xref: /freebsd/sys/fs/msdosfs/msdosfs_fat.c (revision 53e1cbefe4b81e6ecdad529fcd252600f838cf69)
1 /* $FreeBSD$ */
2 /*	$NetBSD: msdosfs_fat.c,v 1.28 1997/11/17 15:36:49 ws Exp $	*/
3 
4 /*-
5  * SPDX-License-Identifier: BSD-4-Clause
6  *
7  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
8  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
9  * All rights reserved.
10  * Original code by Paul Popelka (paulp@uts.amdahl.com) (see below).
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by TooLs GmbH.
23  * 4. The name of TooLs GmbH may not be used to endorse or promote products
24  *    derived from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
27  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
28  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
29  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
30  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
31  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
32  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
33  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
34  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
35  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36  */
37 /*-
38  * Written by Paul Popelka (paulp@uts.amdahl.com)
39  *
40  * You can do anything you want with this software, just don't say you wrote
41  * it, and don't remove this notice.
42  *
43  * This software is provided "as is".
44  *
45  * The author supplies this software to be publicly redistributed on the
46  * understanding that the author is not responsible for the correct
47  * functioning of this software in any circumstances and is not liable for
48  * any damages caused by this software.
49  *
50  * October 1992
51  */
52 
53 #include <sys/param.h>
54 #include <sys/systm.h>
55 #include <sys/buf.h>
56 #include <sys/mount.h>
57 #include <sys/vmmeter.h>
58 #include <sys/vnode.h>
59 
60 #include <fs/msdosfs/bpb.h>
61 #include <fs/msdosfs/direntry.h>
62 #include <fs/msdosfs/denode.h>
63 #include <fs/msdosfs/fat.h>
64 #include <fs/msdosfs/msdosfsmount.h>
65 
66 #define	FULL_RUN	((u_int)0xffffffff)
67 
68 static int	chainalloc(struct msdosfsmount *pmp, u_long start,
69 		    u_long count, u_long fillwith, u_long *retcluster,
70 		    u_long *got);
71 static int	chainlength(struct msdosfsmount *pmp, u_long start,
72 		    u_long count);
73 static void	fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp,
74 		    u_long *sizep, u_long *bop);
75 static int	fatchain(struct msdosfsmount *pmp, u_long start, u_long count,
76 		    u_long fillwith);
77 static void	fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp,
78 		    u_long *fsrcnp);
79 static void	updatefats(struct msdosfsmount *pmp, struct buf *bp,
80 		    u_long fatbn);
81 static __inline void
82 		usemap_alloc(struct msdosfsmount *pmp, u_long cn);
83 static int	usemap_free(struct msdosfsmount *pmp, u_long cn);
84 static int	clusteralloc1(struct msdosfsmount *pmp, u_long start,
85 		    u_long count, u_long fillwith, u_long *retcluster,
86 		    u_long *got);
87 
88 static void
89 fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp, u_long *sizep,
90     u_long *bop)
91 {
92 	u_long bn, size;
93 
94 	bn = ofs / pmp->pm_fatblocksize * pmp->pm_fatblocksec;
95 	size = min(pmp->pm_fatblocksec, pmp->pm_FATsecs - bn)
96 	    * DEV_BSIZE;
97 	bn += pmp->pm_fatblk + pmp->pm_curfat * pmp->pm_FATsecs;
98 
99 	if (bnp)
100 		*bnp = bn;
101 	if (sizep)
102 		*sizep = size;
103 	if (bop)
104 		*bop = ofs % pmp->pm_fatblocksize;
105 }
106 
107 /*
108  * Map the logical cluster number of a file into a physical disk sector
109  * that is filesystem relative.
110  *
111  * dep	  - address of denode representing the file of interest
112  * findcn - file relative cluster whose filesystem relative cluster number
113  *	    and/or block number are/is to be found
114  * bnp	  - address of where to place the filesystem relative block number.
115  *	    If this pointer is null then don't return this quantity.
116  * cnp	  - address of where to place the filesystem relative cluster number.
117  *	    If this pointer is null then don't return this quantity.
118  * sp     - pointer to returned block size
119  *
120  * NOTE: Either bnp or cnp must be non-null.
121  * This function has one side effect.  If the requested file relative cluster
122  * is beyond the end of file, then the actual number of clusters in the file
123  * is returned in *cnp.  This is useful for determining how long a directory is.
124  *  If cnp is null, nothing is returned.
125  */
126 int
127 pcbmap(struct denode *dep, u_long findcn, daddr_t *bnp, u_long *cnp, int *sp)
128 {
129 	int error;
130 	u_long i;
131 	u_long cn;
132 	u_long prevcn = 0; /* XXX: prevcn could be used unititialized */
133 	u_long byteoffset;
134 	u_long bn;
135 	u_long bo;
136 	struct buf *bp = NULL;
137 	u_long bp_bn = -1;
138 	struct msdosfsmount *pmp = dep->de_pmp;
139 	u_long bsize;
140 
141 	KASSERT(bnp != NULL || cnp != NULL || sp != NULL,
142 	    ("pcbmap: extra call"));
143 	ASSERT_VOP_ELOCKED(DETOV(dep), "pcbmap");
144 
145 	cn = dep->de_StartCluster;
146 	/*
147 	 * The "file" that makes up the root directory is contiguous,
148 	 * permanently allocated, of fixed size, and is not made up of
149 	 * clusters.  If the cluster number is beyond the end of the root
150 	 * directory, then return the number of clusters in the file.
151 	 */
152 	if (cn == MSDOSFSROOT) {
153 		if (dep->de_Attributes & ATTR_DIRECTORY) {
154 			if (de_cn2off(pmp, findcn) >= dep->de_FileSize) {
155 				if (cnp)
156 					*cnp = de_bn2cn(pmp, pmp->pm_rootdirsize);
157 				return (E2BIG);
158 			}
159 			if (bnp)
160 				*bnp = pmp->pm_rootdirblk + de_cn2bn(pmp, findcn);
161 			if (cnp)
162 				*cnp = MSDOSFSROOT;
163 			if (sp)
164 				*sp = min(pmp->pm_bpcluster,
165 				    dep->de_FileSize - de_cn2off(pmp, findcn));
166 			return (0);
167 		} else {		/* just an empty file */
168 			if (cnp)
169 				*cnp = 0;
170 			return (E2BIG);
171 		}
172 	}
173 
174 	/*
175 	 * All other files do I/O in cluster sized blocks
176 	 */
177 	if (sp)
178 		*sp = pmp->pm_bpcluster;
179 
180 	/*
181 	 * Rummage around in the FAT cache, maybe we can avoid tromping
182 	 * through every FAT entry for the file. And, keep track of how far
183 	 * off the cache was from where we wanted to be.
184 	 */
185 	i = 0;
186 	fc_lookup(dep, findcn, &i, &cn);
187 
188 	/*
189 	 * Handle all other files or directories the normal way.
190 	 */
191 	for (; i < findcn; i++) {
192 		/*
193 		 * Stop with all reserved clusters, not just with EOF.
194 		 */
195 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
196 			goto hiteof;
197 		byteoffset = FATOFS(pmp, cn);
198 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
199 		if (bn != bp_bn) {
200 			if (bp)
201 				brelse(bp);
202 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
203 			if (error) {
204 				return (error);
205 			}
206 			bp_bn = bn;
207 		}
208 		prevcn = cn;
209 		if (bo >= bsize) {
210 			if (bp)
211 				brelse(bp);
212 			return (EIO);
213 		}
214 		if (FAT32(pmp))
215 			cn = getulong(bp->b_data + bo);
216 		else
217 			cn = getushort(bp->b_data + bo);
218 		if (FAT12(pmp) && (prevcn & 1))
219 			cn >>= 4;
220 		cn &= pmp->pm_fatmask;
221 
222 		/*
223 		 * Force the special cluster numbers
224 		 * to be the same for all cluster sizes
225 		 * to let the rest of msdosfs handle
226 		 * all cases the same.
227 		 */
228 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
229 			cn |= ~pmp->pm_fatmask;
230 	}
231 
232 	if (!MSDOSFSEOF(pmp, cn)) {
233 		if (bp)
234 			brelse(bp);
235 		if (bnp)
236 			*bnp = cntobn(pmp, cn);
237 		if (cnp)
238 			*cnp = cn;
239 		fc_setcache(dep, FC_LASTMAP, i, cn);
240 		return (0);
241 	}
242 
243 hiteof:;
244 	if (cnp)
245 		*cnp = i;
246 	if (bp)
247 		brelse(bp);
248 	/* update last file cluster entry in the FAT cache */
249 	fc_setcache(dep, FC_LASTFC, i - 1, prevcn);
250 	return (E2BIG);
251 }
252 
253 /*
254  * Find the closest entry in the FAT cache to the cluster we are looking
255  * for.
256  */
257 static void
258 fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp, u_long *fsrcnp)
259 {
260 	int i;
261 	u_long cn;
262 	struct fatcache *closest = NULL;
263 
264 	ASSERT_VOP_LOCKED(DETOV(dep), "fc_lookup");
265 
266 	for (i = 0; i < FC_SIZE; i++) {
267 		cn = dep->de_fc[i].fc_frcn;
268 		if (cn != FCE_EMPTY && cn <= findcn) {
269 			if (closest == NULL || cn > closest->fc_frcn)
270 				closest = &dep->de_fc[i];
271 		}
272 	}
273 	if (closest) {
274 		*frcnp = closest->fc_frcn;
275 		*fsrcnp = closest->fc_fsrcn;
276 	}
277 }
278 
279 /*
280  * Purge the FAT cache in denode dep of all entries relating to file
281  * relative cluster frcn and beyond.
282  */
283 void
284 fc_purge(struct denode *dep, u_int frcn)
285 {
286 	int i;
287 	struct fatcache *fcp;
288 
289 	ASSERT_VOP_ELOCKED(DETOV(dep), "fc_purge");
290 
291 	fcp = dep->de_fc;
292 	for (i = 0; i < FC_SIZE; i++, fcp++) {
293 		if (fcp->fc_frcn >= frcn)
294 			fcp->fc_frcn = FCE_EMPTY;
295 	}
296 }
297 
298 /*
299  * Update the FAT.
300  * If mirroring the FAT, update all copies, with the first copy as last.
301  * Else update only the current FAT (ignoring the others).
302  *
303  * pmp	 - msdosfsmount structure for filesystem to update
304  * bp	 - addr of modified FAT block
305  * fatbn - block number relative to begin of filesystem of the modified FAT block.
306  */
307 static void
308 updatefats(struct msdosfsmount *pmp, struct buf *bp, u_long fatbn)
309 {
310 	struct buf *bpn;
311 	int cleanfat, i;
312 
313 #ifdef MSDOSFS_DEBUG
314 	printf("updatefats(pmp %p, bp %p, fatbn %lu)\n", pmp, bp, fatbn);
315 #endif
316 
317 	if (pmp->pm_flags & MSDOSFS_FATMIRROR) {
318 		/*
319 		 * Now copy the block(s) of the modified FAT to the other copies of
320 		 * the FAT and write them out.  This is faster than reading in the
321 		 * other FATs and then writing them back out.  This could tie up
322 		 * the FAT for quite a while. Preventing others from accessing it.
323 		 * To prevent us from going after the FAT quite so much we use
324 		 * delayed writes, unless they specified "synchronous" when the
325 		 * filesystem was mounted.  If synch is asked for then use
326 		 * bwrite()'s and really slow things down.
327 		 */
328 		if (fatbn != pmp->pm_fatblk || FAT12(pmp))
329 			cleanfat = 0;
330 		else if (FAT16(pmp))
331 			cleanfat = 16;
332 		else
333 			cleanfat = 32;
334 		for (i = 1; i < pmp->pm_FATs; i++) {
335 			fatbn += pmp->pm_FATsecs;
336 			/* getblk() never fails */
337 			bpn = getblk(pmp->pm_devvp, fatbn, bp->b_bcount,
338 			    0, 0, 0);
339 			memcpy(bpn->b_data, bp->b_data, bp->b_bcount);
340 			/* Force the clean bit on in the other copies. */
341 			if (cleanfat == 16)
342 				((uint8_t *)bpn->b_data)[3] |= 0x80;
343 			else if (cleanfat == 32)
344 				((uint8_t *)bpn->b_data)[7] |= 0x08;
345 			if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
346 				bwrite(bpn);
347 			else
348 				bdwrite(bpn);
349 		}
350 	}
351 
352 	/*
353 	 * Write out the first (or current) FAT last.
354 	 */
355 	if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
356 		bwrite(bp);
357 	else
358 		bdwrite(bp);
359 }
360 
361 /*
362  * Updating entries in 12 bit FATs is a pain in the butt.
363  *
364  * The following picture shows where nibbles go when moving from a 12 bit
365  * cluster number into the appropriate bytes in the FAT.
366  *
367  *	byte m        byte m+1      byte m+2
368  *	+----+----+   +----+----+   +----+----+
369  *	|  0    1 |   |  2    3 |   |  4    5 |   FAT bytes
370  *	+----+----+   +----+----+   +----+----+
371  *
372  *	+----+----+----+   +----+----+----+
373  *	|  3    0    1 |   |  4    5    2 |
374  *	+----+----+----+   +----+----+----+
375  *	cluster n  	   cluster n+1
376  *
377  * Where n is even. m = n + (n >> 2)
378  *
379  */
380 static __inline void
381 usemap_alloc(struct msdosfsmount *pmp, u_long cn)
382 {
383 
384 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
385 
386 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
387 	    pmp->pm_maxcluster));
388 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
389 	    ("usemap_alloc on ro msdosfs mount"));
390 	KASSERT((pmp->pm_inusemap[cn / N_INUSEBITS] &
391 	    (1U << (cn % N_INUSEBITS))) == 0,
392 	    ("Allocating used sector %ld %ld %x", cn, cn % N_INUSEBITS,
393 	    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]));
394 	pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
395 	KASSERT(pmp->pm_freeclustercount > 0, ("usemap_alloc: too little"));
396 	pmp->pm_freeclustercount--;
397 	pmp->pm_flags |= MSDOSFS_FSIMOD;
398 }
399 
400 static int
401 usemap_free(struct msdosfsmount *pmp, u_long cn)
402 {
403 
404 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
405 
406 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
407 	    pmp->pm_maxcluster));
408 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
409 	    ("usemap_free on ro msdosfs mount"));
410 	if ((pmp->pm_inusemap[cn / N_INUSEBITS] &
411 	    (1U << (cn % N_INUSEBITS))) == 0) {
412 		printf("%s: Freeing unused sector %ld %ld %x\n",
413 		    pmp->pm_mountp->mnt_stat.f_mntonname, cn, cn % N_INUSEBITS,
414 		    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]);
415 		msdosfs_integrity_error(pmp);
416 		return (EINTEGRITY);
417 	}
418 	pmp->pm_freeclustercount++;
419 	pmp->pm_flags |= MSDOSFS_FSIMOD;
420 	pmp->pm_inusemap[cn / N_INUSEBITS] &= ~(1U << (cn % N_INUSEBITS));
421 	return (0);
422 }
423 
424 void
425 clusterfree(struct msdosfsmount *pmp, u_long cluster)
426 {
427 	int error;
428 	u_long oldcn;
429 
430 	error = fatentry(FAT_GET_AND_SET, pmp, cluster, &oldcn, MSDOSFSFREE);
431 	if (error != 0)
432 		return;
433 	/*
434 	 * If the cluster was successfully marked free, then update
435 	 * the count of free clusters, and turn off the "allocated"
436 	 * bit in the "in use" cluster bit map.
437 	 */
438 	MSDOSFS_LOCK_MP(pmp);
439 	error = usemap_free(pmp, cluster);
440 	MSDOSFS_UNLOCK_MP(pmp);
441 }
442 
443 /*
444  * Get or Set or 'Get and Set' the cluster'th entry in the FAT.
445  *
446  * function	- whether to get or set a FAT entry
447  * pmp		- address of the msdosfsmount structure for the filesystem
448  *		  whose FAT is to be manipulated.
449  * cn		- which cluster is of interest
450  * oldcontents	- address of a word that is to receive the contents of the
451  *		  cluster'th entry if this is a get function
452  * newcontents	- the new value to be written into the cluster'th element of
453  *		  the FAT if this is a set function.
454  *
455  * This function can also be used to free a cluster by setting the FAT entry
456  * for a cluster to 0.
457  *
458  * All copies of the FAT are updated if this is a set function. NOTE: If
459  * fatentry() marks a cluster as free it does not update the inusemap in
460  * the msdosfsmount structure. This is left to the caller.
461  */
462 int
463 fatentry(int function, struct msdosfsmount *pmp, u_long cn, u_long *oldcontents,
464     u_long newcontents)
465 {
466 	int error;
467 	u_long readcn;
468 	u_long bn, bo, bsize, byteoffset;
469 	struct buf *bp;
470 
471 #ifdef	MSDOSFS_DEBUG
472 	printf("fatentry(func %d, pmp %p, clust %lu, oldcon %p, newcon %lx)\n",
473 	    function, pmp, cn, oldcontents, newcontents);
474 #endif
475 
476 #ifdef DIAGNOSTIC
477 	/*
478 	 * Be sure they asked us to do something.
479 	 */
480 	if ((function & (FAT_SET | FAT_GET)) == 0) {
481 #ifdef MSDOSFS_DEBUG
482 		printf("fatentry(): function code doesn't specify get or set\n");
483 #endif
484 		return (EINVAL);
485 	}
486 
487 	/*
488 	 * If they asked us to return a cluster number but didn't tell us
489 	 * where to put it, give them an error.
490 	 */
491 	if ((function & FAT_GET) && oldcontents == NULL) {
492 #ifdef MSDOSFS_DEBUG
493 		printf("fatentry(): get function with no place to put result\n");
494 #endif
495 		return (EINVAL);
496 	}
497 #endif
498 
499 	/*
500 	 * Be sure the requested cluster is in the filesystem.
501 	 */
502 	if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster)
503 		return (EINVAL);
504 
505 	byteoffset = FATOFS(pmp, cn);
506 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
507 	error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
508 	if (error) {
509 		return (error);
510 	}
511 
512 	if (function & FAT_GET) {
513 		if (FAT32(pmp))
514 			readcn = getulong(bp->b_data + bo);
515 		else
516 			readcn = getushort(bp->b_data + bo);
517 		if (FAT12(pmp) & (cn & 1))
518 			readcn >>= 4;
519 		readcn &= pmp->pm_fatmask;
520 		/* map reserved FAT entries to same values for all FATs */
521 		if ((readcn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
522 			readcn |= ~pmp->pm_fatmask;
523 		*oldcontents = readcn;
524 	}
525 	if (function & FAT_SET) {
526 		switch (pmp->pm_fatmask) {
527 		case FAT12_MASK:
528 			readcn = getushort(bp->b_data + bo);
529 			if (cn & 1) {
530 				readcn &= 0x000f;
531 				readcn |= newcontents << 4;
532 			} else {
533 				readcn &= 0xf000;
534 				readcn |= newcontents & 0xfff;
535 			}
536 			putushort(bp->b_data + bo, readcn);
537 			break;
538 		case FAT16_MASK:
539 			putushort(bp->b_data + bo, newcontents);
540 			break;
541 		case FAT32_MASK:
542 			/*
543 			 * According to spec we have to retain the
544 			 * high order bits of the FAT entry.
545 			 */
546 			readcn = getulong(bp->b_data + bo);
547 			readcn &= ~FAT32_MASK;
548 			readcn |= newcontents & FAT32_MASK;
549 			putulong(bp->b_data + bo, readcn);
550 			break;
551 		}
552 		updatefats(pmp, bp, bn);
553 		bp = NULL;
554 		pmp->pm_fmod = 1;
555 	}
556 	if (bp)
557 		brelse(bp);
558 	return (0);
559 }
560 
561 /*
562  * Update a contiguous cluster chain
563  *
564  * pmp	    - mount point
565  * start    - first cluster of chain
566  * count    - number of clusters in chain
567  * fillwith - what to write into FAT entry of last cluster
568  */
569 static int
570 fatchain(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith)
571 {
572 	int error;
573 	u_long bn, bo, bsize, byteoffset, readcn, newc;
574 	struct buf *bp;
575 
576 #ifdef MSDOSFS_DEBUG
577 	printf("fatchain(pmp %p, start %lu, count %lu, fillwith %lx)\n",
578 	    pmp, start, count, fillwith);
579 #endif
580 	/*
581 	 * Be sure the clusters are in the filesystem.
582 	 */
583 	if (start < CLUST_FIRST || start + count - 1 > pmp->pm_maxcluster)
584 		return (EINVAL);
585 
586 	while (count > 0) {
587 		byteoffset = FATOFS(pmp, start);
588 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
589 		error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
590 		if (error) {
591 			return (error);
592 		}
593 		while (count > 0) {
594 			start++;
595 			newc = --count > 0 ? start : fillwith;
596 			switch (pmp->pm_fatmask) {
597 			case FAT12_MASK:
598 				readcn = getushort(bp->b_data + bo);
599 				if (start & 1) {
600 					readcn &= 0xf000;
601 					readcn |= newc & 0xfff;
602 				} else {
603 					readcn &= 0x000f;
604 					readcn |= newc << 4;
605 				}
606 				putushort(bp->b_data + bo, readcn);
607 				bo++;
608 				if (!(start & 1))
609 					bo++;
610 				break;
611 			case FAT16_MASK:
612 				putushort(bp->b_data + bo, newc);
613 				bo += 2;
614 				break;
615 			case FAT32_MASK:
616 				readcn = getulong(bp->b_data + bo);
617 				readcn &= ~pmp->pm_fatmask;
618 				readcn |= newc & pmp->pm_fatmask;
619 				putulong(bp->b_data + bo, readcn);
620 				bo += 4;
621 				break;
622 			}
623 			if (bo >= bsize)
624 				break;
625 		}
626 		updatefats(pmp, bp, bn);
627 	}
628 	pmp->pm_fmod = 1;
629 	return (0);
630 }
631 
632 /*
633  * Check the length of a free cluster chain starting at start.
634  *
635  * pmp	 - mount point
636  * start - start of chain
637  * count - maximum interesting length
638  */
639 static int
640 chainlength(struct msdosfsmount *pmp, u_long start, u_long count)
641 {
642 	u_long idx, max_idx;
643 	u_int map;
644 	u_long len;
645 
646 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
647 
648 	if (start > pmp->pm_maxcluster)
649 		return (0);
650 	max_idx = pmp->pm_maxcluster / N_INUSEBITS;
651 	idx = start / N_INUSEBITS;
652 	start %= N_INUSEBITS;
653 	map = pmp->pm_inusemap[idx];
654 	map &= ~((1U << start) - 1);
655 	if (map) {
656 		len = ffs(map) - 1 - start;
657 		len = MIN(len, count);
658 		if (start + len > pmp->pm_maxcluster)
659 			len = pmp->pm_maxcluster - start + 1;
660 		return (len);
661 	}
662 	len = N_INUSEBITS - start;
663 	if (len >= count) {
664 		len = count;
665 		if (start + len > pmp->pm_maxcluster)
666 			len = pmp->pm_maxcluster - start + 1;
667 		return (len);
668 	}
669 	while (++idx <= max_idx) {
670 		if (len >= count)
671 			break;
672 		map = pmp->pm_inusemap[idx];
673 		if (map) {
674 			len += ffs(map) - 1;
675 			break;
676 		}
677 		len += N_INUSEBITS;
678 	}
679 	len = MIN(len, count);
680 	if (start + len > pmp->pm_maxcluster)
681 		len = pmp->pm_maxcluster - start + 1;
682 	return (len);
683 }
684 
685 /*
686  * Allocate contigous free clusters.
687  *
688  * pmp	      - mount point.
689  * start      - start of cluster chain.
690  * count      - number of clusters to allocate.
691  * fillwith   - put this value into the FAT entry for the
692  *		last allocated cluster.
693  * retcluster - put the first allocated cluster's number here.
694  * got	      - how many clusters were actually allocated.
695  */
696 static int
697 chainalloc(struct msdosfsmount *pmp, u_long start, u_long count,
698     u_long fillwith, u_long *retcluster, u_long *got)
699 {
700 	int error;
701 	u_long cl, n;
702 
703 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
704 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
705 	    ("chainalloc on ro msdosfs mount"));
706 
707 	for (cl = start, n = count; n-- > 0;)
708 		usemap_alloc(pmp, cl++);
709 	pmp->pm_nxtfree = start + count;
710 	if (pmp->pm_nxtfree > pmp->pm_maxcluster)
711 		pmp->pm_nxtfree = CLUST_FIRST;
712 	pmp->pm_flags |= MSDOSFS_FSIMOD;
713 	error = fatchain(pmp, start, count, fillwith);
714 	if (error != 0) {
715 		for (cl = start, n = count; n-- > 0;)
716 			(void)usemap_free(pmp, cl++);
717 		return (error);
718 	}
719 #ifdef MSDOSFS_DEBUG
720 	printf("clusteralloc(): allocated cluster chain at %lu (%lu clusters)\n",
721 	    start, count);
722 #endif
723 	if (retcluster)
724 		*retcluster = start;
725 	if (got)
726 		*got = count;
727 	return (0);
728 }
729 
730 /*
731  * Allocate contiguous free clusters.
732  *
733  * pmp	      - mount point.
734  * start      - preferred start of cluster chain.
735  * count      - number of clusters requested.
736  * fillwith   - put this value into the FAT entry for the
737  *		last allocated cluster.
738  * retcluster - put the first allocated cluster's number here.
739  * got	      - how many clusters were actually allocated.
740  */
741 int
742 clusteralloc(struct msdosfsmount *pmp, u_long start, u_long count,
743     u_long fillwith, u_long *retcluster, u_long *got)
744 {
745 	int error;
746 
747 	MSDOSFS_LOCK_MP(pmp);
748 	error = clusteralloc1(pmp, start, count, fillwith, retcluster, got);
749 	MSDOSFS_UNLOCK_MP(pmp);
750 	return (error);
751 }
752 
753 static int
754 clusteralloc1(struct msdosfsmount *pmp, u_long start, u_long count,
755     u_long fillwith, u_long *retcluster, u_long *got)
756 {
757 	u_long idx;
758 	u_long len, newst, foundl, cn, l;
759 	u_long foundcn = 0; /* XXX: foundcn could be used unititialized */
760 	u_int map;
761 
762 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
763 
764 #ifdef MSDOSFS_DEBUG
765 	printf("clusteralloc(): find %lu clusters\n", count);
766 #endif
767 	if (start) {
768 		if ((len = chainlength(pmp, start, count)) >= count)
769 			return (chainalloc(pmp, start, count, fillwith, retcluster, got));
770 	} else
771 		len = 0;
772 
773 	newst = pmp->pm_nxtfree;
774 	foundl = 0;
775 
776 	for (cn = newst; cn <= pmp->pm_maxcluster;) {
777 		idx = cn / N_INUSEBITS;
778 		map = pmp->pm_inusemap[idx];
779 		map |= (1U << (cn % N_INUSEBITS)) - 1;
780 		if (map != FULL_RUN) {
781 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
782 			if ((l = chainlength(pmp, cn, count)) >= count)
783 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
784 			if (l > foundl) {
785 				foundcn = cn;
786 				foundl = l;
787 			}
788 			cn += l + 1;
789 			continue;
790 		}
791 		cn += N_INUSEBITS - cn % N_INUSEBITS;
792 	}
793 	for (cn = 0; cn < newst;) {
794 		idx = cn / N_INUSEBITS;
795 		map = pmp->pm_inusemap[idx];
796 		map |= (1U << (cn % N_INUSEBITS)) - 1;
797 		if (map != FULL_RUN) {
798 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
799 			if ((l = chainlength(pmp, cn, count)) >= count)
800 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
801 			if (l > foundl) {
802 				foundcn = cn;
803 				foundl = l;
804 			}
805 			cn += l + 1;
806 			continue;
807 		}
808 		cn += N_INUSEBITS - cn % N_INUSEBITS;
809 	}
810 
811 	if (!foundl)
812 		return (ENOSPC);
813 
814 	if (len)
815 		return (chainalloc(pmp, start, len, fillwith, retcluster, got));
816 	else
817 		return (chainalloc(pmp, foundcn, foundl, fillwith, retcluster, got));
818 }
819 
820 /*
821  * Free a chain of clusters.
822  *
823  * pmp		- address of the msdosfs mount structure for the filesystem
824  *		  containing the cluster chain to be freed.
825  * startcluster - number of the 1st cluster in the chain of clusters to be
826  *		  freed.
827  */
828 int
829 freeclusterchain(struct msdosfsmount *pmp, u_long cluster)
830 {
831 	int error;
832 	struct buf *bp = NULL;
833 	u_long bn, bo, bsize, byteoffset;
834 	u_long readcn, lbn = -1;
835 
836 	MSDOSFS_LOCK_MP(pmp);
837 	while (cluster >= CLUST_FIRST && cluster <= pmp->pm_maxcluster) {
838 		byteoffset = FATOFS(pmp, cluster);
839 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
840 		if (lbn != bn) {
841 			if (bp)
842 				updatefats(pmp, bp, lbn);
843 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
844 			if (error) {
845 				MSDOSFS_UNLOCK_MP(pmp);
846 				return (error);
847 			}
848 			lbn = bn;
849 		}
850 		error = usemap_free(pmp, cluster);
851 		if (error != 0) {
852 			updatefats(pmp, bp, lbn);
853 			MSDOSFS_UNLOCK_MP(pmp);
854 			return (error);
855 		}
856 		switch (pmp->pm_fatmask) {
857 		case FAT12_MASK:
858 			readcn = getushort(bp->b_data + bo);
859 			if (cluster & 1) {
860 				cluster = readcn >> 4;
861 				readcn &= 0x000f;
862 				readcn |= MSDOSFSFREE << 4;
863 			} else {
864 				cluster = readcn;
865 				readcn &= 0xf000;
866 				readcn |= MSDOSFSFREE & 0xfff;
867 			}
868 			putushort(bp->b_data + bo, readcn);
869 			break;
870 		case FAT16_MASK:
871 			cluster = getushort(bp->b_data + bo);
872 			putushort(bp->b_data + bo, MSDOSFSFREE);
873 			break;
874 		case FAT32_MASK:
875 			cluster = getulong(bp->b_data + bo);
876 			putulong(bp->b_data + bo,
877 				 (MSDOSFSFREE & FAT32_MASK) | (cluster & ~FAT32_MASK));
878 			break;
879 		}
880 		cluster &= pmp->pm_fatmask;
881 		if ((cluster | ~pmp->pm_fatmask) >= CLUST_RSRVD)
882 			cluster |= pmp->pm_fatmask;
883 	}
884 	if (bp)
885 		updatefats(pmp, bp, bn);
886 	MSDOSFS_UNLOCK_MP(pmp);
887 	return (0);
888 }
889 
890 /*
891  * Read in FAT blocks looking for free clusters. For every free cluster
892  * found turn off its corresponding bit in the pm_inusemap.
893  */
894 int
895 fillinusemap(struct msdosfsmount *pmp)
896 {
897 	struct buf *bp;
898 	u_long bn, bo, bsize, byteoffset, cn, readcn;
899 	int error;
900 
901 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
902 	bp = NULL;
903 
904 	/*
905 	 * Mark all clusters in use, we mark the free ones in the FAT scan
906 	 * loop further down.
907 	 */
908 	for (cn = 0; cn < (pmp->pm_maxcluster + N_INUSEBITS) / N_INUSEBITS; cn++)
909 		pmp->pm_inusemap[cn] = FULL_RUN;
910 
911 	/*
912 	 * Figure how many free clusters are in the filesystem by ripping
913 	 * through the FAT counting the number of entries whose content is
914 	 * zero.  These represent free clusters.
915 	 */
916 	pmp->pm_freeclustercount = 0;
917 	for (cn = 0; cn <= pmp->pm_maxcluster; cn++) {
918 		byteoffset = FATOFS(pmp, cn);
919 		bo = byteoffset % pmp->pm_fatblocksize;
920 		if (bo == 0) {
921 			/* Read new FAT block */
922 			if (bp != NULL)
923 				brelse(bp);
924 			fatblock(pmp, byteoffset, &bn, &bsize, NULL);
925 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
926 			if (error != 0)
927 				return (error);
928 		}
929 		if (FAT32(pmp))
930 			readcn = getulong(bp->b_data + bo);
931 		else
932 			readcn = getushort(bp->b_data + bo);
933 		if (FAT12(pmp) && (cn & 1))
934 			readcn >>= 4;
935 		readcn &= pmp->pm_fatmask;
936 
937 		/*
938 		 * Check if the FAT ID matches the BPB's media descriptor and
939 		 * all other bits are set to 1.
940 		 */
941 		if (cn == 0 && readcn != ((pmp->pm_fatmask & 0xffffff00) |
942 		    pmp->pm_bpb.bpbMedia)) {
943 #ifdef MSDOSFS_DEBUG
944 			printf("mountmsdosfs(): Media descriptor in BPB"
945 			    "does not match FAT ID\n");
946 #endif
947 			brelse(bp);
948 			return (EINVAL);
949 		} else if (readcn == CLUST_FREE) {
950 			error = usemap_free(pmp, cn);
951 			if (error != 0) {
952 				brelse(bp);
953 				return (error);
954 			}
955 		}
956 	}
957 	if (bp != NULL)
958 		brelse(bp);
959 
960 	for (cn = pmp->pm_maxcluster + 1; cn < (pmp->pm_maxcluster +
961 	    N_INUSEBITS) / N_INUSEBITS; cn++)
962 		pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
963 
964 	return (0);
965 }
966 
967 /*
968  * Allocate a new cluster and chain it onto the end of the file.
969  *
970  * dep	 - the file to extend
971  * count - number of clusters to allocate
972  * bpp	 - where to return the address of the buf header for the first new
973  *	   file block
974  * ncp	 - where to put cluster number of the first newly allocated cluster
975  *	   If this pointer is 0, do not return the cluster number.
976  * flags - see fat.h
977  *
978  * NOTE: This function is not responsible for turning on the DE_UPDATE bit of
979  * the de_flag field of the denode and it does not change the de_FileSize
980  * field.  This is left for the caller to do.
981  */
982 int
983 extendfile(struct denode *dep, u_long count, struct buf **bpp, u_long *ncp,
984     int flags)
985 {
986 	int error;
987 	u_long frcn;
988 	u_long cn, got;
989 	struct msdosfsmount *pmp = dep->de_pmp;
990 	struct buf *bp;
991 	daddr_t blkno;
992 
993 	/*
994 	 * Don't try to extend the root directory
995 	 */
996 	if (dep->de_StartCluster == MSDOSFSROOT
997 	    && (dep->de_Attributes & ATTR_DIRECTORY)) {
998 #ifdef MSDOSFS_DEBUG
999 		printf("extendfile(): attempt to extend root directory\n");
1000 #endif
1001 		return (ENOSPC);
1002 	}
1003 
1004 	/*
1005 	 * If the "file's last cluster" cache entry is empty, and the file
1006 	 * is not empty, then fill the cache entry by calling pcbmap().
1007 	 */
1008 	if (dep->de_fc[FC_LASTFC].fc_frcn == FCE_EMPTY &&
1009 	    dep->de_StartCluster != 0) {
1010 		error = pcbmap(dep, 0xffff, 0, &cn, 0);
1011 		/* we expect it to return E2BIG */
1012 		if (error != E2BIG)
1013 			return (error);
1014 	}
1015 
1016 	dep->de_fc[FC_NEXTTOLASTFC].fc_frcn =
1017 	    dep->de_fc[FC_LASTFC].fc_frcn;
1018 	dep->de_fc[FC_NEXTTOLASTFC].fc_fsrcn =
1019 	    dep->de_fc[FC_LASTFC].fc_fsrcn;
1020 	while (count > 0) {
1021 		/*
1022 		 * Allocate a new cluster chain and cat onto the end of the
1023 		 * file.  If the file is empty we make de_StartCluster point
1024 		 * to the new block.  Note that de_StartCluster being 0 is
1025 		 * sufficient to be sure the file is empty since we exclude
1026 		 * attempts to extend the root directory above, and the root
1027 		 * dir is the only file with a startcluster of 0 that has
1028 		 * blocks allocated (sort of).
1029 		 */
1030 		if (dep->de_StartCluster == 0)
1031 			cn = 0;
1032 		else
1033 			cn = dep->de_fc[FC_LASTFC].fc_fsrcn + 1;
1034 		error = clusteralloc(pmp, cn, count, CLUST_EOFE, &cn, &got);
1035 		if (error)
1036 			return (error);
1037 
1038 		count -= got;
1039 
1040 		/*
1041 		 * Give them the filesystem relative cluster number if they want
1042 		 * it.
1043 		 */
1044 		if (ncp) {
1045 			*ncp = cn;
1046 			ncp = NULL;
1047 		}
1048 
1049 		if (dep->de_StartCluster == 0) {
1050 			dep->de_StartCluster = cn;
1051 			frcn = 0;
1052 		} else {
1053 			error = fatentry(FAT_SET, pmp,
1054 					 dep->de_fc[FC_LASTFC].fc_fsrcn,
1055 					 0, cn);
1056 			if (error) {
1057 				clusterfree(pmp, cn);
1058 				return (error);
1059 			}
1060 			frcn = dep->de_fc[FC_LASTFC].fc_frcn + 1;
1061 		}
1062 
1063 		/*
1064 		 * Update the "last cluster of the file" entry in the
1065 		 * denode's FAT cache.
1066 		 */
1067 		fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
1068 
1069 		if (flags & DE_CLEAR) {
1070 			while (got-- > 0) {
1071 				/*
1072 				 * Get the buf header for the new block of the file.
1073 				 */
1074 				if (dep->de_Attributes & ATTR_DIRECTORY)
1075 					bp = getblk(pmp->pm_devvp,
1076 					    cntobn(pmp, cn++),
1077 					    pmp->pm_bpcluster, 0, 0, 0);
1078 				else {
1079 					bp = getblk(DETOV(dep),
1080 					    frcn++,
1081 					    pmp->pm_bpcluster, 0, 0, 0);
1082 					/*
1083 					 * Do the bmap now, as in msdosfs_write
1084 					 */
1085 					if (pcbmap(dep,
1086 					    bp->b_lblkno,
1087 					    &blkno, 0, 0))
1088 						bp->b_blkno = -1;
1089 					if (bp->b_blkno == -1)
1090 						panic("extendfile: pcbmap");
1091 					else
1092 						bp->b_blkno = blkno;
1093 				}
1094 				clrbuf(bp);
1095 				if (bpp) {
1096 					*bpp = bp;
1097 					bpp = NULL;
1098 				} else {
1099 					bdwrite(bp);
1100 				}
1101 				if (vm_page_count_severe() ||
1102 				    buf_dirty_count_severe())
1103 					vn_fsync_buf(DETOV(dep), MNT_WAIT);
1104 			}
1105 		}
1106 	}
1107 
1108 	return (0);
1109 }
1110 
1111 /*-
1112  * Routine to mark a FAT16 or FAT32 volume as "clean" or "dirty" by
1113  * manipulating the upper bit of the FAT entry for cluster 1.  Note that
1114  * this bit is not defined for FAT12 volumes, which are always assumed to
1115  * be clean.
1116  *
1117  * The fatentry() routine only works on cluster numbers that a file could
1118  * occupy, so it won't manipulate the entry for cluster 1.  So we have to do
1119  * it here.  The code was stolen from fatentry() and tailored for cluster 1.
1120  *
1121  * Inputs:
1122  *	pmp	The MS-DOS volume to mark
1123  *	dirty	Non-zero if the volume should be marked dirty; zero if it
1124  *		should be marked clean
1125  *
1126  * Result:
1127  *	0	Success
1128  *	EROFS	Volume is read-only
1129  *	?	(other errors from called routines)
1130  */
1131 int
1132 markvoldirty_upgrade(struct msdosfsmount *pmp, bool dirty, bool rw_upgrade)
1133 {
1134 	struct buf *bp;
1135 	u_long bn, bo, bsize, byteoffset, fatval;
1136 	int error;
1137 
1138 	/*
1139 	 * FAT12 does not support a "clean" bit, so don't do anything for
1140 	 * FAT12.
1141 	 */
1142 	if (FAT12(pmp))
1143 		return (0);
1144 
1145 	/*
1146 	 * Can't change the bit on a read-only filesystem, except as part of
1147 	 * ro->rw upgrade.
1148 	 */
1149 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) != 0 && !rw_upgrade)
1150 		return (EROFS);
1151 
1152 	/*
1153 	 * Fetch the block containing the FAT entry.  It is given by the
1154 	 * pseudo-cluster 1.
1155 	 */
1156 	byteoffset = FATOFS(pmp, 1);
1157 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
1158 	error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
1159 	if (error)
1160 		return (error);
1161 
1162 	/*
1163 	 * Get the current value of the FAT entry and set/clear the relevant
1164 	 * bit.  Dirty means clear the "clean" bit; clean means set the
1165 	 * "clean" bit.
1166 	 */
1167 	if (FAT32(pmp)) {
1168 		/* FAT32 uses bit 27. */
1169 		fatval = getulong(&bp->b_data[bo]);
1170 		if (dirty)
1171 			fatval &= 0xF7FFFFFF;
1172 		else
1173 			fatval |= 0x08000000;
1174 		putulong(&bp->b_data[bo], fatval);
1175 	} else {
1176 		/* Must be FAT16; use bit 15. */
1177 		fatval = getushort(&bp->b_data[bo]);
1178 		if (dirty)
1179 			fatval &= 0x7FFF;
1180 		else
1181 			fatval |= 0x8000;
1182 		putushort(&bp->b_data[bo], fatval);
1183 	}
1184 
1185 	/*
1186 	 * The concern here is that a devvp may be readonly, without reporting
1187 	 * itself as such through the usual channels.  In that case, we'd like
1188 	 * it if attempting to mount msdosfs rw didn't panic the system.
1189 	 *
1190 	 * markvoldirty is invoked as the first write on backing devvps when
1191 	 * either msdosfs is mounted for the first time, or a ro mount is
1192 	 * upgraded to rw.
1193 	 *
1194 	 * In either event, if a write error occurs dirtying the volume:
1195 	 *   - No user data has been permitted to be written to cache yet.
1196 	 *   - We can abort the high-level operation (mount, or ro->rw) safely.
1197 	 *   - We don't derive any benefit from leaving a zombie dirty buf in
1198 	 *   the cache that can not be cleaned or evicted.
1199 	 *
1200 	 * So, mark B_INVALONERR to have bwrite() -> brelse() detect that
1201 	 * condition and force-invalidate our write to the block if it occurs.
1202 	 *
1203 	 * PR 210316 provides more context on the discovery and diagnosis of
1204 	 * the problem, as well as earlier attempts to solve it.
1205 	 */
1206 	bp->b_flags |= B_INVALONERR;
1207 
1208 	/* Write out the modified FAT block synchronously. */
1209 	return (bwrite(bp));
1210 }
1211