xref: /dragonfly/sys/vfs/msdosfs/msdosfs_fat.c (revision 63e03116)
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/vnode.h>
58 
59 #include <sys/buf2.h>
60 
61 #include <vfs/msdosfs/bpb.h>
62 #include <vfs/msdosfs/direntry.h>
63 #include <vfs/msdosfs/denode.h>
64 #include <vfs/msdosfs/fat.h>
65 #include <vfs/msdosfs/msdosfsmount.h>
66 
67 #define	FULL_RUN	((u_int)0xffffffff)
68 
69 static void	fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp,
70 		    u_long *fsrcnp);
71 static int	clusteralloc1(struct msdosfsmount *pmp, u_long start,
72 		    u_long count, u_long fillwith, u_long *retcluster,
73 		    u_long *got);
74 
75 /*
76  * Given a byte offset `ofs` within FAT, return block number in backing device,
77  * block size, and byte offset within a block in FAT.
78  */
79 static void
80 fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp, u_long *sizep,
81     u_long *bop)
82 {
83 	u_long bn, size;
84 
85 	bn = ofs / pmp->pm_fatblocksize * pmp->pm_fatblocksec;
86 	size = min(pmp->pm_fatblocksec, pmp->pm_FATsecs - bn) * DEV_BSIZE;
87 	bn += pmp->pm_fatblk + pmp->pm_curfat * pmp->pm_FATsecs;
88 
89 	if (bnp)
90 		*bnp = bn;
91 	if (sizep)
92 		*sizep = size;
93 	if (bop)
94 		*bop = ofs % pmp->pm_fatblocksize;
95 }
96 
97 /*
98  * Map the logical cluster number of a file into a physical disk sector
99  * that is filesystem relative.
100  *
101  * dep	  - address of denode representing the file of interest
102  * findcn - file relative cluster whose filesystem relative cluster number
103  *	    and/or block number are/is to be found
104  * bnp	  - address of where to place the filesystem relative block number.
105  *	    If this pointer is null then don't return this quantity.
106  * cnp	  - address of where to place the filesystem relative cluster number.
107  *	    If this pointer is null then don't return this quantity.
108  * sp     - pointer to returned block size
109  *
110  * NOTE: Either bnp or cnp must be non-null.
111  * This function has one side effect.  If the requested file relative cluster
112  * is beyond the end of file, then the actual number of clusters in the file
113  * is returned in *cnp.  This is useful for determining how long a directory is.
114  *  If cnp is null, nothing is returned.
115  */
116 int
117 pcbmap(struct denode *dep, u_long findcn, daddr_t *bnp, u_long *cnp, int *sp)
118 {
119 	int error;
120 	u_long i;
121 	u_long cn;
122 	u_long prevcn = 0; /* XXX: prevcn could be used unititialized */
123 	u_long byteoffset;
124 	u_long bn;
125 	u_long bo;
126 	struct buf *bp = NULL;
127 	u_long bp_bn = -1;
128 	struct msdosfsmount *pmp = dep->de_pmp;
129 	u_long bsize;
130 
131 	KASSERT(bnp != NULL || cnp != NULL || sp != NULL,
132 	    ("pcbmap: extra call"));
133 	ASSERT_VOP_ELOCKED(DETOV(dep), "pcbmap");
134 
135 	cn = dep->de_StartCluster;
136 	/*
137 	 * The "file" that makes up the root directory is contiguous,
138 	 * permanently allocated, of fixed size, and is not made up of
139 	 * clusters.  If the cluster number is beyond the end of the root
140 	 * directory, then return the number of clusters in the file.
141 	 */
142 	if (cn == MSDOSFSROOT) {
143 		if (dep->de_Attributes & ATTR_DIRECTORY) {
144 			if (de_cn2off(pmp, findcn) >= dep->de_FileSize) {
145 				if (cnp)
146 					*cnp = de_bn2cn(pmp,
147 							pmp->pm_rootdirsize);
148 				return (E2BIG);
149 			}
150 			if (bnp)
151 				*bnp = pmp->pm_rootdirblk +
152 					de_cn2bn(pmp, findcn);
153 			if (cnp)
154 				*cnp = MSDOSFSROOT;
155 			if (sp)
156 				*sp = min(pmp->pm_bpcluster,
157 				    dep->de_FileSize - de_cn2off(pmp, findcn));
158 			return (0);
159 		} else {		/* just an empty file */
160 			if (cnp)
161 				*cnp = 0;
162 			return (E2BIG);
163 		}
164 	}
165 
166 	/*
167 	 * All other files do I/O in cluster sized blocks
168 	 */
169 	if (sp)
170 		*sp = pmp->pm_bpcluster;
171 
172 	/*
173 	 * Rummage around in the FAT cache, maybe we can avoid tromping
174 	 * through every FAT entry for the file. And, keep track of how far
175 	 * off the cache was from where we wanted to be.
176 	 */
177 	i = 0;
178 	fc_lookup(dep, findcn, &i, &cn);
179 
180 	/*
181 	 * Handle all other files or directories the normal way.
182 	 */
183 	for (; i < findcn; i++) {
184 		/*
185 		 * Stop with all reserved clusters, not just with EOF.
186 		 */
187 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
188 			goto hiteof;
189 		byteoffset = FATOFS(pmp, cn);
190 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
191 		if (bn != bp_bn) {
192 			if (bp)
193 				brelse(bp);
194 			error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn),
195 				      bsize, &bp);
196 			if (error) {
197 				brelse(bp);
198 				return (error);
199 			}
200 			bp_bn = bn;
201 		}
202 		prevcn = cn;
203 		if (bo >= bsize) {
204 			if (bp)
205 				brelse(bp);
206 			return (EIO);
207 		}
208 		if (FAT32(pmp))
209 			cn = getulong(bp->b_data + bo);
210 		else
211 			cn = getushort(bp->b_data + bo);
212 		if (FAT12(pmp) && (prevcn & 1))
213 			cn >>= 4;
214 		cn &= pmp->pm_fatmask;
215 
216 		/*
217 		 * Force the special cluster numbers
218 		 * to be the same for all cluster sizes
219 		 * to let the rest of msdosfs handle
220 		 * all cases the same.
221 		 */
222 		if ((cn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
223 			cn |= ~pmp->pm_fatmask;
224 	}
225 
226 	if (!MSDOSFSEOF(pmp, cn)) {
227 		if (bp)
228 			brelse(bp);
229 		if (bnp)
230 			*bnp = cntobn(pmp, cn);
231 		if (cnp)
232 			*cnp = cn;
233 		fc_setcache(dep, FC_LASTMAP, i, cn);
234 		return (0);
235 	}
236 
237 hiteof:
238 	if (cnp)
239 		*cnp = i;
240 	if (bp)
241 		brelse(bp);
242 	/* update last file cluster entry in the FAT cache */
243 	fc_setcache(dep, FC_LASTFC, i - 1, prevcn);
244 	return (E2BIG);
245 }
246 
247 /*
248  * Find the closest entry in the FAT cache to the cluster we are looking
249  * for.
250  */
251 static void
252 fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp, u_long *fsrcnp)
253 {
254 	int i;
255 	u_long cn;
256 	struct fatcache *closest = NULL;
257 
258 	ASSERT_VOP_LOCKED(DETOV(dep), "fc_lookup");
259 
260 	for (i = 0; i < FC_SIZE; i++) {
261 		cn = dep->de_fc[i].fc_frcn;
262 		if (cn != FCE_EMPTY && cn <= findcn) {
263 			if (closest == NULL || cn > closest->fc_frcn)
264 				closest = &dep->de_fc[i];
265 		}
266 	}
267 	if (closest) {
268 		*frcnp = closest->fc_frcn;
269 		*fsrcnp = closest->fc_fsrcn;
270 	}
271 }
272 
273 /*
274  * Purge the FAT cache in denode dep of all entries relating to file
275  * relative cluster frcn and beyond.
276  */
277 void
278 fc_purge(struct denode *dep, u_int frcn)
279 {
280 	int i;
281 	struct fatcache *fcp;
282 
283 	ASSERT_VOP_ELOCKED(DETOV(dep), "fc_purge");
284 
285 	fcp = dep->de_fc;
286 	for (i = 0; i < FC_SIZE; i++, fcp++) {
287 		if (fcp->fc_frcn >= frcn)
288 			fcp->fc_frcn = FCE_EMPTY;
289 	}
290 }
291 
292 /*
293  * Update the FAT.
294  * If mirroring the FAT, update all copies, with the first copy as last.
295  * Else update only the current FAT (ignoring the others).
296  *
297  * pmp	 - msdosfsmount structure for filesystem to update
298  * bp	 - addr of modified FAT block
299  * fatbn - block number relative to begin of filesystem of the modified FAT block.
300  */
301 static void
302 updatefats(struct msdosfsmount *pmp, struct buf *bp, u_long fatbn)
303 {
304 	struct buf *bpn;
305 	int cleanfat, i;
306 
307 	mprintf("updatefats(pmp %p, bp %p, fatbn %lu)\n", pmp, bp, fatbn);
308 
309 	if (pmp->pm_flags & MSDOSFS_FATMIRROR) {
310 		/*
311 		 * Now copy the block(s) of the modified FAT to the other copies of
312 		 * the FAT and write them out.  This is faster than reading in the
313 		 * other FATs and then writing them back out.  This could tie up
314 		 * the FAT for quite a while. Preventing others from accessing it.
315 		 * To prevent us from going after the FAT quite so much we use
316 		 * delayed writes, unless they specified "synchronous" when the
317 		 * filesystem was mounted.  If synch is asked for then use
318 		 * bwrite()'s and really slow things down.
319 		 */
320 		if (fatbn != pmp->pm_fatblk || FAT12(pmp))
321 			cleanfat = 0;
322 		else if (FAT16(pmp))
323 			cleanfat = 16;
324 		else
325 			cleanfat = 32;
326 		for (i = 1; i < pmp->pm_FATs; i++) {
327 			fatbn += pmp->pm_FATsecs;
328 			/* getblk() never fails */
329 			bpn = getblk(pmp->pm_devvp, de_bn2doff(pmp, fatbn),
330 				     bp->b_bcount, 0, 0);
331 			memcpy(bpn->b_data, bp->b_data, bp->b_bcount);
332 			/* Force the clean bit on in the other copies. */
333 			if (cleanfat == 16)
334 				((uint8_t *)bpn->b_data)[3] |= 0x80;
335 			else if (cleanfat == 32)
336 				((uint8_t *)bpn->b_data)[7] |= 0x08;
337 			if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
338 				bwrite(bpn);
339 			else
340 				bdwrite(bpn);
341 		}
342 	}
343 
344 	/*
345 	 * Write out the first (or current) FAT last.
346 	 */
347 	if (pmp->pm_mountp->mnt_flag & MNT_SYNCHRONOUS)
348 		bwrite(bp);
349 	else
350 		bdwrite(bp);
351 }
352 
353 /*
354  * Updating entries in 12 bit FATs is a pain in the butt.
355  *
356  * The following picture shows where nibbles go when moving from a 12 bit
357  * cluster number into the appropriate bytes in the FAT.
358  *
359  *	byte m        byte m+1      byte m+2
360  *	+----+----+   +----+----+   +----+----+
361  *	|  0    1 |   |  2    3 |   |  4    5 |   FAT bytes
362  *	+----+----+   +----+----+   +----+----+
363  *
364  *	+----+----+----+   +----+----+----+
365  *	|  3    0    1 |   |  4    5    2 |
366  *	+----+----+----+   +----+----+----+
367  *	cluster n  	   cluster n+1
368  *
369  * Where n is even. m = n + (n >> 2)
370  *
371  */
372 static __inline void
373 usemap_alloc(struct msdosfsmount *pmp, u_long cn)
374 {
375 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
376 
377 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
378 	    pmp->pm_maxcluster));
379 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
380 	    ("usemap_alloc on ro msdosfs mount"));
381 	KASSERT((pmp->pm_inusemap[cn / N_INUSEBITS] &
382 	    (1U << (cn % N_INUSEBITS))) == 0,
383 	    ("Allocating used sector %ld %ld %x", cn, cn % N_INUSEBITS,
384 	    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]));
385 	pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
386 	KASSERT(pmp->pm_freeclustercount > 0, ("usemap_alloc: too little"));
387 	pmp->pm_freeclustercount--;
388 	pmp->pm_flags |= MSDOSFS_FSIMOD;
389 }
390 
391 static int
392 usemap_free(struct msdosfsmount *pmp, u_long cn)
393 {
394 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
395 
396 	KASSERT(cn <= pmp->pm_maxcluster, ("cn too large %lu %lu", cn,
397 	    pmp->pm_maxcluster));
398 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
399 	    ("usemap_free on ro msdosfs mount"));
400 	if ((pmp->pm_inusemap[cn / N_INUSEBITS] &
401 	    (1U << (cn % N_INUSEBITS))) == 0) {
402 		kprintf("%s: Freeing unused sector %ld %ld %x\n",
403 		    pmp->pm_mountp->mnt_stat.f_mntonname, cn, cn % N_INUSEBITS,
404 		    (unsigned)pmp->pm_inusemap[cn / N_INUSEBITS]);
405 		return (EINVAL);
406 	}
407 	pmp->pm_freeclustercount++;
408 	pmp->pm_flags |= MSDOSFS_FSIMOD;
409 	pmp->pm_inusemap[cn / N_INUSEBITS] &= ~(1U << (cn % N_INUSEBITS));
410 	return (0);
411 }
412 
413 void
414 clusterfree(struct msdosfsmount *pmp, u_long cluster)
415 {
416 	int error;
417 	u_long oldcn;
418 
419 	error = fatentry(FAT_GET_AND_SET, pmp, cluster, &oldcn, MSDOSFSFREE);
420 	if (error != 0)
421 		return;
422 	/*
423 	 * If the cluster was successfully marked free, then update
424 	 * the count of free clusters, and turn off the "allocated"
425 	 * bit in the "in use" cluster bit map.
426 	 */
427 	MSDOSFS_LOCK_MP(pmp);
428 	error = usemap_free(pmp, cluster);
429 	MSDOSFS_UNLOCK_MP(pmp);
430 }
431 
432 /*
433  * Get or Set or 'Get and Set' the cluster'th entry in the FAT.
434  *
435  * function	- whether to get or set a FAT entry
436  * pmp		- address of the msdosfsmount structure for the filesystem
437  *		  whose FAT is to be manipulated.
438  * cn		- which cluster is of interest
439  * oldcontents	- address of a word that is to receive the contents of the
440  *		  cluster'th entry if this is a get function
441  * newcontents	- the new value to be written into the cluster'th element of
442  *		  the FAT if this is a set function.
443  *
444  * This function can also be used to free a cluster by setting the FAT entry
445  * for a cluster to 0.
446  *
447  * All copies of the FAT are updated if this is a set function. NOTE: If
448  * fatentry() marks a cluster as free it does not update the inusemap in
449  * the msdosfsmount structure. This is left to the caller.
450  */
451 int
452 fatentry(int function, struct msdosfsmount *pmp, u_long cn, u_long *oldcontents,
453     u_long newcontents)
454 {
455 	int error;
456 	u_long readcn;
457 	u_long bn, bo, bsize, byteoffset;
458 	struct buf *bp;
459 
460 	mprintf("fatentry(func %d, pmp %p, clust %lu, oldcon %p, newcon %lx)\n",
461 		function, pmp, cn, oldcontents, newcontents);
462 
463 #ifdef DIAGNOSTIC
464 	/*
465 	 * Be sure they asked us to do something.
466 	 */
467 	if ((function & (FAT_SET | FAT_GET)) == 0) {
468 		kprintf("fatentry(): function code doesn't specify get or set\n");
469 		return (EINVAL);
470 	}
471 
472 	/*
473 	 * If they asked us to return a cluster number but didn't tell us
474 	 * where to put it, give them an error.
475 	 */
476 	if ((function & FAT_GET) && oldcontents == NULL) {
477 		kprintf("fatentry(): get function with no place to put result\n");
478 		return (EINVAL);
479 	}
480 #endif
481 
482 	/*
483 	 * Be sure the requested cluster is in the filesystem.
484 	 */
485 	if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster)
486 		return (EINVAL);
487 
488 	byteoffset = FATOFS(pmp, cn);
489 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
490 	error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn), bsize, &bp);
491 	if (error) {
492 		brelse(bp);
493 		return (error);
494 	}
495 
496 	if (function & FAT_GET) {
497 		if (FAT32(pmp))
498 			readcn = getulong(bp->b_data + bo);
499 		else
500 			readcn = getushort(bp->b_data + bo);
501 		if (FAT12(pmp) & (cn & 1))
502 			readcn >>= 4;
503 		readcn &= pmp->pm_fatmask;
504 		/* map reserved FAT entries to same values for all FATs */
505 		if ((readcn | ~pmp->pm_fatmask) >= CLUST_RSRVD)
506 			readcn |= ~pmp->pm_fatmask;
507 		*oldcontents = readcn;
508 	}
509 	if (function & FAT_SET) {
510 		switch (pmp->pm_fatmask) {
511 		case FAT12_MASK:
512 			readcn = getushort(bp->b_data + bo);
513 			if (cn & 1) {
514 				readcn &= 0x000f;
515 				readcn |= newcontents << 4;
516 			} else {
517 				readcn &= 0xf000;
518 				readcn |= newcontents & 0xfff;
519 			}
520 			putushort(bp->b_data + bo, readcn);
521 			break;
522 		case FAT16_MASK:
523 			putushort(bp->b_data + bo, newcontents);
524 			break;
525 		case FAT32_MASK:
526 			/*
527 			 * According to spec we have to retain the
528 			 * high order bits of the FAT entry.
529 			 */
530 			readcn = getulong(bp->b_data + bo);
531 			readcn &= ~FAT32_MASK;
532 			readcn |= newcontents & FAT32_MASK;
533 			putulong(bp->b_data + bo, readcn);
534 			break;
535 		}
536 		updatefats(pmp, bp, bn);
537 		bp = NULL;
538 		pmp->pm_fmod = 1;
539 	}
540 	if (bp)
541 		brelse(bp);
542 	return (0);
543 }
544 
545 /*
546  * Update a contiguous cluster chain
547  *
548  * pmp	    - mount point
549  * start    - first cluster of chain
550  * count    - number of clusters in chain
551  * fillwith - what to write into FAT entry of last cluster
552  */
553 static int
554 fatchain(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith)
555 {
556 	int error;
557 	u_long bn, bo, bsize, byteoffset, readcn, newc;
558 	struct buf *bp;
559 
560 	mprintf("fatchain(pmp %p, start %lu, count %lu, fillwith %lx)\n",
561 	    pmp, start, count, fillwith);
562 	/*
563 	 * Be sure the clusters are in the filesystem.
564 	 */
565 	if (start < CLUST_FIRST || start + count - 1 > pmp->pm_maxcluster)
566 		return (EINVAL);
567 
568 	while (count > 0) {
569 		byteoffset = FATOFS(pmp, start);
570 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
571 		error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn), bsize, &bp);
572 		if (error) {
573 			brelse(bp);
574 			return (error);
575 		}
576 		while (count > 0) {
577 			start++; /* Callers must guarantee contiguous free
578 				    clusters. */
579 			newc = --count > 0 ? start : fillwith;
580 			switch (pmp->pm_fatmask) {
581 			case FAT12_MASK:
582 				readcn = getushort(bp->b_data + bo);
583 				if (start & 1) {
584 					readcn &= 0xf000;
585 					readcn |= newc & 0xfff;
586 				} else {
587 					readcn &= 0x000f;
588 					readcn |= newc << 4;
589 				}
590 				putushort(bp->b_data + bo, readcn);
591 				bo++;
592 				if (!(start & 1))
593 					bo++;
594 				break;
595 			case FAT16_MASK:
596 				putushort(bp->b_data + bo, newc);
597 				bo += 2;
598 				break;
599 			case FAT32_MASK:
600 				readcn = getulong(bp->b_data + bo);
601 				readcn &= ~pmp->pm_fatmask;
602 				readcn |= newc & pmp->pm_fatmask;
603 				putulong(bp->b_data + bo, readcn);
604 				bo += 4;
605 				break;
606 			}
607 			if (bo >= bsize)
608 				break;
609 		}
610 		updatefats(pmp, bp, bn);
611 	}
612 	pmp->pm_fmod = 1;
613 	return (0);
614 }
615 
616 /*
617  * Check the length of a free cluster chain starting at start.
618  *
619  * pmp	 - mount point
620  * start - start of chain
621  * count - maximum interesting length
622  */
623 static int
624 chainlength(struct msdosfsmount *pmp, u_long start, u_long count)
625 {
626 	u_long idx, max_idx;
627 	u_int map;
628 	u_long len;
629 
630 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
631 
632 	if (start > pmp->pm_maxcluster)
633 		return (0);
634 	max_idx = pmp->pm_maxcluster / N_INUSEBITS;
635 	idx = start / N_INUSEBITS;
636 	start %= N_INUSEBITS;
637 	map = pmp->pm_inusemap[idx];
638 	map &= ~((1U << start) - 1);
639 	if (map) {
640 		len = ffs(map) - 1 - start;
641 		len = MIN(len, count);
642 		if (start + len > pmp->pm_maxcluster)
643 			len = pmp->pm_maxcluster - start + 1;
644 		return (len);
645 	}
646 	len = N_INUSEBITS - start;
647 	if (len >= count) {
648 		len = count;
649 		if (start + len > pmp->pm_maxcluster)
650 			len = pmp->pm_maxcluster - start + 1;
651 		return (len);
652 	}
653 	while (++idx <= max_idx) {
654 		if (len >= count)
655 			break;
656 		map = pmp->pm_inusemap[idx];
657 		if (map) {
658 			len += ffs(map) - 1;
659 			break;
660 		}
661 		len += N_INUSEBITS;
662 	}
663 	len = MIN(len, count);
664 	if (start + len > pmp->pm_maxcluster)
665 		len = pmp->pm_maxcluster - start + 1;
666 	return (len);
667 }
668 
669 /*
670  * Allocate contigous free clusters.
671  *
672  * pmp	      - mount point.
673  * start      - start of cluster chain.
674  * count      - number of clusters to allocate.
675  * fillwith   - put this value into the FAT entry for the
676  *		last allocated cluster.
677  * retcluster - put the first allocated cluster's number here.
678  * got	      - how many clusters were actually allocated.
679  */
680 static int
681 chainalloc(struct msdosfsmount *pmp, u_long start, u_long count,
682     u_long fillwith, u_long *retcluster, u_long *got)
683 {
684 	int error;
685 	u_long cl, n;
686 
687 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
688 	KASSERT((pmp->pm_flags & MSDOSFSMNT_RONLY) == 0,
689 	    ("chainalloc on ro msdosfs mount"));
690 
691 	for (cl = start, n = count; n-- > 0;)
692 		usemap_alloc(pmp, cl++);
693 
694 	pmp->pm_nxtfree = start + count;
695 	if (pmp->pm_nxtfree > pmp->pm_maxcluster)
696 		pmp->pm_nxtfree = CLUST_FIRST;
697 	pmp->pm_flags |= MSDOSFS_FSIMOD;
698 
699 	error = fatchain(pmp, start, count, fillwith);
700 	if (error != 0) {
701 		for (cl = start, n = count; n-- > 0;)
702 			(void)usemap_free(pmp, cl++);
703 		return (error);
704 	}
705 	mprintf("clusteralloc(): allocated cluster chain at %lu (%lu clusters)\n",
706 	    start, count);
707 	if (retcluster)
708 		*retcluster = start;
709 	if (got)
710 		*got = count;
711 	return (0);
712 }
713 
714 /*
715  * Allocate contiguous free clusters.
716  *
717  * pmp	      - mount point.
718  * start      - preferred start of cluster chain.
719  * count      - number of clusters requested.
720  * fillwith   - put this value into the FAT entry for the
721  *		last allocated cluster.
722  * retcluster - put the first allocated cluster's number here.
723  * got	      - how many clusters were actually allocated.
724  */
725 int
726 clusteralloc(struct msdosfsmount *pmp, u_long start, u_long count,
727     u_long fillwith, u_long *retcluster, u_long *got)
728 {
729 	int error;
730 
731 	MSDOSFS_LOCK_MP(pmp);
732 	error = clusteralloc1(pmp, start, count, fillwith, retcluster, got);
733 	MSDOSFS_UNLOCK_MP(pmp);
734 	return (error);
735 }
736 
737 static int
738 clusteralloc1(struct msdosfsmount *pmp, u_long start, u_long count,
739     u_long fillwith, u_long *retcluster, u_long *got)
740 {
741 	u_long idx;
742 	u_long len, newst, foundl, cn, l;
743 	u_long foundcn = 0; /* XXX: foundcn could be used unititialized */
744 	u_int map;
745 
746 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
747 
748 	mprintf("clusteralloc(): find %lu clusters\n",count);
749 	if (start) {
750 		if ((len = chainlength(pmp, start, count)) >= count)
751 			return (chainalloc(pmp, start, count, fillwith,
752 				retcluster, got));
753 	} else
754 		len = 0;
755 
756 	newst = pmp->pm_nxtfree;
757 	foundl = 0;
758 
759 	for (cn = newst; cn <= pmp->pm_maxcluster;) {
760 		idx = cn / N_INUSEBITS;
761 		map = pmp->pm_inusemap[idx];
762 		map |= (1U << (cn % N_INUSEBITS)) - 1;
763 		if (map != FULL_RUN) {
764 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
765 			if ((l = chainlength(pmp, cn, count)) >= count)
766 				return (chainalloc(pmp, cn, count, fillwith,
767 					retcluster, got));
768 			if (l > foundl) {
769 				foundcn = cn;
770 				foundl = l;
771 			}
772 			cn += l + 1;
773 			continue;
774 		}
775 		cn += N_INUSEBITS - cn % N_INUSEBITS;
776 	}
777 	for (cn = 0; cn < newst;) {
778 		idx = cn / N_INUSEBITS;
779 		map = pmp->pm_inusemap[idx];
780 		map |= (1U << (cn % N_INUSEBITS)) - 1;
781 		if (map != FULL_RUN) {
782 			cn = idx * N_INUSEBITS + ffs(map ^ FULL_RUN) - 1;
783 			if ((l = chainlength(pmp, cn, count)) >= count)
784 				return (chainalloc(pmp, cn, count, fillwith,
785 					retcluster, got));
786 			if (l > foundl) {
787 				foundcn = cn;
788 				foundl = l;
789 			}
790 			cn += l + 1;
791 			continue;
792 		}
793 		cn += N_INUSEBITS - cn % N_INUSEBITS;
794 	}
795 
796 	if (!foundl)
797 		return (ENOSPC);
798 
799 	if (len)
800 		return (chainalloc(pmp, start, len, fillwith, retcluster, got));
801 	else
802 		return (chainalloc(pmp, foundcn, foundl, fillwith, retcluster,
803 			got));
804 }
805 
806 /*
807  * Free a chain of clusters.
808  *
809  * pmp		- address of the msdosfs mount structure for the filesystem
810  *		  containing the cluster chain to be freed.
811  * startcluster - number of the 1st cluster in the chain of clusters to be
812  *		  freed.
813  */
814 int
815 freeclusterchain(struct msdosfsmount *pmp, u_long cluster)
816 {
817 	int error;
818 	struct buf *bp = NULL;
819 	u_long bn, bo, bsize, byteoffset;
820 	u_long readcn, lbn = -1;
821 
822 	MSDOSFS_LOCK_MP(pmp);
823 	while (cluster >= CLUST_FIRST && cluster <= pmp->pm_maxcluster) {
824 		byteoffset = FATOFS(pmp, cluster);
825 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
826 		if (lbn != bn) {
827 			if (bp)
828 				updatefats(pmp, bp, lbn);
829 			error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn),
830 				      bsize, &bp);
831 			if (error) {
832 				brelse(bp);
833 				MSDOSFS_UNLOCK_MP(pmp);
834 				return (error);
835 			}
836 			lbn = bn;
837 		}
838 		error = usemap_free(pmp, cluster);
839 		if (error != 0) {
840 			updatefats(pmp, bp, lbn);
841 			MSDOSFS_UNLOCK_MP(pmp);
842 			return (error);
843 		}
844 		switch (pmp->pm_fatmask) {
845 		case FAT12_MASK:
846 			readcn = getushort(bp->b_data + bo);
847 			if (cluster & 1) {
848 				cluster = readcn >> 4;
849 				readcn &= 0x000f;
850 				readcn |= MSDOSFSFREE << 4;
851 			} else {
852 				cluster = readcn;
853 				readcn &= 0xf000;
854 				readcn |= MSDOSFSFREE & 0xfff;
855 			}
856 			putushort(bp->b_data + bo, readcn);
857 			break;
858 		case FAT16_MASK:
859 			cluster = getushort(bp->b_data + bo);
860 			putushort(bp->b_data + bo, MSDOSFSFREE);
861 			break;
862 		case FAT32_MASK:
863 			cluster = getulong(bp->b_data + bo);
864 			putulong(bp->b_data + bo,
865 				 (MSDOSFSFREE & FAT32_MASK) |
866 				 (cluster & ~FAT32_MASK));
867 			break;
868 		}
869 		cluster &= pmp->pm_fatmask;
870 		if ((cluster | ~pmp->pm_fatmask) >= CLUST_RSRVD)
871 			cluster |= pmp->pm_fatmask;
872 	}
873 	if (bp)
874 		updatefats(pmp, bp, bn);
875 	MSDOSFS_UNLOCK_MP(pmp);
876 	return (0);
877 }
878 
879 /*
880  * Read in FAT blocks looking for free clusters. For every free cluster
881  * found turn off its corresponding bit in the pm_inusemap.
882  */
883 int
884 fillinusemap(struct msdosfsmount *pmp)
885 {
886 	struct buf *bp;
887 	u_long bn, bo, bsize, byteoffset, cn, readcn;
888 	int error;
889 
890 	MSDOSFS_ASSERT_MP_LOCKED(pmp);
891 	bp = NULL;
892 
893 	/*
894 	 * Mark all clusters in use, we mark the free ones in the FAT scan
895 	 * loop further down.
896 	 */
897 	for (cn = 0; cn < (pmp->pm_maxcluster + N_INUSEBITS) / N_INUSEBITS; cn++)
898 		pmp->pm_inusemap[cn] = FULL_RUN;
899 
900 	/*
901 	 * Figure how many free clusters are in the filesystem by ripping
902 	 * through the FAT counting the number of entries whose content is
903 	 * zero.  These represent free clusters.
904 	 */
905 	pmp->pm_freeclustercount = 0;
906 	for (cn = 0; cn <= pmp->pm_maxcluster; cn++) {
907 		byteoffset = FATOFS(pmp, cn);
908 		bo = byteoffset % pmp->pm_fatblocksize;
909 		if (bo == 0) {
910 			/* Read new FAT block */
911 			if (bp != NULL)
912 				brelse(bp);
913 			fatblock(pmp, byteoffset, &bn, &bsize, NULL);
914 			error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn),
915 				      bsize, &bp);
916 			if (error != 0) {
917 				brelse(bp);
918 				return (error);
919 			}
920 		}
921 		if (FAT32(pmp))
922 			readcn = getulong(bp->b_data + bo);
923 		else
924 			readcn = getushort(bp->b_data + bo);
925 		if (FAT12(pmp) && (cn & 1))
926 			readcn >>= 4;
927 		readcn &= pmp->pm_fatmask;
928 
929 		/*
930 		 * Check if the FAT ID matches the BPB's media descriptor and
931 		 * all other bits are set to 1.
932 		 */
933 		if (cn == 0 && readcn != ((pmp->pm_fatmask & 0xffffff00) |
934 		    pmp->pm_bpb.bpbMedia)) {
935 			mprintf("fillinusemap(): Media descriptor in BPB "
936 			    "does not match FAT ID\n");
937 			brelse(bp);
938 			return (EINVAL);
939 		} else if (readcn == CLUST_FREE) {
940 			error = usemap_free(pmp, cn);
941 			if (error != 0) {
942 				brelse(bp);
943 				return (error);
944 			}
945 		}
946 	}
947 	if (bp != NULL)
948 		brelse(bp);
949 
950 	for (cn = pmp->pm_maxcluster + 1; cn < (pmp->pm_maxcluster +
951 	    N_INUSEBITS) / N_INUSEBITS; cn++)
952 		pmp->pm_inusemap[cn / N_INUSEBITS] |= 1U << (cn % N_INUSEBITS);
953 
954 	return (0);
955 }
956 
957 /*
958  * Allocate a new cluster and chain it onto the end of the file.
959  *
960  * dep	 - the file to extend
961  * count - number of clusters to allocate
962  * bpp	 - where to return the address of the buf header for the first new
963  *	   file block
964  * ncp	 - where to put cluster number of the first newly allocated cluster
965  *	   If this pointer is 0, do not return the cluster number.
966  * flags - see fat.h
967  *
968  * NOTE: This function is not responsible for turning on the DE_UPDATE bit of
969  * the de_flag field of the denode and it does not change the de_FileSize
970  * field.  This is left for the caller to do.
971  */
972 int
973 extendfile(struct denode *dep, u_long count, struct buf **bpp, u_long *ncp,
974     int flags)
975 {
976 	int error;
977 	u_long frcn;
978 	u_long cn, got;
979 	struct msdosfsmount *pmp = dep->de_pmp;
980 	struct buf *bp;
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 		kprintf("extendfile(): attempt to extend root directory\n");
988 		return (ENOSPC);
989 	}
990 
991 	/*
992 	 * If the "file's last cluster" cache entry is empty, and the file
993 	 * is not empty, then fill the cache entry by calling pcbmap().
994 	 */
995 	if (dep->de_fc[FC_LASTFC].fc_frcn == FCE_EMPTY &&
996 	    dep->de_StartCluster != 0) {
997 		error = pcbmap(dep, 0xffff, NULL, &cn, NULL);
998 		/* we expect it to return E2BIG */
999 		if (error != E2BIG)
1000 			return (error);
1001 	}
1002 
1003 	dep->de_fc[FC_NEXTTOLASTFC].fc_frcn =
1004 	    dep->de_fc[FC_LASTFC].fc_frcn;
1005 	dep->de_fc[FC_NEXTTOLASTFC].fc_fsrcn =
1006 	    dep->de_fc[FC_LASTFC].fc_fsrcn;
1007 	while (count > 0) {
1008 		/*
1009 		 * Allocate a new cluster chain and cat onto the end of the
1010 		 * file.  If the file is empty we make de_StartCluster point
1011 		 * to the new block.  Note that de_StartCluster being 0 is
1012 		 * sufficient to be sure the file is empty since we exclude
1013 		 * attempts to extend the root directory above, and the root
1014 		 * dir is the only file with a startcluster of 0 that has
1015 		 * blocks allocated (sort of).
1016 		 */
1017 		if (dep->de_StartCluster == 0)
1018 			cn = 0;
1019 		else
1020 			cn = dep->de_fc[FC_LASTFC].fc_fsrcn + 1;
1021 		error = clusteralloc(pmp, cn, count, CLUST_EOFE, &cn, &got);
1022 		if (error)
1023 			return (error);
1024 
1025 		count -= got;
1026 
1027 		/*
1028 		 * Give them the filesystem relative cluster number if they want
1029 		 * it.
1030 		 */
1031 		if (ncp) {
1032 			*ncp = cn;
1033 			ncp = NULL;
1034 		}
1035 
1036 		if (dep->de_StartCluster == 0) {
1037 			dep->de_StartCluster = cn;
1038 			frcn = 0;
1039 		} else {
1040 			error = fatentry(FAT_SET, pmp,
1041 					 dep->de_fc[FC_LASTFC].fc_fsrcn,
1042 					 0, cn);
1043 			if (error) {
1044 				clusterfree(pmp, cn);
1045 				return (error);
1046 			}
1047 			frcn = dep->de_fc[FC_LASTFC].fc_frcn + 1;
1048 		}
1049 
1050 		/*
1051 		 * Update the "last cluster of the file" entry in the
1052 		 * denode's FAT cache.
1053 		 */
1054 		fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
1055 
1056 		if (flags & DE_CLEAR) {
1057 			while (got-- > 0) {
1058 				/*
1059 				 * Get the buf header for the new block of the file.
1060 				 */
1061 				if (dep->de_Attributes & ATTR_DIRECTORY) {
1062 					bp = getblk(pmp->pm_devvp,
1063 						    de_bn2doff(pmp, cntobn(pmp, cn)),
1064 						    pmp->pm_bpcluster, 0, 0);
1065 					++cn;
1066 				} else {
1067 					daddr_t dblkno;
1068 					u_long findcn;
1069 
1070 					bp = getblk(DETOV(dep),
1071 						    de_cn2doff(pmp, frcn),
1072 						    pmp->pm_bpcluster, 0, 0);
1073 					++frcn;
1074 					/*
1075 					 * Convert bio1 offset to file relative
1076 					 * cluster number.
1077 					 */
1078 					findcn = de_bn2cn(pmp,
1079 					    (daddr_t)(bp->b_bio1.bio_offset >>
1080 					    pmp->pm_bnshift));
1081 					/*
1082 					 * Do the bmap now, as in msdosfs_write
1083 					 */
1084 					if (pcbmap(dep, findcn, &dblkno, NULL,
1085 					    NULL)) {
1086 						bp->b_bio2.bio_offset = NOOFFSET;
1087 					} else {
1088 						bp->b_bio2.bio_offset = de_bn2doff(pmp, dblkno);
1089 					}
1090 					if (bp->b_bio2.bio_offset == NOOFFSET)
1091 						panic("extendfile: pcbmap");
1092 				}
1093 				clrbuf(bp);
1094 				if (bpp) {
1095 					*bpp = bp;
1096 					bpp = NULL;
1097 				} else {
1098 					bdwrite(bp);
1099 				}
1100 			}
1101 		}
1102 	}
1103 
1104 	return (0);
1105 }
1106 
1107 /*-
1108  * Routine to mark a FAT16 or FAT32 volume as "clean" or "dirty" by
1109  * manipulating the upper bit of the FAT entry for cluster 1.  Note that
1110  * this bit is not defined for FAT12 volumes, which are always assumed to
1111  * be clean.
1112  *
1113  * The fatentry() routine only works on cluster numbers that a file could
1114  * occupy, so it won't manipulate the entry for cluster 1.  So we have to do
1115  * it here.  The code was stolen from fatentry() and tailored for cluster 1.
1116  *
1117  * Inputs:
1118  *	pmp	The MS-DOS volume to mark
1119  *	dirty	Non-zero if the volume should be marked dirty; zero if it
1120  *		should be marked clean
1121  *
1122  * Result:
1123  *	0	Success
1124  *	EROFS	Volume is read-only
1125  *	?	(other errors from called routines)
1126  */
1127 int
1128 markvoldirty_upgrade(struct msdosfsmount *pmp, bool dirty, bool rw_upgrade)
1129 {
1130 	struct buf *bp;
1131 	u_long bn, bo, bsize, byteoffset, fatval;
1132 	int error;
1133 
1134 	/*
1135 	 * FAT12 does not support a "clean" bit, so don't do anything for
1136 	 * FAT12.
1137 	 */
1138 	if (FAT12(pmp))
1139 		return (0);
1140 
1141 	/*
1142 	 * Can't change the bit on a read-only filesystem, except as part of
1143 	 * ro->rw upgrade.
1144 	 */
1145 	if ((pmp->pm_flags & MSDOSFSMNT_RONLY) != 0 && !rw_upgrade)
1146 		return (EROFS);
1147 
1148 	/*
1149 	 * Fetch the block containing the FAT entry.  It is given by the
1150 	 * pseudo-cluster 1.
1151 	 */
1152 	byteoffset = FATOFS(pmp, 1);
1153 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
1154 	error = bread(pmp->pm_devvp, de_bn2doff(pmp, bn), bsize, &bp);
1155 	if (error)
1156 		return (error);
1157 
1158 	/*
1159 	 * Get the current value of the FAT entry and set/clear the relevant
1160 	 * bit.  Dirty means clear the "clean" bit; clean means set the
1161 	 * "clean" bit.
1162 	 */
1163 	if (FAT32(pmp)) {
1164 		/* FAT32 uses bit 27. */
1165 		fatval = getulong(&bp->b_data[bo]);
1166 		if (dirty)
1167 			fatval &= 0xF7FFFFFF;
1168 		else
1169 			fatval |= 0x08000000;
1170 		putulong(&bp->b_data[bo], fatval);
1171 	} else {
1172 		/* Must be FAT16; use bit 15. */
1173 		fatval = getushort(&bp->b_data[bo]);
1174 		if (dirty)
1175 			fatval &= 0x7FFF;
1176 		else
1177 			fatval |= 0x8000;
1178 		putushort(&bp->b_data[bo], fatval);
1179 	}
1180 #if 0
1181 	/*
1182 	 * The concern here is that a devvp may be readonly, without reporting
1183 	 * itself as such through the usual channels.  In that case, we'd like
1184 	 * it if attempting to mount msdosfs rw didn't panic the system.
1185 	 *
1186 	 * markvoldirty is invoked as the first write on backing devvps when
1187 	 * either msdosfs is mounted for the first time, or a ro mount is
1188 	 * upgraded to rw.
1189 	 *
1190 	 * In either event, if a write error occurs dirtying the volume:
1191 	 *   - No user data has been permitted to be written to cache yet.
1192 	 *   - We can abort the high-level operation (mount, or ro->rw) safely.
1193 	 *   - We don't derive any benefit from leaving a zombie dirty buf in
1194 	 *   the cache that can not be cleaned or evicted.
1195 	 *
1196 	 * So, mark B_INVALONERR to have bwrite() -> brelse() detect that
1197 	 * condition and force-invalidate our write to the block if it occurs.
1198 	 *
1199 	 * PR 210316 provides more context on the discovery and diagnosis of
1200 	 * the problem, as well as earlier attempts to solve it.
1201 	 */
1202 	bp->b_flags |= B_INVALONERR;
1203 #endif
1204 	/* Write out the modified FAT block synchronously. */
1205 	return (bwrite(bp));
1206 }
1207