xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision 5bd73b51076b5cb5a2c9810f76c1d7ed20c4460e)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $	*/
3 
4 /*-
5  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
6  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
7  * All rights reserved.
8  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by TooLs GmbH.
21  * 4. The name of TooLs GmbH may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 /*-
36  * Written by Paul Popelka (paulp@uts.amdahl.com)
37  *
38  * You can do anything you want with this software, just don't say you wrote
39  * it, and don't remove this notice.
40  *
41  * This software is provided "as is".
42  *
43  * The author supplies this software to be publicly redistributed on the
44  * understanding that the author is not responsible for the correct
45  * functioning of this software in any circumstances and is not liable for
46  * any damages caused by this software.
47  *
48  * October 1992
49  */
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/clock.h>
56 #include <sys/dirent.h>
57 #include <sys/lock.h>
58 #include <sys/lockf.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/priv.h>
64 #include <sys/stat.h>
65 #include <sys/unistd.h>
66 #include <sys/vnode.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 
71 #include <fs/msdosfs/bpb.h>
72 #include <fs/msdosfs/direntry.h>
73 #include <fs/msdosfs/denode.h>
74 #include <fs/msdosfs/fat.h>
75 #include <fs/msdosfs/msdosfsmount.h>
76 
77 #define	DOS_FILESIZE_MAX	0xffffffff
78 
79 /*
80  * Prototypes for MSDOSFS vnode operations
81  */
82 static vop_create_t	msdosfs_create;
83 static vop_mknod_t	msdosfs_mknod;
84 static vop_open_t	msdosfs_open;
85 static vop_close_t	msdosfs_close;
86 static vop_access_t	msdosfs_access;
87 static vop_getattr_t	msdosfs_getattr;
88 static vop_setattr_t	msdosfs_setattr;
89 static vop_read_t	msdosfs_read;
90 static vop_write_t	msdosfs_write;
91 static vop_fsync_t	msdosfs_fsync;
92 static vop_remove_t	msdosfs_remove;
93 static vop_link_t	msdosfs_link;
94 static vop_rename_t	msdosfs_rename;
95 static vop_mkdir_t	msdosfs_mkdir;
96 static vop_rmdir_t	msdosfs_rmdir;
97 static vop_symlink_t	msdosfs_symlink;
98 static vop_readdir_t	msdosfs_readdir;
99 static vop_bmap_t	msdosfs_bmap;
100 static vop_strategy_t	msdosfs_strategy;
101 static vop_print_t	msdosfs_print;
102 static vop_pathconf_t	msdosfs_pathconf;
103 static vop_vptofh_t	msdosfs_vptofh;
104 
105 /*
106  * Some general notes:
107  *
108  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
109  * read/written using the vnode for the filesystem. Blocks that represent
110  * the contents of a file are read/written using the vnode for the file
111  * (including directories when they are read/written as files). This
112  * presents problems for the dos filesystem because data that should be in
113  * an inode (if dos had them) resides in the directory itself.  Since we
114  * must update directory entries without the benefit of having the vnode
115  * for the directory we must use the vnode for the filesystem.  This means
116  * that when a directory is actually read/written (via read, write, or
117  * readdir, or seek) we must use the vnode for the filesystem instead of
118  * the vnode for the directory as would happen in ufs. This is to insure we
119  * retreive the correct block from the buffer cache since the hash value is
120  * based upon the vnode address and the desired block number.
121  */
122 
123 /*
124  * Create a regular file. On entry the directory to contain the file being
125  * created is locked.  We must release before we return. We must also free
126  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
127  * only if the SAVESTART bit in cn_flags is clear on success.
128  */
129 static int
130 msdosfs_create(ap)
131 	struct vop_create_args /* {
132 		struct vnode *a_dvp;
133 		struct vnode **a_vpp;
134 		struct componentname *a_cnp;
135 		struct vattr *a_vap;
136 	} */ *ap;
137 {
138 	struct componentname *cnp = ap->a_cnp;
139 	struct denode ndirent;
140 	struct denode *dep;
141 	struct denode *pdep = VTODE(ap->a_dvp);
142 	struct timespec ts;
143 	int error;
144 
145 #ifdef MSDOSFS_DEBUG
146 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
147 #endif
148 
149 	/*
150 	 * If this is the root directory and there is no space left we
151 	 * can't do anything.  This is because the root directory can not
152 	 * change size.
153 	 */
154 	if (pdep->de_StartCluster == MSDOSFSROOT
155 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
156 		error = ENOSPC;
157 		goto bad;
158 	}
159 
160 	/*
161 	 * Create a directory entry for the file, then call createde() to
162 	 * have it installed. NOTE: DOS files are always executable.  We
163 	 * use the absence of the owner write bit to make the file
164 	 * readonly.
165 	 */
166 #ifdef DIAGNOSTIC
167 	if ((cnp->cn_flags & HASBUF) == 0)
168 		panic("msdosfs_create: no name");
169 #endif
170 	bzero(&ndirent, sizeof(ndirent));
171 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
172 	if (error)
173 		goto bad;
174 
175 	ndirent.de_Attributes = ATTR_ARCHIVE;
176 	ndirent.de_LowerCase = 0;
177 	ndirent.de_StartCluster = 0;
178 	ndirent.de_FileSize = 0;
179 	ndirent.de_pmp = pdep->de_pmp;
180 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
181 	getnanotime(&ts);
182 	DETIMES(&ndirent, &ts, &ts, &ts);
183 	error = createde(&ndirent, pdep, &dep, cnp);
184 	if (error)
185 		goto bad;
186 	*ap->a_vpp = DETOV(dep);
187 	return (0);
188 
189 bad:
190 	return (error);
191 }
192 
193 static int
194 msdosfs_mknod(ap)
195 	struct vop_mknod_args /* {
196 		struct vnode *a_dvp;
197 		struct vnode **a_vpp;
198 		struct componentname *a_cnp;
199 		struct vattr *a_vap;
200 	} */ *ap;
201 {
202 
203     return (EINVAL);
204 }
205 
206 static int
207 msdosfs_open(ap)
208 	struct vop_open_args /* {
209 		struct vnode *a_vp;
210 		int a_mode;
211 		struct ucred *a_cred;
212 		struct thread *a_td;
213 		struct file *a_fp;
214 	} */ *ap;
215 {
216 	struct denode *dep = VTODE(ap->a_vp);
217 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
218 	return 0;
219 }
220 
221 static int
222 msdosfs_close(ap)
223 	struct vop_close_args /* {
224 		struct vnode *a_vp;
225 		int a_fflag;
226 		struct ucred *a_cred;
227 		struct thread *a_td;
228 	} */ *ap;
229 {
230 	struct vnode *vp = ap->a_vp;
231 	struct denode *dep = VTODE(vp);
232 	struct timespec ts;
233 
234 	VI_LOCK(vp);
235 	if (vp->v_usecount > 1) {
236 		getnanotime(&ts);
237 		DETIMES(dep, &ts, &ts, &ts);
238 	}
239 	VI_UNLOCK(vp);
240 	return 0;
241 }
242 
243 static int
244 msdosfs_access(ap)
245 	struct vop_access_args /* {
246 		struct vnode *a_vp;
247 		accmode_t a_accmode;
248 		struct ucred *a_cred;
249 		struct thread *a_td;
250 	} */ *ap;
251 {
252 	struct vnode *vp = ap->a_vp;
253 	struct denode *dep = VTODE(ap->a_vp);
254 	struct msdosfsmount *pmp = dep->de_pmp;
255 	mode_t file_mode;
256 	accmode_t accmode = ap->a_accmode;
257 
258 	file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
259 	file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
260 
261 	/*
262 	 * Disallow writing to directories and regular files if the
263 	 * filesystem is read-only.
264 	 */
265 	if (accmode & VWRITE) {
266 		switch (vp->v_type) {
267 		case VREG:
268 		case VDIR:
269 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
270 				return (EROFS);
271 			break;
272 		default:
273 			break;
274 		}
275 	}
276 
277 	return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
278 	    ap->a_accmode, ap->a_cred, NULL));
279 }
280 
281 static int
282 msdosfs_getattr(ap)
283 	struct vop_getattr_args /* {
284 		struct vnode *a_vp;
285 		struct vattr *a_vap;
286 		struct ucred *a_cred;
287 	} */ *ap;
288 {
289 	struct denode *dep = VTODE(ap->a_vp);
290 	struct msdosfsmount *pmp = dep->de_pmp;
291 	struct vattr *vap = ap->a_vap;
292 	mode_t mode;
293 	struct timespec ts;
294 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
295 	uint64_t fileid;
296 
297 	getnanotime(&ts);
298 	DETIMES(dep, &ts, &ts, &ts);
299 	vap->va_fsid = dev2udev(pmp->pm_dev);
300 	/*
301 	 * The following computation of the fileid must be the same as that
302 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
303 	 * doesn't work.
304 	 */
305 	if (dep->de_Attributes & ATTR_DIRECTORY) {
306 		fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
307 		    dirsperblk;
308 		if (dep->de_StartCluster == MSDOSFSROOT)
309 			fileid = 1;
310 	} else {
311 		fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
312 		    dirsperblk;
313 		if (dep->de_dirclust == MSDOSFSROOT)
314 			fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
315 		fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
316 	}
317 
318 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
319 		vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
320 	else
321 		vap->va_fileid = (long)fileid;
322 
323 	mode = S_IRWXU|S_IRWXG|S_IRWXO;
324 	vap->va_mode = mode &
325 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
326 	vap->va_uid = pmp->pm_uid;
327 	vap->va_gid = pmp->pm_gid;
328 	vap->va_nlink = 1;
329 	vap->va_rdev = NODEV;
330 	vap->va_size = dep->de_FileSize;
331 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
332 	vap->va_ctime = vap->va_mtime;
333 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
334 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
335 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
336 		    0, &vap->va_birthtime);
337 	} else {
338 		vap->va_atime = vap->va_mtime;
339 		vap->va_birthtime.tv_sec = -1;
340 		vap->va_birthtime.tv_nsec = 0;
341 	}
342 	vap->va_flags = 0;
343 	if (dep->de_Attributes & ATTR_ARCHIVE)
344 		vap->va_flags |= UF_ARCHIVE;
345 	if (dep->de_Attributes & ATTR_HIDDEN)
346 		vap->va_flags |= UF_HIDDEN;
347 	if (dep->de_Attributes & ATTR_READONLY)
348 		vap->va_flags |= UF_READONLY;
349 	if (dep->de_Attributes & ATTR_SYSTEM)
350 		vap->va_flags |= UF_SYSTEM;
351 	vap->va_gen = 0;
352 	vap->va_blocksize = pmp->pm_bpcluster;
353 	vap->va_bytes =
354 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
355 	vap->va_type = ap->a_vp->v_type;
356 	vap->va_filerev = dep->de_modrev;
357 	return (0);
358 }
359 
360 static int
361 msdosfs_setattr(ap)
362 	struct vop_setattr_args /* {
363 		struct vnode *a_vp;
364 		struct vattr *a_vap;
365 		struct ucred *a_cred;
366 	} */ *ap;
367 {
368 	struct vnode *vp = ap->a_vp;
369 	struct denode *dep = VTODE(ap->a_vp);
370 	struct msdosfsmount *pmp = dep->de_pmp;
371 	struct vattr *vap = ap->a_vap;
372 	struct ucred *cred = ap->a_cred;
373 	struct thread *td = curthread;
374 	int error = 0;
375 
376 #ifdef MSDOSFS_DEBUG
377 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
378 	    ap->a_vp, vap, cred);
379 #endif
380 
381 	/*
382 	 * Check for unsettable attributes.
383 	 */
384 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
385 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
386 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
387 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
388 #ifdef MSDOSFS_DEBUG
389 		printf("msdosfs_setattr(): returning EINVAL\n");
390 		printf("    va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
391 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
392 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
393 		    vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
394 		printf("    va_uid %x, va_gid %x\n",
395 		    vap->va_uid, vap->va_gid);
396 #endif
397 		return (EINVAL);
398 	}
399 
400 	/*
401 	 * We don't allow setting attributes on the root directory.
402 	 * The special case for the root directory is because before
403 	 * FAT32, the root directory didn't have an entry for itself
404 	 * (and was otherwise special).  With FAT32, the root
405 	 * directory is not so special, but still doesn't have an
406 	 * entry for itself.
407 	 */
408 	if (vp->v_vflag & VV_ROOT)
409 		return (EINVAL);
410 
411 	if (vap->va_flags != VNOVAL) {
412 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
413 			return (EROFS);
414 		if (cred->cr_uid != pmp->pm_uid) {
415 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
416 			if (error)
417 				return (error);
418 		}
419 		/*
420 		 * We are very inconsistent about handling unsupported
421 		 * attributes.  We ignored the access time and the
422 		 * read and execute bits.  We were strict for the other
423 		 * attributes.
424 		 */
425 		if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
426 		    UF_SYSTEM))
427 			return EOPNOTSUPP;
428 		if (vap->va_flags & UF_ARCHIVE)
429 			dep->de_Attributes |= ATTR_ARCHIVE;
430 		else
431 			dep->de_Attributes &= ~ATTR_ARCHIVE;
432 		if (vap->va_flags & UF_HIDDEN)
433 			dep->de_Attributes |= ATTR_HIDDEN;
434 		else
435 			dep->de_Attributes &= ~ATTR_HIDDEN;
436 		/* We don't allow changing the readonly bit on directories. */
437 		if (vp->v_type != VDIR) {
438 			if (vap->va_flags & UF_READONLY)
439 				dep->de_Attributes |= ATTR_READONLY;
440 			else
441 				dep->de_Attributes &= ~ATTR_READONLY;
442 		}
443 		if (vap->va_flags & UF_SYSTEM)
444 			dep->de_Attributes |= ATTR_SYSTEM;
445 		else
446 			dep->de_Attributes &= ~ATTR_SYSTEM;
447 		dep->de_flag |= DE_MODIFIED;
448 	}
449 
450 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
451 		uid_t uid;
452 		gid_t gid;
453 
454 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
455 			return (EROFS);
456 		uid = vap->va_uid;
457 		if (uid == (uid_t)VNOVAL)
458 			uid = pmp->pm_uid;
459 		gid = vap->va_gid;
460 		if (gid == (gid_t)VNOVAL)
461 			gid = pmp->pm_gid;
462 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
463 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
464 			error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
465 			if (error)
466 				return (error);
467 		}
468 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
469 			return EINVAL;
470 	}
471 
472 	if (vap->va_size != VNOVAL) {
473 		switch (vp->v_type) {
474 		case VDIR:
475 			return (EISDIR);
476 		case VREG:
477 			/*
478 			 * Truncation is only supported for regular files,
479 			 * Disallow it if the filesystem is read-only.
480 			 */
481 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
482 				return (EROFS);
483 			break;
484 		default:
485 			/*
486 			 * According to POSIX, the result is unspecified
487 			 * for file types other than regular files,
488 			 * directories and shared memory objects.  We
489 			 * don't support any file types except regular
490 			 * files and directories in this file system, so
491 			 * this (default) case is unreachable and can do
492 			 * anything.  Keep falling through to detrunc()
493 			 * for now.
494 			 */
495 			break;
496 		}
497 		error = detrunc(dep, vap->va_size, 0, cred);
498 		if (error)
499 			return error;
500 	}
501 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
502 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
503 			return (EROFS);
504 		error = vn_utimes_perm(vp, vap, cred, td);
505 		if (error != 0)
506 			return (error);
507 		if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
508 		    vap->va_atime.tv_sec != VNOVAL) {
509 			dep->de_flag &= ~DE_ACCESS;
510 			timespec2fattime(&vap->va_atime, 0,
511 			    &dep->de_ADate, NULL, NULL);
512 		}
513 		if (vap->va_mtime.tv_sec != VNOVAL) {
514 			dep->de_flag &= ~DE_UPDATE;
515 			timespec2fattime(&vap->va_mtime, 0,
516 			    &dep->de_MDate, &dep->de_MTime, NULL);
517 		}
518 		/*
519 		 * We don't set the archive bit when modifying the time of
520 		 * a directory to emulate the Windows/DOS behavior.
521 		 */
522 		if (vp->v_type != VDIR)
523 			dep->de_Attributes |= ATTR_ARCHIVE;
524 		dep->de_flag |= DE_MODIFIED;
525 	}
526 	/*
527 	 * DOS files only have the ability to have their writability
528 	 * attribute set, so we use the owner write bit to set the readonly
529 	 * attribute.
530 	 */
531 	if (vap->va_mode != (mode_t)VNOVAL) {
532 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
533 			return (EROFS);
534 		if (cred->cr_uid != pmp->pm_uid) {
535 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
536 			if (error)
537 				return (error);
538 		}
539 		if (vp->v_type != VDIR) {
540 			/* We ignore the read and execute bits. */
541 			if (vap->va_mode & VWRITE)
542 				dep->de_Attributes &= ~ATTR_READONLY;
543 			else
544 				dep->de_Attributes |= ATTR_READONLY;
545 			dep->de_Attributes |= ATTR_ARCHIVE;
546 			dep->de_flag |= DE_MODIFIED;
547 		}
548 	}
549 	return (deupdat(dep, 0));
550 }
551 
552 static int
553 msdosfs_read(ap)
554 	struct vop_read_args /* {
555 		struct vnode *a_vp;
556 		struct uio *a_uio;
557 		int a_ioflag;
558 		struct ucred *a_cred;
559 	} */ *ap;
560 {
561 	int error = 0;
562 	int blsize;
563 	int isadir;
564 	ssize_t orig_resid;
565 	u_int n;
566 	u_long diff;
567 	u_long on;
568 	daddr_t lbn;
569 	daddr_t rablock;
570 	int rasize;
571 	int seqcount;
572 	struct buf *bp;
573 	struct vnode *vp = ap->a_vp;
574 	struct denode *dep = VTODE(vp);
575 	struct msdosfsmount *pmp = dep->de_pmp;
576 	struct uio *uio = ap->a_uio;
577 
578 	/*
579 	 * If they didn't ask for any data, then we are done.
580 	 */
581 	orig_resid = uio->uio_resid;
582 	if (orig_resid == 0)
583 		return (0);
584 
585 	/*
586 	 * The caller is supposed to ensure that
587 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
588 	 * We don't need to check for large offsets as in ffs because
589 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
590 	 * offsets cannot cause overflow even in theory.
591 	 */
592 
593 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
594 
595 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
596 	do {
597 		if (uio->uio_offset >= dep->de_FileSize)
598 			break;
599 		lbn = de_cluster(pmp, uio->uio_offset);
600 		rablock = lbn + 1;
601 		blsize = pmp->pm_bpcluster;
602 		on = uio->uio_offset & pmp->pm_crbomask;
603 		/*
604 		 * If we are operating on a directory file then be sure to
605 		 * do i/o with the vnode for the filesystem instead of the
606 		 * vnode for the directory.
607 		 */
608 		if (isadir) {
609 			/* convert cluster # to block # */
610 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
611 			if (error == E2BIG) {
612 				error = EINVAL;
613 				break;
614 			} else if (error)
615 				break;
616 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
617 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
618 			error = bread(vp, lbn, blsize, NOCRED, &bp);
619 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
620 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
621 			    NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
622 		} else if (seqcount > 1) {
623 			rasize = blsize;
624 			error = breadn(vp, lbn,
625 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
626 		} else {
627 			error = bread(vp, lbn, blsize, NOCRED, &bp);
628 		}
629 		if (error) {
630 			brelse(bp);
631 			break;
632 		}
633 		diff = pmp->pm_bpcluster - on;
634 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
635 		diff = dep->de_FileSize - uio->uio_offset;
636 		if (diff < n)
637 			n = diff;
638 		diff = blsize - bp->b_resid;
639 		if (diff < n)
640 			n = diff;
641 		error = uiomove(bp->b_data + on, (int) n, uio);
642 		brelse(bp);
643 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
644 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
645 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
646 		dep->de_flag |= DE_ACCESS;
647 	return (error);
648 }
649 
650 /*
651  * Write data to a file or directory.
652  */
653 static int
654 msdosfs_write(ap)
655 	struct vop_write_args /* {
656 		struct vnode *a_vp;
657 		struct uio *a_uio;
658 		int a_ioflag;
659 		struct ucred *a_cred;
660 	} */ *ap;
661 {
662 	int n;
663 	int croffset;
664 	ssize_t resid;
665 	u_long osize;
666 	int error = 0;
667 	u_long count;
668 	int seqcount;
669 	daddr_t bn, lastcn;
670 	struct buf *bp;
671 	int ioflag = ap->a_ioflag;
672 	struct uio *uio = ap->a_uio;
673 	struct vnode *vp = ap->a_vp;
674 	struct vnode *thisvp;
675 	struct denode *dep = VTODE(vp);
676 	struct msdosfsmount *pmp = dep->de_pmp;
677 	struct ucred *cred = ap->a_cred;
678 
679 #ifdef MSDOSFS_DEBUG
680 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
681 	    vp, uio, ioflag, cred);
682 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
683 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
684 #endif
685 
686 	switch (vp->v_type) {
687 	case VREG:
688 		if (ioflag & IO_APPEND)
689 			uio->uio_offset = dep->de_FileSize;
690 		thisvp = vp;
691 		break;
692 	case VDIR:
693 		return EISDIR;
694 	default:
695 		panic("msdosfs_write(): bad file type");
696 	}
697 
698 	/*
699 	 * This is needed (unlike in ffs_write()) because we extend the
700 	 * file outside of the loop but we don't want to extend the file
701 	 * for writes of 0 bytes.
702 	 */
703 	if (uio->uio_resid == 0)
704 		return (0);
705 
706 	/*
707 	 * The caller is supposed to ensure that
708 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
709 	 */
710 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
711 		return (EFBIG);
712 
713 	/*
714 	 * If they've exceeded their filesize limit, tell them about it.
715 	 */
716 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
717 		return (EFBIG);
718 
719 	/*
720 	 * If the offset we are starting the write at is beyond the end of
721 	 * the file, then they've done a seek.  Unix filesystems allow
722 	 * files with holes in them, DOS doesn't so we must fill the hole
723 	 * with zeroed blocks.
724 	 */
725 	if (uio->uio_offset > dep->de_FileSize) {
726 		error = deextend(dep, uio->uio_offset, cred);
727 		if (error)
728 			return (error);
729 	}
730 
731 	/*
732 	 * Remember some values in case the write fails.
733 	 */
734 	resid = uio->uio_resid;
735 	osize = dep->de_FileSize;
736 
737 	/*
738 	 * If we write beyond the end of the file, extend it to its ultimate
739 	 * size ahead of the time to hopefully get a contiguous area.
740 	 */
741 	if (uio->uio_offset + resid > osize) {
742 		count = de_clcount(pmp, uio->uio_offset + resid) -
743 			de_clcount(pmp, osize);
744 		error = extendfile(dep, count, NULL, NULL, 0);
745 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
746 			goto errexit;
747 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
748 	} else
749 		lastcn = de_clcount(pmp, osize) - 1;
750 
751 	seqcount = ioflag >> IO_SEQSHIFT;
752 	do {
753 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
754 			error = ENOSPC;
755 			break;
756 		}
757 
758 		croffset = uio->uio_offset & pmp->pm_crbomask;
759 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
760 		if (uio->uio_offset + n > dep->de_FileSize) {
761 			dep->de_FileSize = uio->uio_offset + n;
762 			/* The object size needs to be set before buffer is allocated */
763 			vnode_pager_setsize(vp, dep->de_FileSize);
764 		}
765 
766 		bn = de_cluster(pmp, uio->uio_offset);
767 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
768 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
769 			> de_cluster(pmp, uio->uio_offset)
770 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
771 			/*
772 			 * If either the whole cluster gets written,
773 			 * or we write the cluster from its start beyond EOF,
774 			 * then no need to read data from disk.
775 			 */
776 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
777 			vfs_bio_clrbuf(bp);
778 			/*
779 			 * Do the bmap now, since pcbmap needs buffers
780 			 * for the fat table. (see msdosfs_strategy)
781 			 */
782 			if (bp->b_blkno == bp->b_lblkno) {
783 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
784 				if (error)
785 					bp->b_blkno = -1;
786 				else
787 					bp->b_blkno = bn;
788 			}
789 			if (bp->b_blkno == -1) {
790 				brelse(bp);
791 				if (!error)
792 					error = EIO;		/* XXX */
793 				break;
794 			}
795 		} else {
796 			/*
797 			 * The block we need to write into exists, so read it in.
798 			 */
799 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
800 			if (error) {
801 				brelse(bp);
802 				break;
803 			}
804 		}
805 
806 		/*
807 		 * Should these vnode_pager_* functions be done on dir
808 		 * files?
809 		 */
810 
811 		/*
812 		 * Copy the data from user space into the buf header.
813 		 */
814 		error = uiomove(bp->b_data + croffset, n, uio);
815 		if (error) {
816 			brelse(bp);
817 			break;
818 		}
819 
820 		/* Prepare for clustered writes in some else clauses. */
821 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
822 			bp->b_flags |= B_CLUSTEROK;
823 
824 		/*
825 		 * If IO_SYNC, then each buffer is written synchronously.
826 		 * Otherwise, if we have a severe page deficiency then
827 		 * write the buffer asynchronously.  Otherwise, if on a
828 		 * cluster boundary then write the buffer asynchronously,
829 		 * combining it with contiguous clusters if permitted and
830 		 * possible, since we don't expect more writes into this
831 		 * buffer soon.  Otherwise, do a delayed write because we
832 		 * expect more writes into this buffer soon.
833 		 */
834 		if (ioflag & IO_SYNC)
835 			(void)bwrite(bp);
836 		else if (vm_page_count_severe() || buf_dirty_count_severe())
837 			bawrite(bp);
838 		else if (n + croffset == pmp->pm_bpcluster) {
839 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
840 				cluster_write(vp, bp, dep->de_FileSize,
841 				    seqcount, 0);
842 			else
843 				bawrite(bp);
844 		} else
845 			bdwrite(bp);
846 		dep->de_flag |= DE_UPDATE;
847 	} while (error == 0 && uio->uio_resid > 0);
848 
849 	/*
850 	 * If the write failed and they want us to, truncate the file back
851 	 * to the size it was before the write was attempted.
852 	 */
853 errexit:
854 	if (error) {
855 		if (ioflag & IO_UNIT) {
856 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
857 			uio->uio_offset -= resid - uio->uio_resid;
858 			uio->uio_resid = resid;
859 		} else {
860 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
861 			if (uio->uio_resid != resid)
862 				error = 0;
863 		}
864 	} else if (ioflag & IO_SYNC)
865 		error = deupdat(dep, 1);
866 	return (error);
867 }
868 
869 /*
870  * Flush the blocks of a file to disk.
871  */
872 static int
873 msdosfs_fsync(ap)
874 	struct vop_fsync_args /* {
875 		struct vnode *a_vp;
876 		struct ucred *a_cred;
877 		int a_waitfor;
878 		struct thread *a_td;
879 	} */ *ap;
880 {
881 	struct vnode *devvp;
882 	int allerror, error;
883 
884 	vop_stdfsync(ap);
885 
886 	/*
887 	* If the syncing request comes from fsync(2), sync the entire
888 	* FAT and any other metadata that happens to be on devvp.  We
889 	* need this mainly for the FAT.  We write the FAT sloppily, and
890 	* syncing it all now is the best we can easily do to get all
891 	* directory entries associated with the file (not just the file)
892 	* fully synced.  The other metadata includes critical metadata
893 	* for all directory entries, but only in the MNT_ASYNC case.  We
894 	* will soon sync all metadata in the file's directory entry.
895 	* Non-critical metadata for associated directory entries only
896 	* gets synced accidentally, as in most file systems.
897 	*/
898 	if (ap->a_waitfor == MNT_WAIT) {
899 		devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
900 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
901 		allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
902 		VOP_UNLOCK(devvp, 0);
903 	} else
904 		allerror = 0;
905 
906 	error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
907 	if (allerror == 0)
908 		allerror = error;
909 	return (allerror);
910 }
911 
912 static int
913 msdosfs_remove(ap)
914 	struct vop_remove_args /* {
915 		struct vnode *a_dvp;
916 		struct vnode *a_vp;
917 		struct componentname *a_cnp;
918 	} */ *ap;
919 {
920 	struct denode *dep = VTODE(ap->a_vp);
921 	struct denode *ddep = VTODE(ap->a_dvp);
922 	int error;
923 
924 	if (ap->a_vp->v_type == VDIR)
925 		error = EPERM;
926 	else
927 		error = removede(ddep, dep);
928 #ifdef MSDOSFS_DEBUG
929 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
930 #endif
931 	return (error);
932 }
933 
934 /*
935  * DOS filesystems don't know what links are.
936  */
937 static int
938 msdosfs_link(ap)
939 	struct vop_link_args /* {
940 		struct vnode *a_tdvp;
941 		struct vnode *a_vp;
942 		struct componentname *a_cnp;
943 	} */ *ap;
944 {
945 	return (EOPNOTSUPP);
946 }
947 
948 /*
949  * Renames on files require moving the denode to a new hash queue since the
950  * denode's location is used to compute which hash queue to put the file
951  * in. Unless it is a rename in place.  For example "mv a b".
952  *
953  * What follows is the basic algorithm:
954  *
955  * if (file move) {
956  *	if (dest file exists) {
957  *		remove dest file
958  *	}
959  *	if (dest and src in same directory) {
960  *		rewrite name in existing directory slot
961  *	} else {
962  *		write new entry in dest directory
963  *		update offset and dirclust in denode
964  *		move denode to new hash chain
965  *		clear old directory entry
966  *	}
967  * } else {
968  *	directory move
969  *	if (dest directory exists) {
970  *		if (dest is not empty) {
971  *			return ENOTEMPTY
972  *		}
973  *		remove dest directory
974  *	}
975  *	if (dest and src in same directory) {
976  *		rewrite name in existing entry
977  *	} else {
978  *		be sure dest is not a child of src directory
979  *		write entry in dest directory
980  *		update "." and ".." in moved directory
981  *		clear old directory entry for moved directory
982  *	}
983  * }
984  *
985  * On entry:
986  *	source's parent directory is unlocked
987  *	source file or directory is unlocked
988  *	destination's parent directory is locked
989  *	destination file or directory is locked if it exists
990  *
991  * On exit:
992  *	all denodes should be released
993  */
994 static int
995 msdosfs_rename(ap)
996 	struct vop_rename_args /* {
997 		struct vnode *a_fdvp;
998 		struct vnode *a_fvp;
999 		struct componentname *a_fcnp;
1000 		struct vnode *a_tdvp;
1001 		struct vnode *a_tvp;
1002 		struct componentname *a_tcnp;
1003 	} */ *ap;
1004 {
1005 	struct vnode *tdvp = ap->a_tdvp;
1006 	struct vnode *fvp = ap->a_fvp;
1007 	struct vnode *fdvp = ap->a_fdvp;
1008 	struct vnode *tvp = ap->a_tvp;
1009 	struct componentname *tcnp = ap->a_tcnp;
1010 	struct componentname *fcnp = ap->a_fcnp;
1011 	struct denode *ip, *xp, *dp, *zp;
1012 	u_char toname[12], oldname[11];
1013 	u_long from_diroffset, to_diroffset;
1014 	u_char to_count;
1015 	int doingdirectory = 0, newparent = 0;
1016 	int error;
1017 	u_long cn, pcl;
1018 	daddr_t bn;
1019 	struct denode *fddep;	/* from file's parent directory	 */
1020 	struct msdosfsmount *pmp;
1021 	struct direntry *dotdotp;
1022 	struct buf *bp;
1023 
1024 	fddep = VTODE(ap->a_fdvp);
1025 	pmp = fddep->de_pmp;
1026 
1027 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1028 
1029 #ifdef DIAGNOSTIC
1030 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1031 	    (fcnp->cn_flags & HASBUF) == 0)
1032 		panic("msdosfs_rename: no name");
1033 #endif
1034 	/*
1035 	 * Check for cross-device rename.
1036 	 */
1037 	if (fvp->v_mount != tdvp->v_mount ||
1038 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1039 		error = EXDEV;
1040 abortit:
1041 		if (tdvp == tvp)
1042 			vrele(tdvp);
1043 		else
1044 			vput(tdvp);
1045 		if (tvp)
1046 			vput(tvp);
1047 		vrele(fdvp);
1048 		vrele(fvp);
1049 		return (error);
1050 	}
1051 
1052 	/*
1053 	 * If source and dest are the same, do nothing.
1054 	 */
1055 	if (tvp == fvp) {
1056 		error = 0;
1057 		goto abortit;
1058 	}
1059 
1060 	error = vn_lock(fvp, LK_EXCLUSIVE);
1061 	if (error)
1062 		goto abortit;
1063 	dp = VTODE(fdvp);
1064 	ip = VTODE(fvp);
1065 
1066 	/*
1067 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1068 	 * leads to a crippled directory tree.  It's pretty tough to do a
1069 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1070 	 * doesn't work if the ".." entry is missing.
1071 	 */
1072 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1073 		/*
1074 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1075 		 */
1076 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1077 		    dp == ip ||
1078 		    (fcnp->cn_flags & ISDOTDOT) ||
1079 		    (tcnp->cn_flags & ISDOTDOT) ||
1080 		    (ip->de_flag & DE_RENAME)) {
1081 			VOP_UNLOCK(fvp, 0);
1082 			error = EINVAL;
1083 			goto abortit;
1084 		}
1085 		ip->de_flag |= DE_RENAME;
1086 		doingdirectory++;
1087 	}
1088 
1089 	/*
1090 	 * When the target exists, both the directory
1091 	 * and target vnodes are returned locked.
1092 	 */
1093 	dp = VTODE(tdvp);
1094 	xp = tvp ? VTODE(tvp) : NULL;
1095 	/*
1096 	 * Remember direntry place to use for destination
1097 	 */
1098 	to_diroffset = dp->de_fndoffset;
1099 	to_count = dp->de_fndcnt;
1100 
1101 	/*
1102 	 * If ".." must be changed (ie the directory gets a new
1103 	 * parent) then the source directory must not be in the
1104 	 * directory hierarchy above the target, as this would
1105 	 * orphan everything below the source directory. Also
1106 	 * the user must have write permission in the source so
1107 	 * as to be able to change "..". We must repeat the call
1108 	 * to namei, as the parent directory is unlocked by the
1109 	 * call to doscheckpath().
1110 	 */
1111 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1112 	VOP_UNLOCK(fvp, 0);
1113 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1114 		newparent = 1;
1115 	if (doingdirectory && newparent) {
1116 		if (error)	/* write access check above */
1117 			goto bad;
1118 		if (xp != NULL)
1119 			vput(tvp);
1120 		/*
1121 		 * doscheckpath() vput()'s dp,
1122 		 * so we have to do a relookup afterwards
1123 		 */
1124 		error = doscheckpath(ip, dp);
1125 		if (error)
1126 			goto out;
1127 		if ((tcnp->cn_flags & SAVESTART) == 0)
1128 			panic("msdosfs_rename: lost to startdir");
1129 		error = relookup(tdvp, &tvp, tcnp);
1130 		if (error)
1131 			goto out;
1132 		dp = VTODE(tdvp);
1133 		xp = tvp ? VTODE(tvp) : NULL;
1134 	}
1135 
1136 	if (xp != NULL) {
1137 		/*
1138 		 * Target must be empty if a directory and have no links
1139 		 * to it. Also, ensure source and target are compatible
1140 		 * (both directories, or both not directories).
1141 		 */
1142 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1143 			if (!dosdirempty(xp)) {
1144 				error = ENOTEMPTY;
1145 				goto bad;
1146 			}
1147 			if (!doingdirectory) {
1148 				error = ENOTDIR;
1149 				goto bad;
1150 			}
1151 			cache_purge(tdvp);
1152 		} else if (doingdirectory) {
1153 			error = EISDIR;
1154 			goto bad;
1155 		}
1156 		error = removede(dp, xp);
1157 		if (error)
1158 			goto bad;
1159 		vput(tvp);
1160 		xp = NULL;
1161 	}
1162 
1163 	/*
1164 	 * Convert the filename in tcnp into a dos filename. We copy this
1165 	 * into the denode and directory entry for the destination
1166 	 * file/directory.
1167 	 */
1168 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1169 	if (error)
1170 		goto abortit;
1171 
1172 	/*
1173 	 * Since from wasn't locked at various places above,
1174 	 * have to do a relookup here.
1175 	 */
1176 	fcnp->cn_flags &= ~MODMASK;
1177 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1178 	if ((fcnp->cn_flags & SAVESTART) == 0)
1179 		panic("msdosfs_rename: lost from startdir");
1180 	if (!newparent)
1181 		VOP_UNLOCK(tdvp, 0);
1182 	if (relookup(fdvp, &fvp, fcnp) == 0)
1183 		vrele(fdvp);
1184 	if (fvp == NULL) {
1185 		/*
1186 		 * From name has disappeared.
1187 		 */
1188 		if (doingdirectory)
1189 			panic("rename: lost dir entry");
1190 		if (newparent)
1191 			VOP_UNLOCK(tdvp, 0);
1192 		vrele(tdvp);
1193 		vrele(ap->a_fvp);
1194 		return 0;
1195 	}
1196 	xp = VTODE(fvp);
1197 	zp = VTODE(fdvp);
1198 	from_diroffset = zp->de_fndoffset;
1199 
1200 	/*
1201 	 * Ensure that the directory entry still exists and has not
1202 	 * changed till now. If the source is a file the entry may
1203 	 * have been unlinked or renamed. In either case there is
1204 	 * no further work to be done. If the source is a directory
1205 	 * then it cannot have been rmdir'ed or renamed; this is
1206 	 * prohibited by the DE_RENAME flag.
1207 	 */
1208 	if (xp != ip) {
1209 		if (doingdirectory)
1210 			panic("rename: lost dir entry");
1211 		VOP_UNLOCK(fvp, 0);
1212 		if (newparent)
1213 			VOP_UNLOCK(fdvp, 0);
1214 		vrele(ap->a_fvp);
1215 		xp = NULL;
1216 	} else {
1217 		vrele(fvp);
1218 		xp = NULL;
1219 
1220 		/*
1221 		 * First write a new entry in the destination
1222 		 * directory and mark the entry in the source directory
1223 		 * as deleted.  Then move the denode to the correct hash
1224 		 * chain for its new location in the filesystem.  And, if
1225 		 * we moved a directory, then update its .. entry to point
1226 		 * to the new parent directory.
1227 		 */
1228 		bcopy(ip->de_Name, oldname, 11);
1229 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1230 		dp->de_fndoffset = to_diroffset;
1231 		dp->de_fndcnt = to_count;
1232 		error = createde(ip, dp, (struct denode **)0, tcnp);
1233 		if (error) {
1234 			bcopy(oldname, ip->de_Name, 11);
1235 			if (newparent)
1236 				VOP_UNLOCK(fdvp, 0);
1237 			VOP_UNLOCK(fvp, 0);
1238 			goto bad;
1239 		}
1240 		/*
1241 		 * If ip is for a directory, then its name should always
1242 		 * be "." since it is for the directory entry in the
1243 		 * directory itself (msdosfs_lookup() always translates
1244 		 * to the "." entry so as to get a unique denode, except
1245 		 * for the root directory there are different
1246 		 * complications).  However, we just corrupted its name
1247 		 * to pass the correct name to createde().  Undo this.
1248 		 */
1249 		if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1250 			bcopy(oldname, ip->de_Name, 11);
1251 		ip->de_refcnt++;
1252 		zp->de_fndoffset = from_diroffset;
1253 		error = removede(zp, ip);
1254 		if (error) {
1255 			/* XXX should downgrade to ro here, fs is corrupt */
1256 			if (newparent)
1257 				VOP_UNLOCK(fdvp, 0);
1258 			VOP_UNLOCK(fvp, 0);
1259 			goto bad;
1260 		}
1261 		if (!doingdirectory) {
1262 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1263 				       &ip->de_dirclust, 0);
1264 			if (error) {
1265 				/* XXX should downgrade to ro here, fs is corrupt */
1266 				if (newparent)
1267 					VOP_UNLOCK(fdvp, 0);
1268 				VOP_UNLOCK(fvp, 0);
1269 				goto bad;
1270 			}
1271 			if (ip->de_dirclust == MSDOSFSROOT)
1272 				ip->de_diroffset = to_diroffset;
1273 			else
1274 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1275 		}
1276 		reinsert(ip);
1277 		if (newparent)
1278 			VOP_UNLOCK(fdvp, 0);
1279 	}
1280 
1281 	/*
1282 	 * If we moved a directory to a new parent directory, then we must
1283 	 * fixup the ".." entry in the moved directory.
1284 	 */
1285 	if (doingdirectory && newparent) {
1286 		cn = ip->de_StartCluster;
1287 		if (cn == MSDOSFSROOT) {
1288 			/* this should never happen */
1289 			panic("msdosfs_rename(): updating .. in root directory?");
1290 		} else
1291 			bn = cntobn(pmp, cn);
1292 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1293 			      NOCRED, &bp);
1294 		if (error) {
1295 			/* XXX should downgrade to ro here, fs is corrupt */
1296 			brelse(bp);
1297 			VOP_UNLOCK(fvp, 0);
1298 			goto bad;
1299 		}
1300 		dotdotp = (struct direntry *)bp->b_data + 1;
1301 		pcl = dp->de_StartCluster;
1302 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1303 			pcl = MSDOSFSROOT;
1304 		putushort(dotdotp->deStartCluster, pcl);
1305 		if (FAT32(pmp))
1306 			putushort(dotdotp->deHighClust, pcl >> 16);
1307 		if (DOINGASYNC(fvp))
1308 			bdwrite(bp);
1309 		else if ((error = bwrite(bp)) != 0) {
1310 			/* XXX should downgrade to ro here, fs is corrupt */
1311 			VOP_UNLOCK(fvp, 0);
1312 			goto bad;
1313 		}
1314 	}
1315 
1316 	/*
1317 	 * The msdosfs lookup is case insensitive. Several aliases may
1318 	 * be inserted for a single directory entry. As a consequnce,
1319 	 * name cache purge done by lookup for fvp when DELETE op for
1320 	 * namei is specified, might be not enough to expunge all
1321 	 * namecache entries that were installed for this direntry.
1322 	 */
1323 	cache_purge(fvp);
1324 	VOP_UNLOCK(fvp, 0);
1325 bad:
1326 	if (xp)
1327 		vput(tvp);
1328 	vput(tdvp);
1329 out:
1330 	ip->de_flag &= ~DE_RENAME;
1331 	vrele(fdvp);
1332 	vrele(fvp);
1333 	return (error);
1334 
1335 }
1336 
1337 static struct {
1338 	struct direntry dot;
1339 	struct direntry dotdot;
1340 } dosdirtemplate = {
1341 	{	".          ",				/* the . entry */
1342 		ATTR_DIRECTORY,				/* file attribute */
1343 		0,					/* reserved */
1344 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1345 		{ 0, 0 },				/* access date */
1346 		{ 0, 0 },				/* high bits of start cluster */
1347 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1348 		{ 0, 0 },				/* startcluster */
1349 		{ 0, 0, 0, 0 }				/* filesize */
1350 	},
1351 	{	"..         ",				/* the .. entry */
1352 		ATTR_DIRECTORY,				/* file attribute */
1353 		0,					/* reserved */
1354 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1355 		{ 0, 0 },				/* access date */
1356 		{ 0, 0 },				/* high bits of start cluster */
1357 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1358 		{ 0, 0 },				/* startcluster */
1359 		{ 0, 0, 0, 0 }				/* filesize */
1360 	}
1361 };
1362 
1363 static int
1364 msdosfs_mkdir(ap)
1365 	struct vop_mkdir_args /* {
1366 		struct vnode *a_dvp;
1367 		struct vnode **a_vpp;
1368 		struvt componentname *a_cnp;
1369 		struct vattr *a_vap;
1370 	} */ *ap;
1371 {
1372 	struct componentname *cnp = ap->a_cnp;
1373 	struct denode *dep;
1374 	struct denode *pdep = VTODE(ap->a_dvp);
1375 	struct direntry *denp;
1376 	struct msdosfsmount *pmp = pdep->de_pmp;
1377 	struct buf *bp;
1378 	u_long newcluster, pcl;
1379 	int bn;
1380 	int error;
1381 	struct denode ndirent;
1382 	struct timespec ts;
1383 
1384 	/*
1385 	 * If this is the root directory and there is no space left we
1386 	 * can't do anything.  This is because the root directory can not
1387 	 * change size.
1388 	 */
1389 	if (pdep->de_StartCluster == MSDOSFSROOT
1390 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1391 		error = ENOSPC;
1392 		goto bad2;
1393 	}
1394 
1395 	/*
1396 	 * Allocate a cluster to hold the about to be created directory.
1397 	 */
1398 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1399 	if (error)
1400 		goto bad2;
1401 
1402 	bzero(&ndirent, sizeof(ndirent));
1403 	ndirent.de_pmp = pmp;
1404 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1405 	getnanotime(&ts);
1406 	DETIMES(&ndirent, &ts, &ts, &ts);
1407 
1408 	/*
1409 	 * Now fill the cluster with the "." and ".." entries. And write
1410 	 * the cluster to disk.  This way it is there for the parent
1411 	 * directory to be pointing at if there were a crash.
1412 	 */
1413 	bn = cntobn(pmp, newcluster);
1414 	/* always succeeds */
1415 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1416 	bzero(bp->b_data, pmp->pm_bpcluster);
1417 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1418 	denp = (struct direntry *)bp->b_data;
1419 	putushort(denp[0].deStartCluster, newcluster);
1420 	putushort(denp[0].deCDate, ndirent.de_CDate);
1421 	putushort(denp[0].deCTime, ndirent.de_CTime);
1422 	denp[0].deCHundredth = ndirent.de_CHun;
1423 	putushort(denp[0].deADate, ndirent.de_ADate);
1424 	putushort(denp[0].deMDate, ndirent.de_MDate);
1425 	putushort(denp[0].deMTime, ndirent.de_MTime);
1426 	pcl = pdep->de_StartCluster;
1427 	/*
1428 	 * Although the root directory has a non-magic starting cluster
1429 	 * number for FAT32, chkdsk and fsck_msdosfs still require
1430 	 * references to it in dotdot entries to be magic.
1431 	 */
1432 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1433 		pcl = MSDOSFSROOT;
1434 	putushort(denp[1].deStartCluster, pcl);
1435 	putushort(denp[1].deCDate, ndirent.de_CDate);
1436 	putushort(denp[1].deCTime, ndirent.de_CTime);
1437 	denp[1].deCHundredth = ndirent.de_CHun;
1438 	putushort(denp[1].deADate, ndirent.de_ADate);
1439 	putushort(denp[1].deMDate, ndirent.de_MDate);
1440 	putushort(denp[1].deMTime, ndirent.de_MTime);
1441 	if (FAT32(pmp)) {
1442 		putushort(denp[0].deHighClust, newcluster >> 16);
1443 		putushort(denp[1].deHighClust, pcl >> 16);
1444 	}
1445 
1446 	if (DOINGASYNC(ap->a_dvp))
1447 		bdwrite(bp);
1448 	else if ((error = bwrite(bp)) != 0)
1449 		goto bad;
1450 
1451 	/*
1452 	 * Now build up a directory entry pointing to the newly allocated
1453 	 * cluster.  This will be written to an empty slot in the parent
1454 	 * directory.
1455 	 */
1456 #ifdef DIAGNOSTIC
1457 	if ((cnp->cn_flags & HASBUF) == 0)
1458 		panic("msdosfs_mkdir: no name");
1459 #endif
1460 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1461 	if (error)
1462 		goto bad;
1463 
1464 	ndirent.de_Attributes = ATTR_DIRECTORY;
1465 	ndirent.de_LowerCase = 0;
1466 	ndirent.de_StartCluster = newcluster;
1467 	ndirent.de_FileSize = 0;
1468 	error = createde(&ndirent, pdep, &dep, cnp);
1469 	if (error)
1470 		goto bad;
1471 	*ap->a_vpp = DETOV(dep);
1472 	return (0);
1473 
1474 bad:
1475 	clusterfree(pmp, newcluster, NULL);
1476 bad2:
1477 	return (error);
1478 }
1479 
1480 static int
1481 msdosfs_rmdir(ap)
1482 	struct vop_rmdir_args /* {
1483 		struct vnode *a_dvp;
1484 		struct vnode *a_vp;
1485 		struct componentname *a_cnp;
1486 	} */ *ap;
1487 {
1488 	struct vnode *vp = ap->a_vp;
1489 	struct vnode *dvp = ap->a_dvp;
1490 	struct componentname *cnp = ap->a_cnp;
1491 	struct denode *ip, *dp;
1492 	int error;
1493 
1494 	ip = VTODE(vp);
1495 	dp = VTODE(dvp);
1496 
1497 	/*
1498 	 * Verify the directory is empty (and valid).
1499 	 * (Rmdir ".." won't be valid since
1500 	 *  ".." will contain a reference to
1501 	 *  the current directory and thus be
1502 	 *  non-empty.)
1503 	 */
1504 	error = 0;
1505 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1506 		error = ENOTEMPTY;
1507 		goto out;
1508 	}
1509 	/*
1510 	 * Delete the entry from the directory.  For dos filesystems this
1511 	 * gets rid of the directory entry on disk, the in memory copy
1512 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1513 	 * being found by deget().  When the vput() on dep is done we give
1514 	 * up access and eventually msdosfs_reclaim() will be called which
1515 	 * will remove it from the denode cache.
1516 	 */
1517 	error = removede(dp, ip);
1518 	if (error)
1519 		goto out;
1520 	/*
1521 	 * This is where we decrement the link count in the parent
1522 	 * directory.  Since dos filesystems don't do this we just purge
1523 	 * the name cache.
1524 	 */
1525 	cache_purge(dvp);
1526 	/*
1527 	 * Truncate the directory that is being deleted.
1528 	 */
1529 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1530 	cache_purge(vp);
1531 
1532 out:
1533 	return (error);
1534 }
1535 
1536 /*
1537  * DOS filesystems don't know what symlinks are.
1538  */
1539 static int
1540 msdosfs_symlink(ap)
1541 	struct vop_symlink_args /* {
1542 		struct vnode *a_dvp;
1543 		struct vnode **a_vpp;
1544 		struct componentname *a_cnp;
1545 		struct vattr *a_vap;
1546 		char *a_target;
1547 	} */ *ap;
1548 {
1549 	return (EOPNOTSUPP);
1550 }
1551 
1552 static int
1553 msdosfs_readdir(ap)
1554 	struct vop_readdir_args /* {
1555 		struct vnode *a_vp;
1556 		struct uio *a_uio;
1557 		struct ucred *a_cred;
1558 		int *a_eofflag;
1559 		int *a_ncookies;
1560 		u_long **a_cookies;
1561 	} */ *ap;
1562 {
1563 	struct mbnambuf nb;
1564 	int error = 0;
1565 	int diff;
1566 	long n;
1567 	int blsize;
1568 	long on;
1569 	u_long cn;
1570 	uint64_t fileno;
1571 	u_long dirsperblk;
1572 	long bias = 0;
1573 	daddr_t bn, lbn;
1574 	struct buf *bp;
1575 	struct denode *dep = VTODE(ap->a_vp);
1576 	struct msdosfsmount *pmp = dep->de_pmp;
1577 	struct direntry *dentp;
1578 	struct dirent dirbuf;
1579 	struct uio *uio = ap->a_uio;
1580 	u_long *cookies = NULL;
1581 	int ncookies = 0;
1582 	off_t offset, off;
1583 	int chksum = -1;
1584 
1585 #ifdef MSDOSFS_DEBUG
1586 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1587 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1588 #endif
1589 
1590 	/*
1591 	 * msdosfs_readdir() won't operate properly on regular files since
1592 	 * it does i/o only with the filesystem vnode, and hence can
1593 	 * retrieve the wrong block from the buffer cache for a plain file.
1594 	 * So, fail attempts to readdir() on a plain file.
1595 	 */
1596 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1597 		return (ENOTDIR);
1598 
1599 	/*
1600 	 * To be safe, initialize dirbuf
1601 	 */
1602 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1603 
1604 	/*
1605 	 * If the user buffer is smaller than the size of one dos directory
1606 	 * entry or the file offset is not a multiple of the size of a
1607 	 * directory entry, then we fail the read.
1608 	 */
1609 	off = offset = uio->uio_offset;
1610 	if (uio->uio_resid < sizeof(struct direntry) ||
1611 	    (offset & (sizeof(struct direntry) - 1)))
1612 		return (EINVAL);
1613 
1614 	if (ap->a_ncookies) {
1615 		ncookies = uio->uio_resid / 16;
1616 		cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1617 		       M_WAITOK);
1618 		*ap->a_cookies = cookies;
1619 		*ap->a_ncookies = ncookies;
1620 	}
1621 
1622 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1623 
1624 	/*
1625 	 * If they are reading from the root directory then, we simulate
1626 	 * the . and .. entries since these don't exist in the root
1627 	 * directory.  We also set the offset bias to make up for having to
1628 	 * simulate these entries. By this I mean that at file offset 64 we
1629 	 * read the first entry in the root directory that lives on disk.
1630 	 */
1631 	if (dep->de_StartCluster == MSDOSFSROOT
1632 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1633 #if 0
1634 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1635 		    offset);
1636 #endif
1637 		bias = 2 * sizeof(struct direntry);
1638 		if (offset < bias) {
1639 			for (n = (int)offset / sizeof(struct direntry);
1640 			     n < 2; n++) {
1641 				if (FAT32(pmp))
1642 					fileno = (uint64_t)cntobn(pmp,
1643 								 pmp->pm_rootdirblk)
1644 							  * dirsperblk;
1645 				else
1646 					fileno = 1;
1647 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1648 					dirbuf.d_fileno =
1649 					    msdosfs_fileno_map(pmp->pm_mountp,
1650 					    fileno);
1651 				} else {
1652 
1653 					dirbuf.d_fileno = (uint32_t)fileno;
1654 				}
1655 				dirbuf.d_type = DT_DIR;
1656 				switch (n) {
1657 				case 0:
1658 					dirbuf.d_namlen = 1;
1659 					strcpy(dirbuf.d_name, ".");
1660 					break;
1661 				case 1:
1662 					dirbuf.d_namlen = 2;
1663 					strcpy(dirbuf.d_name, "..");
1664 					break;
1665 				}
1666 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1667 				if (uio->uio_resid < dirbuf.d_reclen)
1668 					goto out;
1669 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1670 				if (error)
1671 					goto out;
1672 				offset += sizeof(struct direntry);
1673 				off = offset;
1674 				if (cookies) {
1675 					*cookies++ = offset;
1676 					if (--ncookies <= 0)
1677 						goto out;
1678 				}
1679 			}
1680 		}
1681 	}
1682 
1683 	mbnambuf_init(&nb);
1684 	off = offset;
1685 	while (uio->uio_resid > 0) {
1686 		lbn = de_cluster(pmp, offset - bias);
1687 		on = (offset - bias) & pmp->pm_crbomask;
1688 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1689 		diff = dep->de_FileSize - (offset - bias);
1690 		if (diff <= 0)
1691 			break;
1692 		n = min(n, diff);
1693 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1694 		if (error)
1695 			break;
1696 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1697 		if (error) {
1698 			brelse(bp);
1699 			return (error);
1700 		}
1701 		n = min(n, blsize - bp->b_resid);
1702 		if (n == 0) {
1703 			brelse(bp);
1704 			return (EIO);
1705 		}
1706 
1707 		/*
1708 		 * Convert from dos directory entries to fs-independent
1709 		 * directory entries.
1710 		 */
1711 		for (dentp = (struct direntry *)(bp->b_data + on);
1712 		     (char *)dentp < bp->b_data + on + n;
1713 		     dentp++, offset += sizeof(struct direntry)) {
1714 #if 0
1715 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1716 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1717 #endif
1718 			/*
1719 			 * If this is an unused entry, we can stop.
1720 			 */
1721 			if (dentp->deName[0] == SLOT_EMPTY) {
1722 				brelse(bp);
1723 				goto out;
1724 			}
1725 			/*
1726 			 * Skip deleted entries.
1727 			 */
1728 			if (dentp->deName[0] == SLOT_DELETED) {
1729 				chksum = -1;
1730 				mbnambuf_init(&nb);
1731 				continue;
1732 			}
1733 
1734 			/*
1735 			 * Handle Win95 long directory entries
1736 			 */
1737 			if (dentp->deAttributes == ATTR_WIN95) {
1738 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1739 					continue;
1740 				chksum = win2unixfn(&nb,
1741 				    (struct winentry *)dentp, chksum, pmp);
1742 				continue;
1743 			}
1744 
1745 			/*
1746 			 * Skip volume labels
1747 			 */
1748 			if (dentp->deAttributes & ATTR_VOLUME) {
1749 				chksum = -1;
1750 				mbnambuf_init(&nb);
1751 				continue;
1752 			}
1753 			/*
1754 			 * This computation of d_fileno must match
1755 			 * the computation of va_fileid in
1756 			 * msdosfs_getattr.
1757 			 */
1758 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1759 				fileno = getushort(dentp->deStartCluster);
1760 				if (FAT32(pmp))
1761 					fileno |= getushort(dentp->deHighClust) << 16;
1762 				/* if this is the root directory */
1763 				if (fileno == MSDOSFSROOT)
1764 					if (FAT32(pmp))
1765 						fileno = (uint64_t)cntobn(pmp,
1766 								pmp->pm_rootdirblk)
1767 							 * dirsperblk;
1768 					else
1769 						fileno = 1;
1770 				else
1771 					fileno = (uint64_t)cntobn(pmp, fileno) *
1772 					    dirsperblk;
1773 				dirbuf.d_type = DT_DIR;
1774 			} else {
1775 				fileno = (uoff_t)offset /
1776 				    sizeof(struct direntry);
1777 				dirbuf.d_type = DT_REG;
1778 			}
1779 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1780 				dirbuf.d_fileno =
1781 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1782 			} else
1783 				dirbuf.d_fileno = (uint32_t)fileno;
1784 
1785 			if (chksum != winChksum(dentp->deName)) {
1786 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1787 				    (u_char *)dirbuf.d_name,
1788 				    dentp->deLowerCase |
1789 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1790 					(LCASE_BASE | LCASE_EXT) : 0),
1791 				    pmp);
1792 				mbnambuf_init(&nb);
1793 			} else
1794 				mbnambuf_flush(&nb, &dirbuf);
1795 			chksum = -1;
1796 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1797 			if (uio->uio_resid < dirbuf.d_reclen) {
1798 				brelse(bp);
1799 				goto out;
1800 			}
1801 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1802 			if (error) {
1803 				brelse(bp);
1804 				goto out;
1805 			}
1806 			if (cookies) {
1807 				*cookies++ = offset + sizeof(struct direntry);
1808 				if (--ncookies <= 0) {
1809 					brelse(bp);
1810 					goto out;
1811 				}
1812 			}
1813 			off = offset + sizeof(struct direntry);
1814 		}
1815 		brelse(bp);
1816 	}
1817 out:
1818 	/* Subtract unused cookies */
1819 	if (ap->a_ncookies)
1820 		*ap->a_ncookies -= ncookies;
1821 
1822 	uio->uio_offset = off;
1823 
1824 	/*
1825 	 * Set the eofflag (NFS uses it)
1826 	 */
1827 	if (ap->a_eofflag) {
1828 		if (dep->de_FileSize - (offset - bias) <= 0)
1829 			*ap->a_eofflag = 1;
1830 		else
1831 			*ap->a_eofflag = 0;
1832 	}
1833 	return (error);
1834 }
1835 
1836 /*-
1837  * a_vp   - pointer to the file's vnode
1838  * a_bn   - logical block number within the file (cluster number for us)
1839  * a_bop  - where to return the bufobj of the special file containing the fs
1840  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1841  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1842  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1843  *          blocks whose physical blocks (together with a_bn's physical block)
1844  *          are contiguous.
1845  * a_runb - where to return the "run before" a_bn.
1846  */
1847 static int
1848 msdosfs_bmap(ap)
1849 	struct vop_bmap_args /* {
1850 		struct vnode *a_vp;
1851 		daddr_t a_bn;
1852 		struct bufobj **a_bop;
1853 		daddr_t *a_bnp;
1854 		int *a_runp;
1855 		int *a_runb;
1856 	} */ *ap;
1857 {
1858 	struct denode *dep;
1859 	struct mount *mp;
1860 	struct msdosfsmount *pmp;
1861 	struct vnode *vp;
1862 	daddr_t runbn;
1863 	u_long cn;
1864 	int bnpercn, error, maxio, maxrun, run;
1865 
1866 	vp = ap->a_vp;
1867 	dep = VTODE(vp);
1868 	pmp = dep->de_pmp;
1869 	if (ap->a_bop != NULL)
1870 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1871 	if (ap->a_bnp == NULL)
1872 		return (0);
1873 	if (ap->a_runp != NULL)
1874 		*ap->a_runp = 0;
1875 	if (ap->a_runb != NULL)
1876 		*ap->a_runb = 0;
1877 	cn = ap->a_bn;
1878 	if (cn != ap->a_bn)
1879 		return (EFBIG);
1880 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1881 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1882 		return (error);
1883 
1884 	mp = vp->v_mount;
1885 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1886 	bnpercn = de_cn2bn(pmp, 1);
1887 	if (ap->a_runp != NULL) {
1888 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1889 		for (run = 1; run <= maxrun; run++) {
1890 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1891 			    runbn != *ap->a_bnp + run * bnpercn)
1892 				break;
1893 		}
1894 		*ap->a_runp = run - 1;
1895 	}
1896 	if (ap->a_runb != NULL) {
1897 		maxrun = ulmin(maxio - 1, cn);
1898 		for (run = 1; run < maxrun; run++) {
1899 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1900 			    runbn != *ap->a_bnp - run * bnpercn)
1901 				break;
1902 		}
1903 		*ap->a_runb = run - 1;
1904 	}
1905 	return (0);
1906 }
1907 
1908 static int
1909 msdosfs_strategy(ap)
1910 	struct vop_strategy_args /* {
1911 		struct vnode *a_vp;
1912 		struct buf *a_bp;
1913 	} */ *ap;
1914 {
1915 	struct buf *bp = ap->a_bp;
1916 	struct denode *dep = VTODE(ap->a_vp);
1917 	struct bufobj *bo;
1918 	int error = 0;
1919 	daddr_t blkno;
1920 
1921 	/*
1922 	 * If we don't already know the filesystem relative block number
1923 	 * then get it using pcbmap().  If pcbmap() returns the block
1924 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1925 	 * don't allow files with holes, so we shouldn't ever see this.
1926 	 */
1927 	if (bp->b_blkno == bp->b_lblkno) {
1928 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1929 		bp->b_blkno = blkno;
1930 		if (error) {
1931 			bp->b_error = error;
1932 			bp->b_ioflags |= BIO_ERROR;
1933 			bufdone(bp);
1934 			return (0);
1935 		}
1936 		if ((long)bp->b_blkno == -1)
1937 			vfs_bio_clrbuf(bp);
1938 	}
1939 	if (bp->b_blkno == -1) {
1940 		bufdone(bp);
1941 		return (0);
1942 	}
1943 	/*
1944 	 * Read/write the block from/to the disk that contains the desired
1945 	 * file block.
1946 	 */
1947 	bp->b_iooffset = dbtob(bp->b_blkno);
1948 	bo = dep->de_pmp->pm_bo;
1949 	BO_STRATEGY(bo, bp);
1950 	return (0);
1951 }
1952 
1953 static int
1954 msdosfs_print(ap)
1955 	struct vop_print_args /* {
1956 		struct vnode *vp;
1957 	} */ *ap;
1958 {
1959 	struct denode *dep = VTODE(ap->a_vp);
1960 
1961 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1962 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1963 	printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1964 	return (0);
1965 }
1966 
1967 static int
1968 msdosfs_pathconf(ap)
1969 	struct vop_pathconf_args /* {
1970 		struct vnode *a_vp;
1971 		int a_name;
1972 		int *a_retval;
1973 	} */ *ap;
1974 {
1975 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1976 
1977 	switch (ap->a_name) {
1978 	case _PC_LINK_MAX:
1979 		*ap->a_retval = 1;
1980 		return (0);
1981 	case _PC_NAME_MAX:
1982 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1983 		return (0);
1984 	case _PC_PATH_MAX:
1985 		*ap->a_retval = PATH_MAX;
1986 		return (0);
1987 	case _PC_CHOWN_RESTRICTED:
1988 		*ap->a_retval = 1;
1989 		return (0);
1990 	case _PC_NO_TRUNC:
1991 		*ap->a_retval = 0;
1992 		return (0);
1993 	default:
1994 		return (EINVAL);
1995 	}
1996 	/* NOTREACHED */
1997 }
1998 
1999 static int
2000 msdosfs_vptofh(ap)
2001 	struct vop_vptofh_args /* {
2002 		struct vnode *a_vp;
2003 		struct fid *a_fhp;
2004 	} */ *ap;
2005 {
2006 	struct denode *dep;
2007 	struct defid *defhp;
2008 
2009 	dep = VTODE(ap->a_vp);
2010 	defhp = (struct defid *)ap->a_fhp;
2011 	defhp->defid_len = sizeof(struct defid);
2012 	defhp->defid_dirclust = dep->de_dirclust;
2013 	defhp->defid_dirofs = dep->de_diroffset;
2014 	/* defhp->defid_gen = dep->de_gen; */
2015 	return (0);
2016 }
2017 
2018 /* Global vfs data structures for msdosfs */
2019 struct vop_vector msdosfs_vnodeops = {
2020 	.vop_default =		&default_vnodeops,
2021 
2022 	.vop_access =		msdosfs_access,
2023 	.vop_bmap =		msdosfs_bmap,
2024 	.vop_cachedlookup =	msdosfs_lookup,
2025 	.vop_open =		msdosfs_open,
2026 	.vop_close =		msdosfs_close,
2027 	.vop_create =		msdosfs_create,
2028 	.vop_fsync =		msdosfs_fsync,
2029 	.vop_getattr =		msdosfs_getattr,
2030 	.vop_inactive =		msdosfs_inactive,
2031 	.vop_link =		msdosfs_link,
2032 	.vop_lookup =		vfs_cache_lookup,
2033 	.vop_mkdir =		msdosfs_mkdir,
2034 	.vop_mknod =		msdosfs_mknod,
2035 	.vop_pathconf =		msdosfs_pathconf,
2036 	.vop_print =		msdosfs_print,
2037 	.vop_read =		msdosfs_read,
2038 	.vop_readdir =		msdosfs_readdir,
2039 	.vop_reclaim =		msdosfs_reclaim,
2040 	.vop_remove =		msdosfs_remove,
2041 	.vop_rename =		msdosfs_rename,
2042 	.vop_rmdir =		msdosfs_rmdir,
2043 	.vop_setattr =		msdosfs_setattr,
2044 	.vop_strategy =		msdosfs_strategy,
2045 	.vop_symlink =		msdosfs_symlink,
2046 	.vop_write =		msdosfs_write,
2047 	.vop_vptofh =		msdosfs_vptofh,
2048 };
2049