xref: /freebsd/sys/kern/vfs_default.c (revision 123af6ec70016f5556da5972d4d63c7d175c06d3)
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 		case _PC_ACL_EXTENDED:
486 		case _PC_ACL_NFS4:
487 		case _PC_CAP_PRESENT:
488 		case _PC_INF_PRESENT:
489 		case _PC_MAC_PRESENT:
490 			*ap->a_retval = 0;
491 			return (0);
492 		default:
493 			return (EINVAL);
494 	}
495 	/* NOTREACHED */
496 }
497 
498 /*
499  * Standard lock, unlock and islocked functions.
500  */
501 int
502 vop_stdlock(ap)
503 	struct vop_lock1_args /* {
504 		struct vnode *a_vp;
505 		int a_flags;
506 		char *file;
507 		int line;
508 	} */ *ap;
509 {
510 	struct vnode *vp = ap->a_vp;
511 	struct mtx *ilk;
512 
513 	ilk = VI_MTX(vp);
514 	return (lockmgr_lock_fast_path(vp->v_vnlock, ap->a_flags,
515 	    &ilk->lock_object, ap->a_file, ap->a_line));
516 }
517 
518 /* See above. */
519 int
520 vop_stdunlock(ap)
521 	struct vop_unlock_args /* {
522 		struct vnode *a_vp;
523 		int a_flags;
524 	} */ *ap;
525 {
526 	struct vnode *vp = ap->a_vp;
527 	struct mtx *ilk;
528 
529 	ilk = VI_MTX(vp);
530 	return (lockmgr_unlock_fast_path(vp->v_vnlock, ap->a_flags,
531 	    &ilk->lock_object));
532 }
533 
534 /* See above. */
535 int
536 vop_stdislocked(ap)
537 	struct vop_islocked_args /* {
538 		struct vnode *a_vp;
539 	} */ *ap;
540 {
541 
542 	return (lockstatus(ap->a_vp->v_vnlock));
543 }
544 
545 /*
546  * Return true for select/poll.
547  */
548 int
549 vop_nopoll(ap)
550 	struct vop_poll_args /* {
551 		struct vnode *a_vp;
552 		int  a_events;
553 		struct ucred *a_cred;
554 		struct thread *a_td;
555 	} */ *ap;
556 {
557 
558 	return (poll_no_poll(ap->a_events));
559 }
560 
561 /*
562  * Implement poll for local filesystems that support it.
563  */
564 int
565 vop_stdpoll(ap)
566 	struct vop_poll_args /* {
567 		struct vnode *a_vp;
568 		int  a_events;
569 		struct ucred *a_cred;
570 		struct thread *a_td;
571 	} */ *ap;
572 {
573 	if (ap->a_events & ~POLLSTANDARD)
574 		return (vn_pollrecord(ap->a_vp, ap->a_td, ap->a_events));
575 	return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM));
576 }
577 
578 /*
579  * Return our mount point, as we will take charge of the writes.
580  */
581 int
582 vop_stdgetwritemount(ap)
583 	struct vop_getwritemount_args /* {
584 		struct vnode *a_vp;
585 		struct mount **a_mpp;
586 	} */ *ap;
587 {
588 	struct mount *mp;
589 
590 	/*
591 	 * XXX Since this is called unlocked we may be recycled while
592 	 * attempting to ref the mount.  If this is the case or mountpoint
593 	 * will be set to NULL.  We only have to prevent this call from
594 	 * returning with a ref to an incorrect mountpoint.  It is not
595 	 * harmful to return with a ref to our previous mountpoint.
596 	 */
597 	mp = ap->a_vp->v_mount;
598 	if (mp != NULL) {
599 		vfs_ref(mp);
600 		if (mp != ap->a_vp->v_mount) {
601 			vfs_rel(mp);
602 			mp = NULL;
603 		}
604 	}
605 	*(ap->a_mpp) = mp;
606 	return (0);
607 }
608 
609 /* XXX Needs good comment and VOP_BMAP(9) manpage */
610 int
611 vop_stdbmap(ap)
612 	struct vop_bmap_args /* {
613 		struct vnode *a_vp;
614 		daddr_t  a_bn;
615 		struct bufobj **a_bop;
616 		daddr_t *a_bnp;
617 		int *a_runp;
618 		int *a_runb;
619 	} */ *ap;
620 {
621 
622 	if (ap->a_bop != NULL)
623 		*ap->a_bop = &ap->a_vp->v_bufobj;
624 	if (ap->a_bnp != NULL)
625 		*ap->a_bnp = ap->a_bn * btodb(ap->a_vp->v_mount->mnt_stat.f_iosize);
626 	if (ap->a_runp != NULL)
627 		*ap->a_runp = 0;
628 	if (ap->a_runb != NULL)
629 		*ap->a_runb = 0;
630 	return (0);
631 }
632 
633 int
634 vop_stdfsync(ap)
635 	struct vop_fsync_args /* {
636 		struct vnode *a_vp;
637 		int a_waitfor;
638 		struct thread *a_td;
639 	} */ *ap;
640 {
641 	struct vnode *vp;
642 	struct buf *bp, *nbp;
643 	struct bufobj *bo;
644 	struct mount *mp;
645 	int error, maxretry;
646 
647 	error = 0;
648 	maxretry = 10000;     /* large, arbitrarily chosen */
649 	vp = ap->a_vp;
650 	mp = NULL;
651 	if (vp->v_type == VCHR) {
652 		VI_LOCK(vp);
653 		mp = vp->v_rdev->si_mountpt;
654 		VI_UNLOCK(vp);
655 	}
656 	bo = &vp->v_bufobj;
657 	BO_LOCK(bo);
658 loop1:
659 	/*
660 	 * MARK/SCAN initialization to avoid infinite loops.
661 	 */
662         TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) {
663                 bp->b_vflags &= ~BV_SCANNED;
664 		bp->b_error = 0;
665 	}
666 
667 	/*
668 	 * Flush all dirty buffers associated with a vnode.
669 	 */
670 loop2:
671 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
672 		if ((bp->b_vflags & BV_SCANNED) != 0)
673 			continue;
674 		bp->b_vflags |= BV_SCANNED;
675 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
676 			if (ap->a_waitfor != MNT_WAIT)
677 				continue;
678 			if (BUF_LOCK(bp,
679 			    LK_EXCLUSIVE | LK_INTERLOCK | LK_SLEEPFAIL,
680 			    BO_LOCKPTR(bo)) != 0) {
681 				BO_LOCK(bo);
682 				goto loop1;
683 			}
684 			BO_LOCK(bo);
685 		}
686 		BO_UNLOCK(bo);
687 		KASSERT(bp->b_bufobj == bo,
688 		    ("bp %p wrong b_bufobj %p should be %p",
689 		    bp, bp->b_bufobj, bo));
690 		if ((bp->b_flags & B_DELWRI) == 0)
691 			panic("fsync: not dirty");
692 		if ((vp->v_object != NULL) && (bp->b_flags & B_CLUSTEROK)) {
693 			vfs_bio_awrite(bp);
694 		} else {
695 			bremfree(bp);
696 			bawrite(bp);
697 		}
698 		if (maxretry < 1000)
699 			pause("dirty", hz < 1000 ? 1 : hz / 1000);
700 		BO_LOCK(bo);
701 		goto loop2;
702 	}
703 
704 	/*
705 	 * If synchronous the caller expects us to completely resolve all
706 	 * dirty buffers in the system.  Wait for in-progress I/O to
707 	 * complete (which could include background bitmap writes), then
708 	 * retry if dirty blocks still exist.
709 	 */
710 	if (ap->a_waitfor == MNT_WAIT) {
711 		bufobj_wwait(bo, 0, 0);
712 		if (bo->bo_dirty.bv_cnt > 0) {
713 			/*
714 			 * If we are unable to write any of these buffers
715 			 * then we fail now rather than trying endlessly
716 			 * to write them out.
717 			 */
718 			TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs)
719 				if ((error = bp->b_error) != 0)
720 					break;
721 			if ((mp != NULL && mp->mnt_secondary_writes > 0) ||
722 			    (error == 0 && --maxretry >= 0))
723 				goto loop1;
724 			if (error == 0)
725 				error = EAGAIN;
726 		}
727 	}
728 	BO_UNLOCK(bo);
729 	if (error != 0)
730 		vn_printf(vp, "fsync: giving up on dirty (error = %d) ", error);
731 
732 	return (error);
733 }
734 
735 static int
736 vop_stdfdatasync(struct vop_fdatasync_args *ap)
737 {
738 
739 	return (VOP_FSYNC(ap->a_vp, MNT_WAIT, ap->a_td));
740 }
741 
742 int
743 vop_stdfdatasync_buf(struct vop_fdatasync_args *ap)
744 {
745 	struct vop_fsync_args apf;
746 
747 	apf.a_vp = ap->a_vp;
748 	apf.a_waitfor = MNT_WAIT;
749 	apf.a_td = ap->a_td;
750 	return (vop_stdfsync(&apf));
751 }
752 
753 /* XXX Needs good comment and more info in the manpage (VOP_GETPAGES(9)). */
754 int
755 vop_stdgetpages(ap)
756 	struct vop_getpages_args /* {
757 		struct vnode *a_vp;
758 		vm_page_t *a_m;
759 		int a_count;
760 		int *a_rbehind;
761 		int *a_rahead;
762 	} */ *ap;
763 {
764 
765 	return vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
766 	    ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL);
767 }
768 
769 static int
770 vop_stdgetpages_async(struct vop_getpages_async_args *ap)
771 {
772 	int error;
773 
774 	error = VOP_GETPAGES(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind,
775 	    ap->a_rahead);
776 	ap->a_iodone(ap->a_arg, ap->a_m, ap->a_count, error);
777 	return (error);
778 }
779 
780 int
781 vop_stdkqfilter(struct vop_kqfilter_args *ap)
782 {
783 	return vfs_kqfilter(ap);
784 }
785 
786 /* XXX Needs good comment and more info in the manpage (VOP_PUTPAGES(9)). */
787 int
788 vop_stdputpages(ap)
789 	struct vop_putpages_args /* {
790 		struct vnode *a_vp;
791 		vm_page_t *a_m;
792 		int a_count;
793 		int a_sync;
794 		int *a_rtvals;
795 	} */ *ap;
796 {
797 
798 	return vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
799 	     ap->a_sync, ap->a_rtvals);
800 }
801 
802 int
803 vop_stdvptofh(struct vop_vptofh_args *ap)
804 {
805 	return (EOPNOTSUPP);
806 }
807 
808 int
809 vop_stdvptocnp(struct vop_vptocnp_args *ap)
810 {
811 	struct vnode *vp = ap->a_vp;
812 	struct vnode **dvp = ap->a_vpp;
813 	struct ucred *cred = ap->a_cred;
814 	char *buf = ap->a_buf;
815 	int *buflen = ap->a_buflen;
816 	char *dirbuf, *cpos;
817 	int i, error, eofflag, dirbuflen, flags, locked, len, covered;
818 	off_t off;
819 	ino_t fileno;
820 	struct vattr va;
821 	struct nameidata nd;
822 	struct thread *td;
823 	struct dirent *dp;
824 	struct vnode *mvp;
825 
826 	i = *buflen;
827 	error = 0;
828 	covered = 0;
829 	td = curthread;
830 
831 	if (vp->v_type != VDIR)
832 		return (ENOENT);
833 
834 	error = VOP_GETATTR(vp, &va, cred);
835 	if (error)
836 		return (error);
837 
838 	VREF(vp);
839 	locked = VOP_ISLOCKED(vp);
840 	VOP_UNLOCK(vp, 0);
841 	NDINIT_ATVP(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
842 	    "..", vp, td);
843 	flags = FREAD;
844 	error = vn_open_cred(&nd, &flags, 0, VN_OPEN_NOAUDIT, cred, NULL);
845 	if (error) {
846 		vn_lock(vp, locked | LK_RETRY);
847 		return (error);
848 	}
849 	NDFREE(&nd, NDF_ONLY_PNBUF);
850 
851 	mvp = *dvp = nd.ni_vp;
852 
853 	if (vp->v_mount != (*dvp)->v_mount &&
854 	    ((*dvp)->v_vflag & VV_ROOT) &&
855 	    ((*dvp)->v_mount->mnt_flag & MNT_UNION)) {
856 		*dvp = (*dvp)->v_mount->mnt_vnodecovered;
857 		VREF(mvp);
858 		VOP_UNLOCK(mvp, 0);
859 		vn_close(mvp, FREAD, cred, td);
860 		VREF(*dvp);
861 		vn_lock(*dvp, LK_SHARED | LK_RETRY);
862 		covered = 1;
863 	}
864 
865 	fileno = va.va_fileid;
866 
867 	dirbuflen = DEV_BSIZE;
868 	if (dirbuflen < va.va_blocksize)
869 		dirbuflen = va.va_blocksize;
870 	dirbuf = (char *)malloc(dirbuflen, M_TEMP, M_WAITOK);
871 
872 	if ((*dvp)->v_type != VDIR) {
873 		error = ENOENT;
874 		goto out;
875 	}
876 
877 	off = 0;
878 	len = 0;
879 	do {
880 		/* call VOP_READDIR of parent */
881 		error = get_next_dirent(*dvp, &dp, dirbuf, dirbuflen, &off,
882 					&cpos, &len, &eofflag, td);
883 		if (error)
884 			goto out;
885 
886 		if ((dp->d_type != DT_WHT) &&
887 		    (dp->d_fileno == fileno)) {
888 			if (covered) {
889 				VOP_UNLOCK(*dvp, 0);
890 				vn_lock(mvp, LK_SHARED | LK_RETRY);
891 				if (dirent_exists(mvp, dp->d_name, td)) {
892 					error = ENOENT;
893 					VOP_UNLOCK(mvp, 0);
894 					vn_lock(*dvp, LK_SHARED | LK_RETRY);
895 					goto out;
896 				}
897 				VOP_UNLOCK(mvp, 0);
898 				vn_lock(*dvp, LK_SHARED | LK_RETRY);
899 			}
900 			i -= dp->d_namlen;
901 
902 			if (i < 0) {
903 				error = ENOMEM;
904 				goto out;
905 			}
906 			if (dp->d_namlen == 1 && dp->d_name[0] == '.') {
907 				error = ENOENT;
908 			} else {
909 				bcopy(dp->d_name, buf + i, dp->d_namlen);
910 				error = 0;
911 			}
912 			goto out;
913 		}
914 	} while (len > 0 || !eofflag);
915 	error = ENOENT;
916 
917 out:
918 	free(dirbuf, M_TEMP);
919 	if (!error) {
920 		*buflen = i;
921 		vref(*dvp);
922 	}
923 	if (covered) {
924 		vput(*dvp);
925 		vrele(mvp);
926 	} else {
927 		VOP_UNLOCK(mvp, 0);
928 		vn_close(mvp, FREAD, cred, td);
929 	}
930 	vn_lock(vp, locked | LK_RETRY);
931 	return (error);
932 }
933 
934 int
935 vop_stdallocate(struct vop_allocate_args *ap)
936 {
937 #ifdef __notyet__
938 	struct statfs *sfs;
939 	off_t maxfilesize = 0;
940 #endif
941 	struct iovec aiov;
942 	struct vattr vattr, *vap;
943 	struct uio auio;
944 	off_t fsize, len, cur, offset;
945 	uint8_t *buf;
946 	struct thread *td;
947 	struct vnode *vp;
948 	size_t iosize;
949 	int error;
950 
951 	buf = NULL;
952 	error = 0;
953 	td = curthread;
954 	vap = &vattr;
955 	vp = ap->a_vp;
956 	len = *ap->a_len;
957 	offset = *ap->a_offset;
958 
959 	error = VOP_GETATTR(vp, vap, td->td_ucred);
960 	if (error != 0)
961 		goto out;
962 	fsize = vap->va_size;
963 	iosize = vap->va_blocksize;
964 	if (iosize == 0)
965 		iosize = BLKDEV_IOSIZE;
966 	if (iosize > MAXPHYS)
967 		iosize = MAXPHYS;
968 	buf = malloc(iosize, M_TEMP, M_WAITOK);
969 
970 #ifdef __notyet__
971 	/*
972 	 * Check if the filesystem sets f_maxfilesize; if not use
973 	 * VOP_SETATTR to perform the check.
974 	 */
975 	sfs = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK);
976 	error = VFS_STATFS(vp->v_mount, sfs, td);
977 	if (error == 0)
978 		maxfilesize = sfs->f_maxfilesize;
979 	free(sfs, M_STATFS);
980 	if (error != 0)
981 		goto out;
982 	if (maxfilesize) {
983 		if (offset > maxfilesize || len > maxfilesize ||
984 		    offset + len > maxfilesize) {
985 			error = EFBIG;
986 			goto out;
987 		}
988 	} else
989 #endif
990 	if (offset + len > vap->va_size) {
991 		/*
992 		 * Test offset + len against the filesystem's maxfilesize.
993 		 */
994 		VATTR_NULL(vap);
995 		vap->va_size = offset + len;
996 		error = VOP_SETATTR(vp, vap, td->td_ucred);
997 		if (error != 0)
998 			goto out;
999 		VATTR_NULL(vap);
1000 		vap->va_size = fsize;
1001 		error = VOP_SETATTR(vp, vap, td->td_ucred);
1002 		if (error != 0)
1003 			goto out;
1004 	}
1005 
1006 	for (;;) {
1007 		/*
1008 		 * Read and write back anything below the nominal file
1009 		 * size.  There's currently no way outside the filesystem
1010 		 * to know whether this area is sparse or not.
1011 		 */
1012 		cur = iosize;
1013 		if ((offset % iosize) != 0)
1014 			cur -= (offset % iosize);
1015 		if (cur > len)
1016 			cur = len;
1017 		if (offset < fsize) {
1018 			aiov.iov_base = buf;
1019 			aiov.iov_len = cur;
1020 			auio.uio_iov = &aiov;
1021 			auio.uio_iovcnt = 1;
1022 			auio.uio_offset = offset;
1023 			auio.uio_resid = cur;
1024 			auio.uio_segflg = UIO_SYSSPACE;
1025 			auio.uio_rw = UIO_READ;
1026 			auio.uio_td = td;
1027 			error = VOP_READ(vp, &auio, 0, td->td_ucred);
1028 			if (error != 0)
1029 				break;
1030 			if (auio.uio_resid > 0) {
1031 				bzero(buf + cur - auio.uio_resid,
1032 				    auio.uio_resid);
1033 			}
1034 		} else {
1035 			bzero(buf, cur);
1036 		}
1037 
1038 		aiov.iov_base = buf;
1039 		aiov.iov_len = cur;
1040 		auio.uio_iov = &aiov;
1041 		auio.uio_iovcnt = 1;
1042 		auio.uio_offset = offset;
1043 		auio.uio_resid = cur;
1044 		auio.uio_segflg = UIO_SYSSPACE;
1045 		auio.uio_rw = UIO_WRITE;
1046 		auio.uio_td = td;
1047 
1048 		error = VOP_WRITE(vp, &auio, 0, td->td_ucred);
1049 		if (error != 0)
1050 			break;
1051 
1052 		len -= cur;
1053 		offset += cur;
1054 		if (len == 0)
1055 			break;
1056 		if (should_yield())
1057 			break;
1058 	}
1059 
1060  out:
1061 	*ap->a_len = len;
1062 	*ap->a_offset = offset;
1063 	free(buf, M_TEMP);
1064 	return (error);
1065 }
1066 
1067 int
1068 vop_stdadvise(struct vop_advise_args *ap)
1069 {
1070 	struct vnode *vp;
1071 	struct bufobj *bo;
1072 	daddr_t startn, endn;
1073 	off_t bstart, bend, start, end;
1074 	int bsize, error;
1075 
1076 	vp = ap->a_vp;
1077 	switch (ap->a_advice) {
1078 	case POSIX_FADV_WILLNEED:
1079 		/*
1080 		 * Do nothing for now.  Filesystems should provide a
1081 		 * custom method which starts an asynchronous read of
1082 		 * the requested region.
1083 		 */
1084 		error = 0;
1085 		break;
1086 	case POSIX_FADV_DONTNEED:
1087 		error = 0;
1088 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1089 		if (vp->v_iflag & VI_DOOMED) {
1090 			VOP_UNLOCK(vp, 0);
1091 			break;
1092 		}
1093 
1094 		/*
1095 		 * Round to block boundaries (and later possibly further to
1096 		 * page boundaries).  Applications cannot reasonably be aware
1097 		 * of the boundaries, and the rounding must be to expand at
1098 		 * both extremities to cover enough.  It still doesn't cover
1099 		 * read-ahead.  For partial blocks, this gives unnecessary
1100 		 * discarding of buffers but is efficient enough since the
1101 		 * pages usually remain in VMIO for some time.
1102 		 */
1103 		bsize = vp->v_bufobj.bo_bsize;
1104 		bstart = rounddown(ap->a_start, bsize);
1105 		bend = roundup(ap->a_end, bsize);
1106 
1107 		/*
1108 		 * Deactivate pages in the specified range from the backing VM
1109 		 * object.  Pages that are resident in the buffer cache will
1110 		 * remain wired until their corresponding buffers are released
1111 		 * below.
1112 		 */
1113 		if (vp->v_object != NULL) {
1114 			start = trunc_page(bstart);
1115 			end = round_page(bend);
1116 			VM_OBJECT_RLOCK(vp->v_object);
1117 			vm_object_page_noreuse(vp->v_object, OFF_TO_IDX(start),
1118 			    OFF_TO_IDX(end));
1119 			VM_OBJECT_RUNLOCK(vp->v_object);
1120 		}
1121 
1122 		bo = &vp->v_bufobj;
1123 		BO_RLOCK(bo);
1124 		startn = bstart / bsize;
1125 		endn = bend / bsize;
1126 		error = bnoreuselist(&bo->bo_clean, bo, startn, endn);
1127 		if (error == 0)
1128 			error = bnoreuselist(&bo->bo_dirty, bo, startn, endn);
1129 		BO_RUNLOCK(bo);
1130 		VOP_UNLOCK(vp, 0);
1131 		break;
1132 	default:
1133 		error = EINVAL;
1134 		break;
1135 	}
1136 	return (error);
1137 }
1138 
1139 int
1140 vop_stdunp_bind(struct vop_unp_bind_args *ap)
1141 {
1142 
1143 	ap->a_vp->v_unpcb = ap->a_unpcb;
1144 	return (0);
1145 }
1146 
1147 int
1148 vop_stdunp_connect(struct vop_unp_connect_args *ap)
1149 {
1150 
1151 	*ap->a_unpcb = ap->a_vp->v_unpcb;
1152 	return (0);
1153 }
1154 
1155 int
1156 vop_stdunp_detach(struct vop_unp_detach_args *ap)
1157 {
1158 
1159 	ap->a_vp->v_unpcb = NULL;
1160 	return (0);
1161 }
1162 
1163 static int
1164 vop_stdis_text(struct vop_is_text_args *ap)
1165 {
1166 
1167 	return ((ap->a_vp->v_vflag & VV_TEXT) != 0);
1168 }
1169 
1170 static int
1171 vop_stdset_text(struct vop_set_text_args *ap)
1172 {
1173 
1174 	ap->a_vp->v_vflag |= VV_TEXT;
1175 	return (0);
1176 }
1177 
1178 static int
1179 vop_stdunset_text(struct vop_unset_text_args *ap)
1180 {
1181 
1182 	ap->a_vp->v_vflag &= ~VV_TEXT;
1183 	return (0);
1184 }
1185 
1186 static int
1187 vop_stdget_writecount(struct vop_get_writecount_args *ap)
1188 {
1189 
1190 	*ap->a_writecount = ap->a_vp->v_writecount;
1191 	return (0);
1192 }
1193 
1194 static int
1195 vop_stdadd_writecount(struct vop_add_writecount_args *ap)
1196 {
1197 
1198 	ap->a_vp->v_writecount += ap->a_inc;
1199 	return (0);
1200 }
1201 
1202 /*
1203  * vfs default ops
1204  * used to fill the vfs function table to get reasonable default return values.
1205  */
1206 int
1207 vfs_stdroot (mp, flags, vpp)
1208 	struct mount *mp;
1209 	int flags;
1210 	struct vnode **vpp;
1211 {
1212 
1213 	return (EOPNOTSUPP);
1214 }
1215 
1216 int
1217 vfs_stdstatfs (mp, sbp)
1218 	struct mount *mp;
1219 	struct statfs *sbp;
1220 {
1221 
1222 	return (EOPNOTSUPP);
1223 }
1224 
1225 int
1226 vfs_stdquotactl (mp, cmds, uid, arg)
1227 	struct mount *mp;
1228 	int cmds;
1229 	uid_t uid;
1230 	void *arg;
1231 {
1232 
1233 	return (EOPNOTSUPP);
1234 }
1235 
1236 int
1237 vfs_stdsync(mp, waitfor)
1238 	struct mount *mp;
1239 	int waitfor;
1240 {
1241 	struct vnode *vp, *mvp;
1242 	struct thread *td;
1243 	int error, lockreq, allerror = 0;
1244 
1245 	td = curthread;
1246 	lockreq = LK_EXCLUSIVE | LK_INTERLOCK;
1247 	if (waitfor != MNT_WAIT)
1248 		lockreq |= LK_NOWAIT;
1249 	/*
1250 	 * Force stale buffer cache information to be flushed.
1251 	 */
1252 loop:
1253 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1254 		if (vp->v_bufobj.bo_dirty.bv_cnt == 0) {
1255 			VI_UNLOCK(vp);
1256 			continue;
1257 		}
1258 		if ((error = vget(vp, lockreq, td)) != 0) {
1259 			if (error == ENOENT) {
1260 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1261 				goto loop;
1262 			}
1263 			continue;
1264 		}
1265 		error = VOP_FSYNC(vp, waitfor, td);
1266 		if (error)
1267 			allerror = error;
1268 		vput(vp);
1269 	}
1270 	return (allerror);
1271 }
1272 
1273 int
1274 vfs_stdnosync (mp, waitfor)
1275 	struct mount *mp;
1276 	int waitfor;
1277 {
1278 
1279 	return (0);
1280 }
1281 
1282 int
1283 vfs_stdvget (mp, ino, flags, vpp)
1284 	struct mount *mp;
1285 	ino_t ino;
1286 	int flags;
1287 	struct vnode **vpp;
1288 {
1289 
1290 	return (EOPNOTSUPP);
1291 }
1292 
1293 int
1294 vfs_stdfhtovp (mp, fhp, flags, vpp)
1295 	struct mount *mp;
1296 	struct fid *fhp;
1297 	int flags;
1298 	struct vnode **vpp;
1299 {
1300 
1301 	return (EOPNOTSUPP);
1302 }
1303 
1304 int
1305 vfs_stdinit (vfsp)
1306 	struct vfsconf *vfsp;
1307 {
1308 
1309 	return (0);
1310 }
1311 
1312 int
1313 vfs_stduninit (vfsp)
1314 	struct vfsconf *vfsp;
1315 {
1316 
1317 	return(0);
1318 }
1319 
1320 int
1321 vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)
1322 	struct mount *mp;
1323 	int cmd;
1324 	struct vnode *filename_vp;
1325 	int attrnamespace;
1326 	const char *attrname;
1327 {
1328 
1329 	if (filename_vp != NULL)
1330 		VOP_UNLOCK(filename_vp, 0);
1331 	return (EOPNOTSUPP);
1332 }
1333 
1334 int
1335 vfs_stdsysctl(mp, op, req)
1336 	struct mount *mp;
1337 	fsctlop_t op;
1338 	struct sysctl_req *req;
1339 {
1340 
1341 	return (EOPNOTSUPP);
1342 }
1343 
1344 static vop_bypass_t *
1345 bp_by_off(struct vop_vector *vop, struct vop_generic_args *a)
1346 {
1347 
1348 	return (*(vop_bypass_t **)((char *)vop + a->a_desc->vdesc_vop_offset));
1349 }
1350 
1351 int
1352 vop_sigdefer(struct vop_vector *vop, struct vop_generic_args *a)
1353 {
1354 	vop_bypass_t *bp;
1355 	int prev_stops, rc;
1356 
1357 	for (; vop != NULL; vop = vop->vop_default) {
1358 		bp = bp_by_off(vop, a);
1359 		if (bp != NULL)
1360 			break;
1361 
1362 		/*
1363 		 * Bypass is not really supported.  It is done for
1364 		 * fallback to unimplemented vops in the default
1365 		 * vector.
1366 		 */
1367 		bp = vop->vop_bypass;
1368 		if (bp != NULL)
1369 			break;
1370 	}
1371 	MPASS(bp != NULL);
1372 
1373 	prev_stops = sigdeferstop(SIGDEFERSTOP_SILENT);
1374 	rc = bp(a);
1375 	sigallowstop(prev_stops);
1376 	return (rc);
1377 }
1378