xref: /freebsd/sys/fs/msdosfs/msdosfs_fat.c (revision 3cbb4cc200f8a0ad7ed08233425ea54524a21f1c)
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 __inline void
84 		usemap_free(struct msdosfsmount *pmp, u_long cn);
85 static int	clusteralloc1(struct msdosfsmount *pmp, u_long start,
86 		    u_long count, u_long fillwith, u_long *retcluster,
87 		    u_long *got);
88 
89 static void
90 fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp, u_long *sizep,
91     u_long *bop)
92 {
93 	u_long bn, size;
94 
95 	bn = ofs / pmp->pm_fatblocksize * pmp->pm_fatblocksec;
96 	size = min(pmp->pm_fatblocksec, pmp->pm_FATsecs - bn)
97 	    * DEV_BSIZE;
98 	bn += pmp->pm_fatblk + pmp->pm_curfat * pmp->pm_FATsecs;
99 
100 	if (bnp)
101 		*bnp = bn;
102 	if (sizep)
103 		*sizep = size;
104 	if (bop)
105 		*bop = ofs % pmp->pm_fatblocksize;
106 }
107 
108 /*
109  * Map the logical cluster number of a file into a physical disk sector
110  * that is filesystem relative.
111  *
112  * dep	  - address of denode representing the file of interest
113  * findcn - file relative cluster whose filesystem relative cluster number
114  *	    and/or block number are/is to be found
115  * bnp	  - address of where to place the filesystem relative block number.
116  *	    If this pointer is null then don't return this quantity.
117  * cnp	  - address of where to place the filesystem relative cluster number.
118  *	    If this pointer is null then don't return this quantity.
119  * sp     - pointer to returned block size
120  *
121  * NOTE: Either bnp or cnp must be non-null.
122  * This function has one side effect.  If the requested file relative cluster
123  * is beyond the end of file, then the actual number of clusters in the file
124  * is returned in *cnp.  This is useful for determining how long a directory is.
125  *  If cnp is null, nothing is returned.
126  */
127 int
128 pcbmap(struct denode *dep, u_long findcn, daddr_t *bnp, u_long *cnp, int *sp)
129 {
130 	int error;
131 	u_long i;
132 	u_long cn;
133 	u_long prevcn = 0; /* XXX: prevcn could be used unititialized */
134 	u_long byteoffset;
135 	u_long bn;
136 	u_long bo;
137 	struct buf *bp = NULL;
138 	u_long bp_bn = -1;
139 	struct msdosfsmount *pmp = dep->de_pmp;
140 	u_long bsize;
141 
142 	KASSERT(bnp != NULL || cnp != NULL || sp != NULL,
143 	    ("pcbmap: extra call"));
144 	ASSERT_VOP_ELOCKED(DETOV(dep), "pcbmap");
145 
146 	cn = dep->de_StartCluster;
147 	/*
148 	 * The "file" that makes up the root directory is contiguous,
149 	 * permanently allocated, of fixed size, and is not made up of
150 	 * clusters.  If the cluster number is beyond the end of the root
151 	 * directory, then return the number of clusters in the file.
152 	 */
153 	if (cn == MSDOSFSROOT) {
154 		if (dep->de_Attributes & ATTR_DIRECTORY) {
155 			if (de_cn2off(pmp, findcn) >= dep->de_FileSize) {
156 				if (cnp)
157 					*cnp = de_bn2cn(pmp, pmp->pm_rootdirsize);
158 				return (E2BIG);
159 			}
160 			if (bnp)
161 				*bnp = pmp->pm_rootdirblk + de_cn2bn(pmp, findcn);
162 			if (cnp)
163 				*cnp = MSDOSFSROOT;
164 			if (sp)
165 				*sp = min(pmp->pm_bpcluster,
166 				    dep->de_FileSize - de_cn2off(pmp, findcn));
167 			return (0);
168 		} else {		/* just an empty file */
169 			if (cnp)
170 				*cnp = 0;
171 			return (E2BIG);
172 		}
173 	}
174 
175 	/*
176 	 * All other files do I/O in cluster sized blocks
177 	 */
178 	if (sp)
179 		*sp = pmp->pm_bpcluster;
180 
181 	/*
182 	 * Rummage around in the FAT cache, maybe we can avoid tromping
183 	 * through every FAT entry for the file. And, keep track of how far
184 	 * off the cache was from where we wanted to be.
185 	 */
186 	i = 0;
187 	fc_lookup(dep, findcn, &i, &cn);
188 
189 	/*
190 	 * Handle all other files or directories the normal way.
191 	 */
192 	for (; i < findcn; i++) {
193 		/*
194 		 * Stop with all reserved clusters, not just with EOF.
195 		 */
196 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
197 			goto hiteof;
198 		byteoffset = FATOFS(pmp, cn);
199 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
200 		if (bn != bp_bn) {
201 			if (bp)
202 				brelse(bp);
203 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
204 			if (error) {
205 				return (error);
206 			}
207 			bp_bn = bn;
208 		}
209 		prevcn = cn;
210 		if (bo >= bsize) {
211 			if (bp)
212 				brelse(bp);
213 			return (EIO);
214 		}
215 		if (FAT32(pmp))
216 			cn = getulong(bp->b_data + bo);
217 		else
218 			cn = getushort(bp->b_data + bo);
219 		if (FAT12(pmp) && (prevcn & 1))
220 			cn >>= 4;
221 		cn &= pmp->pm_fatmask;
222 
223 		/*
224 		 * Force the special cluster numbers
225 		 * to be the same for all cluster sizes
226 		 * to let the rest of msdosfs handle
227 		 * all cases the same.
228 		 */
229 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
230 			cn |= ~pmp->pm_fatmask;
231 	}
232 
233 	if (!MSDOSFSEOF(pmp, cn)) {
234 		if (bp)
235 			brelse(bp);
236 		if (bnp)
237 			*bnp = cntobn(pmp, cn);
238 		if (cnp)
239 			*cnp = cn;
240 		fc_setcache(dep, FC_LASTMAP, i, cn);
241 		return (0);
242 	}
243 
244 hiteof:;
245 	if (cnp)
246 		*cnp = i;
247 	if (bp)
248 		brelse(bp);
249 	/* update last file cluster entry in the FAT cache */
250 	fc_setcache(dep, FC_LASTFC, i - 1, prevcn);
251 	return (E2BIG);
252 }
253 
254 /*
255  * Find the closest entry in the FAT cache to the cluster we are looking
256  * for.
257  */
258 static void
259 fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp, u_long *fsrcnp)
260 {
261 	int i;
262 	u_long cn;
263 	struct fatcache *closest = NULL;
264 
265 	ASSERT_VOP_LOCKED(DETOV(dep), "fc_lookup");
266 
267 	for (i = 0; i < FC_SIZE; i++) {
268 		cn = dep->de_fc[i].fc_frcn;
269 		if (cn != FCE_EMPTY && cn <= findcn) {
270 			if (closest == NULL || cn > closest->fc_frcn)
271 				closest = &dep->de_fc[i];
272 		}
273 	}
274 	if (closest) {
275 		*frcnp = closest->fc_frcn;
276 		*fsrcnp = closest->fc_fsrcn;
277 	}
278 }
279 
280 /*
281  * Purge the FAT cache in denode dep of all entries relating to file
282  * relative cluster frcn and beyond.
283  */
284 void
285 fc_purge(struct denode *dep, u_int frcn)
286 {
287 	int i;
288 	struct fatcache *fcp;
289 
290 	ASSERT_VOP_ELOCKED(DETOV(dep), "fc_purge");
291 
292 	fcp = dep->de_fc;
293 	for (i = 0; i < FC_SIZE; i++, fcp++) {
294 		if (fcp->fc_frcn >= frcn)
295 			fcp->fc_frcn = FCE_EMPTY;
296 	}
297 }
298 
299 /*
300  * Update the FAT.
301  * If mirroring the FAT, update all copies, with the first copy as last.
302  * Else update only the current FAT (ignoring the others).
303  *
304  * pmp	 - msdosfsmount structure for filesystem to update
305  * bp	 - addr of modified FAT block
306  * fatbn - block number relative to begin of filesystem of the modified FAT block.
307  */
308 static void
309 updatefats(struct msdosfsmount *pmp, struct buf *bp, u_long fatbn)
310 {
311 	struct buf *bpn;
312 	int cleanfat, i;
313 
314 #ifdef MSDOSFS_DEBUG
315 	printf("updatefats(pmp %p, bp %p, fatbn %lu)\n", pmp, bp, fatbn);
316 #endif
317 
318 	if (pmp->pm_flags & MSDOSFS_FATMIRROR) {
319 		/*
320 		 * Now copy the block(s) of the modified FAT to the other copies of
321 		 * the FAT and write them out.  This is faster than reading in the
322 		 * other FATs and then writing them back out.  This could tie up
323 		 * the FAT for quite a while. Preventing others from accessing it.
324 		 * To prevent us from going after the FAT quite so much we use
325 		 * delayed writes, unless they specified "synchronous" when the
326 		 * filesystem was mounted.  If synch is asked for then use
327 		 * bwrite()'s and really slow things down.
328 		 */
329 		if (fatbn != pmp->pm_fatblk || FAT12(pmp))
330 			cleanfat = 0;
331 		else if (FAT16(pmp))
332 			cleanfat = 16;
333 		else
334 			cleanfat = 32;
335 		for (i = 1; i < pmp->pm_FATs; i++) {
336 			fatbn += pmp->pm_FATsecs;
337 			/* getblk() never fails */
338 			bpn = getblk(pmp->pm_devvp, fatbn, bp->b_bcount,
339 			    0, 0, 0);
340 			memcpy(bpn->b_data, bp->b_data, bp->b_bcount);
341 			/* Force the clean bit on in the other copies. */
342 			if (cleanfat == 16)
343 				((uint8_t *)bpn->b_data)[3] |= 0x80;
344 			else if (cleanfat == 32)
345 				((uint8_t *)bpn->b_data)[7] |= 0x08;
346 			if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
347 				bwrite(bpn);
348 			else
349 				bdwrite(bpn);
350 		}
351 	}
352 
353 	/*
354 	 * Write out the first (or current) FAT last.
355 	 */
356 	if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
357 		bwrite(bp);
358 	else
359 		bdwrite(bp);
360 }
361 
362 /*
363  * Updating entries in 12 bit FATs is a pain in the butt.
364  *
365  * The following picture shows where nibbles go when moving from a 12 bit
366  * cluster number into the appropriate bytes in the FAT.
367  *
368  *	byte m        byte m+1      byte m+2
369  *	+----+----+   +----+----+   +----+----+
370  *	|  0    1 |   |  2    3 |   |  4    5 |   FAT bytes
371  *	+----+----+   +----+----+   +----+----+
372  *
373  *	+----+----+----+   +----+----+----+
374  *	|  3    0    1 |   |  4    5    2 |
375  *	+----+----+----+   +----+----+----+
376  *	cluster n  	   cluster n+1
377  *
378  * Where n is even. m = n + (n >> 2)
379  *
380  */
381 static __inline void
382 usemap_alloc(struct msdosfsmount *pmp, u_long cn)
383 {
384 
385 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
386 
387 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
388 	    pmp->pm_maxcluster));
389 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
390 	    ("usemap_alloc on ro msdosfs mount"));
391 	KASSERT((pmp->pm_inusemap[cn / N_INUSEBITS] &
392 	    (1U << (cn % N_INUSEBITS))) == 0,
393 	    ("Allocating used sector %ld %ld %x", cn, cn % N_INUSEBITS,
394 	    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]));
395 	pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
396 	KASSERT(pmp->pm_freeclustercount > 0, ("usemap_alloc: too little"));
397 	pmp->pm_freeclustercount--;
398 	pmp->pm_flags |= MSDOSFS_FSIMOD;
399 }
400 
401 static __inline void
402 usemap_free(struct msdosfsmount *pmp, u_long cn)
403 {
404 
405 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
406 
407 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
408 	    pmp->pm_maxcluster));
409 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
410 	    ("usemap_free on ro msdosfs mount"));
411 	pmp->pm_freeclustercount++;
412 	pmp->pm_flags |= MSDOSFS_FSIMOD;
413 	KASSERT((pmp->pm_inusemap[cn / N_INUSEBITS] &
414 	    (1U << (cn % N_INUSEBITS))) != 0,
415 	    ("Freeing unused sector %ld %ld %x", cn, cn % N_INUSEBITS,
416 	    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]));
417 	pmp->pm_inusemap[cn / N_INUSEBITS] &= ~(1U << (cn % N_INUSEBITS));
418 }
419 
420 int
421 clusterfree(struct msdosfsmount *pmp, u_long cluster, u_long *oldcnp)
422 {
423 	int error;
424 	u_long oldcn;
425 
426 	error = fatentry(FAT_GET_AND_SET, pmp, cluster, &oldcn, MSDOSFSFREE);
427 	if (error)
428 		return (error);
429 	/*
430 	 * If the cluster was successfully marked free, then update
431 	 * the count of free clusters, and turn off the "allocated"
432 	 * bit in the "in use" cluster bit map.
433 	 */
434 	MSDOSFS_LOCK_MP(pmp);
435 	usemap_free(pmp, cluster);
436 	MSDOSFS_UNLOCK_MP(pmp);
437 	if (oldcnp)
438 		*oldcnp = oldcn;
439 	return (0);
440 }
441 
442 /*
443  * Get or Set or 'Get and Set' the cluster'th entry in the FAT.
444  *
445  * function	- whether to get or set a FAT entry
446  * pmp		- address of the msdosfsmount structure for the filesystem
447  *		  whose FAT is to be manipulated.
448  * cn		- which cluster is of interest
449  * oldcontents	- address of a word that is to receive the contents of the
450  *		  cluster'th entry if this is a get function
451  * newcontents	- the new value to be written into the cluster'th element of
452  *		  the FAT if this is a set function.
453  *
454  * This function can also be used to free a cluster by setting the FAT entry
455  * for a cluster to 0.
456  *
457  * All copies of the FAT are updated if this is a set function. NOTE: If
458  * fatentry() marks a cluster as free it does not update the inusemap in
459  * the msdosfsmount structure. This is left to the caller.
460  */
461 int
462 fatentry(int function, struct msdosfsmount *pmp, u_long cn, u_long *oldcontents,
463     u_long newcontents)
464 {
465 	int error;
466 	u_long readcn;
467 	u_long bn, bo, bsize, byteoffset;
468 	struct buf *bp;
469 
470 #ifdef	MSDOSFS_DEBUG
471 	printf("fatentry(func %d, pmp %p, clust %lu, oldcon %p, newcon %lx)\n",
472 	    function, pmp, cn, oldcontents, newcontents);
473 #endif
474 
475 #ifdef DIAGNOSTIC
476 	/*
477 	 * Be sure they asked us to do something.
478 	 */
479 	if ((function & (FAT_SET | FAT_GET)) == 0) {
480 #ifdef MSDOSFS_DEBUG
481 		printf("fatentry(): function code doesn't specify get or set\n");
482 #endif
483 		return (EINVAL);
484 	}
485 
486 	/*
487 	 * If they asked us to return a cluster number but didn't tell us
488 	 * where to put it, give them an error.
489 	 */
490 	if ((function & FAT_GET) && oldcontents == NULL) {
491 #ifdef MSDOSFS_DEBUG
492 		printf("fatentry(): get function with no place to put result\n");
493 #endif
494 		return (EINVAL);
495 	}
496 #endif
497 
498 	/*
499 	 * Be sure the requested cluster is in the filesystem.
500 	 */
501 	if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster)
502 		return (EINVAL);
503 
504 	byteoffset = FATOFS(pmp, cn);
505 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
506 	error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
507 	if (error) {
508 		return (error);
509 	}
510 
511 	if (function & FAT_GET) {
512 		if (FAT32(pmp))
513 			readcn = getulong(bp->b_data + bo);
514 		else
515 			readcn = getushort(bp->b_data + bo);
516 		if (FAT12(pmp) & (cn & 1))
517 			readcn >>= 4;
518 		readcn &= pmp->pm_fatmask;
519 		/* map reserved FAT entries to same values for all FATs */
520 		if ((readcn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
521 			readcn |= ~pmp->pm_fatmask;
522 		*oldcontents = readcn;
523 	}
524 	if (function & FAT_SET) {
525 		switch (pmp->pm_fatmask) {
526 		case FAT12_MASK:
527 			readcn = getushort(bp->b_data + bo);
528 			if (cn & 1) {
529 				readcn &= 0x000f;
530 				readcn |= newcontents << 4;
531 			} else {
532 				readcn &= 0xf000;
533 				readcn |= newcontents & 0xfff;
534 			}
535 			putushort(bp->b_data + bo, readcn);
536 			break;
537 		case FAT16_MASK:
538 			putushort(bp->b_data + bo, newcontents);
539 			break;
540 		case FAT32_MASK:
541 			/*
542 			 * According to spec we have to retain the
543 			 * high order bits of the FAT entry.
544 			 */
545 			readcn = getulong(bp->b_data + bo);
546 			readcn &= ~FAT32_MASK;
547 			readcn |= newcontents & FAT32_MASK;
548 			putulong(bp->b_data + bo, readcn);
549 			break;
550 		}
551 		updatefats(pmp, bp, bn);
552 		bp = NULL;
553 		pmp->pm_fmod = 1;
554 	}
555 	if (bp)
556 		brelse(bp);
557 	return (0);
558 }
559 
560 /*
561  * Update a contiguous cluster chain
562  *
563  * pmp	    - mount point
564  * start    - first cluster of chain
565  * count    - number of clusters in chain
566  * fillwith - what to write into FAT entry of last cluster
567  */
568 static int
569 fatchain(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith)
570 {
571 	int error;
572 	u_long bn, bo, bsize, byteoffset, readcn, newc;
573 	struct buf *bp;
574 
575 #ifdef MSDOSFS_DEBUG
576 	printf("fatchain(pmp %p, start %lu, count %lu, fillwith %lx)\n",
577 	    pmp, start, count, fillwith);
578 #endif
579 	/*
580 	 * Be sure the clusters are in the filesystem.
581 	 */
582 	if (start < CLUST_FIRST || start + count - 1 > pmp->pm_maxcluster)
583 		return (EINVAL);
584 
585 	while (count > 0) {
586 		byteoffset = FATOFS(pmp, start);
587 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
588 		error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
589 		if (error) {
590 			return (error);
591 		}
592 		while (count > 0) {
593 			start++;
594 			newc = --count > 0 ? start : fillwith;
595 			switch (pmp->pm_fatmask) {
596 			case FAT12_MASK:
597 				readcn = getushort(bp->b_data + bo);
598 				if (start & 1) {
599 					readcn &= 0xf000;
600 					readcn |= newc & 0xfff;
601 				} else {
602 					readcn &= 0x000f;
603 					readcn |= newc << 4;
604 				}
605 				putushort(bp->b_data + bo, readcn);
606 				bo++;
607 				if (!(start & 1))
608 					bo++;
609 				break;
610 			case FAT16_MASK:
611 				putushort(bp->b_data + bo, newc);
612 				bo += 2;
613 				break;
614 			case FAT32_MASK:
615 				readcn = getulong(bp->b_data + bo);
616 				readcn &= ~pmp->pm_fatmask;
617 				readcn |= newc & pmp->pm_fatmask;
618 				putulong(bp->b_data + bo, readcn);
619 				bo += 4;
620 				break;
621 			}
622 			if (bo >= bsize)
623 				break;
624 		}
625 		updatefats(pmp, bp, bn);
626 	}
627 	pmp->pm_fmod = 1;
628 	return (0);
629 }
630 
631 /*
632  * Check the length of a free cluster chain starting at start.
633  *
634  * pmp	 - mount point
635  * start - start of chain
636  * count - maximum interesting length
637  */
638 static int
639 chainlength(struct msdosfsmount *pmp, u_long start, u_long count)
640 {
641 	u_long idx, max_idx;
642 	u_int map;
643 	u_long len;
644 
645 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
646 
647 	if (start > pmp->pm_maxcluster)
648 		return (0);
649 	max_idx = pmp->pm_maxcluster / N_INUSEBITS;
650 	idx = start / N_INUSEBITS;
651 	start %= N_INUSEBITS;
652 	map = pmp->pm_inusemap[idx];
653 	map &= ~((1U << start) - 1);
654 	if (map) {
655 		len = ffs(map) - 1 - start;
656 		len = MIN(len, count);
657 		if (start + len > pmp->pm_maxcluster)
658 			len = pmp->pm_maxcluster - start + 1;
659 		return (len);
660 	}
661 	len = N_INUSEBITS - start;
662 	if (len >= count) {
663 		len = count;
664 		if (start + len > pmp->pm_maxcluster)
665 			len = pmp->pm_maxcluster - start + 1;
666 		return (len);
667 	}
668 	while (++idx <= max_idx) {
669 		if (len >= count)
670 			break;
671 		map = pmp->pm_inusemap[idx];
672 		if (map) {
673 			len += ffs(map) - 1;
674 			break;
675 		}
676 		len += N_INUSEBITS;
677 	}
678 	len = MIN(len, count);
679 	if (start + len > pmp->pm_maxcluster)
680 		len = pmp->pm_maxcluster - start + 1;
681 	return (len);
682 }
683 
684 /*
685  * Allocate contigous free clusters.
686  *
687  * pmp	      - mount point.
688  * start      - start of cluster chain.
689  * count      - number of clusters to allocate.
690  * fillwith   - put this value into the FAT entry for the
691  *		last allocated cluster.
692  * retcluster - put the first allocated cluster's number here.
693  * got	      - how many clusters were actually allocated.
694  */
695 static int
696 chainalloc(struct msdosfsmount *pmp, u_long start, u_long count,
697     u_long fillwith, u_long *retcluster, u_long *got)
698 {
699 	int error;
700 	u_long cl, n;
701 
702 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
703 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
704 	    ("chainalloc on ro msdosfs mount"));
705 
706 	for (cl = start, n = count; n-- > 0;)
707 		usemap_alloc(pmp, cl++);
708 	pmp->pm_nxtfree = start + count;
709 	if (pmp->pm_nxtfree > pmp->pm_maxcluster)
710 		pmp->pm_nxtfree = CLUST_FIRST;
711 	pmp->pm_flags |= MSDOSFS_FSIMOD;
712 	error = fatchain(pmp, start, count, fillwith);
713 	if (error != 0) {
714 		for (cl = start, n = count; n-- > 0;)
715 			usemap_free(pmp, cl++);
716 		return (error);
717 	}
718 #ifdef MSDOSFS_DEBUG
719 	printf("clusteralloc(): allocated cluster chain at %lu (%lu clusters)\n",
720 	    start, count);
721 #endif
722 	if (retcluster)
723 		*retcluster = start;
724 	if (got)
725 		*got = count;
726 	return (0);
727 }
728 
729 /*
730  * Allocate contiguous free clusters.
731  *
732  * pmp	      - mount point.
733  * start      - preferred start of cluster chain.
734  * count      - number of clusters requested.
735  * fillwith   - put this value into the FAT entry for the
736  *		last allocated cluster.
737  * retcluster - put the first allocated cluster's number here.
738  * got	      - how many clusters were actually allocated.
739  */
740 int
741 clusteralloc(struct msdosfsmount *pmp, u_long start, u_long count,
742     u_long fillwith, u_long *retcluster, u_long *got)
743 {
744 	int error;
745 
746 	MSDOSFS_LOCK_MP(pmp);
747 	error = clusteralloc1(pmp, start, count, fillwith, retcluster, got);
748 	MSDOSFS_UNLOCK_MP(pmp);
749 	return (error);
750 }
751 
752 static int
753 clusteralloc1(struct msdosfsmount *pmp, u_long start, u_long count,
754     u_long fillwith, u_long *retcluster, u_long *got)
755 {
756 	u_long idx;
757 	u_long len, newst, foundl, cn, l;
758 	u_long foundcn = 0; /* XXX: foundcn could be used unititialized */
759 	u_int map;
760 
761 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
762 
763 #ifdef MSDOSFS_DEBUG
764 	printf("clusteralloc(): find %lu clusters\n", count);
765 #endif
766 	if (start) {
767 		if ((len = chainlength(pmp, start, count)) >= count)
768 			return (chainalloc(pmp, start, count, fillwith, retcluster, got));
769 	} else
770 		len = 0;
771 
772 	newst = pmp->pm_nxtfree;
773 	foundl = 0;
774 
775 	for (cn = newst; cn <= pmp->pm_maxcluster;) {
776 		idx = cn / N_INUSEBITS;
777 		map = pmp->pm_inusemap[idx];
778 		map |= (1U << (cn % N_INUSEBITS)) - 1;
779 		if (map != FULL_RUN) {
780 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
781 			if ((l = chainlength(pmp, cn, count)) >= count)
782 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
783 			if (l > foundl) {
784 				foundcn = cn;
785 				foundl = l;
786 			}
787 			cn += l + 1;
788 			continue;
789 		}
790 		cn += N_INUSEBITS - cn % N_INUSEBITS;
791 	}
792 	for (cn = 0; cn < newst;) {
793 		idx = cn / N_INUSEBITS;
794 		map = pmp->pm_inusemap[idx];
795 		map |= (1U << (cn % N_INUSEBITS)) - 1;
796 		if (map != FULL_RUN) {
797 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
798 			if ((l = chainlength(pmp, cn, count)) >= count)
799 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
800 			if (l > foundl) {
801 				foundcn = cn;
802 				foundl = l;
803 			}
804 			cn += l + 1;
805 			continue;
806 		}
807 		cn += N_INUSEBITS - cn % N_INUSEBITS;
808 	}
809 
810 	if (!foundl)
811 		return (ENOSPC);
812 
813 	if (len)
814 		return (chainalloc(pmp, start, len, fillwith, retcluster, got));
815 	else
816 		return (chainalloc(pmp, foundcn, foundl, fillwith, retcluster, got));
817 }
818 
819 /*
820  * Free a chain of clusters.
821  *
822  * pmp		- address of the msdosfs mount structure for the filesystem
823  *		  containing the cluster chain to be freed.
824  * startcluster - number of the 1st cluster in the chain of clusters to be
825  *		  freed.
826  */
827 int
828 freeclusterchain(struct msdosfsmount *pmp, u_long cluster)
829 {
830 	int error;
831 	struct buf *bp = NULL;
832 	u_long bn, bo, bsize, byteoffset;
833 	u_long readcn, lbn = -1;
834 
835 	MSDOSFS_LOCK_MP(pmp);
836 	while (cluster >= CLUST_FIRST && cluster <= pmp->pm_maxcluster) {
837 		byteoffset = FATOFS(pmp, cluster);
838 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
839 		if (lbn != bn) {
840 			if (bp)
841 				updatefats(pmp, bp, lbn);
842 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
843 			if (error) {
844 				MSDOSFS_UNLOCK_MP(pmp);
845 				return (error);
846 			}
847 			lbn = bn;
848 		}
849 		usemap_free(pmp, cluster);
850 		switch (pmp->pm_fatmask) {
851 		case FAT12_MASK:
852 			readcn = getushort(bp->b_data + bo);
853 			if (cluster & 1) {
854 				cluster = readcn >> 4;
855 				readcn &= 0x000f;
856 				readcn |= MSDOSFSFREE << 4;
857 			} else {
858 				cluster = readcn;
859 				readcn &= 0xf000;
860 				readcn |= MSDOSFSFREE & 0xfff;
861 			}
862 			putushort(bp->b_data + bo, readcn);
863 			break;
864 		case FAT16_MASK:
865 			cluster = getushort(bp->b_data + bo);
866 			putushort(bp->b_data + bo, MSDOSFSFREE);
867 			break;
868 		case FAT32_MASK:
869 			cluster = getulong(bp->b_data + bo);
870 			putulong(bp->b_data + bo,
871 				 (MSDOSFSFREE & FAT32_MASK) | (cluster & ~FAT32_MASK));
872 			break;
873 		}
874 		cluster &= pmp->pm_fatmask;
875 		if ((cluster | ~pmp->pm_fatmask) >= CLUST_RSRVD)
876 			cluster |= pmp->pm_fatmask;
877 	}
878 	if (bp)
879 		updatefats(pmp, bp, bn);
880 	MSDOSFS_UNLOCK_MP(pmp);
881 	return (0);
882 }
883 
884 /*
885  * Read in FAT blocks looking for free clusters. For every free cluster
886  * found turn off its corresponding bit in the pm_inusemap.
887  */
888 int
889 fillinusemap(struct msdosfsmount *pmp)
890 {
891 	struct buf *bp;
892 	u_long bn, bo, bsize, byteoffset, cn, readcn;
893 	int error;
894 
895 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
896 	bp = NULL;
897 
898 	/*
899 	 * Mark all clusters in use, we mark the free ones in the FAT scan
900 	 * loop further down.
901 	 */
902 	for (cn = 0; cn < (pmp->pm_maxcluster + N_INUSEBITS) / N_INUSEBITS; cn++)
903 		pmp->pm_inusemap[cn] = FULL_RUN;
904 
905 	/*
906 	 * Figure how many free clusters are in the filesystem by ripping
907 	 * through the FAT counting the number of entries whose content is
908 	 * zero.  These represent free clusters.
909 	 */
910 	pmp->pm_freeclustercount = 0;
911 	for (cn = 0; cn <= pmp->pm_maxcluster; cn++) {
912 		byteoffset = FATOFS(pmp, cn);
913 		bo = byteoffset % pmp->pm_fatblocksize;
914 		if (bo == 0) {
915 			/* Read new FAT block */
916 			if (bp != NULL)
917 				brelse(bp);
918 			fatblock(pmp, byteoffset, &bn, &bsize, NULL);
919 			error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
920 			if (error != 0)
921 				return (error);
922 		}
923 		if (FAT32(pmp))
924 			readcn = getulong(bp->b_data + bo);
925 		else
926 			readcn = getushort(bp->b_data + bo);
927 		if (FAT12(pmp) && (cn & 1))
928 			readcn >>= 4;
929 		readcn &= pmp->pm_fatmask;
930 
931 		/*
932 		 * Check if the FAT ID matches the BPB's media descriptor and
933 		 * all other bits are set to 1.
934 		 */
935 		if (cn == 0 && readcn != ((pmp->pm_fatmask & 0xffffff00) |
936 		    pmp->pm_bpb.bpbMedia)) {
937 #ifdef MSDOSFS_DEBUG
938 			printf("mountmsdosfs(): Media descriptor in BPB"
939 			    "does not match FAT ID\n");
940 #endif
941 			brelse(bp);
942 			return (EINVAL);
943 		} else if (readcn == CLUST_FREE)
944 			usemap_free(pmp, cn);
945 	}
946 	if (bp != NULL)
947 		brelse(bp);
948 
949 	for (cn = pmp->pm_maxcluster + 1; cn < (pmp->pm_maxcluster +
950 	    N_INUSEBITS) / N_INUSEBITS; cn++)
951 		pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
952 
953 	return (0);
954 }
955 
956 /*
957  * Allocate a new cluster and chain it onto the end of the file.
958  *
959  * dep	 - the file to extend
960  * count - number of clusters to allocate
961  * bpp	 - where to return the address of the buf header for the first new
962  *	   file block
963  * ncp	 - where to put cluster number of the first newly allocated cluster
964  *	   If this pointer is 0, do not return the cluster number.
965  * flags - see fat.h
966  *
967  * NOTE: This function is not responsible for turning on the DE_UPDATE bit of
968  * the de_flag field of the denode and it does not change the de_FileSize
969  * field.  This is left for the caller to do.
970  */
971 int
972 extendfile(struct denode *dep, u_long count, struct buf **bpp, u_long *ncp,
973     int flags)
974 {
975 	int error;
976 	u_long frcn;
977 	u_long cn, got;
978 	struct msdosfsmount *pmp = dep->de_pmp;
979 	struct buf *bp;
980 	daddr_t blkno;
981 
982 	/*
983 	 * Don't try to extend the root directory
984 	 */
985 	if (dep->de_StartCluster == MSDOSFSROOT
986 	    && (dep->de_Attributes & ATTR_DIRECTORY)) {
987 #ifdef MSDOSFS_DEBUG
988 		printf("extendfile(): attempt to extend root directory\n");
989 #endif
990 		return (ENOSPC);
991 	}
992 
993 	/*
994 	 * If the "file's last cluster" cache entry is empty, and the file
995 	 * is not empty, then fill the cache entry by calling pcbmap().
996 	 */
997 	if (dep->de_fc[FC_LASTFC].fc_frcn == FCE_EMPTY &&
998 	    dep->de_StartCluster != 0) {
999 		error = pcbmap(dep, 0xffff, 0, &cn, 0);
1000 		/* we expect it to return E2BIG */
1001 		if (error != E2BIG)
1002 			return (error);
1003 	}
1004 
1005 	dep->de_fc[FC_NEXTTOLASTFC].fc_frcn =
1006 	    dep->de_fc[FC_LASTFC].fc_frcn;
1007 	dep->de_fc[FC_NEXTTOLASTFC].fc_fsrcn =
1008 	    dep->de_fc[FC_LASTFC].fc_fsrcn;
1009 	while (count > 0) {
1010 		/*
1011 		 * Allocate a new cluster chain and cat onto the end of the
1012 		 * file.  If the file is empty we make de_StartCluster point
1013 		 * to the new block.  Note that de_StartCluster being 0 is
1014 		 * sufficient to be sure the file is empty since we exclude
1015 		 * attempts to extend the root directory above, and the root
1016 		 * dir is the only file with a startcluster of 0 that has
1017 		 * blocks allocated (sort of).
1018 		 */
1019 		if (dep->de_StartCluster == 0)
1020 			cn = 0;
1021 		else
1022 			cn = dep->de_fc[FC_LASTFC].fc_fsrcn + 1;
1023 		error = clusteralloc(pmp, cn, count, CLUST_EOFE, &cn, &got);
1024 		if (error)
1025 			return (error);
1026 
1027 		count -= got;
1028 
1029 		/*
1030 		 * Give them the filesystem relative cluster number if they want
1031 		 * it.
1032 		 */
1033 		if (ncp) {
1034 			*ncp = cn;
1035 			ncp = NULL;
1036 		}
1037 
1038 		if (dep->de_StartCluster == 0) {
1039 			dep->de_StartCluster = cn;
1040 			frcn = 0;
1041 		} else {
1042 			error = fatentry(FAT_SET, pmp,
1043 					 dep->de_fc[FC_LASTFC].fc_fsrcn,
1044 					 0, cn);
1045 			if (error) {
1046 				clusterfree(pmp, cn, NULL);
1047 				return (error);
1048 			}
1049 			frcn = dep->de_fc[FC_LASTFC].fc_frcn + 1;
1050 		}
1051 
1052 		/*
1053 		 * Update the "last cluster of the file" entry in the
1054 		 * denode's FAT cache.
1055 		 */
1056 		fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
1057 
1058 		if (flags & DE_CLEAR) {
1059 			while (got-- > 0) {
1060 				/*
1061 				 * Get the buf header for the new block of the file.
1062 				 */
1063 				if (dep->de_Attributes & ATTR_DIRECTORY)
1064 					bp = getblk(pmp->pm_devvp,
1065 					    cntobn(pmp, cn++),
1066 					    pmp->pm_bpcluster, 0, 0, 0);
1067 				else {
1068 					bp = getblk(DETOV(dep),
1069 					    frcn++,
1070 					    pmp->pm_bpcluster, 0, 0, 0);
1071 					/*
1072 					 * Do the bmap now, as in msdosfs_write
1073 					 */
1074 					if (pcbmap(dep,
1075 					    bp->b_lblkno,
1076 					    &blkno, 0, 0))
1077 						bp->b_blkno = -1;
1078 					if (bp->b_blkno == -1)
1079 						panic("extendfile: pcbmap");
1080 					else
1081 						bp->b_blkno = blkno;
1082 				}
1083 				clrbuf(bp);
1084 				if (bpp) {
1085 					*bpp = bp;
1086 					bpp = NULL;
1087 				} else {
1088 					bdwrite(bp);
1089 				}
1090 				if (vm_page_count_severe() ||
1091 				    buf_dirty_count_severe())
1092 					vn_fsync_buf(DETOV(dep), MNT_WAIT);
1093 			}
1094 		}
1095 	}
1096 
1097 	return (0);
1098 }
1099 
1100 /*-
1101  * Routine to mark a FAT16 or FAT32 volume as "clean" or "dirty" by
1102  * manipulating the upper bit of the FAT entry for cluster 1.  Note that
1103  * this bit is not defined for FAT12 volumes, which are always assumed to
1104  * be clean.
1105  *
1106  * The fatentry() routine only works on cluster numbers that a file could
1107  * occupy, so it won't manipulate the entry for cluster 1.  So we have to do
1108  * it here.  The code was stolen from fatentry() and tailored for cluster 1.
1109  *
1110  * Inputs:
1111  *	pmp	The MS-DOS volume to mark
1112  *	dirty	Non-zero if the volume should be marked dirty; zero if it
1113  *		should be marked clean
1114  *
1115  * Result:
1116  *	0	Success
1117  *	EROFS	Volume is read-only
1118  *	?	(other errors from called routines)
1119  */
1120 int
1121 markvoldirty_upgrade(struct msdosfsmount *pmp, bool dirty, bool rw_upgrade)
1122 {
1123 	struct buf *bp;
1124 	u_long bn, bo, bsize, byteoffset, fatval;
1125 	int error;
1126 
1127 	/*
1128 	 * FAT12 does not support a "clean" bit, so don't do anything for
1129 	 * FAT12.
1130 	 */
1131 	if (FAT12(pmp))
1132 		return (0);
1133 
1134 	/*
1135 	 * Can't change the bit on a read-only filesystem, except as part of
1136 	 * ro->rw upgrade.
1137 	 */
1138 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) != 0 && !rw_upgrade)
1139 		return (EROFS);
1140 
1141 	/*
1142 	 * Fetch the block containing the FAT entry.  It is given by the
1143 	 * pseudo-cluster 1.
1144 	 */
1145 	byteoffset = FATOFS(pmp, 1);
1146 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
1147 	error = bread(pmp->pm_devvp, bn, bsize, NOCRED, &bp);
1148 	if (error)
1149 		return (error);
1150 
1151 	/*
1152 	 * Get the current value of the FAT entry and set/clear the relevant
1153 	 * bit.  Dirty means clear the "clean" bit; clean means set the
1154 	 * "clean" bit.
1155 	 */
1156 	if (FAT32(pmp)) {
1157 		/* FAT32 uses bit 27. */
1158 		fatval = getulong(&bp->b_data[bo]);
1159 		if (dirty)
1160 			fatval &= 0xF7FFFFFF;
1161 		else
1162 			fatval |= 0x08000000;
1163 		putulong(&bp->b_data[bo], fatval);
1164 	} else {
1165 		/* Must be FAT16; use bit 15. */
1166 		fatval = getushort(&bp->b_data[bo]);
1167 		if (dirty)
1168 			fatval &= 0x7FFF;
1169 		else
1170 			fatval |= 0x8000;
1171 		putushort(&bp->b_data[bo], fatval);
1172 	}
1173 
1174 	/*
1175 	 * The concern here is that a devvp may be readonly, without reporting
1176 	 * itself as such through the usual channels.  In that case, we'd like
1177 	 * it if attempting to mount msdosfs rw didn't panic the system.
1178 	 *
1179 	 * markvoldirty is invoked as the first write on backing devvps when
1180 	 * either msdosfs is mounted for the first time, or a ro mount is
1181 	 * upgraded to rw.
1182 	 *
1183 	 * In either event, if a write error occurs dirtying the volume:
1184 	 *   - No user data has been permitted to be written to cache yet.
1185 	 *   - We can abort the high-level operation (mount, or ro->rw) safely.
1186 	 *   - We don't derive any benefit from leaving a zombie dirty buf in
1187 	 *   the cache that can not be cleaned or evicted.
1188 	 *
1189 	 * So, mark B_INVALONERR to have bwrite() -> brelse() detect that
1190 	 * condition and force-invalidate our write to the block if it occurs.
1191 	 *
1192 	 * PR 210316 provides more context on the discovery and diagnosis of
1193 	 * the problem, as well as earlier attempts to solve it.
1194 	 */
1195 	bp->b_flags |= B_INVALONERR;
1196 
1197 	/* Write out the modified FAT block synchronously. */
1198 	return (bwrite(bp));
1199 }
1200