xref: /netbsd-src/sys/fs/msdosfs/msdosfs_lookup.c (revision 274254cdae52594c1aa480a736aef78313d15c9c)
1 /*	$NetBSD: msdosfs_lookup.c,v 1.19 2009/03/14 21:04:23 dsl Exp $	*/
2 
3 /*-
4  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
5  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
6  * All rights reserved.
7  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by TooLs GmbH.
20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
21  *    derived from this software without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33  */
34 /*
35  * Written by Paul Popelka (paulp@uts.amdahl.com)
36  *
37  * You can do anything you want with this software, just don't say you wrote
38  * it, and don't remove this notice.
39  *
40  * This software is provided "as is".
41  *
42  * The author supplies this software to be publicly redistributed on the
43  * understanding that the author is not responsible for the correct
44  * functioning of this software in any circumstances and is not liable for
45  * any damages caused by this software.
46  *
47  * October 1992
48  */
49 
50 #include <sys/cdefs.h>
51 __KERNEL_RCSID(0, "$NetBSD: msdosfs_lookup.c,v 1.19 2009/03/14 21:04:23 dsl Exp $");
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/namei.h>
56 #include <sys/buf.h>
57 #include <sys/vnode.h>
58 #include <sys/mount.h>
59 #include <sys/dirent.h>
60 #include <sys/kauth.h>
61 
62 #include <fs/msdosfs/bpb.h>
63 #include <fs/msdosfs/direntry.h>
64 #include <fs/msdosfs/denode.h>
65 #include <fs/msdosfs/msdosfsmount.h>
66 #include <fs/msdosfs/fat.h>
67 
68 /*
69  * When we search a directory the blocks containing directory entries are
70  * read and examined.  The directory entries contain information that would
71  * normally be in the inode of a unix filesystem.  This means that some of
72  * a directory's contents may also be in memory resident denodes (sort of
73  * an inode).  This can cause problems if we are searching while some other
74  * process is modifying a directory.  To prevent one process from accessing
75  * incompletely modified directory information we depend upon being the
76  * sole owner of a directory block.  bread/brelse provide this service.
77  * This being the case, when a process modifies a directory it must first
78  * acquire the disk block that contains the directory entry to be modified.
79  * Then update the disk block and the denode, and then write the disk block
80  * out to disk.  This way disk blocks containing directory entries and in
81  * memory denode's will be in synch.
82  */
83 int
84 msdosfs_lookup(void *v)
85 {
86 	struct vop_lookup_args /* {
87 		struct vnode *a_dvp;
88 		struct vnode **a_vpp;
89 		struct componentname *a_cnp;
90 	} */ *ap = v;
91 	struct vnode *vdp = ap->a_dvp;
92 	struct vnode **vpp = ap->a_vpp;
93 	struct componentname *cnp = ap->a_cnp;
94 	daddr_t bn;
95 	int error;
96 	int slotcount;
97 	int slotoffset = 0;
98 	int frcn;
99 	u_long cluster;
100 	int blkoff;
101 	int diroff;
102 	int blsize;
103 	int isadir;		/* ~0 if found direntry is a directory	 */
104 	u_long scn;		/* starting cluster number		 */
105 	struct vnode *pdp;
106 	struct denode *dp;
107 	struct denode *tdp;
108 	struct msdosfsmount *pmp;
109 	struct buf *bp = 0;
110 	struct direntry *dep;
111 	u_char dosfilename[12];
112 	int flags;
113 	int nameiop = cnp->cn_nameiop;
114 	int wincnt = 1;
115 	int chksum = -1, chksum_ok;
116 	int olddos = 1;
117 
118 	flags = cnp->cn_flags;
119 
120 #ifdef MSDOSFS_DEBUG
121 	printf("msdosfs_lookup(): looking for %.*s\n",
122 		(int)cnp->cn_namelen, cnp->cn_nameptr);
123 #endif
124 	dp = VTODE(vdp);
125 	pmp = dp->de_pmp;
126 	*vpp = NULL;
127 #ifdef MSDOSFS_DEBUG
128 	printf("msdosfs_lookup(): vdp %p, dp %p, Attr %02x\n",
129 	    vdp, dp, dp->de_Attributes);
130 #endif
131 
132 	/*
133 	 * Check accessiblity of directory.
134 	 */
135 	if ((error = VOP_ACCESS(vdp, VEXEC, cnp->cn_cred)) != 0)
136 		return (error);
137 
138 	if ((flags & ISLASTCN) && (vdp->v_mount->mnt_flag & MNT_RDONLY) &&
139 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
140 		return (EROFS);
141 
142 	/*
143 	 * We now have a segment name to search for, and a directory to search.
144 	 *
145 	 * Before tediously performing a linear scan of the directory,
146 	 * check the name cache to see if the directory/name pair
147 	 * we are looking for is known already.
148 	 */
149 	if ((error = cache_lookup(vdp, vpp, cnp)) >= 0)
150 		return (error);
151 
152 	/*
153 	 * If they are going after the . or .. entry in the root directory,
154 	 * they won't find it.  DOS filesystems don't have them in the root
155 	 * directory.  So, we fake it. deget() is in on this scam too.
156 	 */
157 	if ((vdp->v_vflag & VV_ROOT) && cnp->cn_nameptr[0] == '.' &&
158 	    (cnp->cn_namelen == 1 ||
159 		(cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.'))) {
160 		isadir = ATTR_DIRECTORY;
161 		scn = MSDOSFSROOT;
162 #ifdef MSDOSFS_DEBUG
163 		printf("msdosfs_lookup(): looking for . or .. in root directory\n");
164 #endif
165 		cluster = MSDOSFSROOT;
166 		blkoff = MSDOSFSROOT_OFS;
167 		goto foundroot;
168 	}
169 
170 	switch (unix2dosfn((const u_char *)cnp->cn_nameptr, dosfilename,
171 	    cnp->cn_namelen, 0)) {
172 	case 0:
173 		return (EINVAL);
174 	case 1:
175 		break;
176 	case 2:
177 		wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr,
178 		    cnp->cn_namelen) + 1;
179 		break;
180 	case 3:
181 		olddos = 0;
182 		wincnt = winSlotCnt((const u_char *)cnp->cn_nameptr,
183 		    cnp->cn_namelen) + 1;
184 		break;
185 	}
186 	if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
187 		wincnt = 1;
188 
189 	/*
190 	 * Suppress search for slots unless creating
191 	 * file and at end of pathname, in which case
192 	 * we watch for a place to put the new file in
193 	 * case it doesn't already exist.
194 	 */
195 	slotcount = wincnt;
196 	if ((nameiop == CREATE || nameiop == RENAME) &&
197 	    (flags & ISLASTCN))
198 		slotcount = 0;
199 
200 #ifdef MSDOSFS_DEBUG
201 	printf("msdosfs_lookup(): dos filename: %s\n", dosfilename);
202 #endif
203 	/*
204 	 * Search the directory pointed at by vdp for the name pointed at
205 	 * by cnp->cn_nameptr.
206 	 */
207 	tdp = NULL;
208 	/*
209 	 * The outer loop ranges over the clusters that make up the
210 	 * directory.  Note that the root directory is different from all
211 	 * other directories.  It has a fixed number of blocks that are not
212 	 * part of the pool of allocatable clusters.  So, we treat it a
213 	 * little differently. The root directory starts at "cluster" 0.
214 	 */
215 	diroff = 0;
216 	for (frcn = 0; diroff < dp->de_FileSize; frcn++) {
217 		if ((error = pcbmap(dp, frcn, &bn, &cluster, &blsize)) != 0) {
218 			if (error == E2BIG)
219 				break;
220 			return (error);
221 		}
222 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
223 		    0, &bp);
224 		if (error) {
225 			brelse(bp, 0);
226 			return (error);
227 		}
228 		for (blkoff = 0; blkoff < blsize;
229 		     blkoff += sizeof(struct direntry),
230 		     diroff += sizeof(struct direntry)) {
231 			dep = (struct direntry *)((char *)bp->b_data + blkoff);
232 			/*
233 			 * If the slot is empty and we are still looking
234 			 * for an empty then remember this one.  If the
235 			 * slot is not empty then check to see if it
236 			 * matches what we are looking for.  If the slot
237 			 * has never been filled with anything, then the
238 			 * remainder of the directory has never been used,
239 			 * so there is no point in searching it.
240 			 */
241 			if (dep->deName[0] == SLOT_EMPTY ||
242 			    dep->deName[0] == SLOT_DELETED) {
243 				/*
244 				 * Drop memory of previous long matches
245 				 */
246 				chksum = -1;
247 
248 				if (slotcount < wincnt) {
249 					slotcount++;
250 					slotoffset = diroff;
251 				}
252 				if (dep->deName[0] == SLOT_EMPTY) {
253 					brelse(bp, 0);
254 					goto notfound;
255 				}
256 			} else {
257 				/*
258 				 * If there wasn't enough space for our
259 				 * winentries, forget about the empty space
260 				 */
261 				if (slotcount < wincnt)
262 					slotcount = 0;
263 
264 				/*
265 				 * Check for Win95 long filename entry
266 				 */
267 				if (dep->deAttributes == ATTR_WIN95) {
268 					if (pmp->pm_flags & MSDOSFSMNT_SHORTNAME)
269 						continue;
270 
271 					chksum = winChkName((const u_char *)cnp->cn_nameptr,
272 							    cnp->cn_namelen,
273 							    (struct winentry *)dep,
274 							    chksum);
275 					continue;
276 				}
277 
278 				/*
279 				 * Ignore volume labels (anywhere, not just
280 				 * the root directory).
281 				 */
282 				if (dep->deAttributes & ATTR_VOLUME) {
283 					chksum = -1;
284 					continue;
285 				}
286 
287 				/*
288 				 * Check for a checksum or name match
289 				 */
290 				chksum_ok = (chksum == winChksum(dep->deName));
291 				if (!chksum_ok
292 				    && (!olddos || memcmp(dosfilename, dep->deName, 11))) {
293 					chksum = -1;
294 					continue;
295 				}
296 #ifdef MSDOSFS_DEBUG
297 				printf("msdosfs_lookup(): match blkoff %d, diroff %d\n",
298 				    blkoff, diroff);
299 #endif
300 				/*
301 				 * Remember where this directory
302 				 * entry came from for whoever did
303 				 * this lookup.
304 				 */
305 				dp->de_fndoffset = diroff;
306 				if (chksum_ok && nameiop == RENAME) {
307 					/*
308 					 * Target had correct long name
309 					 * directory entries, reuse them
310 					 * as needed.
311 					 */
312 					dp->de_fndcnt = wincnt - 1;
313 				} else {
314 					/*
315 					 * Long name directory entries
316 					 * not present or corrupt, can only
317 					 * reuse dos directory entry.
318 					 */
319 					dp->de_fndcnt = 0;
320 				}
321 
322 				goto found;
323 			}
324 		}	/* for (blkoff = 0; .... */
325 		/*
326 		 * Release the buffer holding the directory cluster just
327 		 * searched.
328 		 */
329 		brelse(bp, 0);
330 	}	/* for (frcn = 0; ; frcn++) */
331 
332 notfound:
333 	/*
334 	 * We hold no disk buffers at this point.
335 	 */
336 
337 	/*
338 	 * If we get here we didn't find the entry we were looking for. But
339 	 * that's ok if we are creating or renaming and are at the end of
340 	 * the pathname and the directory hasn't been removed.
341 	 */
342 #ifdef MSDOSFS_DEBUG
343 	printf("msdosfs_lookup(): op %d, refcnt %ld, slotcount %d, slotoffset %d\n",
344 	    nameiop, dp->de_refcnt, slotcount, slotoffset);
345 #endif
346 	if ((nameiop == CREATE || nameiop == RENAME) &&
347 	    (flags & ISLASTCN) && dp->de_refcnt != 0) {
348 		/*
349 		 * Access for write is interpreted as allowing
350 		 * creation of files in the directory.
351 		 */
352 		error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred);
353 		if (error)
354 			return (error);
355 
356 		/*
357 		 * Fixup the slot description to point to the place where
358 		 * we might put the new DOS direntry (putting the Win95
359 		 * long name entries before that)
360 		 */
361 		if (!slotcount) {
362 			slotcount = 1;
363 			slotoffset = diroff;
364 		}
365 		if (wincnt > slotcount) {
366 			slotoffset +=
367 				sizeof(struct direntry) * (wincnt - slotcount);
368 		}
369 
370 		/*
371 		 * Return an indication of where the new directory
372 		 * entry should be put.
373 		 */
374 		dp->de_fndoffset = slotoffset;
375 		dp->de_fndcnt = wincnt - 1;
376 
377 		/*
378 		 * We return with the directory locked, so that
379 		 * the parameters we set up above will still be
380 		 * valid if we actually decide to do a direnter().
381 		 * We return ni_vp == NULL to indicate that the entry
382 		 * does not currently exist; we leave a pointer to
383 		 * the (locked) directory inode in ndp->ni_dvp.
384 		 * The pathname buffer is saved so that the name
385 		 * can be obtained later.
386 		 *
387 		 * NB - if the directory is unlocked, then this
388 		 * information cannot be used.
389 		 */
390 		cnp->cn_flags |= SAVENAME;
391 		return (EJUSTRETURN);
392 	}
393 
394 #if 0
395 	/*
396 	 * Insert name into cache (as non-existent) if appropriate.
397 	 *
398 	 * XXX Negative caching is broken for msdosfs because the name
399 	 * cache doesn't understand peculiarities such as case insensitivity
400 	 * and 8.3 filenames.  Hence, it may not invalidate all negative
401 	 * entries if a file with this name is later created.
402 	 * e.g. creating a file 'foo' won't invalidate a negative entry
403 	 * for 'FOO'.
404 	 */
405 	if ((cnp->cn_flags & MAKEENTRY) && nameiop != CREATE)
406 		cache_enter(vdp, *vpp, cnp);
407 #endif
408 
409 	return (ENOENT);
410 
411 found:
412 	/*
413 	 * NOTE:  We still have the buffer with matched directory entry at
414 	 * this point.
415 	 */
416 	isadir = dep->deAttributes & ATTR_DIRECTORY;
417 	scn = getushort(dep->deStartCluster);
418 	if (FAT32(pmp)) {
419 		scn |= getushort(dep->deHighClust) << 16;
420 		if (scn == pmp->pm_rootdirblk) {
421 			/*
422 			 * There should actually be 0 here.
423 			 * Just ignore the error.
424 			 */
425 			scn = MSDOSFSROOT;
426 		}
427 	}
428 
429 	if (isadir) {
430 		cluster = scn;
431 		if (cluster == MSDOSFSROOT)
432 			blkoff = MSDOSFSROOT_OFS;
433 		else
434 			blkoff = 0;
435 	} else if (cluster == MSDOSFSROOT)
436 		blkoff = diroff;
437 
438 	/*
439 	 * Now release buf to allow deget to read the entry again.
440 	 * Reserving it here and giving it to deget could result
441 	 * in a deadlock.
442 	 */
443 	brelse(bp, 0);
444 
445 foundroot:
446 	/*
447 	 * If we entered at foundroot, then we are looking for the . or ..
448 	 * entry of the filesystems root directory.  isadir and scn were
449 	 * setup before jumping here.  And, bp is already null.
450 	 */
451 	if (FAT32(pmp) && scn == MSDOSFSROOT)
452 		scn = pmp->pm_rootdirblk;
453 
454 	/*
455 	 * If deleting, and at end of pathname, return
456 	 * parameters which can be used to remove file.
457 	 * Lock the inode, being careful with ".".
458 	 */
459 	if (nameiop == DELETE && (flags & ISLASTCN)) {
460 		/*
461 		 * Don't allow deleting the root.
462 		 */
463 		if (blkoff == MSDOSFSROOT_OFS)
464 			return EROFS;			/* really? XXX */
465 
466 		/*
467 		 * Write access to directory required to delete files.
468 		 */
469 		error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred);
470 		if (error)
471 			return (error);
472 
473 		/*
474 		 * Return pointer to current entry in dp->i_offset.
475 		 * Save directory inode pointer in ndp->ni_dvp for dirremove().
476 		 */
477 		if (dp->de_StartCluster == scn && isadir) {	/* "." */
478 			VREF(vdp);
479 			*vpp = vdp;
480 			return (0);
481 		}
482 		if ((error = deget(pmp, cluster, blkoff, &tdp)) != 0)
483 			return (error);
484 		*vpp = DETOV(tdp);
485 		return (0);
486 	}
487 
488 	/*
489 	 * If rewriting (RENAME), return the inode and the
490 	 * information required to rewrite the present directory
491 	 * Must get inode of directory entry to verify it's a
492 	 * regular file, or empty directory.
493 	 */
494 	if (nameiop == RENAME && (flags & ISLASTCN)) {
495 
496 		if (vdp->v_mount->mnt_flag & MNT_RDONLY)
497 			return (EROFS);
498 
499 		if (blkoff == MSDOSFSROOT_OFS)
500 			return EROFS;				/* really? XXX */
501 
502 		error = VOP_ACCESS(vdp, VWRITE, cnp->cn_cred);
503 		if (error)
504 			return (error);
505 
506 		/*
507 		 * Careful about locking second inode.
508 		 * This can only occur if the target is ".".
509 		 */
510 		if (dp->de_StartCluster == scn && isadir)
511 			return (EISDIR);
512 
513 		if ((error = deget(pmp, cluster, blkoff, &tdp)) != 0)
514 			return (error);
515 		*vpp = DETOV(tdp);
516 		cnp->cn_flags |= SAVENAME;
517 		return (0);
518 	}
519 
520 	/*
521 	 * Step through the translation in the name.  We do not `vput' the
522 	 * directory because we may need it again if a symbolic link
523 	 * is relative to the current directory.  Instead we save it
524 	 * unlocked as "pdp".  We must get the target inode before unlocking
525 	 * the directory to insure that the inode will not be removed
526 	 * before we get it.  We prevent deadlock by always fetching
527 	 * inodes from the root, moving down the directory tree. Thus
528 	 * when following backward pointers ".." we must unlock the
529 	 * parent directory before getting the requested directory.
530 	 * There is a potential race condition here if both the current
531 	 * and parent directories are removed before the VFS_VGET for the
532 	 * inode associated with ".." returns.  We hope that this occurs
533 	 * infrequently since we cannot avoid this race condition without
534 	 * implementing a sophisticated deadlock detection algorithm.
535 	 * Note also that this simple deadlock detection scheme will not
536 	 * work if the file system has any hard links other than ".."
537 	 * that point backwards in the directory structure.
538 	 */
539 	pdp = vdp;
540 	if (flags & ISDOTDOT) {
541 		VOP_UNLOCK(pdp, 0);	/* race to get the inode */
542 		error = deget(pmp, cluster, blkoff, &tdp);
543 		vn_lock(pdp, LK_EXCLUSIVE | LK_RETRY);
544 		if (error) {
545 			return error;
546 		}
547 		*vpp = DETOV(tdp);
548 	} else if (dp->de_StartCluster == scn && isadir) {
549 		VREF(vdp);	/* we want ourself, ie "." */
550 		*vpp = vdp;
551 	} else {
552 		if ((error = deget(pmp, cluster, blkoff, &tdp)) != 0)
553 			return (error);
554 		*vpp = DETOV(tdp);
555 	}
556 
557 	/*
558 	 * Insert name into cache if appropriate.
559 	 */
560 	if (cnp->cn_flags & MAKEENTRY)
561 		cache_enter(vdp, *vpp, cnp);
562 
563 	return (0);
564 }
565 
566 /*
567  * dep  - directory entry to copy into the directory
568  * ddep - directory to add to
569  * depp - return the address of the denode for the created directory entry
570  *	  if depp != 0
571  * cnp  - componentname needed for Win95 long filenames
572  */
573 int
574 createde(struct denode *dep, struct denode *ddep, struct denode **depp, struct componentname *cnp)
575 {
576 	int error, rberror;
577 	u_long dirclust, clusoffset;
578 	u_long fndoffset, havecnt=0, wcnt=1;
579 	struct direntry *ndep;
580 	struct msdosfsmount *pmp = ddep->de_pmp;
581 	struct buf *bp;
582 	daddr_t bn;
583 	int blsize, i;
584 	int async = ddep->de_pmp->pm_mountp->mnt_flag & MNT_ASYNC;
585 
586 #ifdef MSDOSFS_DEBUG
587 	printf("createde(dep %p, ddep %p, depp %p, cnp %p)\n",
588 	    dep, ddep, depp, cnp);
589 #endif
590 
591 	/*
592 	 * If no space left in the directory then allocate another cluster
593 	 * and chain it onto the end of the file.  There is one exception
594 	 * to this.  That is, if the root directory has no more space it
595 	 * can NOT be expanded.  extendfile() checks for and fails attempts
596 	 * to extend the root directory.  We just return an error in that
597 	 * case.
598 	 */
599 	if (ddep->de_fndoffset >= ddep->de_FileSize) {
600 		u_long needlen = ddep->de_fndoffset + sizeof(struct direntry)
601 		    - ddep->de_FileSize;
602 		dirclust = de_clcount(pmp, needlen);
603 		if ((error = extendfile(ddep, dirclust, 0, 0, DE_CLEAR)) != 0) {
604 			(void)detrunc(ddep, ddep->de_FileSize, 0, NOCRED);
605 			goto err_norollback;
606 		}
607 
608 		/*
609 		 * Update the size of the directory
610 		 */
611 		ddep->de_FileSize += de_cn2off(pmp, dirclust);
612 	}
613 
614 	/*
615 	 * We just read in the cluster with space.  Copy the new directory
616 	 * entry in.  Then write it to disk. NOTE:  DOS directories
617 	 * do not get smaller as clusters are emptied.
618 	 */
619 	error = pcbmap(ddep, de_cluster(pmp, ddep->de_fndoffset),
620 		       &bn, &dirclust, &blsize);
621 	if (error)
622 		goto err_norollback;
623 	clusoffset = ddep->de_fndoffset;
624 	if (dirclust != MSDOSFSROOT)
625 		clusoffset &= pmp->pm_crbomask;
626 	if ((error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
627 	    B_MODIFY, &bp)) != 0) {
628 		brelse(bp, 0);
629 		goto err_norollback;
630 	}
631 	ndep = bptoep(pmp, bp, clusoffset);
632 
633 	DE_EXTERNALIZE(ndep, dep);
634 
635 	/*
636 	 * Now write the Win95 long name
637 	 */
638 	if (ddep->de_fndcnt > 0) {
639 		u_int8_t chksum = winChksum(ndep->deName);
640 		const u_char *un = (const u_char *)cnp->cn_nameptr;
641 		int unlen = cnp->cn_namelen;
642 		u_long xhavecnt;
643 
644 		fndoffset = ddep->de_fndoffset;
645 		xhavecnt = ddep->de_fndcnt + 1;
646 
647 		for(; wcnt < xhavecnt; wcnt++) {
648 			if ((fndoffset & pmp->pm_crbomask) == 0) {
649 				/* we should never get here if ddep is root
650 				 * directory */
651 
652 				if (async)
653 					(void) bdwrite(bp);
654 				else if ((error = bwrite(bp)) != 0)
655 					goto rollback;
656 
657 				fndoffset -= sizeof(struct direntry);
658 				error = pcbmap(ddep,
659 					       de_cluster(pmp, fndoffset),
660 					       &bn, 0, &blsize);
661 				if (error)
662 					goto rollback;
663 
664 				error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn),
665 				    blsize, NOCRED, B_MODIFY, &bp);
666 				if (error) {
667 					brelse(bp, 0);
668 					goto rollback;
669 				}
670 				ndep = bptoep(pmp, bp,
671 						fndoffset & pmp->pm_crbomask);
672 			} else {
673 				ndep--;
674 				fndoffset -= sizeof(struct direntry);
675 			}
676 			if (!unix2winfn(un, unlen, (struct winentry *)ndep,
677 						wcnt, chksum))
678 				break;
679 		}
680 	}
681 
682 	if (async)
683 		bdwrite(bp);
684 	else if ((error = bwrite(bp)) != 0)
685 		goto rollback;
686 
687 	/*
688 	 * If they want us to return with the denode gotten.
689 	 */
690 	if (depp) {
691 		u_long diroffset = clusoffset;
692 		if (dep->de_Attributes & ATTR_DIRECTORY) {
693 			dirclust = dep->de_StartCluster;
694 			if (FAT32(pmp) && dirclust == pmp->pm_rootdirblk)
695 				dirclust = MSDOSFSROOT;
696 			if (dirclust == MSDOSFSROOT)
697 				diroffset = MSDOSFSROOT_OFS;
698 			else
699 				diroffset = 0;
700 		}
701 		return deget(pmp, dirclust, diroffset, depp);
702 	}
703 
704 	return 0;
705 
706     rollback:
707 	/*
708 	 * Mark all slots modified so far as deleted. Note that we
709 	 * can't just call removede(), since directory is not in
710 	 * consistent state.
711 	 */
712 	fndoffset = ddep->de_fndoffset;
713 	rberror = pcbmap(ddep, de_cluster(pmp, fndoffset),
714 	       &bn, NULL, &blsize);
715 	if (rberror)
716 		goto err_norollback;
717 	if ((rberror = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
718 	    B_MODIFY, &bp)) != 0) {
719 		brelse(bp, 0);
720 		goto err_norollback;
721 	}
722 	ndep = bptoep(pmp, bp, clusoffset);
723 
724 	havecnt = ddep->de_fndcnt + 1;
725 	for(i=wcnt; i <= havecnt; i++) {
726 		/* mark entry as deleted */
727 		ndep->deName[0] = SLOT_DELETED;
728 
729 		if ((fndoffset & pmp->pm_crbomask) == 0) {
730 			/* we should never get here if ddep is root
731 			 * directory */
732 
733 			if (async)
734 				bdwrite(bp);
735 			else if ((rberror = bwrite(bp)) != 0)
736 				goto err_norollback;
737 
738 			fndoffset -= sizeof(struct direntry);
739 			rberror = pcbmap(ddep,
740 				       de_cluster(pmp, fndoffset),
741 				       &bn, 0, &blsize);
742 			if (rberror)
743 				goto err_norollback;
744 
745 			rberror = bread(pmp->pm_devvp, de_bn2kb(pmp, bn),
746 			    blsize, NOCRED, B_MODIFY, &bp);
747 			if (rberror) {
748 				brelse(bp, 0);
749 				goto err_norollback;
750 			}
751 			ndep = bptoep(pmp, bp, fndoffset);
752 		} else {
753 			ndep--;
754 			fndoffset -= sizeof(struct direntry);
755 		}
756 	}
757 
758 	/* ignore any further error */
759 	if (async)
760 		(void) bdwrite(bp);
761 	else
762 		(void) bwrite(bp);
763 
764     err_norollback:
765 	return error;
766 }
767 
768 /*
769  * Be sure a directory is empty except for "." and "..". Return 1 if empty,
770  * return 0 if not empty or error.
771  */
772 int
773 dosdirempty(struct denode *dep)
774 {
775 	int blsize;
776 	int error;
777 	u_long cn;
778 	daddr_t bn;
779 	struct buf *bp;
780 	struct msdosfsmount *pmp = dep->de_pmp;
781 	struct direntry *dentp;
782 
783 	/*
784 	 * Since the filesize field in directory entries for a directory is
785 	 * zero, we just have to feel our way through the directory until
786 	 * we hit end of file.
787 	 */
788 	for (cn = 0;; cn++) {
789 		if ((error = pcbmap(dep, cn, &bn, 0, &blsize)) != 0) {
790 			if (error == E2BIG)
791 				return (1);	/* it's empty */
792 			return (0);
793 		}
794 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
795 		    0, &bp);
796 		if (error) {
797 			brelse(bp, 0);
798 			return (0);
799 		}
800 		for (dentp = (struct direntry *)bp->b_data;
801 		     (char *)dentp < (char *)bp->b_data + blsize;
802 		     dentp++) {
803 			if (dentp->deName[0] != SLOT_DELETED &&
804 			    (dentp->deAttributes & ATTR_VOLUME) == 0) {
805 				/*
806 				 * In dos directories an entry whose name
807 				 * starts with SLOT_EMPTY (0) starts the
808 				 * beginning of the unused part of the
809 				 * directory, so we can just return that it
810 				 * is empty.
811 				 */
812 				if (dentp->deName[0] == SLOT_EMPTY) {
813 					brelse(bp, 0);
814 					return (1);
815 				}
816 				/*
817 				 * Any names other than "." and ".." in a
818 				 * directory mean it is not empty.
819 				 */
820 				if (memcmp(dentp->deName, ".          ", 11) &&
821 				    memcmp(dentp->deName, "..         ", 11)) {
822 					brelse(bp, 0);
823 #ifdef MSDOSFS_DEBUG
824 					printf("dosdirempty(): found %.11s, %d, %d\n",
825 					    dentp->deName, dentp->deName[0],
826 						dentp->deName[1]);
827 #endif
828 					return (0);	/* not empty */
829 				}
830 			}
831 		}
832 		brelse(bp, 0);
833 	}
834 	/* NOTREACHED */
835 }
836 
837 /*
838  * Check to see if the directory described by target is in some
839  * subdirectory of source.  This prevents something like the following from
840  * succeeding and leaving a bunch or files and directories orphaned. mv
841  * /a/b/c /a/b/c/d/e/f Where c and f are directories.
842  *
843  * source - the inode for /a/b/c
844  * target - the inode for /a/b/c/d/e/f
845  *
846  * Returns 0 if target is NOT a subdirectory of source.
847  * Otherwise returns a non-zero error number.
848  * The target inode is always unlocked on return.
849  */
850 int
851 doscheckpath(struct denode *source, struct denode *target)
852 {
853 	u_long scn;
854 	struct msdosfsmount *pmp;
855 	struct direntry *ep;
856 	struct denode *dep;
857 	struct buf *bp = NULL;
858 	int error = 0;
859 
860 	dep = target;
861 	if ((target->de_Attributes & ATTR_DIRECTORY) == 0 ||
862 	    (source->de_Attributes & ATTR_DIRECTORY) == 0) {
863 		error = ENOTDIR;
864 		goto out;
865 	}
866 	if (dep->de_StartCluster == source->de_StartCluster) {
867 		error = EEXIST;
868 		goto out;
869 	}
870 	if (dep->de_StartCluster == MSDOSFSROOT)
871 		goto out;
872 	pmp = dep->de_pmp;
873 #ifdef	DIAGNOSTIC
874 	if (pmp != source->de_pmp)
875 		panic("doscheckpath: source and target on different filesystems");
876 #endif
877 	if (FAT32(pmp) && dep->de_StartCluster == pmp->pm_rootdirblk)
878 		goto out;
879 
880 	for (;;) {
881 		if ((dep->de_Attributes & ATTR_DIRECTORY) == 0) {
882 			error = ENOTDIR;
883 			break;
884 		}
885 		scn = dep->de_StartCluster;
886 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, cntobn(pmp, scn)),
887 			      pmp->pm_bpcluster, NOCRED, 0, &bp);
888 		if (error)
889 			break;
890 
891 		ep = (struct direntry *) bp->b_data + 1;
892 		if ((ep->deAttributes & ATTR_DIRECTORY) == 0 ||
893 		    memcmp(ep->deName, "..         ", 11) != 0) {
894 			error = ENOTDIR;
895 			break;
896 		}
897 		scn = getushort(ep->deStartCluster);
898 		if (FAT32(pmp))
899 			scn |= getushort(ep->deHighClust) << 16;
900 
901 		if (scn == source->de_StartCluster) {
902 			error = EINVAL;
903 			break;
904 		}
905 		if (scn == MSDOSFSROOT)
906 			break;
907 		if (FAT32(pmp) && scn == pmp->pm_rootdirblk) {
908 			/*
909 			 * scn should be 0 in this case,
910 			 * but we silently ignore the error.
911 			 */
912 			break;
913 		}
914 
915 		vput(DETOV(dep));
916 		brelse(bp, 0);
917 		bp = NULL;
918 		/* NOTE: deget() clears dep on error */
919 		if ((error = deget(pmp, scn, 0, &dep)) != 0)
920 			break;
921 	}
922 out:
923 	if (bp)
924 		brelse(bp, 0);
925 	if (error == ENOTDIR)
926 		printf("doscheckpath(): .. not a directory?\n");
927 	if (dep != NULL)
928 		vput(DETOV(dep));
929 	return (error);
930 }
931 
932 /*
933  * Read in the disk block containing the directory entry (dirclu, dirofs)
934  * and return the address of the buf header, and the address of the
935  * directory entry within the block.
936  */
937 int
938 readep(struct msdosfsmount *pmp, u_long dirclust, u_long diroffset, struct buf **bpp, struct direntry **epp)
939 {
940 	int error;
941 	daddr_t bn;
942 	int blsize;
943 
944 	blsize = pmp->pm_bpcluster;
945 	if (dirclust == MSDOSFSROOT
946 	    && de_blk(pmp, diroffset + blsize) > pmp->pm_rootdirsize)
947 		blsize = de_bn2off(pmp, pmp->pm_rootdirsize) & pmp->pm_crbomask;
948 	bn = detobn(pmp, dirclust, diroffset);
949 	if ((error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
950 	    0, bpp)) != 0) {
951 		brelse(*bpp, 0);
952 		*bpp = NULL;
953 		return (error);
954 	}
955 	if (epp)
956 		*epp = bptoep(pmp, *bpp, diroffset);
957 	return (0);
958 }
959 
960 /*
961  * Read in the disk block containing the directory entry dep came from and
962  * return the address of the buf header, and the address of the directory
963  * entry within the block.
964  */
965 int
966 readde(struct denode *dep, struct buf **bpp, struct direntry **epp)
967 {
968 	return (readep(dep->de_pmp, dep->de_dirclust, dep->de_diroffset,
969 			bpp, epp));
970 }
971 
972 /*
973  * Remove a directory entry. At this point the file represented by the
974  * directory entry to be removed is still full length until noone has it
975  * open.  When the file no longer being used msdosfs_inactive() is called
976  * and will truncate the file to 0 length.  When the vnode containing the
977  * denode is needed for some other purpose by VFS it will call
978  * msdosfs_reclaim() which will remove the denode from the denode cache.
979  */
980 int
981 removede(struct denode *pdep, struct denode *dep)
982 	/* pdep:	 directory where the entry is removed */
983 	/* dep:	 file to be removed */
984 {
985 	int error;
986 	struct direntry *ep;
987 	struct buf *bp;
988 	daddr_t bn;
989 	int blsize;
990 	struct msdosfsmount *pmp = pdep->de_pmp;
991 	u_long offset = pdep->de_fndoffset;
992 	int async = pdep->de_pmp->pm_mountp->mnt_flag & MNT_ASYNC;
993 
994 #ifdef MSDOSFS_DEBUG
995 	printf("removede(): filename %s, dep %p, offset %08lx\n",
996 	    dep->de_Name, dep, offset);
997 #endif
998 
999 	dep->de_refcnt--;
1000 	offset += sizeof(struct direntry);
1001 	do {
1002 		offset -= sizeof(struct direntry);
1003 		error = pcbmap(pdep, de_cluster(pmp, offset), &bn, 0, &blsize);
1004 		if (error)
1005 			return error;
1006 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
1007 		    B_MODIFY, &bp);
1008 		if (error) {
1009 			brelse(bp, 0);
1010 			return error;
1011 		}
1012 		ep = bptoep(pmp, bp, offset);
1013 		/*
1014 		 * Check whether, if we came here the second time, i.e.
1015 		 * when underflowing into the previous block, the last
1016 		 * entry in this block is a longfilename entry, too.
1017 		 */
1018 		if (ep->deAttributes != ATTR_WIN95
1019 		    && offset != pdep->de_fndoffset) {
1020 			brelse(bp, 0);
1021 			break;
1022 		}
1023 		offset += sizeof(struct direntry);
1024 		while (1) {
1025 			/*
1026 			 * We are a bit agressive here in that we delete any Win95
1027 			 * entries preceding this entry, not just the ones we "own".
1028 			 * Since these presumably aren't valid anyway,
1029 			 * there should be no harm.
1030 			 */
1031 			offset -= sizeof(struct direntry);
1032 			ep--->deName[0] = SLOT_DELETED;
1033 			if ((pmp->pm_flags & MSDOSFSMNT_NOWIN95)
1034 			    || !(offset & pmp->pm_crbomask)
1035 			    || ep->deAttributes != ATTR_WIN95)
1036 				break;
1037 		}
1038 		if (async)
1039 			bdwrite(bp);
1040 		else if ((error = bwrite(bp)) != 0)
1041 			return error;
1042 	} while (!(pmp->pm_flags & MSDOSFSMNT_NOWIN95)
1043 	    && !(offset & pmp->pm_crbomask)
1044 	    && offset);
1045 	return 0;
1046 }
1047 
1048 /*
1049  * Create a unique DOS name in dvp
1050  */
1051 int
1052 uniqdosname(struct denode *dep, struct componentname *cnp, u_char *cp)
1053 {
1054 	struct msdosfsmount *pmp = dep->de_pmp;
1055 	struct direntry *dentp;
1056 	int gen;
1057 	int blsize;
1058 	u_long cn;
1059 	daddr_t bn;
1060 	struct buf *bp;
1061 	int error;
1062 
1063 	for (gen = 1;; gen++) {
1064 		/*
1065 		 * Generate DOS name with generation number
1066 		 */
1067 		if (!unix2dosfn((const u_char *)cnp->cn_nameptr, cp,
1068 		    cnp->cn_namelen, gen))
1069 			return gen == 1 ? EINVAL : EEXIST;
1070 
1071 		/*
1072 		 * Now look for a dir entry with this exact name
1073 		 */
1074 		for (cn = error = 0; !error; cn++) {
1075 			if ((error = pcbmap(dep, cn, &bn, 0, &blsize)) != 0) {
1076 				if (error == E2BIG)	/* EOF reached and not found */
1077 					return 0;
1078 				return error;
1079 			}
1080 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize,
1081 			    NOCRED, 0, &bp);
1082 			if (error) {
1083 				brelse(bp, 0);
1084 				return error;
1085 			}
1086 			for (dentp = (struct direntry *)bp->b_data;
1087 			     (char *)dentp < (char *)bp->b_data + blsize;
1088 			     dentp++) {
1089 				if (dentp->deName[0] == SLOT_EMPTY) {
1090 					/*
1091 					 * Last used entry and not found
1092 					 */
1093 					brelse(bp, 0);
1094 					return 0;
1095 				}
1096 				/*
1097 				 * Ignore volume labels and Win95 entries
1098 				 */
1099 				if (dentp->deAttributes & ATTR_VOLUME)
1100 					continue;
1101 				if (!memcmp(dentp->deName, cp, 11)) {
1102 					error = EEXIST;
1103 					break;
1104 				}
1105 			}
1106 			brelse(bp, 0);
1107 		}
1108 	}
1109 }
1110 
1111 /*
1112  * Find any Win'95 long filename entry in directory dep
1113  */
1114 int
1115 findwin95(struct denode *dep)
1116 {
1117 	struct msdosfsmount *pmp = dep->de_pmp;
1118 	struct direntry *dentp;
1119 	int blsize, win95;
1120 	u_long cn;
1121 	daddr_t bn;
1122 	struct buf *bp;
1123 
1124 	win95 = 1;
1125 	/*
1126 	 * Read through the directory looking for Win'95 entries
1127 	 * XXX Note: Error currently handled just as EOF
1128 	 */
1129 	for (cn = 0;; cn++) {
1130 		if (pcbmap(dep, cn, &bn, 0, &blsize))
1131 			return win95;
1132 		if (bread(pmp->pm_devvp, de_bn2kb(pmp, bn), blsize, NOCRED,
1133 		    0, &bp)) {
1134 			brelse(bp, 0);
1135 			return win95;
1136 		}
1137 		for (dentp = (struct direntry *)bp->b_data;
1138 		     (char *)dentp < (char *)bp->b_data + blsize;
1139 		     dentp++) {
1140 			if (dentp->deName[0] == SLOT_EMPTY) {
1141 				/*
1142 				 * Last used entry and not found
1143 				 */
1144 				brelse(bp, 0);
1145 				return win95;
1146 			}
1147 			if (dentp->deName[0] == SLOT_DELETED) {
1148 				/*
1149 				 * Ignore deleted files
1150 				 * Note: might be an indication of Win'95
1151 				 * anyway	XXX
1152 				 */
1153 				continue;
1154 			}
1155 			if (dentp->deAttributes == ATTR_WIN95) {
1156 				brelse(bp, 0);
1157 				return 1;
1158 			}
1159 			win95 = 0;
1160 		}
1161 		brelse(bp, 0);
1162 	}
1163 }
1164