xref: /freebsd/sys/kern/vfs_default.c (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
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_stdunset_text(struct vop_unset_text_args *ap);
85 static int vop_stdadd_writecount(struct vop_add_writecount_args *ap);
86 static int vop_stdcopy_file_range(struct vop_copy_file_range_args *ap);
87 static int vop_stdfdatasync(struct vop_fdatasync_args *ap);
88 static int vop_stdgetpages_async(struct vop_getpages_async_args *ap);
89 
90 /*
91  * This vnode table stores what we want to do if the filesystem doesn't
92  * implement a particular VOP.
93  *
94  * If there is no specific entry here, we will return EOPNOTSUPP.
95  *
96  * Note that every filesystem has to implement either vop_access
97  * or vop_accessx; failing to do so will result in immediate crash
98  * due to stack overflow, as vop_stdaccess() calls vop_stdaccessx(),
99  * which calls vop_stdaccess() etc.
100  */
101 
102 struct vop_vector default_vnodeops = {
103 	.vop_default =		NULL,
104 	.vop_bypass =		VOP_EOPNOTSUPP,
105 
106 	.vop_access =		vop_stdaccess,
107 	.vop_accessx =		vop_stdaccessx,
108 	.vop_advise =		vop_stdadvise,
109 	.vop_advlock =		vop_stdadvlock,
110 	.vop_advlockasync =	vop_stdadvlockasync,
111 	.vop_advlockpurge =	vop_stdadvlockpurge,
112 	.vop_allocate =		vop_stdallocate,
113 	.vop_bmap =		vop_stdbmap,
114 	.vop_close =		VOP_NULL,
115 	.vop_fsync =		VOP_NULL,
116 	.vop_fdatasync =	vop_stdfdatasync,
117 	.vop_getpages =		vop_stdgetpages,
118 	.vop_getpages_async =	vop_stdgetpages_async,
119 	.vop_getwritemount = 	vop_stdgetwritemount,
120 	.vop_inactive =		VOP_NULL,
121 	.vop_ioctl =		VOP_ENOTTY,
122 	.vop_kqfilter =		vop_stdkqfilter,
123 	.vop_islocked =		vop_stdislocked,
124 	.vop_lock1 =		vop_stdlock,
125 	.vop_lookup =		vop_nolookup,
126 	.vop_open =		VOP_NULL,
127 	.vop_pathconf =		VOP_EINVAL,
128 	.vop_poll =		vop_nopoll,
129 	.vop_putpages =		vop_stdputpages,
130 	.vop_readlink =		VOP_EINVAL,
131 	.vop_rename =		vop_norename,
132 	.vop_revoke =		VOP_PANIC,
133 	.vop_strategy =		vop_nostrategy,
134 	.vop_unlock =		vop_stdunlock,
135 	.vop_vptocnp =		vop_stdvptocnp,
136 	.vop_vptofh =		vop_stdvptofh,
137 	.vop_unp_bind =		vop_stdunp_bind,
138 	.vop_unp_connect =	vop_stdunp_connect,
139 	.vop_unp_detach =	vop_stdunp_detach,
140 	.vop_is_text =		vop_stdis_text,
141 	.vop_set_text =		vop_stdset_text,
142 	.vop_unset_text =	vop_stdunset_text,
143 	.vop_add_writecount =	vop_stdadd_writecount,
144 	.vop_copy_file_range =	vop_stdcopy_file_range,
145 };
146 
147 /*
148  * Series of placeholder functions for various error returns for
149  * VOPs.
150  */
151 
152 int
153 vop_eopnotsupp(struct vop_generic_args *ap)
154 {
155 	/*
156 	printf("vop_notsupp[%s]\n", ap->a_desc->vdesc_name);
157 	*/
158 
159 	return (EOPNOTSUPP);
160 }
161 
162 int
163 vop_ebadf(struct vop_generic_args *ap)
164 {
165 
166 	return (EBADF);
167 }
168 
169 int
170 vop_enotty(struct vop_generic_args *ap)
171 {
172 
173 	return (ENOTTY);
174 }
175 
176 int
177 vop_einval(struct vop_generic_args *ap)
178 {
179 
180 	return (EINVAL);
181 }
182 
183 int
184 vop_enoent(struct vop_generic_args *ap)
185 {
186 
187 	return (ENOENT);
188 }
189 
190 int
191 vop_null(struct vop_generic_args *ap)
192 {
193 
194 	return (0);
195 }
196 
197 /*
198  * Helper function to panic on some bad VOPs in some filesystems.
199  */
200 int
201 vop_panic(struct vop_generic_args *ap)
202 {
203 
204 	panic("filesystem goof: vop_panic[%s]", ap->a_desc->vdesc_name);
205 }
206 
207 /*
208  * vop_std<something> and vop_no<something> are default functions for use by
209  * filesystems that need the "default reasonable" implementation for a
210  * particular operation.
211  *
212  * The documentation for the operations they implement exists (if it exists)
213  * in the VOP_<SOMETHING>(9) manpage (all uppercase).
214  */
215 
216 /*
217  * Default vop for filesystems that do not support name lookup
218  */
219 static int
220 vop_nolookup(ap)
221 	struct vop_lookup_args /* {
222 		struct vnode *a_dvp;
223 		struct vnode **a_vpp;
224 		struct componentname *a_cnp;
225 	} */ *ap;
226 {
227 
228 	*ap->a_vpp = NULL;
229 	return (ENOTDIR);
230 }
231 
232 /*
233  * vop_norename:
234  *
235  * Handle unlock and reference counting for arguments of vop_rename
236  * for filesystems that do not implement rename operation.
237  */
238 static int
239 vop_norename(struct vop_rename_args *ap)
240 {
241 
242 	vop_rename_fail(ap);
243 	return (EOPNOTSUPP);
244 }
245 
246 /*
247  *	vop_nostrategy:
248  *
249  *	Strategy routine for VFS devices that have none.
250  *
251  *	BIO_ERROR and B_INVAL must be cleared prior to calling any strategy
252  *	routine.  Typically this is done for a BIO_READ strategy call.
253  *	Typically B_INVAL is assumed to already be clear prior to a write
254  *	and should not be cleared manually unless you just made the buffer
255  *	invalid.  BIO_ERROR should be cleared either way.
256  */
257 
258 static int
259 vop_nostrategy (struct vop_strategy_args *ap)
260 {
261 	printf("No strategy for buffer at %p\n", ap->a_bp);
262 	vn_printf(ap->a_vp, "vnode ");
263 	ap->a_bp->b_ioflags |= BIO_ERROR;
264 	ap->a_bp->b_error = EOPNOTSUPP;
265 	bufdone(ap->a_bp);
266 	return (EOPNOTSUPP);
267 }
268 
269 static int
270 get_next_dirent(struct vnode *vp, struct dirent **dpp, char *dirbuf,
271 		int dirbuflen, off_t *off, char **cpos, int *len,
272 		int *eofflag, struct thread *td)
273 {
274 	int error, reclen;
275 	struct uio uio;
276 	struct iovec iov;
277 	struct dirent *dp;
278 
279 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
280 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
281 
282 	if (*len == 0) {
283 		iov.iov_base = dirbuf;
284 		iov.iov_len = dirbuflen;
285 
286 		uio.uio_iov = &iov;
287 		uio.uio_iovcnt = 1;
288 		uio.uio_offset = *off;
289 		uio.uio_resid = dirbuflen;
290 		uio.uio_segflg = UIO_SYSSPACE;
291 		uio.uio_rw = UIO_READ;
292 		uio.uio_td = td;
293 
294 		*eofflag = 0;
295 
296 #ifdef MAC
297 		error = mac_vnode_check_readdir(td->td_ucred, vp);
298 		if (error == 0)
299 #endif
300 			error = VOP_READDIR(vp, &uio, td->td_ucred, eofflag,
301 		    		NULL, NULL);
302 		if (error)
303 			return (error);
304 
305 		*off = uio.uio_offset;
306 
307 		*cpos = dirbuf;
308 		*len = (dirbuflen - uio.uio_resid);
309 
310 		if (*len == 0)
311 			return (ENOENT);
312 	}
313 
314 	dp = (struct dirent *)(*cpos);
315 	reclen = dp->d_reclen;
316 	*dpp = dp;
317 
318 	/* check for malformed directory.. */
319 	if (reclen < DIRENT_MINSIZE)
320 		return (EINVAL);
321 
322 	*cpos += reclen;
323 	*len -= reclen;
324 
325 	return (0);
326 }
327 
328 /*
329  * Check if a named file exists in a given directory vnode.
330  */
331 static int
332 dirent_exists(struct vnode *vp, const char *dirname, struct thread *td)
333 {
334 	char *dirbuf, *cpos;
335 	int error, eofflag, dirbuflen, len, found;
336 	off_t off;
337 	struct dirent *dp;
338 	struct vattr va;
339 
340 	KASSERT(VOP_ISLOCKED(vp), ("vp %p is not locked", vp));
341 	KASSERT(vp->v_type == VDIR, ("vp %p is not a directory", vp));
342 
343 	found = 0;
344 
345 	error = VOP_GETATTR(vp, &va, td->td_ucred);
346 	if (error)
347 		return (found);
348 
349 	dirbuflen = DEV_BSIZE;
350 	if (dirbuflen < va.va_blocksize)
351 		dirbuflen = va.va_blocksize;
352 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
353 
354 	off = 0;
355 	len = 0;
356 	do {
357 		error = get_next_dirent(vp, &dp, dirbuf, dirbuflen, &off,
358 					&cpos, &len, &eofflag, td);
359 		if (error)
360 			goto out;
361 
362 		if (dp->d_type != DT_WHT && dp->d_fileno != 0 &&
363 		    strcmp(dp->d_name, dirname) == 0) {
364 			found = 1;
365 			goto out;
366 		}
367 	} while (len > 0 || !eofflag);
368 
369 out:
370 	free(dirbuf, M_TEMP);
371 	return (found);
372 }
373 
374 int
375 vop_stdaccess(struct vop_access_args *ap)
376 {
377 
378 	KASSERT((ap->a_accmode & ~(VEXEC | VWRITE | VREAD | VADMIN |
379 	    VAPPEND)) == 0, ("invalid bit in accmode"));
380 
381 	return (VOP_ACCESSX(ap->a_vp, ap->a_accmode, ap->a_cred, ap->a_td));
382 }
383 
384 int
385 vop_stdaccessx(struct vop_accessx_args *ap)
386 {
387 	int error;
388 	accmode_t accmode = ap->a_accmode;
389 
390 	error = vfs_unixify_accmode(&accmode);
391 	if (error != 0)
392 		return (error);
393 
394 	if (accmode == 0)
395 		return (0);
396 
397 	return (VOP_ACCESS(ap->a_vp, accmode, ap->a_cred, ap->a_td));
398 }
399 
400 /*
401  * Advisory record locking support
402  */
403 int
404 vop_stdadvlock(struct vop_advlock_args *ap)
405 {
406 	struct vnode *vp;
407 	struct vattr vattr;
408 	int error;
409 
410 	vp = ap->a_vp;
411 	if (ap->a_fl->l_whence == SEEK_END) {
412 		/*
413 		 * The NFSv4 server must avoid doing a vn_lock() here, since it
414 		 * can deadlock the nfsd threads, due to a LOR.  Fortunately
415 		 * the NFSv4 server always uses SEEK_SET and this code is
416 		 * only required for the SEEK_END case.
417 		 */
418 		vn_lock(vp, LK_SHARED | LK_RETRY);
419 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
420 		VOP_UNLOCK(vp, 0);
421 		if (error)
422 			return (error);
423 	} else
424 		vattr.va_size = 0;
425 
426 	return (lf_advlock(ap, &(vp->v_lockf), vattr.va_size));
427 }
428 
429 int
430 vop_stdadvlockasync(struct vop_advlockasync_args *ap)
431 {
432 	struct vnode *vp;
433 	struct vattr vattr;
434 	int error;
435 
436 	vp = ap->a_vp;
437 	if (ap->a_fl->l_whence == SEEK_END) {
438 		/* The size argument is only needed for SEEK_END. */
439 		vn_lock(vp, LK_SHARED | LK_RETRY);
440 		error = VOP_GETATTR(vp, &vattr, curthread->td_ucred);
441 		VOP_UNLOCK(vp, 0);
442 		if (error)
443 			return (error);
444 	} else
445 		vattr.va_size = 0;
446 
447 	return (lf_advlockasync(ap, &(vp->v_lockf), vattr.va_size));
448 }
449 
450 int
451 vop_stdadvlockpurge(struct vop_advlockpurge_args *ap)
452 {
453 	struct vnode *vp;
454 
455 	vp = ap->a_vp;
456 	lf_purgelocks(vp, &vp->v_lockf);
457 	return (0);
458 }
459 
460 /*
461  * vop_stdpathconf:
462  *
463  * Standard implementation of POSIX pathconf, to get information about limits
464  * for a filesystem.
465  * Override per filesystem for the case where the filesystem has smaller
466  * limits.
467  */
468 int
469 vop_stdpathconf(ap)
470 	struct vop_pathconf_args /* {
471 	struct vnode *a_vp;
472 	int a_name;
473 	int *a_retval;
474 	} */ *ap;
475 {
476 
477 	switch (ap->a_name) {
478 		case _PC_ASYNC_IO:
479 			*ap->a_retval = _POSIX_ASYNCHRONOUS_IO;
480 			return (0);
481 		case _PC_PATH_MAX:
482 			*ap->a_retval = PATH_MAX;
483 			return (0);
484 		case _PC_ACL_EXTENDED:
485 		case _PC_ACL_NFS4:
486 		case _PC_CAP_PRESENT:
487 		case _PC_INF_PRESENT:
488 		case _PC_MAC_PRESENT:
489 			*ap->a_retval = 0;
490 			return (0);
491 		default:
492 			return (EINVAL);
493 	}
494 	/* NOTREACHED */
495 }
496 
497 /*
498  * Standard lock, unlock and islocked functions.
499  */
500 int
501 vop_stdlock(ap)
502 	struct vop_lock1_args /* {
503 		struct vnode *a_vp;
504 		int a_flags;
505 		char *file;
506 		int line;
507 	} */ *ap;
508 {
509 	struct vnode *vp = ap->a_vp;
510 	struct mtx *ilk;
511 
512 	ilk = VI_MTX(vp);
513 	return (lockmgr_lock_fast_path(vp->v_vnlock, ap->a_flags,
514 	    &ilk->lock_object, ap->a_file, ap->a_line));
515 }
516 
517 /* See above. */
518 int
519 vop_stdunlock(ap)
520 	struct vop_unlock_args /* {
521 		struct vnode *a_vp;
522 		int a_flags;
523 	} */ *ap;
524 {
525 	struct vnode *vp = ap->a_vp;
526 	struct mtx *ilk;
527 
528 	ilk = VI_MTX(vp);
529 	return (lockmgr_unlock_fast_path(vp->v_vnlock, ap->a_flags,
530 	    &ilk->lock_object));
531 }
532 
533 /* See above. */
534 int
535 vop_stdislocked(ap)
536 	struct vop_islocked_args /* {
537 		struct vnode *a_vp;
538 	} */ *ap;
539 {
540 
541 	return (lockstatus(ap->a_vp->v_vnlock));
542 }
543 
544 /*
545  * Return true for select/poll.
546  */
547 int
548 vop_nopoll(ap)
549 	struct vop_poll_args /* {
550 		struct vnode *a_vp;
551 		int  a_events;
552 		struct ucred *a_cred;
553 		struct thread *a_td;
554 	} */ *ap;
555 {
556 
557 	return (poll_no_poll(ap->a_events));
558 }
559 
560 /*
561  * Implement poll for local filesystems that support it.
562  */
563 int
564 vop_stdpoll(ap)
565 	struct vop_poll_args /* {
566 		struct vnode *a_vp;
567 		int  a_events;
568 		struct ucred *a_cred;
569 		struct thread *a_td;
570 	} */ *ap;
571 {
572 	if (ap->a_events & ~POLLSTANDARD)
573 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
574 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
575 }
576 
577 /*
578  * Return our mount point, as we will take charge of the writes.
579  */
580 int
581 vop_stdgetwritemount(ap)
582 	struct vop_getwritemount_args /* {
583 		struct vnode *a_vp;
584 		struct mount **a_mpp;
585 	} */ *ap;
586 {
587 	struct mount *mp;
588 
589 	/*
590 	 * XXX Since this is called unlocked we may be recycled while
591 	 * attempting to ref the mount.  If this is the case or mountpoint
592 	 * will be set to NULL.  We only have to prevent this call from
593 	 * returning with a ref to an incorrect mountpoint.  It is not
594 	 * harmful to return with a ref to our previous mountpoint.
595 	 */
596 	mp = ap->a_vp->v_mount;
597 	if (mp != NULL) {
598 		vfs_ref(mp);
599 		if (mp != ap->a_vp->v_mount) {
600 			vfs_rel(mp);
601 			mp = NULL;
602 		}
603 	}
604 	*(ap->a_mpp) = mp;
605 	return (0);
606 }
607 
608 /*
609  * If the file system doesn't implement VOP_BMAP, then return sensible defaults:
610  * - Return the vnode's bufobj instead of any underlying device's bufobj
611  * - Calculate the physical block number as if there were equal size
612  *   consecutive blocks, but
613  * - Report no contiguous runs of blocks.
614  */
615 int
616 vop_stdbmap(ap)
617 	struct vop_bmap_args /* {
618 		struct vnode *a_vp;
619 		daddr_t  a_bn;
620 		struct bufobj **a_bop;
621 		daddr_t *a_bnp;
622 		int *a_runp;
623 		int *a_runb;
624 	} */ *ap;
625 {
626 
627 	if (ap->a_bop != NULL)
628 		*ap->a_bop = &ap->a_vp->v_bufobj;
629 	if (ap->a_bnp != NULL)
630 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
631 	if (ap->a_runp != NULL)
632 		*ap->a_runp = 0;
633 	if (ap->a_runb != NULL)
634 		*ap->a_runb = 0;
635 	return (0);
636 }
637 
638 int
639 vop_stdfsync(ap)
640 	struct vop_fsync_args /* {
641 		struct vnode *a_vp;
642 		int a_waitfor;
643 		struct thread *a_td;
644 	} */ *ap;
645 {
646 
647 	return (vn_fsync_buf(ap->a_vp, ap->a_waitfor));
648 }
649 
650 static int
651 vop_stdfdatasync(struct vop_fdatasync_args *ap)
652 {
653 
654 	return (VOP_FSYNC(ap->a_vp, MNT_WAIT, ap->a_td));
655 }
656 
657 int
658 vop_stdfdatasync_buf(struct vop_fdatasync_args *ap)
659 {
660 
661 	return (vn_fsync_buf(ap->a_vp, MNT_WAIT));
662 }
663 
664 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
665 int
666 vop_stdgetpages(ap)
667 	struct vop_getpages_args /* {
668 		struct vnode *a_vp;
669 		vm_page_t *a_m;
670 		int a_count;
671 		int *a_rbehind;
672 		int *a_rahead;
673 	} */ *ap;
674 {
675 
676 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
677 	    ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL);
678 }
679 
680 static int
681 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
682 {
683 	int error;
684 
685 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
686 	    ap->a_rahead);
687 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
688 	return (error);
689 }
690 
691 int
692 vop_stdkqfilter(struct vop_kqfilter_args *ap)
693 {
694 	return vfs_kqfilter(ap);
695 }
696 
697 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
698 int
699 vop_stdputpages(ap)
700 	struct vop_putpages_args /* {
701 		struct vnode *a_vp;
702 		vm_page_t *a_m;
703 		int a_count;
704 		int a_sync;
705 		int *a_rtvals;
706 	} */ *ap;
707 {
708 
709 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
710 	     ap->a_sync, ap->a_rtvals);
711 }
712 
713 int
714 vop_stdvptofh(struct vop_vptofh_args *ap)
715 {
716 	return (EOPNOTSUPP);
717 }
718 
719 int
720 vop_stdvptocnp(struct vop_vptocnp_args *ap)
721 {
722 	struct vnode *vp = ap->a_vp;
723 	struct vnode **dvp = ap->a_vpp;
724 	struct ucred *cred = ap->a_cred;
725 	char *buf = ap->a_buf;
726 	int *buflen = ap->a_buflen;
727 	char *dirbuf, *cpos;
728 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
729 	off_t off;
730 	ino_t fileno;
731 	struct vattr va;
732 	struct nameidata nd;
733 	struct thread *td;
734 	struct dirent *dp;
735 	struct vnode *mvp;
736 
737 	i = *buflen;
738 	error = 0;
739 	covered = 0;
740 	td = curthread;
741 
742 	if (vp->v_type != VDIR)
743 		return (ENOENT);
744 
745 	error = VOP_GETATTR(vp, &va, cred);
746 	if (error)
747 		return (error);
748 
749 	VREF(vp);
750 	locked = VOP_ISLOCKED(vp);
751 	VOP_UNLOCK(vp, 0);
752 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
753 	    "..", vp, td);
754 	flags = FREAD;
755 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
756 	if (error) {
757 		vn_lock(vp, locked | LK_RETRY);
758 		return (error);
759 	}
760 	NDFREE(&nd, NDF_ONLY_PNBUF);
761 
762 	mvp = *dvp = nd.ni_vp;
763 
764 	if (vp->v_mount != (*dvp)->v_mount &&
765 	    ((*dvp)->v_vflag & VV_ROOT) &&
766 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
767 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
768 		VREF(mvp);
769 		VOP_UNLOCK(mvp, 0);
770 		vn_close(mvp, FREAD, cred, td);
771 		VREF(*dvp);
772 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
773 		covered = 1;
774 	}
775 
776 	fileno = va.va_fileid;
777 
778 	dirbuflen = DEV_BSIZE;
779 	if (dirbuflen < va.va_blocksize)
780 		dirbuflen = va.va_blocksize;
781 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
782 
783 	if ((*dvp)->v_type != VDIR) {
784 		error = ENOENT;
785 		goto out;
786 	}
787 
788 	off = 0;
789 	len = 0;
790 	do {
791 		/* call VOP_READDIR of parent */
792 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
793 					&cpos, &len, &eofflag, td);
794 		if (error)
795 			goto out;
796 
797 		if ((dp->d_type != DT_WHT) &&
798 		    (dp->d_fileno == fileno)) {
799 			if (covered) {
800 				VOP_UNLOCK(*dvp, 0);
801 				vn_lock(mvp, LK_SHARED | LK_RETRY);
802 				if (dirent_exists(mvp, dp->d_name, td)) {
803 					error = ENOENT;
804 					VOP_UNLOCK(mvp, 0);
805 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
806 					goto out;
807 				}
808 				VOP_UNLOCK(mvp, 0);
809 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
810 			}
811 			i -= dp->d_namlen;
812 
813 			if (i < 0) {
814 				error = ENOMEM;
815 				goto out;
816 			}
817 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
818 				error = ENOENT;
819 			} else {
820 				bcopy(dp->d_name, buf + i, dp->d_namlen);
821 				error = 0;
822 			}
823 			goto out;
824 		}
825 	} while (len > 0 || !eofflag);
826 	error = ENOENT;
827 
828 out:
829 	free(dirbuf, M_TEMP);
830 	if (!error) {
831 		*buflen = i;
832 		vref(*dvp);
833 	}
834 	if (covered) {
835 		vput(*dvp);
836 		vrele(mvp);
837 	} else {
838 		VOP_UNLOCK(mvp, 0);
839 		vn_close(mvp, FREAD, cred, td);
840 	}
841 	vn_lock(vp, locked | LK_RETRY);
842 	return (error);
843 }
844 
845 int
846 vop_stdallocate(struct vop_allocate_args *ap)
847 {
848 #ifdef __notyet__
849 	struct statfs *sfs;
850 	off_t maxfilesize = 0;
851 #endif
852 	struct iovec aiov;
853 	struct vattr vattr, *vap;
854 	struct uio auio;
855 	off_t fsize, len, cur, offset;
856 	uint8_t *buf;
857 	struct thread *td;
858 	struct vnode *vp;
859 	size_t iosize;
860 	int error;
861 
862 	buf = NULL;
863 	error = 0;
864 	td = curthread;
865 	vap = &vattr;
866 	vp = ap->a_vp;
867 	len = *ap->a_len;
868 	offset = *ap->a_offset;
869 
870 	error = VOP_GETATTR(vp, vap, td->td_ucred);
871 	if (error != 0)
872 		goto out;
873 	fsize = vap->va_size;
874 	iosize = vap->va_blocksize;
875 	if (iosize == 0)
876 		iosize = BLKDEV_IOSIZE;
877 	if (iosize > MAXPHYS)
878 		iosize = MAXPHYS;
879 	buf = malloc(iosize, M_TEMP, M_WAITOK);
880 
881 #ifdef __notyet__
882 	/*
883 	 * Check if the filesystem sets f_maxfilesize; if not use
884 	 * VOP_SETATTR to perform the check.
885 	 */
886 	sfs = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
887 	error = VFS_STATFS(vp->v_mount, sfs, td);
888 	if (error == 0)
889 		maxfilesize = sfs->f_maxfilesize;
890 	free(sfs, M_STATFS);
891 	if (error != 0)
892 		goto out;
893 	if (maxfilesize) {
894 		if (offset > maxfilesize || len > maxfilesize ||
895 		    offset + len > maxfilesize) {
896 			error = EFBIG;
897 			goto out;
898 		}
899 	} else
900 #endif
901 	if (offset + len > vap->va_size) {
902 		/*
903 		 * Test offset + len against the filesystem's maxfilesize.
904 		 */
905 		VATTR_NULL(vap);
906 		vap->va_size = offset + len;
907 		error = VOP_SETATTR(vp, vap, td->td_ucred);
908 		if (error != 0)
909 			goto out;
910 		VATTR_NULL(vap);
911 		vap->va_size = fsize;
912 		error = VOP_SETATTR(vp, vap, td->td_ucred);
913 		if (error != 0)
914 			goto out;
915 	}
916 
917 	for (;;) {
918 		/*
919 		 * Read and write back anything below the nominal file
920 		 * size.  There's currently no way outside the filesystem
921 		 * to know whether this area is sparse or not.
922 		 */
923 		cur = iosize;
924 		if ((offset % iosize) != 0)
925 			cur -= (offset % iosize);
926 		if (cur > len)
927 			cur = len;
928 		if (offset < fsize) {
929 			aiov.iov_base = buf;
930 			aiov.iov_len = cur;
931 			auio.uio_iov = &aiov;
932 			auio.uio_iovcnt = 1;
933 			auio.uio_offset = offset;
934 			auio.uio_resid = cur;
935 			auio.uio_segflg = UIO_SYSSPACE;
936 			auio.uio_rw = UIO_READ;
937 			auio.uio_td = td;
938 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
939 			if (error != 0)
940 				break;
941 			if (auio.uio_resid > 0) {
942 				bzero(buf + cur - auio.uio_resid,
943 				    auio.uio_resid);
944 			}
945 		} else {
946 			bzero(buf, cur);
947 		}
948 
949 		aiov.iov_base = buf;
950 		aiov.iov_len = cur;
951 		auio.uio_iov = &aiov;
952 		auio.uio_iovcnt = 1;
953 		auio.uio_offset = offset;
954 		auio.uio_resid = cur;
955 		auio.uio_segflg = UIO_SYSSPACE;
956 		auio.uio_rw = UIO_WRITE;
957 		auio.uio_td = td;
958 
959 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
960 		if (error != 0)
961 			break;
962 
963 		len -= cur;
964 		offset += cur;
965 		if (len == 0)
966 			break;
967 		if (should_yield())
968 			break;
969 	}
970 
971  out:
972 	*ap->a_len = len;
973 	*ap->a_offset = offset;
974 	free(buf, M_TEMP);
975 	return (error);
976 }
977 
978 int
979 vop_stdadvise(struct vop_advise_args *ap)
980 {
981 	struct vnode *vp;
982 	struct bufobj *bo;
983 	daddr_t startn, endn;
984 	off_t bstart, bend, start, end;
985 	int bsize, error;
986 
987 	vp = ap->a_vp;
988 	switch (ap->a_advice) {
989 	case POSIX_FADV_WILLNEED:
990 		/*
991 		 * Do nothing for now.  Filesystems should provide a
992 		 * custom method which starts an asynchronous read of
993 		 * the requested region.
994 		 */
995 		error = 0;
996 		break;
997 	case POSIX_FADV_DONTNEED:
998 		error = 0;
999 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1000 		if (vp->v_iflag & VI_DOOMED) {
1001 			VOP_UNLOCK(vp, 0);
1002 			break;
1003 		}
1004 
1005 		/*
1006 		 * Round to block boundaries (and later possibly further to
1007 		 * page boundaries).  Applications cannot reasonably be aware
1008 		 * of the boundaries, and the rounding must be to expand at
1009 		 * both extremities to cover enough.  It still doesn't cover
1010 		 * read-ahead.  For partial blocks, this gives unnecessary
1011 		 * discarding of buffers but is efficient enough since the
1012 		 * pages usually remain in VMIO for some time.
1013 		 */
1014 		bsize = vp->v_bufobj.bo_bsize;
1015 		bstart = rounddown(ap->a_start, bsize);
1016 		bend = roundup(ap->a_end, bsize);
1017 
1018 		/*
1019 		 * Deactivate pages in the specified range from the backing VM
1020 		 * object.  Pages that are resident in the buffer cache will
1021 		 * remain wired until their corresponding buffers are released
1022 		 * below.
1023 		 */
1024 		if (vp->v_object != NULL) {
1025 			start = trunc_page(bstart);
1026 			end = round_page(bend);
1027 			VM_OBJECT_RLOCK(vp->v_object);
1028 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1029 			    OFF_TO_IDX(end));
1030 			VM_OBJECT_RUNLOCK(vp->v_object);
1031 		}
1032 
1033 		bo = &vp->v_bufobj;
1034 		BO_RLOCK(bo);
1035 		startn = bstart / bsize;
1036 		endn = bend / bsize;
1037 		error = bnoreuselist(&bo->bo_clean, bo, startn, endn);
1038 		if (error == 0)
1039 			error = bnoreuselist(&bo->bo_dirty, bo, startn, endn);
1040 		BO_RUNLOCK(bo);
1041 		VOP_UNLOCK(vp, 0);
1042 		break;
1043 	default:
1044 		error = EINVAL;
1045 		break;
1046 	}
1047 	return (error);
1048 }
1049 
1050 int
1051 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1052 {
1053 
1054 	ap->a_vp->v_unpcb = ap->a_unpcb;
1055 	return (0);
1056 }
1057 
1058 int
1059 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1060 {
1061 
1062 	*ap->a_unpcb = ap->a_vp->v_unpcb;
1063 	return (0);
1064 }
1065 
1066 int
1067 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1068 {
1069 
1070 	ap->a_vp->v_unpcb = NULL;
1071 	return (0);
1072 }
1073 
1074 static int
1075 vop_stdis_text(struct vop_is_text_args *ap)
1076 {
1077 
1078 	return (ap->a_vp->v_writecount < 0);
1079 }
1080 
1081 int
1082 vop_stdset_text(struct vop_set_text_args *ap)
1083 {
1084 	struct vnode *vp;
1085 	int error;
1086 
1087 	vp = ap->a_vp;
1088 	VI_LOCK(vp);
1089 	if (vp->v_writecount > 0) {
1090 		error = ETXTBSY;
1091 	} else {
1092 		vp->v_writecount--;
1093 		error = 0;
1094 	}
1095 	VI_UNLOCK(vp);
1096 	return (error);
1097 }
1098 
1099 static int
1100 vop_stdunset_text(struct vop_unset_text_args *ap)
1101 {
1102 	struct vnode *vp;
1103 	int error;
1104 
1105 	vp = ap->a_vp;
1106 	VI_LOCK(vp);
1107 	if (vp->v_writecount < 0) {
1108 		vp->v_writecount++;
1109 		error = 0;
1110 	} else {
1111 		error = EINVAL;
1112 	}
1113 	VI_UNLOCK(vp);
1114 	return (error);
1115 }
1116 
1117 static int
1118 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1119 {
1120 	struct vnode *vp;
1121 	int error;
1122 
1123 	vp = ap->a_vp;
1124 	VI_LOCK_FLAGS(vp, MTX_DUPOK);
1125 	if (vp->v_writecount < 0) {
1126 		error = ETXTBSY;
1127 	} else {
1128 		VNASSERT(vp->v_writecount + ap->a_inc >= 0, vp,
1129 		    ("neg writecount increment %d", ap->a_inc));
1130 		vp->v_writecount += ap->a_inc;
1131 		error = 0;
1132 	}
1133 	VI_UNLOCK(vp);
1134 	return (error);
1135 }
1136 
1137 /*
1138  * vfs default ops
1139  * used to fill the vfs function table to get reasonable default return values.
1140  */
1141 int
1142 vfs_stdroot (mp, flags, vpp)
1143 	struct mount *mp;
1144 	int flags;
1145 	struct vnode **vpp;
1146 {
1147 
1148 	return (EOPNOTSUPP);
1149 }
1150 
1151 int
1152 vfs_stdstatfs (mp, sbp)
1153 	struct mount *mp;
1154 	struct statfs *sbp;
1155 {
1156 
1157 	return (EOPNOTSUPP);
1158 }
1159 
1160 int
1161 vfs_stdquotactl (mp, cmds, uid, arg)
1162 	struct mount *mp;
1163 	int cmds;
1164 	uid_t uid;
1165 	void *arg;
1166 {
1167 
1168 	return (EOPNOTSUPP);
1169 }
1170 
1171 int
1172 vfs_stdsync(mp, waitfor)
1173 	struct mount *mp;
1174 	int waitfor;
1175 {
1176 	struct vnode *vp, *mvp;
1177 	struct thread *td;
1178 	int error, lockreq, allerror = 0;
1179 
1180 	td = curthread;
1181 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1182 	if (waitfor != MNT_WAIT)
1183 		lockreq |= LK_NOWAIT;
1184 	/*
1185 	 * Force stale buffer cache information to be flushed.
1186 	 */
1187 loop:
1188 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1189 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1190 			VI_UNLOCK(vp);
1191 			continue;
1192 		}
1193 		if ((error = vget(vp, lockreq, td)) != 0) {
1194 			if (error == ENOENT) {
1195 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1196 				goto loop;
1197 			}
1198 			continue;
1199 		}
1200 		error = VOP_FSYNC(vp, waitfor, td);
1201 		if (error)
1202 			allerror = error;
1203 		vput(vp);
1204 	}
1205 	return (allerror);
1206 }
1207 
1208 int
1209 vfs_stdnosync (mp, waitfor)
1210 	struct mount *mp;
1211 	int waitfor;
1212 {
1213 
1214 	return (0);
1215 }
1216 
1217 static int
1218 vop_stdcopy_file_range(struct vop_copy_file_range_args *ap)
1219 {
1220 	int error;
1221 
1222 	error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
1223 	    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags, ap->a_incred,
1224 	    ap->a_outcred, ap->a_fsizetd);
1225 	return (error);
1226 }
1227 
1228 int
1229 vfs_stdvget (mp, ino, flags, vpp)
1230 	struct mount *mp;
1231 	ino_t ino;
1232 	int flags;
1233 	struct vnode **vpp;
1234 {
1235 
1236 	return (EOPNOTSUPP);
1237 }
1238 
1239 int
1240 vfs_stdfhtovp (mp, fhp, flags, vpp)
1241 	struct mount *mp;
1242 	struct fid *fhp;
1243 	int flags;
1244 	struct vnode **vpp;
1245 {
1246 
1247 	return (EOPNOTSUPP);
1248 }
1249 
1250 int
1251 vfs_stdinit (vfsp)
1252 	struct vfsconf *vfsp;
1253 {
1254 
1255 	return (0);
1256 }
1257 
1258 int
1259 vfs_stduninit (vfsp)
1260 	struct vfsconf *vfsp;
1261 {
1262 
1263 	return(0);
1264 }
1265 
1266 int
1267 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1268 	struct mount *mp;
1269 	int cmd;
1270 	struct vnode *filename_vp;
1271 	int attrnamespace;
1272 	const char *attrname;
1273 {
1274 
1275 	if (filename_vp != NULL)
1276 		VOP_UNLOCK(filename_vp, 0);
1277 	return (EOPNOTSUPP);
1278 }
1279 
1280 int
1281 vfs_stdsysctl(mp, op, req)
1282 	struct mount *mp;
1283 	fsctlop_t op;
1284 	struct sysctl_req *req;
1285 {
1286 
1287 	return (EOPNOTSUPP);
1288 }
1289 
1290 static vop_bypass_t *
1291 bp_by_off(struct vop_vector *vop, struct vop_generic_args *a)
1292 {
1293 
1294 	return (*(vop_bypass_t **)((char *)vop + a->a_desc->vdesc_vop_offset));
1295 }
1296 
1297 int
1298 vop_sigdefer(struct vop_vector *vop, struct vop_generic_args *a)
1299 {
1300 	vop_bypass_t *bp;
1301 	int prev_stops, rc;
1302 
1303 	for (; vop != NULL; vop = vop->vop_default) {
1304 		bp = bp_by_off(vop, a);
1305 		if (bp != NULL)
1306 			break;
1307 
1308 		/*
1309 		 * Bypass is not really supported.  It is done for
1310 		 * fallback to unimplemented vops in the default
1311 		 * vector.
1312 		 */
1313 		bp = vop->vop_bypass;
1314 		if (bp != NULL)
1315 			break;
1316 	}
1317 	MPASS(bp != NULL);
1318 
1319 	prev_stops = sigdeferstop(SIGDEFERSTOP_SILENT);
1320 	rc = bp(a);
1321 	sigallowstop(prev_stops);
1322 	return (rc);
1323 }
1324