xref: /netbsd-src/sys/fs/msdosfs/msdosfs_fat.c (revision 82d56013d7b633d116a93943de88e08335357a7c)
1 /*	$NetBSD: msdosfs_fat.c,v 1.35 2020/09/07 01:35:25 mrg 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 #if HAVE_NBTOOL_CONFIG_H
51 #include "nbtool_config.h"
52 #endif
53 
54 #include <sys/cdefs.h>
55 __KERNEL_RCSID(0, "$NetBSD: msdosfs_fat.c,v 1.35 2020/09/07 01:35:25 mrg Exp $");
56 
57 /*
58  * kernel include files.
59  */
60 #include <sys/param.h>
61 #include <sys/file.h>
62 #ifdef _KERNEL
63 #include <sys/mount.h>		/* to define statvfs structure */
64 #include <sys/errno.h>
65 #include <sys/systm.h>
66 #include <sys/kauth.h>
67 #include <sys/dirent.h>
68 #include <sys/namei.h>
69 #include <sys/buf.h>
70 #include <sys/vnode.h>		/* to define vattr structure */
71 #else
72 #include <strings.h>
73 #include <ffs/buf.h>
74 #endif
75 
76 /*
77  * msdosfs include files.
78  */
79 #include <fs/msdosfs/bpb.h>
80 #include <fs/msdosfs/msdosfsmount.h>
81 #include <fs/msdosfs/direntry.h>
82 #include <fs/msdosfs/denode.h>
83 #include <fs/msdosfs/fat.h>
84 
85 /*
86  * Fat cache stats.
87  */
88 int fc_fileextends;		/* # of file extends			 */
89 int fc_lfcempty;		/* # of time last file cluster cache entry
90 				 * was empty */
91 int fc_bmapcalls;		/* # of times pcbmap was called		 */
92 
93 #define	LMMAX	20
94 int fc_lmdistance[LMMAX];	/* counters for how far off the last
95 				 * cluster mapped entry was. */
96 int fc_largedistance;		/* off by more than LMMAX		 */
97 int fc_wherefrom, fc_whereto, fc_lastclust;
98 int pm_fatblocksize;
99 
100 #ifdef MSDOSFS_DEBUG
101 #define DPRINTF(a) printf a
102 #else
103 #define DPRINTF(a)
104 #endif
105 #ifdef MSDOSFS_DEBUG
106 void print_fat_stats(void);
107 
108 void
109 print_fat_stats(void)
110 {
111 	int i;
112 
113 	printf("fc_fileextends=%d fc_lfcempty=%d fc_bmapcalls=%d "
114 	    "fc_largedistance=%d [%d->%d=%d] fc_lastclust=%d pm_fatblocksize=%d\n",
115 	    fc_fileextends, fc_lfcempty, fc_bmapcalls, fc_largedistance,
116 	    fc_wherefrom, fc_whereto, fc_whereto-fc_wherefrom,
117 	    fc_lastclust, pm_fatblocksize);
118 
119 	fc_fileextends = fc_lfcempty = fc_bmapcalls = 0;
120 	fc_wherefrom = fc_whereto = fc_lastclust = 0;
121 
122 	for (i = 0; i < LMMAX; i++) {
123 		printf("%d:%d ", i, fc_lmdistance[i]);
124 	fc_lmdistance[i] = 0;
125 	}
126 
127 	printf("\n");
128 }
129 #endif
130 
131 static void fatblock(struct msdosfsmount *, u_long, u_long *, u_long *,
132 			  u_long *);
133 void updatefats(struct msdosfsmount *, struct buf *, u_long);
134 static inline void usemap_free(struct msdosfsmount *, u_long);
135 static inline void usemap_alloc(struct msdosfsmount *, u_long);
136 static int fatchain(struct msdosfsmount *, u_long, u_long, u_long);
137 int chainlength(struct msdosfsmount *, u_long, u_long);
138 int chainalloc(struct msdosfsmount *, u_long, u_long, u_long, u_long *,
139 		    u_long *);
140 
141 static void
142 fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp, u_long *sizep, u_long *bop)
143 {
144 	u_long bn, size;
145 
146 	bn = ofs / pmp->pm_fatblocksize * pmp->pm_fatblocksec;
147 	size = uimin(pmp->pm_fatblocksec, pmp->pm_FATsecs - bn)
148 	    * pmp->pm_BytesPerSec;
149 	bn += pmp->pm_fatblk + pmp->pm_curfat * pmp->pm_FATsecs;
150 
151 	DPRINTF(("%s(ofs=%lu bn=%lu, size=%lu, bo=%lu)\n", __func__, ofs, bn,
152 	    size, ofs % pmp->pm_fatblocksize));
153 	if (bnp)
154 		*bnp = bn;
155 	if (sizep)
156 		*sizep = size;
157 	if (bop)
158 		*bop = ofs % pmp->pm_fatblocksize;
159 
160 	pm_fatblocksize = pmp->pm_fatblocksize;
161 }
162 
163 /*
164  * Map the logical cluster number of a file into a physical disk sector
165  * that is filesystem relative.
166  *
167  * dep	  - address of denode representing the file of interest
168  * findcn - file relative cluster whose filesystem relative cluster number
169  *	    and/or block number are/is to be found
170  * bnp	  - address of where to place the file system relative block number.
171  *	    If this pointer is null then don't return this quantity.
172  * cnp	  - address of where to place the file system relative cluster number.
173  *	    If this pointer is null then don't return this quantity.
174  *
175  * NOTE: Either bnp or cnp must be non-null.
176  * This function has one side effect.  If the requested file relative cluster
177  * is beyond the end of file, then the actual number of clusters in the file
178  * is returned in *cnp.  This is useful for determining how long a directory is.
179  *  If cnp is null, nothing is returned.
180  */
181 int
182 pcbmap(struct denode *dep, u_long findcn, daddr_t *bnp, u_long *cnp, int *sp)
183 	/* findcn:		 file relative cluster to get		 */
184 	/* bnp:		 returned filesys rel sector number	 */
185 	/* cnp:		 returned cluster number		 */
186 	/* sp:		 returned block size			 */
187 {
188 	int error;
189 	u_long i;
190 	u_long cn;
191 	u_long prevcn = 0; /* XXX: prevcn could be used unititialized */
192 	u_long byteoffset;
193 	u_long bn;
194 	u_long bo;
195 	struct buf *bp = NULL;
196 	u_long bp_bn = -1;
197 	struct msdosfsmount *pmp = dep->de_pmp;
198 	u_long bsize;
199 
200 	fc_bmapcalls++;
201 
202 	/*
203 	 * If they don't give us someplace to return a value then don't
204 	 * bother doing anything.
205 	 */
206 	if (bnp == NULL && cnp == NULL && sp == NULL)
207 		return (0);
208 
209 	cn = dep->de_StartCluster;
210 	DPRINTF(("%s(start cluster=%lu)\n", __func__, cn));
211 	/*
212 	 * The "file" that makes up the root directory is contiguous,
213 	 * permanently allocated, of fixed size, and is not made up of
214 	 * clusters.  If the cluster number is beyond the end of the root
215 	 * directory, then return the number of clusters in the file.
216 	 */
217 	if (cn == MSDOSFSROOT) {
218 		if (dep->de_Attributes & ATTR_DIRECTORY) {
219 			if (de_cn2off(pmp, findcn) >= dep->de_FileSize) {
220 				if (cnp)
221 					*cnp = de_bn2cn(pmp, pmp->pm_rootdirsize);
222 				DPRINTF(("%s(root, %lu ETOOBIG)\n", __func__,
223 				    de_cn2off(pmp, findcn)));
224 				return (E2BIG);
225 			}
226 			if (bnp)
227 				*bnp = pmp->pm_rootdirblk + de_cn2bn(pmp, findcn);
228 			if (cnp)
229 				*cnp = MSDOSFSROOT;
230 			if (sp)
231 				*sp = uimin(pmp->pm_bpcluster,
232 				    dep->de_FileSize - de_cn2off(pmp, findcn));
233 			DPRINTF(("%s(root, bn=%lu, cn=%u)\n", __func__,
234 			    pmp->pm_rootdirblk + de_cn2bn(pmp, findcn),
235 			    MSDOSFSROOT));
236 			return (0);
237 		} else {		/* just an empty file */
238 			if (cnp)
239 				*cnp = 0;
240 			DPRINTF(("%s(root, empty ETOOBIG)\n", __func__));
241 			return (E2BIG);
242 		}
243 	}
244 
245 	/*
246 	 * All other files do I/O in cluster sized blocks
247 	 */
248 	if (sp)
249 		*sp = pmp->pm_bpcluster;
250 
251 	/*
252 	 * Rummage around in the FAT cache, maybe we can avoid tromping
253 	 * thru every FAT entry for the file. And, keep track of how far
254 	 * off the cache was from where we wanted to be.
255 	 */
256 	i = 0;
257 	fc_lookup(dep, findcn, &i, &cn);
258 	DPRINTF(("%s(bpcluster=%lu i=%lu cn=%lu\n", __func__, pmp->pm_bpcluster,
259 	    i, cn));
260 	if ((bn = findcn - i) >= LMMAX) {
261 		fc_largedistance++;
262 		fc_wherefrom = i;
263 		fc_whereto = findcn;
264 		fc_lastclust = dep->de_fc[FC_LASTFC].fc_frcn;
265 	} else
266 		fc_lmdistance[bn]++;
267 
268 	/*
269 	 * Handle all other files or directories the normal way.
270 	 */
271 	for (; i < findcn; i++) {
272 		/*
273 		 * Stop with all reserved clusters, not just with EOF.
274 		 */
275 		if (cn >= (CLUST_RSRVD & pmp->pm_fatmask))
276 			goto hiteof;
277 
278 		/*
279 		 * Also stop when cluster is not in the filesystem
280 		 */
281 		if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster) {
282 			DPRINTF(("%s(cn, %lu not in %lu..%lu)\n", __func__,
283 				cn, (u_long)CLUST_FIRST, pmp->pm_maxcluster));
284 			if (bp)
285 				brelse(bp, 0);
286 			return (EINVAL);
287 		}
288 
289 		byteoffset = FATOFS(pmp, cn);
290 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
291 		if (bn != bp_bn) {
292 			if (bp)
293 				brelse(bp, 0);
294 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
295 			    0, &bp);
296 			if (error) {
297 				DPRINTF(("%s(bread, %d)\n", __func__, error));
298 				return (error);
299 			}
300 			bp_bn = bn;
301 		}
302 		prevcn = cn;
303 		if (bo >= bsize) {
304 			if (bp)
305 				brelse(bp, 0);
306 			DPRINTF(("%s(block, %lu >= %lu)\n", __func__, bo,
307 			    bsize));
308 			return (EIO);
309 		}
310 		KASSERT(bp != NULL);
311 		if (FAT32(pmp))
312 			cn = getulong((char *)bp->b_data + bo);
313 		else
314 			cn = getushort((char *)bp->b_data + bo);
315 		if (FAT12(pmp) && (prevcn & 1))
316 			cn >>= 4;
317 		DPRINTF(("%s(cn=%lu masked=%lu)\n", __func__, cn,
318 		    cn & pmp->pm_fatmask));
319 		cn &= pmp->pm_fatmask;
320 	}
321 
322 	if (!MSDOSFSEOF(cn, pmp->pm_fatmask)) {
323 		if (bp)
324 			brelse(bp, 0);
325 		if (bnp)
326 			*bnp = cntobn(pmp, cn);
327 		if (cnp)
328 			*cnp = cn;
329 		DPRINTF(("%s(bn=%lu, cn=%lu)\n", __func__, cntobn(pmp, cn),
330 		    cn));
331 		fc_setcache(dep, FC_LASTMAP, i, cn);
332 		return (0);
333 	}
334 
335 hiteof:;
336 	if (cnp)
337 		*cnp = i;
338 	if (bp)
339 		brelse(bp, 0);
340 	/* update last file cluster entry in the FAT cache */
341 	fc_setcache(dep, FC_LASTFC, i - 1, prevcn);
342 	DPRINTF(("%s(eof, %lu)\n", __func__, i));
343 	return (E2BIG);
344 }
345 
346 /*
347  * Find the closest entry in the FAT cache to the cluster we are looking
348  * for.
349  */
350 void
351 fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp, u_long *fsrcnp)
352 {
353 	int i;
354 	u_long cn;
355 	struct fatcache *closest = 0;
356 
357 	for (i = 0; i < FC_SIZE; i++) {
358 		cn = dep->de_fc[i].fc_frcn;
359 		if (cn != FCE_EMPTY && cn <= findcn) {
360 			if (closest == 0 || cn > closest->fc_frcn)
361 				closest = &dep->de_fc[i];
362 		}
363 	}
364 	if (closest) {
365 		*frcnp = closest->fc_frcn;
366 		*fsrcnp = closest->fc_fsrcn;
367 	}
368 }
369 
370 /*
371  * Purge the FAT cache in denode dep of all entries relating to file
372  * relative cluster frcn and beyond.
373  */
374 void
375 fc_purge(struct denode *dep, u_int frcn)
376 {
377 	int i;
378 	struct fatcache *fcp;
379 
380 	fcp = dep->de_fc;
381 	for (i = 0; i < FC_SIZE; i++, fcp++) {
382 		if (fcp->fc_frcn >= frcn)
383 			fcp->fc_frcn = FCE_EMPTY;
384 	}
385 }
386 
387 /*
388  * Update the FAT.
389  * If mirroring the FAT, update all copies, with the first copy as last.
390  * Else update only the current FAT (ignoring the others).
391  *
392  * pmp	 - msdosfsmount structure for filesystem to update
393  * bp	 - addr of modified FAT block
394  * fatbn - block number relative to begin of filesystem of the modified FAT block.
395  */
396 void
397 updatefats(struct msdosfsmount *pmp, struct buf *bp, u_long fatbn)
398 {
399 	int i, error;
400 	struct buf *bpn;
401 
402 	DPRINTF(("%s(pmp %p, bp %p, fatbn %lu)\n", __func__, pmp, bp, fatbn));
403 
404 	/*
405 	 * If we have an FSInfo block, update it.
406 	 */
407 	if (pmp->pm_fsinfo) {
408 		u_long cn = pmp->pm_nxtfree;
409 
410 		if (pmp->pm_freeclustercount
411 		    && (pmp->pm_inusemap[cn / N_INUSEBITS]
412 			& (1U << (cn % N_INUSEBITS)))) {
413 			/*
414 			 * The cluster indicated in FSInfo isn't free
415 			 * any longer.  Got get a new free one.
416 			 */
417 			for (cn = 0; cn < pmp->pm_maxcluster; cn++)
418 				if (pmp->pm_inusemap[cn / N_INUSEBITS] != (u_int)-1)
419 					break;
420 			pmp->pm_nxtfree = cn
421 				+ ffs(pmp->pm_inusemap[cn / N_INUSEBITS]
422 				      ^ (u_int)-1) - 1;
423 		}
424 		/*
425 		 * XXX  If the fsinfo block is stored on media with
426 		 *      2KB or larger sectors, is the fsinfo structure
427 		 *      padded at the end or in the middle?
428 		 */
429 		if (bread(pmp->pm_devvp, de_bn2kb(pmp, pmp->pm_fsinfo),
430 		    pmp->pm_BytesPerSec, B_MODIFY, &bpn) != 0) {
431 			/*
432 			 * Ignore the error, but turn off FSInfo update for the future.
433 			 */
434 			pmp->pm_fsinfo = 0;
435 		} else {
436 			struct fsinfo *fp = (struct fsinfo *)bpn->b_data;
437 
438 			putulong(fp->fsinfree, pmp->pm_freeclustercount);
439 			putulong(fp->fsinxtfree, pmp->pm_nxtfree);
440 			if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT)
441 				bwrite(bpn);
442 			else
443 				bdwrite(bpn);
444 		}
445 	}
446 
447 	if (pmp->pm_flags & MSDOSFS_FATMIRROR) {
448 		/*
449 		 * Now copy the block(s) of the modified FAT to the other copies of
450 		 * the FAT and write them out.  This is faster than reading in the
451 		 * other FATs and then writing them back out.  This could tie up
452 		 * the FAT for quite a while. Preventing others from accessing it.
453 		 * To prevent us from going after the FAT quite so much we use
454 		 * delayed writes, unless they specified "synchronous" when the
455 		 * filesystem was mounted.  If synch is asked for then use
456 		 * bwrite()'s and really slow things down.
457 		 */
458 		for (i = 1; i < pmp->pm_FATs; i++) {
459 			fatbn += pmp->pm_FATsecs;
460 			/* getblk() never fails */
461 			bpn = getblk(pmp->pm_devvp, de_bn2kb(pmp, fatbn),
462 			    bp->b_bcount, 0, 0);
463 			memcpy(bpn->b_data, bp->b_data, bp->b_bcount);
464 			if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT) {
465 				error = bwrite(bpn);
466 				if (error)
467 					printf("%s: copy FAT %d (error=%d)\n",
468 						 __func__, i, error);
469 			} else
470 				bdwrite(bpn);
471 		}
472 	}
473 
474 	/*
475 	 * Write out the first (or current) FAT last.
476 	 */
477 	if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT) {
478 		error =  bwrite(bp);
479 		if (error)
480 			printf("%s: write FAT (error=%d)\n",
481 				__func__, error);
482 	} else
483 		bdwrite(bp);
484 	/*
485 	 * Maybe update fsinfo sector here?
486 	 */
487 }
488 
489 /*
490  * Updating entries in 12 bit FATs is a pain in the butt.
491  *
492  * The following picture shows where nibbles go when moving from a 12 bit
493  * cluster number into the appropriate bytes in the FAT.
494  *
495  *	byte m        byte m+1      byte m+2
496  *	+----+----+   +----+----+   +----+----+
497  *	|  0    1 |   |  2    3 |   |  4    5 |   FAT bytes
498  *	+----+----+   +----+----+   +----+----+
499  *
500  *	+----+----+----+   +----+----+----+
501  *	|  3    0    1 |   |  4    5    2 |
502  *	+----+----+----+   +----+----+----+
503  *	cluster n  	   cluster n+1
504  *
505  * Where n is even. m = n + (n >> 2)
506  *
507  */
508 static inline void
509 usemap_alloc(struct msdosfsmount *pmp, u_long cn)
510 {
511 
512 	pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
513 	pmp->pm_freeclustercount--;
514 }
515 
516 static inline void
517 usemap_free(struct msdosfsmount *pmp, u_long cn)
518 {
519 
520 	pmp->pm_freeclustercount++;
521 	pmp->pm_inusemap[cn / N_INUSEBITS] &= ~(1U << (cn % N_INUSEBITS));
522 }
523 
524 int
525 clusterfree(struct msdosfsmount *pmp, u_long cluster, u_long *oldcnp)
526 {
527 	int error;
528 	u_long oldcn;
529 
530 	usemap_free(pmp, cluster);
531 	error = fatentry(FAT_GET_AND_SET, pmp, cluster, &oldcn, MSDOSFSFREE);
532 	if (error) {
533 		usemap_alloc(pmp, cluster);
534 		return (error);
535 	}
536 	/*
537 	 * If the cluster was successfully marked free, then update
538 	 * the count of free clusters, and turn off the "allocated"
539 	 * bit in the "in use" cluster bit map.
540 	 */
541 	if (oldcnp)
542 		*oldcnp = oldcn;
543 	return (0);
544 }
545 
546 /*
547  * Get or Set or 'Get and Set' the cluster'th entry in the FAT.
548  *
549  * function	- whether to get or set a fat entry
550  * pmp		- address of the msdosfsmount structure for the filesystem
551  *		  whose FAT is to be manipulated.
552  * cn		- which cluster is of interest
553  * oldcontents	- address of a word that is to receive the contents of the
554  *		  cluster'th entry if this is a get function
555  * newcontents	- the new value to be written into the cluster'th element of
556  *		  the FAT if this is a set function.
557  *
558  * This function can also be used to free a cluster by setting the FAT entry
559  * for a cluster to 0.
560  *
561  * All copies of the FAT are updated if this is a set function. NOTE: If
562  * fatentry() marks a cluster as free it does not update the inusemap in
563  * the msdosfsmount structure. This is left to the caller.
564  */
565 int
566 fatentry(int function, struct msdosfsmount *pmp, u_long cn, u_long *oldcontents, u_long newcontents)
567 {
568 	int error;
569 	u_long readcn;
570 	u_long bn, bo, bsize, byteoffset;
571 	struct buf *bp;
572 
573 	DPRINTF(("%s(func %d, pmp %p, clust %lu, oldcon %p, newcon " "%lx)\n",
574 	    __func__, function, pmp, cn, oldcontents, newcontents));
575 
576 #ifdef DIAGNOSTIC
577 	/*
578 	 * Be sure they asked us to do something.
579 	 */
580 	if ((function & (FAT_SET | FAT_GET)) == 0) {
581 		DPRINTF(("%s(): function code doesn't specify get or set\n",
582 		    __func__));
583 		return (EINVAL);
584 	}
585 
586 	/*
587 	 * If they asked us to return a cluster number but didn't tell us
588 	 * where to put it, give them an error.
589 	 */
590 	if ((function & FAT_GET) && oldcontents == NULL) {
591 		DPRINTF(("%s(): get function with no place to put result\n",
592 			__func__));
593 		return (EINVAL);
594 	}
595 #endif
596 
597 	/*
598 	 * Be sure the requested cluster is in the filesystem.
599 	 */
600 	if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster)
601 		return (EINVAL);
602 
603 	byteoffset = FATOFS(pmp, cn);
604 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
605 	if ((error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
606 	    0, &bp)) != 0) {
607 		return (error);
608 	}
609 
610 	if (function & FAT_GET) {
611 		if (FAT32(pmp))
612 			readcn = getulong((char *)bp->b_data + bo);
613 		else
614 			readcn = getushort((char *)bp->b_data + bo);
615 		if (FAT12(pmp) & (cn & 1))
616 			readcn >>= 4;
617 		readcn &= pmp->pm_fatmask;
618 		*oldcontents = readcn;
619 	}
620 	if (function & FAT_SET) {
621 		switch (pmp->pm_fatmask) {
622 		case FAT12_MASK:
623 			readcn = getushort((char *)bp->b_data + bo);
624 			if (cn & 1) {
625 				readcn &= 0x000f;
626 				readcn |= newcontents << 4;
627 			} else {
628 				readcn &= 0xf000;
629 				readcn |= newcontents & 0xfff;
630 			}
631 			putushort((char *)bp->b_data + bo, readcn);
632 			break;
633 		case FAT16_MASK:
634 			putushort((char *)bp->b_data + bo, newcontents);
635 			break;
636 		case FAT32_MASK:
637 			/*
638 			 * According to spec we have to retain the
639 			 * high order bits of the FAT entry.
640 			 */
641 			readcn = getulong((char *)bp->b_data + bo);
642 			readcn &= ~FAT32_MASK;
643 			readcn |= newcontents & FAT32_MASK;
644 			putulong((char *)bp->b_data + bo, readcn);
645 			break;
646 		}
647 		updatefats(pmp, bp, bn);
648 		bp = NULL;
649 		pmp->pm_fmod = 1;
650 	}
651 	if (bp)
652 		brelse(bp, 0);
653 	return (0);
654 }
655 
656 /*
657  * Update a contiguous cluster chain
658  *
659  * pmp	    - mount point
660  * start    - first cluster of chain
661  * count    - number of clusters in chain
662  * fillwith - what to write into FAT entry of last cluster
663  */
664 static int
665 fatchain(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith)
666 {
667 	int error;
668 	u_long bn, bo, bsize, byteoffset, readcn, newc;
669 	struct buf *bp;
670 
671 	DPRINTF(("%s(pmp %p, start %lu, count %lu, fillwith %lx)\n", __func__,
672 	    pmp, start, count, fillwith));
673 	/*
674 	 * Be sure the clusters are in the filesystem.
675 	 */
676 	if (start < CLUST_FIRST || start + count - 1 > pmp->pm_maxcluster)
677 		return (EINVAL);
678 
679 	while (count > 0) {
680 		byteoffset = FATOFS(pmp, start);
681 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
682 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
683 		    B_MODIFY, &bp);
684 		if (error) {
685 			return (error);
686 		}
687 		while (count > 0) {
688 			start++;
689 			newc = --count > 0 ? start : fillwith;
690 			switch (pmp->pm_fatmask) {
691 			case FAT12_MASK:
692 				readcn = getushort((char *)bp->b_data + bo);
693 				if (start & 1) {
694 					readcn &= 0xf000;
695 					readcn |= newc & 0xfff;
696 				} else {
697 					readcn &= 0x000f;
698 					readcn |= newc << 4;
699 				}
700 				putushort((char *)bp->b_data + bo, readcn);
701 				bo++;
702 				if (!(start & 1))
703 					bo++;
704 				break;
705 			case FAT16_MASK:
706 				putushort((char *)bp->b_data + bo, newc);
707 				bo += 2;
708 				break;
709 			case FAT32_MASK:
710 				readcn = getulong((char *)bp->b_data + bo);
711 				readcn &= ~pmp->pm_fatmask;
712 				readcn |= newc & pmp->pm_fatmask;
713 				putulong((char *)bp->b_data + bo, readcn);
714 				bo += 4;
715 				break;
716 			}
717 			if (bo >= bsize)
718 				break;
719 		}
720 		updatefats(pmp, bp, bn);
721 	}
722 	pmp->pm_fmod = 1;
723 	return (0);
724 }
725 
726 /*
727  * Check the length of a free cluster chain starting at start.
728  *
729  * pmp	 - mount point
730  * start - start of chain
731  * count - maximum interesting length
732  */
733 int
734 chainlength(struct msdosfsmount *pmp, u_long start, u_long count)
735 {
736 	u_long idx, max_idx;
737 	u_int map;
738 	u_long len;
739 
740 	max_idx = pmp->pm_maxcluster / N_INUSEBITS;
741 	idx = start / N_INUSEBITS;
742 	start %= N_INUSEBITS;
743 	map = pmp->pm_inusemap[idx];
744 	map &= ~((1U << start) - 1);
745 	if (map) {
746 		len = ffs(map) - 1 - start;
747 		return (len > count ? count : len);
748 	}
749 	len = N_INUSEBITS - start;
750 	if (len >= count)
751 		return (count);
752 	while (++idx <= max_idx) {
753 		if (len >= count)
754 			break;
755 		if ((map = pmp->pm_inusemap[idx]) != 0) {
756 			len +=  ffs(map) - 1;
757 			break;
758 		}
759 		len += N_INUSEBITS;
760 	}
761 	return (len > count ? count : len);
762 }
763 
764 /*
765  * Allocate contigous free clusters.
766  *
767  * pmp	      - mount point.
768  * start      - start of cluster chain.
769  * count      - number of clusters to allocate.
770  * fillwith   - put this value into the FAT entry for the
771  *		last allocated cluster.
772  * retcluster - put the first allocated cluster's number here.
773  * got	      - how many clusters were actually allocated.
774  */
775 int
776 chainalloc(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith, u_long *retcluster, u_long *got)
777 {
778 	int error;
779 	u_long cl, n;
780 
781 	for (cl = start, n = count; n-- > 0;)
782 		usemap_alloc(pmp, cl++);
783 	if ((error = fatchain(pmp, start, count, fillwith)) != 0)
784 		return (error);
785 
786 	DPRINTF(("%s(): allocated cluster chain at %lu (%lu clusters)\n",
787 	    __func__, start, count));
788 	if (retcluster)
789 		*retcluster = start;
790 	if (got)
791 		*got = count;
792 	return (0);
793 }
794 
795 /*
796  * Allocate contiguous free clusters.
797  *
798  * pmp	      - mount point.
799  * start      - preferred start of cluster chain.
800  * count      - number of clusters requested.
801  * fillwith   - put this value into the FAT entry for the
802  *		last allocated cluster.
803  * retcluster - put the first allocated cluster's number here.
804  * got	      - how many clusters were actually allocated.
805  */
806 int
807 clusteralloc(struct msdosfsmount *pmp, u_long start, u_long count, u_long *retcluster, u_long *got)
808 {
809 	u_long idx;
810 	u_long len, newst, foundl, cn, l;
811 	u_long foundcn = 0; /* XXX: foundcn could be used unititialized */
812 	u_long fillwith = CLUST_EOFE;
813 	u_int map;
814 
815 	DPRINTF(("%s(): find %lu clusters\n", __func__, count));
816 	if (start) {
817 		if ((len = chainlength(pmp, start, count)) >= count)
818 			return (chainalloc(pmp, start, count, fillwith, retcluster, got));
819 	} else {
820 		/*
821 		 * This is a new file, initialize start
822 		 */
823 		struct timeval tv;
824 
825 		microtime(&tv);
826 		start = (tv.tv_usec >> 10) | tv.tv_usec;
827 		len = 0;
828 	}
829 
830 	/*
831 	 * Start at a (pseudo) random place to maximize cluster runs
832 	 * under multiple writers.
833 	 */
834 	newst = (start * 1103515245 + 12345) % (pmp->pm_maxcluster + 1);
835 	foundl = 0;
836 
837 	for (cn = newst; cn <= pmp->pm_maxcluster;) {
838 		idx = cn / N_INUSEBITS;
839 		map = pmp->pm_inusemap[idx];
840 		map |= (1U << (cn % N_INUSEBITS)) - 1;
841 		if (map != (u_int)-1) {
842 			cn = idx * N_INUSEBITS + ffs(map^(u_int)-1) - 1;
843 			if ((l = chainlength(pmp, cn, count)) >= count)
844 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
845 			if (l > foundl) {
846 				foundcn = cn;
847 				foundl = l;
848 			}
849 			cn += l + 1;
850 			continue;
851 		}
852 		cn += N_INUSEBITS - cn % N_INUSEBITS;
853 	}
854 	for (cn = 0; cn < newst;) {
855 		idx = cn / N_INUSEBITS;
856 		map = pmp->pm_inusemap[idx];
857 		map |= (1U << (cn % N_INUSEBITS)) - 1;
858 		if (map != (u_int)-1) {
859 			cn = idx * N_INUSEBITS + ffs(map^(u_int)-1) - 1;
860 			if ((l = chainlength(pmp, cn, count)) >= count)
861 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
862 			if (l > foundl) {
863 				foundcn = cn;
864 				foundl = l;
865 			}
866 			cn += l + 1;
867 			continue;
868 		}
869 		cn += N_INUSEBITS - cn % N_INUSEBITS;
870 	}
871 
872 	if (!foundl)
873 		return (ENOSPC);
874 
875 	if (len)
876 		return (chainalloc(pmp, start, len, fillwith, retcluster, got));
877 	else
878 		return (chainalloc(pmp, foundcn, foundl, fillwith, retcluster, got));
879 }
880 
881 
882 /*
883  * Free a chain of clusters.
884  *
885  * pmp		- address of the msdosfs mount structure for the filesystem
886  *		  containing the cluster chain to be freed.
887  * startcluster - number of the 1st cluster in the chain of clusters to be
888  *		  freed.
889  */
890 int
891 freeclusterchain(struct msdosfsmount *pmp, u_long cluster)
892 {
893 	int error;
894 	struct buf *bp = NULL;
895 	u_long bn, bo, bsize, byteoffset;
896 	u_long readcn, lbn = -1;
897 
898 	bn = 0; /* XXXgcc */
899 	while (cluster >= CLUST_FIRST && cluster <= pmp->pm_maxcluster) {
900 		byteoffset = FATOFS(pmp, cluster);
901 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
902 		if (lbn != bn) {
903 			if (bp)
904 				updatefats(pmp, bp, lbn);
905 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
906 			    B_MODIFY, &bp);
907 			if (error) {
908 				return (error);
909 			}
910 			lbn = bn;
911 		}
912 		usemap_free(pmp, cluster);
913 		KASSERT(bp != NULL);
914 		switch (pmp->pm_fatmask) {
915 		case FAT12_MASK:
916 			readcn = getushort((char *)bp->b_data + bo);
917 			if (cluster & 1) {
918 				cluster = readcn >> 4;
919 				readcn &= 0x000f;
920 				readcn |= MSDOSFSFREE << 4;
921 			} else {
922 				cluster = readcn;
923 				readcn &= 0xf000;
924 				readcn |= MSDOSFSFREE & 0xfff;
925 			}
926 			putushort((char *)bp->b_data + bo, readcn);
927 			break;
928 		case FAT16_MASK:
929 			cluster = getushort((char *)bp->b_data + bo);
930 			putushort((char *)bp->b_data + bo, MSDOSFSFREE);
931 			break;
932 		case FAT32_MASK:
933 			cluster = getulong((char *)bp->b_data + bo);
934 			putulong((char *)bp->b_data + bo,
935 				 (MSDOSFSFREE & FAT32_MASK) | (cluster & ~FAT32_MASK));
936 			break;
937 		}
938 		cluster &= pmp->pm_fatmask;
939 	}
940 	if (bp)
941 		updatefats(pmp, bp, bn);
942 	return (0);
943 }
944 
945 /*
946  * Read in FAT blocks looking for free clusters. For every free cluster
947  * found turn off its corresponding bit in the pm_inusemap.
948  */
949 int
950 fillinusemap(struct msdosfsmount *pmp)
951 {
952 	struct buf *bp = NULL;
953 	u_long cn, readcn;
954 	int error;
955 	u_long bn, bo, bsize, byteoffset;
956 
957 	/*
958 	 * Mark all clusters in use, we mark the free ones in the FAT scan
959 	 * loop further down.
960 	 */
961 	for (cn = 0; cn < (pmp->pm_maxcluster + N_INUSEBITS) / N_INUSEBITS; cn++)
962 		pmp->pm_inusemap[cn] = (u_int)-1;
963 
964 	/*
965 	 * Figure how many free clusters are in the filesystem by ripping
966 	 * through the FAT counting the number of entries whose content is
967 	 * zero.  These represent free clusters.
968 	 */
969 	pmp->pm_freeclustercount = 0;
970 	for (cn = CLUST_FIRST; cn <= pmp->pm_maxcluster; cn++) {
971 		byteoffset = FATOFS(pmp, cn);
972 		bo = byteoffset % pmp->pm_fatblocksize;
973 		if (!bo || !bp) {
974 			/* Read new FAT block */
975 			if (bp)
976 				brelse(bp, 0);
977 			fatblock(pmp, byteoffset, &bn, &bsize, NULL);
978 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
979 			    0, &bp);
980 			if (error) {
981 				return (error);
982 			}
983 		}
984 		if (FAT32(pmp))
985 			readcn = getulong((char *)bp->b_data + bo);
986 		else
987 			readcn = getushort((char *)bp->b_data + bo);
988 		if (FAT12(pmp) && (cn & 1))
989 			readcn >>= 4;
990 		readcn &= pmp->pm_fatmask;
991 
992 		if (readcn == 0)
993 			usemap_free(pmp, cn);
994 	}
995 	if (bp)
996 		brelse(bp, 0);
997 	return (0);
998 }
999 
1000 /*
1001  * Allocate a new cluster and chain it onto the end of the file.
1002  *
1003  * dep	 - the file to extend
1004  * count - number of clusters to allocate
1005  * bpp	 - where to return the address of the buf header for the first new
1006  *	   file block
1007  * ncp	 - where to put cluster number of the first newly allocated cluster
1008  *	   If this pointer is 0, do not return the cluster number.
1009  * flags - see fat.h
1010  *
1011  * NOTE: This function is not responsible for turning on the DE_UPDATE bit of
1012  * the de_flag field of the denode and it does not change the de_FileSize
1013  * field.  This is left for the caller to do.
1014  */
1015 
1016 int
1017 extendfile(struct denode *dep, u_long count, struct buf **bpp, u_long *ncp, int flags)
1018 {
1019 	int error;
1020 	u_long frcn = 0, cn, got;
1021 	struct msdosfsmount *pmp = dep->de_pmp;
1022 	struct buf *bp;
1023 
1024 	/*
1025 	 * Don't try to extend the root directory
1026 	 */
1027 	if (dep->de_StartCluster == MSDOSFSROOT
1028 	    && (dep->de_Attributes & ATTR_DIRECTORY)) {
1029 		DPRINTF(("%s(): attempt to extend root directory\n", __func__));
1030 		return (ENOSPC);
1031 	}
1032 
1033 	/*
1034 	 * If the "file's last cluster" cache entry is empty, and the file
1035 	 * is not empty, then fill the cache entry by calling pcbmap().
1036 	 */
1037 	fc_fileextends++;
1038 	if (dep->de_fc[FC_LASTFC].fc_frcn == FCE_EMPTY &&
1039 	    dep->de_StartCluster != 0) {
1040 		fc_lfcempty++;
1041 		error = pcbmap(dep, CLUST_END, 0, &cn, 0);
1042 		/* we expect it to return E2BIG */
1043 		if (error != E2BIG)
1044 			return (error);
1045 	}
1046 
1047 	fc_last_to_nexttolast(dep);
1048 
1049 	while (count > 0) {
1050 
1051 		/*
1052 		 * Allocate a new cluster chain and cat onto the end of the
1053 		 * file.  If the file is empty we make de_StartCluster point
1054 		 * to the new block.  Note that de_StartCluster being 0 is
1055 		 * sufficient to be sure the file is empty since we exclude
1056 		 * attempts to extend the root directory above, and the root
1057 		 * dir is the only file with a startcluster of 0 that has
1058 		 * blocks allocated (sort of).
1059 		 */
1060 
1061 		if (dep->de_StartCluster == 0)
1062 			cn = 0;
1063 		else
1064 			cn = dep->de_fc[FC_LASTFC].fc_fsrcn + 1;
1065 		error = clusteralloc(pmp, cn, count, &cn, &got);
1066 		if (error)
1067 			return (error);
1068 
1069 		count -= got;
1070 
1071 		/*
1072 		 * Give them the filesystem relative cluster number if they want
1073 		 * it.
1074 		 */
1075 		if (ncp) {
1076 			*ncp = cn;
1077 			ncp = NULL;
1078 		}
1079 
1080 		if (dep->de_StartCluster == 0) {
1081 			dep->de_StartCluster = cn;
1082 			frcn = 0;
1083 		} else {
1084 			error = fatentry(FAT_SET, pmp,
1085 					 dep->de_fc[FC_LASTFC].fc_fsrcn,
1086 					 0, cn);
1087 			if (error) {
1088 				clusterfree(pmp, cn, NULL);
1089 				return (error);
1090 			}
1091 			frcn = dep->de_fc[FC_LASTFC].fc_frcn + 1;
1092 		}
1093 
1094 		/*
1095 		 * Update the "last cluster of the file" entry in the
1096 		 * denode's FAT cache.
1097 		 */
1098 
1099 		fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
1100 		if ((flags & DE_CLEAR) &&
1101 		    (dep->de_Attributes & ATTR_DIRECTORY)) {
1102 			while (got-- > 0) {
1103 				bp = getblk(pmp->pm_devvp,
1104 				    de_bn2kb(pmp, cntobn(pmp, cn++)),
1105 				    pmp->pm_bpcluster, 0, 0);
1106 				clrbuf(bp);
1107 				if (bpp) {
1108 					*bpp = bp;
1109 						bpp = NULL;
1110 				} else {
1111 					bdwrite(bp);
1112 				}
1113 			}
1114 		}
1115 	}
1116 
1117 	return (0);
1118 }
1119