xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision f05cddf9)
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 = (ap->a_vap->va_mode & VWRITE) ?
176 				ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
177 	ndirent.de_LowerCase = 0;
178 	ndirent.de_StartCluster = 0;
179 	ndirent.de_FileSize = 0;
180 	ndirent.de_pmp = pdep->de_pmp;
181 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
182 	getnanotime(&ts);
183 	DETIMES(&ndirent, &ts, &ts, &ts);
184 	error = createde(&ndirent, pdep, &dep, cnp);
185 	if (error)
186 		goto bad;
187 	*ap->a_vpp = DETOV(dep);
188 	return (0);
189 
190 bad:
191 	return (error);
192 }
193 
194 static int
195 msdosfs_mknod(ap)
196 	struct vop_mknod_args /* {
197 		struct vnode *a_dvp;
198 		struct vnode **a_vpp;
199 		struct componentname *a_cnp;
200 		struct vattr *a_vap;
201 	} */ *ap;
202 {
203 
204     return (EINVAL);
205 }
206 
207 static int
208 msdosfs_open(ap)
209 	struct vop_open_args /* {
210 		struct vnode *a_vp;
211 		int a_mode;
212 		struct ucred *a_cred;
213 		struct thread *a_td;
214 		struct file *a_fp;
215 	} */ *ap;
216 {
217 	struct denode *dep = VTODE(ap->a_vp);
218 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
219 	return 0;
220 }
221 
222 static int
223 msdosfs_close(ap)
224 	struct vop_close_args /* {
225 		struct vnode *a_vp;
226 		int a_fflag;
227 		struct ucred *a_cred;
228 		struct thread *a_td;
229 	} */ *ap;
230 {
231 	struct vnode *vp = ap->a_vp;
232 	struct denode *dep = VTODE(vp);
233 	struct timespec ts;
234 
235 	VI_LOCK(vp);
236 	if (vp->v_usecount > 1) {
237 		getnanotime(&ts);
238 		DETIMES(dep, &ts, &ts, &ts);
239 	}
240 	VI_UNLOCK(vp);
241 	return 0;
242 }
243 
244 static int
245 msdosfs_access(ap)
246 	struct vop_access_args /* {
247 		struct vnode *a_vp;
248 		accmode_t a_accmode;
249 		struct ucred *a_cred;
250 		struct thread *a_td;
251 	} */ *ap;
252 {
253 	struct vnode *vp = ap->a_vp;
254 	struct denode *dep = VTODE(ap->a_vp);
255 	struct msdosfsmount *pmp = dep->de_pmp;
256 	mode_t file_mode;
257 	accmode_t accmode = ap->a_accmode;
258 
259 	file_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
260 	    ((dep->de_Attributes & ATTR_READONLY) ? 0 : (S_IWUSR|S_IWGRP|S_IWOTH));
261 	file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
262 
263 	/*
264 	 * Disallow writing to directories and regular files if the
265 	 * filesystem is read-only.
266 	 */
267 	if (accmode & VWRITE) {
268 		switch (vp->v_type) {
269 		case VDIR:
270 		case VREG:
271 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
272 				return (EROFS);
273 			break;
274 		default:
275 			break;
276 		}
277 	}
278 
279 	return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
280 	    ap->a_accmode, ap->a_cred, NULL));
281 }
282 
283 static int
284 msdosfs_getattr(ap)
285 	struct vop_getattr_args /* {
286 		struct vnode *a_vp;
287 		struct vattr *a_vap;
288 		struct ucred *a_cred;
289 	} */ *ap;
290 {
291 	struct denode *dep = VTODE(ap->a_vp);
292 	struct msdosfsmount *pmp = dep->de_pmp;
293 	struct vattr *vap = ap->a_vap;
294 	mode_t mode;
295 	struct timespec ts;
296 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
297 	uint64_t fileid;
298 
299 	getnanotime(&ts);
300 	DETIMES(dep, &ts, &ts, &ts);
301 	vap->va_fsid = dev2udev(pmp->pm_dev);
302 	/*
303 	 * The following computation of the fileid must be the same as that
304 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
305 	 * doesn't work.
306 	 */
307 	if (dep->de_Attributes & ATTR_DIRECTORY) {
308 		fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
309 		    dirsperblk;
310 		if (dep->de_StartCluster == MSDOSFSROOT)
311 			fileid = 1;
312 	} else {
313 		fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
314 		    dirsperblk;
315 		if (dep->de_dirclust == MSDOSFSROOT)
316 			fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
317 		fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
318 	}
319 
320 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
321 		vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
322 	else
323 		vap->va_fileid = (long)fileid;
324 
325 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
326 		mode = S_IRWXU|S_IRWXG|S_IRWXO;
327 	else
328 		mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
329 	vap->va_mode = mode &
330 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
331 	vap->va_uid = pmp->pm_uid;
332 	vap->va_gid = pmp->pm_gid;
333 	vap->va_nlink = 1;
334 	vap->va_rdev = NODEV;
335 	vap->va_size = dep->de_FileSize;
336 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
337 	vap->va_ctime = vap->va_mtime;
338 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
339 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
340 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
341 		    0, &vap->va_birthtime);
342 	} else {
343 		vap->va_atime = vap->va_mtime;
344 		vap->va_birthtime.tv_sec = -1;
345 		vap->va_birthtime.tv_nsec = 0;
346 	}
347 	vap->va_flags = 0;
348 	if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
349 		vap->va_flags |= SF_ARCHIVED;
350 	vap->va_gen = 0;
351 	vap->va_blocksize = pmp->pm_bpcluster;
352 	vap->va_bytes =
353 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
354 	vap->va_type = ap->a_vp->v_type;
355 	vap->va_filerev = dep->de_modrev;
356 	return (0);
357 }
358 
359 static int
360 msdosfs_setattr(ap)
361 	struct vop_setattr_args /* {
362 		struct vnode *a_vp;
363 		struct vattr *a_vap;
364 		struct ucred *a_cred;
365 	} */ *ap;
366 {
367 	struct vnode *vp = ap->a_vp;
368 	struct denode *dep = VTODE(ap->a_vp);
369 	struct msdosfsmount *pmp = dep->de_pmp;
370 	struct vattr *vap = ap->a_vap;
371 	struct ucred *cred = ap->a_cred;
372 	struct thread *td = curthread;
373 	int error = 0;
374 
375 #ifdef MSDOSFS_DEBUG
376 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
377 	    ap->a_vp, vap, cred);
378 #endif
379 
380 	/*
381 	 * Check for unsettable attributes.
382 	 */
383 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
384 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
385 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
386 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
387 #ifdef MSDOSFS_DEBUG
388 		printf("msdosfs_setattr(): returning EINVAL\n");
389 		printf("    va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
390 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
391 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
392 		    vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
393 		printf("    va_uid %x, va_gid %x\n",
394 		    vap->va_uid, vap->va_gid);
395 #endif
396 		return (EINVAL);
397 	}
398 	if (vap->va_flags != VNOVAL) {
399 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
400 			return (EROFS);
401 		if (cred->cr_uid != pmp->pm_uid) {
402 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
403 			if (error)
404 				return (error);
405 		}
406 		/*
407 		 * We are very inconsistent about handling unsupported
408 		 * attributes.  We ignored the access time and the
409 		 * read and execute bits.  We were strict for the other
410 		 * attributes.
411 		 *
412 		 * Here we are strict, stricter than ufs in not allowing
413 		 * users to attempt to set SF_SETTABLE bits or anyone to
414 		 * set unsupported bits.  However, we ignore attempts to
415 		 * set ATTR_ARCHIVE for directories `cp -pr' from a more
416 		 * sensible filesystem attempts it a lot.
417 		 */
418 		if (vap->va_flags & SF_SETTABLE) {
419 			error = priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0);
420 			if (error)
421 				return (error);
422 		}
423 		if (vap->va_flags & ~SF_ARCHIVED)
424 			return EOPNOTSUPP;
425 		if (vap->va_flags & SF_ARCHIVED)
426 			dep->de_Attributes &= ~ATTR_ARCHIVE;
427 		else if (!(dep->de_Attributes & ATTR_DIRECTORY))
428 			dep->de_Attributes |= ATTR_ARCHIVE;
429 		dep->de_flag |= DE_MODIFIED;
430 	}
431 
432 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
433 		uid_t uid;
434 		gid_t gid;
435 
436 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
437 			return (EROFS);
438 		uid = vap->va_uid;
439 		if (uid == (uid_t)VNOVAL)
440 			uid = pmp->pm_uid;
441 		gid = vap->va_gid;
442 		if (gid == (gid_t)VNOVAL)
443 			gid = pmp->pm_gid;
444 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
445 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
446 			error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
447 			if (error)
448 				return (error);
449 		}
450 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
451 			return EINVAL;
452 	}
453 
454 	if (vap->va_size != VNOVAL) {
455 		switch (vp->v_type) {
456 		case VDIR:
457 			return (EISDIR);
458 		case VREG:
459 			/*
460 			 * Truncation is only supported for regular files,
461 			 * Disallow it if the filesystem is read-only.
462 			 */
463 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
464 				return (EROFS);
465 			break;
466 		default:
467 			/*
468 			 * According to POSIX, the result is unspecified
469 			 * for file types other than regular files,
470 			 * directories and shared memory objects.  We
471 			 * don't support any file types except regular
472 			 * files and directories in this file system, so
473 			 * this (default) case is unreachable and can do
474 			 * anything.  Keep falling through to detrunc()
475 			 * for now.
476 			 */
477 			break;
478 		}
479 		error = detrunc(dep, vap->va_size, 0, cred);
480 		if (error)
481 			return error;
482 	}
483 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
484 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
485 			return (EROFS);
486 		if (vap->va_vaflags & VA_UTIMES_NULL) {
487 			error = VOP_ACCESS(vp, VADMIN, cred, td);
488 			if (error)
489 				error = VOP_ACCESS(vp, VWRITE, cred, td);
490 		} else
491 			error = VOP_ACCESS(vp, VADMIN, cred, td);
492 		if (vp->v_type != VDIR) {
493 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
494 			    vap->va_atime.tv_sec != VNOVAL) {
495 				dep->de_flag &= ~DE_ACCESS;
496 				timespec2fattime(&vap->va_atime, 0,
497 				    &dep->de_ADate, NULL, NULL);
498 			}
499 			if (vap->va_mtime.tv_sec != VNOVAL) {
500 				dep->de_flag &= ~DE_UPDATE;
501 				timespec2fattime(&vap->va_mtime, 0,
502 				    &dep->de_MDate, &dep->de_MTime, NULL);
503 			}
504 			dep->de_Attributes |= ATTR_ARCHIVE;
505 			dep->de_flag |= DE_MODIFIED;
506 		}
507 	}
508 	/*
509 	 * DOS files only have the ability to have their writability
510 	 * attribute set, so we use the owner write bit to set the readonly
511 	 * attribute.
512 	 */
513 	if (vap->va_mode != (mode_t)VNOVAL) {
514 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
515 			return (EROFS);
516 		if (cred->cr_uid != pmp->pm_uid) {
517 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
518 			if (error)
519 				return (error);
520 		}
521 		if (vp->v_type != VDIR) {
522 			/* We ignore the read and execute bits. */
523 			if (vap->va_mode & VWRITE)
524 				dep->de_Attributes &= ~ATTR_READONLY;
525 			else
526 				dep->de_Attributes |= ATTR_READONLY;
527 			dep->de_Attributes |= ATTR_ARCHIVE;
528 			dep->de_flag |= DE_MODIFIED;
529 		}
530 	}
531 	return (deupdat(dep, 0));
532 }
533 
534 static int
535 msdosfs_read(ap)
536 	struct vop_read_args /* {
537 		struct vnode *a_vp;
538 		struct uio *a_uio;
539 		int a_ioflag;
540 		struct ucred *a_cred;
541 	} */ *ap;
542 {
543 	int error = 0;
544 	int blsize;
545 	int isadir;
546 	ssize_t orig_resid;
547 	u_int n;
548 	u_long diff;
549 	u_long on;
550 	daddr_t lbn;
551 	daddr_t rablock;
552 	int rasize;
553 	int seqcount;
554 	struct buf *bp;
555 	struct vnode *vp = ap->a_vp;
556 	struct denode *dep = VTODE(vp);
557 	struct msdosfsmount *pmp = dep->de_pmp;
558 	struct uio *uio = ap->a_uio;
559 
560 	/*
561 	 * If they didn't ask for any data, then we are done.
562 	 */
563 	orig_resid = uio->uio_resid;
564 	if (orig_resid == 0)
565 		return (0);
566 
567 	/*
568 	 * The caller is supposed to ensure that
569 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
570 	 * We don't need to check for large offsets as in ffs because
571 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
572 	 * offsets cannot cause overflow even in theory.
573 	 */
574 
575 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
576 
577 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
578 	do {
579 		if (uio->uio_offset >= dep->de_FileSize)
580 			break;
581 		lbn = de_cluster(pmp, uio->uio_offset);
582 		rablock = lbn + 1;
583 		blsize = pmp->pm_bpcluster;
584 		on = uio->uio_offset & pmp->pm_crbomask;
585 		/*
586 		 * If we are operating on a directory file then be sure to
587 		 * do i/o with the vnode for the filesystem instead of the
588 		 * vnode for the directory.
589 		 */
590 		if (isadir) {
591 			/* convert cluster # to block # */
592 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
593 			if (error == E2BIG) {
594 				error = EINVAL;
595 				break;
596 			} else if (error)
597 				break;
598 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
599 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
600 			error = bread(vp, lbn, blsize, NOCRED, &bp);
601 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
602 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
603 			    NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
604 		} else if (seqcount > 1) {
605 			rasize = blsize;
606 			error = breadn(vp, lbn,
607 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
608 		} else {
609 			error = bread(vp, lbn, blsize, NOCRED, &bp);
610 		}
611 		if (error) {
612 			brelse(bp);
613 			break;
614 		}
615 		diff = pmp->pm_bpcluster - on;
616 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
617 		diff = dep->de_FileSize - uio->uio_offset;
618 		if (diff < n)
619 			n = diff;
620 		diff = blsize - bp->b_resid;
621 		if (diff < n)
622 			n = diff;
623 		error = uiomove(bp->b_data + on, (int) n, uio);
624 		brelse(bp);
625 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
626 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
627 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
628 		dep->de_flag |= DE_ACCESS;
629 	return (error);
630 }
631 
632 /*
633  * Write data to a file or directory.
634  */
635 static int
636 msdosfs_write(ap)
637 	struct vop_write_args /* {
638 		struct vnode *a_vp;
639 		struct uio *a_uio;
640 		int a_ioflag;
641 		struct ucred *a_cred;
642 	} */ *ap;
643 {
644 	int n;
645 	int croffset;
646 	ssize_t resid;
647 	u_long osize;
648 	int error = 0;
649 	u_long count;
650 	int seqcount;
651 	daddr_t bn, lastcn;
652 	struct buf *bp;
653 	int ioflag = ap->a_ioflag;
654 	struct uio *uio = ap->a_uio;
655 	struct vnode *vp = ap->a_vp;
656 	struct vnode *thisvp;
657 	struct denode *dep = VTODE(vp);
658 	struct msdosfsmount *pmp = dep->de_pmp;
659 	struct ucred *cred = ap->a_cred;
660 
661 #ifdef MSDOSFS_DEBUG
662 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
663 	    vp, uio, ioflag, cred);
664 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
665 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
666 #endif
667 
668 	switch (vp->v_type) {
669 	case VREG:
670 		if (ioflag & IO_APPEND)
671 			uio->uio_offset = dep->de_FileSize;
672 		thisvp = vp;
673 		break;
674 	case VDIR:
675 		return EISDIR;
676 	default:
677 		panic("msdosfs_write(): bad file type");
678 	}
679 
680 	/*
681 	 * This is needed (unlike in ffs_write()) because we extend the
682 	 * file outside of the loop but we don't want to extend the file
683 	 * for writes of 0 bytes.
684 	 */
685 	if (uio->uio_resid == 0)
686 		return (0);
687 
688 	/*
689 	 * The caller is supposed to ensure that
690 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
691 	 */
692 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
693 		return (EFBIG);
694 
695 	/*
696 	 * If they've exceeded their filesize limit, tell them about it.
697 	 */
698 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
699 		return (EFBIG);
700 
701 	/*
702 	 * If the offset we are starting the write at is beyond the end of
703 	 * the file, then they've done a seek.  Unix filesystems allow
704 	 * files with holes in them, DOS doesn't so we must fill the hole
705 	 * with zeroed blocks.
706 	 */
707 	if (uio->uio_offset > dep->de_FileSize) {
708 		error = deextend(dep, uio->uio_offset, cred);
709 		if (error)
710 			return (error);
711 	}
712 
713 	/*
714 	 * Remember some values in case the write fails.
715 	 */
716 	resid = uio->uio_resid;
717 	osize = dep->de_FileSize;
718 
719 	/*
720 	 * If we write beyond the end of the file, extend it to its ultimate
721 	 * size ahead of the time to hopefully get a contiguous area.
722 	 */
723 	if (uio->uio_offset + resid > osize) {
724 		count = de_clcount(pmp, uio->uio_offset + resid) -
725 			de_clcount(pmp, osize);
726 		error = extendfile(dep, count, NULL, NULL, 0);
727 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
728 			goto errexit;
729 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
730 	} else
731 		lastcn = de_clcount(pmp, osize) - 1;
732 
733 	seqcount = ioflag >> IO_SEQSHIFT;
734 	do {
735 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
736 			error = ENOSPC;
737 			break;
738 		}
739 
740 		croffset = uio->uio_offset & pmp->pm_crbomask;
741 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
742 		if (uio->uio_offset + n > dep->de_FileSize) {
743 			dep->de_FileSize = uio->uio_offset + n;
744 			/* The object size needs to be set before buffer is allocated */
745 			vnode_pager_setsize(vp, dep->de_FileSize);
746 		}
747 
748 		bn = de_cluster(pmp, uio->uio_offset);
749 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
750 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
751 			> de_cluster(pmp, uio->uio_offset)
752 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
753 			/*
754 			 * If either the whole cluster gets written,
755 			 * or we write the cluster from its start beyond EOF,
756 			 * then no need to read data from disk.
757 			 */
758 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
759 			vfs_bio_clrbuf(bp);
760 			/*
761 			 * Do the bmap now, since pcbmap needs buffers
762 			 * for the fat table. (see msdosfs_strategy)
763 			 */
764 			if (bp->b_blkno == bp->b_lblkno) {
765 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
766 				if (error)
767 					bp->b_blkno = -1;
768 				else
769 					bp->b_blkno = bn;
770 			}
771 			if (bp->b_blkno == -1) {
772 				brelse(bp);
773 				if (!error)
774 					error = EIO;		/* XXX */
775 				break;
776 			}
777 		} else {
778 			/*
779 			 * The block we need to write into exists, so read it in.
780 			 */
781 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
782 			if (error) {
783 				brelse(bp);
784 				break;
785 			}
786 		}
787 
788 		/*
789 		 * Should these vnode_pager_* functions be done on dir
790 		 * files?
791 		 */
792 
793 		/*
794 		 * Copy the data from user space into the buf header.
795 		 */
796 		error = uiomove(bp->b_data + croffset, n, uio);
797 		if (error) {
798 			brelse(bp);
799 			break;
800 		}
801 
802 		/* Prepare for clustered writes in some else clauses. */
803 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
804 			bp->b_flags |= B_CLUSTEROK;
805 
806 		/*
807 		 * If IO_SYNC, then each buffer is written synchronously.
808 		 * Otherwise, if we have a severe page deficiency then
809 		 * write the buffer asynchronously.  Otherwise, if on a
810 		 * cluster boundary then write the buffer asynchronously,
811 		 * combining it with contiguous clusters if permitted and
812 		 * possible, since we don't expect more writes into this
813 		 * buffer soon.  Otherwise, do a delayed write because we
814 		 * expect more writes into this buffer soon.
815 		 */
816 		if (ioflag & IO_SYNC)
817 			(void)bwrite(bp);
818 		else if (vm_page_count_severe() || buf_dirty_count_severe())
819 			bawrite(bp);
820 		else if (n + croffset == pmp->pm_bpcluster) {
821 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
822 				cluster_write(vp, bp, dep->de_FileSize,
823 				    seqcount, 0);
824 			else
825 				bawrite(bp);
826 		} else
827 			bdwrite(bp);
828 		dep->de_flag |= DE_UPDATE;
829 	} while (error == 0 && uio->uio_resid > 0);
830 
831 	/*
832 	 * If the write failed and they want us to, truncate the file back
833 	 * to the size it was before the write was attempted.
834 	 */
835 errexit:
836 	if (error) {
837 		if (ioflag & IO_UNIT) {
838 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
839 			uio->uio_offset -= resid - uio->uio_resid;
840 			uio->uio_resid = resid;
841 		} else {
842 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
843 			if (uio->uio_resid != resid)
844 				error = 0;
845 		}
846 	} else if (ioflag & IO_SYNC)
847 		error = deupdat(dep, 1);
848 	return (error);
849 }
850 
851 /*
852  * Flush the blocks of a file to disk.
853  */
854 static int
855 msdosfs_fsync(ap)
856 	struct vop_fsync_args /* {
857 		struct vnode *a_vp;
858 		struct ucred *a_cred;
859 		int a_waitfor;
860 		struct thread *a_td;
861 	} */ *ap;
862 {
863 	struct vnode *devvp;
864 	int allerror, error;
865 
866 	vop_stdfsync(ap);
867 
868 	/*
869 	* If the syncing request comes from fsync(2), sync the entire
870 	* FAT and any other metadata that happens to be on devvp.  We
871 	* need this mainly for the FAT.  We write the FAT sloppily, and
872 	* syncing it all now is the best we can easily do to get all
873 	* directory entries associated with the file (not just the file)
874 	* fully synced.  The other metadata includes critical metadata
875 	* for all directory entries, but only in the MNT_ASYNC case.  We
876 	* will soon sync all metadata in the file's directory entry.
877 	* Non-critical metadata for associated directory entries only
878 	* gets synced accidentally, as in most file systems.
879 	*/
880 	if (ap->a_waitfor == MNT_WAIT) {
881 		devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
882 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
883 		allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
884 		VOP_UNLOCK(devvp, 0);
885 	} else
886 		allerror = 0;
887 
888 	error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
889 	if (allerror == 0)
890 		allerror = error;
891 	return (allerror);
892 }
893 
894 static int
895 msdosfs_remove(ap)
896 	struct vop_remove_args /* {
897 		struct vnode *a_dvp;
898 		struct vnode *a_vp;
899 		struct componentname *a_cnp;
900 	} */ *ap;
901 {
902 	struct denode *dep = VTODE(ap->a_vp);
903 	struct denode *ddep = VTODE(ap->a_dvp);
904 	int error;
905 
906 	if (ap->a_vp->v_type == VDIR)
907 		error = EPERM;
908 	else
909 		error = removede(ddep, dep);
910 #ifdef MSDOSFS_DEBUG
911 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
912 #endif
913 	return (error);
914 }
915 
916 /*
917  * DOS filesystems don't know what links are.
918  */
919 static int
920 msdosfs_link(ap)
921 	struct vop_link_args /* {
922 		struct vnode *a_tdvp;
923 		struct vnode *a_vp;
924 		struct componentname *a_cnp;
925 	} */ *ap;
926 {
927 	return (EOPNOTSUPP);
928 }
929 
930 /*
931  * Renames on files require moving the denode to a new hash queue since the
932  * denode's location is used to compute which hash queue to put the file
933  * in. Unless it is a rename in place.  For example "mv a b".
934  *
935  * What follows is the basic algorithm:
936  *
937  * if (file move) {
938  *	if (dest file exists) {
939  *		remove dest file
940  *	}
941  *	if (dest and src in same directory) {
942  *		rewrite name in existing directory slot
943  *	} else {
944  *		write new entry in dest directory
945  *		update offset and dirclust in denode
946  *		move denode to new hash chain
947  *		clear old directory entry
948  *	}
949  * } else {
950  *	directory move
951  *	if (dest directory exists) {
952  *		if (dest is not empty) {
953  *			return ENOTEMPTY
954  *		}
955  *		remove dest directory
956  *	}
957  *	if (dest and src in same directory) {
958  *		rewrite name in existing entry
959  *	} else {
960  *		be sure dest is not a child of src directory
961  *		write entry in dest directory
962  *		update "." and ".." in moved directory
963  *		clear old directory entry for moved directory
964  *	}
965  * }
966  *
967  * On entry:
968  *	source's parent directory is unlocked
969  *	source file or directory is unlocked
970  *	destination's parent directory is locked
971  *	destination file or directory is locked if it exists
972  *
973  * On exit:
974  *	all denodes should be released
975  */
976 static int
977 msdosfs_rename(ap)
978 	struct vop_rename_args /* {
979 		struct vnode *a_fdvp;
980 		struct vnode *a_fvp;
981 		struct componentname *a_fcnp;
982 		struct vnode *a_tdvp;
983 		struct vnode *a_tvp;
984 		struct componentname *a_tcnp;
985 	} */ *ap;
986 {
987 	struct vnode *tdvp = ap->a_tdvp;
988 	struct vnode *fvp = ap->a_fvp;
989 	struct vnode *fdvp = ap->a_fdvp;
990 	struct vnode *tvp = ap->a_tvp;
991 	struct componentname *tcnp = ap->a_tcnp;
992 	struct componentname *fcnp = ap->a_fcnp;
993 	struct denode *ip, *xp, *dp, *zp;
994 	u_char toname[12], oldname[11];
995 	u_long from_diroffset, to_diroffset;
996 	u_char to_count;
997 	int doingdirectory = 0, newparent = 0;
998 	int error;
999 	u_long cn, pcl;
1000 	daddr_t bn;
1001 	struct denode *fddep;	/* from file's parent directory	 */
1002 	struct msdosfsmount *pmp;
1003 	struct direntry *dotdotp;
1004 	struct buf *bp;
1005 
1006 	fddep = VTODE(ap->a_fdvp);
1007 	pmp = fddep->de_pmp;
1008 
1009 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1010 
1011 #ifdef DIAGNOSTIC
1012 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1013 	    (fcnp->cn_flags & HASBUF) == 0)
1014 		panic("msdosfs_rename: no name");
1015 #endif
1016 	/*
1017 	 * Check for cross-device rename.
1018 	 */
1019 	if (fvp->v_mount != tdvp->v_mount ||
1020 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1021 		error = EXDEV;
1022 abortit:
1023 		if (tdvp == tvp)
1024 			vrele(tdvp);
1025 		else
1026 			vput(tdvp);
1027 		if (tvp)
1028 			vput(tvp);
1029 		vrele(fdvp);
1030 		vrele(fvp);
1031 		return (error);
1032 	}
1033 
1034 	/*
1035 	 * If source and dest are the same, do nothing.
1036 	 */
1037 	if (tvp == fvp) {
1038 		error = 0;
1039 		goto abortit;
1040 	}
1041 
1042 	error = vn_lock(fvp, LK_EXCLUSIVE);
1043 	if (error)
1044 		goto abortit;
1045 	dp = VTODE(fdvp);
1046 	ip = VTODE(fvp);
1047 
1048 	/*
1049 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1050 	 * leads to a crippled directory tree.  It's pretty tough to do a
1051 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1052 	 * doesn't work if the ".." entry is missing.
1053 	 */
1054 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1055 		/*
1056 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1057 		 */
1058 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1059 		    dp == ip ||
1060 		    (fcnp->cn_flags & ISDOTDOT) ||
1061 		    (tcnp->cn_flags & ISDOTDOT) ||
1062 		    (ip->de_flag & DE_RENAME)) {
1063 			VOP_UNLOCK(fvp, 0);
1064 			error = EINVAL;
1065 			goto abortit;
1066 		}
1067 		ip->de_flag |= DE_RENAME;
1068 		doingdirectory++;
1069 	}
1070 
1071 	/*
1072 	 * When the target exists, both the directory
1073 	 * and target vnodes are returned locked.
1074 	 */
1075 	dp = VTODE(tdvp);
1076 	xp = tvp ? VTODE(tvp) : NULL;
1077 	/*
1078 	 * Remember direntry place to use for destination
1079 	 */
1080 	to_diroffset = dp->de_fndoffset;
1081 	to_count = dp->de_fndcnt;
1082 
1083 	/*
1084 	 * If ".." must be changed (ie the directory gets a new
1085 	 * parent) then the source directory must not be in the
1086 	 * directory hierarchy above the target, as this would
1087 	 * orphan everything below the source directory. Also
1088 	 * the user must have write permission in the source so
1089 	 * as to be able to change "..". We must repeat the call
1090 	 * to namei, as the parent directory is unlocked by the
1091 	 * call to doscheckpath().
1092 	 */
1093 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1094 	VOP_UNLOCK(fvp, 0);
1095 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1096 		newparent = 1;
1097 	if (doingdirectory && newparent) {
1098 		if (error)	/* write access check above */
1099 			goto bad;
1100 		if (xp != NULL)
1101 			vput(tvp);
1102 		/*
1103 		 * doscheckpath() vput()'s dp,
1104 		 * so we have to do a relookup afterwards
1105 		 */
1106 		error = doscheckpath(ip, dp);
1107 		if (error)
1108 			goto out;
1109 		if ((tcnp->cn_flags & SAVESTART) == 0)
1110 			panic("msdosfs_rename: lost to startdir");
1111 		error = relookup(tdvp, &tvp, tcnp);
1112 		if (error)
1113 			goto out;
1114 		dp = VTODE(tdvp);
1115 		xp = tvp ? VTODE(tvp) : NULL;
1116 	}
1117 
1118 	if (xp != NULL) {
1119 		/*
1120 		 * Target must be empty if a directory and have no links
1121 		 * to it. Also, ensure source and target are compatible
1122 		 * (both directories, or both not directories).
1123 		 */
1124 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1125 			if (!dosdirempty(xp)) {
1126 				error = ENOTEMPTY;
1127 				goto bad;
1128 			}
1129 			if (!doingdirectory) {
1130 				error = ENOTDIR;
1131 				goto bad;
1132 			}
1133 			cache_purge(tdvp);
1134 		} else if (doingdirectory) {
1135 			error = EISDIR;
1136 			goto bad;
1137 		}
1138 		error = removede(dp, xp);
1139 		if (error)
1140 			goto bad;
1141 		vput(tvp);
1142 		xp = NULL;
1143 	}
1144 
1145 	/*
1146 	 * Convert the filename in tcnp into a dos filename. We copy this
1147 	 * into the denode and directory entry for the destination
1148 	 * file/directory.
1149 	 */
1150 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1151 	if (error)
1152 		goto abortit;
1153 
1154 	/*
1155 	 * Since from wasn't locked at various places above,
1156 	 * have to do a relookup here.
1157 	 */
1158 	fcnp->cn_flags &= ~MODMASK;
1159 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1160 	if ((fcnp->cn_flags & SAVESTART) == 0)
1161 		panic("msdosfs_rename: lost from startdir");
1162 	if (!newparent)
1163 		VOP_UNLOCK(tdvp, 0);
1164 	if (relookup(fdvp, &fvp, fcnp) == 0)
1165 		vrele(fdvp);
1166 	if (fvp == NULL) {
1167 		/*
1168 		 * From name has disappeared.
1169 		 */
1170 		if (doingdirectory)
1171 			panic("rename: lost dir entry");
1172 		if (newparent)
1173 			VOP_UNLOCK(tdvp, 0);
1174 		vrele(tdvp);
1175 		vrele(ap->a_fvp);
1176 		return 0;
1177 	}
1178 	xp = VTODE(fvp);
1179 	zp = VTODE(fdvp);
1180 	from_diroffset = zp->de_fndoffset;
1181 
1182 	/*
1183 	 * Ensure that the directory entry still exists and has not
1184 	 * changed till now. If the source is a file the entry may
1185 	 * have been unlinked or renamed. In either case there is
1186 	 * no further work to be done. If the source is a directory
1187 	 * then it cannot have been rmdir'ed or renamed; this is
1188 	 * prohibited by the DE_RENAME flag.
1189 	 */
1190 	if (xp != ip) {
1191 		if (doingdirectory)
1192 			panic("rename: lost dir entry");
1193 		VOP_UNLOCK(fvp, 0);
1194 		if (newparent)
1195 			VOP_UNLOCK(fdvp, 0);
1196 		vrele(ap->a_fvp);
1197 		xp = NULL;
1198 	} else {
1199 		vrele(fvp);
1200 		xp = NULL;
1201 
1202 		/*
1203 		 * First write a new entry in the destination
1204 		 * directory and mark the entry in the source directory
1205 		 * as deleted.  Then move the denode to the correct hash
1206 		 * chain for its new location in the filesystem.  And, if
1207 		 * we moved a directory, then update its .. entry to point
1208 		 * to the new parent directory.
1209 		 */
1210 		bcopy(ip->de_Name, oldname, 11);
1211 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1212 		dp->de_fndoffset = to_diroffset;
1213 		dp->de_fndcnt = to_count;
1214 		error = createde(ip, dp, (struct denode **)0, tcnp);
1215 		if (error) {
1216 			bcopy(oldname, ip->de_Name, 11);
1217 			if (newparent)
1218 				VOP_UNLOCK(fdvp, 0);
1219 			VOP_UNLOCK(fvp, 0);
1220 			goto bad;
1221 		}
1222 		ip->de_refcnt++;
1223 		zp->de_fndoffset = from_diroffset;
1224 		error = removede(zp, ip);
1225 		if (error) {
1226 			/* XXX should downgrade to ro here, fs is corrupt */
1227 			if (newparent)
1228 				VOP_UNLOCK(fdvp, 0);
1229 			VOP_UNLOCK(fvp, 0);
1230 			goto bad;
1231 		}
1232 		if (!doingdirectory) {
1233 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1234 				       &ip->de_dirclust, 0);
1235 			if (error) {
1236 				/* XXX should downgrade to ro here, fs is corrupt */
1237 				if (newparent)
1238 					VOP_UNLOCK(fdvp, 0);
1239 				VOP_UNLOCK(fvp, 0);
1240 				goto bad;
1241 			}
1242 			if (ip->de_dirclust == MSDOSFSROOT)
1243 				ip->de_diroffset = to_diroffset;
1244 			else
1245 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1246 		}
1247 		reinsert(ip);
1248 		if (newparent)
1249 			VOP_UNLOCK(fdvp, 0);
1250 	}
1251 
1252 	/*
1253 	 * If we moved a directory to a new parent directory, then we must
1254 	 * fixup the ".." entry in the moved directory.
1255 	 */
1256 	if (doingdirectory && newparent) {
1257 		cn = ip->de_StartCluster;
1258 		if (cn == MSDOSFSROOT) {
1259 			/* this should never happen */
1260 			panic("msdosfs_rename(): updating .. in root directory?");
1261 		} else
1262 			bn = cntobn(pmp, cn);
1263 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1264 			      NOCRED, &bp);
1265 		if (error) {
1266 			/* XXX should downgrade to ro here, fs is corrupt */
1267 			brelse(bp);
1268 			VOP_UNLOCK(fvp, 0);
1269 			goto bad;
1270 		}
1271 		dotdotp = (struct direntry *)bp->b_data + 1;
1272 		pcl = dp->de_StartCluster;
1273 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1274 			pcl = MSDOSFSROOT;
1275 		putushort(dotdotp->deStartCluster, pcl);
1276 		if (FAT32(pmp))
1277 			putushort(dotdotp->deHighClust, pcl >> 16);
1278 		if (DOINGASYNC(fvp))
1279 			bdwrite(bp);
1280 		else if ((error = bwrite(bp)) != 0) {
1281 			/* XXX should downgrade to ro here, fs is corrupt */
1282 			VOP_UNLOCK(fvp, 0);
1283 			goto bad;
1284 		}
1285 	}
1286 
1287 	/*
1288 	 * The msdosfs lookup is case insensitive. Several aliases may
1289 	 * be inserted for a single directory entry. As a consequnce,
1290 	 * name cache purge done by lookup for fvp when DELETE op for
1291 	 * namei is specified, might be not enough to expunge all
1292 	 * namecache entries that were installed for this direntry.
1293 	 */
1294 	cache_purge(fvp);
1295 	VOP_UNLOCK(fvp, 0);
1296 bad:
1297 	if (xp)
1298 		vput(tvp);
1299 	vput(tdvp);
1300 out:
1301 	ip->de_flag &= ~DE_RENAME;
1302 	vrele(fdvp);
1303 	vrele(fvp);
1304 	return (error);
1305 
1306 }
1307 
1308 static struct {
1309 	struct direntry dot;
1310 	struct direntry dotdot;
1311 } dosdirtemplate = {
1312 	{	".          ",				/* the . entry */
1313 		ATTR_DIRECTORY,				/* file attribute */
1314 		0,					/* reserved */
1315 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1316 		{ 0, 0 },				/* access date */
1317 		{ 0, 0 },				/* high bits of start cluster */
1318 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1319 		{ 0, 0 },				/* startcluster */
1320 		{ 0, 0, 0, 0 }				/* filesize */
1321 	},
1322 	{	"..         ",				/* the .. entry */
1323 		ATTR_DIRECTORY,				/* file attribute */
1324 		0,					/* reserved */
1325 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1326 		{ 0, 0 },				/* access date */
1327 		{ 0, 0 },				/* high bits of start cluster */
1328 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1329 		{ 0, 0 },				/* startcluster */
1330 		{ 0, 0, 0, 0 }				/* filesize */
1331 	}
1332 };
1333 
1334 static int
1335 msdosfs_mkdir(ap)
1336 	struct vop_mkdir_args /* {
1337 		struct vnode *a_dvp;
1338 		struct vnode **a_vpp;
1339 		struvt componentname *a_cnp;
1340 		struct vattr *a_vap;
1341 	} */ *ap;
1342 {
1343 	struct componentname *cnp = ap->a_cnp;
1344 	struct denode *dep;
1345 	struct denode *pdep = VTODE(ap->a_dvp);
1346 	struct direntry *denp;
1347 	struct msdosfsmount *pmp = pdep->de_pmp;
1348 	struct buf *bp;
1349 	u_long newcluster, pcl;
1350 	int bn;
1351 	int error;
1352 	struct denode ndirent;
1353 	struct timespec ts;
1354 
1355 	/*
1356 	 * If this is the root directory and there is no space left we
1357 	 * can't do anything.  This is because the root directory can not
1358 	 * change size.
1359 	 */
1360 	if (pdep->de_StartCluster == MSDOSFSROOT
1361 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1362 		error = ENOSPC;
1363 		goto bad2;
1364 	}
1365 
1366 	/*
1367 	 * Allocate a cluster to hold the about to be created directory.
1368 	 */
1369 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1370 	if (error)
1371 		goto bad2;
1372 
1373 	bzero(&ndirent, sizeof(ndirent));
1374 	ndirent.de_pmp = pmp;
1375 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1376 	getnanotime(&ts);
1377 	DETIMES(&ndirent, &ts, &ts, &ts);
1378 
1379 	/*
1380 	 * Now fill the cluster with the "." and ".." entries. And write
1381 	 * the cluster to disk.  This way it is there for the parent
1382 	 * directory to be pointing at if there were a crash.
1383 	 */
1384 	bn = cntobn(pmp, newcluster);
1385 	/* always succeeds */
1386 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1387 	bzero(bp->b_data, pmp->pm_bpcluster);
1388 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1389 	denp = (struct direntry *)bp->b_data;
1390 	putushort(denp[0].deStartCluster, newcluster);
1391 	putushort(denp[0].deCDate, ndirent.de_CDate);
1392 	putushort(denp[0].deCTime, ndirent.de_CTime);
1393 	denp[0].deCHundredth = ndirent.de_CHun;
1394 	putushort(denp[0].deADate, ndirent.de_ADate);
1395 	putushort(denp[0].deMDate, ndirent.de_MDate);
1396 	putushort(denp[0].deMTime, ndirent.de_MTime);
1397 	pcl = pdep->de_StartCluster;
1398 	/*
1399 	 * Although the root directory has a non-magic starting cluster
1400 	 * number for FAT32, chkdsk and fsck_msdosfs still require
1401 	 * references to it in dotdot entries to be magic.
1402 	 */
1403 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1404 		pcl = MSDOSFSROOT;
1405 	putushort(denp[1].deStartCluster, pcl);
1406 	putushort(denp[1].deCDate, ndirent.de_CDate);
1407 	putushort(denp[1].deCTime, ndirent.de_CTime);
1408 	denp[1].deCHundredth = ndirent.de_CHun;
1409 	putushort(denp[1].deADate, ndirent.de_ADate);
1410 	putushort(denp[1].deMDate, ndirent.de_MDate);
1411 	putushort(denp[1].deMTime, ndirent.de_MTime);
1412 	if (FAT32(pmp)) {
1413 		putushort(denp[0].deHighClust, newcluster >> 16);
1414 		putushort(denp[1].deHighClust, pcl >> 16);
1415 	}
1416 
1417 	if (DOINGASYNC(ap->a_dvp))
1418 		bdwrite(bp);
1419 	else if ((error = bwrite(bp)) != 0)
1420 		goto bad;
1421 
1422 	/*
1423 	 * Now build up a directory entry pointing to the newly allocated
1424 	 * cluster.  This will be written to an empty slot in the parent
1425 	 * directory.
1426 	 */
1427 #ifdef DIAGNOSTIC
1428 	if ((cnp->cn_flags & HASBUF) == 0)
1429 		panic("msdosfs_mkdir: no name");
1430 #endif
1431 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1432 	if (error)
1433 		goto bad;
1434 
1435 	ndirent.de_Attributes = ATTR_DIRECTORY;
1436 	ndirent.de_LowerCase = 0;
1437 	ndirent.de_StartCluster = newcluster;
1438 	ndirent.de_FileSize = 0;
1439 	error = createde(&ndirent, pdep, &dep, cnp);
1440 	if (error)
1441 		goto bad;
1442 	*ap->a_vpp = DETOV(dep);
1443 	return (0);
1444 
1445 bad:
1446 	clusterfree(pmp, newcluster, NULL);
1447 bad2:
1448 	return (error);
1449 }
1450 
1451 static int
1452 msdosfs_rmdir(ap)
1453 	struct vop_rmdir_args /* {
1454 		struct vnode *a_dvp;
1455 		struct vnode *a_vp;
1456 		struct componentname *a_cnp;
1457 	} */ *ap;
1458 {
1459 	struct vnode *vp = ap->a_vp;
1460 	struct vnode *dvp = ap->a_dvp;
1461 	struct componentname *cnp = ap->a_cnp;
1462 	struct denode *ip, *dp;
1463 	int error;
1464 
1465 	ip = VTODE(vp);
1466 	dp = VTODE(dvp);
1467 
1468 	/*
1469 	 * Verify the directory is empty (and valid).
1470 	 * (Rmdir ".." won't be valid since
1471 	 *  ".." will contain a reference to
1472 	 *  the current directory and thus be
1473 	 *  non-empty.)
1474 	 */
1475 	error = 0;
1476 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1477 		error = ENOTEMPTY;
1478 		goto out;
1479 	}
1480 	/*
1481 	 * Delete the entry from the directory.  For dos filesystems this
1482 	 * gets rid of the directory entry on disk, the in memory copy
1483 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1484 	 * being found by deget().  When the vput() on dep is done we give
1485 	 * up access and eventually msdosfs_reclaim() will be called which
1486 	 * will remove it from the denode cache.
1487 	 */
1488 	error = removede(dp, ip);
1489 	if (error)
1490 		goto out;
1491 	/*
1492 	 * This is where we decrement the link count in the parent
1493 	 * directory.  Since dos filesystems don't do this we just purge
1494 	 * the name cache.
1495 	 */
1496 	cache_purge(dvp);
1497 	/*
1498 	 * Truncate the directory that is being deleted.
1499 	 */
1500 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1501 	cache_purge(vp);
1502 
1503 out:
1504 	return (error);
1505 }
1506 
1507 /*
1508  * DOS filesystems don't know what symlinks are.
1509  */
1510 static int
1511 msdosfs_symlink(ap)
1512 	struct vop_symlink_args /* {
1513 		struct vnode *a_dvp;
1514 		struct vnode **a_vpp;
1515 		struct componentname *a_cnp;
1516 		struct vattr *a_vap;
1517 		char *a_target;
1518 	} */ *ap;
1519 {
1520 	return (EOPNOTSUPP);
1521 }
1522 
1523 static int
1524 msdosfs_readdir(ap)
1525 	struct vop_readdir_args /* {
1526 		struct vnode *a_vp;
1527 		struct uio *a_uio;
1528 		struct ucred *a_cred;
1529 		int *a_eofflag;
1530 		int *a_ncookies;
1531 		u_long **a_cookies;
1532 	} */ *ap;
1533 {
1534 	struct mbnambuf nb;
1535 	int error = 0;
1536 	int diff;
1537 	long n;
1538 	int blsize;
1539 	long on;
1540 	u_long cn;
1541 	uint64_t fileno;
1542 	u_long dirsperblk;
1543 	long bias = 0;
1544 	daddr_t bn, lbn;
1545 	struct buf *bp;
1546 	struct denode *dep = VTODE(ap->a_vp);
1547 	struct msdosfsmount *pmp = dep->de_pmp;
1548 	struct direntry *dentp;
1549 	struct dirent dirbuf;
1550 	struct uio *uio = ap->a_uio;
1551 	u_long *cookies = NULL;
1552 	int ncookies = 0;
1553 	off_t offset, off;
1554 	int chksum = -1;
1555 
1556 #ifdef MSDOSFS_DEBUG
1557 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1558 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1559 #endif
1560 
1561 	/*
1562 	 * msdosfs_readdir() won't operate properly on regular files since
1563 	 * it does i/o only with the filesystem vnode, and hence can
1564 	 * retrieve the wrong block from the buffer cache for a plain file.
1565 	 * So, fail attempts to readdir() on a plain file.
1566 	 */
1567 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1568 		return (ENOTDIR);
1569 
1570 	/*
1571 	 * To be safe, initialize dirbuf
1572 	 */
1573 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1574 
1575 	/*
1576 	 * If the user buffer is smaller than the size of one dos directory
1577 	 * entry or the file offset is not a multiple of the size of a
1578 	 * directory entry, then we fail the read.
1579 	 */
1580 	off = offset = uio->uio_offset;
1581 	if (uio->uio_resid < sizeof(struct direntry) ||
1582 	    (offset & (sizeof(struct direntry) - 1)))
1583 		return (EINVAL);
1584 
1585 	if (ap->a_ncookies) {
1586 		ncookies = uio->uio_resid / 16;
1587 		cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1588 		       M_WAITOK);
1589 		*ap->a_cookies = cookies;
1590 		*ap->a_ncookies = ncookies;
1591 	}
1592 
1593 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1594 
1595 	/*
1596 	 * If they are reading from the root directory then, we simulate
1597 	 * the . and .. entries since these don't exist in the root
1598 	 * directory.  We also set the offset bias to make up for having to
1599 	 * simulate these entries. By this I mean that at file offset 64 we
1600 	 * read the first entry in the root directory that lives on disk.
1601 	 */
1602 	if (dep->de_StartCluster == MSDOSFSROOT
1603 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1604 #if 0
1605 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1606 		    offset);
1607 #endif
1608 		bias = 2 * sizeof(struct direntry);
1609 		if (offset < bias) {
1610 			for (n = (int)offset / sizeof(struct direntry);
1611 			     n < 2; n++) {
1612 				if (FAT32(pmp))
1613 					fileno = (uint64_t)cntobn(pmp,
1614 								 pmp->pm_rootdirblk)
1615 							  * dirsperblk;
1616 				else
1617 					fileno = 1;
1618 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1619 					dirbuf.d_fileno =
1620 					    msdosfs_fileno_map(pmp->pm_mountp,
1621 					    fileno);
1622 				} else {
1623 
1624 					dirbuf.d_fileno = (uint32_t)fileno;
1625 				}
1626 				dirbuf.d_type = DT_DIR;
1627 				switch (n) {
1628 				case 0:
1629 					dirbuf.d_namlen = 1;
1630 					strcpy(dirbuf.d_name, ".");
1631 					break;
1632 				case 1:
1633 					dirbuf.d_namlen = 2;
1634 					strcpy(dirbuf.d_name, "..");
1635 					break;
1636 				}
1637 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1638 				if (uio->uio_resid < dirbuf.d_reclen)
1639 					goto out;
1640 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1641 				if (error)
1642 					goto out;
1643 				offset += sizeof(struct direntry);
1644 				off = offset;
1645 				if (cookies) {
1646 					*cookies++ = offset;
1647 					if (--ncookies <= 0)
1648 						goto out;
1649 				}
1650 			}
1651 		}
1652 	}
1653 
1654 	mbnambuf_init(&nb);
1655 	off = offset;
1656 	while (uio->uio_resid > 0) {
1657 		lbn = de_cluster(pmp, offset - bias);
1658 		on = (offset - bias) & pmp->pm_crbomask;
1659 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1660 		diff = dep->de_FileSize - (offset - bias);
1661 		if (diff <= 0)
1662 			break;
1663 		n = min(n, diff);
1664 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1665 		if (error)
1666 			break;
1667 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1668 		if (error) {
1669 			brelse(bp);
1670 			return (error);
1671 		}
1672 		n = min(n, blsize - bp->b_resid);
1673 		if (n == 0) {
1674 			brelse(bp);
1675 			return (EIO);
1676 		}
1677 
1678 		/*
1679 		 * Convert from dos directory entries to fs-independent
1680 		 * directory entries.
1681 		 */
1682 		for (dentp = (struct direntry *)(bp->b_data + on);
1683 		     (char *)dentp < bp->b_data + on + n;
1684 		     dentp++, offset += sizeof(struct direntry)) {
1685 #if 0
1686 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1687 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1688 #endif
1689 			/*
1690 			 * If this is an unused entry, we can stop.
1691 			 */
1692 			if (dentp->deName[0] == SLOT_EMPTY) {
1693 				brelse(bp);
1694 				goto out;
1695 			}
1696 			/*
1697 			 * Skip deleted entries.
1698 			 */
1699 			if (dentp->deName[0] == SLOT_DELETED) {
1700 				chksum = -1;
1701 				mbnambuf_init(&nb);
1702 				continue;
1703 			}
1704 
1705 			/*
1706 			 * Handle Win95 long directory entries
1707 			 */
1708 			if (dentp->deAttributes == ATTR_WIN95) {
1709 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1710 					continue;
1711 				chksum = win2unixfn(&nb,
1712 				    (struct winentry *)dentp, chksum, pmp);
1713 				continue;
1714 			}
1715 
1716 			/*
1717 			 * Skip volume labels
1718 			 */
1719 			if (dentp->deAttributes & ATTR_VOLUME) {
1720 				chksum = -1;
1721 				mbnambuf_init(&nb);
1722 				continue;
1723 			}
1724 			/*
1725 			 * This computation of d_fileno must match
1726 			 * the computation of va_fileid in
1727 			 * msdosfs_getattr.
1728 			 */
1729 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1730 				fileno = getushort(dentp->deStartCluster);
1731 				if (FAT32(pmp))
1732 					fileno |= getushort(dentp->deHighClust) << 16;
1733 				/* if this is the root directory */
1734 				if (fileno == MSDOSFSROOT)
1735 					if (FAT32(pmp))
1736 						fileno = (uint64_t)cntobn(pmp,
1737 								pmp->pm_rootdirblk)
1738 							 * dirsperblk;
1739 					else
1740 						fileno = 1;
1741 				else
1742 					fileno = (uint64_t)cntobn(pmp, fileno) *
1743 					    dirsperblk;
1744 				dirbuf.d_type = DT_DIR;
1745 			} else {
1746 				fileno = (uoff_t)offset /
1747 				    sizeof(struct direntry);
1748 				dirbuf.d_type = DT_REG;
1749 			}
1750 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1751 				dirbuf.d_fileno =
1752 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1753 			} else
1754 				dirbuf.d_fileno = (uint32_t)fileno;
1755 
1756 			if (chksum != winChksum(dentp->deName)) {
1757 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1758 				    (u_char *)dirbuf.d_name,
1759 				    dentp->deLowerCase |
1760 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1761 					(LCASE_BASE | LCASE_EXT) : 0),
1762 				    pmp);
1763 				mbnambuf_init(&nb);
1764 			} else
1765 				mbnambuf_flush(&nb, &dirbuf);
1766 			chksum = -1;
1767 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1768 			if (uio->uio_resid < dirbuf.d_reclen) {
1769 				brelse(bp);
1770 				goto out;
1771 			}
1772 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1773 			if (error) {
1774 				brelse(bp);
1775 				goto out;
1776 			}
1777 			if (cookies) {
1778 				*cookies++ = offset + sizeof(struct direntry);
1779 				if (--ncookies <= 0) {
1780 					brelse(bp);
1781 					goto out;
1782 				}
1783 			}
1784 			off = offset + sizeof(struct direntry);
1785 		}
1786 		brelse(bp);
1787 	}
1788 out:
1789 	/* Subtract unused cookies */
1790 	if (ap->a_ncookies)
1791 		*ap->a_ncookies -= ncookies;
1792 
1793 	uio->uio_offset = off;
1794 
1795 	/*
1796 	 * Set the eofflag (NFS uses it)
1797 	 */
1798 	if (ap->a_eofflag) {
1799 		if (dep->de_FileSize - (offset - bias) <= 0)
1800 			*ap->a_eofflag = 1;
1801 		else
1802 			*ap->a_eofflag = 0;
1803 	}
1804 	return (error);
1805 }
1806 
1807 /*-
1808  * a_vp   - pointer to the file's vnode
1809  * a_bn   - logical block number within the file (cluster number for us)
1810  * a_bop  - where to return the bufobj of the special file containing the fs
1811  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1812  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1813  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1814  *          blocks whose physical blocks (together with a_bn's physical block)
1815  *          are contiguous.
1816  * a_runb - where to return the "run before" a_bn.
1817  */
1818 static int
1819 msdosfs_bmap(ap)
1820 	struct vop_bmap_args /* {
1821 		struct vnode *a_vp;
1822 		daddr_t a_bn;
1823 		struct bufobj **a_bop;
1824 		daddr_t *a_bnp;
1825 		int *a_runp;
1826 		int *a_runb;
1827 	} */ *ap;
1828 {
1829 	struct denode *dep;
1830 	struct mount *mp;
1831 	struct msdosfsmount *pmp;
1832 	struct vnode *vp;
1833 	daddr_t runbn;
1834 	u_long cn;
1835 	int bnpercn, error, maxio, maxrun, run;
1836 
1837 	vp = ap->a_vp;
1838 	dep = VTODE(vp);
1839 	pmp = dep->de_pmp;
1840 	if (ap->a_bop != NULL)
1841 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1842 	if (ap->a_bnp == NULL)
1843 		return (0);
1844 	if (ap->a_runp != NULL)
1845 		*ap->a_runp = 0;
1846 	if (ap->a_runb != NULL)
1847 		*ap->a_runb = 0;
1848 	cn = ap->a_bn;
1849 	if (cn != ap->a_bn)
1850 		return (EFBIG);
1851 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1852 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1853 		return (error);
1854 
1855 	mp = vp->v_mount;
1856 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1857 	bnpercn = de_cn2bn(pmp, 1);
1858 	if (ap->a_runp != NULL) {
1859 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1860 		for (run = 1; run <= maxrun; run++) {
1861 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1862 			    runbn != *ap->a_bnp + run * bnpercn)
1863 				break;
1864 		}
1865 		*ap->a_runp = run - 1;
1866 	}
1867 	if (ap->a_runb != NULL) {
1868 		maxrun = ulmin(maxio - 1, cn);
1869 		for (run = 1; run < maxrun; run++) {
1870 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1871 			    runbn != *ap->a_bnp - run * bnpercn)
1872 				break;
1873 		}
1874 		*ap->a_runb = run - 1;
1875 	}
1876 	return (0);
1877 }
1878 
1879 static int
1880 msdosfs_strategy(ap)
1881 	struct vop_strategy_args /* {
1882 		struct vnode *a_vp;
1883 		struct buf *a_bp;
1884 	} */ *ap;
1885 {
1886 	struct buf *bp = ap->a_bp;
1887 	struct denode *dep = VTODE(ap->a_vp);
1888 	struct bufobj *bo;
1889 	int error = 0;
1890 	daddr_t blkno;
1891 
1892 	/*
1893 	 * If we don't already know the filesystem relative block number
1894 	 * then get it using pcbmap().  If pcbmap() returns the block
1895 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1896 	 * don't allow files with holes, so we shouldn't ever see this.
1897 	 */
1898 	if (bp->b_blkno == bp->b_lblkno) {
1899 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1900 		bp->b_blkno = blkno;
1901 		if (error) {
1902 			bp->b_error = error;
1903 			bp->b_ioflags |= BIO_ERROR;
1904 			bufdone(bp);
1905 			return (0);
1906 		}
1907 		if ((long)bp->b_blkno == -1)
1908 			vfs_bio_clrbuf(bp);
1909 	}
1910 	if (bp->b_blkno == -1) {
1911 		bufdone(bp);
1912 		return (0);
1913 	}
1914 	/*
1915 	 * Read/write the block from/to the disk that contains the desired
1916 	 * file block.
1917 	 */
1918 	bp->b_iooffset = dbtob(bp->b_blkno);
1919 	bo = dep->de_pmp->pm_bo;
1920 	BO_STRATEGY(bo, bp);
1921 	return (0);
1922 }
1923 
1924 static int
1925 msdosfs_print(ap)
1926 	struct vop_print_args /* {
1927 		struct vnode *vp;
1928 	} */ *ap;
1929 {
1930 	struct denode *dep = VTODE(ap->a_vp);
1931 
1932 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1933 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1934 	printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1935 	return (0);
1936 }
1937 
1938 static int
1939 msdosfs_pathconf(ap)
1940 	struct vop_pathconf_args /* {
1941 		struct vnode *a_vp;
1942 		int a_name;
1943 		int *a_retval;
1944 	} */ *ap;
1945 {
1946 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1947 
1948 	switch (ap->a_name) {
1949 	case _PC_LINK_MAX:
1950 		*ap->a_retval = 1;
1951 		return (0);
1952 	case _PC_NAME_MAX:
1953 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1954 		return (0);
1955 	case _PC_PATH_MAX:
1956 		*ap->a_retval = PATH_MAX;
1957 		return (0);
1958 	case _PC_CHOWN_RESTRICTED:
1959 		*ap->a_retval = 1;
1960 		return (0);
1961 	case _PC_NO_TRUNC:
1962 		*ap->a_retval = 0;
1963 		return (0);
1964 	default:
1965 		return (EINVAL);
1966 	}
1967 	/* NOTREACHED */
1968 }
1969 
1970 static int
1971 msdosfs_vptofh(ap)
1972 	struct vop_vptofh_args /* {
1973 		struct vnode *a_vp;
1974 		struct fid *a_fhp;
1975 	} */ *ap;
1976 {
1977 	struct denode *dep;
1978 	struct defid *defhp;
1979 
1980 	dep = VTODE(ap->a_vp);
1981 	defhp = (struct defid *)ap->a_fhp;
1982 	defhp->defid_len = sizeof(struct defid);
1983 	defhp->defid_dirclust = dep->de_dirclust;
1984 	defhp->defid_dirofs = dep->de_diroffset;
1985 	/* defhp->defid_gen = dep->de_gen; */
1986 	return (0);
1987 }
1988 
1989 /* Global vfs data structures for msdosfs */
1990 struct vop_vector msdosfs_vnodeops = {
1991 	.vop_default =		&default_vnodeops,
1992 
1993 	.vop_access =		msdosfs_access,
1994 	.vop_bmap =		msdosfs_bmap,
1995 	.vop_cachedlookup =	msdosfs_lookup,
1996 	.vop_open =		msdosfs_open,
1997 	.vop_close =		msdosfs_close,
1998 	.vop_create =		msdosfs_create,
1999 	.vop_fsync =		msdosfs_fsync,
2000 	.vop_getattr =		msdosfs_getattr,
2001 	.vop_inactive =		msdosfs_inactive,
2002 	.vop_link =		msdosfs_link,
2003 	.vop_lookup =		vfs_cache_lookup,
2004 	.vop_mkdir =		msdosfs_mkdir,
2005 	.vop_mknod =		msdosfs_mknod,
2006 	.vop_pathconf =		msdosfs_pathconf,
2007 	.vop_print =		msdosfs_print,
2008 	.vop_read =		msdosfs_read,
2009 	.vop_readdir =		msdosfs_readdir,
2010 	.vop_reclaim =		msdosfs_reclaim,
2011 	.vop_remove =		msdosfs_remove,
2012 	.vop_rename =		msdosfs_rename,
2013 	.vop_rmdir =		msdosfs_rmdir,
2014 	.vop_setattr =		msdosfs_setattr,
2015 	.vop_strategy =		msdosfs_strategy,
2016 	.vop_symlink =		msdosfs_symlink,
2017 	.vop_write =		msdosfs_write,
2018 	.vop_vptofh =		msdosfs_vptofh,
2019 };
2020