xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision 3332f1b444d4a73238e9f59cca27bfc95fe936bd)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $	*/
3 
4 /*-
5  * SPDX-License-Identifier: BSD-4-Clause
6  *
7  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
8  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
9  * All rights reserved.
10  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
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. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by TooLs GmbH.
23  * 4. The name of TooLs GmbH may not be used to endorse or promote products
24  *    derived from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
28  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
29  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
32  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
34  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
35  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36  */
37 /*-
38  * Written by Paul Popelka (paulp@uts.amdahl.com)
39  *
40  * You can do anything you want with this software, just don't say you wrote
41  * it, and don't remove this notice.
42  *
43  * This software is provided "as is".
44  *
45  * The author supplies this software to be publicly redistributed on the
46  * understanding that the author is not responsible for the correct
47  * functioning of this software in any circumstances and is not liable for
48  * any damages caused by this software.
49  *
50  * October 1992
51  */
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/bio.h>
56 #include <sys/buf.h>
57 #include <sys/clock.h>
58 #include <sys/dirent.h>
59 #include <sys/lock.h>
60 #include <sys/lockf.h>
61 #include <sys/malloc.h>
62 #include <sys/mount.h>
63 #include <sys/mutex.h>
64 #include <sys/namei.h>
65 #include <sys/priv.h>
66 #include <sys/stat.h>
67 #include <sys/sysctl.h>
68 #include <sys/unistd.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vnode_pager.h>
75 
76 #include <fs/msdosfs/bpb.h>
77 #include <fs/msdosfs/direntry.h>
78 #include <fs/msdosfs/denode.h>
79 #include <fs/msdosfs/fat.h>
80 #include <fs/msdosfs/msdosfsmount.h>
81 
82 /*
83  * Prototypes for MSDOSFS vnode operations
84  */
85 static vop_create_t	msdosfs_create;
86 static vop_mknod_t	msdosfs_mknod;
87 static vop_open_t	msdosfs_open;
88 static vop_close_t	msdosfs_close;
89 static vop_access_t	msdosfs_access;
90 static vop_getattr_t	msdosfs_getattr;
91 static vop_setattr_t	msdosfs_setattr;
92 static vop_read_t	msdosfs_read;
93 static vop_write_t	msdosfs_write;
94 static vop_fsync_t	msdosfs_fsync;
95 static vop_remove_t	msdosfs_remove;
96 static vop_link_t	msdosfs_link;
97 static vop_rename_t	msdosfs_rename;
98 static vop_mkdir_t	msdosfs_mkdir;
99 static vop_rmdir_t	msdosfs_rmdir;
100 static vop_symlink_t	msdosfs_symlink;
101 static vop_readdir_t	msdosfs_readdir;
102 static vop_bmap_t	msdosfs_bmap;
103 static vop_getpages_t	msdosfs_getpages;
104 static vop_strategy_t	msdosfs_strategy;
105 static vop_print_t	msdosfs_print;
106 static vop_pathconf_t	msdosfs_pathconf;
107 static vop_vptofh_t	msdosfs_vptofh;
108 
109 /*
110  * Some general notes:
111  *
112  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
113  * read/written using the vnode for the filesystem. Blocks that represent
114  * the contents of a file are read/written using the vnode for the file
115  * (including directories when they are read/written as files). This
116  * presents problems for the dos filesystem because data that should be in
117  * an inode (if dos had them) resides in the directory itself.  Since we
118  * must update directory entries without the benefit of having the vnode
119  * for the directory we must use the vnode for the filesystem.  This means
120  * that when a directory is actually read/written (via read, write, or
121  * readdir, or seek) we must use the vnode for the filesystem instead of
122  * the vnode for the directory as would happen in ufs. This is to insure we
123  * retrieve the correct block from the buffer cache since the hash value is
124  * based upon the vnode address and the desired block number.
125  */
126 
127 /*
128  * Create a regular file. On entry the directory to contain the file being
129  * created is locked.  We must release before we return. We must also free
130  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
131  * only if the SAVESTART bit in cn_flags is clear on success.
132  */
133 static int
134 msdosfs_create(struct vop_create_args *ap)
135 {
136 	struct componentname *cnp = ap->a_cnp;
137 	struct denode ndirent;
138 	struct denode *dep;
139 	struct denode *pdep = VTODE(ap->a_dvp);
140 	struct timespec ts;
141 	int error;
142 
143 #ifdef MSDOSFS_DEBUG
144 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
145 #endif
146 
147 	/*
148 	 * If this is the root directory and there is no space left we
149 	 * can't do anything.  This is because the root directory can not
150 	 * change size.
151 	 */
152 	if (pdep->de_StartCluster == MSDOSFSROOT
153 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
154 		error = ENOSPC;
155 		goto bad;
156 	}
157 
158 	/*
159 	 * Create a directory entry for the file, then call createde() to
160 	 * have it installed. NOTE: DOS files are always executable.  We
161 	 * use the absence of the owner write bit to make the file
162 	 * readonly.
163 	 */
164 #ifdef DIAGNOSTIC
165 	if ((cnp->cn_flags & HASBUF) == 0)
166 		panic("msdosfs_create: no name");
167 #endif
168 	memset(&ndirent, 0, sizeof(ndirent));
169 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
170 	if (error)
171 		goto bad;
172 
173 	ndirent.de_Attributes = ATTR_ARCHIVE;
174 	ndirent.de_LowerCase = 0;
175 	ndirent.de_StartCluster = 0;
176 	ndirent.de_FileSize = 0;
177 	ndirent.de_pmp = pdep->de_pmp;
178 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
179 	vfs_timestamp(&ts);
180 	DETIMES(&ndirent, &ts, &ts, &ts);
181 	error = createde(&ndirent, pdep, &dep, cnp);
182 	if (error)
183 		goto bad;
184 	*ap->a_vpp = DETOV(dep);
185 	if ((cnp->cn_flags & MAKEENTRY) != 0)
186 		cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
187 	return (0);
188 
189 bad:
190 	return (error);
191 }
192 
193 static int
194 msdosfs_mknod(struct vop_mknod_args *ap)
195 {
196 
197     return (EINVAL);
198 }
199 
200 static int
201 msdosfs_open(struct vop_open_args *ap)
202 {
203 	struct denode *dep = VTODE(ap->a_vp);
204 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
205 	return 0;
206 }
207 
208 static int
209 msdosfs_close(struct vop_close_args *ap)
210 {
211 	struct vnode *vp = ap->a_vp;
212 	struct denode *dep = VTODE(vp);
213 	struct timespec ts;
214 
215 	VI_LOCK(vp);
216 	if (vp->v_usecount > 1) {
217 		vfs_timestamp(&ts);
218 		DETIMES(dep, &ts, &ts, &ts);
219 	}
220 	VI_UNLOCK(vp);
221 	return 0;
222 }
223 
224 static int
225 msdosfs_access(struct vop_access_args *ap)
226 {
227 	struct vnode *vp = ap->a_vp;
228 	struct denode *dep = VTODE(ap->a_vp);
229 	struct msdosfsmount *pmp = dep->de_pmp;
230 	mode_t file_mode;
231 	accmode_t accmode = ap->a_accmode;
232 
233 	file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
234 	file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
235 
236 	/*
237 	 * Disallow writing to directories and regular files if the
238 	 * filesystem is read-only.
239 	 */
240 	if (accmode & VWRITE) {
241 		switch (vp->v_type) {
242 		case VREG:
243 		case VDIR:
244 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
245 				return (EROFS);
246 			break;
247 		default:
248 			break;
249 		}
250 	}
251 
252 	return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
253 	    ap->a_accmode, ap->a_cred));
254 }
255 
256 static int
257 msdosfs_getattr(struct vop_getattr_args *ap)
258 {
259 	struct denode *dep = VTODE(ap->a_vp);
260 	struct msdosfsmount *pmp = dep->de_pmp;
261 	struct vattr *vap = ap->a_vap;
262 	mode_t mode;
263 	struct timespec ts;
264 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
265 	uint64_t fileid;
266 
267 	vfs_timestamp(&ts);
268 	DETIMES(dep, &ts, &ts, &ts);
269 	vap->va_fsid = dev2udev(pmp->pm_dev);
270 	/*
271 	 * The following computation of the fileid must be the same as that
272 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
273 	 * doesn't work.
274 	 */
275 	if (dep->de_Attributes & ATTR_DIRECTORY) {
276 		fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
277 		    dirsperblk;
278 		if (dep->de_StartCluster == MSDOSFSROOT)
279 			fileid = 1;
280 	} else {
281 		fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
282 		    dirsperblk;
283 		if (dep->de_dirclust == MSDOSFSROOT)
284 			fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
285 		fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
286 	}
287 	vap->va_fileid = fileid;
288 
289 	mode = S_IRWXU|S_IRWXG|S_IRWXO;
290 	if (dep->de_Attributes & ATTR_READONLY)
291 		mode &= ~(S_IWUSR|S_IWGRP|S_IWOTH);
292 	vap->va_mode = mode &
293 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
294 	vap->va_uid = pmp->pm_uid;
295 	vap->va_gid = pmp->pm_gid;
296 	vap->va_nlink = 1;
297 	vap->va_rdev = NODEV;
298 	vap->va_size = dep->de_FileSize;
299 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
300 	vap->va_ctime = vap->va_mtime;
301 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
302 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
303 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
304 		    0, &vap->va_birthtime);
305 	} else {
306 		vap->va_atime = vap->va_mtime;
307 		vap->va_birthtime.tv_sec = -1;
308 		vap->va_birthtime.tv_nsec = 0;
309 	}
310 	vap->va_flags = 0;
311 	if (dep->de_Attributes & ATTR_ARCHIVE)
312 		vap->va_flags |= UF_ARCHIVE;
313 	if (dep->de_Attributes & ATTR_HIDDEN)
314 		vap->va_flags |= UF_HIDDEN;
315 	if (dep->de_Attributes & ATTR_READONLY)
316 		vap->va_flags |= UF_READONLY;
317 	if (dep->de_Attributes & ATTR_SYSTEM)
318 		vap->va_flags |= UF_SYSTEM;
319 	vap->va_gen = 0;
320 	vap->va_blocksize = pmp->pm_bpcluster;
321 	vap->va_bytes =
322 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
323 	vap->va_type = ap->a_vp->v_type;
324 	vap->va_filerev = dep->de_modrev;
325 	return (0);
326 }
327 
328 static int
329 msdosfs_setattr(struct vop_setattr_args *ap)
330 {
331 	struct vnode *vp = ap->a_vp;
332 	struct denode *dep = VTODE(ap->a_vp);
333 	struct msdosfsmount *pmp = dep->de_pmp;
334 	struct vattr *vap = ap->a_vap;
335 	struct ucred *cred = ap->a_cred;
336 	struct thread *td = curthread;
337 	int error = 0;
338 
339 #ifdef MSDOSFS_DEBUG
340 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
341 	    ap->a_vp, vap, cred);
342 #endif
343 
344 	/*
345 	 * Check for unsettable attributes.
346 	 */
347 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
348 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
349 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
350 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
351 #ifdef MSDOSFS_DEBUG
352 		printf("msdosfs_setattr(): returning EINVAL\n");
353 		printf("    va_type %d, va_nlink %llx, va_fsid %llx, va_fileid %llx\n",
354 		    vap->va_type, (unsigned long long)vap->va_nlink,
355 		    (unsigned long long)vap->va_fsid,
356 		    (unsigned long long)vap->va_fileid);
357 		printf("    va_blocksize %lx, va_rdev %llx, va_bytes %llx, va_gen %lx\n",
358 		    vap->va_blocksize, (unsigned long long)vap->va_rdev,
359 		    (unsigned long long)vap->va_bytes, vap->va_gen);
360 		printf("    va_uid %x, va_gid %x\n",
361 		    vap->va_uid, vap->va_gid);
362 #endif
363 		return (EINVAL);
364 	}
365 
366 	/*
367 	 * We don't allow setting attributes on the root directory.
368 	 * The special case for the root directory is because before
369 	 * FAT32, the root directory didn't have an entry for itself
370 	 * (and was otherwise special).  With FAT32, the root
371 	 * directory is not so special, but still doesn't have an
372 	 * entry for itself.
373 	 */
374 	if (vp->v_vflag & VV_ROOT)
375 		return (EINVAL);
376 
377 	if (vap->va_flags != VNOVAL) {
378 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
379 			return (EROFS);
380 		if (cred->cr_uid != pmp->pm_uid) {
381 			error = priv_check_cred(cred, PRIV_VFS_ADMIN);
382 			if (error)
383 				return (error);
384 		}
385 		/*
386 		 * We are very inconsistent about handling unsupported
387 		 * attributes.  We ignored the access time and the
388 		 * read and execute bits.  We were strict for the other
389 		 * attributes.
390 		 */
391 		if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
392 		    UF_SYSTEM))
393 			return EOPNOTSUPP;
394 		if (vap->va_flags & UF_ARCHIVE)
395 			dep->de_Attributes |= ATTR_ARCHIVE;
396 		else
397 			dep->de_Attributes &= ~ATTR_ARCHIVE;
398 		if (vap->va_flags & UF_HIDDEN)
399 			dep->de_Attributes |= ATTR_HIDDEN;
400 		else
401 			dep->de_Attributes &= ~ATTR_HIDDEN;
402 		/* We don't allow changing the readonly bit on directories. */
403 		if (vp->v_type != VDIR) {
404 			if (vap->va_flags & UF_READONLY)
405 				dep->de_Attributes |= ATTR_READONLY;
406 			else
407 				dep->de_Attributes &= ~ATTR_READONLY;
408 		}
409 		if (vap->va_flags & UF_SYSTEM)
410 			dep->de_Attributes |= ATTR_SYSTEM;
411 		else
412 			dep->de_Attributes &= ~ATTR_SYSTEM;
413 		dep->de_flag |= DE_MODIFIED;
414 	}
415 
416 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
417 		uid_t uid;
418 		gid_t gid;
419 
420 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
421 			return (EROFS);
422 		uid = vap->va_uid;
423 		if (uid == (uid_t)VNOVAL)
424 			uid = pmp->pm_uid;
425 		gid = vap->va_gid;
426 		if (gid == (gid_t)VNOVAL)
427 			gid = pmp->pm_gid;
428 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
429 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
430 			error = priv_check_cred(cred, PRIV_VFS_CHOWN);
431 			if (error)
432 				return (error);
433 		}
434 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
435 			return EINVAL;
436 	}
437 
438 	if (vap->va_size != VNOVAL) {
439 		switch (vp->v_type) {
440 		case VDIR:
441 			return (EISDIR);
442 		case VREG:
443 			/*
444 			 * Truncation is only supported for regular files,
445 			 * Disallow it if the filesystem is read-only.
446 			 */
447 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
448 				return (EROFS);
449 			break;
450 		default:
451 			/*
452 			 * According to POSIX, the result is unspecified
453 			 * for file types other than regular files,
454 			 * directories and shared memory objects.  We
455 			 * don't support any file types except regular
456 			 * files and directories in this file system, so
457 			 * this (default) case is unreachable and can do
458 			 * anything.  Keep falling through to detrunc()
459 			 * for now.
460 			 */
461 			break;
462 		}
463 		error = detrunc(dep, vap->va_size, 0, cred);
464 		if (error)
465 			return error;
466 	}
467 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
468 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
469 			return (EROFS);
470 		error = vn_utimes_perm(vp, vap, cred, td);
471 		if (error != 0)
472 			return (error);
473 		if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
474 		    vap->va_atime.tv_sec != VNOVAL) {
475 			dep->de_flag &= ~DE_ACCESS;
476 			timespec2fattime(&vap->va_atime, 0,
477 			    &dep->de_ADate, NULL, NULL);
478 		}
479 		if (vap->va_mtime.tv_sec != VNOVAL) {
480 			dep->de_flag &= ~DE_UPDATE;
481 			timespec2fattime(&vap->va_mtime, 0,
482 			    &dep->de_MDate, &dep->de_MTime, NULL);
483 		}
484 		/*
485 		 * We don't set the archive bit when modifying the time of
486 		 * a directory to emulate the Windows/DOS behavior.
487 		 */
488 		if (vp->v_type != VDIR)
489 			dep->de_Attributes |= ATTR_ARCHIVE;
490 		dep->de_flag |= DE_MODIFIED;
491 	}
492 	/*
493 	 * DOS files only have the ability to have their writability
494 	 * attribute set, so we use the owner write bit to set the readonly
495 	 * attribute.
496 	 */
497 	if (vap->va_mode != (mode_t)VNOVAL) {
498 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
499 			return (EROFS);
500 		if (cred->cr_uid != pmp->pm_uid) {
501 			error = priv_check_cred(cred, PRIV_VFS_ADMIN);
502 			if (error)
503 				return (error);
504 		}
505 		if (vp->v_type != VDIR) {
506 			/* We ignore the read and execute bits. */
507 			if (vap->va_mode & S_IWUSR)
508 				dep->de_Attributes &= ~ATTR_READONLY;
509 			else
510 				dep->de_Attributes |= ATTR_READONLY;
511 			dep->de_Attributes |= ATTR_ARCHIVE;
512 			dep->de_flag |= DE_MODIFIED;
513 		}
514 	}
515 	return (deupdat(dep, 0));
516 }
517 
518 static int
519 msdosfs_read(struct vop_read_args *ap)
520 {
521 	int error = 0;
522 	int blsize;
523 	int isadir;
524 	ssize_t orig_resid;
525 	u_int n;
526 	u_long diff;
527 	u_long on;
528 	daddr_t lbn;
529 	daddr_t rablock;
530 	int rasize;
531 	int seqcount;
532 	struct buf *bp;
533 	struct vnode *vp = ap->a_vp;
534 	struct denode *dep = VTODE(vp);
535 	struct msdosfsmount *pmp = dep->de_pmp;
536 	struct uio *uio = ap->a_uio;
537 
538 	/*
539 	 * If they didn't ask for any data, then we are done.
540 	 */
541 	orig_resid = uio->uio_resid;
542 	if (orig_resid == 0)
543 		return (0);
544 
545 	/*
546 	 * The caller is supposed to ensure that
547 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
548 	 * We don't need to check for large offsets as in ffs because
549 	 * dep->de_FileSize <= MSDOSFS_FILESIZE_MAX < OFF_MAX, so large
550 	 * offsets cannot cause overflow even in theory.
551 	 */
552 
553 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
554 
555 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
556 	do {
557 		if (uio->uio_offset >= dep->de_FileSize)
558 			break;
559 		lbn = de_cluster(pmp, uio->uio_offset);
560 		rablock = lbn + 1;
561 		blsize = pmp->pm_bpcluster;
562 		on = uio->uio_offset & pmp->pm_crbomask;
563 		/*
564 		 * If we are operating on a directory file then be sure to
565 		 * do i/o with the vnode for the filesystem instead of the
566 		 * vnode for the directory.
567 		 */
568 		if (isadir) {
569 			/* convert cluster # to block # */
570 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
571 			if (error == E2BIG) {
572 				error = EINVAL;
573 				break;
574 			} else if (error)
575 				break;
576 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
577 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
578 			error = bread(vp, lbn, blsize, NOCRED, &bp);
579 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
580 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
581 			    NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
582 		} else if (seqcount > 1) {
583 			rasize = blsize;
584 			error = breadn(vp, lbn,
585 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
586 		} else {
587 			error = bread(vp, lbn, blsize, NOCRED, &bp);
588 		}
589 		if (error) {
590 			brelse(bp);
591 			break;
592 		}
593 		diff = pmp->pm_bpcluster - on;
594 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
595 		diff = dep->de_FileSize - uio->uio_offset;
596 		if (diff < n)
597 			n = diff;
598 		diff = blsize - bp->b_resid;
599 		if (diff < n)
600 			n = diff;
601 		error = vn_io_fault_uiomove(bp->b_data + on, (int) n, uio);
602 		brelse(bp);
603 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
604 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
605 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
606 		dep->de_flag |= DE_ACCESS;
607 	return (error);
608 }
609 
610 /*
611  * Write data to a file or directory.
612  */
613 static int
614 msdosfs_write(struct vop_write_args *ap)
615 {
616 	int n;
617 	int croffset;
618 	ssize_t resid;
619 	u_long osize;
620 	int error = 0;
621 	u_long count;
622 	int seqcount;
623 	daddr_t bn, lastcn;
624 	struct buf *bp;
625 	int ioflag = ap->a_ioflag;
626 	struct uio *uio = ap->a_uio;
627 	struct vnode *vp = ap->a_vp;
628 	struct vnode *thisvp;
629 	struct denode *dep = VTODE(vp);
630 	struct msdosfsmount *pmp = dep->de_pmp;
631 	struct ucred *cred = ap->a_cred;
632 
633 #ifdef MSDOSFS_DEBUG
634 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
635 	    vp, uio, ioflag, cred);
636 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
637 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
638 #endif
639 
640 	switch (vp->v_type) {
641 	case VREG:
642 		if (ioflag & IO_APPEND)
643 			uio->uio_offset = dep->de_FileSize;
644 		thisvp = vp;
645 		break;
646 	case VDIR:
647 		return EISDIR;
648 	default:
649 		panic("msdosfs_write(): bad file type");
650 	}
651 
652 	/*
653 	 * This is needed (unlike in ffs_write()) because we extend the
654 	 * file outside of the loop but we don't want to extend the file
655 	 * for writes of 0 bytes.
656 	 */
657 	if (uio->uio_resid == 0)
658 		return (0);
659 
660 	/*
661 	 * The caller is supposed to ensure that
662 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
663 	 */
664 	if ((uoff_t)uio->uio_offset + uio->uio_resid > MSDOSFS_FILESIZE_MAX)
665 		return (EFBIG);
666 
667 	/*
668 	 * If they've exceeded their filesize limit, tell them about it.
669 	 */
670 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
671 		return (EFBIG);
672 
673 	/*
674 	 * If the offset we are starting the write at is beyond the end of
675 	 * the file, then they've done a seek.  Unix filesystems allow
676 	 * files with holes in them, DOS doesn't so we must fill the hole
677 	 * with zeroed blocks.
678 	 */
679 	if (uio->uio_offset > dep->de_FileSize) {
680 		error = deextend(dep, uio->uio_offset, cred);
681 		if (error)
682 			return (error);
683 	}
684 
685 	/*
686 	 * Remember some values in case the write fails.
687 	 */
688 	resid = uio->uio_resid;
689 	osize = dep->de_FileSize;
690 
691 	/*
692 	 * If we write beyond the end of the file, extend it to its ultimate
693 	 * size ahead of the time to hopefully get a contiguous area.
694 	 */
695 	if (uio->uio_offset + resid > osize) {
696 		count = de_clcount(pmp, uio->uio_offset + resid) -
697 			de_clcount(pmp, osize);
698 		error = extendfile(dep, count, NULL, NULL, 0);
699 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
700 			goto errexit;
701 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
702 	} else
703 		lastcn = de_clcount(pmp, osize) - 1;
704 
705 	seqcount = ioflag >> IO_SEQSHIFT;
706 	do {
707 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
708 			error = ENOSPC;
709 			break;
710 		}
711 
712 		croffset = uio->uio_offset & pmp->pm_crbomask;
713 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
714 		if (uio->uio_offset + n > dep->de_FileSize) {
715 			dep->de_FileSize = uio->uio_offset + n;
716 			/* The object size needs to be set before buffer is allocated */
717 			vnode_pager_setsize(vp, dep->de_FileSize);
718 		}
719 
720 		bn = de_cluster(pmp, uio->uio_offset);
721 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
722 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
723 			> de_cluster(pmp, uio->uio_offset)
724 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
725 			/*
726 			 * If either the whole cluster gets written,
727 			 * or we write the cluster from its start beyond EOF,
728 			 * then no need to read data from disk.
729 			 */
730 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
731 			/*
732 			 * This call to vfs_bio_clrbuf() ensures that
733 			 * even if vn_io_fault_uiomove() below faults,
734 			 * garbage from the newly instantiated buffer
735 			 * is not exposed to the userspace via mmap().
736 			 */
737 			vfs_bio_clrbuf(bp);
738 			/*
739 			 * Do the bmap now, since pcbmap needs buffers
740 			 * for the FAT table. (see msdosfs_strategy)
741 			 */
742 			if (bp->b_blkno == bp->b_lblkno) {
743 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
744 				if (error)
745 					bp->b_blkno = -1;
746 				else
747 					bp->b_blkno = bn;
748 			}
749 			if (bp->b_blkno == -1) {
750 				brelse(bp);
751 				if (!error)
752 					error = EIO;		/* XXX */
753 				break;
754 			}
755 		} else {
756 			/*
757 			 * The block we need to write into exists, so read it in.
758 			 */
759 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
760 			if (error) {
761 				break;
762 			}
763 		}
764 
765 		/*
766 		 * Should these vnode_pager_* functions be done on dir
767 		 * files?
768 		 */
769 
770 		/*
771 		 * Copy the data from user space into the buf header.
772 		 */
773 		error = vn_io_fault_uiomove(bp->b_data + croffset, n, uio);
774 		if (error) {
775 			brelse(bp);
776 			break;
777 		}
778 
779 		/* Prepare for clustered writes in some else clauses. */
780 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
781 			bp->b_flags |= B_CLUSTEROK;
782 
783 		/*
784 		 * If IO_SYNC, then each buffer is written synchronously.
785 		 * Otherwise, if we have a severe page deficiency then
786 		 * write the buffer asynchronously.  Otherwise, if on a
787 		 * cluster boundary then write the buffer asynchronously,
788 		 * combining it with contiguous clusters if permitted and
789 		 * possible, since we don't expect more writes into this
790 		 * buffer soon.  Otherwise, do a delayed write because we
791 		 * expect more writes into this buffer soon.
792 		 */
793 		if (ioflag & IO_SYNC)
794 			(void)bwrite(bp);
795 		else if (vm_page_count_severe() || buf_dirty_count_severe())
796 			bawrite(bp);
797 		else if (n + croffset == pmp->pm_bpcluster) {
798 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
799 				cluster_write(vp, &dep->de_clusterw, bp,
800 				    dep->de_FileSize, seqcount, 0);
801 			else
802 				bawrite(bp);
803 		} else
804 			bdwrite(bp);
805 		dep->de_flag |= DE_UPDATE;
806 	} while (error == 0 && uio->uio_resid > 0);
807 
808 	/*
809 	 * If the write failed and they want us to, truncate the file back
810 	 * to the size it was before the write was attempted.
811 	 */
812 errexit:
813 	if (error) {
814 		if (ioflag & IO_UNIT) {
815 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
816 			uio->uio_offset -= resid - uio->uio_resid;
817 			uio->uio_resid = resid;
818 		} else {
819 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
820 			if (uio->uio_resid != resid)
821 				error = 0;
822 		}
823 	} else if (ioflag & IO_SYNC)
824 		error = deupdat(dep, 1);
825 	return (error);
826 }
827 
828 /*
829  * Flush the blocks of a file to disk.
830  */
831 static int
832 msdosfs_fsync(struct vop_fsync_args *ap)
833 {
834 	struct vnode *devvp;
835 	int allerror, error;
836 
837 	vop_stdfsync(ap);
838 
839 	/*
840 	* If the syncing request comes from fsync(2), sync the entire
841 	* FAT and any other metadata that happens to be on devvp.  We
842 	* need this mainly for the FAT.  We write the FAT sloppily, and
843 	* syncing it all now is the best we can easily do to get all
844 	* directory entries associated with the file (not just the file)
845 	* fully synced.  The other metadata includes critical metadata
846 	* for all directory entries, but only in the MNT_ASYNC case.  We
847 	* will soon sync all metadata in the file's directory entry.
848 	* Non-critical metadata for associated directory entries only
849 	* gets synced accidentally, as in most file systems.
850 	*/
851 	if (ap->a_waitfor != MNT_NOWAIT) {
852 		devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
853 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
854 		allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
855 		VOP_UNLOCK(devvp);
856 	} else
857 		allerror = 0;
858 
859 	error = deupdat(VTODE(ap->a_vp), ap->a_waitfor != MNT_NOWAIT);
860 	if (allerror == 0)
861 		allerror = error;
862 	return (allerror);
863 }
864 
865 static int
866 msdosfs_remove(struct vop_remove_args *ap)
867 {
868 	struct denode *dep = VTODE(ap->a_vp);
869 	struct denode *ddep = VTODE(ap->a_dvp);
870 	int error;
871 
872 	if (ap->a_vp->v_type == VDIR)
873 		error = EPERM;
874 	else
875 		error = removede(ddep, dep);
876 #ifdef MSDOSFS_DEBUG
877 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
878 #endif
879 	return (error);
880 }
881 
882 /*
883  * DOS filesystems don't know what links are.
884  */
885 static int
886 msdosfs_link(struct vop_link_args *ap)
887 {
888 	return (EOPNOTSUPP);
889 }
890 
891 /*
892  * Renames on files require moving the denode to a new hash queue since the
893  * denode's location is used to compute which hash queue to put the file
894  * in. Unless it is a rename in place.  For example "mv a b".
895  *
896  * What follows is the basic algorithm:
897  *
898  * if (file move) {
899  *	if (dest file exists) {
900  *		remove dest file
901  *	}
902  *	if (dest and src in same directory) {
903  *		rewrite name in existing directory slot
904  *	} else {
905  *		write new entry in dest directory
906  *		update offset and dirclust in denode
907  *		move denode to new hash chain
908  *		clear old directory entry
909  *	}
910  * } else {
911  *	directory move
912  *	if (dest directory exists) {
913  *		if (dest is not empty) {
914  *			return ENOTEMPTY
915  *		}
916  *		remove dest directory
917  *	}
918  *	if (dest and src in same directory) {
919  *		rewrite name in existing entry
920  *	} else {
921  *		be sure dest is not a child of src directory
922  *		write entry in dest directory
923  *		update "." and ".." in moved directory
924  *		clear old directory entry for moved directory
925  *	}
926  * }
927  *
928  * On entry:
929  *	source's parent directory is unlocked
930  *	source file or directory is unlocked
931  *	destination's parent directory is locked
932  *	destination file or directory is locked if it exists
933  *
934  * On exit:
935  *	all denodes should be released
936  */
937 static int
938 msdosfs_rename(struct vop_rename_args *ap)
939 {
940 	struct vnode *fdvp, *fvp, *tdvp, *tvp, *vp;
941 	struct componentname *fcnp, *tcnp;
942 	struct denode *fdip, *fip, *tdip, *tip, *nip;
943 	u_char toname[12], oldname[11];
944 	u_long to_diroffset;
945 	bool checkpath_locked, doingdirectory, newparent;
946 	int error;
947 	u_long cn, pcl, blkoff;
948 	daddr_t bn, wait_scn, scn;
949 	struct msdosfsmount *pmp;
950 	struct direntry *dotdotp;
951 	struct buf *bp;
952 
953 	tdvp = ap->a_tdvp;
954 	fvp = ap->a_fvp;
955 	fdvp = ap->a_fdvp;
956 	tvp = ap->a_tvp;
957 	tcnp = ap->a_tcnp;
958 	fcnp = ap->a_fcnp;
959 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
960 
961 #ifdef DIAGNOSTIC
962 	if ((tcnp->cn_flags & HASBUF) == 0 ||
963 	    (fcnp->cn_flags & HASBUF) == 0)
964 		panic("msdosfs_rename: no name");
965 #endif
966 	/*
967 	 * Check for cross-device rename.
968 	 */
969 	if (fvp->v_mount != tdvp->v_mount ||
970 	    (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
971 		error = EXDEV;
972 		goto abortit;
973 	}
974 
975 	/*
976 	 * If source and dest are the same, do nothing.
977 	 */
978 	if (tvp == fvp) {
979 		error = 0;
980 		goto abortit;
981 	}
982 
983 	/*
984 	 * When the target exists, both the directory
985 	 * and target vnodes are passed locked.
986 	 */
987 	VOP_UNLOCK(tdvp);
988 	if (tvp != NULL && tvp != tdvp)
989 		VOP_UNLOCK(tvp);
990 
991 	checkpath_locked = false;
992 
993 relock:
994 	doingdirectory = newparent = false;
995 
996 	error = vn_lock(fdvp, LK_EXCLUSIVE);
997 	if (error != 0)
998 		goto releout;
999 	if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1000 		VOP_UNLOCK(fdvp);
1001 		error = vn_lock(tdvp, LK_EXCLUSIVE);
1002 		if (error != 0)
1003 			goto releout;
1004 		VOP_UNLOCK(tdvp);
1005 		goto relock;
1006 	}
1007 
1008 	error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff);
1009 	if (error != 0) {
1010 		VOP_UNLOCK(fdvp);
1011 		VOP_UNLOCK(tdvp);
1012 		goto releout;
1013 	}
1014 	error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT, &nip);
1015 	if (error != 0) {
1016 		VOP_UNLOCK(fdvp);
1017 		VOP_UNLOCK(tdvp);
1018 		if (error != EBUSY)
1019 			goto releout;
1020 		error = deget(pmp, scn, blkoff, LK_EXCLUSIVE, &nip);
1021 		if (error != 0)
1022 			goto releout;
1023 		vp = fvp;
1024 		fvp = DETOV(nip);
1025 		VOP_UNLOCK(fvp);
1026 		vrele(vp);
1027 		goto relock;
1028 	}
1029 	vrele(fvp);
1030 	fvp = DETOV(nip);
1031 
1032 	error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
1033 	if (error != 0 && error != EJUSTRETURN) {
1034 		VOP_UNLOCK(fdvp);
1035 		VOP_UNLOCK(tdvp);
1036 		VOP_UNLOCK(fvp);
1037 		goto releout;
1038 	}
1039 	if (error == EJUSTRETURN && tvp != NULL) {
1040 		vrele(tvp);
1041 		tvp = NULL;
1042 	}
1043 	if (error == 0) {
1044 		nip = NULL;
1045 		error = deget(pmp, scn, blkoff, LK_EXCLUSIVE | LK_NOWAIT,
1046 		    &nip);
1047 		if (tvp != NULL) {
1048 			vrele(tvp);
1049 			tvp = NULL;
1050 		}
1051 		if (error != 0) {
1052 			VOP_UNLOCK(fdvp);
1053 			VOP_UNLOCK(tdvp);
1054 			VOP_UNLOCK(fvp);
1055 			if (error != EBUSY)
1056 				goto releout;
1057 			error = deget(pmp, scn, blkoff, LK_EXCLUSIVE,
1058 			    &nip);
1059 			if (error != 0)
1060 				goto releout;
1061 			vput(DETOV(nip));
1062 			goto relock;
1063 		}
1064 		tvp = DETOV(nip);
1065 	}
1066 
1067 	fdip = VTODE(fdvp);
1068 	fip = VTODE(fvp);
1069 	tdip = VTODE(tdvp);
1070 	tip = tvp != NULL ? VTODE(tvp) : NULL;
1071 
1072 	/*
1073 	 * Remember direntry place to use for destination
1074 	 */
1075 	to_diroffset = tdip->de_fndoffset;
1076 
1077 	/*
1078 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1079 	 * leads to a crippled directory tree.  It's pretty tough to do a
1080 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1081 	 * doesn't work if the ".." entry is missing.
1082 	 */
1083 	if ((fip->de_Attributes & ATTR_DIRECTORY) != 0) {
1084 		/*
1085 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1086 		 */
1087 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1088 		    fdip == fip ||
1089 		    (fcnp->cn_flags & ISDOTDOT) != 0 ||
1090 		    (tcnp->cn_flags & ISDOTDOT) != 0) {
1091 			error = EINVAL;
1092 			goto unlock;
1093 		}
1094 		doingdirectory = true;
1095 	}
1096 
1097 	/*
1098 	 * If ".." must be changed (ie the directory gets a new
1099 	 * parent) then the source directory must not be in the
1100 	 * directory hierarchy above the target, as this would
1101 	 * orphan everything below the source directory. Also
1102 	 * the user must have write permission in the source so
1103 	 * as to be able to change "..". We must repeat the call
1104 	 * to namei, as the parent directory is unlocked by the
1105 	 * call to doscheckpath().
1106 	 */
1107 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
1108 	if (fdip->de_StartCluster != tdip->de_StartCluster)
1109 		newparent = true;
1110 	if (doingdirectory && newparent) {
1111 		if (error != 0)	/* write access check above */
1112 			goto unlock;
1113 		lockmgr(&pmp->pm_checkpath_lock, LK_EXCLUSIVE, NULL);
1114 		checkpath_locked = true;
1115 		error = doscheckpath(fip, tdip, &wait_scn);
1116 		if (wait_scn != 0) {
1117 			lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
1118 			checkpath_locked = false;
1119 			VOP_UNLOCK(fdvp);
1120 			VOP_UNLOCK(tdvp);
1121 			VOP_UNLOCK(fvp);
1122 			if (tvp != NULL && tvp != tdvp)
1123 				VOP_UNLOCK(tvp);
1124 			error = deget(pmp, wait_scn, 0, LK_EXCLUSIVE,
1125 			    &nip);
1126 			if (error == 0) {
1127 				vput(DETOV(nip));
1128 				goto relock;
1129 			}
1130 		}
1131 		if (error != 0)
1132 			goto unlock;
1133 		if ((tcnp->cn_flags & SAVESTART) == 0)
1134 			panic("msdosfs_rename: lost to startdir");
1135 	}
1136 
1137 	if (tip != NULL) {
1138 		/*
1139 		 * Target must be empty if a directory and have no links
1140 		 * to it. Also, ensure source and target are compatible
1141 		 * (both directories, or both not directories).
1142 		 */
1143 		if ((tip->de_Attributes & ATTR_DIRECTORY) != 0) {
1144 			if (!dosdirempty(tip)) {
1145 				error = ENOTEMPTY;
1146 				goto unlock;
1147 			}
1148 			if (!doingdirectory) {
1149 				error = ENOTDIR;
1150 				goto unlock;
1151 			}
1152 			cache_purge(tdvp);
1153 		} else if (doingdirectory) {
1154 			error = EISDIR;
1155 			goto unlock;
1156 		}
1157 		error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
1158 		MPASS(error == 0);
1159 		error = removede(tdip, tip);
1160 		if (error != 0)
1161 			goto unlock;
1162 		vput(tvp);
1163 		tvp = NULL;
1164 		tip = NULL;
1165 	}
1166 
1167 	/*
1168 	 * Convert the filename in tcnp into a dos filename. We copy this
1169 	 * into the denode and directory entry for the destination
1170 	 * file/directory.
1171 	 */
1172 	error = uniqdosname(tdip, tcnp, toname);
1173 	if (error != 0)
1174 		goto unlock;
1175 
1176 	/*
1177 	 * First write a new entry in the destination
1178 	 * directory and mark the entry in the source directory
1179 	 * as deleted.  Then move the denode to the correct hash
1180 	 * chain for its new location in the filesystem.  And, if
1181 	 * we moved a directory, then update its .. entry to point
1182 	 * to the new parent directory.
1183 	 */
1184 	memcpy(oldname, fip->de_Name, 11);
1185 	memcpy(fip->de_Name, toname, 11);	/* update denode */
1186 	error = msdosfs_lookup_ino(tdvp, NULL, tcnp, &scn, &blkoff);
1187 	MPASS(error == EJUSTRETURN);
1188 	error = createde(fip, tdip, NULL, tcnp);
1189 	if (error != 0) {
1190 		memcpy(fip->de_Name, oldname, 11);
1191 		goto unlock;
1192 	}
1193 
1194 	/*
1195 	 * If fip is for a directory, then its name should always
1196 	 * be "." since it is for the directory entry in the
1197 	 * directory itself (msdosfs_lookup() always translates
1198 	 * to the "." entry so as to get a unique denode, except
1199 	 * for the root directory there are different
1200 	 * complications).  However, we just corrupted its name
1201 	 * to pass the correct name to createde().  Undo this.
1202 	 */
1203 	if ((fip->de_Attributes & ATTR_DIRECTORY) != 0)
1204 		memcpy(fip->de_Name, oldname, 11);
1205 	fip->de_refcnt++;
1206 	error = msdosfs_lookup_ino(fdvp, NULL, fcnp, &scn, &blkoff);
1207 	MPASS(error == 0);
1208 	error = removede(fdip, fip);
1209 	if (error != 0) {
1210 		/* XXX should downgrade to ro here, fs is corrupt */
1211 		goto unlock;
1212 	}
1213 	if (!doingdirectory) {
1214 		error = pcbmap(tdip, de_cluster(pmp, to_diroffset), 0,
1215 		    &fip->de_dirclust, 0);
1216 		if (error != 0) {
1217 			/*
1218 			 * XXX should downgrade to ro here,
1219 			 * fs is corrupt
1220 			 */
1221 			goto unlock;
1222 		}
1223 		if (fip->de_dirclust == MSDOSFSROOT)
1224 			fip->de_diroffset = to_diroffset;
1225 		else
1226 			fip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1227 	}
1228 	reinsert(fip);
1229 
1230 	/*
1231 	 * If we moved a directory to a new parent directory, then we must
1232 	 * fixup the ".." entry in the moved directory.
1233 	 */
1234 	if (doingdirectory && newparent) {
1235 		cn = fip->de_StartCluster;
1236 		if (cn == MSDOSFSROOT) {
1237 			/* this should never happen */
1238 			panic("msdosfs_rename(): updating .. in root directory?");
1239 		} else
1240 			bn = cntobn(pmp, cn);
1241 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1242 		    NOCRED, &bp);
1243 		if (error != 0) {
1244 			/* XXX should downgrade to ro here, fs is corrupt */
1245 			goto unlock;
1246 		}
1247 		dotdotp = (struct direntry *)bp->b_data + 1;
1248 		pcl = tdip->de_StartCluster;
1249 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1250 			pcl = MSDOSFSROOT;
1251 		putushort(dotdotp->deStartCluster, pcl);
1252 		if (FAT32(pmp))
1253 			putushort(dotdotp->deHighClust, pcl >> 16);
1254 		if (DOINGASYNC(fvp))
1255 			bdwrite(bp);
1256 		else if ((error = bwrite(bp)) != 0) {
1257 			/* XXX should downgrade to ro here, fs is corrupt */
1258 			goto unlock;
1259 		}
1260 	}
1261 
1262 	/*
1263 	 * The msdosfs lookup is case insensitive. Several aliases may
1264 	 * be inserted for a single directory entry. As a consequnce,
1265 	 * name cache purge done by lookup for fvp when DELETE op for
1266 	 * namei is specified, might be not enough to expunge all
1267 	 * namecache entries that were installed for this direntry.
1268 	 */
1269 	cache_purge(fvp);
1270 
1271 unlock:
1272 	if (checkpath_locked)
1273 		lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
1274 	vput(fdvp);
1275 	vput(fvp);
1276 	if (tvp != NULL) {
1277 		if (tvp != tdvp)
1278 			vput(tvp);
1279 		else
1280 			vrele(tvp);
1281 	}
1282 	vput(tdvp);
1283 	return (error);
1284 releout:
1285 	MPASS(!checkpath_locked);
1286 	vrele(tdvp);
1287 	if (tvp != NULL)
1288 		vrele(tvp);
1289 	vrele(fdvp);
1290 	vrele(fvp);
1291 	return (error);
1292 abortit:
1293 	if (tdvp == tvp)
1294 		vrele(tdvp);
1295 	else
1296 		vput(tdvp);
1297 	if (tvp != NULL)
1298 		vput(tvp);
1299 	vrele(fdvp);
1300 	vrele(fvp);
1301 	return (error);
1302 }
1303 
1304 static struct {
1305 	struct direntry dot;
1306 	struct direntry dotdot;
1307 } dosdirtemplate = {
1308 	{	".          ",				/* the . entry */
1309 		ATTR_DIRECTORY,				/* file attribute */
1310 		0,					/* reserved */
1311 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1312 		{ 0, 0 },				/* access date */
1313 		{ 0, 0 },				/* high bits of start cluster */
1314 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1315 		{ 0, 0 },				/* startcluster */
1316 		{ 0, 0, 0, 0 }				/* filesize */
1317 	},
1318 	{	"..         ",				/* the .. entry */
1319 		ATTR_DIRECTORY,				/* file attribute */
1320 		0,					/* reserved */
1321 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1322 		{ 0, 0 },				/* access date */
1323 		{ 0, 0 },				/* high bits of start cluster */
1324 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1325 		{ 0, 0 },				/* startcluster */
1326 		{ 0, 0, 0, 0 }				/* filesize */
1327 	}
1328 };
1329 
1330 static int
1331 msdosfs_mkdir(struct vop_mkdir_args *ap)
1332 {
1333 	struct componentname *cnp = ap->a_cnp;
1334 	struct denode *dep;
1335 	struct denode *pdep = VTODE(ap->a_dvp);
1336 	struct direntry *denp;
1337 	struct msdosfsmount *pmp = pdep->de_pmp;
1338 	struct buf *bp;
1339 	u_long newcluster, pcl;
1340 	int bn;
1341 	int error;
1342 	struct denode ndirent;
1343 	struct timespec ts;
1344 
1345 	/*
1346 	 * If this is the root directory and there is no space left we
1347 	 * can't do anything.  This is because the root directory can not
1348 	 * change size.
1349 	 */
1350 	if (pdep->de_StartCluster == MSDOSFSROOT
1351 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1352 		error = ENOSPC;
1353 		goto bad2;
1354 	}
1355 
1356 	/*
1357 	 * Allocate a cluster to hold the about to be created directory.
1358 	 */
1359 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1360 	if (error)
1361 		goto bad2;
1362 
1363 	memset(&ndirent, 0, sizeof(ndirent));
1364 	ndirent.de_pmp = pmp;
1365 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1366 	vfs_timestamp(&ts);
1367 	DETIMES(&ndirent, &ts, &ts, &ts);
1368 
1369 	/*
1370 	 * Now fill the cluster with the "." and ".." entries. And write
1371 	 * the cluster to disk.  This way it is there for the parent
1372 	 * directory to be pointing at if there were a crash.
1373 	 */
1374 	bn = cntobn(pmp, newcluster);
1375 	/* always succeeds */
1376 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1377 	memset(bp->b_data, 0, pmp->pm_bpcluster);
1378 	memcpy(bp->b_data, &dosdirtemplate, sizeof dosdirtemplate);
1379 	denp = (struct direntry *)bp->b_data;
1380 	putushort(denp[0].deStartCluster, newcluster);
1381 	putushort(denp[0].deCDate, ndirent.de_CDate);
1382 	putushort(denp[0].deCTime, ndirent.de_CTime);
1383 	denp[0].deCHundredth = ndirent.de_CHun;
1384 	putushort(denp[0].deADate, ndirent.de_ADate);
1385 	putushort(denp[0].deMDate, ndirent.de_MDate);
1386 	putushort(denp[0].deMTime, ndirent.de_MTime);
1387 	pcl = pdep->de_StartCluster;
1388 	/*
1389 	 * Although the root directory has a non-magic starting cluster
1390 	 * number for FAT32, chkdsk and fsck_msdosfs still require
1391 	 * references to it in dotdot entries to be magic.
1392 	 */
1393 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1394 		pcl = MSDOSFSROOT;
1395 	putushort(denp[1].deStartCluster, pcl);
1396 	putushort(denp[1].deCDate, ndirent.de_CDate);
1397 	putushort(denp[1].deCTime, ndirent.de_CTime);
1398 	denp[1].deCHundredth = ndirent.de_CHun;
1399 	putushort(denp[1].deADate, ndirent.de_ADate);
1400 	putushort(denp[1].deMDate, ndirent.de_MDate);
1401 	putushort(denp[1].deMTime, ndirent.de_MTime);
1402 	if (FAT32(pmp)) {
1403 		putushort(denp[0].deHighClust, newcluster >> 16);
1404 		putushort(denp[1].deHighClust, pcl >> 16);
1405 	}
1406 
1407 	if (DOINGASYNC(ap->a_dvp))
1408 		bdwrite(bp);
1409 	else if ((error = bwrite(bp)) != 0)
1410 		goto bad;
1411 
1412 	/*
1413 	 * Now build up a directory entry pointing to the newly allocated
1414 	 * cluster.  This will be written to an empty slot in the parent
1415 	 * directory.
1416 	 */
1417 #ifdef DIAGNOSTIC
1418 	if ((cnp->cn_flags & HASBUF) == 0)
1419 		panic("msdosfs_mkdir: no name");
1420 #endif
1421 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1422 	if (error)
1423 		goto bad;
1424 
1425 	ndirent.de_Attributes = ATTR_DIRECTORY;
1426 	ndirent.de_LowerCase = 0;
1427 	ndirent.de_StartCluster = newcluster;
1428 	ndirent.de_FileSize = 0;
1429 	error = createde(&ndirent, pdep, &dep, cnp);
1430 	if (error)
1431 		goto bad;
1432 	*ap->a_vpp = DETOV(dep);
1433 	return (0);
1434 
1435 bad:
1436 	clusterfree(pmp, newcluster, NULL);
1437 bad2:
1438 	return (error);
1439 }
1440 
1441 static int
1442 msdosfs_rmdir(struct vop_rmdir_args *ap)
1443 {
1444 	struct vnode *vp = ap->a_vp;
1445 	struct vnode *dvp = ap->a_dvp;
1446 	struct componentname *cnp = ap->a_cnp;
1447 	struct denode *ip, *dp;
1448 	int error;
1449 
1450 	ip = VTODE(vp);
1451 	dp = VTODE(dvp);
1452 
1453 	/*
1454 	 * Verify the directory is empty (and valid).
1455 	 * (Rmdir ".." won't be valid since
1456 	 *  ".." will contain a reference to
1457 	 *  the current directory and thus be
1458 	 *  non-empty.)
1459 	 */
1460 	error = 0;
1461 	if (!dosdirempty(ip)) {
1462 		error = ENOTEMPTY;
1463 		goto out;
1464 	}
1465 	/*
1466 	 * Delete the entry from the directory.  For dos filesystems this
1467 	 * gets rid of the directory entry on disk, the in memory copy
1468 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1469 	 * being found by deget().  When the vput() on dep is done we give
1470 	 * up access and eventually msdosfs_reclaim() will be called which
1471 	 * will remove it from the denode cache.
1472 	 */
1473 	error = removede(dp, ip);
1474 	if (error)
1475 		goto out;
1476 	/*
1477 	 * This is where we decrement the link count in the parent
1478 	 * directory.  Since dos filesystems don't do this we just purge
1479 	 * the name cache.
1480 	 */
1481 	cache_purge(dvp);
1482 	/*
1483 	 * Truncate the directory that is being deleted.
1484 	 */
1485 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1486 	cache_purge(vp);
1487 
1488 out:
1489 	return (error);
1490 }
1491 
1492 /*
1493  * DOS filesystems don't know what symlinks are.
1494  */
1495 static int
1496 msdosfs_symlink(struct vop_symlink_args *ap)
1497 {
1498 	return (EOPNOTSUPP);
1499 }
1500 
1501 static int
1502 msdosfs_readdir(struct vop_readdir_args *ap)
1503 {
1504 	struct mbnambuf nb;
1505 	int error = 0;
1506 	int diff;
1507 	long n;
1508 	int blsize;
1509 	long on;
1510 	u_long cn;
1511 	u_long dirsperblk;
1512 	long bias = 0;
1513 	daddr_t bn, lbn;
1514 	struct buf *bp;
1515 	struct denode *dep = VTODE(ap->a_vp);
1516 	struct msdosfsmount *pmp = dep->de_pmp;
1517 	struct direntry *dentp;
1518 	struct dirent dirbuf;
1519 	struct uio *uio = ap->a_uio;
1520 	u_long *cookies = NULL;
1521 	int ncookies = 0;
1522 	off_t offset, off;
1523 	int chksum = -1;
1524 
1525 #ifdef MSDOSFS_DEBUG
1526 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1527 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1528 #endif
1529 
1530 	/*
1531 	 * msdosfs_readdir() won't operate properly on regular files since
1532 	 * it does i/o only with the filesystem vnode, and hence can
1533 	 * retrieve the wrong block from the buffer cache for a plain file.
1534 	 * So, fail attempts to readdir() on a plain file.
1535 	 */
1536 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1537 		return (ENOTDIR);
1538 
1539 	/*
1540 	 * To be safe, initialize dirbuf
1541 	 */
1542 	memset(dirbuf.d_name, 0, sizeof(dirbuf.d_name));
1543 
1544 	/*
1545 	 * If the user buffer is smaller than the size of one dos directory
1546 	 * entry or the file offset is not a multiple of the size of a
1547 	 * directory entry, then we fail the read.
1548 	 */
1549 	off = offset = uio->uio_offset;
1550 	if (uio->uio_resid < sizeof(struct direntry) ||
1551 	    (offset & (sizeof(struct direntry) - 1)))
1552 		return (EINVAL);
1553 
1554 	if (ap->a_ncookies) {
1555 		ncookies = uio->uio_resid / 16;
1556 		cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1557 		       M_WAITOK);
1558 		*ap->a_cookies = cookies;
1559 		*ap->a_ncookies = ncookies;
1560 	}
1561 
1562 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1563 
1564 	/*
1565 	 * If they are reading from the root directory then, we simulate
1566 	 * the . and .. entries since these don't exist in the root
1567 	 * directory.  We also set the offset bias to make up for having to
1568 	 * simulate these entries. By this I mean that at file offset 64 we
1569 	 * read the first entry in the root directory that lives on disk.
1570 	 */
1571 	if (dep->de_StartCluster == MSDOSFSROOT
1572 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1573 #if 0
1574 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1575 		    offset);
1576 #endif
1577 		bias = 2 * sizeof(struct direntry);
1578 		if (offset < bias) {
1579 			for (n = (int)offset / sizeof(struct direntry);
1580 			     n < 2; n++) {
1581 				dirbuf.d_fileno = FAT32(pmp) ?
1582 				    (uint64_t)cntobn(pmp, pmp->pm_rootdirblk) *
1583 				    dirsperblk : 1;
1584 				dirbuf.d_type = DT_DIR;
1585 				switch (n) {
1586 				case 0:
1587 					dirbuf.d_namlen = 1;
1588 					dirbuf.d_name[0] = '.';
1589 					break;
1590 				case 1:
1591 					dirbuf.d_namlen = 2;
1592 					dirbuf.d_name[0] = '.';
1593 					dirbuf.d_name[1] = '.';
1594 					break;
1595 				}
1596 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1597 				/* NOTE: d_off is the offset of the *next* entry. */
1598 				dirbuf.d_off = offset + sizeof(struct direntry);
1599 				dirent_terminate(&dirbuf);
1600 				if (uio->uio_resid < dirbuf.d_reclen)
1601 					goto out;
1602 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1603 				if (error)
1604 					goto out;
1605 				offset += sizeof(struct direntry);
1606 				off = offset;
1607 				if (cookies) {
1608 					*cookies++ = offset;
1609 					if (--ncookies <= 0)
1610 						goto out;
1611 				}
1612 			}
1613 		}
1614 	}
1615 
1616 	mbnambuf_init(&nb);
1617 	off = offset;
1618 	while (uio->uio_resid > 0) {
1619 		lbn = de_cluster(pmp, offset - bias);
1620 		on = (offset - bias) & pmp->pm_crbomask;
1621 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1622 		diff = dep->de_FileSize - (offset - bias);
1623 		if (diff <= 0)
1624 			break;
1625 		n = min(n, diff);
1626 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1627 		if (error)
1628 			break;
1629 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1630 		if (error) {
1631 			return (error);
1632 		}
1633 		n = min(n, blsize - bp->b_resid);
1634 		if (n == 0) {
1635 			brelse(bp);
1636 			return (EIO);
1637 		}
1638 
1639 		/*
1640 		 * Convert from dos directory entries to fs-independent
1641 		 * directory entries.
1642 		 */
1643 		for (dentp = (struct direntry *)(bp->b_data + on);
1644 		     (char *)dentp < bp->b_data + on + n;
1645 		     dentp++, offset += sizeof(struct direntry)) {
1646 #if 0
1647 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1648 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1649 #endif
1650 			/*
1651 			 * If this is an unused entry, we can stop.
1652 			 */
1653 			if (dentp->deName[0] == SLOT_EMPTY) {
1654 				brelse(bp);
1655 				goto out;
1656 			}
1657 			/*
1658 			 * Skip deleted entries.
1659 			 */
1660 			if (dentp->deName[0] == SLOT_DELETED) {
1661 				chksum = -1;
1662 				mbnambuf_init(&nb);
1663 				continue;
1664 			}
1665 
1666 			/*
1667 			 * Handle Win95 long directory entries
1668 			 */
1669 			if (dentp->deAttributes == ATTR_WIN95) {
1670 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1671 					continue;
1672 				chksum = win2unixfn(&nb,
1673 				    (struct winentry *)dentp, chksum, pmp);
1674 				continue;
1675 			}
1676 
1677 			/*
1678 			 * Skip volume labels
1679 			 */
1680 			if (dentp->deAttributes & ATTR_VOLUME) {
1681 				chksum = -1;
1682 				mbnambuf_init(&nb);
1683 				continue;
1684 			}
1685 			/*
1686 			 * This computation of d_fileno must match
1687 			 * the computation of va_fileid in
1688 			 * msdosfs_getattr.
1689 			 */
1690 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1691 				cn = getushort(dentp->deStartCluster);
1692 				if (FAT32(pmp)) {
1693 					cn |= getushort(dentp->deHighClust) <<
1694 					    16;
1695 					if (cn == MSDOSFSROOT)
1696 						cn = pmp->pm_rootdirblk;
1697 				}
1698 				if (cn == MSDOSFSROOT && !FAT32(pmp))
1699 					dirbuf.d_fileno = 1;
1700 				else
1701 					dirbuf.d_fileno = cntobn(pmp, cn) *
1702 					    dirsperblk;
1703 				dirbuf.d_type = DT_DIR;
1704 			} else {
1705 				dirbuf.d_fileno = (uoff_t)offset /
1706 				    sizeof(struct direntry);
1707 				dirbuf.d_type = DT_REG;
1708 			}
1709 
1710 			if (chksum != winChksum(dentp->deName)) {
1711 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1712 				    (u_char *)dirbuf.d_name,
1713 				    dentp->deLowerCase |
1714 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1715 					(LCASE_BASE | LCASE_EXT) : 0),
1716 				    pmp);
1717 				mbnambuf_init(&nb);
1718 			} else
1719 				mbnambuf_flush(&nb, &dirbuf);
1720 			chksum = -1;
1721 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1722 			/* NOTE: d_off is the offset of the *next* entry. */
1723 			dirbuf.d_off = offset + sizeof(struct direntry);
1724 			dirent_terminate(&dirbuf);
1725 			if (uio->uio_resid < dirbuf.d_reclen) {
1726 				brelse(bp);
1727 				goto out;
1728 			}
1729 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1730 			if (error) {
1731 				brelse(bp);
1732 				goto out;
1733 			}
1734 			if (cookies) {
1735 				*cookies++ = offset + sizeof(struct direntry);
1736 				if (--ncookies <= 0) {
1737 					brelse(bp);
1738 					goto out;
1739 				}
1740 			}
1741 			off = offset + sizeof(struct direntry);
1742 		}
1743 		brelse(bp);
1744 	}
1745 out:
1746 	/* Subtract unused cookies */
1747 	if (ap->a_ncookies)
1748 		*ap->a_ncookies -= ncookies;
1749 
1750 	uio->uio_offset = off;
1751 
1752 	/*
1753 	 * Set the eofflag (NFS uses it)
1754 	 */
1755 	if (ap->a_eofflag) {
1756 		if (dep->de_FileSize - (offset - bias) <= 0)
1757 			*ap->a_eofflag = 1;
1758 		else
1759 			*ap->a_eofflag = 0;
1760 	}
1761 	return (error);
1762 }
1763 
1764 /*-
1765  * a_vp   - pointer to the file's vnode
1766  * a_bn   - logical block number within the file (cluster number for us)
1767  * a_bop  - where to return the bufobj of the special file containing the fs
1768  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1769  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1770  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1771  *          blocks whose physical blocks (together with a_bn's physical block)
1772  *          are contiguous.
1773  * a_runb - where to return the "run before" a_bn.
1774  */
1775 static int
1776 msdosfs_bmap(struct vop_bmap_args *ap)
1777 {
1778 	struct fatcache savefc;
1779 	struct denode *dep;
1780 	struct mount *mp;
1781 	struct msdosfsmount *pmp;
1782 	struct vnode *vp;
1783 	daddr_t runbn;
1784 	u_long cn;
1785 	int bnpercn, error, maxio, maxrun, run;
1786 
1787 	vp = ap->a_vp;
1788 	dep = VTODE(vp);
1789 	pmp = dep->de_pmp;
1790 	if (ap->a_bop != NULL)
1791 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1792 	if (ap->a_bnp == NULL)
1793 		return (0);
1794 	if (ap->a_runp != NULL)
1795 		*ap->a_runp = 0;
1796 	if (ap->a_runb != NULL)
1797 		*ap->a_runb = 0;
1798 	cn = ap->a_bn;
1799 	if (cn != ap->a_bn)
1800 		return (EFBIG);
1801 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1802 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1803 		return (error);
1804 
1805 	/*
1806 	 * Prepare to back out updates of the fatchain cache after the one
1807 	 * for the first block done by pcbmap() above.  Without the backout,
1808 	 * then whenever the caller doesn't do i/o to all of the blocks that
1809 	 * we find, the single useful cache entry would be too far in advance
1810 	 * of the actual i/o to work for the next sequential i/o.  Then the
1811 	 * FAT would be searched from the beginning.  With the backout, the
1812 	 * FAT is searched starting at most a few blocks early.  This wastes
1813 	 * much less time.  Time is also wasted finding more blocks than the
1814 	 * caller will do i/o to.  This is necessary because the runlength
1815 	 * parameters are output-only.
1816 	 */
1817 	savefc = dep->de_fc[FC_LASTMAP];
1818 
1819 	mp = vp->v_mount;
1820 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1821 	bnpercn = de_cn2bn(pmp, 1);
1822 	if (ap->a_runp != NULL) {
1823 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1824 		for (run = 1; run <= maxrun; run++) {
1825 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1826 			    runbn != *ap->a_bnp + run * bnpercn)
1827 				break;
1828 		}
1829 		*ap->a_runp = run - 1;
1830 	}
1831 	if (ap->a_runb != NULL) {
1832 		maxrun = ulmin(maxio - 1, cn);
1833 		for (run = 1; run < maxrun; run++) {
1834 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1835 			    runbn != *ap->a_bnp - run * bnpercn)
1836 				break;
1837 		}
1838 		*ap->a_runb = run - 1;
1839 	}
1840 	dep->de_fc[FC_LASTMAP] = savefc;
1841 	return (0);
1842 }
1843 
1844 SYSCTL_NODE(_vfs, OID_AUTO, msdosfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1845     "msdos filesystem");
1846 static int use_buf_pager = 1;
1847 SYSCTL_INT(_vfs_msdosfs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN,
1848     &use_buf_pager, 0,
1849     "Use buffer pager instead of bmap");
1850 
1851 static daddr_t
1852 msdosfs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off)
1853 {
1854 
1855 	return (de_cluster(VTODE(vp)->de_pmp, off));
1856 }
1857 
1858 static int
1859 msdosfs_gbp_getblksz(struct vnode *vp, daddr_t lbn, long *sz)
1860 {
1861 
1862 	*sz = VTODE(vp)->de_pmp->pm_bpcluster;
1863 	return (0);
1864 }
1865 
1866 static int
1867 msdosfs_getpages(struct vop_getpages_args *ap)
1868 {
1869 
1870 	if (use_buf_pager)
1871 		return (vfs_bio_getpages(ap->a_vp, ap->a_m, ap->a_count,
1872 		    ap->a_rbehind, ap->a_rahead, msdosfs_gbp_getblkno,
1873 		    msdosfs_gbp_getblksz));
1874 	return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count,
1875 	    ap->a_rbehind, ap->a_rahead, NULL, NULL));
1876 }
1877 
1878 static int
1879 msdosfs_strategy(struct vop_strategy_args *ap)
1880 {
1881 	struct buf *bp = ap->a_bp;
1882 	struct denode *dep = VTODE(ap->a_vp);
1883 	struct bufobj *bo;
1884 	int error = 0;
1885 	daddr_t blkno;
1886 
1887 	/*
1888 	 * If we don't already know the filesystem relative block number
1889 	 * then get it using pcbmap().  If pcbmap() returns the block
1890 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1891 	 * don't allow files with holes, so we shouldn't ever see this.
1892 	 */
1893 	if (bp->b_blkno == bp->b_lblkno) {
1894 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1895 		bp->b_blkno = blkno;
1896 		if (error) {
1897 			bp->b_error = error;
1898 			bp->b_ioflags |= BIO_ERROR;
1899 			bufdone(bp);
1900 			return (0);
1901 		}
1902 		if ((long)bp->b_blkno == -1)
1903 			vfs_bio_clrbuf(bp);
1904 	}
1905 	if (bp->b_blkno == -1) {
1906 		bufdone(bp);
1907 		return (0);
1908 	}
1909 	/*
1910 	 * Read/write the block from/to the disk that contains the desired
1911 	 * file block.
1912 	 */
1913 	bp->b_iooffset = dbtob(bp->b_blkno);
1914 	bo = dep->de_pmp->pm_bo;
1915 	BO_STRATEGY(bo, bp);
1916 	return (0);
1917 }
1918 
1919 static int
1920 msdosfs_print(struct vop_print_args *ap)
1921 {
1922 	struct denode *dep = VTODE(ap->a_vp);
1923 
1924 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1925 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1926 	printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1927 	return (0);
1928 }
1929 
1930 static int
1931 msdosfs_pathconf(struct vop_pathconf_args *ap)
1932 {
1933 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1934 
1935 	switch (ap->a_name) {
1936 	case _PC_FILESIZEBITS:
1937 		*ap->a_retval = 32;
1938 		return (0);
1939 	case _PC_LINK_MAX:
1940 		*ap->a_retval = 1;
1941 		return (0);
1942 	case _PC_NAME_MAX:
1943 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1944 		return (0);
1945 	case _PC_CHOWN_RESTRICTED:
1946 		*ap->a_retval = 1;
1947 		return (0);
1948 	case _PC_NO_TRUNC:
1949 		*ap->a_retval = 0;
1950 		return (0);
1951 	default:
1952 		return (vop_stdpathconf(ap));
1953 	}
1954 	/* NOTREACHED */
1955 }
1956 
1957 static int
1958 msdosfs_vptofh(struct vop_vptofh_args *ap)
1959 {
1960 	struct denode *dep;
1961 	struct defid *defhp;
1962 
1963 	dep = VTODE(ap->a_vp);
1964 	defhp = (struct defid *)ap->a_fhp;
1965 	defhp->defid_len = sizeof(struct defid);
1966 	defhp->defid_dirclust = dep->de_dirclust;
1967 	defhp->defid_dirofs = dep->de_diroffset;
1968 	/* defhp->defid_gen = dep->de_gen; */
1969 	return (0);
1970 }
1971 
1972 /* Global vfs data structures for msdosfs */
1973 struct vop_vector msdosfs_vnodeops = {
1974 	.vop_default =		&default_vnodeops,
1975 
1976 	.vop_access =		msdosfs_access,
1977 	.vop_bmap =		msdosfs_bmap,
1978 	.vop_getpages =		msdosfs_getpages,
1979 	.vop_cachedlookup =	msdosfs_lookup,
1980 	.vop_open =		msdosfs_open,
1981 	.vop_close =		msdosfs_close,
1982 	.vop_create =		msdosfs_create,
1983 	.vop_fsync =		msdosfs_fsync,
1984 	.vop_fdatasync =	vop_stdfdatasync_buf,
1985 	.vop_getattr =		msdosfs_getattr,
1986 	.vop_inactive =		msdosfs_inactive,
1987 	.vop_link =		msdosfs_link,
1988 	.vop_lookup =		vfs_cache_lookup,
1989 	.vop_mkdir =		msdosfs_mkdir,
1990 	.vop_mknod =		msdosfs_mknod,
1991 	.vop_pathconf =		msdosfs_pathconf,
1992 	.vop_print =		msdosfs_print,
1993 	.vop_read =		msdosfs_read,
1994 	.vop_readdir =		msdosfs_readdir,
1995 	.vop_reclaim =		msdosfs_reclaim,
1996 	.vop_remove =		msdosfs_remove,
1997 	.vop_rename =		msdosfs_rename,
1998 	.vop_rmdir =		msdosfs_rmdir,
1999 	.vop_setattr =		msdosfs_setattr,
2000 	.vop_strategy =		msdosfs_strategy,
2001 	.vop_symlink =		msdosfs_symlink,
2002 	.vop_write =		msdosfs_write,
2003 	.vop_vptofh =		msdosfs_vptofh,
2004 };
2005 VFS_VOP_VECTOR_REGISTER(msdosfs_vnodeops);
2006