xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
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 		if (vap->va_vaflags & VA_UTIMES_NULL) {
505 			error = VOP_ACCESS(vp, VADMIN, cred, td);
506 			if (error)
507 				error = VOP_ACCESS(vp, VWRITE, cred, td);
508 		} else
509 			error = VOP_ACCESS(vp, VADMIN, cred, td);
510 		if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
511 		    vap->va_atime.tv_sec != VNOVAL) {
512 			dep->de_flag &= ~DE_ACCESS;
513 			timespec2fattime(&vap->va_atime, 0,
514 			    &dep->de_ADate, NULL, NULL);
515 		}
516 		if (vap->va_mtime.tv_sec != VNOVAL) {
517 			dep->de_flag &= ~DE_UPDATE;
518 			timespec2fattime(&vap->va_mtime, 0,
519 			    &dep->de_MDate, &dep->de_MTime, NULL);
520 		}
521 		/*
522 		 * We don't set the archive bit when modifying the time of
523 		 * a directory to emulate the Windows/DOS behavior.
524 		 */
525 		if (vp->v_type != VDIR)
526 			dep->de_Attributes |= ATTR_ARCHIVE;
527 		dep->de_flag |= DE_MODIFIED;
528 	}
529 	/*
530 	 * DOS files only have the ability to have their writability
531 	 * attribute set, so we use the owner write bit to set the readonly
532 	 * attribute.
533 	 */
534 	if (vap->va_mode != (mode_t)VNOVAL) {
535 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
536 			return (EROFS);
537 		if (cred->cr_uid != pmp->pm_uid) {
538 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
539 			if (error)
540 				return (error);
541 		}
542 		if (vp->v_type != VDIR) {
543 			/* We ignore the read and execute bits. */
544 			if (vap->va_mode & VWRITE)
545 				dep->de_Attributes &= ~ATTR_READONLY;
546 			else
547 				dep->de_Attributes |= ATTR_READONLY;
548 			dep->de_Attributes |= ATTR_ARCHIVE;
549 			dep->de_flag |= DE_MODIFIED;
550 		}
551 	}
552 	return (deupdat(dep, 0));
553 }
554 
555 static int
556 msdosfs_read(ap)
557 	struct vop_read_args /* {
558 		struct vnode *a_vp;
559 		struct uio *a_uio;
560 		int a_ioflag;
561 		struct ucred *a_cred;
562 	} */ *ap;
563 {
564 	int error = 0;
565 	int blsize;
566 	int isadir;
567 	ssize_t orig_resid;
568 	u_int n;
569 	u_long diff;
570 	u_long on;
571 	daddr_t lbn;
572 	daddr_t rablock;
573 	int rasize;
574 	int seqcount;
575 	struct buf *bp;
576 	struct vnode *vp = ap->a_vp;
577 	struct denode *dep = VTODE(vp);
578 	struct msdosfsmount *pmp = dep->de_pmp;
579 	struct uio *uio = ap->a_uio;
580 
581 	/*
582 	 * If they didn't ask for any data, then we are done.
583 	 */
584 	orig_resid = uio->uio_resid;
585 	if (orig_resid == 0)
586 		return (0);
587 
588 	/*
589 	 * The caller is supposed to ensure that
590 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
591 	 * We don't need to check for large offsets as in ffs because
592 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
593 	 * offsets cannot cause overflow even in theory.
594 	 */
595 
596 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
597 
598 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
599 	do {
600 		if (uio->uio_offset >= dep->de_FileSize)
601 			break;
602 		lbn = de_cluster(pmp, uio->uio_offset);
603 		rablock = lbn + 1;
604 		blsize = pmp->pm_bpcluster;
605 		on = uio->uio_offset & pmp->pm_crbomask;
606 		/*
607 		 * If we are operating on a directory file then be sure to
608 		 * do i/o with the vnode for the filesystem instead of the
609 		 * vnode for the directory.
610 		 */
611 		if (isadir) {
612 			/* convert cluster # to block # */
613 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
614 			if (error == E2BIG) {
615 				error = EINVAL;
616 				break;
617 			} else if (error)
618 				break;
619 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
620 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
621 			error = bread(vp, lbn, blsize, NOCRED, &bp);
622 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
623 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
624 			    NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
625 		} else if (seqcount > 1) {
626 			rasize = blsize;
627 			error = breadn(vp, lbn,
628 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
629 		} else {
630 			error = bread(vp, lbn, blsize, NOCRED, &bp);
631 		}
632 		if (error) {
633 			brelse(bp);
634 			break;
635 		}
636 		diff = pmp->pm_bpcluster - on;
637 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
638 		diff = dep->de_FileSize - uio->uio_offset;
639 		if (diff < n)
640 			n = diff;
641 		diff = blsize - bp->b_resid;
642 		if (diff < n)
643 			n = diff;
644 		error = uiomove(bp->b_data + on, (int) n, uio);
645 		brelse(bp);
646 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
647 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
648 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
649 		dep->de_flag |= DE_ACCESS;
650 	return (error);
651 }
652 
653 /*
654  * Write data to a file or directory.
655  */
656 static int
657 msdosfs_write(ap)
658 	struct vop_write_args /* {
659 		struct vnode *a_vp;
660 		struct uio *a_uio;
661 		int a_ioflag;
662 		struct ucred *a_cred;
663 	} */ *ap;
664 {
665 	int n;
666 	int croffset;
667 	ssize_t resid;
668 	u_long osize;
669 	int error = 0;
670 	u_long count;
671 	int seqcount;
672 	daddr_t bn, lastcn;
673 	struct buf *bp;
674 	int ioflag = ap->a_ioflag;
675 	struct uio *uio = ap->a_uio;
676 	struct vnode *vp = ap->a_vp;
677 	struct vnode *thisvp;
678 	struct denode *dep = VTODE(vp);
679 	struct msdosfsmount *pmp = dep->de_pmp;
680 	struct ucred *cred = ap->a_cred;
681 
682 #ifdef MSDOSFS_DEBUG
683 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
684 	    vp, uio, ioflag, cred);
685 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
686 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
687 #endif
688 
689 	switch (vp->v_type) {
690 	case VREG:
691 		if (ioflag & IO_APPEND)
692 			uio->uio_offset = dep->de_FileSize;
693 		thisvp = vp;
694 		break;
695 	case VDIR:
696 		return EISDIR;
697 	default:
698 		panic("msdosfs_write(): bad file type");
699 	}
700 
701 	/*
702 	 * This is needed (unlike in ffs_write()) because we extend the
703 	 * file outside of the loop but we don't want to extend the file
704 	 * for writes of 0 bytes.
705 	 */
706 	if (uio->uio_resid == 0)
707 		return (0);
708 
709 	/*
710 	 * The caller is supposed to ensure that
711 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
712 	 */
713 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
714 		return (EFBIG);
715 
716 	/*
717 	 * If they've exceeded their filesize limit, tell them about it.
718 	 */
719 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
720 		return (EFBIG);
721 
722 	/*
723 	 * If the offset we are starting the write at is beyond the end of
724 	 * the file, then they've done a seek.  Unix filesystems allow
725 	 * files with holes in them, DOS doesn't so we must fill the hole
726 	 * with zeroed blocks.
727 	 */
728 	if (uio->uio_offset > dep->de_FileSize) {
729 		error = deextend(dep, uio->uio_offset, cred);
730 		if (error)
731 			return (error);
732 	}
733 
734 	/*
735 	 * Remember some values in case the write fails.
736 	 */
737 	resid = uio->uio_resid;
738 	osize = dep->de_FileSize;
739 
740 	/*
741 	 * If we write beyond the end of the file, extend it to its ultimate
742 	 * size ahead of the time to hopefully get a contiguous area.
743 	 */
744 	if (uio->uio_offset + resid > osize) {
745 		count = de_clcount(pmp, uio->uio_offset + resid) -
746 			de_clcount(pmp, osize);
747 		error = extendfile(dep, count, NULL, NULL, 0);
748 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
749 			goto errexit;
750 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
751 	} else
752 		lastcn = de_clcount(pmp, osize) - 1;
753 
754 	seqcount = ioflag >> IO_SEQSHIFT;
755 	do {
756 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
757 			error = ENOSPC;
758 			break;
759 		}
760 
761 		croffset = uio->uio_offset & pmp->pm_crbomask;
762 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
763 		if (uio->uio_offset + n > dep->de_FileSize) {
764 			dep->de_FileSize = uio->uio_offset + n;
765 			/* The object size needs to be set before buffer is allocated */
766 			vnode_pager_setsize(vp, dep->de_FileSize);
767 		}
768 
769 		bn = de_cluster(pmp, uio->uio_offset);
770 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
771 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
772 			> de_cluster(pmp, uio->uio_offset)
773 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
774 			/*
775 			 * If either the whole cluster gets written,
776 			 * or we write the cluster from its start beyond EOF,
777 			 * then no need to read data from disk.
778 			 */
779 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
780 			vfs_bio_clrbuf(bp);
781 			/*
782 			 * Do the bmap now, since pcbmap needs buffers
783 			 * for the fat table. (see msdosfs_strategy)
784 			 */
785 			if (bp->b_blkno == bp->b_lblkno) {
786 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
787 				if (error)
788 					bp->b_blkno = -1;
789 				else
790 					bp->b_blkno = bn;
791 			}
792 			if (bp->b_blkno == -1) {
793 				brelse(bp);
794 				if (!error)
795 					error = EIO;		/* XXX */
796 				break;
797 			}
798 		} else {
799 			/*
800 			 * The block we need to write into exists, so read it in.
801 			 */
802 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
803 			if (error) {
804 				brelse(bp);
805 				break;
806 			}
807 		}
808 
809 		/*
810 		 * Should these vnode_pager_* functions be done on dir
811 		 * files?
812 		 */
813 
814 		/*
815 		 * Copy the data from user space into the buf header.
816 		 */
817 		error = uiomove(bp->b_data + croffset, n, uio);
818 		if (error) {
819 			brelse(bp);
820 			break;
821 		}
822 
823 		/* Prepare for clustered writes in some else clauses. */
824 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
825 			bp->b_flags |= B_CLUSTEROK;
826 
827 		/*
828 		 * If IO_SYNC, then each buffer is written synchronously.
829 		 * Otherwise, if we have a severe page deficiency then
830 		 * write the buffer asynchronously.  Otherwise, if on a
831 		 * cluster boundary then write the buffer asynchronously,
832 		 * combining it with contiguous clusters if permitted and
833 		 * possible, since we don't expect more writes into this
834 		 * buffer soon.  Otherwise, do a delayed write because we
835 		 * expect more writes into this buffer soon.
836 		 */
837 		if (ioflag & IO_SYNC)
838 			(void)bwrite(bp);
839 		else if (vm_page_count_severe() || buf_dirty_count_severe())
840 			bawrite(bp);
841 		else if (n + croffset == pmp->pm_bpcluster) {
842 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
843 				cluster_write(vp, bp, dep->de_FileSize,
844 				    seqcount, 0);
845 			else
846 				bawrite(bp);
847 		} else
848 			bdwrite(bp);
849 		dep->de_flag |= DE_UPDATE;
850 	} while (error == 0 && uio->uio_resid > 0);
851 
852 	/*
853 	 * If the write failed and they want us to, truncate the file back
854 	 * to the size it was before the write was attempted.
855 	 */
856 errexit:
857 	if (error) {
858 		if (ioflag & IO_UNIT) {
859 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
860 			uio->uio_offset -= resid - uio->uio_resid;
861 			uio->uio_resid = resid;
862 		} else {
863 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
864 			if (uio->uio_resid != resid)
865 				error = 0;
866 		}
867 	} else if (ioflag & IO_SYNC)
868 		error = deupdat(dep, 1);
869 	return (error);
870 }
871 
872 /*
873  * Flush the blocks of a file to disk.
874  */
875 static int
876 msdosfs_fsync(ap)
877 	struct vop_fsync_args /* {
878 		struct vnode *a_vp;
879 		struct ucred *a_cred;
880 		int a_waitfor;
881 		struct thread *a_td;
882 	} */ *ap;
883 {
884 	struct vnode *devvp;
885 	int allerror, error;
886 
887 	vop_stdfsync(ap);
888 
889 	/*
890 	* If the syncing request comes from fsync(2), sync the entire
891 	* FAT and any other metadata that happens to be on devvp.  We
892 	* need this mainly for the FAT.  We write the FAT sloppily, and
893 	* syncing it all now is the best we can easily do to get all
894 	* directory entries associated with the file (not just the file)
895 	* fully synced.  The other metadata includes critical metadata
896 	* for all directory entries, but only in the MNT_ASYNC case.  We
897 	* will soon sync all metadata in the file's directory entry.
898 	* Non-critical metadata for associated directory entries only
899 	* gets synced accidentally, as in most file systems.
900 	*/
901 	if (ap->a_waitfor == MNT_WAIT) {
902 		devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
903 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
904 		allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
905 		VOP_UNLOCK(devvp, 0);
906 	} else
907 		allerror = 0;
908 
909 	error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
910 	if (allerror == 0)
911 		allerror = error;
912 	return (allerror);
913 }
914 
915 static int
916 msdosfs_remove(ap)
917 	struct vop_remove_args /* {
918 		struct vnode *a_dvp;
919 		struct vnode *a_vp;
920 		struct componentname *a_cnp;
921 	} */ *ap;
922 {
923 	struct denode *dep = VTODE(ap->a_vp);
924 	struct denode *ddep = VTODE(ap->a_dvp);
925 	int error;
926 
927 	if (ap->a_vp->v_type == VDIR)
928 		error = EPERM;
929 	else
930 		error = removede(ddep, dep);
931 #ifdef MSDOSFS_DEBUG
932 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
933 #endif
934 	return (error);
935 }
936 
937 /*
938  * DOS filesystems don't know what links are.
939  */
940 static int
941 msdosfs_link(ap)
942 	struct vop_link_args /* {
943 		struct vnode *a_tdvp;
944 		struct vnode *a_vp;
945 		struct componentname *a_cnp;
946 	} */ *ap;
947 {
948 	return (EOPNOTSUPP);
949 }
950 
951 /*
952  * Renames on files require moving the denode to a new hash queue since the
953  * denode's location is used to compute which hash queue to put the file
954  * in. Unless it is a rename in place.  For example "mv a b".
955  *
956  * What follows is the basic algorithm:
957  *
958  * if (file move) {
959  *	if (dest file exists) {
960  *		remove dest file
961  *	}
962  *	if (dest and src in same directory) {
963  *		rewrite name in existing directory slot
964  *	} else {
965  *		write new entry in dest directory
966  *		update offset and dirclust in denode
967  *		move denode to new hash chain
968  *		clear old directory entry
969  *	}
970  * } else {
971  *	directory move
972  *	if (dest directory exists) {
973  *		if (dest is not empty) {
974  *			return ENOTEMPTY
975  *		}
976  *		remove dest directory
977  *	}
978  *	if (dest and src in same directory) {
979  *		rewrite name in existing entry
980  *	} else {
981  *		be sure dest is not a child of src directory
982  *		write entry in dest directory
983  *		update "." and ".." in moved directory
984  *		clear old directory entry for moved directory
985  *	}
986  * }
987  *
988  * On entry:
989  *	source's parent directory is unlocked
990  *	source file or directory is unlocked
991  *	destination's parent directory is locked
992  *	destination file or directory is locked if it exists
993  *
994  * On exit:
995  *	all denodes should be released
996  */
997 static int
998 msdosfs_rename(ap)
999 	struct vop_rename_args /* {
1000 		struct vnode *a_fdvp;
1001 		struct vnode *a_fvp;
1002 		struct componentname *a_fcnp;
1003 		struct vnode *a_tdvp;
1004 		struct vnode *a_tvp;
1005 		struct componentname *a_tcnp;
1006 	} */ *ap;
1007 {
1008 	struct vnode *tdvp = ap->a_tdvp;
1009 	struct vnode *fvp = ap->a_fvp;
1010 	struct vnode *fdvp = ap->a_fdvp;
1011 	struct vnode *tvp = ap->a_tvp;
1012 	struct componentname *tcnp = ap->a_tcnp;
1013 	struct componentname *fcnp = ap->a_fcnp;
1014 	struct denode *ip, *xp, *dp, *zp;
1015 	u_char toname[12], oldname[11];
1016 	u_long from_diroffset, to_diroffset;
1017 	u_char to_count;
1018 	int doingdirectory = 0, newparent = 0;
1019 	int error;
1020 	u_long cn, pcl;
1021 	daddr_t bn;
1022 	struct denode *fddep;	/* from file's parent directory	 */
1023 	struct msdosfsmount *pmp;
1024 	struct direntry *dotdotp;
1025 	struct buf *bp;
1026 
1027 	fddep = VTODE(ap->a_fdvp);
1028 	pmp = fddep->de_pmp;
1029 
1030 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1031 
1032 #ifdef DIAGNOSTIC
1033 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1034 	    (fcnp->cn_flags & HASBUF) == 0)
1035 		panic("msdosfs_rename: no name");
1036 #endif
1037 	/*
1038 	 * Check for cross-device rename.
1039 	 */
1040 	if (fvp->v_mount != tdvp->v_mount ||
1041 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1042 		error = EXDEV;
1043 abortit:
1044 		if (tdvp == tvp)
1045 			vrele(tdvp);
1046 		else
1047 			vput(tdvp);
1048 		if (tvp)
1049 			vput(tvp);
1050 		vrele(fdvp);
1051 		vrele(fvp);
1052 		return (error);
1053 	}
1054 
1055 	/*
1056 	 * If source and dest are the same, do nothing.
1057 	 */
1058 	if (tvp == fvp) {
1059 		error = 0;
1060 		goto abortit;
1061 	}
1062 
1063 	error = vn_lock(fvp, LK_EXCLUSIVE);
1064 	if (error)
1065 		goto abortit;
1066 	dp = VTODE(fdvp);
1067 	ip = VTODE(fvp);
1068 
1069 	/*
1070 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1071 	 * leads to a crippled directory tree.  It's pretty tough to do a
1072 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1073 	 * doesn't work if the ".." entry is missing.
1074 	 */
1075 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1076 		/*
1077 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1078 		 */
1079 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1080 		    dp == ip ||
1081 		    (fcnp->cn_flags & ISDOTDOT) ||
1082 		    (tcnp->cn_flags & ISDOTDOT) ||
1083 		    (ip->de_flag & DE_RENAME)) {
1084 			VOP_UNLOCK(fvp, 0);
1085 			error = EINVAL;
1086 			goto abortit;
1087 		}
1088 		ip->de_flag |= DE_RENAME;
1089 		doingdirectory++;
1090 	}
1091 
1092 	/*
1093 	 * When the target exists, both the directory
1094 	 * and target vnodes are returned locked.
1095 	 */
1096 	dp = VTODE(tdvp);
1097 	xp = tvp ? VTODE(tvp) : NULL;
1098 	/*
1099 	 * Remember direntry place to use for destination
1100 	 */
1101 	to_diroffset = dp->de_fndoffset;
1102 	to_count = dp->de_fndcnt;
1103 
1104 	/*
1105 	 * If ".." must be changed (ie the directory gets a new
1106 	 * parent) then the source directory must not be in the
1107 	 * directory hierarchy above the target, as this would
1108 	 * orphan everything below the source directory. Also
1109 	 * the user must have write permission in the source so
1110 	 * as to be able to change "..". We must repeat the call
1111 	 * to namei, as the parent directory is unlocked by the
1112 	 * call to doscheckpath().
1113 	 */
1114 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1115 	VOP_UNLOCK(fvp, 0);
1116 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1117 		newparent = 1;
1118 	if (doingdirectory && newparent) {
1119 		if (error)	/* write access check above */
1120 			goto bad;
1121 		if (xp != NULL)
1122 			vput(tvp);
1123 		/*
1124 		 * doscheckpath() vput()'s dp,
1125 		 * so we have to do a relookup afterwards
1126 		 */
1127 		error = doscheckpath(ip, dp);
1128 		if (error)
1129 			goto out;
1130 		if ((tcnp->cn_flags & SAVESTART) == 0)
1131 			panic("msdosfs_rename: lost to startdir");
1132 		error = relookup(tdvp, &tvp, tcnp);
1133 		if (error)
1134 			goto out;
1135 		dp = VTODE(tdvp);
1136 		xp = tvp ? VTODE(tvp) : NULL;
1137 	}
1138 
1139 	if (xp != NULL) {
1140 		/*
1141 		 * Target must be empty if a directory and have no links
1142 		 * to it. Also, ensure source and target are compatible
1143 		 * (both directories, or both not directories).
1144 		 */
1145 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1146 			if (!dosdirempty(xp)) {
1147 				error = ENOTEMPTY;
1148 				goto bad;
1149 			}
1150 			if (!doingdirectory) {
1151 				error = ENOTDIR;
1152 				goto bad;
1153 			}
1154 			cache_purge(tdvp);
1155 		} else if (doingdirectory) {
1156 			error = EISDIR;
1157 			goto bad;
1158 		}
1159 		error = removede(dp, xp);
1160 		if (error)
1161 			goto bad;
1162 		vput(tvp);
1163 		xp = NULL;
1164 	}
1165 
1166 	/*
1167 	 * Convert the filename in tcnp into a dos filename. We copy this
1168 	 * into the denode and directory entry for the destination
1169 	 * file/directory.
1170 	 */
1171 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1172 	if (error)
1173 		goto abortit;
1174 
1175 	/*
1176 	 * Since from wasn't locked at various places above,
1177 	 * have to do a relookup here.
1178 	 */
1179 	fcnp->cn_flags &= ~MODMASK;
1180 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1181 	if ((fcnp->cn_flags & SAVESTART) == 0)
1182 		panic("msdosfs_rename: lost from startdir");
1183 	if (!newparent)
1184 		VOP_UNLOCK(tdvp, 0);
1185 	if (relookup(fdvp, &fvp, fcnp) == 0)
1186 		vrele(fdvp);
1187 	if (fvp == NULL) {
1188 		/*
1189 		 * From name has disappeared.
1190 		 */
1191 		if (doingdirectory)
1192 			panic("rename: lost dir entry");
1193 		if (newparent)
1194 			VOP_UNLOCK(tdvp, 0);
1195 		vrele(tdvp);
1196 		vrele(ap->a_fvp);
1197 		return 0;
1198 	}
1199 	xp = VTODE(fvp);
1200 	zp = VTODE(fdvp);
1201 	from_diroffset = zp->de_fndoffset;
1202 
1203 	/*
1204 	 * Ensure that the directory entry still exists and has not
1205 	 * changed till now. If the source is a file the entry may
1206 	 * have been unlinked or renamed. In either case there is
1207 	 * no further work to be done. If the source is a directory
1208 	 * then it cannot have been rmdir'ed or renamed; this is
1209 	 * prohibited by the DE_RENAME flag.
1210 	 */
1211 	if (xp != ip) {
1212 		if (doingdirectory)
1213 			panic("rename: lost dir entry");
1214 		VOP_UNLOCK(fvp, 0);
1215 		if (newparent)
1216 			VOP_UNLOCK(fdvp, 0);
1217 		vrele(ap->a_fvp);
1218 		xp = NULL;
1219 	} else {
1220 		vrele(fvp);
1221 		xp = NULL;
1222 
1223 		/*
1224 		 * First write a new entry in the destination
1225 		 * directory and mark the entry in the source directory
1226 		 * as deleted.  Then move the denode to the correct hash
1227 		 * chain for its new location in the filesystem.  And, if
1228 		 * we moved a directory, then update its .. entry to point
1229 		 * to the new parent directory.
1230 		 */
1231 		bcopy(ip->de_Name, oldname, 11);
1232 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1233 		dp->de_fndoffset = to_diroffset;
1234 		dp->de_fndcnt = to_count;
1235 		error = createde(ip, dp, (struct denode **)0, tcnp);
1236 		if (error) {
1237 			bcopy(oldname, ip->de_Name, 11);
1238 			if (newparent)
1239 				VOP_UNLOCK(fdvp, 0);
1240 			VOP_UNLOCK(fvp, 0);
1241 			goto bad;
1242 		}
1243 		/*
1244 		 * If ip is for a directory, then its name should always
1245 		 * be "." since it is for the directory entry in the
1246 		 * directory itself (msdosfs_lookup() always translates
1247 		 * to the "." entry so as to get a unique denode, except
1248 		 * for the root directory there are different
1249 		 * complications).  However, we just corrupted its name
1250 		 * to pass the correct name to createde().  Undo this.
1251 		 */
1252 		if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1253 			bcopy(oldname, ip->de_Name, 11);
1254 		ip->de_refcnt++;
1255 		zp->de_fndoffset = from_diroffset;
1256 		error = removede(zp, ip);
1257 		if (error) {
1258 			/* XXX should downgrade to ro here, fs is corrupt */
1259 			if (newparent)
1260 				VOP_UNLOCK(fdvp, 0);
1261 			VOP_UNLOCK(fvp, 0);
1262 			goto bad;
1263 		}
1264 		if (!doingdirectory) {
1265 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1266 				       &ip->de_dirclust, 0);
1267 			if (error) {
1268 				/* XXX should downgrade to ro here, fs is corrupt */
1269 				if (newparent)
1270 					VOP_UNLOCK(fdvp, 0);
1271 				VOP_UNLOCK(fvp, 0);
1272 				goto bad;
1273 			}
1274 			if (ip->de_dirclust == MSDOSFSROOT)
1275 				ip->de_diroffset = to_diroffset;
1276 			else
1277 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1278 		}
1279 		reinsert(ip);
1280 		if (newparent)
1281 			VOP_UNLOCK(fdvp, 0);
1282 	}
1283 
1284 	/*
1285 	 * If we moved a directory to a new parent directory, then we must
1286 	 * fixup the ".." entry in the moved directory.
1287 	 */
1288 	if (doingdirectory && newparent) {
1289 		cn = ip->de_StartCluster;
1290 		if (cn == MSDOSFSROOT) {
1291 			/* this should never happen */
1292 			panic("msdosfs_rename(): updating .. in root directory?");
1293 		} else
1294 			bn = cntobn(pmp, cn);
1295 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1296 			      NOCRED, &bp);
1297 		if (error) {
1298 			/* XXX should downgrade to ro here, fs is corrupt */
1299 			brelse(bp);
1300 			VOP_UNLOCK(fvp, 0);
1301 			goto bad;
1302 		}
1303 		dotdotp = (struct direntry *)bp->b_data + 1;
1304 		pcl = dp->de_StartCluster;
1305 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1306 			pcl = MSDOSFSROOT;
1307 		putushort(dotdotp->deStartCluster, pcl);
1308 		if (FAT32(pmp))
1309 			putushort(dotdotp->deHighClust, pcl >> 16);
1310 		if (DOINGASYNC(fvp))
1311 			bdwrite(bp);
1312 		else if ((error = bwrite(bp)) != 0) {
1313 			/* XXX should downgrade to ro here, fs is corrupt */
1314 			VOP_UNLOCK(fvp, 0);
1315 			goto bad;
1316 		}
1317 	}
1318 
1319 	/*
1320 	 * The msdosfs lookup is case insensitive. Several aliases may
1321 	 * be inserted for a single directory entry. As a consequnce,
1322 	 * name cache purge done by lookup for fvp when DELETE op for
1323 	 * namei is specified, might be not enough to expunge all
1324 	 * namecache entries that were installed for this direntry.
1325 	 */
1326 	cache_purge(fvp);
1327 	VOP_UNLOCK(fvp, 0);
1328 bad:
1329 	if (xp)
1330 		vput(tvp);
1331 	vput(tdvp);
1332 out:
1333 	ip->de_flag &= ~DE_RENAME;
1334 	vrele(fdvp);
1335 	vrele(fvp);
1336 	return (error);
1337 
1338 }
1339 
1340 static struct {
1341 	struct direntry dot;
1342 	struct direntry dotdot;
1343 } dosdirtemplate = {
1344 	{	".          ",				/* the . entry */
1345 		ATTR_DIRECTORY,				/* file attribute */
1346 		0,					/* reserved */
1347 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1348 		{ 0, 0 },				/* access date */
1349 		{ 0, 0 },				/* high bits of start cluster */
1350 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1351 		{ 0, 0 },				/* startcluster */
1352 		{ 0, 0, 0, 0 }				/* filesize */
1353 	},
1354 	{	"..         ",				/* the .. entry */
1355 		ATTR_DIRECTORY,				/* file attribute */
1356 		0,					/* reserved */
1357 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1358 		{ 0, 0 },				/* access date */
1359 		{ 0, 0 },				/* high bits of start cluster */
1360 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1361 		{ 0, 0 },				/* startcluster */
1362 		{ 0, 0, 0, 0 }				/* filesize */
1363 	}
1364 };
1365 
1366 static int
1367 msdosfs_mkdir(ap)
1368 	struct vop_mkdir_args /* {
1369 		struct vnode *a_dvp;
1370 		struct vnode **a_vpp;
1371 		struvt componentname *a_cnp;
1372 		struct vattr *a_vap;
1373 	} */ *ap;
1374 {
1375 	struct componentname *cnp = ap->a_cnp;
1376 	struct denode *dep;
1377 	struct denode *pdep = VTODE(ap->a_dvp);
1378 	struct direntry *denp;
1379 	struct msdosfsmount *pmp = pdep->de_pmp;
1380 	struct buf *bp;
1381 	u_long newcluster, pcl;
1382 	int bn;
1383 	int error;
1384 	struct denode ndirent;
1385 	struct timespec ts;
1386 
1387 	/*
1388 	 * If this is the root directory and there is no space left we
1389 	 * can't do anything.  This is because the root directory can not
1390 	 * change size.
1391 	 */
1392 	if (pdep->de_StartCluster == MSDOSFSROOT
1393 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1394 		error = ENOSPC;
1395 		goto bad2;
1396 	}
1397 
1398 	/*
1399 	 * Allocate a cluster to hold the about to be created directory.
1400 	 */
1401 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1402 	if (error)
1403 		goto bad2;
1404 
1405 	bzero(&ndirent, sizeof(ndirent));
1406 	ndirent.de_pmp = pmp;
1407 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1408 	getnanotime(&ts);
1409 	DETIMES(&ndirent, &ts, &ts, &ts);
1410 
1411 	/*
1412 	 * Now fill the cluster with the "." and ".." entries. And write
1413 	 * the cluster to disk.  This way it is there for the parent
1414 	 * directory to be pointing at if there were a crash.
1415 	 */
1416 	bn = cntobn(pmp, newcluster);
1417 	/* always succeeds */
1418 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1419 	bzero(bp->b_data, pmp->pm_bpcluster);
1420 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1421 	denp = (struct direntry *)bp->b_data;
1422 	putushort(denp[0].deStartCluster, newcluster);
1423 	putushort(denp[0].deCDate, ndirent.de_CDate);
1424 	putushort(denp[0].deCTime, ndirent.de_CTime);
1425 	denp[0].deCHundredth = ndirent.de_CHun;
1426 	putushort(denp[0].deADate, ndirent.de_ADate);
1427 	putushort(denp[0].deMDate, ndirent.de_MDate);
1428 	putushort(denp[0].deMTime, ndirent.de_MTime);
1429 	pcl = pdep->de_StartCluster;
1430 	/*
1431 	 * Although the root directory has a non-magic starting cluster
1432 	 * number for FAT32, chkdsk and fsck_msdosfs still require
1433 	 * references to it in dotdot entries to be magic.
1434 	 */
1435 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1436 		pcl = MSDOSFSROOT;
1437 	putushort(denp[1].deStartCluster, pcl);
1438 	putushort(denp[1].deCDate, ndirent.de_CDate);
1439 	putushort(denp[1].deCTime, ndirent.de_CTime);
1440 	denp[1].deCHundredth = ndirent.de_CHun;
1441 	putushort(denp[1].deADate, ndirent.de_ADate);
1442 	putushort(denp[1].deMDate, ndirent.de_MDate);
1443 	putushort(denp[1].deMTime, ndirent.de_MTime);
1444 	if (FAT32(pmp)) {
1445 		putushort(denp[0].deHighClust, newcluster >> 16);
1446 		putushort(denp[1].deHighClust, pcl >> 16);
1447 	}
1448 
1449 	if (DOINGASYNC(ap->a_dvp))
1450 		bdwrite(bp);
1451 	else if ((error = bwrite(bp)) != 0)
1452 		goto bad;
1453 
1454 	/*
1455 	 * Now build up a directory entry pointing to the newly allocated
1456 	 * cluster.  This will be written to an empty slot in the parent
1457 	 * directory.
1458 	 */
1459 #ifdef DIAGNOSTIC
1460 	if ((cnp->cn_flags & HASBUF) == 0)
1461 		panic("msdosfs_mkdir: no name");
1462 #endif
1463 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1464 	if (error)
1465 		goto bad;
1466 
1467 	ndirent.de_Attributes = ATTR_DIRECTORY;
1468 	ndirent.de_LowerCase = 0;
1469 	ndirent.de_StartCluster = newcluster;
1470 	ndirent.de_FileSize = 0;
1471 	error = createde(&ndirent, pdep, &dep, cnp);
1472 	if (error)
1473 		goto bad;
1474 	*ap->a_vpp = DETOV(dep);
1475 	return (0);
1476 
1477 bad:
1478 	clusterfree(pmp, newcluster, NULL);
1479 bad2:
1480 	return (error);
1481 }
1482 
1483 static int
1484 msdosfs_rmdir(ap)
1485 	struct vop_rmdir_args /* {
1486 		struct vnode *a_dvp;
1487 		struct vnode *a_vp;
1488 		struct componentname *a_cnp;
1489 	} */ *ap;
1490 {
1491 	struct vnode *vp = ap->a_vp;
1492 	struct vnode *dvp = ap->a_dvp;
1493 	struct componentname *cnp = ap->a_cnp;
1494 	struct denode *ip, *dp;
1495 	int error;
1496 
1497 	ip = VTODE(vp);
1498 	dp = VTODE(dvp);
1499 
1500 	/*
1501 	 * Verify the directory is empty (and valid).
1502 	 * (Rmdir ".." won't be valid since
1503 	 *  ".." will contain a reference to
1504 	 *  the current directory and thus be
1505 	 *  non-empty.)
1506 	 */
1507 	error = 0;
1508 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1509 		error = ENOTEMPTY;
1510 		goto out;
1511 	}
1512 	/*
1513 	 * Delete the entry from the directory.  For dos filesystems this
1514 	 * gets rid of the directory entry on disk, the in memory copy
1515 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1516 	 * being found by deget().  When the vput() on dep is done we give
1517 	 * up access and eventually msdosfs_reclaim() will be called which
1518 	 * will remove it from the denode cache.
1519 	 */
1520 	error = removede(dp, ip);
1521 	if (error)
1522 		goto out;
1523 	/*
1524 	 * This is where we decrement the link count in the parent
1525 	 * directory.  Since dos filesystems don't do this we just purge
1526 	 * the name cache.
1527 	 */
1528 	cache_purge(dvp);
1529 	/*
1530 	 * Truncate the directory that is being deleted.
1531 	 */
1532 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1533 	cache_purge(vp);
1534 
1535 out:
1536 	return (error);
1537 }
1538 
1539 /*
1540  * DOS filesystems don't know what symlinks are.
1541  */
1542 static int
1543 msdosfs_symlink(ap)
1544 	struct vop_symlink_args /* {
1545 		struct vnode *a_dvp;
1546 		struct vnode **a_vpp;
1547 		struct componentname *a_cnp;
1548 		struct vattr *a_vap;
1549 		char *a_target;
1550 	} */ *ap;
1551 {
1552 	return (EOPNOTSUPP);
1553 }
1554 
1555 static int
1556 msdosfs_readdir(ap)
1557 	struct vop_readdir_args /* {
1558 		struct vnode *a_vp;
1559 		struct uio *a_uio;
1560 		struct ucred *a_cred;
1561 		int *a_eofflag;
1562 		int *a_ncookies;
1563 		u_long **a_cookies;
1564 	} */ *ap;
1565 {
1566 	struct mbnambuf nb;
1567 	int error = 0;
1568 	int diff;
1569 	long n;
1570 	int blsize;
1571 	long on;
1572 	u_long cn;
1573 	uint64_t fileno;
1574 	u_long dirsperblk;
1575 	long bias = 0;
1576 	daddr_t bn, lbn;
1577 	struct buf *bp;
1578 	struct denode *dep = VTODE(ap->a_vp);
1579 	struct msdosfsmount *pmp = dep->de_pmp;
1580 	struct direntry *dentp;
1581 	struct dirent dirbuf;
1582 	struct uio *uio = ap->a_uio;
1583 	u_long *cookies = NULL;
1584 	int ncookies = 0;
1585 	off_t offset, off;
1586 	int chksum = -1;
1587 
1588 #ifdef MSDOSFS_DEBUG
1589 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1590 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1591 #endif
1592 
1593 	/*
1594 	 * msdosfs_readdir() won't operate properly on regular files since
1595 	 * it does i/o only with the filesystem vnode, and hence can
1596 	 * retrieve the wrong block from the buffer cache for a plain file.
1597 	 * So, fail attempts to readdir() on a plain file.
1598 	 */
1599 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1600 		return (ENOTDIR);
1601 
1602 	/*
1603 	 * To be safe, initialize dirbuf
1604 	 */
1605 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1606 
1607 	/*
1608 	 * If the user buffer is smaller than the size of one dos directory
1609 	 * entry or the file offset is not a multiple of the size of a
1610 	 * directory entry, then we fail the read.
1611 	 */
1612 	off = offset = uio->uio_offset;
1613 	if (uio->uio_resid < sizeof(struct direntry) ||
1614 	    (offset & (sizeof(struct direntry) - 1)))
1615 		return (EINVAL);
1616 
1617 	if (ap->a_ncookies) {
1618 		ncookies = uio->uio_resid / 16;
1619 		cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1620 		       M_WAITOK);
1621 		*ap->a_cookies = cookies;
1622 		*ap->a_ncookies = ncookies;
1623 	}
1624 
1625 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1626 
1627 	/*
1628 	 * If they are reading from the root directory then, we simulate
1629 	 * the . and .. entries since these don't exist in the root
1630 	 * directory.  We also set the offset bias to make up for having to
1631 	 * simulate these entries. By this I mean that at file offset 64 we
1632 	 * read the first entry in the root directory that lives on disk.
1633 	 */
1634 	if (dep->de_StartCluster == MSDOSFSROOT
1635 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1636 #if 0
1637 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1638 		    offset);
1639 #endif
1640 		bias = 2 * sizeof(struct direntry);
1641 		if (offset < bias) {
1642 			for (n = (int)offset / sizeof(struct direntry);
1643 			     n < 2; n++) {
1644 				if (FAT32(pmp))
1645 					fileno = (uint64_t)cntobn(pmp,
1646 								 pmp->pm_rootdirblk)
1647 							  * dirsperblk;
1648 				else
1649 					fileno = 1;
1650 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1651 					dirbuf.d_fileno =
1652 					    msdosfs_fileno_map(pmp->pm_mountp,
1653 					    fileno);
1654 				} else {
1655 
1656 					dirbuf.d_fileno = (uint32_t)fileno;
1657 				}
1658 				dirbuf.d_type = DT_DIR;
1659 				switch (n) {
1660 				case 0:
1661 					dirbuf.d_namlen = 1;
1662 					strcpy(dirbuf.d_name, ".");
1663 					break;
1664 				case 1:
1665 					dirbuf.d_namlen = 2;
1666 					strcpy(dirbuf.d_name, "..");
1667 					break;
1668 				}
1669 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1670 				if (uio->uio_resid < dirbuf.d_reclen)
1671 					goto out;
1672 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1673 				if (error)
1674 					goto out;
1675 				offset += sizeof(struct direntry);
1676 				off = offset;
1677 				if (cookies) {
1678 					*cookies++ = offset;
1679 					if (--ncookies <= 0)
1680 						goto out;
1681 				}
1682 			}
1683 		}
1684 	}
1685 
1686 	mbnambuf_init(&nb);
1687 	off = offset;
1688 	while (uio->uio_resid > 0) {
1689 		lbn = de_cluster(pmp, offset - bias);
1690 		on = (offset - bias) & pmp->pm_crbomask;
1691 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1692 		diff = dep->de_FileSize - (offset - bias);
1693 		if (diff <= 0)
1694 			break;
1695 		n = min(n, diff);
1696 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1697 		if (error)
1698 			break;
1699 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1700 		if (error) {
1701 			brelse(bp);
1702 			return (error);
1703 		}
1704 		n = min(n, blsize - bp->b_resid);
1705 		if (n == 0) {
1706 			brelse(bp);
1707 			return (EIO);
1708 		}
1709 
1710 		/*
1711 		 * Convert from dos directory entries to fs-independent
1712 		 * directory entries.
1713 		 */
1714 		for (dentp = (struct direntry *)(bp->b_data + on);
1715 		     (char *)dentp < bp->b_data + on + n;
1716 		     dentp++, offset += sizeof(struct direntry)) {
1717 #if 0
1718 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1719 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1720 #endif
1721 			/*
1722 			 * If this is an unused entry, we can stop.
1723 			 */
1724 			if (dentp->deName[0] == SLOT_EMPTY) {
1725 				brelse(bp);
1726 				goto out;
1727 			}
1728 			/*
1729 			 * Skip deleted entries.
1730 			 */
1731 			if (dentp->deName[0] == SLOT_DELETED) {
1732 				chksum = -1;
1733 				mbnambuf_init(&nb);
1734 				continue;
1735 			}
1736 
1737 			/*
1738 			 * Handle Win95 long directory entries
1739 			 */
1740 			if (dentp->deAttributes == ATTR_WIN95) {
1741 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1742 					continue;
1743 				chksum = win2unixfn(&nb,
1744 				    (struct winentry *)dentp, chksum, pmp);
1745 				continue;
1746 			}
1747 
1748 			/*
1749 			 * Skip volume labels
1750 			 */
1751 			if (dentp->deAttributes & ATTR_VOLUME) {
1752 				chksum = -1;
1753 				mbnambuf_init(&nb);
1754 				continue;
1755 			}
1756 			/*
1757 			 * This computation of d_fileno must match
1758 			 * the computation of va_fileid in
1759 			 * msdosfs_getattr.
1760 			 */
1761 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1762 				fileno = getushort(dentp->deStartCluster);
1763 				if (FAT32(pmp))
1764 					fileno |= getushort(dentp->deHighClust) << 16;
1765 				/* if this is the root directory */
1766 				if (fileno == MSDOSFSROOT)
1767 					if (FAT32(pmp))
1768 						fileno = (uint64_t)cntobn(pmp,
1769 								pmp->pm_rootdirblk)
1770 							 * dirsperblk;
1771 					else
1772 						fileno = 1;
1773 				else
1774 					fileno = (uint64_t)cntobn(pmp, fileno) *
1775 					    dirsperblk;
1776 				dirbuf.d_type = DT_DIR;
1777 			} else {
1778 				fileno = (uoff_t)offset /
1779 				    sizeof(struct direntry);
1780 				dirbuf.d_type = DT_REG;
1781 			}
1782 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1783 				dirbuf.d_fileno =
1784 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1785 			} else
1786 				dirbuf.d_fileno = (uint32_t)fileno;
1787 
1788 			if (chksum != winChksum(dentp->deName)) {
1789 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1790 				    (u_char *)dirbuf.d_name,
1791 				    dentp->deLowerCase |
1792 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1793 					(LCASE_BASE | LCASE_EXT) : 0),
1794 				    pmp);
1795 				mbnambuf_init(&nb);
1796 			} else
1797 				mbnambuf_flush(&nb, &dirbuf);
1798 			chksum = -1;
1799 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1800 			if (uio->uio_resid < dirbuf.d_reclen) {
1801 				brelse(bp);
1802 				goto out;
1803 			}
1804 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1805 			if (error) {
1806 				brelse(bp);
1807 				goto out;
1808 			}
1809 			if (cookies) {
1810 				*cookies++ = offset + sizeof(struct direntry);
1811 				if (--ncookies <= 0) {
1812 					brelse(bp);
1813 					goto out;
1814 				}
1815 			}
1816 			off = offset + sizeof(struct direntry);
1817 		}
1818 		brelse(bp);
1819 	}
1820 out:
1821 	/* Subtract unused cookies */
1822 	if (ap->a_ncookies)
1823 		*ap->a_ncookies -= ncookies;
1824 
1825 	uio->uio_offset = off;
1826 
1827 	/*
1828 	 * Set the eofflag (NFS uses it)
1829 	 */
1830 	if (ap->a_eofflag) {
1831 		if (dep->de_FileSize - (offset - bias) <= 0)
1832 			*ap->a_eofflag = 1;
1833 		else
1834 			*ap->a_eofflag = 0;
1835 	}
1836 	return (error);
1837 }
1838 
1839 /*-
1840  * a_vp   - pointer to the file's vnode
1841  * a_bn   - logical block number within the file (cluster number for us)
1842  * a_bop  - where to return the bufobj of the special file containing the fs
1843  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1844  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1845  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1846  *          blocks whose physical blocks (together with a_bn's physical block)
1847  *          are contiguous.
1848  * a_runb - where to return the "run before" a_bn.
1849  */
1850 static int
1851 msdosfs_bmap(ap)
1852 	struct vop_bmap_args /* {
1853 		struct vnode *a_vp;
1854 		daddr_t a_bn;
1855 		struct bufobj **a_bop;
1856 		daddr_t *a_bnp;
1857 		int *a_runp;
1858 		int *a_runb;
1859 	} */ *ap;
1860 {
1861 	struct denode *dep;
1862 	struct mount *mp;
1863 	struct msdosfsmount *pmp;
1864 	struct vnode *vp;
1865 	daddr_t runbn;
1866 	u_long cn;
1867 	int bnpercn, error, maxio, maxrun, run;
1868 
1869 	vp = ap->a_vp;
1870 	dep = VTODE(vp);
1871 	pmp = dep->de_pmp;
1872 	if (ap->a_bop != NULL)
1873 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1874 	if (ap->a_bnp == NULL)
1875 		return (0);
1876 	if (ap->a_runp != NULL)
1877 		*ap->a_runp = 0;
1878 	if (ap->a_runb != NULL)
1879 		*ap->a_runb = 0;
1880 	cn = ap->a_bn;
1881 	if (cn != ap->a_bn)
1882 		return (EFBIG);
1883 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1884 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1885 		return (error);
1886 
1887 	mp = vp->v_mount;
1888 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1889 	bnpercn = de_cn2bn(pmp, 1);
1890 	if (ap->a_runp != NULL) {
1891 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1892 		for (run = 1; run <= maxrun; run++) {
1893 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1894 			    runbn != *ap->a_bnp + run * bnpercn)
1895 				break;
1896 		}
1897 		*ap->a_runp = run - 1;
1898 	}
1899 	if (ap->a_runb != NULL) {
1900 		maxrun = ulmin(maxio - 1, cn);
1901 		for (run = 1; run < maxrun; run++) {
1902 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1903 			    runbn != *ap->a_bnp - run * bnpercn)
1904 				break;
1905 		}
1906 		*ap->a_runb = run - 1;
1907 	}
1908 	return (0);
1909 }
1910 
1911 static int
1912 msdosfs_strategy(ap)
1913 	struct vop_strategy_args /* {
1914 		struct vnode *a_vp;
1915 		struct buf *a_bp;
1916 	} */ *ap;
1917 {
1918 	struct buf *bp = ap->a_bp;
1919 	struct denode *dep = VTODE(ap->a_vp);
1920 	struct bufobj *bo;
1921 	int error = 0;
1922 	daddr_t blkno;
1923 
1924 	/*
1925 	 * If we don't already know the filesystem relative block number
1926 	 * then get it using pcbmap().  If pcbmap() returns the block
1927 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1928 	 * don't allow files with holes, so we shouldn't ever see this.
1929 	 */
1930 	if (bp->b_blkno == bp->b_lblkno) {
1931 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1932 		bp->b_blkno = blkno;
1933 		if (error) {
1934 			bp->b_error = error;
1935 			bp->b_ioflags |= BIO_ERROR;
1936 			bufdone(bp);
1937 			return (0);
1938 		}
1939 		if ((long)bp->b_blkno == -1)
1940 			vfs_bio_clrbuf(bp);
1941 	}
1942 	if (bp->b_blkno == -1) {
1943 		bufdone(bp);
1944 		return (0);
1945 	}
1946 	/*
1947 	 * Read/write the block from/to the disk that contains the desired
1948 	 * file block.
1949 	 */
1950 	bp->b_iooffset = dbtob(bp->b_blkno);
1951 	bo = dep->de_pmp->pm_bo;
1952 	BO_STRATEGY(bo, bp);
1953 	return (0);
1954 }
1955 
1956 static int
1957 msdosfs_print(ap)
1958 	struct vop_print_args /* {
1959 		struct vnode *vp;
1960 	} */ *ap;
1961 {
1962 	struct denode *dep = VTODE(ap->a_vp);
1963 
1964 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1965 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1966 	printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1967 	return (0);
1968 }
1969 
1970 static int
1971 msdosfs_pathconf(ap)
1972 	struct vop_pathconf_args /* {
1973 		struct vnode *a_vp;
1974 		int a_name;
1975 		int *a_retval;
1976 	} */ *ap;
1977 {
1978 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1979 
1980 	switch (ap->a_name) {
1981 	case _PC_LINK_MAX:
1982 		*ap->a_retval = 1;
1983 		return (0);
1984 	case _PC_NAME_MAX:
1985 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1986 		return (0);
1987 	case _PC_PATH_MAX:
1988 		*ap->a_retval = PATH_MAX;
1989 		return (0);
1990 	case _PC_CHOWN_RESTRICTED:
1991 		*ap->a_retval = 1;
1992 		return (0);
1993 	case _PC_NO_TRUNC:
1994 		*ap->a_retval = 0;
1995 		return (0);
1996 	default:
1997 		return (EINVAL);
1998 	}
1999 	/* NOTREACHED */
2000 }
2001 
2002 static int
2003 msdosfs_vptofh(ap)
2004 	struct vop_vptofh_args /* {
2005 		struct vnode *a_vp;
2006 		struct fid *a_fhp;
2007 	} */ *ap;
2008 {
2009 	struct denode *dep;
2010 	struct defid *defhp;
2011 
2012 	dep = VTODE(ap->a_vp);
2013 	defhp = (struct defid *)ap->a_fhp;
2014 	defhp->defid_len = sizeof(struct defid);
2015 	defhp->defid_dirclust = dep->de_dirclust;
2016 	defhp->defid_dirofs = dep->de_diroffset;
2017 	/* defhp->defid_gen = dep->de_gen; */
2018 	return (0);
2019 }
2020 
2021 /* Global vfs data structures for msdosfs */
2022 struct vop_vector msdosfs_vnodeops = {
2023 	.vop_default =		&default_vnodeops,
2024 
2025 	.vop_access =		msdosfs_access,
2026 	.vop_bmap =		msdosfs_bmap,
2027 	.vop_cachedlookup =	msdosfs_lookup,
2028 	.vop_open =		msdosfs_open,
2029 	.vop_close =		msdosfs_close,
2030 	.vop_create =		msdosfs_create,
2031 	.vop_fsync =		msdosfs_fsync,
2032 	.vop_getattr =		msdosfs_getattr,
2033 	.vop_inactive =		msdosfs_inactive,
2034 	.vop_link =		msdosfs_link,
2035 	.vop_lookup =		vfs_cache_lookup,
2036 	.vop_mkdir =		msdosfs_mkdir,
2037 	.vop_mknod =		msdosfs_mknod,
2038 	.vop_pathconf =		msdosfs_pathconf,
2039 	.vop_print =		msdosfs_print,
2040 	.vop_read =		msdosfs_read,
2041 	.vop_readdir =		msdosfs_readdir,
2042 	.vop_reclaim =		msdosfs_reclaim,
2043 	.vop_remove =		msdosfs_remove,
2044 	.vop_rename =		msdosfs_rename,
2045 	.vop_rmdir =		msdosfs_rmdir,
2046 	.vop_setattr =		msdosfs_setattr,
2047 	.vop_strategy =		msdosfs_strategy,
2048 	.vop_symlink =		msdosfs_symlink,
2049 	.vop_write =		msdosfs_write,
2050 	.vop_vptofh =		msdosfs_vptofh,
2051 };
2052