xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision 995dc984471c92c03daad19a1d35af46c086ef3e)
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_advlock_t	msdosfs_advlock;
86 static vop_create_t	msdosfs_create;
87 static vop_mknod_t	msdosfs_mknod;
88 static vop_open_t	msdosfs_open;
89 static vop_close_t	msdosfs_close;
90 static vop_access_t	msdosfs_access;
91 static vop_getattr_t	msdosfs_getattr;
92 static vop_setattr_t	msdosfs_setattr;
93 static vop_read_t	msdosfs_read;
94 static vop_write_t	msdosfs_write;
95 static vop_fsync_t	msdosfs_fsync;
96 static vop_remove_t	msdosfs_remove;
97 static vop_link_t	msdosfs_link;
98 static vop_rename_t	msdosfs_rename;
99 static vop_mkdir_t	msdosfs_mkdir;
100 static vop_rmdir_t	msdosfs_rmdir;
101 static vop_symlink_t	msdosfs_symlink;
102 static vop_readdir_t	msdosfs_readdir;
103 static vop_bmap_t	msdosfs_bmap;
104 static vop_strategy_t	msdosfs_strategy;
105 static vop_print_t	msdosfs_print;
106 static vop_pathconf_t	msdosfs_pathconf;
107 static vop_vptofh_t	msdosfs_vptofh;
108 
109 /*
110  * Some general notes:
111  *
112  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
113  * read/written using the vnode for the filesystem. Blocks that represent
114  * the contents of a file are read/written using the vnode for the file
115  * (including directories when they are read/written as files). This
116  * presents problems for the dos filesystem because data that should be in
117  * an inode (if dos had them) resides in the directory itself.  Since we
118  * must update directory entries without the benefit of having the vnode
119  * for the directory we must use the vnode for the filesystem.  This means
120  * that when a directory is actually read/written (via read, write, or
121  * readdir, or seek) we must use the vnode for the filesystem instead of
122  * the vnode for the directory as would happen in ufs. This is to insure we
123  * retreive the correct block from the buffer cache since the hash value is
124  * based upon the vnode address and the desired block number.
125  */
126 
127 /*
128  * Create a regular file. On entry the directory to contain the file being
129  * created is locked.  We must release before we return. We must also free
130  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
131  * only if the SAVESTART bit in cn_flags is clear on success.
132  */
133 static int
134 msdosfs_create(ap)
135 	struct vop_create_args /* {
136 		struct vnode *a_dvp;
137 		struct vnode **a_vpp;
138 		struct componentname *a_cnp;
139 		struct vattr *a_vap;
140 	} */ *ap;
141 {
142 	struct componentname *cnp = ap->a_cnp;
143 	struct denode ndirent;
144 	struct denode *dep;
145 	struct denode *pdep = VTODE(ap->a_dvp);
146 	struct timespec ts;
147 	int error;
148 
149 #ifdef MSDOSFS_DEBUG
150 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
151 #endif
152 
153 	/*
154 	 * If this is the root directory and there is no space left we
155 	 * can't do anything.  This is because the root directory can not
156 	 * change size.
157 	 */
158 	if (pdep->de_StartCluster == MSDOSFSROOT
159 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
160 		error = ENOSPC;
161 		goto bad;
162 	}
163 
164 	/*
165 	 * Create a directory entry for the file, then call createde() to
166 	 * have it installed. NOTE: DOS files are always executable.  We
167 	 * use the absence of the owner write bit to make the file
168 	 * readonly.
169 	 */
170 #ifdef DIAGNOSTIC
171 	if ((cnp->cn_flags & HASBUF) == 0)
172 		panic("msdosfs_create: no name");
173 #endif
174 	bzero(&ndirent, sizeof(ndirent));
175 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
176 	if (error)
177 		goto bad;
178 
179 	ndirent.de_Attributes = (ap->a_vap->va_mode & VWRITE) ?
180 				ATTR_ARCHIVE : ATTR_ARCHIVE | ATTR_READONLY;
181 	ndirent.de_LowerCase = 0;
182 	ndirent.de_StartCluster = 0;
183 	ndirent.de_FileSize = 0;
184 	ndirent.de_dev = pdep->de_dev;
185 	ndirent.de_pmp = pdep->de_pmp;
186 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
187 	getnanotime(&ts);
188 	DETIMES(&ndirent, &ts, &ts, &ts);
189 	error = createde(&ndirent, pdep, &dep, cnp);
190 	if (error)
191 		goto bad;
192 	*ap->a_vpp = DETOV(dep);
193 	return (0);
194 
195 bad:
196 	return (error);
197 }
198 
199 static int
200 msdosfs_mknod(ap)
201 	struct vop_mknod_args /* {
202 		struct vnode *a_dvp;
203 		struct vnode **a_vpp;
204 		struct componentname *a_cnp;
205 		struct vattr *a_vap;
206 	} */ *ap;
207 {
208 
209     return (EINVAL);
210 }
211 
212 static int
213 msdosfs_open(ap)
214 	struct vop_open_args /* {
215 		struct vnode *a_vp;
216 		int a_mode;
217 		struct ucred *a_cred;
218 		struct thread *a_td;
219 		int a_fdidx;
220 	} */ *ap;
221 {
222 	struct denode *dep = VTODE(ap->a_vp);
223 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
224 	return 0;
225 }
226 
227 static int
228 msdosfs_close(ap)
229 	struct vop_close_args /* {
230 		struct vnode *a_vp;
231 		int a_fflag;
232 		struct ucred *a_cred;
233 		struct thread *a_td;
234 	} */ *ap;
235 {
236 	struct vnode *vp = ap->a_vp;
237 	struct denode *dep = VTODE(vp);
238 	struct timespec ts;
239 
240 	VI_LOCK(vp);
241 	if (vp->v_usecount > 1) {
242 		getnanotime(&ts);
243 		DETIMES(dep, &ts, &ts, &ts);
244 	}
245 	VI_UNLOCK(vp);
246 	return 0;
247 }
248 
249 static int
250 msdosfs_access(ap)
251 	struct vop_access_args /* {
252 		struct vnode *a_vp;
253 		int a_mode;
254 		struct ucred *a_cred;
255 		struct thread *a_td;
256 	} */ *ap;
257 {
258 	struct vnode *vp = ap->a_vp;
259 	struct denode *dep = VTODE(ap->a_vp);
260 	struct msdosfsmount *pmp = dep->de_pmp;
261 	mode_t file_mode, mode = ap->a_mode;
262 
263 	file_mode = (S_IXUSR|S_IXGRP|S_IXOTH) | (S_IRUSR|S_IRGRP|S_IROTH) |
264 	    ((dep->de_Attributes & ATTR_READONLY) ? 0 : (S_IWUSR|S_IWGRP|S_IWOTH));
265 	file_mode &= (vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
266 
267 	/*
268 	 * Disallow writing to directories and regular files if the
269 	 * filesystem is read-only.
270 	 */
271 	if (mode & VWRITE) {
272 		switch (vp->v_type) {
273 		case VDIR:
274 		case VREG:
275 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
276 				return (EROFS);
277 			break;
278 		default:
279 			break;
280 		}
281 	}
282 
283 	return (vaccess(vp->v_type, file_mode, pmp->pm_uid, pmp->pm_gid,
284 	    ap->a_mode, ap->a_cred, NULL));
285 }
286 
287 static int
288 msdosfs_getattr(ap)
289 	struct vop_getattr_args /* {
290 		struct vnode *a_vp;
291 		struct vattr *a_vap;
292 		struct ucred *a_cred;
293 		struct thread *a_td;
294 	} */ *ap;
295 {
296 	struct denode *dep = VTODE(ap->a_vp);
297 	struct msdosfsmount *pmp = dep->de_pmp;
298 	struct vattr *vap = ap->a_vap;
299 	mode_t mode;
300 	struct timespec ts;
301 	u_long dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
302 	uint64_t fileid;
303 
304 	getnanotime(&ts);
305 	DETIMES(dep, &ts, &ts, &ts);
306 	vap->va_fsid = dev2udev(dep->de_dev);
307 	/*
308 	 * The following computation of the fileid must be the same as that
309 	 * used in msdosfs_readdir() to compute d_fileno. If not, pwd
310 	 * doesn't work.
311 	 */
312 	if (dep->de_Attributes & ATTR_DIRECTORY) {
313 		fileid = (uint64_t)cntobn(pmp, dep->de_StartCluster) *
314 		    dirsperblk;
315 		if (dep->de_StartCluster == MSDOSFSROOT)
316 			fileid = 1;
317 	} else {
318 		fileid = (uint64_t)cntobn(pmp, dep->de_dirclust) *
319 		    dirsperblk;
320 		if (dep->de_dirclust == MSDOSFSROOT)
321 			fileid = (uint64_t)roottobn(pmp, 0) * dirsperblk;
322 		fileid += (uoff_t)dep->de_diroffset / sizeof(struct direntry);
323 	}
324 
325 	if (pmp->pm_flags & MSDOSFS_LARGEFS)
326 		vap->va_fileid = msdosfs_fileno_map(pmp->pm_mountp, fileid);
327 	else
328 		vap->va_fileid = (long)fileid;
329 
330 	if ((dep->de_Attributes & ATTR_READONLY) == 0)
331 		mode = S_IRWXU|S_IRWXG|S_IRWXO;
332 	else
333 		mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
334 	vap->va_mode = mode &
335 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
336 	vap->va_uid = pmp->pm_uid;
337 	vap->va_gid = pmp->pm_gid;
338 	vap->va_nlink = 1;
339 	vap->va_rdev = 0;
340 	vap->va_size = dep->de_FileSize;
341 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
342 	vap->va_ctime = vap->va_mtime;
343 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
344 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
345 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
346 		    0, &vap->va_birthtime);
347 	} else {
348 		vap->va_atime = vap->va_mtime;
349 		vap->va_birthtime.tv_sec = -1;
350 		vap->va_birthtime.tv_nsec = 0;
351 	}
352 	vap->va_flags = 0;
353 	if ((dep->de_Attributes & ATTR_ARCHIVE) == 0)
354 		vap->va_flags |= SF_ARCHIVED;
355 	vap->va_gen = 0;
356 	vap->va_blocksize = pmp->pm_bpcluster;
357 	vap->va_bytes =
358 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
359 	vap->va_type = ap->a_vp->v_type;
360 	vap->va_filerev = dep->de_modrev;
361 	return (0);
362 }
363 
364 static int
365 msdosfs_setattr(ap)
366 	struct vop_setattr_args /* {
367 		struct vnode *a_vp;
368 		struct vattr *a_vap;
369 		struct ucred *a_cred;
370 		struct thread *a_td;
371 	} */ *ap;
372 {
373 	struct vnode *vp = ap->a_vp;
374 	struct denode *dep = VTODE(ap->a_vp);
375 	struct msdosfsmount *pmp = dep->de_pmp;
376 	struct vattr *vap = ap->a_vap;
377 	struct ucred *cred = ap->a_cred;
378 	int error = 0;
379 
380 #ifdef MSDOSFS_DEBUG
381 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p, p %p\n",
382 	    ap->a_vp, vap, cred, ap->a_td);
383 #endif
384 
385 	/*
386 	 * Check for unsettable attributes.
387 	 */
388 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
389 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
390 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
391 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
392 #ifdef MSDOSFS_DEBUG
393 		printf("msdosfs_setattr(): returning EINVAL\n");
394 		printf("    va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
395 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
396 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
397 		    vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
398 		printf("    va_uid %x, va_gid %x\n",
399 		    vap->va_uid, vap->va_gid);
400 #endif
401 		return (EINVAL);
402 	}
403 	if (vap->va_flags != VNOVAL) {
404 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
405 			return (EROFS);
406 		if (cred->cr_uid != pmp->pm_uid) {
407 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
408 			if (error)
409 				return (error);
410 		}
411 		/*
412 		 * We are very inconsistent about handling unsupported
413 		 * attributes.  We ignored the access time and the
414 		 * read and execute bits.  We were strict for the other
415 		 * attributes.
416 		 *
417 		 * Here we are strict, stricter than ufs in not allowing
418 		 * users to attempt to set SF_SETTABLE bits or anyone to
419 		 * set unsupported bits.  However, we ignore attempts to
420 		 * set ATTR_ARCHIVE for directories `cp -pr' from a more
421 		 * sensible filesystem attempts it a lot.
422 		 */
423 		if (vap->va_flags & SF_SETTABLE) {
424 			error = priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0);
425 			if (error)
426 				return (error);
427 		}
428 		if (vap->va_flags & ~SF_ARCHIVED)
429 			return EOPNOTSUPP;
430 		if (vap->va_flags & SF_ARCHIVED)
431 			dep->de_Attributes &= ~ATTR_ARCHIVE;
432 		else if (!(dep->de_Attributes & ATTR_DIRECTORY))
433 			dep->de_Attributes |= ATTR_ARCHIVE;
434 		dep->de_flag |= DE_MODIFIED;
435 	}
436 
437 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
438 		uid_t uid;
439 		gid_t gid;
440 
441 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
442 			return (EROFS);
443 		uid = vap->va_uid;
444 		if (uid == (uid_t)VNOVAL)
445 			uid = pmp->pm_uid;
446 		gid = vap->va_gid;
447 		if (gid == (gid_t)VNOVAL)
448 			gid = pmp->pm_gid;
449 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
450 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
451 			error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
452 			if (error)
453 				return (error);
454 		}
455 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
456 			return EINVAL;
457 	}
458 
459 	if (vap->va_size != VNOVAL) {
460 		switch (vp->v_type) {
461 		case VDIR:
462 			return (EISDIR);
463 		case VREG:
464 			/*
465 			 * Truncation is only supported for regular files,
466 			 * Disallow it if the filesystem is read-only.
467 			 */
468 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
469 				return (EROFS);
470 			break;
471 		default:
472 			/*
473 			 * According to POSIX, the result is unspecified
474 			 * for file types other than regular files,
475 			 * directories and shared memory objects.  We
476 			 * don't support any file types except regular
477 			 * files and directories in this file system, so
478 			 * this (default) case is unreachable and can do
479 			 * anything.  Keep falling through to detrunc()
480 			 * for now.
481 			 */
482 			break;
483 		}
484 		error = detrunc(dep, vap->va_size, 0, cred, ap->a_td);
485 		if (error)
486 			return error;
487 	}
488 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
489 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
490 			return (EROFS);
491 		if (vap->va_vaflags & VA_UTIMES_NULL) {
492 			error = VOP_ACCESS(vp, VADMIN, cred, ap->a_td);
493 			if (error)
494 				error = VOP_ACCESS(vp, VWRITE, cred,
495 				    ap->a_td);
496 		} else
497 			error = VOP_ACCESS(vp, VADMIN, cred, ap->a_td);
498 		if (vp->v_type != VDIR) {
499 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
500 			    vap->va_atime.tv_sec != VNOVAL) {
501 				dep->de_flag &= ~DE_ACCESS;
502 				timespec2fattime(&vap->va_atime, 0,
503 				    &dep->de_ADate, NULL, NULL);
504 			}
505 			if (vap->va_mtime.tv_sec != VNOVAL) {
506 				dep->de_flag &= ~DE_UPDATE;
507 				timespec2fattime(&vap->va_mtime, 0,
508 				    &dep->de_MDate, &dep->de_MTime, NULL);
509 			}
510 			dep->de_Attributes |= ATTR_ARCHIVE;
511 			dep->de_flag |= DE_MODIFIED;
512 		}
513 	}
514 	/*
515 	 * DOS files only have the ability to have their writability
516 	 * attribute set, so we use the owner write bit to set the readonly
517 	 * attribute.
518 	 */
519 	if (vap->va_mode != (mode_t)VNOVAL) {
520 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
521 			return (EROFS);
522 		if (cred->cr_uid != pmp->pm_uid) {
523 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
524 			if (error)
525 				return (error);
526 		}
527 		if (vp->v_type != VDIR) {
528 			/* We ignore the read and execute bits. */
529 			if (vap->va_mode & VWRITE)
530 				dep->de_Attributes &= ~ATTR_READONLY;
531 			else
532 				dep->de_Attributes |= ATTR_READONLY;
533 			dep->de_Attributes |= ATTR_ARCHIVE;
534 			dep->de_flag |= DE_MODIFIED;
535 		}
536 	}
537 	return (deupdat(dep, 0));
538 }
539 
540 static int
541 msdosfs_read(ap)
542 	struct vop_read_args /* {
543 		struct vnode *a_vp;
544 		struct uio *a_uio;
545 		int a_ioflag;
546 		struct ucred *a_cred;
547 	} */ *ap;
548 {
549 	int error = 0;
550 	int blsize;
551 	int isadir;
552 	int orig_resid;
553 	u_int n;
554 	u_long diff;
555 	u_long on;
556 	daddr_t lbn;
557 	daddr_t rablock;
558 	int rasize;
559 	int seqcount;
560 	struct buf *bp;
561 	struct vnode *vp = ap->a_vp;
562 	struct denode *dep = VTODE(vp);
563 	struct msdosfsmount *pmp = dep->de_pmp;
564 	struct uio *uio = ap->a_uio;
565 
566 	/*
567 	 * If they didn't ask for any data, then we are done.
568 	 */
569 	orig_resid = uio->uio_resid;
570 	if (orig_resid == 0)
571 		return (0);
572 
573 	/*
574 	 * The caller is supposed to ensure that
575 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
576 	 * We don't need to check for large offsets as in ffs because
577 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
578 	 * offsets cannot cause overflow even in theory.
579 	 */
580 
581 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
582 
583 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
584 	do {
585 		if (uio->uio_offset >= dep->de_FileSize)
586 			break;
587 		lbn = de_cluster(pmp, uio->uio_offset);
588 		rablock = lbn + 1;
589 		blsize = pmp->pm_bpcluster;
590 		on = uio->uio_offset & pmp->pm_crbomask;
591 		/*
592 		 * If we are operating on a directory file then be sure to
593 		 * do i/o with the vnode for the filesystem instead of the
594 		 * vnode for the directory.
595 		 */
596 		if (isadir) {
597 			/* convert cluster # to block # */
598 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
599 			if (error == E2BIG) {
600 				error = EINVAL;
601 				break;
602 			} else if (error)
603 				break;
604 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
605 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
606 			error = bread(vp, lbn, blsize, NOCRED, &bp);
607 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
608 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
609 			    NOCRED, on + uio->uio_resid, seqcount, &bp);
610 		} else if (seqcount > 1) {
611 			rasize = blsize;
612 			error = breadn(vp, lbn,
613 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
614 		} else {
615 			error = bread(vp, lbn, blsize, NOCRED, &bp);
616 		}
617 		if (error) {
618 			brelse(bp);
619 			break;
620 		}
621 		diff = pmp->pm_bpcluster - on;
622 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
623 		diff = dep->de_FileSize - uio->uio_offset;
624 		if (diff < n)
625 			n = diff;
626 		diff = blsize - bp->b_resid;
627 		if (diff < n)
628 			n = diff;
629 		error = uiomove(bp->b_data + on, (int) n, uio);
630 		brelse(bp);
631 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
632 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
633 	    (vp->v_mount->mnt_flag & MNT_NOATIME) == 0)
634 		dep->de_flag |= DE_ACCESS;
635 	return (error);
636 }
637 
638 /*
639  * Write data to a file or directory.
640  */
641 static int
642 msdosfs_write(ap)
643 	struct vop_write_args /* {
644 		struct vnode *a_vp;
645 		struct uio *a_uio;
646 		int a_ioflag;
647 		struct ucred *a_cred;
648 	} */ *ap;
649 {
650 	int n;
651 	int croffset;
652 	int resid;
653 	u_long osize;
654 	int error = 0;
655 	u_long count;
656 	int seqcount;
657 	daddr_t bn, lastcn;
658 	struct buf *bp;
659 	int ioflag = ap->a_ioflag;
660 	struct uio *uio = ap->a_uio;
661 	struct thread *td = uio->uio_td;
662 	struct vnode *vp = ap->a_vp;
663 	struct vnode *thisvp;
664 	struct denode *dep = VTODE(vp);
665 	struct msdosfsmount *pmp = dep->de_pmp;
666 	struct ucred *cred = ap->a_cred;
667 
668 #ifdef MSDOSFS_DEBUG
669 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
670 	    vp, uio, ioflag, cred);
671 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
672 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
673 #endif
674 
675 	switch (vp->v_type) {
676 	case VREG:
677 		if (ioflag & IO_APPEND)
678 			uio->uio_offset = dep->de_FileSize;
679 		thisvp = vp;
680 		break;
681 	case VDIR:
682 		return EISDIR;
683 	default:
684 		panic("msdosfs_write(): bad file type");
685 	}
686 
687 	/*
688 	 * This is needed (unlike in ffs_write()) because we extend the
689 	 * file outside of the loop but we don't want to extend the file
690 	 * for writes of 0 bytes.
691 	 */
692 	if (uio->uio_resid == 0)
693 		return (0);
694 
695 	/*
696 	 * The caller is supposed to ensure that
697 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
698 	 */
699 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
700 		return (EFBIG);
701 
702 	/*
703 	 * If they've exceeded their filesize limit, tell them about it.
704 	 */
705 	if (td != NULL) {
706 		PROC_LOCK(td->td_proc);
707 		if ((uoff_t)uio->uio_offset + uio->uio_resid >
708 		    lim_cur(td->td_proc, RLIMIT_FSIZE)) {
709 			psignal(td->td_proc, SIGXFSZ);
710 			PROC_UNLOCK(td->td_proc);
711 			return (EFBIG);
712 		}
713 		PROC_UNLOCK(td->td_proc);
714 	}
715 
716 	/*
717 	 * If the offset we are starting the write at is beyond the end of
718 	 * the file, then they've done a seek.  Unix filesystems allow
719 	 * files with holes in them, DOS doesn't so we must fill the hole
720 	 * with zeroed blocks.
721 	 */
722 	if (uio->uio_offset > dep->de_FileSize) {
723 		error = deextend(dep, uio->uio_offset, cred);
724 		if (error)
725 			return (error);
726 	}
727 
728 	/*
729 	 * Remember some values in case the write fails.
730 	 */
731 	resid = uio->uio_resid;
732 	osize = dep->de_FileSize;
733 
734 	/*
735 	 * If we write beyond the end of the file, extend it to its ultimate
736 	 * size ahead of the time to hopefully get a contiguous area.
737 	 */
738 	if (uio->uio_offset + resid > osize) {
739 		count = de_clcount(pmp, uio->uio_offset + resid) -
740 			de_clcount(pmp, osize);
741 		error = extendfile(dep, count, NULL, NULL, 0);
742 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
743 			goto errexit;
744 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
745 	} else
746 		lastcn = de_clcount(pmp, osize) - 1;
747 
748 	seqcount = ioflag >> IO_SEQSHIFT;
749 	do {
750 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
751 			error = ENOSPC;
752 			break;
753 		}
754 
755 		croffset = uio->uio_offset & pmp->pm_crbomask;
756 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
757 		if (uio->uio_offset + n > dep->de_FileSize) {
758 			dep->de_FileSize = uio->uio_offset + n;
759 			/* The object size needs to be set before buffer is allocated */
760 			vnode_pager_setsize(vp, dep->de_FileSize);
761 		}
762 
763 		bn = de_cluster(pmp, uio->uio_offset);
764 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
765 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
766 			> de_cluster(pmp, uio->uio_offset)
767 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
768 			/*
769 			 * If either the whole cluster gets written,
770 			 * or we write the cluster from its start beyond EOF,
771 			 * then no need to read data from disk.
772 			 */
773 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
774 			vfs_bio_clrbuf(bp);
775 			/*
776 			 * Do the bmap now, since pcbmap needs buffers
777 			 * for the fat table. (see msdosfs_strategy)
778 			 */
779 			if (bp->b_blkno == bp->b_lblkno) {
780 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
781 				if (error)
782 					bp->b_blkno = -1;
783 				else
784 					bp->b_blkno = bn;
785 			}
786 			if (bp->b_blkno == -1) {
787 				brelse(bp);
788 				if (!error)
789 					error = EIO;		/* XXX */
790 				break;
791 			}
792 		} else {
793 			/*
794 			 * The block we need to write into exists, so read it in.
795 			 */
796 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
797 			if (error) {
798 				brelse(bp);
799 				break;
800 			}
801 		}
802 
803 		/*
804 		 * Should these vnode_pager_* functions be done on dir
805 		 * files?
806 		 */
807 
808 		/*
809 		 * Copy the data from user space into the buf header.
810 		 */
811 		error = uiomove(bp->b_data + croffset, n, uio);
812 		if (error) {
813 			brelse(bp);
814 			break;
815 		}
816 
817 		/* Prepare for clustered writes in some else clauses. */
818 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
819 			bp->b_flags |= B_CLUSTEROK;
820 
821 		/*
822 		 * If IO_SYNC, then each buffer is written synchronously.
823 		 * Otherwise, if we have a severe page deficiency then
824 		 * write the buffer asynchronously.  Otherwise, if on a
825 		 * cluster boundary then write the buffer asynchronously,
826 		 * combining it with contiguous clusters if permitted and
827 		 * possible, since we don't expect more writes into this
828 		 * buffer soon.  Otherwise, do a delayed write because we
829 		 * expect more writes into this buffer soon.
830 		 */
831 		if (ioflag & IO_SYNC)
832 			(void)bwrite(bp);
833 		else if (vm_page_count_severe() || buf_dirty_count_severe())
834 			bawrite(bp);
835 		else if (n + croffset == pmp->pm_bpcluster) {
836 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
837 				cluster_write(vp, bp, dep->de_FileSize,
838 				    seqcount);
839 			else
840 				bawrite(bp);
841 		} else
842 			bdwrite(bp);
843 		dep->de_flag |= DE_UPDATE;
844 	} while (error == 0 && uio->uio_resid > 0);
845 
846 	/*
847 	 * If the write failed and they want us to, truncate the file back
848 	 * to the size it was before the write was attempted.
849 	 */
850 errexit:
851 	if (error) {
852 		if (ioflag & IO_UNIT) {
853 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED, NULL);
854 			uio->uio_offset -= resid - uio->uio_resid;
855 			uio->uio_resid = resid;
856 		} else {
857 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED, NULL);
858 			if (uio->uio_resid != resid)
859 				error = 0;
860 		}
861 	} else if (ioflag & IO_SYNC)
862 		error = deupdat(dep, 1);
863 	return (error);
864 }
865 
866 /*
867  * Flush the blocks of a file to disk.
868  *
869  * This function is worthless for vnodes that represent directories. Maybe we
870  * could just do a sync if they try an fsync on a directory file.
871  */
872 static int
873 msdosfs_fsync(ap)
874 	struct vop_fsync_args /* {
875 		struct vnode *a_vp;
876 		struct ucred *a_cred;
877 		int a_waitfor;
878 		struct thread *a_td;
879 	} */ *ap;
880 {
881 
882 	vop_stdfsync(ap);
883 	return (deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT));
884 }
885 
886 static int
887 msdosfs_remove(ap)
888 	struct vop_remove_args /* {
889 		struct vnode *a_dvp;
890 		struct vnode *a_vp;
891 		struct componentname *a_cnp;
892 	} */ *ap;
893 {
894 	struct denode *dep = VTODE(ap->a_vp);
895 	struct denode *ddep = VTODE(ap->a_dvp);
896 	int error;
897 
898 	if (ap->a_vp->v_type == VDIR)
899 		error = EPERM;
900 	else
901 		error = removede(ddep, dep);
902 #ifdef MSDOSFS_DEBUG
903 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
904 #endif
905 	return (error);
906 }
907 
908 /*
909  * DOS filesystems don't know what links are.
910  */
911 static int
912 msdosfs_link(ap)
913 	struct vop_link_args /* {
914 		struct vnode *a_tdvp;
915 		struct vnode *a_vp;
916 		struct componentname *a_cnp;
917 	} */ *ap;
918 {
919 	return (EOPNOTSUPP);
920 }
921 
922 /*
923  * Renames on files require moving the denode to a new hash queue since the
924  * denode's location is used to compute which hash queue to put the file
925  * in. Unless it is a rename in place.  For example "mv a b".
926  *
927  * What follows is the basic algorithm:
928  *
929  * if (file move) {
930  *	if (dest file exists) {
931  *		remove dest file
932  *	}
933  *	if (dest and src in same directory) {
934  *		rewrite name in existing directory slot
935  *	} else {
936  *		write new entry in dest directory
937  *		update offset and dirclust in denode
938  *		move denode to new hash chain
939  *		clear old directory entry
940  *	}
941  * } else {
942  *	directory move
943  *	if (dest directory exists) {
944  *		if (dest is not empty) {
945  *			return ENOTEMPTY
946  *		}
947  *		remove dest directory
948  *	}
949  *	if (dest and src in same directory) {
950  *		rewrite name in existing entry
951  *	} else {
952  *		be sure dest is not a child of src directory
953  *		write entry in dest directory
954  *		update "." and ".." in moved directory
955  *		clear old directory entry for moved directory
956  *	}
957  * }
958  *
959  * On entry:
960  *	source's parent directory is unlocked
961  *	source file or directory is unlocked
962  *	destination's parent directory is locked
963  *	destination file or directory is locked if it exists
964  *
965  * On exit:
966  *	all denodes should be released
967  */
968 static int
969 msdosfs_rename(ap)
970 	struct vop_rename_args /* {
971 		struct vnode *a_fdvp;
972 		struct vnode *a_fvp;
973 		struct componentname *a_fcnp;
974 		struct vnode *a_tdvp;
975 		struct vnode *a_tvp;
976 		struct componentname *a_tcnp;
977 	} */ *ap;
978 {
979 	struct vnode *tdvp = ap->a_tdvp;
980 	struct vnode *fvp = ap->a_fvp;
981 	struct vnode *fdvp = ap->a_fdvp;
982 	struct vnode *tvp = ap->a_tvp;
983 	struct componentname *tcnp = ap->a_tcnp;
984 	struct componentname *fcnp = ap->a_fcnp;
985 	struct denode *ip, *xp, *dp, *zp;
986 	u_char toname[11], oldname[11];
987 	u_long from_diroffset, to_diroffset;
988 	u_char to_count;
989 	int doingdirectory = 0, newparent = 0;
990 	int error;
991 	u_long cn;
992 	daddr_t bn;
993 	struct denode *fddep;	/* from file's parent directory	 */
994 	struct msdosfsmount *pmp;
995 	struct direntry *dotdotp;
996 	struct buf *bp;
997 
998 	fddep = VTODE(ap->a_fdvp);
999 	pmp = fddep->de_pmp;
1000 
1001 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1002 
1003 #ifdef DIAGNOSTIC
1004 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1005 	    (fcnp->cn_flags & HASBUF) == 0)
1006 		panic("msdosfs_rename: no name");
1007 #endif
1008 	/*
1009 	 * Check for cross-device rename.
1010 	 */
1011 	if (fvp->v_mount != tdvp->v_mount ||
1012 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1013 		error = EXDEV;
1014 abortit:
1015 		if (tdvp == tvp)
1016 			vrele(tdvp);
1017 		else
1018 			vput(tdvp);
1019 		if (tvp)
1020 			vput(tvp);
1021 		vrele(fdvp);
1022 		vrele(fvp);
1023 		return (error);
1024 	}
1025 
1026 	/*
1027 	 * If source and dest are the same, do nothing.
1028 	 */
1029 	if (tvp == fvp) {
1030 		error = 0;
1031 		goto abortit;
1032 	}
1033 
1034 	error = vn_lock(fvp, LK_EXCLUSIVE);
1035 	if (error)
1036 		goto abortit;
1037 	dp = VTODE(fdvp);
1038 	ip = VTODE(fvp);
1039 
1040 	/*
1041 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1042 	 * leads to a crippled directory tree.  It's pretty tough to do a
1043 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1044 	 * doesn't work if the ".." entry is missing.
1045 	 */
1046 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1047 		/*
1048 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1049 		 */
1050 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1051 		    dp == ip ||
1052 		    (fcnp->cn_flags & ISDOTDOT) ||
1053 		    (tcnp->cn_flags & ISDOTDOT) ||
1054 		    (ip->de_flag & DE_RENAME)) {
1055 			VOP_UNLOCK(fvp, 0);
1056 			error = EINVAL;
1057 			goto abortit;
1058 		}
1059 		ip->de_flag |= DE_RENAME;
1060 		doingdirectory++;
1061 	}
1062 
1063 	/*
1064 	 * When the target exists, both the directory
1065 	 * and target vnodes are returned locked.
1066 	 */
1067 	dp = VTODE(tdvp);
1068 	xp = tvp ? VTODE(tvp) : NULL;
1069 	/*
1070 	 * Remember direntry place to use for destination
1071 	 */
1072 	to_diroffset = dp->de_fndoffset;
1073 	to_count = dp->de_fndcnt;
1074 
1075 	/*
1076 	 * If ".." must be changed (ie the directory gets a new
1077 	 * parent) then the source directory must not be in the
1078 	 * directory heirarchy above the target, as this would
1079 	 * orphan everything below the source directory. Also
1080 	 * the user must have write permission in the source so
1081 	 * as to be able to change "..". We must repeat the call
1082 	 * to namei, as the parent directory is unlocked by the
1083 	 * call to doscheckpath().
1084 	 */
1085 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1086 	VOP_UNLOCK(fvp, 0);
1087 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1088 		newparent = 1;
1089 	if (doingdirectory && newparent) {
1090 		if (error)	/* write access check above */
1091 			goto bad;
1092 		if (xp != NULL)
1093 			vput(tvp);
1094 		/*
1095 		 * doscheckpath() vput()'s dp,
1096 		 * so we have to do a relookup afterwards
1097 		 */
1098 		error = doscheckpath(ip, dp);
1099 		if (error)
1100 			goto out;
1101 		if ((tcnp->cn_flags & SAVESTART) == 0)
1102 			panic("msdosfs_rename: lost to startdir");
1103 		error = relookup(tdvp, &tvp, tcnp);
1104 		if (error)
1105 			goto out;
1106 		dp = VTODE(tdvp);
1107 		xp = tvp ? VTODE(tvp) : NULL;
1108 	}
1109 
1110 	if (xp != NULL) {
1111 		/*
1112 		 * Target must be empty if a directory and have no links
1113 		 * to it. Also, ensure source and target are compatible
1114 		 * (both directories, or both not directories).
1115 		 */
1116 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1117 			if (!dosdirempty(xp)) {
1118 				error = ENOTEMPTY;
1119 				goto bad;
1120 			}
1121 			if (!doingdirectory) {
1122 				error = ENOTDIR;
1123 				goto bad;
1124 			}
1125 			cache_purge(tdvp);
1126 		} else if (doingdirectory) {
1127 			error = EISDIR;
1128 			goto bad;
1129 		}
1130 		error = removede(dp, xp);
1131 		if (error)
1132 			goto bad;
1133 		vput(tvp);
1134 		xp = NULL;
1135 	}
1136 
1137 	/*
1138 	 * Convert the filename in tcnp into a dos filename. We copy this
1139 	 * into the denode and directory entry for the destination
1140 	 * file/directory.
1141 	 */
1142 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1143 	if (error)
1144 		goto abortit;
1145 
1146 	/*
1147 	 * Since from wasn't locked at various places above,
1148 	 * have to do a relookup here.
1149 	 */
1150 	fcnp->cn_flags &= ~MODMASK;
1151 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1152 	if ((fcnp->cn_flags & SAVESTART) == 0)
1153 		panic("msdosfs_rename: lost from startdir");
1154 	if (!newparent)
1155 		VOP_UNLOCK(tdvp, 0);
1156 	if (relookup(fdvp, &fvp, fcnp) == 0)
1157 		vrele(fdvp);
1158 	if (fvp == NULL) {
1159 		/*
1160 		 * From name has disappeared.
1161 		 */
1162 		if (doingdirectory)
1163 			panic("rename: lost dir entry");
1164 		if (newparent)
1165 			VOP_UNLOCK(tdvp, 0);
1166 		vrele(tdvp);
1167 		vrele(ap->a_fvp);
1168 		return 0;
1169 	}
1170 	xp = VTODE(fvp);
1171 	zp = VTODE(fdvp);
1172 	from_diroffset = zp->de_fndoffset;
1173 
1174 	/*
1175 	 * Ensure that the directory entry still exists and has not
1176 	 * changed till now. If the source is a file the entry may
1177 	 * have been unlinked or renamed. In either case there is
1178 	 * no further work to be done. If the source is a directory
1179 	 * then it cannot have been rmdir'ed or renamed; this is
1180 	 * prohibited by the DE_RENAME flag.
1181 	 */
1182 	if (xp != ip) {
1183 		if (doingdirectory)
1184 			panic("rename: lost dir entry");
1185 		VOP_UNLOCK(fvp, 0);
1186 		if (newparent)
1187 			VOP_UNLOCK(fdvp, 0);
1188 		vrele(ap->a_fvp);
1189 		xp = NULL;
1190 	} else {
1191 		vrele(fvp);
1192 		xp = NULL;
1193 
1194 		/*
1195 		 * First write a new entry in the destination
1196 		 * directory and mark the entry in the source directory
1197 		 * as deleted.  Then move the denode to the correct hash
1198 		 * chain for its new location in the filesystem.  And, if
1199 		 * we moved a directory, then update its .. entry to point
1200 		 * to the new parent directory.
1201 		 */
1202 		bcopy(ip->de_Name, oldname, 11);
1203 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1204 		dp->de_fndoffset = to_diroffset;
1205 		dp->de_fndcnt = to_count;
1206 		error = createde(ip, dp, (struct denode **)0, tcnp);
1207 		if (error) {
1208 			bcopy(oldname, ip->de_Name, 11);
1209 			if (newparent)
1210 				VOP_UNLOCK(fdvp, 0);
1211 			VOP_UNLOCK(fvp, 0);
1212 			goto bad;
1213 		}
1214 		ip->de_refcnt++;
1215 		zp->de_fndoffset = from_diroffset;
1216 		error = removede(zp, ip);
1217 		if (error) {
1218 			/* XXX should downgrade to ro here, fs is corrupt */
1219 			if (newparent)
1220 				VOP_UNLOCK(fdvp, 0);
1221 			VOP_UNLOCK(fvp, 0);
1222 			goto bad;
1223 		}
1224 		if (!doingdirectory) {
1225 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1226 				       &ip->de_dirclust, 0);
1227 			if (error) {
1228 				/* XXX should downgrade to ro here, fs is corrupt */
1229 				if (newparent)
1230 					VOP_UNLOCK(fdvp, 0);
1231 				VOP_UNLOCK(fvp, 0);
1232 				goto bad;
1233 			}
1234 			if (ip->de_dirclust == MSDOSFSROOT)
1235 				ip->de_diroffset = to_diroffset;
1236 			else
1237 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1238 		}
1239 		reinsert(ip);
1240 		if (newparent)
1241 			VOP_UNLOCK(fdvp, 0);
1242 	}
1243 
1244 	/*
1245 	 * If we moved a directory to a new parent directory, then we must
1246 	 * fixup the ".." entry in the moved directory.
1247 	 */
1248 	if (doingdirectory && newparent) {
1249 		cn = ip->de_StartCluster;
1250 		if (cn == MSDOSFSROOT) {
1251 			/* this should never happen */
1252 			panic("msdosfs_rename(): updating .. in root directory?");
1253 		} else
1254 			bn = cntobn(pmp, cn);
1255 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1256 			      NOCRED, &bp);
1257 		if (error) {
1258 			/* XXX should downgrade to ro here, fs is corrupt */
1259 			brelse(bp);
1260 			VOP_UNLOCK(fvp, 0);
1261 			goto bad;
1262 		}
1263 		dotdotp = (struct direntry *)bp->b_data + 1;
1264 		putushort(dotdotp->deStartCluster, dp->de_StartCluster);
1265 		if (FAT32(pmp))
1266 			putushort(dotdotp->deHighClust, dp->de_StartCluster >> 16);
1267 		if (fvp->v_mount->mnt_flag & MNT_ASYNC)
1268 			bdwrite(bp);
1269 		else if ((error = bwrite(bp)) != 0) {
1270 			/* XXX should downgrade to ro here, fs is corrupt */
1271 			VOP_UNLOCK(fvp, 0);
1272 			goto bad;
1273 		}
1274 	}
1275 
1276 	VOP_UNLOCK(fvp, 0);
1277 bad:
1278 	if (xp)
1279 		vput(tvp);
1280 	vput(tdvp);
1281 out:
1282 	ip->de_flag &= ~DE_RENAME;
1283 	vrele(fdvp);
1284 	vrele(fvp);
1285 	return (error);
1286 
1287 }
1288 
1289 static struct {
1290 	struct direntry dot;
1291 	struct direntry dotdot;
1292 } dosdirtemplate = {
1293 	{	".       ", "   ",			/* the . entry */
1294 		ATTR_DIRECTORY,				/* file attribute */
1295 		0,					/* reserved */
1296 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1297 		{ 0, 0 },				/* access date */
1298 		{ 0, 0 },				/* high bits of start cluster */
1299 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1300 		{ 0, 0 },				/* startcluster */
1301 		{ 0, 0, 0, 0 }				/* filesize */
1302 	},
1303 	{	"..      ", "   ",			/* the .. entry */
1304 		ATTR_DIRECTORY,				/* file attribute */
1305 		0,					/* reserved */
1306 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1307 		{ 0, 0 },				/* access date */
1308 		{ 0, 0 },				/* high bits of start cluster */
1309 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1310 		{ 0, 0 },				/* startcluster */
1311 		{ 0, 0, 0, 0 }				/* filesize */
1312 	}
1313 };
1314 
1315 static int
1316 msdosfs_mkdir(ap)
1317 	struct vop_mkdir_args /* {
1318 		struct vnode *a_dvp;
1319 		struct vnode **a_vpp;
1320 		struvt componentname *a_cnp;
1321 		struct vattr *a_vap;
1322 	} */ *ap;
1323 {
1324 	struct componentname *cnp = ap->a_cnp;
1325 	struct denode *dep;
1326 	struct denode *pdep = VTODE(ap->a_dvp);
1327 	struct direntry *denp;
1328 	struct msdosfsmount *pmp = pdep->de_pmp;
1329 	struct buf *bp;
1330 	u_long newcluster, pcl;
1331 	int bn;
1332 	int error;
1333 	struct denode ndirent;
1334 	struct timespec ts;
1335 
1336 	/*
1337 	 * If this is the root directory and there is no space left we
1338 	 * can't do anything.  This is because the root directory can not
1339 	 * change size.
1340 	 */
1341 	if (pdep->de_StartCluster == MSDOSFSROOT
1342 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1343 		error = ENOSPC;
1344 		goto bad2;
1345 	}
1346 
1347 	/*
1348 	 * Allocate a cluster to hold the about to be created directory.
1349 	 */
1350 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1351 	if (error)
1352 		goto bad2;
1353 
1354 	bzero(&ndirent, sizeof(ndirent));
1355 	ndirent.de_pmp = pmp;
1356 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1357 	getnanotime(&ts);
1358 	DETIMES(&ndirent, &ts, &ts, &ts);
1359 
1360 	/*
1361 	 * Now fill the cluster with the "." and ".." entries. And write
1362 	 * the cluster to disk.  This way it is there for the parent
1363 	 * directory to be pointing at if there were a crash.
1364 	 */
1365 	bn = cntobn(pmp, newcluster);
1366 	/* always succeeds */
1367 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1368 	bzero(bp->b_data, pmp->pm_bpcluster);
1369 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1370 	denp = (struct direntry *)bp->b_data;
1371 	putushort(denp[0].deStartCluster, newcluster);
1372 	putushort(denp[0].deCDate, ndirent.de_CDate);
1373 	putushort(denp[0].deCTime, ndirent.de_CTime);
1374 	denp[0].deCHundredth = ndirent.de_CHun;
1375 	putushort(denp[0].deADate, ndirent.de_ADate);
1376 	putushort(denp[0].deMDate, ndirent.de_MDate);
1377 	putushort(denp[0].deMTime, ndirent.de_MTime);
1378 	pcl = pdep->de_StartCluster;
1379 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1380 		pcl = 0;
1381 	putushort(denp[1].deStartCluster, pcl);
1382 	putushort(denp[1].deCDate, ndirent.de_CDate);
1383 	putushort(denp[1].deCTime, ndirent.de_CTime);
1384 	denp[1].deCHundredth = ndirent.de_CHun;
1385 	putushort(denp[1].deADate, ndirent.de_ADate);
1386 	putushort(denp[1].deMDate, ndirent.de_MDate);
1387 	putushort(denp[1].deMTime, ndirent.de_MTime);
1388 	if (FAT32(pmp)) {
1389 		putushort(denp[0].deHighClust, newcluster >> 16);
1390 		putushort(denp[1].deHighClust, pdep->de_StartCluster >> 16);
1391 	}
1392 
1393 	if (ap->a_dvp->v_mount->mnt_flag & MNT_ASYNC)
1394 		bdwrite(bp);
1395 	else if ((error = bwrite(bp)) != 0)
1396 		goto bad;
1397 
1398 	/*
1399 	 * Now build up a directory entry pointing to the newly allocated
1400 	 * cluster.  This will be written to an empty slot in the parent
1401 	 * directory.
1402 	 */
1403 #ifdef DIAGNOSTIC
1404 	if ((cnp->cn_flags & HASBUF) == 0)
1405 		panic("msdosfs_mkdir: no name");
1406 #endif
1407 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1408 	if (error)
1409 		goto bad;
1410 
1411 	ndirent.de_Attributes = ATTR_DIRECTORY;
1412 	ndirent.de_LowerCase = 0;
1413 	ndirent.de_StartCluster = newcluster;
1414 	ndirent.de_FileSize = 0;
1415 	ndirent.de_dev = pdep->de_dev;
1416 	error = createde(&ndirent, pdep, &dep, cnp);
1417 	if (error)
1418 		goto bad;
1419 	*ap->a_vpp = DETOV(dep);
1420 	return (0);
1421 
1422 bad:
1423 	clusterfree(pmp, newcluster, NULL);
1424 bad2:
1425 	return (error);
1426 }
1427 
1428 static int
1429 msdosfs_rmdir(ap)
1430 	struct vop_rmdir_args /* {
1431 		struct vnode *a_dvp;
1432 		struct vnode *a_vp;
1433 		struct componentname *a_cnp;
1434 	} */ *ap;
1435 {
1436 	struct vnode *vp = ap->a_vp;
1437 	struct vnode *dvp = ap->a_dvp;
1438 	struct componentname *cnp = ap->a_cnp;
1439 	struct denode *ip, *dp;
1440 	struct thread *td = cnp->cn_thread;
1441 	int error;
1442 
1443 	ip = VTODE(vp);
1444 	dp = VTODE(dvp);
1445 
1446 	/*
1447 	 * Verify the directory is empty (and valid).
1448 	 * (Rmdir ".." won't be valid since
1449 	 *  ".." will contain a reference to
1450 	 *  the current directory and thus be
1451 	 *  non-empty.)
1452 	 */
1453 	error = 0;
1454 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1455 		error = ENOTEMPTY;
1456 		goto out;
1457 	}
1458 	/*
1459 	 * Delete the entry from the directory.  For dos filesystems this
1460 	 * gets rid of the directory entry on disk, the in memory copy
1461 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1462 	 * being found by deget().  When the vput() on dep is done we give
1463 	 * up access and eventually msdosfs_reclaim() will be called which
1464 	 * will remove it from the denode cache.
1465 	 */
1466 	error = removede(dp, ip);
1467 	if (error)
1468 		goto out;
1469 	/*
1470 	 * This is where we decrement the link count in the parent
1471 	 * directory.  Since dos filesystems don't do this we just purge
1472 	 * the name cache.
1473 	 */
1474 	cache_purge(dvp);
1475 	VOP_UNLOCK(dvp, 0);
1476 	/*
1477 	 * Truncate the directory that is being deleted.
1478 	 */
1479 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred, td);
1480 	cache_purge(vp);
1481 
1482 	vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1483 out:
1484 	return (error);
1485 }
1486 
1487 /*
1488  * DOS filesystems don't know what symlinks are.
1489  */
1490 static int
1491 msdosfs_symlink(ap)
1492 	struct vop_symlink_args /* {
1493 		struct vnode *a_dvp;
1494 		struct vnode **a_vpp;
1495 		struct componentname *a_cnp;
1496 		struct vattr *a_vap;
1497 		char *a_target;
1498 	} */ *ap;
1499 {
1500 	return (EOPNOTSUPP);
1501 }
1502 
1503 static int
1504 msdosfs_readdir(ap)
1505 	struct vop_readdir_args /* {
1506 		struct vnode *a_vp;
1507 		struct uio *a_uio;
1508 		struct ucred *a_cred;
1509 		int *a_eofflag;
1510 		int *a_ncookies;
1511 		u_long **a_cookies;
1512 	} */ *ap;
1513 {
1514 	struct mbnambuf nb;
1515 	int error = 0;
1516 	int diff;
1517 	long n;
1518 	int blsize;
1519 	long on;
1520 	u_long cn;
1521 	uint64_t fileno;
1522 	u_long dirsperblk;
1523 	long bias = 0;
1524 	daddr_t bn, lbn;
1525 	struct buf *bp;
1526 	struct denode *dep = VTODE(ap->a_vp);
1527 	struct msdosfsmount *pmp = dep->de_pmp;
1528 	struct direntry *dentp;
1529 	struct dirent dirbuf;
1530 	struct uio *uio = ap->a_uio;
1531 	u_long *cookies = NULL;
1532 	int ncookies = 0;
1533 	off_t offset, off;
1534 	int chksum = -1;
1535 
1536 #ifdef MSDOSFS_DEBUG
1537 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1538 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1539 #endif
1540 
1541 	/*
1542 	 * msdosfs_readdir() won't operate properly on regular files since
1543 	 * it does i/o only with the the filesystem vnode, and hence can
1544 	 * retrieve the wrong block from the buffer cache for a plain file.
1545 	 * So, fail attempts to readdir() on a plain file.
1546 	 */
1547 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1548 		return (ENOTDIR);
1549 
1550 	/*
1551 	 * To be safe, initialize dirbuf
1552 	 */
1553 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1554 
1555 	/*
1556 	 * If the user buffer is smaller than the size of one dos directory
1557 	 * entry or the file offset is not a multiple of the size of a
1558 	 * directory entry, then we fail the read.
1559 	 */
1560 	off = offset = uio->uio_offset;
1561 	if (uio->uio_resid < sizeof(struct direntry) ||
1562 	    (offset & (sizeof(struct direntry) - 1)))
1563 		return (EINVAL);
1564 
1565 	if (ap->a_ncookies) {
1566 		ncookies = uio->uio_resid / 16;
1567 		MALLOC(cookies, u_long *, ncookies * sizeof(u_long), M_TEMP,
1568 		       M_WAITOK);
1569 		*ap->a_cookies = cookies;
1570 		*ap->a_ncookies = ncookies;
1571 	}
1572 
1573 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1574 
1575 	/*
1576 	 * If they are reading from the root directory then, we simulate
1577 	 * the . and .. entries since these don't exist in the root
1578 	 * directory.  We also set the offset bias to make up for having to
1579 	 * simulate these entries. By this I mean that at file offset 64 we
1580 	 * read the first entry in the root directory that lives on disk.
1581 	 */
1582 	if (dep->de_StartCluster == MSDOSFSROOT
1583 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1584 #if 0
1585 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1586 		    offset);
1587 #endif
1588 		bias = 2 * sizeof(struct direntry);
1589 		if (offset < bias) {
1590 			for (n = (int)offset / sizeof(struct direntry);
1591 			     n < 2; n++) {
1592 				if (FAT32(pmp))
1593 					fileno = (uint64_t)cntobn(pmp,
1594 								 pmp->pm_rootdirblk)
1595 							  * dirsperblk;
1596 				else
1597 					fileno = 1;
1598 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1599 					dirbuf.d_fileno =
1600 					    msdosfs_fileno_map(pmp->pm_mountp,
1601 					    fileno);
1602 				} else {
1603 
1604 					dirbuf.d_fileno = (uint32_t)fileno;
1605 				}
1606 				dirbuf.d_type = DT_DIR;
1607 				switch (n) {
1608 				case 0:
1609 					dirbuf.d_namlen = 1;
1610 					strcpy(dirbuf.d_name, ".");
1611 					break;
1612 				case 1:
1613 					dirbuf.d_namlen = 2;
1614 					strcpy(dirbuf.d_name, "..");
1615 					break;
1616 				}
1617 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1618 				if (uio->uio_resid < dirbuf.d_reclen)
1619 					goto out;
1620 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1621 				if (error)
1622 					goto out;
1623 				offset += sizeof(struct direntry);
1624 				off = offset;
1625 				if (cookies) {
1626 					*cookies++ = offset;
1627 					if (--ncookies <= 0)
1628 						goto out;
1629 				}
1630 			}
1631 		}
1632 	}
1633 
1634 	mbnambuf_init(&nb);
1635 	off = offset;
1636 	while (uio->uio_resid > 0) {
1637 		lbn = de_cluster(pmp, offset - bias);
1638 		on = (offset - bias) & pmp->pm_crbomask;
1639 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1640 		diff = dep->de_FileSize - (offset - bias);
1641 		if (diff <= 0)
1642 			break;
1643 		n = min(n, diff);
1644 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1645 		if (error)
1646 			break;
1647 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1648 		if (error) {
1649 			brelse(bp);
1650 			return (error);
1651 		}
1652 		n = min(n, blsize - bp->b_resid);
1653 		if (n == 0) {
1654 			brelse(bp);
1655 			return (EIO);
1656 		}
1657 
1658 		/*
1659 		 * Convert from dos directory entries to fs-independent
1660 		 * directory entries.
1661 		 */
1662 		for (dentp = (struct direntry *)(bp->b_data + on);
1663 		     (char *)dentp < bp->b_data + on + n;
1664 		     dentp++, offset += sizeof(struct direntry)) {
1665 #if 0
1666 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1667 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1668 #endif
1669 			/*
1670 			 * If this is an unused entry, we can stop.
1671 			 */
1672 			if (dentp->deName[0] == SLOT_EMPTY) {
1673 				brelse(bp);
1674 				goto out;
1675 			}
1676 			/*
1677 			 * Skip deleted entries.
1678 			 */
1679 			if (dentp->deName[0] == SLOT_DELETED) {
1680 				chksum = -1;
1681 				mbnambuf_init(&nb);
1682 				continue;
1683 			}
1684 
1685 			/*
1686 			 * Handle Win95 long directory entries
1687 			 */
1688 			if (dentp->deAttributes == ATTR_WIN95) {
1689 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1690 					continue;
1691 				chksum = win2unixfn(&nb,
1692 				    (struct winentry *)dentp, chksum, pmp);
1693 				continue;
1694 			}
1695 
1696 			/*
1697 			 * Skip volume labels
1698 			 */
1699 			if (dentp->deAttributes & ATTR_VOLUME) {
1700 				chksum = -1;
1701 				mbnambuf_init(&nb);
1702 				continue;
1703 			}
1704 			/*
1705 			 * This computation of d_fileno must match
1706 			 * the computation of va_fileid in
1707 			 * msdosfs_getattr.
1708 			 */
1709 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1710 				fileno = getushort(dentp->deStartCluster);
1711 				if (FAT32(pmp))
1712 					fileno |= getushort(dentp->deHighClust) << 16;
1713 				/* if this is the root directory */
1714 				if (fileno == MSDOSFSROOT)
1715 					if (FAT32(pmp))
1716 						fileno = (uint64_t)cntobn(pmp,
1717 								pmp->pm_rootdirblk)
1718 							 * dirsperblk;
1719 					else
1720 						fileno = 1;
1721 				else
1722 					fileno = (uint64_t)cntobn(pmp, fileno) *
1723 					    dirsperblk;
1724 				dirbuf.d_type = DT_DIR;
1725 			} else {
1726 				fileno = (uoff_t)offset /
1727 				    sizeof(struct direntry);
1728 				dirbuf.d_type = DT_REG;
1729 			}
1730 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1731 				dirbuf.d_fileno =
1732 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1733 			} else
1734 				dirbuf.d_fileno = (uint32_t)fileno;
1735 
1736 			if (chksum != winChksum(dentp)) {
1737 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1738 				    (u_char *)dirbuf.d_name,
1739 				    dentp->deLowerCase |
1740 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1741 					(LCASE_BASE | LCASE_EXT) : 0),
1742 				    pmp);
1743 				mbnambuf_init(&nb);
1744 			} else
1745 				mbnambuf_flush(&nb, &dirbuf);
1746 			chksum = -1;
1747 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1748 			if (uio->uio_resid < dirbuf.d_reclen) {
1749 				brelse(bp);
1750 				goto out;
1751 			}
1752 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1753 			if (error) {
1754 				brelse(bp);
1755 				goto out;
1756 			}
1757 			if (cookies) {
1758 				*cookies++ = offset + sizeof(struct direntry);
1759 				if (--ncookies <= 0) {
1760 					brelse(bp);
1761 					goto out;
1762 				}
1763 			}
1764 			off = offset + sizeof(struct direntry);
1765 		}
1766 		brelse(bp);
1767 	}
1768 out:
1769 	/* Subtract unused cookies */
1770 	if (ap->a_ncookies)
1771 		*ap->a_ncookies -= ncookies;
1772 
1773 	uio->uio_offset = off;
1774 
1775 	/*
1776 	 * Set the eofflag (NFS uses it)
1777 	 */
1778 	if (ap->a_eofflag) {
1779 		if (dep->de_FileSize - (offset - bias) <= 0)
1780 			*ap->a_eofflag = 1;
1781 		else
1782 			*ap->a_eofflag = 0;
1783 	}
1784 	return (error);
1785 }
1786 
1787 /*-
1788  * a_vp   - pointer to the file's vnode
1789  * a_bn   - logical block number within the file (cluster number for us)
1790  * a_bop  - where to return the bufobj of the special file containing the fs
1791  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1792  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1793  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1794  *          blocks whose physical blocks (together with a_bn's physical block)
1795  *          are contiguous.
1796  * a_runb - where to return the "run before" a_bn.
1797  */
1798 static int
1799 msdosfs_bmap(ap)
1800 	struct vop_bmap_args /* {
1801 		struct vnode *a_vp;
1802 		daddr_t a_bn;
1803 		struct bufobj **a_bop;
1804 		daddr_t *a_bnp;
1805 		int *a_runp;
1806 		int *a_runb;
1807 	} */ *ap;
1808 {
1809 	struct denode *dep;
1810 	struct mount *mp;
1811 	struct msdosfsmount *pmp;
1812 	struct vnode *vp;
1813 	daddr_t runbn;
1814 	u_long cn;
1815 	int bnpercn, error, maxio, maxrun, run;
1816 
1817 	vp = ap->a_vp;
1818 	dep = VTODE(vp);
1819 	pmp = dep->de_pmp;
1820 	if (ap->a_bop != NULL)
1821 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1822 	if (ap->a_bnp == NULL)
1823 		return (0);
1824 	if (ap->a_runp != NULL)
1825 		*ap->a_runp = 0;
1826 	if (ap->a_runb != NULL)
1827 		*ap->a_runb = 0;
1828 	cn = ap->a_bn;
1829 	if (cn != ap->a_bn)
1830 		return (EFBIG);
1831 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1832 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1833 		return (error);
1834 
1835 	mp = vp->v_mount;
1836 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1837 	bnpercn = de_cn2bn(pmp, 1);
1838 	if (ap->a_runp != NULL) {
1839 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1840 		for (run = 1; run <= maxrun; run++) {
1841 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1842 			    runbn != *ap->a_bnp + run * bnpercn)
1843 				break;
1844 		}
1845 		*ap->a_runp = run - 1;
1846 	}
1847 	if (ap->a_runb != NULL) {
1848 		maxrun = ulmin(maxio - 1, cn);
1849 		for (run = 1; run < maxrun; run++) {
1850 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1851 			    runbn != *ap->a_bnp - run * bnpercn)
1852 				break;
1853 		}
1854 		*ap->a_runb = run - 1;
1855 	}
1856 	return (0);
1857 }
1858 
1859 static int
1860 msdosfs_strategy(ap)
1861 	struct vop_strategy_args /* {
1862 		struct vnode *a_vp;
1863 		struct buf *a_bp;
1864 	} */ *ap;
1865 {
1866 	struct buf *bp = ap->a_bp;
1867 	struct denode *dep = VTODE(ap->a_vp);
1868 	struct bufobj *bo;
1869 	int error = 0;
1870 	daddr_t blkno;
1871 
1872 	/*
1873 	 * If we don't already know the filesystem relative block number
1874 	 * then get it using pcbmap().  If pcbmap() returns the block
1875 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1876 	 * don't allow files with holes, so we shouldn't ever see this.
1877 	 */
1878 	if (bp->b_blkno == bp->b_lblkno) {
1879 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1880 		bp->b_blkno = blkno;
1881 		if (error) {
1882 			bp->b_error = error;
1883 			bp->b_ioflags |= BIO_ERROR;
1884 			bufdone(bp);
1885 			return (error);
1886 		}
1887 		if ((long)bp->b_blkno == -1)
1888 			vfs_bio_clrbuf(bp);
1889 	}
1890 	if (bp->b_blkno == -1) {
1891 		bufdone(bp);
1892 		return (0);
1893 	}
1894 	/*
1895 	 * Read/write the block from/to the disk that contains the desired
1896 	 * file block.
1897 	 */
1898 	bp->b_iooffset = dbtob(bp->b_blkno);
1899 	bo = dep->de_pmp->pm_bo;
1900 	BO_STRATEGY(bo, bp);
1901 	return (0);
1902 }
1903 
1904 static int
1905 msdosfs_print(ap)
1906 	struct vop_print_args /* {
1907 		struct vnode *vp;
1908 	} */ *ap;
1909 {
1910 	struct denode *dep = VTODE(ap->a_vp);
1911 
1912 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1913 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1914 	printf("on dev %s\n", devtoname(dep->de_dev));
1915 	return (0);
1916 }
1917 
1918 static int
1919 msdosfs_pathconf(ap)
1920 	struct vop_pathconf_args /* {
1921 		struct vnode *a_vp;
1922 		int a_name;
1923 		int *a_retval;
1924 	} */ *ap;
1925 {
1926 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1927 
1928 	switch (ap->a_name) {
1929 	case _PC_LINK_MAX:
1930 		*ap->a_retval = 1;
1931 		return (0);
1932 	case _PC_NAME_MAX:
1933 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1934 		return (0);
1935 	case _PC_PATH_MAX:
1936 		*ap->a_retval = PATH_MAX;
1937 		return (0);
1938 	case _PC_CHOWN_RESTRICTED:
1939 		*ap->a_retval = 1;
1940 		return (0);
1941 	case _PC_NO_TRUNC:
1942 		*ap->a_retval = 0;
1943 		return (0);
1944 	default:
1945 		return (EINVAL);
1946 	}
1947 	/* NOTREACHED */
1948 }
1949 
1950 static int
1951 msdosfs_advlock(ap)
1952 	struct vop_advlock_args /* {
1953 		struct vnode *a_vp;
1954 		u_char a_id;
1955 		int a_op;
1956 		struct flock *a_fl;
1957 		int a_flags;
1958 	} */ *ap;
1959 {
1960 	struct denode *dep = VTODE(ap->a_vp);
1961 
1962 	return (lf_advlock(ap, &dep->de_lockf, dep->de_FileSize));
1963 }
1964 
1965 static int
1966 msdosfs_vptofh(ap)
1967 	struct vop_vptofh_args /* {
1968 		struct vnode *a_vp;
1969 		struct fid *a_fhp;
1970 	} */ *ap;
1971 {
1972 	struct denode *dep;
1973 	struct defid *defhp;
1974 
1975 	dep = VTODE(ap->a_vp);
1976 	defhp = (struct defid *)ap->a_fhp;
1977 	defhp->defid_len = sizeof(struct defid);
1978 	defhp->defid_dirclust = dep->de_dirclust;
1979 	defhp->defid_dirofs = dep->de_diroffset;
1980 	/* defhp->defid_gen = dep->de_gen; */
1981 	return (0);
1982 }
1983 
1984 /* Global vfs data structures for msdosfs */
1985 struct vop_vector msdosfs_vnodeops = {
1986 	.vop_default =		&default_vnodeops,
1987 
1988 	.vop_access =		msdosfs_access,
1989 	.vop_advlock =		msdosfs_advlock,
1990 	.vop_bmap =		msdosfs_bmap,
1991 	.vop_cachedlookup =	msdosfs_lookup,
1992 	.vop_open =		msdosfs_open,
1993 	.vop_close =		msdosfs_close,
1994 	.vop_create =		msdosfs_create,
1995 	.vop_fsync =		msdosfs_fsync,
1996 	.vop_getattr =		msdosfs_getattr,
1997 	.vop_inactive =		msdosfs_inactive,
1998 	.vop_link =		msdosfs_link,
1999 	.vop_lookup =		vfs_cache_lookup,
2000 	.vop_mkdir =		msdosfs_mkdir,
2001 	.vop_mknod =		msdosfs_mknod,
2002 	.vop_pathconf =		msdosfs_pathconf,
2003 	.vop_print =		msdosfs_print,
2004 	.vop_read =		msdosfs_read,
2005 	.vop_readdir =		msdosfs_readdir,
2006 	.vop_reclaim =		msdosfs_reclaim,
2007 	.vop_remove =		msdosfs_remove,
2008 	.vop_rename =		msdosfs_rename,
2009 	.vop_rmdir =		msdosfs_rmdir,
2010 	.vop_setattr =		msdosfs_setattr,
2011 	.vop_strategy =		msdosfs_strategy,
2012 	.vop_symlink =		msdosfs_symlink,
2013 	.vop_write =		msdosfs_write,
2014 	.vop_vptofh =		msdosfs_vptofh,
2015 };
2016