Home | History | Annotate | Line # | Download | only in msdosfs
msdosfs_fat.c revision 1.31
      1 /*	$NetBSD: msdosfs_fat.c,v 1.31 2016/05/07 16:43:02 mlelstv Exp $	*/
      2 
      3 /*-
      4  * Copyright (C) 1994, 1995, 1997 Wolfgang Solfrank.
      5  * Copyright (C) 1994, 1995, 1997 TooLs GmbH.
      6  * All rights reserved.
      7  * Original code by Paul Popelka (paulp (at) uts.amdahl.com) (see below).
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     12  * 1. Redistributions of source code must retain the above copyright
     13  *    notice, this list of conditions and the following disclaimer.
     14  * 2. Redistributions in binary form must reproduce the above copyright
     15  *    notice, this list of conditions and the following disclaimer in the
     16  *    documentation and/or other materials provided with the distribution.
     17  * 3. All advertising materials mentioning features or use of this software
     18  *    must display the following acknowledgement:
     19  *	This product includes software developed by TooLs GmbH.
     20  * 4. The name of TooLs GmbH may not be used to endorse or promote products
     21  *    derived from this software without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
     24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     26  * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
     27  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
     28  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
     29  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
     30  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
     31  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
     32  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     33  */
     34 /*
     35  * Written by Paul Popelka (paulp (at) uts.amdahl.com)
     36  *
     37  * You can do anything you want with this software, just don't say you wrote
     38  * it, and don't remove this notice.
     39  *
     40  * This software is provided "as is".
     41  *
     42  * The author supplies this software to be publicly redistributed on the
     43  * understanding that the author is not responsible for the correct
     44  * functioning of this software in any circumstances and is not liable for
     45  * any damages caused by this software.
     46  *
     47  * October 1992
     48  */
     49 
     50 #if HAVE_NBTOOL_CONFIG_H
     51 #include "nbtool_config.h"
     52 #endif
     53 
     54 #include <sys/cdefs.h>
     55 __KERNEL_RCSID(0, "$NetBSD: msdosfs_fat.c,v 1.31 2016/05/07 16:43:02 mlelstv Exp $");
     56 
     57 /*
     58  * kernel include files.
     59  */
     60 #include <sys/param.h>
     61 #include <sys/file.h>
     62 #ifdef _KERNEL
     63 #include <sys/mount.h>		/* to define statvfs structure */
     64 #include <sys/errno.h>
     65 #include <sys/systm.h>
     66 #include <sys/kauth.h>
     67 #include <sys/dirent.h>
     68 #include <sys/namei.h>
     69 #include <sys/buf.h>
     70 #include <sys/vnode.h>		/* to define vattr structure */
     71 #else
     72 #include <ffs/buf.h>
     73 #endif
     74 
     75 /*
     76  * msdosfs include files.
     77  */
     78 #include <fs/msdosfs/bpb.h>
     79 #include <fs/msdosfs/msdosfsmount.h>
     80 #include <fs/msdosfs/direntry.h>
     81 #include <fs/msdosfs/denode.h>
     82 #include <fs/msdosfs/fat.h>
     83 
     84 /*
     85  * Fat cache stats.
     86  */
     87 int fc_fileextends;		/* # of file extends			 */
     88 int fc_lfcempty;		/* # of time last file cluster cache entry
     89 				 * was empty */
     90 int fc_bmapcalls;		/* # of times pcbmap was called		 */
     91 
     92 #define	LMMAX	20
     93 int fc_lmdistance[LMMAX];	/* counters for how far off the last
     94 				 * cluster mapped entry was. */
     95 int fc_largedistance;		/* off by more than LMMAX		 */
     96 int fc_wherefrom, fc_whereto, fc_lastclust;
     97 int pm_fatblocksize;
     98 
     99 #ifdef MSDOSFS_DEBUG
    100 #define DPRINTF(a) printf a
    101 #else
    102 #define DPRINTF(a)
    103 #endif
    104 #ifdef MSDOSFS_DEBUG
    105 void print_fat_stats(void);
    106 
    107 void
    108 print_fat_stats(void)
    109 {
    110 	int i;
    111 
    112 	printf("fc_fileextends=%d fc_lfcempty=%d fc_bmapcalls=%d "
    113 	    "fc_largedistance=%d [%d->%d=%d] fc_lastclust=%d pm_fatblocksize=%d\n",
    114 	    fc_fileextends, fc_lfcempty, fc_bmapcalls, fc_largedistance,
    115 	    fc_wherefrom, fc_whereto, fc_whereto-fc_wherefrom,
    116 	    fc_lastclust, pm_fatblocksize);
    117 
    118 	fc_fileextends = fc_lfcempty = fc_bmapcalls = 0;
    119 	fc_wherefrom = fc_whereto = fc_lastclust = 0;
    120 
    121 	for (i = 0; i < LMMAX; i++) {
    122 		printf("%d:%d ", i, fc_lmdistance[i]);
    123 	fc_lmdistance[i] = 0;
    124 	}
    125 
    126 	printf("\n");
    127 }
    128 #endif
    129 
    130 static void fatblock(struct msdosfsmount *, u_long, u_long *, u_long *,
    131 			  u_long *);
    132 void updatefats(struct msdosfsmount *, struct buf *, u_long);
    133 static inline void usemap_free(struct msdosfsmount *, u_long);
    134 static inline void usemap_alloc(struct msdosfsmount *, u_long);
    135 static int fatchain(struct msdosfsmount *, u_long, u_long, u_long);
    136 int chainlength(struct msdosfsmount *, u_long, u_long);
    137 int chainalloc(struct msdosfsmount *, u_long, u_long, u_long, u_long *,
    138 		    u_long *);
    139 
    140 static void
    141 fatblock(struct msdosfsmount *pmp, u_long ofs, u_long *bnp, u_long *sizep, u_long *bop)
    142 {
    143 	u_long bn, size;
    144 
    145 	bn = ofs / pmp->pm_fatblocksize * pmp->pm_fatblocksec;
    146 	size = min(pmp->pm_fatblocksec, pmp->pm_FATsecs - bn)
    147 	    * pmp->pm_BytesPerSec;
    148 	bn += pmp->pm_fatblk + pmp->pm_curfat * pmp->pm_FATsecs;
    149 
    150 	DPRINTF(("%s(ofs=%lu bn=%lu, size=%lu, bo=%lu)\n", __func__, ofs, bn,
    151 	    size, ofs % pmp->pm_fatblocksize));
    152 	if (bnp)
    153 		*bnp = bn;
    154 	if (sizep)
    155 		*sizep = size;
    156 	if (bop)
    157 		*bop = ofs % pmp->pm_fatblocksize;
    158 
    159 	pm_fatblocksize = pmp->pm_fatblocksize;
    160 }
    161 
    162 /*
    163  * Map the logical cluster number of a file into a physical disk sector
    164  * that is filesystem relative.
    165  *
    166  * dep	  - address of denode representing the file of interest
    167  * findcn - file relative cluster whose filesystem relative cluster number
    168  *	    and/or block number are/is to be found
    169  * bnp	  - address of where to place the file system relative block number.
    170  *	    If this pointer is null then don't return this quantity.
    171  * cnp	  - address of where to place the file system relative cluster number.
    172  *	    If this pointer is null then don't return this quantity.
    173  *
    174  * NOTE: Either bnp or cnp must be non-null.
    175  * This function has one side effect.  If the requested file relative cluster
    176  * is beyond the end of file, then the actual number of clusters in the file
    177  * is returned in *cnp.  This is useful for determining how long a directory is.
    178  *  If cnp is null, nothing is returned.
    179  */
    180 int
    181 pcbmap(struct denode *dep, u_long findcn, daddr_t *bnp, u_long *cnp, int *sp)
    182 	/* findcn:		 file relative cluster to get		 */
    183 	/* bnp:		 returned filesys rel sector number	 */
    184 	/* cnp:		 returned cluster number		 */
    185 	/* sp:		 returned block size			 */
    186 {
    187 	int error;
    188 	u_long i;
    189 	u_long cn;
    190 	u_long prevcn = 0; /* XXX: prevcn could be used unititialized */
    191 	u_long byteoffset;
    192 	u_long bn;
    193 	u_long bo;
    194 	struct buf *bp = NULL;
    195 	u_long bp_bn = -1;
    196 	struct msdosfsmount *pmp = dep->de_pmp;
    197 	u_long bsize;
    198 
    199 	fc_bmapcalls++;
    200 
    201 	/*
    202 	 * If they don't give us someplace to return a value then don't
    203 	 * bother doing anything.
    204 	 */
    205 	if (bnp == NULL && cnp == NULL && sp == NULL)
    206 		return (0);
    207 
    208 	cn = dep->de_StartCluster;
    209 	DPRINTF(("%s(start cluster=%lu)\n", __func__, cn));
    210 	/*
    211 	 * The "file" that makes up the root directory is contiguous,
    212 	 * permanently allocated, of fixed size, and is not made up of
    213 	 * clusters.  If the cluster number is beyond the end of the root
    214 	 * directory, then return the number of clusters in the file.
    215 	 */
    216 	if (cn == MSDOSFSROOT) {
    217 		if (dep->de_Attributes & ATTR_DIRECTORY) {
    218 			if (de_cn2off(pmp, findcn) >= dep->de_FileSize) {
    219 				if (cnp)
    220 					*cnp = de_bn2cn(pmp, pmp->pm_rootdirsize);
    221 				DPRINTF(("%s(root, %lu ETOOBIG)\n", __func__,
    222 				    de_cn2off(pmp, findcn)));
    223 				return (E2BIG);
    224 			}
    225 			if (bnp)
    226 				*bnp = pmp->pm_rootdirblk + de_cn2bn(pmp, findcn);
    227 			if (cnp)
    228 				*cnp = MSDOSFSROOT;
    229 			if (sp)
    230 				*sp = min(pmp->pm_bpcluster,
    231 				    dep->de_FileSize - de_cn2off(pmp, findcn));
    232 			DPRINTF(("%s(root, bn=%lu, cn=%u)\n", __func__,
    233 			    pmp->pm_rootdirblk + de_cn2bn(pmp, findcn),
    234 			    MSDOSFSROOT));
    235 			return (0);
    236 		} else {		/* just an empty file */
    237 			if (cnp)
    238 				*cnp = 0;
    239 			DPRINTF(("%s(root, empty ETOOBIG)\n", __func__));
    240 			return (E2BIG);
    241 		}
    242 	}
    243 
    244 	/*
    245 	 * All other files do I/O in cluster sized blocks
    246 	 */
    247 	if (sp)
    248 		*sp = pmp->pm_bpcluster;
    249 
    250 	/*
    251 	 * Rummage around in the FAT cache, maybe we can avoid tromping
    252 	 * thru every FAT entry for the file. And, keep track of how far
    253 	 * off the cache was from where we wanted to be.
    254 	 */
    255 	i = 0;
    256 	fc_lookup(dep, findcn, &i, &cn);
    257 	DPRINTF(("%s(bpcluster=%lu i=%lu cn=%lu\n", __func__, pmp->pm_bpcluster,
    258 	    i, cn));
    259 	if ((bn = findcn - i) >= LMMAX) {
    260 		fc_largedistance++;
    261 		fc_wherefrom = i;
    262 		fc_whereto = findcn;
    263 		fc_lastclust = dep->de_fc[FC_LASTFC].fc_frcn;
    264 	} else
    265 		fc_lmdistance[bn]++;
    266 
    267 	/*
    268 	 * Handle all other files or directories the normal way.
    269 	 */
    270 	for (; i < findcn; i++) {
    271 		/*
    272 		 * Stop with all reserved clusters, not just with EOF.
    273 		 */
    274 		if (cn >= (CLUST_RSRVD & pmp->pm_fatmask))
    275 			goto hiteof;
    276 
    277 		/*
    278 		 * Also stop when cluster is not in the filesystem
    279 		 */
    280 		if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster) {
    281 			DPRINTF(("%s(cn, %lu not in %lu..%lu)\n", __func__,
    282 				cn, (u_long)CLUST_FIRST, pmp->pm_maxcluster));
    283 			if (bp)
    284 				brelse(bp, 0);
    285 			return (EINVAL);
    286 		}
    287 
    288 		byteoffset = FATOFS(pmp, cn);
    289 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
    290 		if (bn != bp_bn) {
    291 			if (bp)
    292 				brelse(bp, 0);
    293 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
    294 			    0, &bp);
    295 			if (error) {
    296 				DPRINTF(("%s(bread, %d)\n", __func__, error));
    297 				return (error);
    298 			}
    299 			bp_bn = bn;
    300 		}
    301 		prevcn = cn;
    302 		if (bo >= bsize) {
    303 			if (bp)
    304 				brelse(bp, 0);
    305 			DPRINTF(("%s(block, %lu >= %lu)\n", __func__, bo,
    306 			    bsize));
    307 			return (EIO);
    308 		}
    309 		KASSERT(bp != NULL);
    310 		if (FAT32(pmp))
    311 			cn = getulong((char *)bp->b_data + bo);
    312 		else
    313 			cn = getushort((char *)bp->b_data + bo);
    314 		if (FAT12(pmp) && (prevcn & 1))
    315 			cn >>= 4;
    316 		DPRINTF(("%s(cn=%lu masked=%lu)\n", __func__, cn,
    317 		    cn & pmp->pm_fatmask));
    318 		cn &= pmp->pm_fatmask;
    319 	}
    320 
    321 	if (!MSDOSFSEOF(cn, pmp->pm_fatmask)) {
    322 		if (bp)
    323 			brelse(bp, 0);
    324 		if (bnp)
    325 			*bnp = cntobn(pmp, cn);
    326 		if (cnp)
    327 			*cnp = cn;
    328 		DPRINTF(("%s(bn=%lu, cn=%lu)\n", __func__, cntobn(pmp, cn),
    329 		    cn));
    330 		fc_setcache(dep, FC_LASTMAP, i, cn);
    331 		return (0);
    332 	}
    333 
    334 hiteof:;
    335 	if (cnp)
    336 		*cnp = i;
    337 	if (bp)
    338 		brelse(bp, 0);
    339 	/* update last file cluster entry in the FAT cache */
    340 	fc_setcache(dep, FC_LASTFC, i - 1, prevcn);
    341 	DPRINTF(("%s(eof, %lu)\n", __func__, i));
    342 	return (E2BIG);
    343 }
    344 
    345 /*
    346  * Find the closest entry in the FAT cache to the cluster we are looking
    347  * for.
    348  */
    349 void
    350 fc_lookup(struct denode *dep, u_long findcn, u_long *frcnp, u_long *fsrcnp)
    351 {
    352 	int i;
    353 	u_long cn;
    354 	struct fatcache *closest = 0;
    355 
    356 	for (i = 0; i < FC_SIZE; i++) {
    357 		cn = dep->de_fc[i].fc_frcn;
    358 		if (cn != FCE_EMPTY && cn <= findcn) {
    359 			if (closest == 0 || cn > closest->fc_frcn)
    360 				closest = &dep->de_fc[i];
    361 		}
    362 	}
    363 	if (closest) {
    364 		*frcnp = closest->fc_frcn;
    365 		*fsrcnp = closest->fc_fsrcn;
    366 	}
    367 }
    368 
    369 /*
    370  * Purge the FAT cache in denode dep of all entries relating to file
    371  * relative cluster frcn and beyond.
    372  */
    373 void
    374 fc_purge(struct denode *dep, u_int frcn)
    375 {
    376 	int i;
    377 	struct fatcache *fcp;
    378 
    379 	fcp = dep->de_fc;
    380 	for (i = 0; i < FC_SIZE; i++, fcp++) {
    381 		if (fcp->fc_frcn >= frcn)
    382 			fcp->fc_frcn = FCE_EMPTY;
    383 	}
    384 }
    385 
    386 /*
    387  * Update the FAT.
    388  * If mirroring the FAT, update all copies, with the first copy as last.
    389  * Else update only the current FAT (ignoring the others).
    390  *
    391  * pmp	 - msdosfsmount structure for filesystem to update
    392  * bp	 - addr of modified FAT block
    393  * fatbn - block number relative to begin of filesystem of the modified FAT block.
    394  */
    395 void
    396 updatefats(struct msdosfsmount *pmp, struct buf *bp, u_long fatbn)
    397 {
    398 	int i, error;
    399 	struct buf *bpn;
    400 
    401 	DPRINTF(("%s(pmp %p, bp %p, fatbn %lu)\n", __func__, pmp, bp, fatbn));
    402 
    403 	/*
    404 	 * If we have an FSInfo block, update it.
    405 	 */
    406 	if (pmp->pm_fsinfo) {
    407 		u_long cn = pmp->pm_nxtfree;
    408 
    409 		if (pmp->pm_freeclustercount
    410 		    && (pmp->pm_inusemap[cn / N_INUSEBITS]
    411 			& (1 << (cn % N_INUSEBITS)))) {
    412 			/*
    413 			 * The cluster indicated in FSInfo isn't free
    414 			 * any longer.  Got get a new free one.
    415 			 */
    416 			for (cn = 0; cn < pmp->pm_maxcluster; cn++)
    417 				if (pmp->pm_inusemap[cn / N_INUSEBITS] != (u_int)-1)
    418 					break;
    419 			pmp->pm_nxtfree = cn
    420 				+ ffs(pmp->pm_inusemap[cn / N_INUSEBITS]
    421 				      ^ (u_int)-1) - 1;
    422 		}
    423 		/*
    424 		 * XXX  If the fsinfo block is stored on media with
    425 		 *      2KB or larger sectors, is the fsinfo structure
    426 		 *      padded at the end or in the middle?
    427 		 */
    428 		if (bread(pmp->pm_devvp, de_bn2kb(pmp, pmp->pm_fsinfo),
    429 		    pmp->pm_BytesPerSec, B_MODIFY, &bpn) != 0) {
    430 			/*
    431 			 * Ignore the error, but turn off FSInfo update for the future.
    432 			 */
    433 			pmp->pm_fsinfo = 0;
    434 		} else {
    435 			struct fsinfo *fp = (struct fsinfo *)bpn->b_data;
    436 
    437 			putulong(fp->fsinfree, pmp->pm_freeclustercount);
    438 			putulong(fp->fsinxtfree, pmp->pm_nxtfree);
    439 			if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT)
    440 				bwrite(bpn);
    441 			else
    442 				bdwrite(bpn);
    443 		}
    444 	}
    445 
    446 	if (pmp->pm_flags & MSDOSFS_FATMIRROR) {
    447 		/*
    448 		 * Now copy the block(s) of the modified FAT to the other copies of
    449 		 * the FAT and write them out.  This is faster than reading in the
    450 		 * other FATs and then writing them back out.  This could tie up
    451 		 * the FAT for quite a while. Preventing others from accessing it.
    452 		 * To prevent us from going after the FAT quite so much we use
    453 		 * delayed writes, unless they specified "synchronous" when the
    454 		 * filesystem was mounted.  If synch is asked for then use
    455 		 * bwrite()'s and really slow things down.
    456 		 */
    457 		for (i = 1; i < pmp->pm_FATs; i++) {
    458 			fatbn += pmp->pm_FATsecs;
    459 			/* getblk() never fails */
    460 			bpn = getblk(pmp->pm_devvp, de_bn2kb(pmp, fatbn),
    461 			    bp->b_bcount, 0, 0);
    462 			memcpy(bpn->b_data, bp->b_data, bp->b_bcount);
    463 			if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT) {
    464 				error = bwrite(bpn);
    465 				if (error)
    466 					printf("%s: copy FAT %d (error=%d)\n",
    467 						 __func__, i, error);
    468 			} else
    469 				bdwrite(bpn);
    470 		}
    471 	}
    472 
    473 	/*
    474 	 * Write out the first (or current) FAT last.
    475 	 */
    476 	if (pmp->pm_flags & MSDOSFSMNT_WAITONFAT) {
    477 		error =  bwrite(bp);
    478 		if (error)
    479 			printf("%s: write FAT (error=%d)\n",
    480 				__func__, error);
    481 	} else
    482 		bdwrite(bp);
    483 	/*
    484 	 * Maybe update fsinfo sector here?
    485 	 */
    486 }
    487 
    488 /*
    489  * Updating entries in 12 bit FATs is a pain in the butt.
    490  *
    491  * The following picture shows where nibbles go when moving from a 12 bit
    492  * cluster number into the appropriate bytes in the FAT.
    493  *
    494  *	byte m        byte m+1      byte m+2
    495  *	+----+----+   +----+----+   +----+----+
    496  *	|  0    1 |   |  2    3 |   |  4    5 |   FAT bytes
    497  *	+----+----+   +----+----+   +----+----+
    498  *
    499  *	+----+----+----+   +----+----+----+
    500  *	|  3    0    1 |   |  4    5    2 |
    501  *	+----+----+----+   +----+----+----+
    502  *	cluster n  	   cluster n+1
    503  *
    504  * Where n is even. m = n + (n >> 2)
    505  *
    506  */
    507 static inline void
    508 usemap_alloc(struct msdosfsmount *pmp, u_long cn)
    509 {
    510 
    511 	pmp->pm_inusemap[cn / N_INUSEBITS] |= 1 << (cn % N_INUSEBITS);
    512 	pmp->pm_freeclustercount--;
    513 }
    514 
    515 static inline void
    516 usemap_free(struct msdosfsmount *pmp, u_long cn)
    517 {
    518 
    519 	pmp->pm_freeclustercount++;
    520 	pmp->pm_inusemap[cn / N_INUSEBITS] &= ~(1 << (cn % N_INUSEBITS));
    521 }
    522 
    523 int
    524 clusterfree(struct msdosfsmount *pmp, u_long cluster, u_long *oldcnp)
    525 {
    526 	int error;
    527 	u_long oldcn;
    528 
    529 	usemap_free(pmp, cluster);
    530 	error = fatentry(FAT_GET_AND_SET, pmp, cluster, &oldcn, MSDOSFSFREE);
    531 	if (error) {
    532 		usemap_alloc(pmp, cluster);
    533 		return (error);
    534 	}
    535 	/*
    536 	 * If the cluster was successfully marked free, then update
    537 	 * the count of free clusters, and turn off the "allocated"
    538 	 * bit in the "in use" cluster bit map.
    539 	 */
    540 	if (oldcnp)
    541 		*oldcnp = oldcn;
    542 	return (0);
    543 }
    544 
    545 /*
    546  * Get or Set or 'Get and Set' the cluster'th entry in the FAT.
    547  *
    548  * function	- whether to get or set a fat entry
    549  * pmp		- address of the msdosfsmount structure for the filesystem
    550  *		  whose FAT is to be manipulated.
    551  * cn		- which cluster is of interest
    552  * oldcontents	- address of a word that is to receive the contents of the
    553  *		  cluster'th entry if this is a get function
    554  * newcontents	- the new value to be written into the cluster'th element of
    555  *		  the FAT if this is a set function.
    556  *
    557  * This function can also be used to free a cluster by setting the FAT entry
    558  * for a cluster to 0.
    559  *
    560  * All copies of the FAT are updated if this is a set function. NOTE: If
    561  * fatentry() marks a cluster as free it does not update the inusemap in
    562  * the msdosfsmount structure. This is left to the caller.
    563  */
    564 int
    565 fatentry(int function, struct msdosfsmount *pmp, u_long cn, u_long *oldcontents, u_long newcontents)
    566 {
    567 	int error;
    568 	u_long readcn;
    569 	u_long bn, bo, bsize, byteoffset;
    570 	struct buf *bp;
    571 
    572 	DPRINTF(("%s(func %d, pmp %p, clust %lu, oldcon %p, newcon " "%lx)\n",
    573 	    __func__, function, pmp, cn, oldcontents, newcontents));
    574 
    575 #ifdef DIAGNOSTIC
    576 	/*
    577 	 * Be sure they asked us to do something.
    578 	 */
    579 	if ((function & (FAT_SET | FAT_GET)) == 0) {
    580 		DPRINTF(("%s(): function code doesn't specify get or set\n",
    581 		    __func__));
    582 		return (EINVAL);
    583 	}
    584 
    585 	/*
    586 	 * If they asked us to return a cluster number but didn't tell us
    587 	 * where to put it, give them an error.
    588 	 */
    589 	if ((function & FAT_GET) && oldcontents == NULL) {
    590 		DPRINTF(("%s(): get function with no place to put result\n",
    591 			__func__));
    592 		return (EINVAL);
    593 	}
    594 #endif
    595 
    596 	/*
    597 	 * Be sure the requested cluster is in the filesystem.
    598 	 */
    599 	if (cn < CLUST_FIRST || cn > pmp->pm_maxcluster)
    600 		return (EINVAL);
    601 
    602 	byteoffset = FATOFS(pmp, cn);
    603 	fatblock(pmp, byteoffset, &bn, &bsize, &bo);
    604 	if ((error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
    605 	    0, &bp)) != 0) {
    606 		return (error);
    607 	}
    608 
    609 	if (function & FAT_GET) {
    610 		if (FAT32(pmp))
    611 			readcn = getulong((char *)bp->b_data + bo);
    612 		else
    613 			readcn = getushort((char *)bp->b_data + bo);
    614 		if (FAT12(pmp) & (cn & 1))
    615 			readcn >>= 4;
    616 		readcn &= pmp->pm_fatmask;
    617 		*oldcontents = readcn;
    618 	}
    619 	if (function & FAT_SET) {
    620 		switch (pmp->pm_fatmask) {
    621 		case FAT12_MASK:
    622 			readcn = getushort((char *)bp->b_data + bo);
    623 			if (cn & 1) {
    624 				readcn &= 0x000f;
    625 				readcn |= newcontents << 4;
    626 			} else {
    627 				readcn &= 0xf000;
    628 				readcn |= newcontents & 0xfff;
    629 			}
    630 			putushort((char *)bp->b_data + bo, readcn);
    631 			break;
    632 		case FAT16_MASK:
    633 			putushort((char *)bp->b_data + bo, newcontents);
    634 			break;
    635 		case FAT32_MASK:
    636 			/*
    637 			 * According to spec we have to retain the
    638 			 * high order bits of the FAT entry.
    639 			 */
    640 			readcn = getulong((char *)bp->b_data + bo);
    641 			readcn &= ~FAT32_MASK;
    642 			readcn |= newcontents & FAT32_MASK;
    643 			putulong((char *)bp->b_data + bo, readcn);
    644 			break;
    645 		}
    646 		updatefats(pmp, bp, bn);
    647 		bp = NULL;
    648 		pmp->pm_fmod = 1;
    649 	}
    650 	if (bp)
    651 		brelse(bp, 0);
    652 	return (0);
    653 }
    654 
    655 /*
    656  * Update a contiguous cluster chain
    657  *
    658  * pmp	    - mount point
    659  * start    - first cluster of chain
    660  * count    - number of clusters in chain
    661  * fillwith - what to write into FAT entry of last cluster
    662  */
    663 static int
    664 fatchain(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith)
    665 {
    666 	int error;
    667 	u_long bn, bo, bsize, byteoffset, readcn, newc;
    668 	struct buf *bp;
    669 
    670 	DPRINTF(("%s(pmp %p, start %lu, count %lu, fillwith %lx)\n", __func__,
    671 	    pmp, start, count, fillwith));
    672 	/*
    673 	 * Be sure the clusters are in the filesystem.
    674 	 */
    675 	if (start < CLUST_FIRST || start + count - 1 > pmp->pm_maxcluster)
    676 		return (EINVAL);
    677 
    678 	while (count > 0) {
    679 		byteoffset = FATOFS(pmp, start);
    680 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
    681 		error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
    682 		    B_MODIFY, &bp);
    683 		if (error) {
    684 			return (error);
    685 		}
    686 		while (count > 0) {
    687 			start++;
    688 			newc = --count > 0 ? start : fillwith;
    689 			switch (pmp->pm_fatmask) {
    690 			case FAT12_MASK:
    691 				readcn = getushort((char *)bp->b_data + bo);
    692 				if (start & 1) {
    693 					readcn &= 0xf000;
    694 					readcn |= newc & 0xfff;
    695 				} else {
    696 					readcn &= 0x000f;
    697 					readcn |= newc << 4;
    698 				}
    699 				putushort((char *)bp->b_data + bo, readcn);
    700 				bo++;
    701 				if (!(start & 1))
    702 					bo++;
    703 				break;
    704 			case FAT16_MASK:
    705 				putushort((char *)bp->b_data + bo, newc);
    706 				bo += 2;
    707 				break;
    708 			case FAT32_MASK:
    709 				readcn = getulong((char *)bp->b_data + bo);
    710 				readcn &= ~pmp->pm_fatmask;
    711 				readcn |= newc & pmp->pm_fatmask;
    712 				putulong((char *)bp->b_data + bo, readcn);
    713 				bo += 4;
    714 				break;
    715 			}
    716 			if (bo >= bsize)
    717 				break;
    718 		}
    719 		updatefats(pmp, bp, bn);
    720 	}
    721 	pmp->pm_fmod = 1;
    722 	return (0);
    723 }
    724 
    725 /*
    726  * Check the length of a free cluster chain starting at start.
    727  *
    728  * pmp	 - mount point
    729  * start - start of chain
    730  * count - maximum interesting length
    731  */
    732 int
    733 chainlength(struct msdosfsmount *pmp, u_long start, u_long count)
    734 {
    735 	u_long idx, max_idx;
    736 	u_int map;
    737 	u_long len;
    738 
    739 	max_idx = pmp->pm_maxcluster / N_INUSEBITS;
    740 	idx = start / N_INUSEBITS;
    741 	start %= N_INUSEBITS;
    742 	map = pmp->pm_inusemap[idx];
    743 	map &= ~((1 << start) - 1);
    744 	if (map) {
    745 		len = ffs(map) - 1 - start;
    746 		return (len > count ? count : len);
    747 	}
    748 	len = N_INUSEBITS - start;
    749 	if (len >= count)
    750 		return (count);
    751 	while (++idx <= max_idx) {
    752 		if (len >= count)
    753 			break;
    754 		if ((map = pmp->pm_inusemap[idx]) != 0) {
    755 			len +=  ffs(map) - 1;
    756 			break;
    757 		}
    758 		len += N_INUSEBITS;
    759 	}
    760 	return (len > count ? count : len);
    761 }
    762 
    763 /*
    764  * Allocate contigous free clusters.
    765  *
    766  * pmp	      - mount point.
    767  * start      - start of cluster chain.
    768  * count      - number of clusters to allocate.
    769  * fillwith   - put this value into the FAT entry for the
    770  *		last allocated cluster.
    771  * retcluster - put the first allocated cluster's number here.
    772  * got	      - how many clusters were actually allocated.
    773  */
    774 int
    775 chainalloc(struct msdosfsmount *pmp, u_long start, u_long count, u_long fillwith, u_long *retcluster, u_long *got)
    776 {
    777 	int error;
    778 	u_long cl, n;
    779 
    780 	for (cl = start, n = count; n-- > 0;)
    781 		usemap_alloc(pmp, cl++);
    782 	if ((error = fatchain(pmp, start, count, fillwith)) != 0)
    783 		return (error);
    784 
    785 	DPRINTF(("%s(): allocated cluster chain at %lu (%lu clusters)\n",
    786 	    __func__, start, count));
    787 	if (retcluster)
    788 		*retcluster = start;
    789 	if (got)
    790 		*got = count;
    791 	return (0);
    792 }
    793 
    794 /*
    795  * Allocate contiguous free clusters.
    796  *
    797  * pmp	      - mount point.
    798  * start      - preferred start of cluster chain.
    799  * count      - number of clusters requested.
    800  * fillwith   - put this value into the FAT entry for the
    801  *		last allocated cluster.
    802  * retcluster - put the first allocated cluster's number here.
    803  * got	      - how many clusters were actually allocated.
    804  */
    805 int
    806 clusteralloc(struct msdosfsmount *pmp, u_long start, u_long count, u_long *retcluster, u_long *got)
    807 {
    808 	u_long idx;
    809 	u_long len, newst, foundl, cn, l;
    810 	u_long foundcn = 0; /* XXX: foundcn could be used unititialized */
    811 	u_long fillwith = CLUST_EOFE;
    812 	u_int map;
    813 
    814 	DPRINTF(("%s(): find %lu clusters\n", __func__, count));
    815 	if (start) {
    816 		if ((len = chainlength(pmp, start, count)) >= count)
    817 			return (chainalloc(pmp, start, count, fillwith, retcluster, got));
    818 	} else {
    819 		/*
    820 		 * This is a new file, initialize start
    821 		 */
    822 		struct timeval tv;
    823 
    824 		microtime(&tv);
    825 		start = (tv.tv_usec >> 10) | tv.tv_usec;
    826 		len = 0;
    827 	}
    828 
    829 	/*
    830 	 * Start at a (pseudo) random place to maximize cluster runs
    831 	 * under multiple writers.
    832 	 */
    833 	newst = (start * 1103515245 + 12345) % (pmp->pm_maxcluster + 1);
    834 	foundl = 0;
    835 
    836 	for (cn = newst; cn <= pmp->pm_maxcluster;) {
    837 		idx = cn / N_INUSEBITS;
    838 		map = pmp->pm_inusemap[idx];
    839 		map |= (1 << (cn % N_INUSEBITS)) - 1;
    840 		if (map != (u_int)-1) {
    841 			cn = idx * N_INUSEBITS + ffs(map^(u_int)-1) - 1;
    842 			if ((l = chainlength(pmp, cn, count)) >= count)
    843 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
    844 			if (l > foundl) {
    845 				foundcn = cn;
    846 				foundl = l;
    847 			}
    848 			cn += l + 1;
    849 			continue;
    850 		}
    851 		cn += N_INUSEBITS - cn % N_INUSEBITS;
    852 	}
    853 	for (cn = 0; cn < newst;) {
    854 		idx = cn / N_INUSEBITS;
    855 		map = pmp->pm_inusemap[idx];
    856 		map |= (1 << (cn % N_INUSEBITS)) - 1;
    857 		if (map != (u_int)-1) {
    858 			cn = idx * N_INUSEBITS + ffs(map^(u_int)-1) - 1;
    859 			if ((l = chainlength(pmp, cn, count)) >= count)
    860 				return (chainalloc(pmp, cn, count, fillwith, retcluster, got));
    861 			if (l > foundl) {
    862 				foundcn = cn;
    863 				foundl = l;
    864 			}
    865 			cn += l + 1;
    866 			continue;
    867 		}
    868 		cn += N_INUSEBITS - cn % N_INUSEBITS;
    869 	}
    870 
    871 	if (!foundl)
    872 		return (ENOSPC);
    873 
    874 	if (len)
    875 		return (chainalloc(pmp, start, len, fillwith, retcluster, got));
    876 	else
    877 		return (chainalloc(pmp, foundcn, foundl, fillwith, retcluster, got));
    878 }
    879 
    880 
    881 /*
    882  * Free a chain of clusters.
    883  *
    884  * pmp		- address of the msdosfs mount structure for the filesystem
    885  *		  containing the cluster chain to be freed.
    886  * startcluster - number of the 1st cluster in the chain of clusters to be
    887  *		  freed.
    888  */
    889 int
    890 freeclusterchain(struct msdosfsmount *pmp, u_long cluster)
    891 {
    892 	int error;
    893 	struct buf *bp = NULL;
    894 	u_long bn, bo, bsize, byteoffset;
    895 	u_long readcn, lbn = -1;
    896 
    897 	while (cluster >= CLUST_FIRST && cluster <= pmp->pm_maxcluster) {
    898 		byteoffset = FATOFS(pmp, cluster);
    899 		fatblock(pmp, byteoffset, &bn, &bsize, &bo);
    900 		if (lbn != bn) {
    901 			if (bp)
    902 				updatefats(pmp, bp, lbn);
    903 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
    904 			    B_MODIFY, &bp);
    905 			if (error) {
    906 				return (error);
    907 			}
    908 			lbn = bn;
    909 		}
    910 		usemap_free(pmp, cluster);
    911 		KASSERT(bp != NULL);
    912 		switch (pmp->pm_fatmask) {
    913 		case FAT12_MASK:
    914 			readcn = getushort((char *)bp->b_data + bo);
    915 			if (cluster & 1) {
    916 				cluster = readcn >> 4;
    917 				readcn &= 0x000f;
    918 				readcn |= MSDOSFSFREE << 4;
    919 			} else {
    920 				cluster = readcn;
    921 				readcn &= 0xf000;
    922 				readcn |= MSDOSFSFREE & 0xfff;
    923 			}
    924 			putushort((char *)bp->b_data + bo, readcn);
    925 			break;
    926 		case FAT16_MASK:
    927 			cluster = getushort((char *)bp->b_data + bo);
    928 			putushort((char *)bp->b_data + bo, MSDOSFSFREE);
    929 			break;
    930 		case FAT32_MASK:
    931 			cluster = getulong((char *)bp->b_data + bo);
    932 			putulong((char *)bp->b_data + bo,
    933 				 (MSDOSFSFREE & FAT32_MASK) | (cluster & ~FAT32_MASK));
    934 			break;
    935 		}
    936 		cluster &= pmp->pm_fatmask;
    937 	}
    938 	if (bp)
    939 		updatefats(pmp, bp, bn);
    940 	return (0);
    941 }
    942 
    943 /*
    944  * Read in FAT blocks looking for free clusters. For every free cluster
    945  * found turn off its corresponding bit in the pm_inusemap.
    946  */
    947 int
    948 fillinusemap(struct msdosfsmount *pmp)
    949 {
    950 	struct buf *bp = NULL;
    951 	u_long cn, readcn;
    952 	int error;
    953 	u_long bn, bo, bsize, byteoffset;
    954 
    955 	/*
    956 	 * Mark all clusters in use, we mark the free ones in the FAT scan
    957 	 * loop further down.
    958 	 */
    959 	for (cn = 0; cn < (pmp->pm_maxcluster + N_INUSEBITS) / N_INUSEBITS; cn++)
    960 		pmp->pm_inusemap[cn] = (u_int)-1;
    961 
    962 	/*
    963 	 * Figure how many free clusters are in the filesystem by ripping
    964 	 * through the FAT counting the number of entries whose content is
    965 	 * zero.  These represent free clusters.
    966 	 */
    967 	pmp->pm_freeclustercount = 0;
    968 	for (cn = CLUST_FIRST; cn <= pmp->pm_maxcluster; cn++) {
    969 		byteoffset = FATOFS(pmp, cn);
    970 		bo = byteoffset % pmp->pm_fatblocksize;
    971 		if (!bo || !bp) {
    972 			/* Read new FAT block */
    973 			if (bp)
    974 				brelse(bp, 0);
    975 			fatblock(pmp, byteoffset, &bn, &bsize, NULL);
    976 			error = bread(pmp->pm_devvp, de_bn2kb(pmp, bn), bsize,
    977 			    0, &bp);
    978 			if (error) {
    979 				return (error);
    980 			}
    981 		}
    982 		if (FAT32(pmp))
    983 			readcn = getulong((char *)bp->b_data + bo);
    984 		else
    985 			readcn = getushort((char *)bp->b_data + bo);
    986 		if (FAT12(pmp) && (cn & 1))
    987 			readcn >>= 4;
    988 		readcn &= pmp->pm_fatmask;
    989 
    990 		if (readcn == 0)
    991 			usemap_free(pmp, cn);
    992 	}
    993 	if (bp)
    994 		brelse(bp, 0);
    995 	return (0);
    996 }
    997 
    998 /*
    999  * Allocate a new cluster and chain it onto the end of the file.
   1000  *
   1001  * dep	 - the file to extend
   1002  * count - number of clusters to allocate
   1003  * bpp	 - where to return the address of the buf header for the first new
   1004  *	   file block
   1005  * ncp	 - where to put cluster number of the first newly allocated cluster
   1006  *	   If this pointer is 0, do not return the cluster number.
   1007  * flags - see fat.h
   1008  *
   1009  * NOTE: This function is not responsible for turning on the DE_UPDATE bit of
   1010  * the de_flag field of the denode and it does not change the de_FileSize
   1011  * field.  This is left for the caller to do.
   1012  */
   1013 
   1014 int
   1015 extendfile(struct denode *dep, u_long count, struct buf **bpp, u_long *ncp, int flags)
   1016 {
   1017 	int error;
   1018 	u_long frcn = 0, cn, got;
   1019 	struct msdosfsmount *pmp = dep->de_pmp;
   1020 	struct buf *bp;
   1021 
   1022 	/*
   1023 	 * Don't try to extend the root directory
   1024 	 */
   1025 	if (dep->de_StartCluster == MSDOSFSROOT
   1026 	    && (dep->de_Attributes & ATTR_DIRECTORY)) {
   1027 		DPRINTF(("%s(): attempt to extend root directory\n", __func__));
   1028 		return (ENOSPC);
   1029 	}
   1030 
   1031 	/*
   1032 	 * If the "file's last cluster" cache entry is empty, and the file
   1033 	 * is not empty, then fill the cache entry by calling pcbmap().
   1034 	 */
   1035 	fc_fileextends++;
   1036 	if (dep->de_fc[FC_LASTFC].fc_frcn == FCE_EMPTY &&
   1037 	    dep->de_StartCluster != 0) {
   1038 		fc_lfcempty++;
   1039 		error = pcbmap(dep, CLUST_END, 0, &cn, 0);
   1040 		/* we expect it to return E2BIG */
   1041 		if (error != E2BIG)
   1042 			return (error);
   1043 	}
   1044 
   1045 	fc_last_to_nexttolast(dep);
   1046 
   1047 	while (count > 0) {
   1048 
   1049 		/*
   1050 		 * Allocate a new cluster chain and cat onto the end of the
   1051 		 * file.  If the file is empty we make de_StartCluster point
   1052 		 * to the new block.  Note that de_StartCluster being 0 is
   1053 		 * sufficient to be sure the file is empty since we exclude
   1054 		 * attempts to extend the root directory above, and the root
   1055 		 * dir is the only file with a startcluster of 0 that has
   1056 		 * blocks allocated (sort of).
   1057 		 */
   1058 
   1059 		if (dep->de_StartCluster == 0)
   1060 			cn = 0;
   1061 		else
   1062 			cn = dep->de_fc[FC_LASTFC].fc_fsrcn + 1;
   1063 		error = clusteralloc(pmp, cn, count, &cn, &got);
   1064 		if (error)
   1065 			return (error);
   1066 
   1067 		count -= got;
   1068 
   1069 		/*
   1070 		 * Give them the filesystem relative cluster number if they want
   1071 		 * it.
   1072 		 */
   1073 		if (ncp) {
   1074 			*ncp = cn;
   1075 			ncp = NULL;
   1076 		}
   1077 
   1078 		if (dep->de_StartCluster == 0) {
   1079 			dep->de_StartCluster = cn;
   1080 			frcn = 0;
   1081 		} else {
   1082 			error = fatentry(FAT_SET, pmp,
   1083 					 dep->de_fc[FC_LASTFC].fc_fsrcn,
   1084 					 0, cn);
   1085 			if (error) {
   1086 				clusterfree(pmp, cn, NULL);
   1087 				return (error);
   1088 			}
   1089 			frcn = dep->de_fc[FC_LASTFC].fc_frcn + 1;
   1090 		}
   1091 
   1092 		/*
   1093 		 * Update the "last cluster of the file" entry in the
   1094 		 * denode's FAT cache.
   1095 		 */
   1096 
   1097 		fc_setcache(dep, FC_LASTFC, frcn + got - 1, cn + got - 1);
   1098 		if ((flags & DE_CLEAR) &&
   1099 		    (dep->de_Attributes & ATTR_DIRECTORY)) {
   1100 			while (got-- > 0) {
   1101 				bp = getblk(pmp->pm_devvp,
   1102 				    de_bn2kb(pmp, cntobn(pmp, cn++)),
   1103 				    pmp->pm_bpcluster, 0, 0);
   1104 				clrbuf(bp);
   1105 				if (bpp) {
   1106 					*bpp = bp;
   1107 						bpp = NULL;
   1108 				} else {
   1109 					bdwrite(bp);
   1110 				}
   1111 			}
   1112 		}
   1113 	}
   1114 
   1115 	return (0);
   1116 }
   1117