp);
263 	vn_printf(ap->a_vp, "vnode ");
264 	ap->a_bp->b_ioflags |= BIO_ERROR;
265 	ap->a_bp->b_error = EOPNOTSUPP;
266 	bufdone(ap->a_bp);
267 	return (EOPNOTSUPP);
268 }
269 
270 static int
271 get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf,
272 		int dirbuflen, off_t *off, char **cpos, int *len,
273 		int *eofflag, struct thread *td)
274 {
275 	int error, reclen;
276 	struct uio uio;
277 	struct iovec iov;
278 	struct dirent *dp;
279 
280 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
281 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
282 
283 	if (*len == 0) {
284 		iov.iov_base = dirbuf;
285 		iov.iov_len = dirbuflen;
286 
287 		uio.uio_iov = &iov;
288 		uio.uio_iovcnt = 1;
289 		uio.uio_offset = *off;
290 		uio.uio_resid = dirbuflen;
291 		uio.uio_segflg = UIO_SYSSPACE;
292 		uio.uio_rw = UIO_READ;
293 		uio.uio_td = td;
294 
295 		*eofflag = 0;
296 
297 #ifdef MAC
298 		error = mac_vnode_check_readdir(td->td_ucred, vp);
299 		if (error == 0)
300 #endif
301 			error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag,
302 		    		NULL, NULL);
303 		if (error)
304 			return (error);
305 
306 		*off = uio.uio_offset;
307 
308 		*cpos = dirbuf;
309 		*len = (dirbuflen - uio.uio_resid);
310 
311 		if (*len == 0)
312 			return (ENOENT);
313 	}
314 
315 	dp = (struct dirent *)(*cpos);
316 	reclen = dp->d_reclen;
317 	*dpp = dp;
318 
319 	/* check for malformed directory.. */
320 	if (reclen < DIRENT_MINSIZE)
321 		return (EINVAL);
322 
323 	*cpos += reclen;
324 	*len -= reclen;
325 
326 	return (0);
327 }
328 
329 /*
330  * Check if a named file exists in a given directory vnode.
331  */
332 static int
333 dirent_exists(struct vnode *vp, const char *dirname, struct thread *td)
334 {
335 	char *dirbuf, *cpos;
336 	int error, eofflag, dirbuflen, len, found;
337 	off_t off;
338 	struct dirent *dp;
339 	struct vattr va;
340 
341 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
342 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
343 
344 	found = 0;
345 
346 	error = VOP_GETATTR(vp, &va, td->td_ucred);
347 	if (error)
348 		return (found);
349 
350 	dirbuflen = DEV_BSIZE;
351 	if (dirbuflen < va.va_blocksize)
352 		dirbuflen = va.va_blocksize;
353 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
354 
355 	off = 0;
356 	len = 0;
357 	do {
358 		error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off,
359 					&cpos, &len, &eofflag, td);
360 		if (error)
361 			goto out;
362 
363 		if (dp->d_type != DT_WHT && dp->d_fileno != 0 &&
364 		    strcmp(dp->d_name, dirname) == 0) {
365 			found = 1;
366 			goto out;
367 		}
368 	} while (len > 0 || !eofflag);
369 
370 out:
371 	free(dirbuf, M_TEMP);
372 	return (found);
373 }
374 
375 int
376 vop_stdaccess(struct vop_access_args *ap)
377 {
378 
379 	KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN |
380 	    VAPPEND)) == 0, ("invalid bit in accmode"));
381 
382 	return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td));
383 }
384 
385 int
386 vop_stdaccessx(struct vop_accessx_args *ap)
387 {
388 	int error;
389 	accmode_t accmode = ap->a_accmode;
390 
391 	error = vfs_unixify_accmode(&accmode);
392 	if (error != 0)
393 		return (error);
394 
395 	if (accmode == 0)
396 		return (0);
397 
398 	return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td));
399 }
400 
401 /*
402  * Advisory record locking support
403  */
404 int
405 vop_stdadvlock(struct vop_advlock_args *ap)
406 {
407 	struct vnode *vp;
408 	struct vattr vattr;
409 	int error;
410 
411 	vp = ap->a_vp;
412 	if (ap->a_fl->l_whence == SEEK_END) {
413 		/*
414 		 * The NFSv4 server must avoid doing a vn_lock() here, since it
415 		 * can deadlock the nfsd threads, due to a LOR.  Fortunately
416 		 * the NFSv4 server always uses SEEK_SET and this code is
417 		 * only required for the SEEK_END case.
418 		 */
419 		vn_lock(vp, LK_SHARED | LK_RETRY);
420 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
421 		VOP_UNLOCK(vp, 0);
422 		if (error)
423 			return (error);
424 	} else
425 		vattr.va_size = 0;
426 
427 	return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size));
428 }
429 
430 int
431 vop_stdadvlockasync(struct vop_advlockasync_args *ap)
432 {
433 	struct vnode *vp;
434 	struct vattr vattr;
435 	int error;
436 
437 	vp = ap->a_vp;
438 	if (ap->a_fl->l_whence == SEEK_END) {
439 		/* The size argument is only needed for SEEK_END. */
440 		vn_lock(vp, LK_SHARED | LK_RETRY);
441 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
442 		VOP_UNLOCK(vp, 0);
443 		if (error)
444 			return (error);
445 	} else
446 		vattr.va_size = 0;
447 
448 	return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size));
449 }
450 
451 int
452 vop_stdadvlockpurge(struct vop_advlockpurge_args *ap)
453 {
454 	struct vnode *vp;
455 
456 	vp = ap->a_vp;
457 	lf_purgelocks(vp, &vp->v_lockf);
458 	return (0);
459 }
460 
461 /*
462  * vop_stdpathconf:
463  *
464  * Standard implementation of POSIX pathconf, to get information about limits
465  * for a filesystem.
466  * Override per filesystem for the case where the filesystem has smaller
467  * limits.
468  */
469 int
470 vop_stdpathconf(ap)
471 	struct vop_pathconf_args /* {
472 	struct vnode *a_vp;
473 	int a_name;
474 	int *a_retval;
475 	} */ *ap;
476 {
477 
478 	switch (ap->a_name) {
479 		case _PC_ASYNC_IO:
480 			*ap->a_retval = _POSIX_ASYNCHRONOUS_IO;
481 			return (0);
482 		case _PC_PATH_MAX:
483 			*ap->a_retval = PATH_MAX;
484 			return (0);
485 		default:
486 			return (EINVAL);
487 	}
488 	/* NOTREACHED */
489 }
490 
491 /*
492  * Standard lock, unlock and islocked functions.
493  */
494 int
495 vop_stdlock(ap)
496 	struct vop_lock1_args /* {
497 		struct vnode *a_vp;
498 		int a_flags;
499 		char *file;
500 		int line;
501 	} */ *ap;
502 {
503 	struct vnode *vp = ap->a_vp;
504 	struct mtx *ilk;
505 
506 	ilk = VI_MTX(vp);
507 	return (lockmgr_lock_fast_path(vp->v_vnlock, ap->a_flags,
508 	    (ilk != NULL) ? &ilk->lock_object : NULL, ap->a_file, ap->a_line));
509 }
510 
511 /* See above. */
512 int
513 vop_stdunlock(ap)
514 	struct vop_unlock_args /* {
515 		struct vnode *a_vp;
516 		int a_flags;
517 	} */ *ap;
518 {
519 	struct vnode *vp = ap->a_vp;
520 	struct mtx *ilk;
521 
522 	ilk = VI_MTX(vp);
523 	return (lockmgr_unlock_fast_path(vp->v_vnlock, ap->a_flags,
524 	    (ilk != NULL) ? &ilk->lock_object : NULL));
525 }
526 
527 /* See above. */
528 int
529 vop_stdislocked(ap)
530 	struct vop_islocked_args /* {
531 		struct vnode *a_vp;
532 	} */ *ap;
533 {
534 
535 	return (lockstatus(ap->a_vp->v_vnlock));
536 }
537 
538 /*
539  * Return true for select/poll.
540  */
541 int
542 vop_nopoll(ap)
543 	struct vop_poll_args /* {
544 		struct vnode *a_vp;
545 		int  a_events;
546 		struct ucred *a_cred;
547 		struct thread *a_td;
548 	} */ *ap;
549 {
550 
551 	return (poll_no_poll(ap->a_events));
552 }
553 
554 /*
555  * Implement poll for local filesystems that support it.
556  */
557 int
558 vop_stdpoll(ap)
559 	struct vop_poll_args /* {
560 		struct vnode *a_vp;
561 		int  a_events;
562 		struct ucred *a_cred;
563 		struct thread *a_td;
564 	} */ *ap;
565 {
566 	if (ap->a_events & ~POLLSTANDARD)
567 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
568 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
569 }
570 
571 /*
572  * Return our mount point, as we will take charge of the writes.
573  */
574 int
575 vop_stdgetwritemount(ap)
576 	struct vop_getwritemount_args /* {
577 		struct vnode *a_vp;
578 		struct mount **a_mpp;
579 	} */ *ap;
580 {
581 	struct mount *mp;
582 
583 	/*
584 	 * XXX Since this is called unlocked we may be recycled while
585 	 * attempting to ref the mount.  If this is the case or mountpoint
586 	 * will be set to NULL.  We only have to prevent this call from
587 	 * returning with a ref to an incorrect mountpoint.  It is not
588 	 * harmful to return with a ref to our previous mountpoint.
589 	 */
590 	mp = ap->a_vp->v_mount;
591 	if (mp != NULL) {
592 		vfs_ref(mp);
593 		if (mp != ap->a_vp->v_mount) {
594 			vfs_rel(mp);
595 			mp = NULL;
596 		}
597 	}
598 	*(ap->a_mpp) = mp;
599 	return (0);
600 }
601 
602 /* XXX Needs good comment and VOP_BMAP(9) manpage */
603 int
604 vop_stdbmap(ap)
605 	struct vop_bmap_args /* {
606 		struct vnode *a_vp;
607 		daddr_t  a_bn;
608 		struct bufobj **a_bop;
609 		daddr_t *a_bnp;
610 		int *a_runp;
611 		int *a_runb;
612 	} */ *ap;
613 {
614 
615 	if (ap->a_bop != NULL)
616 		*ap->a_bop = &ap->a_vp->v_bufobj;
617 	if (ap->a_bnp != NULL)
618 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
619 	if (ap->a_runp != NULL)
620 		*ap->a_runp = 0;
621 	if (ap->a_runb != NULL)
622 		*ap->a_runb = 0;
623 	return (0);
624 }
625 
626 int
627 vop_stdfsync(ap)
628 	struct vop_fsync_args /* {
629 		struct vnode *a_vp;
630 		int a_waitfor;
631 		struct thread *a_td;
632 	} */ *ap;
633 {
634 	struct vnode *vp = ap->a_vp;
635 	struct buf *bp;
636 	struct bufobj *bo;
637 	struct buf *nbp;
638 	int error = 0;
639 	int maxretry = 1000;     /* large, arbitrarily chosen */
640 
641 	bo = &vp->v_bufobj;
642 	BO_LOCK(bo);
643 loop1:
644 	/*
645 	 * MARK/SCAN initialization to avoid infinite loops.
646 	 */
647         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
648                 bp->b_vflags &= ~BV_SCANNED;
649 		bp->b_error = 0;
650 	}
651 
652 	/*
653 	 * Flush all dirty buffers associated with a vnode.
654 	 */
655 loop2:
656 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
657 		if ((bp->b_vflags & BV_SCANNED) != 0)
658 			continue;
659 		bp->b_vflags |= BV_SCANNED;
660 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
661 			if (ap->a_waitfor != MNT_WAIT)
662 				continue;
663 			if (BUF_LOCK(bp,
664 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
665 			    BO_LOCKPTR(bo)) != 0) {
666 				BO_LOCK(bo);
667 				goto loop1;
668 			}
669 			BO_LOCK(bo);
670 		}
671 		BO_UNLOCK(bo);
672 		KASSERT(bp->b_bufobj == bo,
673 		    ("bp %p wrong b_bufobj %p should be %p",
674 		    bp, bp->b_bufobj, bo));
675 		if ((bp->b_flags & B_DELWRI) == 0)
676 			panic("fsync: not dirty");
677 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
678 			vfs_bio_awrite(bp);
679 		} else {
680 			bremfree(bp);
681 			bawrite(bp);
682 		}
683 		BO_LOCK(bo);
684 		goto loop2;
685 	}
686 
687 	/*
688 	 * If synchronous the caller expects us to completely resolve all
689 	 * dirty buffers in the system.  Wait for in-progress I/O to
690 	 * complete (which could include background bitmap writes), then
691 	 * retry if dirty blocks still exist.
692 	 */
693 	if (ap->a_waitfor == MNT_WAIT) {
694 		bufobj_wwait(bo, 0, 0);
695 		if (bo->bo_dirty.bv_cnt > 0) {
696 			/*
697 			 * If we are unable to write any of these buffers
698 			 * then we fail now rather than trying endlessly
699 			 * to write them out.
700 			 */
701 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
702 				if ((error = bp->b_error) != 0)
703 					break;
704 			if (error == 0 && --maxretry >= 0)
705 				goto loop1;
706 			error = EAGAIN;
707 		}
708 	}
709 	BO_UNLOCK(bo);
710 	if (error == EAGAIN)
711 		vn_printf(vp, "fsync: giving up on dirty ");
712 
713 	return (error);
714 }
715 
716 static int
717 vop_stdfdatasync(struct vop_fdatasync_args *ap)
718 {
719 
720 	return (VOP_FSYNC(ap->a_vp, MNT_WAIT, ap->a_td));
721 }
722 
723 int
724 vop_stdfdatasync_buf(struct vop_fdatasync_args *ap)
725 {
726 	struct vop_fsync_args apf;
727 
728 	apf.a_vp = ap->a_vp;
729 	apf.a_waitfor = MNT_WAIT;
730 	apf.a_td = ap->a_td;
731 	return (vop_stdfsync(&apf));
732 }
733 
734 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
735 int
736 vop_stdgetpages(ap)
737 	struct vop_getpages_args /* {
738 		struct vnode *a_vp;
739 		vm_page_t *a_m;
740 		int a_count;
741 		int *a_rbehind;
742 		int *a_rahead;
743 	} */ *ap;
744 {
745 
746 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
747 	    ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL);
748 }
749 
750 static int
751 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
752 {
753 	int error;
754 
755 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
756 	    ap->a_rahead);
757 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
758 	return (error);
759 }
760 
761 int
762 vop_stdkqfilter(struct vop_kqfilter_args *ap)
763 {
764 	return vfs_kqfilter(ap);
765 }
766 
767 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
768 int
769 vop_stdputpages(ap)
770 	struct vop_putpages_args /* {
771 		struct vnode *a_vp;
772 		vm_page_t *a_m;
773 		int a_count;
774 		int a_sync;
775 		int *a_rtvals;
776 	} */ *ap;
777 {
778 
779 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
780 	     ap->a_sync, ap->a_rtvals);
781 }
782 
783 int
784 vop_stdvptofh(struct vop_vptofh_args *ap)
785 {
786 	return (EOPNOTSUPP);
787 }
788 
789 int
790 vop_stdvptocnp(struct vop_vptocnp_args *ap)
791 {
792 	struct vnode *vp = ap->a_vp;
793 	struct vnode **dvp = ap->a_vpp;
794 	struct ucred *cred = ap->a_cred;
795 	char *buf = ap->a_buf;
796 	int *buflen = ap->a_buflen;
797 	char *dirbuf, *cpos;
798 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
799 	off_t off;
800 	ino_t fileno;
801 	struct vattr va;
802 	struct nameidata nd;
803 	struct thread *td;
804 	struct dirent *dp;
805 	struct vnode *mvp;
806 
807 	i = *buflen;
808 	error = 0;
809 	covered = 0;
810 	td = curthread;
811 
812 	if (vp->v_type != VDIR)
813 		return (ENOENT);
814 
815 	error = VOP_GETATTR(vp, &va, cred);
816 	if (error)
817 		return (error);
818 
819 	VREF(vp);
820 	locked = VOP_ISLOCKED(vp);
821 	VOP_UNLOCK(vp, 0);
822 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
823 	    "..", vp, td);
824 	flags = FREAD;
825 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
826 	if (error) {
827 		vn_lock(vp, locked | LK_RETRY);
828 		return (error);
829 	}
830 	NDFREE(&nd, NDF_ONLY_PNBUF);
831 
832 	mvp = *dvp = nd.ni_vp;
833 
834 	if (vp->v_mount != (*dvp)->v_mount &&
835 	    ((*dvp)->v_vflag & VV_ROOT) &&
836 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
837 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
838 		VREF(mvp);
839 		VOP_UNLOCK(mvp, 0);
840 		vn_close(mvp, FREAD, cred, td);
841 		VREF(*dvp);
842 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
843 		covered = 1;
844 	}
845 
846 	fileno = va.va_fileid;
847 
848 	dirbuflen = DEV_BSIZE;
849 	if (dirbuflen < va.va_blocksize)
850 		dirbuflen = va.va_blocksize;
851 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
852 
853 	if ((*dvp)->v_type != VDIR) {
854 		error = ENOENT;
855 		goto out;
856 	}
857 
858 	off = 0;
859 	len = 0;
860 	do {
861 		/* call VOP_READDIR of parent */
862 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
863 					&cpos, &len, &eofflag, td);
864 		if (error)
865 			goto out;
866 
867 		if ((dp->d_type != DT_WHT) &&
868 		    (dp->d_fileno == fileno)) {
869 			if (covered) {
870 				VOP_UNLOCK(*dvp, 0);
871 				vn_lock(mvp, LK_SHARED | LK_RETRY);
872 				if (dirent_exists(mvp, dp->d_name, td)) {
873 					error = ENOENT;
874 					VOP_UNLOCK(mvp, 0);
875 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
876 					goto out;
877 				}
878 				VOP_UNLOCK(mvp, 0);
879 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
880 			}
881 			i -= dp->d_namlen;
882 
883 			if (i < 0) {
884 				error = ENOMEM;
885 				goto out;
886 			}
887 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
888 				error = ENOENT;
889 			} else {
890 				bcopy(dp->d_name, buf + i, dp->d_namlen);
891 				error = 0;
892 			}
893 			goto out;
894 		}
895 	} while (len > 0 || !eofflag);
896 	error = ENOENT;
897 
898 out:
899 	free(dirbuf, M_TEMP);
900 	if (!error) {
901 		*buflen = i;
902 		vref(*dvp);
903 	}
904 	if (covered) {
905 		vput(*dvp);
906 		vrele(mvp);
907 	} else {
908 		VOP_UNLOCK(mvp, 0);
909 		vn_close(mvp, FREAD, cred, td);
910 	}
911 	vn_lock(vp, locked | LK_RETRY);
912 	return (error);
913 }
914 
915 int
916 vop_stdallocate(struct vop_allocate_args *ap)
917 {
918 #ifdef __notyet__
919 	struct statfs *sfs;
920 	off_t maxfilesize = 0;
921 #endif
922 	struct iovec aiov;
923 	struct vattr vattr, *vap;
924 	struct uio auio;
925 	off_t fsize, len, cur, offset;
926 	uint8_t *buf;
927 	struct thread *td;
928 	struct vnode *vp;
929 	size_t iosize;
930 	int error;
931 
932 	buf = NULL;
933 	error = 0;
934 	td = curthread;
935 	vap = &vattr;
936 	vp = ap->a_vp;
937 	len = *ap->a_len;
938 	offset = *ap->a_offset;
939 
940 	error = VOP_GETATTR(vp, vap, td->td_ucred);
941 	if (error != 0)
942 		goto out;
943 	fsize = vap->va_size;
944 	iosize = vap->va_blocksize;
945 	if (iosize == 0)
946 		iosize = BLKDEV_IOSIZE;
947 	if (iosize > MAXPHYS)
948 		iosize = MAXPHYS;
949 	buf = malloc(iosize, M_TEMP, M_WAITOK);
950 
951 #ifdef __notyet__
952 	/*
953 	 * Check if the filesystem sets f_maxfilesize; if not use
954 	 * VOP_SETATTR to perform the check.
955 	 */
956 	sfs = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
957 	error = VFS_STATFS(vp->v_mount, sfs, td);
958 	if (error == 0)
959 		maxfilesize = sfs->f_maxfilesize;
960 	free(sfs, M_STATFS);
961 	if (error != 0)
962 		goto out;
963 	if (maxfilesize) {
964 		if (offset > maxfilesize || len > maxfilesize ||
965 		    offset + len > maxfilesize) {
966 			error = EFBIG;
967 			goto out;
968 		}
969 	} else
970 #endif
971 	if (offset + len > vap->va_size) {
972 		/*
973 		 * Test offset + len against the filesystem's maxfilesize.
974 		 */
975 		VATTR_NULL(vap);
976 		vap->va_size = offset + len;
977 		error = VOP_SETATTR(vp, vap, td->td_ucred);
978 		if (error != 0)
979 			goto out;
980 		VATTR_NULL(vap);
981 		vap->va_size = fsize;
982 		error = VOP_SETATTR(vp, vap, td->td_ucred);
983 		if (error != 0)
984 			goto out;
985 	}
986 
987 	for (;;) {
988 		/*
989 		 * Read and write back anything below the nominal file
990 		 * size.  There's currently no way outside the filesystem
991 		 * to know whether this area is sparse or not.
992 		 */
993 		cur = iosize;
994 		if ((offset % iosize) != 0)
995 			cur -= (offset % iosize);
996 		if (cur > len)
997 			cur = len;
998 		if (offset < fsize) {
999 			aiov.iov_base = buf;
1000 			aiov.iov_len = cur;
1001 			auio.uio_iov = &aiov;
1002 			auio.uio_iovcnt = 1;
1003 			auio.uio_offset = offset;
1004 			auio.uio_resid = cur;
1005 			auio.uio_segflg = UIO_SYSSPACE;
1006 			auio.uio_rw = UIO_READ;
1007 			auio.uio_td = td;
1008 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
1009 			if (error != 0)
1010 				break;
1011 			if (auio.uio_resid > 0) {
1012 				bzero(buf + cur - auio.uio_resid,
1013 				    auio.uio_resid);
1014 			}
1015 		} else {
1016 			bzero(buf, cur);
1017 		}
1018 
1019 		aiov.iov_base = buf;
1020 		aiov.iov_len = cur;
1021 		auio.uio_iov = &aiov;
1022 		auio.uio_iovcnt = 1;
1023 		auio.uio_offset = offset;
1024 		auio.uio_resid = cur;
1025 		auio.uio_segflg = UIO_SYSSPACE;
1026 		auio.uio_rw = UIO_WRITE;
1027 		auio.uio_td = td;
1028 
1029 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
1030 		if (error != 0)
1031 			break;
1032 
1033 		len -= cur;
1034 		offset += cur;
1035 		if (len == 0)
1036 			break;
1037 		if (should_yield())
1038 			break;
1039 	}
1040 
1041  out:
1042 	*ap->a_len = len;
1043 	*ap->a_offset = offset;
1044 	free(buf, M_TEMP);
1045 	return (error);
1046 }
1047 
1048 int
1049 vop_stdadvise(struct vop_advise_args *ap)
1050 {
1051 	struct vnode *vp;
1052 	struct bufobj *bo;
1053 	daddr_t startn, endn;
1054 	off_t start, end;
1055 	int bsize, error;
1056 
1057 	vp = ap->a_vp;
1058 	switch (ap->a_advice) {
1059 	case POSIX_FADV_WILLNEED:
1060 		/*
1061 		 * Do nothing for now.  Filesystems should provide a
1062 		 * custom method which starts an asynchronous read of
1063 		 * the requested region.
1064 		 */
1065 		error = 0;
1066 		break;
1067 	case POSIX_FADV_DONTNEED:
1068 		error = 0;
1069 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1070 		if (vp->v_iflag & VI_DOOMED) {
1071 			VOP_UNLOCK(vp, 0);
1072 			break;
1073 		}
1074 
1075 		/*
1076 		 * Deactivate pages in the specified range from the backing VM
1077 		 * object.  Pages that are resident in the buffer cache will
1078 		 * remain wired until their corresponding buffers are released
1079 		 * below.
1080 		 */
1081 		if (vp->v_object != NULL) {
1082 			start = trunc_page(ap->a_start);
1083 			end = round_page(ap->a_end);
1084 			VM_OBJECT_RLOCK(vp->v_object);
1085 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1086 			    OFF_TO_IDX(end));
1087 			VM_OBJECT_RUNLOCK(vp->v_object);
1088 		}
1089 
1090 		bo = &vp->v_bufobj;
1091 		BO_RLOCK(bo);
1092 		bsize = vp->v_bufobj.bo_bsize;
1093 		startn = ap->a_start / bsize;
1094 		endn = ap->a_end / bsize;
1095 		error = bnoreuselist(&bo->bo_clean, bo, startn, endn);
1096 		if (error == 0)
1097 			error = bnoreuselist(&bo->bo_dirty, bo, startn, endn);
1098 		BO_RUNLOCK(bo);
1099 		VOP_UNLOCK(vp, 0);
1100 		break;
1101 	default:
1102 		error = EINVAL;
1103 		break;
1104 	}
1105 	return (error);
1106 }
1107 
1108 int
1109 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1110 {
1111 
1112 	ap->a_vp->v_unpcb = ap->a_unpcb;
1113 	return (0);
1114 }
1115 
1116 int
1117 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1118 {
1119 
1120 	*ap->a_unpcb = ap->a_vp->v_unpcb;
1121 	return (0);
1122 }
1123 
1124 int
1125 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1126 {
1127 
1128 	ap->a_vp->v_unpcb = NULL;
1129 	return (0);
1130 }
1131 
1132 static int
1133 vop_stdis_text(struct vop_is_text_args *ap)
1134 {
1135 
1136 	return ((ap->a_vp->v_vflag & VV_TEXT) != 0);
1137 }
1138 
1139 static int
1140 vop_stdset_text(struct vop_set_text_args *ap)
1141 {
1142 
1143 	ap->a_vp->v_vflag |= VV_TEXT;
1144 	return (0);
1145 }
1146 
1147 static int
1148 vop_stdunset_text(struct vop_unset_text_args *ap)
1149 {
1150 
1151 	ap->a_vp->v_vflag &= ~VV_TEXT;
1152 	return (0);
1153 }
1154 
1155 static int
1156 vop_stdget_writecount(struct vop_get_writecount_args *ap)
1157 {
1158 
1159 	*ap->a_writecount = ap->a_vp->v_writecount;
1160 	return (0);
1161 }
1162 
1163 static int
1164 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1165 {
1166 
1167 	ap->a_vp->v_writecount += ap->a_inc;
1168 	return (0);
1169 }
1170 
1171 /*
1172  * vfs default ops
1173  * used to fill the vfs function table to get reasonable default return values.
1174  */
1175 int
1176 vfs_stdroot (mp, flags, vpp)
1177 	struct mount *mp;
1178 	int flags;
1179 	struct vnode **vpp;
1180 {
1181 
1182 	return (EOPNOTSUPP);
1183 }
1184 
1185 int
1186 vfs_stdstatfs (mp, sbp)
1187 	struct mount *mp;
1188 	struct statfs *sbp;
1189 {
1190 
1191 	return (EOPNOTSUPP);
1192 }
1193 
1194 int
1195 vfs_stdquotactl (mp, cmds, uid, arg)
1196 	struct mount *mp;
1197 	int cmds;
1198 	uid_t uid;
1199 	void *arg;
1200 {
1201 
1202 	return (EOPNOTSUPP);
1203 }
1204 
1205 int
1206 vfs_stdsync(mp, waitfor)
1207 	struct mount *mp;
1208 	int waitfor;
1209 {
1210 	struct vnode *vp, *mvp;
1211 	struct thread *td;
1212 	int error, lockreq, allerror = 0;
1213 
1214 	td = curthread;
1215 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1216 	if (waitfor != MNT_WAIT)
1217 		lockreq |= LK_NOWAIT;
1218 	/*
1219 	 * Force stale buffer cache information to be flushed.
1220 	 */
1221 loop:
1222 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1223 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1224 			VI_UNLOCK(vp);
1225 			continue;
1226 		}
1227 		if ((error = vget(vp, lockreq, td)) != 0) {
1228 			if (error == ENOENT) {
1229 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1230 				goto loop;
1231 			}
1232 			continue;
1233 		}
1234 		error = VOP_FSYNC(vp, waitfor, td);
1235 		if (error)
1236 			allerror = error;
1237 		vput(vp);
1238 	}
1239 	return (allerror);
1240 }
1241 
1242 int
1243 vfs_stdnosync (mp, waitfor)
1244 	struct mount *mp;
1245 	int waitfor;
1246 {
1247 
1248 	return (0);
1249 }
1250 
1251 int
1252 vfs_stdvget (mp, ino, flags, vpp)
1253 	struct mount *mp;
1254 	ino_t ino;
1255 	int flags;
1256 	struct vnode **vpp;
1257 {
1258 
1259 	return (EOPNOTSUPP);
1260 }
1261 
1262 int
1263 vfs_stdfhtovp (mp, fhp, flags, vpp)
1264 	struct mount *mp;
1265 	struct fid *fhp;
1266 	int flags;
1267 	struct vnode **vpp;
1268 {
1269 
1270 	return (EOPNOTSUPP);
1271 }
1272 
1273 int
1274 vfs_stdinit (vfsp)
1275 	struct vfsconf *vfsp;
1276 {
1277 
1278 	return (0);
1279 }
1280 
1281 int
1282 vfs_stduninit (vfsp)
1283 	struct vfsconf *vfsp;
1284 {
1285 
1286 	return(0);
1287 }
1288 
1289 int
1290 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1291 	struct mount *mp;
1292 	int cmd;
1293 	struct vnode *filename_vp;
1294 	int attrnamespace;
1295 	const char *attrname;
1296 {
1297 
1298 	if (filename_vp != NULL)
1299 		VOP_UNLOCK(filename_vp, 0);
1300 	return (EOPNOTSUPP);
1301 }
1302 
1303 int
1304 vfs_stdsysctl(mp, op, req)
1305 	struct mount *mp;
1306 	fsctlop_t op;
1307 	struct sysctl_req *req;
1308 {
1309 
1310 	return (EOPNOTSUPP);
1311 }
1312 
1313 /* end of vfs default ops */
1314