Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.2
      1 /*	$NetBSD: ffs_alloc.c,v 1.2 2001/10/28 13:14:06 lukem Exp $	*/
      2 /* From: NetBSD: ffs_alloc.c,v 1.50 2001/09/06 02:16:01 lukem Exp */
      3 
      4 /*
      5  * Copyright (c) 1982, 1986, 1989, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. All advertising materials mentioning features or use of this software
     17  *    must display the following acknowledgement:
     18  *	This product includes software developed by the University of
     19  *	California, Berkeley and its contributors.
     20  * 4. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  *	@(#)ffs_alloc.c	8.19 (Berkeley) 7/13/95
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 #ifndef __lint
     41 __RCSID("$NetBSD: ffs_alloc.c,v 1.2 2001/10/28 13:14:06 lukem Exp $");
     42 #endif	/* !__lint */
     43 
     44 #include <sys/param.h>
     45 #include <sys/time.h>
     46 
     47 #include <err.h>
     48 #include <errno.h>
     49 
     50 #include <ufs/ufs/ufs_bswap.h>
     51 #include <ufs/ufs/inode.h>
     52 #include <ufs/ffs/fs.h>
     53 
     54 #include "ffs/buf.h"
     55 #include "ffs/ffs_extern.h"
     56 
     57 
     58 static int scanc(u_int, const u_char *, const u_char *, int);
     59 
     60 static ufs_daddr_t ffs_alloccg(struct inode *, int, ufs_daddr_t, int);
     61 static ufs_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs_daddr_t);
     62 static u_long ffs_hashalloc(struct inode *, int, long, int,
     63 		     ufs_daddr_t (*)(struct inode *, int, ufs_daddr_t, int));
     64 static ufs_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs_daddr_t, int);
     65 
     66 /* in ffs_tables.c */
     67 extern const int inside[], around[];
     68 extern const u_char * const fragtbl[];
     69 
     70 /*
     71  * Allocate a block in the file system.
     72  *
     73  * The size of the requested block is given, which must be some
     74  * multiple of fs_fsize and <= fs_bsize.
     75  * A preference may be optionally specified. If a preference is given
     76  * the following hierarchy is used to allocate a block:
     77  *   1) allocate the requested block.
     78  *   2) allocate a rotationally optimal block in the same cylinder.
     79  *   3) allocate a block in the same cylinder group.
     80  *   4) quadradically rehash into other cylinder groups, until an
     81  *      available block is located.
     82  * If no block preference is given the following hierarchy is used
     83  * to allocate a block:
     84  *   1) allocate a block in the cylinder group that contains the
     85  *      inode for the file.
     86  *   2) quadradically rehash into other cylinder groups, until an
     87  *      available block is located.
     88  */
     89 int
     90 ffs_alloc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t bpref, int size,
     91     ufs_daddr_t *bnp)
     92 {
     93 	struct fs *fs = ip->i_fs;
     94 	ufs_daddr_t bno;
     95 	int cg;
     96 
     97 	*bnp = 0;
     98 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
     99 		errx(1, "ffs_alloc: bad size: bsize %d size %d",
    100 		    fs->fs_bsize, size);
    101 	}
    102 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
    103 		goto nospace;
    104 	if (bpref >= fs->fs_size)
    105 		bpref = 0;
    106 	if (bpref == 0)
    107 		cg = ino_to_cg(fs, ip->i_number);
    108 	else
    109 		cg = dtog(fs, bpref);
    110 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
    111 	    			     ffs_alloccg);
    112 	if (bno > 0) {
    113 		ip->i_ffs_blocks += btodb(size);
    114 		*bnp = bno;
    115 		return (0);
    116 	}
    117 nospace:
    118 	return (ENOSPC);
    119 }
    120 
    121 /*
    122  * Select the desired position for the next block in a file.  The file is
    123  * logically divided into sections. The first section is composed of the
    124  * direct blocks. Each additional section contains fs_maxbpg blocks.
    125  *
    126  * If no blocks have been allocated in the first section, the policy is to
    127  * request a block in the same cylinder group as the inode that describes
    128  * the file. If no blocks have been allocated in any other section, the
    129  * policy is to place the section in a cylinder group with a greater than
    130  * average number of free blocks.  An appropriate cylinder group is found
    131  * by using a rotor that sweeps the cylinder groups. When a new group of
    132  * blocks is needed, the sweep begins in the cylinder group following the
    133  * cylinder group from which the previous allocation was made. The sweep
    134  * continues until a cylinder group with greater than the average number
    135  * of free blocks is found. If the allocation is for the first block in an
    136  * indirect block, the information on the previous allocation is unavailable;
    137  * here a best guess is made based upon the logical block number being
    138  * allocated.
    139  *
    140  * If a section is already partially allocated, the policy is to
    141  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    142  * contiguous blocks and the beginning of the next is physically separated
    143  * so that the disk head will be in transit between them for at least
    144  * fs_rotdelay milliseconds.  This is to allow time for the processor to
    145  * schedule another I/O transfer.
    146  */
    147 ufs_daddr_t
    148 ffs_blkpref(struct inode *ip, ufs_daddr_t lbn, int indx, ufs_daddr_t *bap)
    149 {
    150 	struct fs *fs;
    151 	int cg;
    152 	int avgbfree, startcg;
    153 	ufs_daddr_t nextblk;
    154 
    155 	fs = ip->i_fs;
    156 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    157 		if (lbn < NDADDR + NINDIR(fs)) {
    158 			cg = ino_to_cg(fs, ip->i_number);
    159 			return (fs->fs_fpg * cg + fs->fs_frag);
    160 		}
    161 		/*
    162 		 * Find a cylinder with greater than average number of
    163 		 * unused data blocks.
    164 		 */
    165 		if (indx == 0 || bap[indx - 1] == 0)
    166 			startcg =
    167 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    168 		else
    169 			startcg = dtog(fs,
    170 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    171 		startcg %= fs->fs_ncg;
    172 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    173 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    174 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    175 				fs->fs_cgrotor = cg;
    176 				return (fs->fs_fpg * cg + fs->fs_frag);
    177 			}
    178 		for (cg = 0; cg <= startcg; cg++)
    179 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    180 				fs->fs_cgrotor = cg;
    181 				return (fs->fs_fpg * cg + fs->fs_frag);
    182 			}
    183 		return (0);
    184 	}
    185 	/*
    186 	 * One or more previous blocks have been laid out. If less
    187 	 * than fs_maxcontig previous blocks are contiguous, the
    188 	 * next block is requested contiguously, otherwise it is
    189 	 * requested rotationally delayed by fs_rotdelay milliseconds.
    190 	 */
    191 	nextblk = ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    192 	if (indx < fs->fs_maxcontig ||
    193 		ufs_rw32(bap[indx - fs->fs_maxcontig], UFS_FSNEEDSWAP(fs)) +
    194 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
    195 		return (nextblk);
    196 	if (fs->fs_rotdelay != 0)
    197 		/*
    198 		 * Here we convert ms of delay to frags as:
    199 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
    200 		 *	((sect/frag) * (ms/sec))
    201 		 * then round up to the next block.
    202 		 */
    203 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
    204 		    (NSPF(fs) * 1000), fs->fs_frag);
    205 	return (nextblk);
    206 }
    207 
    208 /*
    209  * Implement the cylinder overflow algorithm.
    210  *
    211  * The policy implemented by this algorithm is:
    212  *   1) allocate the block in its requested cylinder group.
    213  *   2) quadradically rehash on the cylinder group number.
    214  *   3) brute force search for a free block.
    215  *
    216  * `size':	size for data blocks, mode for inodes
    217  */
    218 /*VARARGS5*/
    219 static u_long
    220 ffs_hashalloc(struct inode *ip, int cg, long pref, int size,
    221     ufs_daddr_t (*allocator)(struct inode *, int, ufs_daddr_t, int))
    222 {
    223 	struct fs *fs;
    224 	long result;
    225 	int i, icg = cg;
    226 
    227 	fs = ip->i_fs;
    228 	/*
    229 	 * 1: preferred cylinder group
    230 	 */
    231 	result = (*allocator)(ip, cg, pref, size);
    232 	if (result)
    233 		return (result);
    234 	/*
    235 	 * 2: quadratic rehash
    236 	 */
    237 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    238 		cg += i;
    239 		if (cg >= fs->fs_ncg)
    240 			cg -= fs->fs_ncg;
    241 		result = (*allocator)(ip, cg, 0, size);
    242 		if (result)
    243 			return (result);
    244 	}
    245 	/*
    246 	 * 3: brute force search
    247 	 * Note that we start at i == 2, since 0 was checked initially,
    248 	 * and 1 is always checked in the quadratic rehash.
    249 	 */
    250 	cg = (icg + 2) % fs->fs_ncg;
    251 	for (i = 2; i < fs->fs_ncg; i++) {
    252 		result = (*allocator)(ip, cg, 0, size);
    253 		if (result)
    254 			return (result);
    255 		cg++;
    256 		if (cg == fs->fs_ncg)
    257 			cg = 0;
    258 	}
    259 	return (0);
    260 }
    261 
    262 /*
    263  * Determine whether a block can be allocated.
    264  *
    265  * Check to see if a block of the appropriate size is available,
    266  * and if it is, allocate it.
    267  */
    268 static ufs_daddr_t
    269 ffs_alloccg(struct inode *ip, int cg, ufs_daddr_t bpref, int size)
    270 {
    271 	struct cg *cgp;
    272 	struct buf *bp;
    273 	ufs_daddr_t bno, blkno;
    274 	int error, frags, allocsiz, i;
    275 	struct fs *fs = ip->i_fs;
    276 	const int needswap = UFS_FSNEEDSWAP(fs);
    277 
    278 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
    279 		return (0);
    280 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    281 		(int)fs->fs_cgsize, &bp);
    282 	if (error) {
    283 		brelse(bp);
    284 		return (0);
    285 	}
    286 	cgp = (struct cg *)bp->b_data;
    287 	if (!cg_chkmagic(cgp, needswap) ||
    288 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
    289 		brelse(bp);
    290 		return (0);
    291 	}
    292 	if (size == fs->fs_bsize) {
    293 		bno = ffs_alloccgblk(ip, bp, bpref);
    294 		bdwrite(bp);
    295 		return (bno);
    296 	}
    297 	/*
    298 	 * check to see if any fragments are already available
    299 	 * allocsiz is the size which will be allocated, hacking
    300 	 * it down to a smaller size if necessary
    301 	 */
    302 	frags = numfrags(fs, size);
    303 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
    304 		if (cgp->cg_frsum[allocsiz] != 0)
    305 			break;
    306 	if (allocsiz == fs->fs_frag) {
    307 		/*
    308 		 * no fragments were available, so a block will be
    309 		 * allocated, and hacked up
    310 		 */
    311 		if (cgp->cg_cs.cs_nbfree == 0) {
    312 			brelse(bp);
    313 			return (0);
    314 		}
    315 		bno = ffs_alloccgblk(ip, bp, bpref);
    316 		bpref = dtogd(fs, bno);
    317 		for (i = frags; i < fs->fs_frag; i++)
    318 			setbit(cg_blksfree(cgp, needswap), bpref + i);
    319 		i = fs->fs_frag - frags;
    320 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    321 		fs->fs_cstotal.cs_nffree += i;
    322 		fs->fs_cs(fs, cg).cs_nffree += i;
    323 		fs->fs_fmod = 1;
    324 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
    325 		bdwrite(bp);
    326 		return (bno);
    327 	}
    328 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
    329 	for (i = 0; i < frags; i++)
    330 		clrbit(cg_blksfree(cgp, needswap), bno + i);
    331 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
    332 	fs->fs_cstotal.cs_nffree -= frags;
    333 	fs->fs_cs(fs, cg).cs_nffree -= frags;
    334 	fs->fs_fmod = 1;
    335 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
    336 	if (frags != allocsiz)
    337 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
    338 	blkno = cg * fs->fs_fpg + bno;
    339 	bdwrite(bp);
    340 	return blkno;
    341 }
    342 
    343 /*
    344  * Allocate a block in a cylinder group.
    345  *
    346  * This algorithm implements the following policy:
    347  *   1) allocate the requested block.
    348  *   2) allocate a rotationally optimal block in the same cylinder.
    349  *   3) allocate the next available block on the block rotor for the
    350  *      specified cylinder group.
    351  * Note that this routine only allocates fs_bsize blocks; these
    352  * blocks may be fragmented by the routine that allocates them.
    353  */
    354 static ufs_daddr_t
    355 ffs_alloccgblk(struct inode *ip, struct buf *bp, ufs_daddr_t bpref)
    356 {
    357 	struct cg *cgp;
    358 	ufs_daddr_t bno, blkno;
    359 	int cylno, pos, delta;
    360 	short *cylbp;
    361 	int i;
    362 	struct fs *fs = ip->i_fs;
    363 	const int needswap = UFS_FSNEEDSWAP(fs);
    364 
    365 	cgp = (struct cg *)bp->b_data;
    366 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
    367 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
    368 		goto norot;
    369 	}
    370 	bpref = blknum(fs, bpref);
    371 	bpref = dtogd(fs, bpref);
    372 	/*
    373 	 * if the requested block is available, use it
    374 	 */
    375 	if (ffs_isblock(fs, cg_blksfree(cgp, needswap),
    376 		fragstoblks(fs, bpref))) {
    377 		bno = bpref;
    378 		goto gotit;
    379 	}
    380 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
    381 		/*
    382 		 * Block layout information is not available.
    383 		 * Leaving bpref unchanged means we take the
    384 		 * next available free block following the one
    385 		 * we just allocated. Hopefully this will at
    386 		 * least hit a track cache on drives of unknown
    387 		 * geometry (e.g. SCSI).
    388 		 */
    389 		goto norot;
    390 	}
    391 	/*
    392 	 * check for a block available on the same cylinder
    393 	 */
    394 	cylno = cbtocylno(fs, bpref);
    395 	if (cg_blktot(cgp, needswap)[cylno] == 0)
    396 		goto norot;
    397 	/*
    398 	 * check the summary information to see if a block is
    399 	 * available in the requested cylinder starting at the
    400 	 * requested rotational position and proceeding around.
    401 	 */
    402 	cylbp = cg_blks(fs, cgp, cylno, needswap);
    403 	pos = cbtorpos(fs, bpref);
    404 	for (i = pos; i < fs->fs_nrpos; i++)
    405 		if (ufs_rw16(cylbp[i], needswap) > 0)
    406 			break;
    407 	if (i == fs->fs_nrpos)
    408 		for (i = 0; i < pos; i++)
    409 			if (ufs_rw16(cylbp[i], needswap) > 0)
    410 				break;
    411 	if (ufs_rw16(cylbp[i], needswap) > 0) {
    412 		/*
    413 		 * found a rotational position, now find the actual
    414 		 * block. A panic if none is actually there.
    415 		 */
    416 		pos = cylno % fs->fs_cpc;
    417 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
    418 		if (fs_postbl(fs, pos)[i] == -1) {
    419 			errx(1,
    420 			    "ffs_alloccgblk: cyl groups corrupted: pos %d i %d",
    421 			    pos, i);
    422 		}
    423 		for (i = fs_postbl(fs, pos)[i];; ) {
    424 			if (ffs_isblock(fs, cg_blksfree(cgp, needswap), bno + i)) {
    425 				bno = blkstofrags(fs, (bno + i));
    426 				goto gotit;
    427 			}
    428 			delta = fs_rotbl(fs)[i];
    429 			if (delta <= 0 ||
    430 			    delta + i > fragstoblks(fs, fs->fs_fpg))
    431 				break;
    432 			i += delta;
    433 		}
    434 		errx(1, "ffs_alloccgblk: can't find blk in cyl: pos %d i %d",
    435 		    pos, i);
    436 	}
    437 norot:
    438 	/*
    439 	 * no blocks in the requested cylinder, so take next
    440 	 * available one in this cylinder group.
    441 	 */
    442 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
    443 	if (bno < 0)
    444 		return (0);
    445 	cgp->cg_rotor = ufs_rw32(bno, needswap);
    446 gotit:
    447 	blkno = fragstoblks(fs, bno);
    448 	ffs_clrblock(fs, cg_blksfree(cgp, needswap), (long)blkno);
    449 	ffs_clusteracct(fs, cgp, blkno, -1);
    450 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
    451 	fs->fs_cstotal.cs_nbfree--;
    452 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
    453 	cylno = cbtocylno(fs, bno);
    454 	ufs_add16(cg_blks(fs, cgp, cylno, needswap)[cbtorpos(fs, bno)], -1,
    455 	    needswap);
    456 	ufs_add32(cg_blktot(cgp, needswap)[cylno], -1, needswap);
    457 	fs->fs_fmod = 1;
    458 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
    459 	return (blkno);
    460 }
    461 
    462 /*
    463  * Free a block or fragment.
    464  *
    465  * The specified block or fragment is placed back in the
    466  * free map. If a fragment is deallocated, a possible
    467  * block reassembly is checked.
    468  */
    469 void
    470 ffs_blkfree(struct inode *ip, ufs_daddr_t bno, long size)
    471 {
    472 	struct cg *cgp;
    473 	struct buf *bp;
    474 	ufs_daddr_t blkno;
    475 	int i, error, cg, blk, frags, bbase;
    476 	struct fs *fs = ip->i_fs;
    477 	const int needswap = UFS_FSNEEDSWAP(fs);
    478 
    479 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
    480 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
    481 		errx(1, "blkfree: bad size: bno %u bsize %d size %ld",
    482 		    bno, fs->fs_bsize, size);
    483 	}
    484 	cg = dtog(fs, bno);
    485 	if ((u_int)bno >= fs->fs_size) {
    486 		warnx("bad block %d, ino %d\n", bno, ip->i_number);
    487 		return;
    488 	}
    489 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    490 		(int)fs->fs_cgsize, &bp);
    491 	if (error) {
    492 		brelse(bp);
    493 		return;
    494 	}
    495 	cgp = (struct cg *)bp->b_data;
    496 	if (!cg_chkmagic(cgp, needswap)) {
    497 		brelse(bp);
    498 		return;
    499 	}
    500 	bno = dtogd(fs, bno);
    501 	if (size == fs->fs_bsize) {
    502 		blkno = fragstoblks(fs, bno);
    503 		if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    504 			errx(1, "blkfree: freeing free block %d", bno);
    505 		}
    506 		ffs_setblock(fs, cg_blksfree(cgp, needswap), blkno);
    507 		ffs_clusteracct(fs, cgp, blkno, 1);
    508 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    509 		fs->fs_cstotal.cs_nbfree++;
    510 		fs->fs_cs(fs, cg).cs_nbfree++;
    511 		i = cbtocylno(fs, bno);
    512 		ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs, bno)], 1,
    513 		    needswap);
    514 		ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    515 	} else {
    516 		bbase = bno - fragnum(fs, bno);
    517 		/*
    518 		 * decrement the counts associated with the old frags
    519 		 */
    520 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    521 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
    522 		/*
    523 		 * deallocate the fragment
    524 		 */
    525 		frags = numfrags(fs, size);
    526 		for (i = 0; i < frags; i++) {
    527 			if (isset(cg_blksfree(cgp, needswap), bno + i)) {
    528 				errx(1, "blkfree: freeing free frag: block %d",
    529 				    bno + i);
    530 			}
    531 			setbit(cg_blksfree(cgp, needswap), bno + i);
    532 		}
    533 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    534 		fs->fs_cstotal.cs_nffree += i;
    535 		fs->fs_cs(fs, cg).cs_nffree += i;
    536 		/*
    537 		 * add back in counts associated with the new frags
    538 		 */
    539 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    540 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
    541 		/*
    542 		 * if a complete block has been reassembled, account for it
    543 		 */
    544 		blkno = fragstoblks(fs, bbase);
    545 		if (ffs_isblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    546 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
    547 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
    548 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
    549 			ffs_clusteracct(fs, cgp, blkno, 1);
    550 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    551 			fs->fs_cstotal.cs_nbfree++;
    552 			fs->fs_cs(fs, cg).cs_nbfree++;
    553 			i = cbtocylno(fs, bbase);
    554 			ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs,
    555 								bbase)], 1,
    556 			    needswap);
    557 			ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    558 		}
    559 	}
    560 	fs->fs_fmod = 1;
    561 	bdwrite(bp);
    562 }
    563 
    564 
    565 static int
    566 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
    567 {
    568 	const u_char *end = &cp[size];
    569 
    570 	while (cp < end && (table[*cp] & mask) == 0)
    571 		cp++;
    572 	return (end - cp);
    573 }
    574 
    575 /*
    576  * Find a block of the specified size in the specified cylinder group.
    577  *
    578  * It is a panic if a request is made to find a block if none are
    579  * available.
    580  */
    581 static ufs_daddr_t
    582 ffs_mapsearch(struct fs *fs, struct cg *cgp, ufs_daddr_t bpref, int allocsiz)
    583 {
    584 	ufs_daddr_t bno;
    585 	int start, len, loc, i;
    586 	int blk, field, subfield, pos;
    587 	int ostart, olen;
    588 	const int needswap = UFS_FSNEEDSWAP(fs);
    589 
    590 	/*
    591 	 * find the fragment by searching through the free block
    592 	 * map for an appropriate bit pattern
    593 	 */
    594 	if (bpref)
    595 		start = dtogd(fs, bpref) / NBBY;
    596 	else
    597 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
    598 	len = howmany(fs->fs_fpg, NBBY) - start;
    599 	ostart = start;
    600 	olen = len;
    601 	loc = scanc((u_int)len,
    602 		(const u_char *)&cg_blksfree(cgp, needswap)[start],
    603 		(const u_char *)fragtbl[fs->fs_frag],
    604 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    605 	if (loc == 0) {
    606 		len = start + 1;
    607 		start = 0;
    608 		loc = scanc((u_int)len,
    609 			(const u_char *)&cg_blksfree(cgp, needswap)[0],
    610 			(const u_char *)fragtbl[fs->fs_frag],
    611 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    612 		if (loc == 0) {
    613 			errx(1,
    614     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
    615 				ostart, olen,
    616 				ufs_rw32(cgp->cg_freeoff, needswap),
    617 				(long)cg_blksfree(cgp, needswap) - (long)cgp);
    618 			/* NOTREACHED */
    619 		}
    620 	}
    621 	bno = (start + len - loc) * NBBY;
    622 	cgp->cg_frotor = ufs_rw32(bno, needswap);
    623 	/*
    624 	 * found the byte in the map
    625 	 * sift through the bits to find the selected frag
    626 	 */
    627 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
    628 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
    629 		blk <<= 1;
    630 		field = around[allocsiz];
    631 		subfield = inside[allocsiz];
    632 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
    633 			if ((blk & field) == subfield)
    634 				return (bno + pos);
    635 			field <<= 1;
    636 			subfield <<= 1;
    637 		}
    638 	}
    639 	errx(1, "ffs_alloccg: block not in map: bno %d", bno);
    640 	return (-1);
    641 }
    642 
    643 /*
    644  * Update the cluster map because of an allocation or free.
    645  *
    646  * Cnt == 1 means free; cnt == -1 means allocating.
    647  */
    648 void
    649 ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs_daddr_t blkno, int cnt)
    650 {
    651 	int32_t *sump;
    652 	int32_t *lp;
    653 	u_char *freemapp, *mapp;
    654 	int i, start, end, forw, back, map, bit;
    655 	const int needswap = UFS_FSNEEDSWAP(fs);
    656 
    657 	if (fs->fs_contigsumsize <= 0)
    658 		return;
    659 	freemapp = cg_clustersfree(cgp, needswap);
    660 	sump = cg_clustersum(cgp, needswap);
    661 	/*
    662 	 * Allocate or clear the actual block.
    663 	 */
    664 	if (cnt > 0)
    665 		setbit(freemapp, blkno);
    666 	else
    667 		clrbit(freemapp, blkno);
    668 	/*
    669 	 * Find the size of the cluster going forward.
    670 	 */
    671 	start = blkno + 1;
    672 	end = start + fs->fs_contigsumsize;
    673 	if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
    674 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
    675 	mapp = &freemapp[start / NBBY];
    676 	map = *mapp++;
    677 	bit = 1 << (start % NBBY);
    678 	for (i = start; i < end; i++) {
    679 		if ((map & bit) == 0)
    680 			break;
    681 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
    682 			bit <<= 1;
    683 		} else {
    684 			map = *mapp++;
    685 			bit = 1;
    686 		}
    687 	}
    688 	forw = i - start;
    689 	/*
    690 	 * Find the size of the cluster going backward.
    691 	 */
    692 	start = blkno - 1;
    693 	end = start - fs->fs_contigsumsize;
    694 	if (end < 0)
    695 		end = -1;
    696 	mapp = &freemapp[start / NBBY];
    697 	map = *mapp--;
    698 	bit = 1 << (start % NBBY);
    699 	for (i = start; i > end; i--) {
    700 		if ((map & bit) == 0)
    701 			break;
    702 		if ((i & (NBBY - 1)) != 0) {
    703 			bit >>= 1;
    704 		} else {
    705 			map = *mapp--;
    706 			bit = 1 << (NBBY - 1);
    707 		}
    708 	}
    709 	back = start - i;
    710 	/*
    711 	 * Account for old cluster and the possibly new forward and
    712 	 * back clusters.
    713 	 */
    714 	i = back + forw + 1;
    715 	if (i > fs->fs_contigsumsize)
    716 		i = fs->fs_contigsumsize;
    717 	ufs_add32(sump[i], cnt, needswap);
    718 	if (back > 0)
    719 		ufs_add32(sump[back], -cnt, needswap);
    720 	if (forw > 0)
    721 		ufs_add32(sump[forw], -cnt, needswap);
    722 
    723 	/*
    724 	 * Update cluster summary information.
    725 	 */
    726 	lp = &sump[fs->fs_contigsumsize];
    727 	for (i = fs->fs_contigsumsize; i > 0; i--)
    728 		if (ufs_rw32(*lp--, needswap) > 0)
    729 			break;
    730 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
    731 }
    732