xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision d876124d6ae9d56da5b4ff4c6015efd1d0c9222a)
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/proc.h>
65 #include <sys/resourcevar.h>
66 #include <sys/signalvar.h>
67 #include <sys/stat.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_extern.h>
73 
74 #include <fs/msdosfs/bpb.h>
75 #include <fs/msdosfs/direntry.h>
76 #include <fs/msdosfs/denode.h>
77 #include <fs/msdosfs/fat.h>
78 #include <fs/msdosfs/msdosfsmount.h>
79 
80 #define	DOS_FILESIZE_MAX	0xffffffff
81 
82 /*
83  * Prototypes for MSDOSFS vnode operations
84  */
85 static vop_create_t	msdosfs_create;
86 static vop_mknod_t	msdosfs_mknod;
87 static vop_open_t	msdosfs_open;
88 static vop_close_t	msdosfs_close;
89 static vop_access_t	msdosfs_access;
90 static vop_getattr_t	msdosfs_getattr;
91 static vop_setattr_t	msdosfs_setattr;
92 static vop_read_t	msdosfs_read;
93 static vop_write_t	msdosfs_write;
94 static vop_fsync_t	msdosfs_fsync;
95 static vop_remove_t	msdosfs_remove;
96 static vop_link_t	msdosfs_link;
97 static vop_rename_t	msdosfs_rename;
98 static vop_mkdir_t	msdosfs_mkdir;
99 static vop_rmdir_t	msdosfs_rmdir;
100 static vop_symlink_t	msdosfs_symlink;
101 static vop_readdir_t	msdosfs_readdir;
102 static vop_bmap_t	msdosfs_bmap;
103 static vop_strategy_t	msdosfs_strategy;
104 static vop_print_t	msdosfs_print;
105 static vop_pathconf_t	msdosfs_pathconf;
106 static vop_vptofh_t	msdosfs_vptofh;
107 
108 /*
109  * Some general notes:
110  *
111  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
112  * read/written using the vnode for the filesystem. Blocks that represent
113  * the contents of a file are read/written using the vnode for the file
114  * (including directories when they are read/written as files). This
115  * presents problems for the dos filesystem because data that should be in
116  * an inode (if dos had them) resides in the directory itself.  Since we
117  * must update directory entries without the benefit of having the vnode
118  * for the directory we must use the vnode for the filesystem.  This means
119  * that when a directory is actually read/written (via read, write, or
120  * readdir, or seek) we must use the vnode for the filesystem instead of
121  * the vnode for the directory as would happen in ufs. This is to insure we
122  * retreive the correct block from the buffer cache since the hash value is
123  * based upon the vnode address and the desired block number.
124  */
125 
126 /*
127  * Create a regular file. On entry the directory to contain the file being
128  * created is locked.  We must release before we return. We must also free
129  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
130  * only if the SAVESTART bit in cn_flags is clear on success.
131  */
132 static int
133 msdosfs_create(ap)
134 	struct vop_create_args /* {
135 		struct vnode *a_dvp;
136 		struct vnode **a_vpp;
137 		struct componentname *a_cnp;
138 		struct vattr *a_vap;
139 	} */ *ap;
140 {
141 	struct componentname *cnp = ap->a_cnp;
142 	struct denode ndirent;
143 	struct denode *dep;
144 	struct denode *pdep = VTODE(ap->a_dvp);
145 	struct timespec ts;
146 	int error;
147 
148 #ifdef MSDOSFS_DEBUG
149 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
150 #endif
151 
152 	/*
153 	 * If this is the root directory and there is no space left we
154 	 * can't do anything.  This is because the root directory can not
155 	 * change size.
156 	 */
157 	if (pdep->de_StartCluster == MSDOSFSROOT
158 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
159 		error = ENOSPC;
160 		goto bad;
161 	}
162 
163 	/*
164 	 * Create a directory entry for the file, then call createde() to
165 	 * have it installed. NOTE: DOS files are always executable.  We
166 	 * use the absence of the owner write bit to make the file
167 	 * readonly.
168 	 */
169 #ifdef DIAGNOSTIC
170 	if ((cnp->cn_flags & HASBUF) == 0)
171 		panic("msdosfs_create: no name");
172 #endif
173 	bzero(&ndirent, sizeof(ndirent));
174 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
175 	if (error)
176 		goto bad;
177 
178 	ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
179 				ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
180 	ndirent.de_LowerCase = 0;
181 	ndirent.de_StartCluster = 0;
182 	ndirent.de_FileSize = 0;
183 	ndirent.de_dev = pdep->de_dev;
184 	ndirent.de_pmp = pdep->de_pmp;
185 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
186 	getnanotime(&ts);
187 	DETIMES(&ndirent, &ts, &ts, &ts);
188 	error = createde(&ndirent, pdep, &dep, cnp);
189 	if (error)
190 		goto bad;
191 	*ap->a_vpp = DETOV(dep);
192 	return (0);
193 
194 bad:
195 	return (error);
196 }
197 
198 static int
199 msdosfs_mknod(ap)
200 	struct vop_mknod_args /* {
201 		struct vnode *a_dvp;
202 		struct vnode **a_vpp;
203 		struct componentname *a_cnp;
204 		struct vattr *a_vap;
205 	} */ *ap;
206 {
207 
208     return (EINVAL);
209 }
210 
211 static int
212 msdosfs_open(ap)
213 	struct vop_open_args /* {
214 		struct vnode *a_vp;
215 		int a_mode;
216 		struct ucred *a_cred;
217 		struct thread *a_td;
218 		int a_fdidx;
219 	} */ *ap;
220 {
221 	struct denode *dep = VTODE(ap->a_vp);
222 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
223 	return 0;
224 }
225 
226 static int
227 msdosfs_close(ap)
228 	struct vop_close_args /* {
229 		struct vnode *a_vp;
230 		int a_fflag;
231 		struct ucred *a_cred;
232 		struct thread *a_td;
233 	} */ *ap;
234 {
235 	struct vnode *vp = ap->a_vp;
236 	struct denode *dep = VTODE(vp);
237 	struct timespec ts;
238 
239 	VI_LOCK(vp);
240 	if (vp->v_usecount > 1) {
241 		getnanotime(&ts);
242 		DETIMES(dep, &ts, &ts, &ts);
243 	}
244 	VI_UNLOCK(vp);
245 	return 0;
246 }
247 
248 static int
249 msdosfs_access(ap)
250 	struct vop_access_args /* {
251 		struct vnode *a_vp;
252 		int a_mode;
253 		struct ucred *a_cred;
254 		struct thread *a_td;
255 	} */ *ap;
256 {
257 	struct vnode *vp = ap->a_vp;
258 	struct denode *dep = VTODE(ap->a_vp);
259 	struct msdosfsmount *pmp = dep->de_pmp;
260 	mode_t file_mode, mode = ap->a_mode;
261 
262 	file_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
263 	    ((dep->de_Attributes & ATTR_READONLY) ? 0 : (S_IWUSR|S_IWGRP|S_IWOTH));
264 	file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
265 
266 	/*
267 	 * Disallow writing to directories and regular files if the
268 	 * filesystem is read-only.
269 	 */
270 	if (mode & VWRITE) {
271 		switch (vp->v_type) {
272 		case VDIR:
273 		case VREG:
274 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
275 				return (EROFS);
276 			break;
277 		default:
278 			break;
279 		}
280 	}
281 
282 	return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
283 	    ap->a_mode, ap->a_cred, NULL));
284 }
285 
286 static int
287 msdosfs_getattr(ap)
288 	struct vop_getattr_args /* {
289 		struct vnode *a_vp;
290 		struct vattr *a_vap;
291 		struct ucred *a_cred;
292 		struct thread *a_td;
293 	} */ *ap;
294 {
295 	struct denode *dep = VTODE(ap->a_vp);
296 	struct msdosfsmount *pmp = dep->de_pmp;
297 	struct vattr *vap = ap->a_vap;
298 	mode_t mode;
299 	struct timespec ts;
300 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
301 	uint64_t fileid;
302 
303 	getnanotime(&ts);
304 	DETIMES(dep, &ts, &ts, &ts);
305 	vap->va_fsid = dev2udev(dep->de_dev);
306 	/*
307 	 * The following computation of the fileid must be the same as that
308 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
309 	 * doesn't work.
310 	 */
311 	if (dep->de_Attributes & ATTR_DIRECTORY) {
312 		fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
313 		    dirsperblk;
314 		if (dep->de_StartCluster == MSDOSFSROOT)
315 			fileid = 1;
316 	} else {
317 		fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
318 		    dirsperblk;
319 		if (dep->de_dirclust == MSDOSFSROOT)
320 			fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
321 		fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
322 	}
323 
324 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
325 		vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
326 	else
327 		vap->va_fileid = (long)fileid;
328 
329 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
330 		mode = S_IRWXU|S_IRWXG|S_IRWXO;
331 	else
332 		mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
333 	vap->va_mode = mode &
334 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
335 	vap->va_uid = pmp->pm_uid;
336 	vap->va_gid = pmp->pm_gid;
337 	vap->va_nlink = 1;
338 	vap->va_rdev = 0;
339 	vap->va_size = dep->de_FileSize;
340 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
341 	vap->va_ctime = vap->va_mtime;
342 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
343 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
344 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
345 		    0, &vap->va_birthtime);
346 	} else {
347 		vap->va_atime = vap->va_mtime;
348 		vap->va_birthtime.tv_sec = -1;
349 		vap->va_birthtime.tv_nsec = 0;
350 	}
351 	vap->va_flags = 0;
352 	if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
353 		vap->va_flags |= SF_ARCHIVED;
354 	vap->va_gen = 0;
355 	vap->va_blocksize = pmp->pm_bpcluster;
356 	vap->va_bytes =
357 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
358 	vap->va_type = ap->a_vp->v_type;
359 	vap->va_filerev = dep->de_modrev;
360 	return (0);
361 }
362 
363 static int
364 msdosfs_setattr(ap)
365 	struct vop_setattr_args /* {
366 		struct vnode *a_vp;
367 		struct vattr *a_vap;
368 		struct ucred *a_cred;
369 		struct thread *a_td;
370 	} */ *ap;
371 {
372 	struct vnode *vp = ap->a_vp;
373 	struct denode *dep = VTODE(ap->a_vp);
374 	struct msdosfsmount *pmp = dep->de_pmp;
375 	struct vattr *vap = ap->a_vap;
376 	struct ucred *cred = ap->a_cred;
377 	int error = 0;
378 
379 #ifdef MSDOSFS_DEBUG
380 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p, p %p\n",
381 	    ap->a_vp, vap, cred, ap->a_td);
382 #endif
383 
384 	/*
385 	 * Check for unsettable attributes.
386 	 */
387 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
388 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
389 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
390 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
391 #ifdef MSDOSFS_DEBUG
392 		printf("msdosfs_setattr(): returning EINVAL\n");
393 		printf("    va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
394 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
395 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
396 		    vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
397 		printf("    va_uid %x, va_gid %x\n",
398 		    vap->va_uid, vap->va_gid);
399 #endif
400 		return (EINVAL);
401 	}
402 	if (vap->va_flags != VNOVAL) {
403 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
404 			return (EROFS);
405 		if (cred->cr_uid != pmp->pm_uid) {
406 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
407 			if (error)
408 				return (error);
409 		}
410 		/*
411 		 * We are very inconsistent about handling unsupported
412 		 * attributes.  We ignored the access time and the
413 		 * read and execute bits.  We were strict for the other
414 		 * attributes.
415 		 *
416 		 * Here we are strict, stricter than ufs in not allowing
417 		 * users to attempt to set SF_SETTABLE bits or anyone to
418 		 * set unsupported bits.  However, we ignore attempts to
419 		 * set ATTR_ARCHIVE for directories `cp -pr' from a more
420 		 * sensible filesystem attempts it a lot.
421 		 */
422 		if (vap->va_flags & SF_SETTABLE) {
423 			error = priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0);
424 			if (error)
425 				return (error);
426 		}
427 		if (vap->va_flags & ~SF_ARCHIVED)
428 			return EOPNOTSUPP;
429 		if (vap->va_flags & SF_ARCHIVED)
430 			dep->de_Attributes &= ~ATTR_ARCHIVE;
431 		else if (!(dep->de_Attributes & ATTR_DIRECTORY))
432 			dep->de_Attributes |= ATTR_ARCHIVE;
433 		dep->de_flag |= DE_MODIFIED;
434 	}
435 
436 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
437 		uid_t uid;
438 		gid_t gid;
439 
440 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
441 			return (EROFS);
442 		uid = vap->va_uid;
443 		if (uid == (uid_t)VNOVAL)
444 			uid = pmp->pm_uid;
445 		gid = vap->va_gid;
446 		if (gid == (gid_t)VNOVAL)
447 			gid = pmp->pm_gid;
448 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
449 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
450 			error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
451 			if (error)
452 				return (error);
453 		}
454 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
455 			return EINVAL;
456 	}
457 
458 	if (vap->va_size != VNOVAL) {
459 		switch (vp->v_type) {
460 		case VDIR:
461 			return (EISDIR);
462 		case VREG:
463 			/*
464 			 * Truncation is only supported for regular files,
465 			 * Disallow it if the filesystem is read-only.
466 			 */
467 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
468 				return (EROFS);
469 			break;
470 		default:
471 			/*
472 			 * According to POSIX, the result is unspecified
473 			 * for file types other than regular files,
474 			 * directories and shared memory objects.  We
475 			 * don't support any file types except regular
476 			 * files and directories in this file system, so
477 			 * this (default) case is unreachable and can do
478 			 * anything.  Keep falling through to detrunc()
479 			 * for now.
480 			 */
481 			break;
482 		}
483 		error = detrunc(dep, vap->va_size, 0, cred, ap->a_td);
484 		if (error)
485 			return error;
486 	}
487 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
488 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
489 			return (EROFS);
490 		if (vap->va_vaflags & VA_UTIMES_NULL) {
491 			error = VOP_ACCESS(vp, VADMIN, cred, ap->a_td);
492 			if (error)
493 				error = VOP_ACCESS(vp, VWRITE, cred,
494 				    ap->a_td);
495 		} else
496 			error = VOP_ACCESS(vp, VADMIN, cred, ap->a_td);
497 		if (vp->v_type != VDIR) {
498 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
499 			    vap->va_atime.tv_sec != VNOVAL) {
500 				dep->de_flag &= ~DE_ACCESS;
501 				timespec2fattime(&vap->va_atime, 0,
502 				    &dep->de_ADate, NULL, NULL);
503 			}
504 			if (vap->va_mtime.tv_sec != VNOVAL) {
505 				dep->de_flag &= ~DE_UPDATE;
506 				timespec2fattime(&vap->va_mtime, 0,
507 				    &dep->de_MDate, &dep->de_MTime, NULL);
508 			}
509 			dep->de_Attributes |= ATTR_ARCHIVE;
510 			dep->de_flag |= DE_MODIFIED;
511 		}
512 	}
513 	/*
514 	 * DOS files only have the ability to have their writability
515 	 * attribute set, so we use the owner write bit to set the readonly
516 	 * attribute.
517 	 */
518 	if (vap->va_mode != (mode_t)VNOVAL) {
519 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
520 			return (EROFS);
521 		if (cred->cr_uid != pmp->pm_uid) {
522 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
523 			if (error)
524 				return (error);
525 		}
526 		if (vp->v_type != VDIR) {
527 			/* We ignore the read and execute bits. */
528 			if (vap->va_mode & VWRITE)
529 				dep->de_Attributes &= ~ATTR_READONLY;
530 			else
531 				dep->de_Attributes |= ATTR_READONLY;
532 			dep->de_Attributes |= ATTR_ARCHIVE;
533 			dep->de_flag |= DE_MODIFIED;
534 		}
535 	}
536 	return (deupdat(dep, 0));
537 }
538 
539 static int
540 msdosfs_read(ap)
541 	struct vop_read_args /* {
542 		struct vnode *a_vp;
543 		struct uio *a_uio;
544 		int a_ioflag;
545 		struct ucred *a_cred;
546 	} */ *ap;
547 {
548 	int error = 0;
549 	int blsize;
550 	int isadir;
551 	int orig_resid;
552 	u_int n;
553 	u_long diff;
554 	u_long on;
555 	daddr_t lbn;
556 	daddr_t rablock;
557 	int rasize;
558 	int seqcount;
559 	struct buf *bp;
560 	struct vnode *vp = ap->a_vp;
561 	struct denode *dep = VTODE(vp);
562 	struct msdosfsmount *pmp = dep->de_pmp;
563 	struct uio *uio = ap->a_uio;
564 
565 	/*
566 	 * If they didn't ask for any data, then we are done.
567 	 */
568 	orig_resid = uio->uio_resid;
569 	if (orig_resid == 0)
570 		return (0);
571 
572 	/*
573 	 * The caller is supposed to ensure that
574 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
575 	 * We don't need to check for large offsets as in ffs because
576 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
577 	 * offsets cannot cause overflow even in theory.
578 	 */
579 
580 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
581 
582 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
583 	do {
584 		if (uio->uio_offset >= dep->de_FileSize)
585 			break;
586 		lbn = de_cluster(pmp, uio->uio_offset);
587 		rablock = lbn + 1;
588 		blsize = pmp->pm_bpcluster;
589 		on = uio->uio_offset & pmp->pm_crbomask;
590 		/*
591 		 * If we are operating on a directory file then be sure to
592 		 * do i/o with the vnode for the filesystem instead of the
593 		 * vnode for the directory.
594 		 */
595 		if (isadir) {
596 			/* convert cluster # to block # */
597 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
598 			if (error == E2BIG) {
599 				error = EINVAL;
600 				break;
601 			} else if (error)
602 				break;
603 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
604 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
605 			error = bread(vp, lbn, blsize, NOCRED, &bp);
606 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
607 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
608 			    NOCRED, on + uio->uio_resid, seqcount, &bp);
609 		} else if (seqcount > 1) {
610 			rasize = blsize;
611 			error = breadn(vp, lbn,
612 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
613 		} else {
614 			error = bread(vp, lbn, blsize, NOCRED, &bp);
615 		}
616 		if (error) {
617 			brelse(bp);
618 			break;
619 		}
620 		diff = pmp->pm_bpcluster - on;
621 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
622 		diff = dep->de_FileSize - uio->uio_offset;
623 		if (diff < n)
624 			n = diff;
625 		diff = blsize - bp->b_resid;
626 		if (diff < n)
627 			n = diff;
628 		error = uiomove(bp->b_data + on, (int) n, uio);
629 		brelse(bp);
630 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
631 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
632 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
633 		dep->de_flag |= DE_ACCESS;
634 	return (error);
635 }
636 
637 /*
638  * Write data to a file or directory.
639  */
640 static int
641 msdosfs_write(ap)
642 	struct vop_write_args /* {
643 		struct vnode *a_vp;
644 		struct uio *a_uio;
645 		int a_ioflag;
646 		struct ucred *a_cred;
647 	} */ *ap;
648 {
649 	int n;
650 	int croffset;
651 	int resid;
652 	u_long osize;
653 	int error = 0;
654 	u_long count;
655 	int seqcount;
656 	daddr_t bn, lastcn;
657 	struct buf *bp;
658 	int ioflag = ap->a_ioflag;
659 	struct uio *uio = ap->a_uio;
660 	struct thread *td = uio->uio_td;
661 	struct vnode *vp = ap->a_vp;
662 	struct vnode *thisvp;
663 	struct denode *dep = VTODE(vp);
664 	struct msdosfsmount *pmp = dep->de_pmp;
665 	struct ucred *cred = ap->a_cred;
666 
667 #ifdef MSDOSFS_DEBUG
668 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
669 	    vp, uio, ioflag, cred);
670 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
671 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
672 #endif
673 
674 	switch (vp->v_type) {
675 	case VREG:
676 		if (ioflag & IO_APPEND)
677 			uio->uio_offset = dep->de_FileSize;
678 		thisvp = vp;
679 		break;
680 	case VDIR:
681 		return EISDIR;
682 	default:
683 		panic("msdosfs_write(): bad file type");
684 	}
685 
686 	/*
687 	 * This is needed (unlike in ffs_write()) because we extend the
688 	 * file outside of the loop but we don't want to extend the file
689 	 * for writes of 0 bytes.
690 	 */
691 	if (uio->uio_resid == 0)
692 		return (0);
693 
694 	/*
695 	 * The caller is supposed to ensure that
696 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
697 	 */
698 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
699 		return (EFBIG);
700 
701 	/*
702 	 * If they've exceeded their filesize limit, tell them about it.
703 	 */
704 	if (td != NULL) {
705 		PROC_LOCK(td->td_proc);
706 		if ((uoff_t)uio->uio_offset + uio->uio_resid >
707 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
708 			psignal(td->td_proc, SIGXFSZ);
709 			PROC_UNLOCK(td->td_proc);
710 			return (EFBIG);
711 		}
712 		PROC_UNLOCK(td->td_proc);
713 	}
714 
715 	/*
716 	 * If the offset we are starting the write at is beyond the end of
717 	 * the file, then they've done a seek.  Unix filesystems allow
718 	 * files with holes in them, DOS doesn't so we must fill the hole
719 	 * with zeroed blocks.
720 	 */
721 	if (uio->uio_offset > dep->de_FileSize) {
722 		error = deextend(dep, uio->uio_offset, cred);
723 		if (error)
724 			return (error);
725 	}
726 
727 	/*
728 	 * Remember some values in case the write fails.
729 	 */
730 	resid = uio->uio_resid;
731 	osize = dep->de_FileSize;
732 
733 	/*
734 	 * If we write beyond the end of the file, extend it to its ultimate
735 	 * size ahead of the time to hopefully get a contiguous area.
736 	 */
737 	if (uio->uio_offset + resid > osize) {
738 		count = de_clcount(pmp, uio->uio_offset + resid) -
739 			de_clcount(pmp, osize);
740 		error = extendfile(dep, count, NULL, NULL, 0);
741 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
742 			goto errexit;
743 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
744 	} else
745 		lastcn = de_clcount(pmp, osize) - 1;
746 
747 	seqcount = ioflag >> IO_SEQSHIFT;
748 	do {
749 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
750 			error = ENOSPC;
751 			break;
752 		}
753 
754 		croffset = uio->uio_offset & pmp->pm_crbomask;
755 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
756 		if (uio->uio_offset + n > dep->de_FileSize) {
757 			dep->de_FileSize = uio->uio_offset + n;
758 			/* The object size needs to be set before buffer is allocated */
759 			vnode_pager_setsize(vp, dep->de_FileSize);
760 		}
761 
762 		bn = de_cluster(pmp, uio->uio_offset);
763 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
764 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
765 			> de_cluster(pmp, uio->uio_offset)
766 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
767 			/*
768 			 * If either the whole cluster gets written,
769 			 * or we write the cluster from its start beyond EOF,
770 			 * then no need to read data from disk.
771 			 */
772 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
773 			vfs_bio_clrbuf(bp);
774 			/*
775 			 * Do the bmap now, since pcbmap needs buffers
776 			 * for the fat table. (see msdosfs_strategy)
777 			 */
778 			if (bp->b_blkno == bp->b_lblkno) {
779 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
780 				if (error)
781 					bp->b_blkno = -1;
782 				else
783 					bp->b_blkno = bn;
784 			}
785 			if (bp->b_blkno == -1) {
786 				brelse(bp);
787 				if (!error)
788 					error = EIO;		/* XXX */
789 				break;
790 			}
791 		} else {
792 			/*
793 			 * The block we need to write into exists, so read it in.
794 			 */
795 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
796 			if (error) {
797 				brelse(bp);
798 				break;
799 			}
800 		}
801 
802 		/*
803 		 * Should these vnode_pager_* functions be done on dir
804 		 * files?
805 		 */
806 
807 		/*
808 		 * Copy the data from user space into the buf header.
809 		 */
810 		error = uiomove(bp->b_data + croffset, n, uio);
811 		if (error) {
812 			brelse(bp);
813 			break;
814 		}
815 
816 		/* Prepare for clustered writes in some else clauses. */
817 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
818 			bp->b_flags |= B_CLUSTEROK;
819 
820 		/*
821 		 * If IO_SYNC, then each buffer is written synchronously.
822 		 * Otherwise, if we have a severe page deficiency then
823 		 * write the buffer asynchronously.  Otherwise, if on a
824 		 * cluster boundary then write the buffer asynchronously,
825 		 * combining it with contiguous clusters if permitted and
826 		 * possible, since we don't expect more writes into this
827 		 * buffer soon.  Otherwise, do a delayed write because we
828 		 * expect more writes into this buffer soon.
829 		 */
830 		if (ioflag & IO_SYNC)
831 			(void)bwrite(bp);
832 		else if (vm_page_count_severe() || buf_dirty_count_severe())
833 			bawrite(bp);
834 		else if (n + croffset == pmp->pm_bpcluster) {
835 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
836 				cluster_write(vp, bp, dep->de_FileSize,
837 				    seqcount);
838 			else
839 				bawrite(bp);
840 		} else
841 			bdwrite(bp);
842 		dep->de_flag |= DE_UPDATE;
843 	} while (error == 0 && uio->uio_resid > 0);
844 
845 	/*
846 	 * If the write failed and they want us to, truncate the file back
847 	 * to the size it was before the write was attempted.
848 	 */
849 errexit:
850 	if (error) {
851 		if (ioflag & IO_UNIT) {
852 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED, NULL);
853 			uio->uio_offset -= resid - uio->uio_resid;
854 			uio->uio_resid = resid;
855 		} else {
856 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED, NULL);
857 			if (uio->uio_resid != resid)
858 				error = 0;
859 		}
860 	} else if (ioflag & IO_SYNC)
861 		error = deupdat(dep, 1);
862 	return (error);
863 }
864 
865 /*
866  * Flush the blocks of a file to disk.
867  *
868  * This function is worthless for vnodes that represent directories. Maybe we
869  * could just do a sync if they try an fsync on a directory file.
870  */
871 static int
872 msdosfs_fsync(ap)
873 	struct vop_fsync_args /* {
874 		struct vnode *a_vp;
875 		struct ucred *a_cred;
876 		int a_waitfor;
877 		struct thread *a_td;
878 	} */ *ap;
879 {
880 
881 	vop_stdfsync(ap);
882 	return (deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT));
883 }
884 
885 static int
886 msdosfs_remove(ap)
887 	struct vop_remove_args /* {
888 		struct vnode *a_dvp;
889 		struct vnode *a_vp;
890 		struct componentname *a_cnp;
891 	} */ *ap;
892 {
893 	struct denode *dep = VTODE(ap->a_vp);
894 	struct denode *ddep = VTODE(ap->a_dvp);
895 	int error;
896 
897 	if (ap->a_vp->v_type == VDIR)
898 		error = EPERM;
899 	else
900 		error = removede(ddep, dep);
901 #ifdef MSDOSFS_DEBUG
902 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
903 #endif
904 	return (error);
905 }
906 
907 /*
908  * DOS filesystems don't know what links are.
909  */
910 static int
911 msdosfs_link(ap)
912 	struct vop_link_args /* {
913 		struct vnode *a_tdvp;
914 		struct vnode *a_vp;
915 		struct componentname *a_cnp;
916 	} */ *ap;
917 {
918 	return (EOPNOTSUPP);
919 }
920 
921 /*
922  * Renames on files require moving the denode to a new hash queue since the
923  * denode's location is used to compute which hash queue to put the file
924  * in. Unless it is a rename in place.  For example "mv a b".
925  *
926  * What follows is the basic algorithm:
927  *
928  * if (file move) {
929  *	if (dest file exists) {
930  *		remove dest file
931  *	}
932  *	if (dest and src in same directory) {
933  *		rewrite name in existing directory slot
934  *	} else {
935  *		write new entry in dest directory
936  *		update offset and dirclust in denode
937  *		move denode to new hash chain
938  *		clear old directory entry
939  *	}
940  * } else {
941  *	directory move
942  *	if (dest directory exists) {
943  *		if (dest is not empty) {
944  *			return ENOTEMPTY
945  *		}
946  *		remove dest directory
947  *	}
948  *	if (dest and src in same directory) {
949  *		rewrite name in existing entry
950  *	} else {
951  *		be sure dest is not a child of src directory
952  *		write entry in dest directory
953  *		update "." and ".." in moved directory
954  *		clear old directory entry for moved directory
955  *	}
956  * }
957  *
958  * On entry:
959  *	source's parent directory is unlocked
960  *	source file or directory is unlocked
961  *	destination's parent directory is locked
962  *	destination file or directory is locked if it exists
963  *
964  * On exit:
965  *	all denodes should be released
966  */
967 static int
968 msdosfs_rename(ap)
969 	struct vop_rename_args /* {
970 		struct vnode *a_fdvp;
971 		struct vnode *a_fvp;
972 		struct componentname *a_fcnp;
973 		struct vnode *a_tdvp;
974 		struct vnode *a_tvp;
975 		struct componentname *a_tcnp;
976 	} */ *ap;
977 {
978 	struct vnode *tdvp = ap->a_tdvp;
979 	struct vnode *fvp = ap->a_fvp;
980 	struct vnode *fdvp = ap->a_fdvp;
981 	struct vnode *tvp = ap->a_tvp;
982 	struct componentname *tcnp = ap->a_tcnp;
983 	struct componentname *fcnp = ap->a_fcnp;
984 	struct denode *ip, *xp, *dp, *zp;
985 	u_char toname[11], oldname[11];
986 	u_long from_diroffset, to_diroffset;
987 	u_char to_count;
988 	int doingdirectory = 0, newparent = 0;
989 	int error;
990 	u_long cn;
991 	daddr_t bn;
992 	struct denode *fddep;	/* from file's parent directory	 */
993 	struct msdosfsmount *pmp;
994 	struct direntry *dotdotp;
995 	struct buf *bp;
996 
997 	fddep = VTODE(ap->a_fdvp);
998 	pmp = fddep->de_pmp;
999 
1000 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1001 
1002 #ifdef DIAGNOSTIC
1003 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1004 	    (fcnp->cn_flags & HASBUF) == 0)
1005 		panic("msdosfs_rename: no name");
1006 #endif
1007 	/*
1008 	 * Check for cross-device rename.
1009 	 */
1010 	if (fvp->v_mount != tdvp->v_mount ||
1011 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1012 		error = EXDEV;
1013 abortit:
1014 		if (tdvp == tvp)
1015 			vrele(tdvp);
1016 		else
1017 			vput(tdvp);
1018 		if (tvp)
1019 			vput(tvp);
1020 		vrele(fdvp);
1021 		vrele(fvp);
1022 		return (error);
1023 	}
1024 
1025 	/*
1026 	 * If source and dest are the same, do nothing.
1027 	 */
1028 	if (tvp == fvp) {
1029 		error = 0;
1030 		goto abortit;
1031 	}
1032 
1033 	error = vn_lock(fvp, LK_EXCLUSIVE);
1034 	if (error)
1035 		goto abortit;
1036 	dp = VTODE(fdvp);
1037 	ip = VTODE(fvp);
1038 
1039 	/*
1040 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1041 	 * leads to a crippled directory tree.  It's pretty tough to do a
1042 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1043 	 * doesn't work if the ".." entry is missing.
1044 	 */
1045 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1046 		/*
1047 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1048 		 */
1049 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1050 		    dp == ip ||
1051 		    (fcnp->cn_flags & ISDOTDOT) ||
1052 		    (tcnp->cn_flags & ISDOTDOT) ||
1053 		    (ip->de_flag & DE_RENAME)) {
1054 			VOP_UNLOCK(fvp, 0);
1055 			error = EINVAL;
1056 			goto abortit;
1057 		}
1058 		ip->de_flag |= DE_RENAME;
1059 		doingdirectory++;
1060 	}
1061 
1062 	/*
1063 	 * When the target exists, both the directory
1064 	 * and target vnodes are returned locked.
1065 	 */
1066 	dp = VTODE(tdvp);
1067 	xp = tvp ? VTODE(tvp) : NULL;
1068 	/*
1069 	 * Remember direntry place to use for destination
1070 	 */
1071 	to_diroffset = dp->de_fndoffset;
1072 	to_count = dp->de_fndcnt;
1073 
1074 	/*
1075 	 * If ".." must be changed (ie the directory gets a new
1076 	 * parent) then the source directory must not be in the
1077 	 * directory heirarchy above the target, as this would
1078 	 * orphan everything below the source directory. Also
1079 	 * the user must have write permission in the source so
1080 	 * as to be able to change "..". We must repeat the call
1081 	 * to namei, as the parent directory is unlocked by the
1082 	 * call to doscheckpath().
1083 	 */
1084 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1085 	VOP_UNLOCK(fvp, 0);
1086 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1087 		newparent = 1;
1088 	if (doingdirectory && newparent) {
1089 		if (error)	/* write access check above */
1090 			goto bad;
1091 		if (xp != NULL)
1092 			vput(tvp);
1093 		/*
1094 		 * doscheckpath() vput()'s dp,
1095 		 * so we have to do a relookup afterwards
1096 		 */
1097 		error = doscheckpath(ip, dp);
1098 		if (error)
1099 			goto out;
1100 		if ((tcnp->cn_flags & SAVESTART) == 0)
1101 			panic("msdosfs_rename: lost to startdir");
1102 		error = relookup(tdvp, &tvp, tcnp);
1103 		if (error)
1104 			goto out;
1105 		dp = VTODE(tdvp);
1106 		xp = tvp ? VTODE(tvp) : NULL;
1107 	}
1108 
1109 	if (xp != NULL) {
1110 		/*
1111 		 * Target must be empty if a directory and have no links
1112 		 * to it. Also, ensure source and target are compatible
1113 		 * (both directories, or both not directories).
1114 		 */
1115 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1116 			if (!dosdirempty(xp)) {
1117 				error = ENOTEMPTY;
1118 				goto bad;
1119 			}
1120 			if (!doingdirectory) {
1121 				error = ENOTDIR;
1122 				goto bad;
1123 			}
1124 			cache_purge(tdvp);
1125 		} else if (doingdirectory) {
1126 			error = EISDIR;
1127 			goto bad;
1128 		}
1129 		error = removede(dp, xp);
1130 		if (error)
1131 			goto bad;
1132 		vput(tvp);
1133 		xp = NULL;
1134 	}
1135 
1136 	/*
1137 	 * Convert the filename in tcnp into a dos filename. We copy this
1138 	 * into the denode and directory entry for the destination
1139 	 * file/directory.
1140 	 */
1141 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1142 	if (error)
1143 		goto abortit;
1144 
1145 	/*
1146 	 * Since from wasn't locked at various places above,
1147 	 * have to do a relookup here.
1148 	 */
1149 	fcnp->cn_flags &= ~MODMASK;
1150 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1151 	if ((fcnp->cn_flags & SAVESTART) == 0)
1152 		panic("msdosfs_rename: lost from startdir");
1153 	if (!newparent)
1154 		VOP_UNLOCK(tdvp, 0);
1155 	if (relookup(fdvp, &fvp, fcnp) == 0)
1156 		vrele(fdvp);
1157 	if (fvp == NULL) {
1158 		/*
1159 		 * From name has disappeared.
1160 		 */
1161 		if (doingdirectory)
1162 			panic("rename: lost dir entry");
1163 		if (newparent)
1164 			VOP_UNLOCK(tdvp, 0);
1165 		vrele(tdvp);
1166 		vrele(ap->a_fvp);
1167 		return 0;
1168 	}
1169 	xp = VTODE(fvp);
1170 	zp = VTODE(fdvp);
1171 	from_diroffset = zp->de_fndoffset;
1172 
1173 	/*
1174 	 * Ensure that the directory entry still exists and has not
1175 	 * changed till now. If the source is a file the entry may
1176 	 * have been unlinked or renamed. In either case there is
1177 	 * no further work to be done. If the source is a directory
1178 	 * then it cannot have been rmdir'ed or renamed; this is
1179 	 * prohibited by the DE_RENAME flag.
1180 	 */
1181 	if (xp != ip) {
1182 		if (doingdirectory)
1183 			panic("rename: lost dir entry");
1184 		VOP_UNLOCK(fvp, 0);
1185 		if (newparent)
1186 			VOP_UNLOCK(fdvp, 0);
1187 		vrele(ap->a_fvp);
1188 		xp = NULL;
1189 	} else {
1190 		vrele(fvp);
1191 		xp = NULL;
1192 
1193 		/*
1194 		 * First write a new entry in the destination
1195 		 * directory and mark the entry in the source directory
1196 		 * as deleted.  Then move the denode to the correct hash
1197 		 * chain for its new location in the filesystem.  And, if
1198 		 * we moved a directory, then update its .. entry to point
1199 		 * to the new parent directory.
1200 		 */
1201 		bcopy(ip->de_Name, oldname, 11);
1202 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1203 		dp->de_fndoffset = to_diroffset;
1204 		dp->de_fndcnt = to_count;
1205 		error = createde(ip, dp, (struct denode **)0, tcnp);
1206 		if (error) {
1207 			bcopy(oldname, ip->de_Name, 11);
1208 			if (newparent)
1209 				VOP_UNLOCK(fdvp, 0);
1210 			VOP_UNLOCK(fvp, 0);
1211 			goto bad;
1212 		}
1213 		ip->de_refcnt++;
1214 		zp->de_fndoffset = from_diroffset;
1215 		error = removede(zp, ip);
1216 		if (error) {
1217 			/* XXX should downgrade to ro here, fs is corrupt */
1218 			if (newparent)
1219 				VOP_UNLOCK(fdvp, 0);
1220 			VOP_UNLOCK(fvp, 0);
1221 			goto bad;
1222 		}
1223 		if (!doingdirectory) {
1224 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1225 				       &ip->de_dirclust, 0);
1226 			if (error) {
1227 				/* XXX should downgrade to ro here, fs is corrupt */
1228 				if (newparent)
1229 					VOP_UNLOCK(fdvp, 0);
1230 				VOP_UNLOCK(fvp, 0);
1231 				goto bad;
1232 			}
1233 			if (ip->de_dirclust == MSDOSFSROOT)
1234 				ip->de_diroffset = to_diroffset;
1235 			else
1236 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1237 		}
1238 		reinsert(ip);
1239 		if (newparent)
1240 			VOP_UNLOCK(fdvp, 0);
1241 	}
1242 
1243 	/*
1244 	 * If we moved a directory to a new parent directory, then we must
1245 	 * fixup the ".." entry in the moved directory.
1246 	 */
1247 	if (doingdirectory && newparent) {
1248 		cn = ip->de_StartCluster;
1249 		if (cn == MSDOSFSROOT) {
1250 			/* this should never happen */
1251 			panic("msdosfs_rename(): updating .. in root directory?");
1252 		} else
1253 			bn = cntobn(pmp, cn);
1254 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1255 			      NOCRED, &bp);
1256 		if (error) {
1257 			/* XXX should downgrade to ro here, fs is corrupt */
1258 			brelse(bp);
1259 			VOP_UNLOCK(fvp, 0);
1260 			goto bad;
1261 		}
1262 		dotdotp = (struct direntry *)bp->b_data + 1;
1263 		putushort(dotdotp->deStartCluster, dp->de_StartCluster);
1264 		if (FAT32(pmp))
1265 			putushort(dotdotp->deHighClust, dp->de_StartCluster >> 16);
1266 		if (fvp->v_mount->mnt_flag & MNT_ASYNC)
1267 			bdwrite(bp);
1268 		else if ((error = bwrite(bp)) != 0) {
1269 			/* XXX should downgrade to ro here, fs is corrupt */
1270 			VOP_UNLOCK(fvp, 0);
1271 			goto bad;
1272 		}
1273 	}
1274 
1275 	VOP_UNLOCK(fvp, 0);
1276 bad:
1277 	if (xp)
1278 		vput(tvp);
1279 	vput(tdvp);
1280 out:
1281 	ip->de_flag &= ~DE_RENAME;
1282 	vrele(fdvp);
1283 	vrele(fvp);
1284 	return (error);
1285 
1286 }
1287 
1288 static struct {
1289 	struct direntry dot;
1290 	struct direntry dotdot;
1291 } dosdirtemplate = {
1292 	{	".       ", "   ",			/* the . entry */
1293 		ATTR_DIRECTORY,				/* file attribute */
1294 		0,					/* reserved */
1295 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1296 		{ 0, 0 },				/* access date */
1297 		{ 0, 0 },				/* high bits of start cluster */
1298 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1299 		{ 0, 0 },				/* startcluster */
1300 		{ 0, 0, 0, 0 }				/* filesize */
1301 	},
1302 	{	"..      ", "   ",			/* the .. entry */
1303 		ATTR_DIRECTORY,				/* file attribute */
1304 		0,					/* reserved */
1305 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1306 		{ 0, 0 },				/* access date */
1307 		{ 0, 0 },				/* high bits of start cluster */
1308 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1309 		{ 0, 0 },				/* startcluster */
1310 		{ 0, 0, 0, 0 }				/* filesize */
1311 	}
1312 };
1313 
1314 static int
1315 msdosfs_mkdir(ap)
1316 	struct vop_mkdir_args /* {
1317 		struct vnode *a_dvp;
1318 		struct vnode **a_vpp;
1319 		struvt componentname *a_cnp;
1320 		struct vattr *a_vap;
1321 	} */ *ap;
1322 {
1323 	struct componentname *cnp = ap->a_cnp;
1324 	struct denode *dep;
1325 	struct denode *pdep = VTODE(ap->a_dvp);
1326 	struct direntry *denp;
1327 	struct msdosfsmount *pmp = pdep->de_pmp;
1328 	struct buf *bp;
1329 	u_long newcluster, pcl;
1330 	int bn;
1331 	int error;
1332 	struct denode ndirent;
1333 	struct timespec ts;
1334 
1335 	/*
1336 	 * If this is the root directory and there is no space left we
1337 	 * can't do anything.  This is because the root directory can not
1338 	 * change size.
1339 	 */
1340 	if (pdep->de_StartCluster == MSDOSFSROOT
1341 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1342 		error = ENOSPC;
1343 		goto bad2;
1344 	}
1345 
1346 	/*
1347 	 * Allocate a cluster to hold the about to be created directory.
1348 	 */
1349 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1350 	if (error)
1351 		goto bad2;
1352 
1353 	bzero(&ndirent, sizeof(ndirent));
1354 	ndirent.de_pmp = pmp;
1355 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1356 	getnanotime(&ts);
1357 	DETIMES(&ndirent, &ts, &ts, &ts);
1358 
1359 	/*
1360 	 * Now fill the cluster with the "." and ".." entries. And write
1361 	 * the cluster to disk.  This way it is there for the parent
1362 	 * directory to be pointing at if there were a crash.
1363 	 */
1364 	bn = cntobn(pmp, newcluster);
1365 	/* always succeeds */
1366 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1367 	bzero(bp->b_data, pmp->pm_bpcluster);
1368 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1369 	denp = (struct direntry *)bp->b_data;
1370 	putushort(denp[0].deStartCluster, newcluster);
1371 	putushort(denp[0].deCDate, ndirent.de_CDate);
1372 	putushort(denp[0].deCTime, ndirent.de_CTime);
1373 	denp[0].deCHundredth = ndirent.de_CHun;
1374 	putushort(denp[0].deADate, ndirent.de_ADate);
1375 	putushort(denp[0].deMDate, ndirent.de_MDate);
1376 	putushort(denp[0].deMTime, ndirent.de_MTime);
1377 	pcl = pdep->de_StartCluster;
1378 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1379 		pcl = 0;
1380 	putushort(denp[1].deStartCluster, pcl);
1381 	putushort(denp[1].deCDate, ndirent.de_CDate);
1382 	putushort(denp[1].deCTime, ndirent.de_CTime);
1383 	denp[1].deCHundredth = ndirent.de_CHun;
1384 	putushort(denp[1].deADate, ndirent.de_ADate);
1385 	putushort(denp[1].deMDate, ndirent.de_MDate);
1386 	putushort(denp[1].deMTime, ndirent.de_MTime);
1387 	if (FAT32(pmp)) {
1388 		putushort(denp[0].deHighClust, newcluster >> 16);
1389 		putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1390 	}
1391 
1392 	if (ap->a_dvp->v_mount->mnt_flag & MNT_ASYNC)
1393 		bdwrite(bp);
1394 	else if ((error = bwrite(bp)) != 0)
1395 		goto bad;
1396 
1397 	/*
1398 	 * Now build up a directory entry pointing to the newly allocated
1399 	 * cluster.  This will be written to an empty slot in the parent
1400 	 * directory.
1401 	 */
1402 #ifdef DIAGNOSTIC
1403 	if ((cnp->cn_flags & HASBUF) == 0)
1404 		panic("msdosfs_mkdir: no name");
1405 #endif
1406 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1407 	if (error)
1408 		goto bad;
1409 
1410 	ndirent.de_Attributes = ATTR_DIRECTORY;
1411 	ndirent.de_LowerCase = 0;
1412 	ndirent.de_StartCluster = newcluster;
1413 	ndirent.de_FileSize = 0;
1414 	ndirent.de_dev = pdep->de_dev;
1415 	error = createde(&ndirent, pdep, &dep, cnp);
1416 	if (error)
1417 		goto bad;
1418 	*ap->a_vpp = DETOV(dep);
1419 	return (0);
1420 
1421 bad:
1422 	clusterfree(pmp, newcluster, NULL);
1423 bad2:
1424 	return (error);
1425 }
1426 
1427 static int
1428 msdosfs_rmdir(ap)
1429 	struct vop_rmdir_args /* {
1430 		struct vnode *a_dvp;
1431 		struct vnode *a_vp;
1432 		struct componentname *a_cnp;
1433 	} */ *ap;
1434 {
1435 	struct vnode *vp = ap->a_vp;
1436 	struct vnode *dvp = ap->a_dvp;
1437 	struct componentname *cnp = ap->a_cnp;
1438 	struct denode *ip, *dp;
1439 	struct thread *td = cnp->cn_thread;
1440 	int error;
1441 
1442 	ip = VTODE(vp);
1443 	dp = VTODE(dvp);
1444 
1445 	/*
1446 	 * Verify the directory is empty (and valid).
1447 	 * (Rmdir ".." won't be valid since
1448 	 *  ".." will contain a reference to
1449 	 *  the current directory and thus be
1450 	 *  non-empty.)
1451 	 */
1452 	error = 0;
1453 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1454 		error = ENOTEMPTY;
1455 		goto out;
1456 	}
1457 	/*
1458 	 * Delete the entry from the directory.  For dos filesystems this
1459 	 * gets rid of the directory entry on disk, the in memory copy
1460 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1461 	 * being found by deget().  When the vput() on dep is done we give
1462 	 * up access and eventually msdosfs_reclaim() will be called which
1463 	 * will remove it from the denode cache.
1464 	 */
1465 	error = removede(dp, ip);
1466 	if (error)
1467 		goto out;
1468 	/*
1469 	 * This is where we decrement the link count in the parent
1470 	 * directory.  Since dos filesystems don't do this we just purge
1471 	 * the name cache.
1472 	 */
1473 	cache_purge(dvp);
1474 	VOP_UNLOCK(dvp, 0);
1475 	/*
1476 	 * Truncate the directory that is being deleted.
1477 	 */
1478 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred, td);
1479 	cache_purge(vp);
1480 
1481 	vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1482 out:
1483 	return (error);
1484 }
1485 
1486 /*
1487  * DOS filesystems don't know what symlinks are.
1488  */
1489 static int
1490 msdosfs_symlink(ap)
1491 	struct vop_symlink_args /* {
1492 		struct vnode *a_dvp;
1493 		struct vnode **a_vpp;
1494 		struct componentname *a_cnp;
1495 		struct vattr *a_vap;
1496 		char *a_target;
1497 	} */ *ap;
1498 {
1499 	return (EOPNOTSUPP);
1500 }
1501 
1502 static int
1503 msdosfs_readdir(ap)
1504 	struct vop_readdir_args /* {
1505 		struct vnode *a_vp;
1506 		struct uio *a_uio;
1507 		struct ucred *a_cred;
1508 		int *a_eofflag;
1509 		int *a_ncookies;
1510 		u_long **a_cookies;
1511 	} */ *ap;
1512 {
1513 	struct mbnambuf nb;
1514 	int error = 0;
1515 	int diff;
1516 	long n;
1517 	int blsize;
1518 	long on;
1519 	u_long cn;
1520 	uint64_t fileno;
1521 	u_long dirsperblk;
1522 	long bias = 0;
1523 	daddr_t bn, lbn;
1524 	struct buf *bp;
1525 	struct denode *dep = VTODE(ap->a_vp);
1526 	struct msdosfsmount *pmp = dep->de_pmp;
1527 	struct direntry *dentp;
1528 	struct dirent dirbuf;
1529 	struct uio *uio = ap->a_uio;
1530 	u_long *cookies = NULL;
1531 	int ncookies = 0;
1532 	off_t offset, off;
1533 	int chksum = -1;
1534 
1535 #ifdef MSDOSFS_DEBUG
1536 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1537 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1538 #endif
1539 
1540 	/*
1541 	 * msdosfs_readdir() won't operate properly on regular files since
1542 	 * it does i/o only with the the filesystem vnode, and hence can
1543 	 * retrieve the wrong block from the buffer cache for a plain file.
1544 	 * So, fail attempts to readdir() on a plain file.
1545 	 */
1546 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1547 		return (ENOTDIR);
1548 
1549 	/*
1550 	 * To be safe, initialize dirbuf
1551 	 */
1552 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1553 
1554 	/*
1555 	 * If the user buffer is smaller than the size of one dos directory
1556 	 * entry or the file offset is not a multiple of the size of a
1557 	 * directory entry, then we fail the read.
1558 	 */
1559 	off = offset = uio->uio_offset;
1560 	if (uio->uio_resid < sizeof(struct direntry) ||
1561 	    (offset & (sizeof(struct direntry) - 1)))
1562 		return (EINVAL);
1563 
1564 	if (ap->a_ncookies) {
1565 		ncookies = uio->uio_resid / 16;
1566 		MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
1567 		       M_WAITOK);
1568 		*ap->a_cookies = cookies;
1569 		*ap->a_ncookies = ncookies;
1570 	}
1571 
1572 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1573 
1574 	/*
1575 	 * If they are reading from the root directory then, we simulate
1576 	 * the . and .. entries since these don't exist in the root
1577 	 * directory.  We also set the offset bias to make up for having to
1578 	 * simulate these entries. By this I mean that at file offset 64 we
1579 	 * read the first entry in the root directory that lives on disk.
1580 	 */
1581 	if (dep->de_StartCluster == MSDOSFSROOT
1582 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1583 #if 0
1584 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1585 		    offset);
1586 #endif
1587 		bias = 2 * sizeof(struct direntry);
1588 		if (offset < bias) {
1589 			for (n = (int)offset / sizeof(struct direntry);
1590 			     n < 2; n++) {
1591 				if (FAT32(pmp))
1592 					fileno = (uint64_t)cntobn(pmp,
1593 								 pmp->pm_rootdirblk)
1594 							  * dirsperblk;
1595 				else
1596 					fileno = 1;
1597 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1598 					dirbuf.d_fileno =
1599 					    msdosfs_fileno_map(pmp->pm_mountp,
1600 					    fileno);
1601 				} else {
1602 
1603 					dirbuf.d_fileno = (uint32_t)fileno;
1604 				}
1605 				dirbuf.d_type = DT_DIR;
1606 				switch (n) {
1607 				case 0:
1608 					dirbuf.d_namlen = 1;
1609 					strcpy(dirbuf.d_name, ".");
1610 					break;
1611 				case 1:
1612 					dirbuf.d_namlen = 2;
1613 					strcpy(dirbuf.d_name, "..");
1614 					break;
1615 				}
1616 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1617 				if (uio->uio_resid < dirbuf.d_reclen)
1618 					goto out;
1619 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1620 				if (error)
1621 					goto out;
1622 				offset += sizeof(struct direntry);
1623 				off = offset;
1624 				if (cookies) {
1625 					*cookies++ = offset;
1626 					if (--ncookies <= 0)
1627 						goto out;
1628 				}
1629 			}
1630 		}
1631 	}
1632 
1633 	mbnambuf_init(&nb);
1634 	off = offset;
1635 	while (uio->uio_resid > 0) {
1636 		lbn = de_cluster(pmp, offset - bias);
1637 		on = (offset - bias) & pmp->pm_crbomask;
1638 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1639 		diff = dep->de_FileSize - (offset - bias);
1640 		if (diff <= 0)
1641 			break;
1642 		n = min(n, diff);
1643 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1644 		if (error)
1645 			break;
1646 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1647 		if (error) {
1648 			brelse(bp);
1649 			return (error);
1650 		}
1651 		n = min(n, blsize - bp->b_resid);
1652 		if (n == 0) {
1653 			brelse(bp);
1654 			return (EIO);
1655 		}
1656 
1657 		/*
1658 		 * Convert from dos directory entries to fs-independent
1659 		 * directory entries.
1660 		 */
1661 		for (dentp = (struct direntry *)(bp->b_data + on);
1662 		     (char *)dentp < bp->b_data + on + n;
1663 		     dentp++, offset += sizeof(struct direntry)) {
1664 #if 0
1665 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1666 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1667 #endif
1668 			/*
1669 			 * If this is an unused entry, we can stop.
1670 			 */
1671 			if (dentp->deName[0] == SLOT_EMPTY) {
1672 				brelse(bp);
1673 				goto out;
1674 			}
1675 			/*
1676 			 * Skip deleted entries.
1677 			 */
1678 			if (dentp->deName[0] == SLOT_DELETED) {
1679 				chksum = -1;
1680 				mbnambuf_init(&nb);
1681 				continue;
1682 			}
1683 
1684 			/*
1685 			 * Handle Win95 long directory entries
1686 			 */
1687 			if (dentp->deAttributes == ATTR_WIN95) {
1688 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1689 					continue;
1690 				chksum = win2unixfn(&nb,
1691 				    (struct winentry *)dentp, chksum, pmp);
1692 				continue;
1693 			}
1694 
1695 			/*
1696 			 * Skip volume labels
1697 			 */
1698 			if (dentp->deAttributes & ATTR_VOLUME) {
1699 				chksum = -1;
1700 				mbnambuf_init(&nb);
1701 				continue;
1702 			}
1703 			/*
1704 			 * This computation of d_fileno must match
1705 			 * the computation of va_fileid in
1706 			 * msdosfs_getattr.
1707 			 */
1708 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1709 				fileno = getushort(dentp->deStartCluster);
1710 				if (FAT32(pmp))
1711 					fileno |= getushort(dentp->deHighClust) << 16;
1712 				/* if this is the root directory */
1713 				if (fileno == MSDOSFSROOT)
1714 					if (FAT32(pmp))
1715 						fileno = (uint64_t)cntobn(pmp,
1716 								pmp->pm_rootdirblk)
1717 							 * dirsperblk;
1718 					else
1719 						fileno = 1;
1720 				else
1721 					fileno = (uint64_t)cntobn(pmp, fileno) *
1722 					    dirsperblk;
1723 				dirbuf.d_type = DT_DIR;
1724 			} else {
1725 				fileno = (uoff_t)offset /
1726 				    sizeof(struct direntry);
1727 				dirbuf.d_type = DT_REG;
1728 			}
1729 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1730 				dirbuf.d_fileno =
1731 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1732 			} else
1733 				dirbuf.d_fileno = (uint32_t)fileno;
1734 
1735 			if (chksum != winChksum(dentp)) {
1736 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1737 				    (u_char *)dirbuf.d_name,
1738 				    dentp->deLowerCase |
1739 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1740 					(LCASE_BASE | LCASE_EXT) : 0),
1741 				    pmp);
1742 				mbnambuf_init(&nb);
1743 			} else
1744 				mbnambuf_flush(&nb, &dirbuf);
1745 			chksum = -1;
1746 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1747 			if (uio->uio_resid < dirbuf.d_reclen) {
1748 				brelse(bp);
1749 				goto out;
1750 			}
1751 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1752 			if (error) {
1753 				brelse(bp);
1754 				goto out;
1755 			}
1756 			if (cookies) {
1757 				*cookies++ = offset + sizeof(struct direntry);
1758 				if (--ncookies <= 0) {
1759 					brelse(bp);
1760 					goto out;
1761 				}
1762 			}
1763 			off = offset + sizeof(struct direntry);
1764 		}
1765 		brelse(bp);
1766 	}
1767 out:
1768 	/* Subtract unused cookies */
1769 	if (ap->a_ncookies)
1770 		*ap->a_ncookies -= ncookies;
1771 
1772 	uio->uio_offset = off;
1773 
1774 	/*
1775 	 * Set the eofflag (NFS uses it)
1776 	 */
1777 	if (ap->a_eofflag) {
1778 		if (dep->de_FileSize - (offset - bias) <= 0)
1779 			*ap->a_eofflag = 1;
1780 		else
1781 			*ap->a_eofflag = 0;
1782 	}
1783 	return (error);
1784 }
1785 
1786 /*-
1787  * a_vp   - pointer to the file's vnode
1788  * a_bn   - logical block number within the file (cluster number for us)
1789  * a_bop  - where to return the bufobj of the special file containing the fs
1790  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1791  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1792  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1793  *          blocks whose physical blocks (together with a_bn's physical block)
1794  *          are contiguous.
1795  * a_runb - where to return the "run before" a_bn.
1796  */
1797 static int
1798 msdosfs_bmap(ap)
1799 	struct vop_bmap_args /* {
1800 		struct vnode *a_vp;
1801 		daddr_t a_bn;
1802 		struct bufobj **a_bop;
1803 		daddr_t *a_bnp;
1804 		int *a_runp;
1805 		int *a_runb;
1806 	} */ *ap;
1807 {
1808 	struct denode *dep;
1809 	struct mount *mp;
1810 	struct msdosfsmount *pmp;
1811 	struct vnode *vp;
1812 	daddr_t runbn;
1813 	u_long cn;
1814 	int bnpercn, error, maxio, maxrun, run;
1815 
1816 	vp = ap->a_vp;
1817 	dep = VTODE(vp);
1818 	pmp = dep->de_pmp;
1819 	if (ap->a_bop != NULL)
1820 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1821 	if (ap->a_bnp == NULL)
1822 		return (0);
1823 	if (ap->a_runp != NULL)
1824 		*ap->a_runp = 0;
1825 	if (ap->a_runb != NULL)
1826 		*ap->a_runb = 0;
1827 	cn = ap->a_bn;
1828 	if (cn != ap->a_bn)
1829 		return (EFBIG);
1830 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1831 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1832 		return (error);
1833 
1834 	mp = vp->v_mount;
1835 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1836 	bnpercn = de_cn2bn(pmp, 1);
1837 	if (ap->a_runp != NULL) {
1838 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1839 		for (run = 1; run <= maxrun; run++) {
1840 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1841 			    runbn != *ap->a_bnp + run * bnpercn)
1842 				break;
1843 		}
1844 		*ap->a_runp = run - 1;
1845 	}
1846 	if (ap->a_runb != NULL) {
1847 		maxrun = ulmin(maxio - 1, cn);
1848 		for (run = 1; run < maxrun; run++) {
1849 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1850 			    runbn != *ap->a_bnp - run * bnpercn)
1851 				break;
1852 		}
1853 		*ap->a_runb = run - 1;
1854 	}
1855 	return (0);
1856 }
1857 
1858 static int
1859 msdosfs_strategy(ap)
1860 	struct vop_strategy_args /* {
1861 		struct vnode *a_vp;
1862 		struct buf *a_bp;
1863 	} */ *ap;
1864 {
1865 	struct buf *bp = ap->a_bp;
1866 	struct denode *dep = VTODE(ap->a_vp);
1867 	struct bufobj *bo;
1868 	int error = 0;
1869 	daddr_t blkno;
1870 
1871 	/*
1872 	 * If we don't already know the filesystem relative block number
1873 	 * then get it using pcbmap().  If pcbmap() returns the block
1874 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1875 	 * don't allow files with holes, so we shouldn't ever see this.
1876 	 */
1877 	if (bp->b_blkno == bp->b_lblkno) {
1878 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1879 		bp->b_blkno = blkno;
1880 		if (error) {
1881 			bp->b_error = error;
1882 			bp->b_ioflags |= BIO_ERROR;
1883 			bufdone(bp);
1884 			return (error);
1885 		}
1886 		if ((long)bp->b_blkno == -1)
1887 			vfs_bio_clrbuf(bp);
1888 	}
1889 	if (bp->b_blkno == -1) {
1890 		bufdone(bp);
1891 		return (0);
1892 	}
1893 	/*
1894 	 * Read/write the block from/to the disk that contains the desired
1895 	 * file block.
1896 	 */
1897 	bp->b_iooffset = dbtob(bp->b_blkno);
1898 	bo = dep->de_pmp->pm_bo;
1899 	BO_STRATEGY(bo, bp);
1900 	return (0);
1901 }
1902 
1903 static int
1904 msdosfs_print(ap)
1905 	struct vop_print_args /* {
1906 		struct vnode *vp;
1907 	} */ *ap;
1908 {
1909 	struct denode *dep = VTODE(ap->a_vp);
1910 
1911 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1912 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1913 	printf("on dev %s\n", devtoname(dep->de_dev));
1914 	return (0);
1915 }
1916 
1917 static int
1918 msdosfs_pathconf(ap)
1919 	struct vop_pathconf_args /* {
1920 		struct vnode *a_vp;
1921 		int a_name;
1922 		int *a_retval;
1923 	} */ *ap;
1924 {
1925 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1926 
1927 	switch (ap->a_name) {
1928 	case _PC_LINK_MAX:
1929 		*ap->a_retval = 1;
1930 		return (0);
1931 	case _PC_NAME_MAX:
1932 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1933 		return (0);
1934 	case _PC_PATH_MAX:
1935 		*ap->a_retval = PATH_MAX;
1936 		return (0);
1937 	case _PC_CHOWN_RESTRICTED:
1938 		*ap->a_retval = 1;
1939 		return (0);
1940 	case _PC_NO_TRUNC:
1941 		*ap->a_retval = 0;
1942 		return (0);
1943 	default:
1944 		return (EINVAL);
1945 	}
1946 	/* NOTREACHED */
1947 }
1948 
1949 static int
1950 msdosfs_vptofh(ap)
1951 	struct vop_vptofh_args /* {
1952 		struct vnode *a_vp;
1953 		struct fid *a_fhp;
1954 	} */ *ap;
1955 {
1956 	struct denode *dep;
1957 	struct defid *defhp;
1958 
1959 	dep = VTODE(ap->a_vp);
1960 	defhp = (struct defid *)ap->a_fhp;
1961 	defhp->defid_len = sizeof(struct defid);
1962 	defhp->defid_dirclust = dep->de_dirclust;
1963 	defhp->defid_dirofs = dep->de_diroffset;
1964 	/* defhp->defid_gen = dep->de_gen; */
1965 	return (0);
1966 }
1967 
1968 /* Global vfs data structures for msdosfs */
1969 struct vop_vector msdosfs_vnodeops = {
1970 	.vop_default =		&default_vnodeops,
1971 
1972 	.vop_access =		msdosfs_access,
1973 	.vop_bmap =		msdosfs_bmap,
1974 	.vop_cachedlookup =	msdosfs_lookup,
1975 	.vop_open =		msdosfs_open,
1976 	.vop_close =		msdosfs_close,
1977 	.vop_create =		msdosfs_create,
1978 	.vop_fsync =		msdosfs_fsync,
1979 	.vop_getattr =		msdosfs_getattr,
1980 	.vop_inactive =		msdosfs_inactive,
1981 	.vop_link =		msdosfs_link,
1982 	.vop_lookup =		vfs_cache_lookup,
1983 	.vop_mkdir =		msdosfs_mkdir,
1984 	.vop_mknod =		msdosfs_mknod,
1985 	.vop_pathconf =		msdosfs_pathconf,
1986 	.vop_print =		msdosfs_print,
1987 	.vop_read =		msdosfs_read,
1988 	.vop_readdir =		msdosfs_readdir,
1989 	.vop_reclaim =		msdosfs_reclaim,
1990 	.vop_remove =		msdosfs_remove,
1991 	.vop_rename =		msdosfs_rename,
1992 	.vop_rmdir =		msdosfs_rmdir,
1993 	.vop_setattr =		msdosfs_setattr,
1994 	.vop_strategy =		msdosfs_strategy,
1995 	.vop_symlink =		msdosfs_symlink,
1996 	.vop_write =		msdosfs_write,
1997 	.vop_vptofh =		msdosfs_vptofh,
1998 };
1999