xref: /freebsd/sys/fs/msdosfs/msdosfs_vnops.c (revision 2f834a0b41079f9be4dc33ff877d50a5fba869d4)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_vnops.c,v 1.68 1998/02/10 14:10:04 mrg Exp $	*/
3 
4 /*-
5  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
6  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
7  * All rights reserved.
8  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by TooLs GmbH.
21  * 4. The name of TooLs GmbH may not be used to endorse or promote products
22  *    derived from this software without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
25  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
26  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
27  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
28  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
30  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  */
35 /*-
36  * Written by Paul Popelka (paulp@uts.amdahl.com)
37  *
38  * You can do anything you want with this software, just don't say you wrote
39  * it, and don't remove this notice.
40  *
41  * This software is provided "as is".
42  *
43  * The author supplies this software to be publicly redistributed on the
44  * understanding that the author is not responsible for the correct
45  * functioning of this software in any circumstances and is not liable for
46  * any damages caused by this software.
47  *
48  * October 1992
49  */
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/clock.h>
56 #include <sys/dirent.h>
57 #include <sys/lock.h>
58 #include <sys/lockf.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/mutex.h>
62 #include <sys/namei.h>
63 #include <sys/priv.h>
64 #include <sys/stat.h>
65 #include <sys/unistd.h>
66 #include <sys/vnode.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 
71 #include <fs/msdosfs/bpb.h>
72 #include <fs/msdosfs/direntry.h>
73 #include <fs/msdosfs/denode.h>
74 #include <fs/msdosfs/fat.h>
75 #include <fs/msdosfs/msdosfsmount.h>
76 
77 #define	DOS_FILESIZE_MAX	0xffffffff
78 
79 /*
80  * Prototypes for MSDOSFS vnode operations
81  */
82 static vop_create_t	msdosfs_create;
83 static vop_mknod_t	msdosfs_mknod;
84 static vop_open_t	msdosfs_open;
85 static vop_close_t	msdosfs_close;
86 static vop_access_t	msdosfs_access;
87 static vop_getattr_t	msdosfs_getattr;
88 static vop_setattr_t	msdosfs_setattr;
89 static vop_read_t	msdosfs_read;
90 static vop_write_t	msdosfs_write;
91 static vop_fsync_t	msdosfs_fsync;
92 static vop_remove_t	msdosfs_remove;
93 static vop_link_t	msdosfs_link;
94 static vop_rename_t	msdosfs_rename;
95 static vop_mkdir_t	msdosfs_mkdir;
96 static vop_rmdir_t	msdosfs_rmdir;
97 static vop_symlink_t	msdosfs_symlink;
98 static vop_readdir_t	msdosfs_readdir;
99 static vop_bmap_t	msdosfs_bmap;
100 static vop_strategy_t	msdosfs_strategy;
101 static vop_print_t	msdosfs_print;
102 static vop_pathconf_t	msdosfs_pathconf;
103 static vop_vptofh_t	msdosfs_vptofh;
104 
105 /*
106  * Some general notes:
107  *
108  * In the ufs filesystem the inodes, superblocks, and indirect blocks are
109  * read/written using the vnode for the filesystem. Blocks that represent
110  * the contents of a file are read/written using the vnode for the file
111  * (including directories when they are read/written as files). This
112  * presents problems for the dos filesystem because data that should be in
113  * an inode (if dos had them) resides in the directory itself.  Since we
114  * must update directory entries without the benefit of having the vnode
115  * for the directory we must use the vnode for the filesystem.  This means
116  * that when a directory is actually read/written (via read, write, or
117  * readdir, or seek) we must use the vnode for the filesystem instead of
118  * the vnode for the directory as would happen in ufs. This is to insure we
119  * retreive the correct block from the buffer cache since the hash value is
120  * based upon the vnode address and the desired block number.
121  */
122 
123 /*
124  * Create a regular file. On entry the directory to contain the file being
125  * created is locked.  We must release before we return. We must also free
126  * the pathname buffer pointed at by cnp->cn_pnbuf, always on error, or
127  * only if the SAVESTART bit in cn_flags is clear on success.
128  */
129 static int
130 msdosfs_create(ap)
131 	struct vop_create_args /* {
132 		struct vnode *a_dvp;
133 		struct vnode **a_vpp;
134 		struct componentname *a_cnp;
135 		struct vattr *a_vap;
136 	} */ *ap;
137 {
138 	struct componentname *cnp = ap->a_cnp;
139 	struct denode ndirent;
140 	struct denode *dep;
141 	struct denode *pdep = VTODE(ap->a_dvp);
142 	struct timespec ts;
143 	int error;
144 
145 #ifdef MSDOSFS_DEBUG
146 	printf("msdosfs_create(cnp %p, vap %p\n", cnp, ap->a_vap);
147 #endif
148 
149 	/*
150 	 * If this is the root directory and there is no space left we
151 	 * can't do anything.  This is because the root directory can not
152 	 * change size.
153 	 */
154 	if (pdep->de_StartCluster == MSDOSFSROOT
155 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
156 		error = ENOSPC;
157 		goto bad;
158 	}
159 
160 	/*
161 	 * Create a directory entry for the file, then call createde() to
162 	 * have it installed. NOTE: DOS files are always executable.  We
163 	 * use the absence of the owner write bit to make the file
164 	 * readonly.
165 	 */
166 #ifdef DIAGNOSTIC
167 	if ((cnp->cn_flags & HASBUF) == 0)
168 		panic("msdosfs_create: no name");
169 #endif
170 	bzero(&ndirent, sizeof(ndirent));
171 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
172 	if (error)
173 		goto bad;
174 
175 	ndirent.de_Attributes = ATTR_ARCHIVE;
176 	ndirent.de_LowerCase = 0;
177 	ndirent.de_StartCluster = 0;
178 	ndirent.de_FileSize = 0;
179 	ndirent.de_pmp = pdep->de_pmp;
180 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
181 	getnanotime(&ts);
182 	DETIMES(&ndirent, &ts, &ts, &ts);
183 	error = createde(&ndirent, pdep, &dep, cnp);
184 	if (error)
185 		goto bad;
186 	*ap->a_vpp = DETOV(dep);
187 	if ((cnp->cn_flags & MAKEENTRY) != 0)
188 		cache_enter(ap->a_dvp, *ap->a_vpp, cnp);
189 	return (0);
190 
191 bad:
192 	return (error);
193 }
194 
195 static int
196 msdosfs_mknod(ap)
197 	struct vop_mknod_args /* {
198 		struct vnode *a_dvp;
199 		struct vnode **a_vpp;
200 		struct componentname *a_cnp;
201 		struct vattr *a_vap;
202 	} */ *ap;
203 {
204 
205     return (EINVAL);
206 }
207 
208 static int
209 msdosfs_open(ap)
210 	struct vop_open_args /* {
211 		struct vnode *a_vp;
212 		int a_mode;
213 		struct ucred *a_cred;
214 		struct thread *a_td;
215 		struct file *a_fp;
216 	} */ *ap;
217 {
218 	struct denode *dep = VTODE(ap->a_vp);
219 	vnode_create_vobject(ap->a_vp, dep->de_FileSize, ap->a_td);
220 	return 0;
221 }
222 
223 static int
224 msdosfs_close(ap)
225 	struct vop_close_args /* {
226 		struct vnode *a_vp;
227 		int a_fflag;
228 		struct ucred *a_cred;
229 		struct thread *a_td;
230 	} */ *ap;
231 {
232 	struct vnode *vp = ap->a_vp;
233 	struct denode *dep = VTODE(vp);
234 	struct timespec ts;
235 
236 	VI_LOCK(vp);
237 	if (vp->v_usecount > 1) {
238 		getnanotime(&ts);
239 		DETIMES(dep, &ts, &ts, &ts);
240 	}
241 	VI_UNLOCK(vp);
242 	return 0;
243 }
244 
245 static int
246 msdosfs_access(ap)
247 	struct vop_access_args /* {
248 		struct vnode *a_vp;
249 		accmode_t a_accmode;
250 		struct ucred *a_cred;
251 		struct thread *a_td;
252 	} */ *ap;
253 {
254 	struct vnode *vp = ap->a_vp;
255 	struct denode *dep = VTODE(ap->a_vp);
256 	struct msdosfsmount *pmp = dep->de_pmp;
257 	mode_t file_mode;
258 	accmode_t accmode = ap->a_accmode;
259 
260 	file_mode = S_IRWXU|S_IRWXG|S_IRWXO;
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 VREG:
270 		case VDIR:
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 	mode = S_IRWXU|S_IRWXG|S_IRWXO;
326 	vap->va_mode = mode &
327 	    (ap->a_vp->v_type == VDIR ? pmp->pm_dirmask : pmp->pm_mask);
328 	vap->va_uid = pmp->pm_uid;
329 	vap->va_gid = pmp->pm_gid;
330 	vap->va_nlink = 1;
331 	vap->va_rdev = NODEV;
332 	vap->va_size = dep->de_FileSize;
333 	fattime2timespec(dep->de_MDate, dep->de_MTime, 0, 0, &vap->va_mtime);
334 	vap->va_ctime = vap->va_mtime;
335 	if (pmp->pm_flags & MSDOSFSMNT_LONGNAME) {
336 		fattime2timespec(dep->de_ADate, 0, 0, 0, &vap->va_atime);
337 		fattime2timespec(dep->de_CDate, dep->de_CTime, dep->de_CHun,
338 		    0, &vap->va_birthtime);
339 	} else {
340 		vap->va_atime = vap->va_mtime;
341 		vap->va_birthtime.tv_sec = -1;
342 		vap->va_birthtime.tv_nsec = 0;
343 	}
344 	vap->va_flags = 0;
345 	if (dep->de_Attributes & ATTR_ARCHIVE)
346 		vap->va_flags |= UF_ARCHIVE;
347 	if (dep->de_Attributes & ATTR_HIDDEN)
348 		vap->va_flags |= UF_HIDDEN;
349 	if (dep->de_Attributes & ATTR_READONLY)
350 		vap->va_flags |= UF_READONLY;
351 	if (dep->de_Attributes & ATTR_SYSTEM)
352 		vap->va_flags |= UF_SYSTEM;
353 	vap->va_gen = 0;
354 	vap->va_blocksize = pmp->pm_bpcluster;
355 	vap->va_bytes =
356 	    (dep->de_FileSize + pmp->pm_crbomask) & ~pmp->pm_crbomask;
357 	vap->va_type = ap->a_vp->v_type;
358 	vap->va_filerev = dep->de_modrev;
359 	return (0);
360 }
361 
362 static int
363 msdosfs_setattr(ap)
364 	struct vop_setattr_args /* {
365 		struct vnode *a_vp;
366 		struct vattr *a_vap;
367 		struct ucred *a_cred;
368 	} */ *ap;
369 {
370 	struct vnode *vp = ap->a_vp;
371 	struct denode *dep = VTODE(ap->a_vp);
372 	struct msdosfsmount *pmp = dep->de_pmp;
373 	struct vattr *vap = ap->a_vap;
374 	struct ucred *cred = ap->a_cred;
375 	struct thread *td = curthread;
376 	int error = 0;
377 
378 #ifdef MSDOSFS_DEBUG
379 	printf("msdosfs_setattr(): vp %p, vap %p, cred %p\n",
380 	    ap->a_vp, vap, cred);
381 #endif
382 
383 	/*
384 	 * Check for unsettable attributes.
385 	 */
386 	if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
387 	    (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
388 	    (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
389 	    (vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
390 #ifdef MSDOSFS_DEBUG
391 		printf("msdosfs_setattr(): returning EINVAL\n");
392 		printf("    va_type %d, va_nlink %x, va_fsid %lx, va_fileid %lx\n",
393 		    vap->va_type, vap->va_nlink, vap->va_fsid, vap->va_fileid);
394 		printf("    va_blocksize %lx, va_rdev %x, va_bytes %qx, va_gen %lx\n",
395 		    vap->va_blocksize, vap->va_rdev, vap->va_bytes, vap->va_gen);
396 		printf("    va_uid %x, va_gid %x\n",
397 		    vap->va_uid, vap->va_gid);
398 #endif
399 		return (EINVAL);
400 	}
401 
402 	/*
403 	 * We don't allow setting attributes on the root directory.
404 	 * The special case for the root directory is because before
405 	 * FAT32, the root directory didn't have an entry for itself
406 	 * (and was otherwise special).  With FAT32, the root
407 	 * directory is not so special, but still doesn't have an
408 	 * entry for itself.
409 	 */
410 	if (vp->v_vflag & VV_ROOT)
411 		return (EINVAL);
412 
413 	if (vap->va_flags != VNOVAL) {
414 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
415 			return (EROFS);
416 		if (cred->cr_uid != pmp->pm_uid) {
417 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
418 			if (error)
419 				return (error);
420 		}
421 		/*
422 		 * We are very inconsistent about handling unsupported
423 		 * attributes.  We ignored the access time and the
424 		 * read and execute bits.  We were strict for the other
425 		 * attributes.
426 		 */
427 		if (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_READONLY |
428 		    UF_SYSTEM))
429 			return EOPNOTSUPP;
430 		if (vap->va_flags & UF_ARCHIVE)
431 			dep->de_Attributes |= ATTR_ARCHIVE;
432 		else
433 			dep->de_Attributes &= ~ATTR_ARCHIVE;
434 		if (vap->va_flags & UF_HIDDEN)
435 			dep->de_Attributes |= ATTR_HIDDEN;
436 		else
437 			dep->de_Attributes &= ~ATTR_HIDDEN;
438 		/* We don't allow changing the readonly bit on directories. */
439 		if (vp->v_type != VDIR) {
440 			if (vap->va_flags & UF_READONLY)
441 				dep->de_Attributes |= ATTR_READONLY;
442 			else
443 				dep->de_Attributes &= ~ATTR_READONLY;
444 		}
445 		if (vap->va_flags & UF_SYSTEM)
446 			dep->de_Attributes |= ATTR_SYSTEM;
447 		else
448 			dep->de_Attributes &= ~ATTR_SYSTEM;
449 		dep->de_flag |= DE_MODIFIED;
450 	}
451 
452 	if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
453 		uid_t uid;
454 		gid_t gid;
455 
456 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
457 			return (EROFS);
458 		uid = vap->va_uid;
459 		if (uid == (uid_t)VNOVAL)
460 			uid = pmp->pm_uid;
461 		gid = vap->va_gid;
462 		if (gid == (gid_t)VNOVAL)
463 			gid = pmp->pm_gid;
464 		if (cred->cr_uid != pmp->pm_uid || uid != pmp->pm_uid ||
465 		    (gid != pmp->pm_gid && !groupmember(gid, cred))) {
466 			error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0);
467 			if (error)
468 				return (error);
469 		}
470 		if (uid != pmp->pm_uid || gid != pmp->pm_gid)
471 			return EINVAL;
472 	}
473 
474 	if (vap->va_size != VNOVAL) {
475 		switch (vp->v_type) {
476 		case VDIR:
477 			return (EISDIR);
478 		case VREG:
479 			/*
480 			 * Truncation is only supported for regular files,
481 			 * Disallow it if the filesystem is read-only.
482 			 */
483 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
484 				return (EROFS);
485 			break;
486 		default:
487 			/*
488 			 * According to POSIX, the result is unspecified
489 			 * for file types other than regular files,
490 			 * directories and shared memory objects.  We
491 			 * don't support any file types except regular
492 			 * files and directories in this file system, so
493 			 * this (default) case is unreachable and can do
494 			 * anything.  Keep falling through to detrunc()
495 			 * for now.
496 			 */
497 			break;
498 		}
499 		error = detrunc(dep, vap->va_size, 0, cred);
500 		if (error)
501 			return error;
502 	}
503 	if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
504 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
505 			return (EROFS);
506 		error = vn_utimes_perm(vp, vap, cred, td);
507 		if (error != 0)
508 			return (error);
509 		if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95) == 0 &&
510 		    vap->va_atime.tv_sec != VNOVAL) {
511 			dep->de_flag &= ~DE_ACCESS;
512 			timespec2fattime(&vap->va_atime, 0,
513 			    &dep->de_ADate, NULL, NULL);
514 		}
515 		if (vap->va_mtime.tv_sec != VNOVAL) {
516 			dep->de_flag &= ~DE_UPDATE;
517 			timespec2fattime(&vap->va_mtime, 0,
518 			    &dep->de_MDate, &dep->de_MTime, NULL);
519 		}
520 		/*
521 		 * We don't set the archive bit when modifying the time of
522 		 * a directory to emulate the Windows/DOS behavior.
523 		 */
524 		if (vp->v_type != VDIR)
525 			dep->de_Attributes |= ATTR_ARCHIVE;
526 		dep->de_flag |= DE_MODIFIED;
527 	}
528 	/*
529 	 * DOS files only have the ability to have their writability
530 	 * attribute set, so we use the owner write bit to set the readonly
531 	 * attribute.
532 	 */
533 	if (vap->va_mode != (mode_t)VNOVAL) {
534 		if (vp->v_mount->mnt_flag & MNT_RDONLY)
535 			return (EROFS);
536 		if (cred->cr_uid != pmp->pm_uid) {
537 			error = priv_check_cred(cred, PRIV_VFS_ADMIN, 0);
538 			if (error)
539 				return (error);
540 		}
541 		if (vp->v_type != VDIR) {
542 			/* We ignore the read and execute bits. */
543 			if (vap->va_mode & VWRITE)
544 				dep->de_Attributes &= ~ATTR_READONLY;
545 			else
546 				dep->de_Attributes |= ATTR_READONLY;
547 			dep->de_Attributes |= ATTR_ARCHIVE;
548 			dep->de_flag |= DE_MODIFIED;
549 		}
550 	}
551 	return (deupdat(dep, 0));
552 }
553 
554 static int
555 msdosfs_read(ap)
556 	struct vop_read_args /* {
557 		struct vnode *a_vp;
558 		struct uio *a_uio;
559 		int a_ioflag;
560 		struct ucred *a_cred;
561 	} */ *ap;
562 {
563 	int error = 0;
564 	int blsize;
565 	int isadir;
566 	ssize_t orig_resid;
567 	u_int n;
568 	u_long diff;
569 	u_long on;
570 	daddr_t lbn;
571 	daddr_t rablock;
572 	int rasize;
573 	int seqcount;
574 	struct buf *bp;
575 	struct vnode *vp = ap->a_vp;
576 	struct denode *dep = VTODE(vp);
577 	struct msdosfsmount *pmp = dep->de_pmp;
578 	struct uio *uio = ap->a_uio;
579 
580 	/*
581 	 * If they didn't ask for any data, then we are done.
582 	 */
583 	orig_resid = uio->uio_resid;
584 	if (orig_resid == 0)
585 		return (0);
586 
587 	/*
588 	 * The caller is supposed to ensure that
589 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
590 	 * We don't need to check for large offsets as in ffs because
591 	 * dep->de_FileSize <= DOS_FILESIZE_MAX < OFF_MAX, so large
592 	 * offsets cannot cause overflow even in theory.
593 	 */
594 
595 	seqcount = ap->a_ioflag >> IO_SEQSHIFT;
596 
597 	isadir = dep->de_Attributes & ATTR_DIRECTORY;
598 	do {
599 		if (uio->uio_offset >= dep->de_FileSize)
600 			break;
601 		lbn = de_cluster(pmp, uio->uio_offset);
602 		rablock = lbn + 1;
603 		blsize = pmp->pm_bpcluster;
604 		on = uio->uio_offset & pmp->pm_crbomask;
605 		/*
606 		 * If we are operating on a directory file then be sure to
607 		 * do i/o with the vnode for the filesystem instead of the
608 		 * vnode for the directory.
609 		 */
610 		if (isadir) {
611 			/* convert cluster # to block # */
612 			error = pcbmap(dep, lbn, &lbn, 0, &blsize);
613 			if (error == E2BIG) {
614 				error = EINVAL;
615 				break;
616 			} else if (error)
617 				break;
618 			error = bread(pmp->pm_devvp, lbn, blsize, NOCRED, &bp);
619 		} else if (de_cn2off(pmp, rablock) >= dep->de_FileSize) {
620 			error = bread(vp, lbn, blsize, NOCRED, &bp);
621 		} else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
622 			error = cluster_read(vp, dep->de_FileSize, lbn, blsize,
623 			    NOCRED, on + uio->uio_resid, seqcount, 0, &bp);
624 		} else if (seqcount > 1) {
625 			rasize = blsize;
626 			error = breadn(vp, lbn,
627 			    blsize, &rablock, &rasize, 1, NOCRED, &bp);
628 		} else {
629 			error = bread(vp, lbn, blsize, NOCRED, &bp);
630 		}
631 		if (error) {
632 			brelse(bp);
633 			break;
634 		}
635 		diff = pmp->pm_bpcluster - on;
636 		n = diff > uio->uio_resid ? uio->uio_resid : diff;
637 		diff = dep->de_FileSize - uio->uio_offset;
638 		if (diff < n)
639 			n = diff;
640 		diff = blsize - bp->b_resid;
641 		if (diff < n)
642 			n = diff;
643 		error = uiomove(bp->b_data + on, (int) n, uio);
644 		brelse(bp);
645 	} while (error == 0 && uio->uio_resid > 0 && n != 0);
646 	if (!isadir && (error == 0 || uio->uio_resid != orig_resid) &&
647 	    (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
648 		dep->de_flag |= DE_ACCESS;
649 	return (error);
650 }
651 
652 /*
653  * Write data to a file or directory.
654  */
655 static int
656 msdosfs_write(ap)
657 	struct vop_write_args /* {
658 		struct vnode *a_vp;
659 		struct uio *a_uio;
660 		int a_ioflag;
661 		struct ucred *a_cred;
662 	} */ *ap;
663 {
664 	int n;
665 	int croffset;
666 	ssize_t resid;
667 	u_long osize;
668 	int error = 0;
669 	u_long count;
670 	int seqcount;
671 	daddr_t bn, lastcn;
672 	struct buf *bp;
673 	int ioflag = ap->a_ioflag;
674 	struct uio *uio = ap->a_uio;
675 	struct vnode *vp = ap->a_vp;
676 	struct vnode *thisvp;
677 	struct denode *dep = VTODE(vp);
678 	struct msdosfsmount *pmp = dep->de_pmp;
679 	struct ucred *cred = ap->a_cred;
680 
681 #ifdef MSDOSFS_DEBUG
682 	printf("msdosfs_write(vp %p, uio %p, ioflag %x, cred %p\n",
683 	    vp, uio, ioflag, cred);
684 	printf("msdosfs_write(): diroff %lu, dirclust %lu, startcluster %lu\n",
685 	    dep->de_diroffset, dep->de_dirclust, dep->de_StartCluster);
686 #endif
687 
688 	switch (vp->v_type) {
689 	case VREG:
690 		if (ioflag & IO_APPEND)
691 			uio->uio_offset = dep->de_FileSize;
692 		thisvp = vp;
693 		break;
694 	case VDIR:
695 		return EISDIR;
696 	default:
697 		panic("msdosfs_write(): bad file type");
698 	}
699 
700 	/*
701 	 * This is needed (unlike in ffs_write()) because we extend the
702 	 * file outside of the loop but we don't want to extend the file
703 	 * for writes of 0 bytes.
704 	 */
705 	if (uio->uio_resid == 0)
706 		return (0);
707 
708 	/*
709 	 * The caller is supposed to ensure that
710 	 * uio->uio_offset >= 0 and uio->uio_resid >= 0.
711 	 */
712 	if ((uoff_t)uio->uio_offset + uio->uio_resid > DOS_FILESIZE_MAX)
713 		return (EFBIG);
714 
715 	/*
716 	 * If they've exceeded their filesize limit, tell them about it.
717 	 */
718 	if (vn_rlimit_fsize(vp, uio, uio->uio_td))
719 		return (EFBIG);
720 
721 	/*
722 	 * If the offset we are starting the write at is beyond the end of
723 	 * the file, then they've done a seek.  Unix filesystems allow
724 	 * files with holes in them, DOS doesn't so we must fill the hole
725 	 * with zeroed blocks.
726 	 */
727 	if (uio->uio_offset > dep->de_FileSize) {
728 		error = deextend(dep, uio->uio_offset, cred);
729 		if (error)
730 			return (error);
731 	}
732 
733 	/*
734 	 * Remember some values in case the write fails.
735 	 */
736 	resid = uio->uio_resid;
737 	osize = dep->de_FileSize;
738 
739 	/*
740 	 * If we write beyond the end of the file, extend it to its ultimate
741 	 * size ahead of the time to hopefully get a contiguous area.
742 	 */
743 	if (uio->uio_offset + resid > osize) {
744 		count = de_clcount(pmp, uio->uio_offset + resid) -
745 			de_clcount(pmp, osize);
746 		error = extendfile(dep, count, NULL, NULL, 0);
747 		if (error &&  (error != ENOSPC || (ioflag & IO_UNIT)))
748 			goto errexit;
749 		lastcn = dep->de_fc[FC_LASTFC].fc_frcn;
750 	} else
751 		lastcn = de_clcount(pmp, osize) - 1;
752 
753 	seqcount = ioflag >> IO_SEQSHIFT;
754 	do {
755 		if (de_cluster(pmp, uio->uio_offset) > lastcn) {
756 			error = ENOSPC;
757 			break;
758 		}
759 
760 		croffset = uio->uio_offset & pmp->pm_crbomask;
761 		n = min(uio->uio_resid, pmp->pm_bpcluster - croffset);
762 		if (uio->uio_offset + n > dep->de_FileSize) {
763 			dep->de_FileSize = uio->uio_offset + n;
764 			/* The object size needs to be set before buffer is allocated */
765 			vnode_pager_setsize(vp, dep->de_FileSize);
766 		}
767 
768 		bn = de_cluster(pmp, uio->uio_offset);
769 		if ((uio->uio_offset & pmp->pm_crbomask) == 0
770 		    && (de_cluster(pmp, uio->uio_offset + uio->uio_resid)
771 			> de_cluster(pmp, uio->uio_offset)
772 			|| uio->uio_offset + uio->uio_resid >= dep->de_FileSize)) {
773 			/*
774 			 * If either the whole cluster gets written,
775 			 * or we write the cluster from its start beyond EOF,
776 			 * then no need to read data from disk.
777 			 */
778 			bp = getblk(thisvp, bn, pmp->pm_bpcluster, 0, 0, 0);
779 			vfs_bio_clrbuf(bp);
780 			/*
781 			 * Do the bmap now, since pcbmap needs buffers
782 			 * for the fat table. (see msdosfs_strategy)
783 			 */
784 			if (bp->b_blkno == bp->b_lblkno) {
785 				error = pcbmap(dep, bp->b_lblkno, &bn, 0, 0);
786 				if (error)
787 					bp->b_blkno = -1;
788 				else
789 					bp->b_blkno = bn;
790 			}
791 			if (bp->b_blkno == -1) {
792 				brelse(bp);
793 				if (!error)
794 					error = EIO;		/* XXX */
795 				break;
796 			}
797 		} else {
798 			/*
799 			 * The block we need to write into exists, so read it in.
800 			 */
801 			error = bread(thisvp, bn, pmp->pm_bpcluster, cred, &bp);
802 			if (error) {
803 				brelse(bp);
804 				break;
805 			}
806 		}
807 
808 		/*
809 		 * Should these vnode_pager_* functions be done on dir
810 		 * files?
811 		 */
812 
813 		/*
814 		 * Copy the data from user space into the buf header.
815 		 */
816 		error = uiomove(bp->b_data + croffset, n, uio);
817 		if (error) {
818 			brelse(bp);
819 			break;
820 		}
821 
822 		/* Prepare for clustered writes in some else clauses. */
823 		if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
824 			bp->b_flags |= B_CLUSTEROK;
825 
826 		/*
827 		 * If IO_SYNC, then each buffer is written synchronously.
828 		 * Otherwise, if we have a severe page deficiency then
829 		 * write the buffer asynchronously.  Otherwise, if on a
830 		 * cluster boundary then write the buffer asynchronously,
831 		 * combining it with contiguous clusters if permitted and
832 		 * possible, since we don't expect more writes into this
833 		 * buffer soon.  Otherwise, do a delayed write because we
834 		 * expect more writes into this buffer soon.
835 		 */
836 		if (ioflag & IO_SYNC)
837 			(void)bwrite(bp);
838 		else if (vm_page_count_severe() || buf_dirty_count_severe())
839 			bawrite(bp);
840 		else if (n + croffset == pmp->pm_bpcluster) {
841 			if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
842 				cluster_write(vp, bp, dep->de_FileSize,
843 				    seqcount, 0);
844 			else
845 				bawrite(bp);
846 		} else
847 			bdwrite(bp);
848 		dep->de_flag |= DE_UPDATE;
849 	} while (error == 0 && uio->uio_resid > 0);
850 
851 	/*
852 	 * If the write failed and they want us to, truncate the file back
853 	 * to the size it was before the write was attempted.
854 	 */
855 errexit:
856 	if (error) {
857 		if (ioflag & IO_UNIT) {
858 			detrunc(dep, osize, ioflag & IO_SYNC, NOCRED);
859 			uio->uio_offset -= resid - uio->uio_resid;
860 			uio->uio_resid = resid;
861 		} else {
862 			detrunc(dep, dep->de_FileSize, ioflag & IO_SYNC, NOCRED);
863 			if (uio->uio_resid != resid)
864 				error = 0;
865 		}
866 	} else if (ioflag & IO_SYNC)
867 		error = deupdat(dep, 1);
868 	return (error);
869 }
870 
871 /*
872  * Flush the blocks of a file to disk.
873  */
874 static int
875 msdosfs_fsync(ap)
876 	struct vop_fsync_args /* {
877 		struct vnode *a_vp;
878 		struct ucred *a_cred;
879 		int a_waitfor;
880 		struct thread *a_td;
881 	} */ *ap;
882 {
883 	struct vnode *devvp;
884 	int allerror, error;
885 
886 	vop_stdfsync(ap);
887 
888 	/*
889 	* If the syncing request comes from fsync(2), sync the entire
890 	* FAT and any other metadata that happens to be on devvp.  We
891 	* need this mainly for the FAT.  We write the FAT sloppily, and
892 	* syncing it all now is the best we can easily do to get all
893 	* directory entries associated with the file (not just the file)
894 	* fully synced.  The other metadata includes critical metadata
895 	* for all directory entries, but only in the MNT_ASYNC case.  We
896 	* will soon sync all metadata in the file's directory entry.
897 	* Non-critical metadata for associated directory entries only
898 	* gets synced accidentally, as in most file systems.
899 	*/
900 	if (ap->a_waitfor == MNT_WAIT) {
901 		devvp = VTODE(ap->a_vp)->de_pmp->pm_devvp;
902 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
903 		allerror = VOP_FSYNC(devvp, MNT_WAIT, ap->a_td);
904 		VOP_UNLOCK(devvp, 0);
905 	} else
906 		allerror = 0;
907 
908 	error = deupdat(VTODE(ap->a_vp), ap->a_waitfor == MNT_WAIT);
909 	if (allerror == 0)
910 		allerror = error;
911 	return (allerror);
912 }
913 
914 static int
915 msdosfs_remove(ap)
916 	struct vop_remove_args /* {
917 		struct vnode *a_dvp;
918 		struct vnode *a_vp;
919 		struct componentname *a_cnp;
920 	} */ *ap;
921 {
922 	struct denode *dep = VTODE(ap->a_vp);
923 	struct denode *ddep = VTODE(ap->a_dvp);
924 	int error;
925 
926 	if (ap->a_vp->v_type == VDIR)
927 		error = EPERM;
928 	else
929 		error = removede(ddep, dep);
930 #ifdef MSDOSFS_DEBUG
931 	printf("msdosfs_remove(), dep %p, v_usecount %d\n", dep, ap->a_vp->v_usecount);
932 #endif
933 	return (error);
934 }
935 
936 /*
937  * DOS filesystems don't know what links are.
938  */
939 static int
940 msdosfs_link(ap)
941 	struct vop_link_args /* {
942 		struct vnode *a_tdvp;
943 		struct vnode *a_vp;
944 		struct componentname *a_cnp;
945 	} */ *ap;
946 {
947 	return (EOPNOTSUPP);
948 }
949 
950 /*
951  * Renames on files require moving the denode to a new hash queue since the
952  * denode's location is used to compute which hash queue to put the file
953  * in. Unless it is a rename in place.  For example "mv a b".
954  *
955  * What follows is the basic algorithm:
956  *
957  * if (file move) {
958  *	if (dest file exists) {
959  *		remove dest file
960  *	}
961  *	if (dest and src in same directory) {
962  *		rewrite name in existing directory slot
963  *	} else {
964  *		write new entry in dest directory
965  *		update offset and dirclust in denode
966  *		move denode to new hash chain
967  *		clear old directory entry
968  *	}
969  * } else {
970  *	directory move
971  *	if (dest directory exists) {
972  *		if (dest is not empty) {
973  *			return ENOTEMPTY
974  *		}
975  *		remove dest directory
976  *	}
977  *	if (dest and src in same directory) {
978  *		rewrite name in existing entry
979  *	} else {
980  *		be sure dest is not a child of src directory
981  *		write entry in dest directory
982  *		update "." and ".." in moved directory
983  *		clear old directory entry for moved directory
984  *	}
985  * }
986  *
987  * On entry:
988  *	source's parent directory is unlocked
989  *	source file or directory is unlocked
990  *	destination's parent directory is locked
991  *	destination file or directory is locked if it exists
992  *
993  * On exit:
994  *	all denodes should be released
995  */
996 static int
997 msdosfs_rename(ap)
998 	struct vop_rename_args /* {
999 		struct vnode *a_fdvp;
1000 		struct vnode *a_fvp;
1001 		struct componentname *a_fcnp;
1002 		struct vnode *a_tdvp;
1003 		struct vnode *a_tvp;
1004 		struct componentname *a_tcnp;
1005 	} */ *ap;
1006 {
1007 	struct vnode *tdvp = ap->a_tdvp;
1008 	struct vnode *fvp = ap->a_fvp;
1009 	struct vnode *fdvp = ap->a_fdvp;
1010 	struct vnode *tvp = ap->a_tvp;
1011 	struct componentname *tcnp = ap->a_tcnp;
1012 	struct componentname *fcnp = ap->a_fcnp;
1013 	struct denode *ip, *xp, *dp, *zp;
1014 	u_char toname[12], oldname[11];
1015 	u_long from_diroffset, to_diroffset;
1016 	u_char to_count;
1017 	int doingdirectory = 0, newparent = 0;
1018 	int error;
1019 	u_long cn, pcl;
1020 	daddr_t bn;
1021 	struct denode *fddep;	/* from file's parent directory	 */
1022 	struct msdosfsmount *pmp;
1023 	struct direntry *dotdotp;
1024 	struct buf *bp;
1025 
1026 	fddep = VTODE(ap->a_fdvp);
1027 	pmp = fddep->de_pmp;
1028 
1029 	pmp = VFSTOMSDOSFS(fdvp->v_mount);
1030 
1031 #ifdef DIAGNOSTIC
1032 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1033 	    (fcnp->cn_flags & HASBUF) == 0)
1034 		panic("msdosfs_rename: no name");
1035 #endif
1036 	/*
1037 	 * Check for cross-device rename.
1038 	 */
1039 	if (fvp->v_mount != tdvp->v_mount ||
1040 	    (tvp && fvp->v_mount != tvp->v_mount)) {
1041 		error = EXDEV;
1042 abortit:
1043 		if (tdvp == tvp)
1044 			vrele(tdvp);
1045 		else
1046 			vput(tdvp);
1047 		if (tvp)
1048 			vput(tvp);
1049 		vrele(fdvp);
1050 		vrele(fvp);
1051 		return (error);
1052 	}
1053 
1054 	/*
1055 	 * If source and dest are the same, do nothing.
1056 	 */
1057 	if (tvp == fvp) {
1058 		error = 0;
1059 		goto abortit;
1060 	}
1061 
1062 	error = vn_lock(fvp, LK_EXCLUSIVE);
1063 	if (error)
1064 		goto abortit;
1065 	dp = VTODE(fdvp);
1066 	ip = VTODE(fvp);
1067 
1068 	/*
1069 	 * Be sure we are not renaming ".", "..", or an alias of ".". This
1070 	 * leads to a crippled directory tree.  It's pretty tough to do a
1071 	 * "ls" or "pwd" with the "." directory entry missing, and "cd .."
1072 	 * doesn't work if the ".." entry is missing.
1073 	 */
1074 	if (ip->de_Attributes & ATTR_DIRECTORY) {
1075 		/*
1076 		 * Avoid ".", "..", and aliases of "." for obvious reasons.
1077 		 */
1078 		if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1079 		    dp == ip ||
1080 		    (fcnp->cn_flags & ISDOTDOT) ||
1081 		    (tcnp->cn_flags & ISDOTDOT) ||
1082 		    (ip->de_flag & DE_RENAME)) {
1083 			VOP_UNLOCK(fvp, 0);
1084 			error = EINVAL;
1085 			goto abortit;
1086 		}
1087 		ip->de_flag |= DE_RENAME;
1088 		doingdirectory++;
1089 	}
1090 
1091 	/*
1092 	 * When the target exists, both the directory
1093 	 * and target vnodes are returned locked.
1094 	 */
1095 	dp = VTODE(tdvp);
1096 	xp = tvp ? VTODE(tvp) : NULL;
1097 	/*
1098 	 * Remember direntry place to use for destination
1099 	 */
1100 	to_diroffset = dp->de_fndoffset;
1101 	to_count = dp->de_fndcnt;
1102 
1103 	/*
1104 	 * If ".." must be changed (ie the directory gets a new
1105 	 * parent) then the source directory must not be in the
1106 	 * directory hierarchy above the target, as this would
1107 	 * orphan everything below the source directory. Also
1108 	 * the user must have write permission in the source so
1109 	 * as to be able to change "..". We must repeat the call
1110 	 * to namei, as the parent directory is unlocked by the
1111 	 * call to doscheckpath().
1112 	 */
1113 	error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread);
1114 	VOP_UNLOCK(fvp, 0);
1115 	if (VTODE(fdvp)->de_StartCluster != VTODE(tdvp)->de_StartCluster)
1116 		newparent = 1;
1117 	if (doingdirectory && newparent) {
1118 		if (error)	/* write access check above */
1119 			goto bad;
1120 		if (xp != NULL)
1121 			vput(tvp);
1122 		/*
1123 		 * doscheckpath() vput()'s dp,
1124 		 * so we have to do a relookup afterwards
1125 		 */
1126 		error = doscheckpath(ip, dp);
1127 		if (error)
1128 			goto out;
1129 		if ((tcnp->cn_flags & SAVESTART) == 0)
1130 			panic("msdosfs_rename: lost to startdir");
1131 		error = relookup(tdvp, &tvp, tcnp);
1132 		if (error)
1133 			goto out;
1134 		dp = VTODE(tdvp);
1135 		xp = tvp ? VTODE(tvp) : NULL;
1136 	}
1137 
1138 	if (xp != NULL) {
1139 		/*
1140 		 * Target must be empty if a directory and have no links
1141 		 * to it. Also, ensure source and target are compatible
1142 		 * (both directories, or both not directories).
1143 		 */
1144 		if (xp->de_Attributes & ATTR_DIRECTORY) {
1145 			if (!dosdirempty(xp)) {
1146 				error = ENOTEMPTY;
1147 				goto bad;
1148 			}
1149 			if (!doingdirectory) {
1150 				error = ENOTDIR;
1151 				goto bad;
1152 			}
1153 			cache_purge(tdvp);
1154 		} else if (doingdirectory) {
1155 			error = EISDIR;
1156 			goto bad;
1157 		}
1158 		error = removede(dp, xp);
1159 		if (error)
1160 			goto bad;
1161 		vput(tvp);
1162 		xp = NULL;
1163 	}
1164 
1165 	/*
1166 	 * Convert the filename in tcnp into a dos filename. We copy this
1167 	 * into the denode and directory entry for the destination
1168 	 * file/directory.
1169 	 */
1170 	error = uniqdosname(VTODE(tdvp), tcnp, toname);
1171 	if (error)
1172 		goto abortit;
1173 
1174 	/*
1175 	 * Since from wasn't locked at various places above,
1176 	 * have to do a relookup here.
1177 	 */
1178 	fcnp->cn_flags &= ~MODMASK;
1179 	fcnp->cn_flags |= LOCKPARENT | LOCKLEAF;
1180 	if ((fcnp->cn_flags & SAVESTART) == 0)
1181 		panic("msdosfs_rename: lost from startdir");
1182 	if (!newparent)
1183 		VOP_UNLOCK(tdvp, 0);
1184 	if (relookup(fdvp, &fvp, fcnp) == 0)
1185 		vrele(fdvp);
1186 	if (fvp == NULL) {
1187 		/*
1188 		 * From name has disappeared.
1189 		 */
1190 		if (doingdirectory)
1191 			panic("rename: lost dir entry");
1192 		if (newparent)
1193 			VOP_UNLOCK(tdvp, 0);
1194 		vrele(tdvp);
1195 		vrele(ap->a_fvp);
1196 		return 0;
1197 	}
1198 	xp = VTODE(fvp);
1199 	zp = VTODE(fdvp);
1200 	from_diroffset = zp->de_fndoffset;
1201 
1202 	/*
1203 	 * Ensure that the directory entry still exists and has not
1204 	 * changed till now. If the source is a file the entry may
1205 	 * have been unlinked or renamed. In either case there is
1206 	 * no further work to be done. If the source is a directory
1207 	 * then it cannot have been rmdir'ed or renamed; this is
1208 	 * prohibited by the DE_RENAME flag.
1209 	 */
1210 	if (xp != ip) {
1211 		if (doingdirectory)
1212 			panic("rename: lost dir entry");
1213 		VOP_UNLOCK(fvp, 0);
1214 		if (newparent)
1215 			VOP_UNLOCK(fdvp, 0);
1216 		vrele(ap->a_fvp);
1217 		xp = NULL;
1218 	} else {
1219 		vrele(fvp);
1220 		xp = NULL;
1221 
1222 		/*
1223 		 * First write a new entry in the destination
1224 		 * directory and mark the entry in the source directory
1225 		 * as deleted.  Then move the denode to the correct hash
1226 		 * chain for its new location in the filesystem.  And, if
1227 		 * we moved a directory, then update its .. entry to point
1228 		 * to the new parent directory.
1229 		 */
1230 		bcopy(ip->de_Name, oldname, 11);
1231 		bcopy(toname, ip->de_Name, 11);	/* update denode */
1232 		dp->de_fndoffset = to_diroffset;
1233 		dp->de_fndcnt = to_count;
1234 		error = createde(ip, dp, (struct denode **)0, tcnp);
1235 		if (error) {
1236 			bcopy(oldname, ip->de_Name, 11);
1237 			if (newparent)
1238 				VOP_UNLOCK(fdvp, 0);
1239 			VOP_UNLOCK(fvp, 0);
1240 			goto bad;
1241 		}
1242 		/*
1243 		 * If ip is for a directory, then its name should always
1244 		 * be "." since it is for the directory entry in the
1245 		 * directory itself (msdosfs_lookup() always translates
1246 		 * to the "." entry so as to get a unique denode, except
1247 		 * for the root directory there are different
1248 		 * complications).  However, we just corrupted its name
1249 		 * to pass the correct name to createde().  Undo this.
1250 		 */
1251 		if ((ip->de_Attributes & ATTR_DIRECTORY) != 0)
1252 			bcopy(oldname, ip->de_Name, 11);
1253 		ip->de_refcnt++;
1254 		zp->de_fndoffset = from_diroffset;
1255 		error = removede(zp, ip);
1256 		if (error) {
1257 			/* XXX should downgrade to ro here, fs is corrupt */
1258 			if (newparent)
1259 				VOP_UNLOCK(fdvp, 0);
1260 			VOP_UNLOCK(fvp, 0);
1261 			goto bad;
1262 		}
1263 		if (!doingdirectory) {
1264 			error = pcbmap(dp, de_cluster(pmp, to_diroffset), 0,
1265 				       &ip->de_dirclust, 0);
1266 			if (error) {
1267 				/* XXX should downgrade to ro here, fs is corrupt */
1268 				if (newparent)
1269 					VOP_UNLOCK(fdvp, 0);
1270 				VOP_UNLOCK(fvp, 0);
1271 				goto bad;
1272 			}
1273 			if (ip->de_dirclust == MSDOSFSROOT)
1274 				ip->de_diroffset = to_diroffset;
1275 			else
1276 				ip->de_diroffset = to_diroffset & pmp->pm_crbomask;
1277 		}
1278 		reinsert(ip);
1279 		if (newparent)
1280 			VOP_UNLOCK(fdvp, 0);
1281 	}
1282 
1283 	/*
1284 	 * If we moved a directory to a new parent directory, then we must
1285 	 * fixup the ".." entry in the moved directory.
1286 	 */
1287 	if (doingdirectory && newparent) {
1288 		cn = ip->de_StartCluster;
1289 		if (cn == MSDOSFSROOT) {
1290 			/* this should never happen */
1291 			panic("msdosfs_rename(): updating .. in root directory?");
1292 		} else
1293 			bn = cntobn(pmp, cn);
1294 		error = bread(pmp->pm_devvp, bn, pmp->pm_bpcluster,
1295 			      NOCRED, &bp);
1296 		if (error) {
1297 			/* XXX should downgrade to ro here, fs is corrupt */
1298 			brelse(bp);
1299 			VOP_UNLOCK(fvp, 0);
1300 			goto bad;
1301 		}
1302 		dotdotp = (struct direntry *)bp->b_data + 1;
1303 		pcl = dp->de_StartCluster;
1304 		if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1305 			pcl = MSDOSFSROOT;
1306 		putushort(dotdotp->deStartCluster, pcl);
1307 		if (FAT32(pmp))
1308 			putushort(dotdotp->deHighClust, pcl >> 16);
1309 		if (DOINGASYNC(fvp))
1310 			bdwrite(bp);
1311 		else if ((error = bwrite(bp)) != 0) {
1312 			/* XXX should downgrade to ro here, fs is corrupt */
1313 			VOP_UNLOCK(fvp, 0);
1314 			goto bad;
1315 		}
1316 	}
1317 
1318 	/*
1319 	 * The msdosfs lookup is case insensitive. Several aliases may
1320 	 * be inserted for a single directory entry. As a consequnce,
1321 	 * name cache purge done by lookup for fvp when DELETE op for
1322 	 * namei is specified, might be not enough to expunge all
1323 	 * namecache entries that were installed for this direntry.
1324 	 */
1325 	cache_purge(fvp);
1326 	VOP_UNLOCK(fvp, 0);
1327 bad:
1328 	if (xp)
1329 		vput(tvp);
1330 	vput(tdvp);
1331 out:
1332 	ip->de_flag &= ~DE_RENAME;
1333 	vrele(fdvp);
1334 	vrele(fvp);
1335 	return (error);
1336 
1337 }
1338 
1339 static struct {
1340 	struct direntry dot;
1341 	struct direntry dotdot;
1342 } dosdirtemplate = {
1343 	{	".          ",				/* the . entry */
1344 		ATTR_DIRECTORY,				/* file attribute */
1345 		0,					/* reserved */
1346 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1347 		{ 0, 0 },				/* access date */
1348 		{ 0, 0 },				/* high bits of start cluster */
1349 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1350 		{ 0, 0 },				/* startcluster */
1351 		{ 0, 0, 0, 0 }				/* filesize */
1352 	},
1353 	{	"..         ",				/* the .. entry */
1354 		ATTR_DIRECTORY,				/* file attribute */
1355 		0,					/* reserved */
1356 		0, { 0, 0 }, { 0, 0 },			/* create time & date */
1357 		{ 0, 0 },				/* access date */
1358 		{ 0, 0 },				/* high bits of start cluster */
1359 		{ 210, 4 }, { 210, 4 },			/* modify time & date */
1360 		{ 0, 0 },				/* startcluster */
1361 		{ 0, 0, 0, 0 }				/* filesize */
1362 	}
1363 };
1364 
1365 static int
1366 msdosfs_mkdir(ap)
1367 	struct vop_mkdir_args /* {
1368 		struct vnode *a_dvp;
1369 		struct vnode **a_vpp;
1370 		struvt componentname *a_cnp;
1371 		struct vattr *a_vap;
1372 	} */ *ap;
1373 {
1374 	struct componentname *cnp = ap->a_cnp;
1375 	struct denode *dep;
1376 	struct denode *pdep = VTODE(ap->a_dvp);
1377 	struct direntry *denp;
1378 	struct msdosfsmount *pmp = pdep->de_pmp;
1379 	struct buf *bp;
1380 	u_long newcluster, pcl;
1381 	int bn;
1382 	int error;
1383 	struct denode ndirent;
1384 	struct timespec ts;
1385 
1386 	/*
1387 	 * If this is the root directory and there is no space left we
1388 	 * can't do anything.  This is because the root directory can not
1389 	 * change size.
1390 	 */
1391 	if (pdep->de_StartCluster == MSDOSFSROOT
1392 	    && pdep->de_fndoffset >= pdep->de_FileSize) {
1393 		error = ENOSPC;
1394 		goto bad2;
1395 	}
1396 
1397 	/*
1398 	 * Allocate a cluster to hold the about to be created directory.
1399 	 */
1400 	error = clusteralloc(pmp, 0, 1, CLUST_EOFE, &newcluster, NULL);
1401 	if (error)
1402 		goto bad2;
1403 
1404 	bzero(&ndirent, sizeof(ndirent));
1405 	ndirent.de_pmp = pmp;
1406 	ndirent.de_flag = DE_ACCESS | DE_CREATE | DE_UPDATE;
1407 	getnanotime(&ts);
1408 	DETIMES(&ndirent, &ts, &ts, &ts);
1409 
1410 	/*
1411 	 * Now fill the cluster with the "." and ".." entries. And write
1412 	 * the cluster to disk.  This way it is there for the parent
1413 	 * directory to be pointing at if there were a crash.
1414 	 */
1415 	bn = cntobn(pmp, newcluster);
1416 	/* always succeeds */
1417 	bp = getblk(pmp->pm_devvp, bn, pmp->pm_bpcluster, 0, 0, 0);
1418 	bzero(bp->b_data, pmp->pm_bpcluster);
1419 	bcopy(&dosdirtemplate, bp->b_data, sizeof dosdirtemplate);
1420 	denp = (struct direntry *)bp->b_data;
1421 	putushort(denp[0].deStartCluster, newcluster);
1422 	putushort(denp[0].deCDate, ndirent.de_CDate);
1423 	putushort(denp[0].deCTime, ndirent.de_CTime);
1424 	denp[0].deCHundredth = ndirent.de_CHun;
1425 	putushort(denp[0].deADate, ndirent.de_ADate);
1426 	putushort(denp[0].deMDate, ndirent.de_MDate);
1427 	putushort(denp[0].deMTime, ndirent.de_MTime);
1428 	pcl = pdep->de_StartCluster;
1429 	/*
1430 	 * Although the root directory has a non-magic starting cluster
1431 	 * number for FAT32, chkdsk and fsck_msdosfs still require
1432 	 * references to it in dotdot entries to be magic.
1433 	 */
1434 	if (FAT32(pmp) && pcl == pmp->pm_rootdirblk)
1435 		pcl = MSDOSFSROOT;
1436 	putushort(denp[1].deStartCluster, pcl);
1437 	putushort(denp[1].deCDate, ndirent.de_CDate);
1438 	putushort(denp[1].deCTime, ndirent.de_CTime);
1439 	denp[1].deCHundredth = ndirent.de_CHun;
1440 	putushort(denp[1].deADate, ndirent.de_ADate);
1441 	putushort(denp[1].deMDate, ndirent.de_MDate);
1442 	putushort(denp[1].deMTime, ndirent.de_MTime);
1443 	if (FAT32(pmp)) {
1444 		putushort(denp[0].deHighClust, newcluster >> 16);
1445 		putushort(denp[1].deHighClust, pcl >> 16);
1446 	}
1447 
1448 	if (DOINGASYNC(ap->a_dvp))
1449 		bdwrite(bp);
1450 	else if ((error = bwrite(bp)) != 0)
1451 		goto bad;
1452 
1453 	/*
1454 	 * Now build up a directory entry pointing to the newly allocated
1455 	 * cluster.  This will be written to an empty slot in the parent
1456 	 * directory.
1457 	 */
1458 #ifdef DIAGNOSTIC
1459 	if ((cnp->cn_flags & HASBUF) == 0)
1460 		panic("msdosfs_mkdir: no name");
1461 #endif
1462 	error = uniqdosname(pdep, cnp, ndirent.de_Name);
1463 	if (error)
1464 		goto bad;
1465 
1466 	ndirent.de_Attributes = ATTR_DIRECTORY;
1467 	ndirent.de_LowerCase = 0;
1468 	ndirent.de_StartCluster = newcluster;
1469 	ndirent.de_FileSize = 0;
1470 	error = createde(&ndirent, pdep, &dep, cnp);
1471 	if (error)
1472 		goto bad;
1473 	*ap->a_vpp = DETOV(dep);
1474 	return (0);
1475 
1476 bad:
1477 	clusterfree(pmp, newcluster, NULL);
1478 bad2:
1479 	return (error);
1480 }
1481 
1482 static int
1483 msdosfs_rmdir(ap)
1484 	struct vop_rmdir_args /* {
1485 		struct vnode *a_dvp;
1486 		struct vnode *a_vp;
1487 		struct componentname *a_cnp;
1488 	} */ *ap;
1489 {
1490 	struct vnode *vp = ap->a_vp;
1491 	struct vnode *dvp = ap->a_dvp;
1492 	struct componentname *cnp = ap->a_cnp;
1493 	struct denode *ip, *dp;
1494 	int error;
1495 
1496 	ip = VTODE(vp);
1497 	dp = VTODE(dvp);
1498 
1499 	/*
1500 	 * Verify the directory is empty (and valid).
1501 	 * (Rmdir ".." won't be valid since
1502 	 *  ".." will contain a reference to
1503 	 *  the current directory and thus be
1504 	 *  non-empty.)
1505 	 */
1506 	error = 0;
1507 	if (!dosdirempty(ip) || ip->de_flag & DE_RENAME) {
1508 		error = ENOTEMPTY;
1509 		goto out;
1510 	}
1511 	/*
1512 	 * Delete the entry from the directory.  For dos filesystems this
1513 	 * gets rid of the directory entry on disk, the in memory copy
1514 	 * still exists but the de_refcnt is <= 0.  This prevents it from
1515 	 * being found by deget().  When the vput() on dep is done we give
1516 	 * up access and eventually msdosfs_reclaim() will be called which
1517 	 * will remove it from the denode cache.
1518 	 */
1519 	error = removede(dp, ip);
1520 	if (error)
1521 		goto out;
1522 	/*
1523 	 * This is where we decrement the link count in the parent
1524 	 * directory.  Since dos filesystems don't do this we just purge
1525 	 * the name cache.
1526 	 */
1527 	cache_purge(dvp);
1528 	/*
1529 	 * Truncate the directory that is being deleted.
1530 	 */
1531 	error = detrunc(ip, (u_long)0, IO_SYNC, cnp->cn_cred);
1532 	cache_purge(vp);
1533 
1534 out:
1535 	return (error);
1536 }
1537 
1538 /*
1539  * DOS filesystems don't know what symlinks are.
1540  */
1541 static int
1542 msdosfs_symlink(ap)
1543 	struct vop_symlink_args /* {
1544 		struct vnode *a_dvp;
1545 		struct vnode **a_vpp;
1546 		struct componentname *a_cnp;
1547 		struct vattr *a_vap;
1548 		char *a_target;
1549 	} */ *ap;
1550 {
1551 	return (EOPNOTSUPP);
1552 }
1553 
1554 static int
1555 msdosfs_readdir(ap)
1556 	struct vop_readdir_args /* {
1557 		struct vnode *a_vp;
1558 		struct uio *a_uio;
1559 		struct ucred *a_cred;
1560 		int *a_eofflag;
1561 		int *a_ncookies;
1562 		u_long **a_cookies;
1563 	} */ *ap;
1564 {
1565 	struct mbnambuf nb;
1566 	int error = 0;
1567 	int diff;
1568 	long n;
1569 	int blsize;
1570 	long on;
1571 	u_long cn;
1572 	uint64_t fileno;
1573 	u_long dirsperblk;
1574 	long bias = 0;
1575 	daddr_t bn, lbn;
1576 	struct buf *bp;
1577 	struct denode *dep = VTODE(ap->a_vp);
1578 	struct msdosfsmount *pmp = dep->de_pmp;
1579 	struct direntry *dentp;
1580 	struct dirent dirbuf;
1581 	struct uio *uio = ap->a_uio;
1582 	u_long *cookies = NULL;
1583 	int ncookies = 0;
1584 	off_t offset, off;
1585 	int chksum = -1;
1586 
1587 #ifdef MSDOSFS_DEBUG
1588 	printf("msdosfs_readdir(): vp %p, uio %p, cred %p, eofflagp %p\n",
1589 	    ap->a_vp, uio, ap->a_cred, ap->a_eofflag);
1590 #endif
1591 
1592 	/*
1593 	 * msdosfs_readdir() won't operate properly on regular files since
1594 	 * it does i/o only with the filesystem vnode, and hence can
1595 	 * retrieve the wrong block from the buffer cache for a plain file.
1596 	 * So, fail attempts to readdir() on a plain file.
1597 	 */
1598 	if ((dep->de_Attributes & ATTR_DIRECTORY) == 0)
1599 		return (ENOTDIR);
1600 
1601 	/*
1602 	 * To be safe, initialize dirbuf
1603 	 */
1604 	bzero(dirbuf.d_name, sizeof(dirbuf.d_name));
1605 
1606 	/*
1607 	 * If the user buffer is smaller than the size of one dos directory
1608 	 * entry or the file offset is not a multiple of the size of a
1609 	 * directory entry, then we fail the read.
1610 	 */
1611 	off = offset = uio->uio_offset;
1612 	if (uio->uio_resid < sizeof(struct direntry) ||
1613 	    (offset & (sizeof(struct direntry) - 1)))
1614 		return (EINVAL);
1615 
1616 	if (ap->a_ncookies) {
1617 		ncookies = uio->uio_resid / 16;
1618 		cookies = malloc(ncookies * sizeof(u_long), M_TEMP,
1619 		       M_WAITOK);
1620 		*ap->a_cookies = cookies;
1621 		*ap->a_ncookies = ncookies;
1622 	}
1623 
1624 	dirsperblk = pmp->pm_BytesPerSec / sizeof(struct direntry);
1625 
1626 	/*
1627 	 * If they are reading from the root directory then, we simulate
1628 	 * the . and .. entries since these don't exist in the root
1629 	 * directory.  We also set the offset bias to make up for having to
1630 	 * simulate these entries. By this I mean that at file offset 64 we
1631 	 * read the first entry in the root directory that lives on disk.
1632 	 */
1633 	if (dep->de_StartCluster == MSDOSFSROOT
1634 	    || (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)) {
1635 #if 0
1636 		printf("msdosfs_readdir(): going after . or .. in root dir, offset %d\n",
1637 		    offset);
1638 #endif
1639 		bias = 2 * sizeof(struct direntry);
1640 		if (offset < bias) {
1641 			for (n = (int)offset / sizeof(struct direntry);
1642 			     n < 2; n++) {
1643 				if (FAT32(pmp))
1644 					fileno = (uint64_t)cntobn(pmp,
1645 								 pmp->pm_rootdirblk)
1646 							  * dirsperblk;
1647 				else
1648 					fileno = 1;
1649 				if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1650 					dirbuf.d_fileno =
1651 					    msdosfs_fileno_map(pmp->pm_mountp,
1652 					    fileno);
1653 				} else {
1654 
1655 					dirbuf.d_fileno = (uint32_t)fileno;
1656 				}
1657 				dirbuf.d_type = DT_DIR;
1658 				switch (n) {
1659 				case 0:
1660 					dirbuf.d_namlen = 1;
1661 					strcpy(dirbuf.d_name, ".");
1662 					break;
1663 				case 1:
1664 					dirbuf.d_namlen = 2;
1665 					strcpy(dirbuf.d_name, "..");
1666 					break;
1667 				}
1668 				dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1669 				if (uio->uio_resid < dirbuf.d_reclen)
1670 					goto out;
1671 				error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1672 				if (error)
1673 					goto out;
1674 				offset += sizeof(struct direntry);
1675 				off = offset;
1676 				if (cookies) {
1677 					*cookies++ = offset;
1678 					if (--ncookies <= 0)
1679 						goto out;
1680 				}
1681 			}
1682 		}
1683 	}
1684 
1685 	mbnambuf_init(&nb);
1686 	off = offset;
1687 	while (uio->uio_resid > 0) {
1688 		lbn = de_cluster(pmp, offset - bias);
1689 		on = (offset - bias) & pmp->pm_crbomask;
1690 		n = min(pmp->pm_bpcluster - on, uio->uio_resid);
1691 		diff = dep->de_FileSize - (offset - bias);
1692 		if (diff <= 0)
1693 			break;
1694 		n = min(n, diff);
1695 		error = pcbmap(dep, lbn, &bn, &cn, &blsize);
1696 		if (error)
1697 			break;
1698 		error = bread(pmp->pm_devvp, bn, blsize, NOCRED, &bp);
1699 		if (error) {
1700 			brelse(bp);
1701 			return (error);
1702 		}
1703 		n = min(n, blsize - bp->b_resid);
1704 		if (n == 0) {
1705 			brelse(bp);
1706 			return (EIO);
1707 		}
1708 
1709 		/*
1710 		 * Convert from dos directory entries to fs-independent
1711 		 * directory entries.
1712 		 */
1713 		for (dentp = (struct direntry *)(bp->b_data + on);
1714 		     (char *)dentp < bp->b_data + on + n;
1715 		     dentp++, offset += sizeof(struct direntry)) {
1716 #if 0
1717 			printf("rd: dentp %08x prev %08x crnt %08x deName %02x attr %02x\n",
1718 			    dentp, prev, crnt, dentp->deName[0], dentp->deAttributes);
1719 #endif
1720 			/*
1721 			 * If this is an unused entry, we can stop.
1722 			 */
1723 			if (dentp->deName[0] == SLOT_EMPTY) {
1724 				brelse(bp);
1725 				goto out;
1726 			}
1727 			/*
1728 			 * Skip deleted entries.
1729 			 */
1730 			if (dentp->deName[0] == SLOT_DELETED) {
1731 				chksum = -1;
1732 				mbnambuf_init(&nb);
1733 				continue;
1734 			}
1735 
1736 			/*
1737 			 * Handle Win95 long directory entries
1738 			 */
1739 			if (dentp->deAttributes == ATTR_WIN95) {
1740 				if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
1741 					continue;
1742 				chksum = win2unixfn(&nb,
1743 				    (struct winentry *)dentp, chksum, pmp);
1744 				continue;
1745 			}
1746 
1747 			/*
1748 			 * Skip volume labels
1749 			 */
1750 			if (dentp->deAttributes & ATTR_VOLUME) {
1751 				chksum = -1;
1752 				mbnambuf_init(&nb);
1753 				continue;
1754 			}
1755 			/*
1756 			 * This computation of d_fileno must match
1757 			 * the computation of va_fileid in
1758 			 * msdosfs_getattr.
1759 			 */
1760 			if (dentp->deAttributes & ATTR_DIRECTORY) {
1761 				fileno = getushort(dentp->deStartCluster);
1762 				if (FAT32(pmp))
1763 					fileno |= getushort(dentp->deHighClust) << 16;
1764 				/* if this is the root directory */
1765 				if (fileno == MSDOSFSROOT)
1766 					if (FAT32(pmp))
1767 						fileno = (uint64_t)cntobn(pmp,
1768 								pmp->pm_rootdirblk)
1769 							 * dirsperblk;
1770 					else
1771 						fileno = 1;
1772 				else
1773 					fileno = (uint64_t)cntobn(pmp, fileno) *
1774 					    dirsperblk;
1775 				dirbuf.d_type = DT_DIR;
1776 			} else {
1777 				fileno = (uoff_t)offset /
1778 				    sizeof(struct direntry);
1779 				dirbuf.d_type = DT_REG;
1780 			}
1781 			if (pmp->pm_flags & MSDOSFS_LARGEFS) {
1782 				dirbuf.d_fileno =
1783 				    msdosfs_fileno_map(pmp->pm_mountp, fileno);
1784 			} else
1785 				dirbuf.d_fileno = (uint32_t)fileno;
1786 
1787 			if (chksum != winChksum(dentp->deName)) {
1788 				dirbuf.d_namlen = dos2unixfn(dentp->deName,
1789 				    (u_char *)dirbuf.d_name,
1790 				    dentp->deLowerCase |
1791 					((pmp->pm_flags & MSDOSFSMNT_SHORTNAME) ?
1792 					(LCASE_BASE | LCASE_EXT) : 0),
1793 				    pmp);
1794 				mbnambuf_init(&nb);
1795 			} else
1796 				mbnambuf_flush(&nb, &dirbuf);
1797 			chksum = -1;
1798 			dirbuf.d_reclen = GENERIC_DIRSIZ(&dirbuf);
1799 			if (uio->uio_resid < dirbuf.d_reclen) {
1800 				brelse(bp);
1801 				goto out;
1802 			}
1803 			error = uiomove(&dirbuf, dirbuf.d_reclen, uio);
1804 			if (error) {
1805 				brelse(bp);
1806 				goto out;
1807 			}
1808 			if (cookies) {
1809 				*cookies++ = offset + sizeof(struct direntry);
1810 				if (--ncookies <= 0) {
1811 					brelse(bp);
1812 					goto out;
1813 				}
1814 			}
1815 			off = offset + sizeof(struct direntry);
1816 		}
1817 		brelse(bp);
1818 	}
1819 out:
1820 	/* Subtract unused cookies */
1821 	if (ap->a_ncookies)
1822 		*ap->a_ncookies -= ncookies;
1823 
1824 	uio->uio_offset = off;
1825 
1826 	/*
1827 	 * Set the eofflag (NFS uses it)
1828 	 */
1829 	if (ap->a_eofflag) {
1830 		if (dep->de_FileSize - (offset - bias) <= 0)
1831 			*ap->a_eofflag = 1;
1832 		else
1833 			*ap->a_eofflag = 0;
1834 	}
1835 	return (error);
1836 }
1837 
1838 /*-
1839  * a_vp   - pointer to the file's vnode
1840  * a_bn   - logical block number within the file (cluster number for us)
1841  * a_bop  - where to return the bufobj of the special file containing the fs
1842  * a_bnp  - where to return the "physical" block number corresponding to a_bn
1843  *          (relative to the special file; units are blocks of size DEV_BSIZE)
1844  * a_runp - where to return the "run past" a_bn.  This is the count of logical
1845  *          blocks whose physical blocks (together with a_bn's physical block)
1846  *          are contiguous.
1847  * a_runb - where to return the "run before" a_bn.
1848  */
1849 static int
1850 msdosfs_bmap(ap)
1851 	struct vop_bmap_args /* {
1852 		struct vnode *a_vp;
1853 		daddr_t a_bn;
1854 		struct bufobj **a_bop;
1855 		daddr_t *a_bnp;
1856 		int *a_runp;
1857 		int *a_runb;
1858 	} */ *ap;
1859 {
1860 	struct denode *dep;
1861 	struct mount *mp;
1862 	struct msdosfsmount *pmp;
1863 	struct vnode *vp;
1864 	daddr_t runbn;
1865 	u_long cn;
1866 	int bnpercn, error, maxio, maxrun, run;
1867 
1868 	vp = ap->a_vp;
1869 	dep = VTODE(vp);
1870 	pmp = dep->de_pmp;
1871 	if (ap->a_bop != NULL)
1872 		*ap->a_bop = &pmp->pm_devvp->v_bufobj;
1873 	if (ap->a_bnp == NULL)
1874 		return (0);
1875 	if (ap->a_runp != NULL)
1876 		*ap->a_runp = 0;
1877 	if (ap->a_runb != NULL)
1878 		*ap->a_runb = 0;
1879 	cn = ap->a_bn;
1880 	if (cn != ap->a_bn)
1881 		return (EFBIG);
1882 	error = pcbmap(dep, cn, ap->a_bnp, NULL, NULL);
1883 	if (error != 0 || (ap->a_runp == NULL && ap->a_runb == NULL))
1884 		return (error);
1885 
1886 	mp = vp->v_mount;
1887 	maxio = mp->mnt_iosize_max / mp->mnt_stat.f_iosize;
1888 	bnpercn = de_cn2bn(pmp, 1);
1889 	if (ap->a_runp != NULL) {
1890 		maxrun = ulmin(maxio - 1, pmp->pm_maxcluster - cn);
1891 		for (run = 1; run <= maxrun; run++) {
1892 			if (pcbmap(dep, cn + run, &runbn, NULL, NULL) != 0 ||
1893 			    runbn != *ap->a_bnp + run * bnpercn)
1894 				break;
1895 		}
1896 		*ap->a_runp = run - 1;
1897 	}
1898 	if (ap->a_runb != NULL) {
1899 		maxrun = ulmin(maxio - 1, cn);
1900 		for (run = 1; run < maxrun; run++) {
1901 			if (pcbmap(dep, cn - run, &runbn, NULL, NULL) != 0 ||
1902 			    runbn != *ap->a_bnp - run * bnpercn)
1903 				break;
1904 		}
1905 		*ap->a_runb = run - 1;
1906 	}
1907 	return (0);
1908 }
1909 
1910 static int
1911 msdosfs_strategy(ap)
1912 	struct vop_strategy_args /* {
1913 		struct vnode *a_vp;
1914 		struct buf *a_bp;
1915 	} */ *ap;
1916 {
1917 	struct buf *bp = ap->a_bp;
1918 	struct denode *dep = VTODE(ap->a_vp);
1919 	struct bufobj *bo;
1920 	int error = 0;
1921 	daddr_t blkno;
1922 
1923 	/*
1924 	 * If we don't already know the filesystem relative block number
1925 	 * then get it using pcbmap().  If pcbmap() returns the block
1926 	 * number as -1 then we've got a hole in the file.  DOS filesystems
1927 	 * don't allow files with holes, so we shouldn't ever see this.
1928 	 */
1929 	if (bp->b_blkno == bp->b_lblkno) {
1930 		error = pcbmap(dep, bp->b_lblkno, &blkno, 0, 0);
1931 		bp->b_blkno = blkno;
1932 		if (error) {
1933 			bp->b_error = error;
1934 			bp->b_ioflags |= BIO_ERROR;
1935 			bufdone(bp);
1936 			return (0);
1937 		}
1938 		if ((long)bp->b_blkno == -1)
1939 			vfs_bio_clrbuf(bp);
1940 	}
1941 	if (bp->b_blkno == -1) {
1942 		bufdone(bp);
1943 		return (0);
1944 	}
1945 	/*
1946 	 * Read/write the block from/to the disk that contains the desired
1947 	 * file block.
1948 	 */
1949 	bp->b_iooffset = dbtob(bp->b_blkno);
1950 	bo = dep->de_pmp->pm_bo;
1951 	BO_STRATEGY(bo, bp);
1952 	return (0);
1953 }
1954 
1955 static int
1956 msdosfs_print(ap)
1957 	struct vop_print_args /* {
1958 		struct vnode *vp;
1959 	} */ *ap;
1960 {
1961 	struct denode *dep = VTODE(ap->a_vp);
1962 
1963 	printf("\tstartcluster %lu, dircluster %lu, diroffset %lu, ",
1964 	       dep->de_StartCluster, dep->de_dirclust, dep->de_diroffset);
1965 	printf("on dev %s\n", devtoname(dep->de_pmp->pm_dev));
1966 	return (0);
1967 }
1968 
1969 static int
1970 msdosfs_pathconf(ap)
1971 	struct vop_pathconf_args /* {
1972 		struct vnode *a_vp;
1973 		int a_name;
1974 		int *a_retval;
1975 	} */ *ap;
1976 {
1977 	struct msdosfsmount *pmp = VTODE(ap->a_vp)->de_pmp;
1978 
1979 	switch (ap->a_name) {
1980 	case _PC_LINK_MAX:
1981 		*ap->a_retval = 1;
1982 		return (0);
1983 	case _PC_NAME_MAX:
1984 		*ap->a_retval = pmp->pm_flags & MSDOSFSMNT_LONGNAME ? WIN_MAXLEN : 12;
1985 		return (0);
1986 	case _PC_PATH_MAX:
1987 		*ap->a_retval = PATH_MAX;
1988 		return (0);
1989 	case _PC_CHOWN_RESTRICTED:
1990 		*ap->a_retval = 1;
1991 		return (0);
1992 	case _PC_NO_TRUNC:
1993 		*ap->a_retval = 0;
1994 		return (0);
1995 	default:
1996 		return (EINVAL);
1997 	}
1998 	/* NOTREACHED */
1999 }
2000 
2001 static int
2002 msdosfs_vptofh(ap)
2003 	struct vop_vptofh_args /* {
2004 		struct vnode *a_vp;
2005 		struct fid *a_fhp;
2006 	} */ *ap;
2007 {
2008 	struct denode *dep;
2009 	struct defid *defhp;
2010 
2011 	dep = VTODE(ap->a_vp);
2012 	defhp = (struct defid *)ap->a_fhp;
2013 	defhp->defid_len = sizeof(struct defid);
2014 	defhp->defid_dirclust = dep->de_dirclust;
2015 	defhp->defid_dirofs = dep->de_diroffset;
2016 	/* defhp->defid_gen = dep->de_gen; */
2017 	return (0);
2018 }
2019 
2020 /* Global vfs data structures for msdosfs */
2021 struct vop_vector msdosfs_vnodeops = {
2022 	.vop_default =		&default_vnodeops,
2023 
2024 	.vop_access =		msdosfs_access,
2025 	.vop_bmap =		msdosfs_bmap,
2026 	.vop_cachedlookup =	msdosfs_lookup,
2027 	.vop_open =		msdosfs_open,
2028 	.vop_close =		msdosfs_close,
2029 	.vop_create =		msdosfs_create,
2030 	.vop_fsync =		msdosfs_fsync,
2031 	.vop_getattr =		msdosfs_getattr,
2032 	.vop_inactive =		msdosfs_inactive,
2033 	.vop_link =		msdosfs_link,
2034 	.vop_lookup =		vfs_cache_lookup,
2035 	.vop_mkdir =		msdosfs_mkdir,
2036 	.vop_mknod =		msdosfs_mknod,
2037 	.vop_pathconf =		msdosfs_pathconf,
2038 	.vop_print =		msdosfs_print,
2039 	.vop_read =		msdosfs_read,
2040 	.vop_readdir =		msdosfs_readdir,
2041 	.vop_reclaim =		msdosfs_reclaim,
2042 	.vop_remove =		msdosfs_remove,
2043 	.vop_rename =		msdosfs_rename,
2044 	.vop_rmdir =		msdosfs_rmdir,
2045 	.vop_setattr =		msdosfs_setattr,
2046 	.vop_strategy =		msdosfs_strategy,
2047 	.vop_symlink =		msdosfs_symlink,
2048 	.vop_write =		msdosfs_write,
2049 	.vop_vptofh =		msdosfs_vptofh,
2050 };
2051