Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.11
      1 /*	$NetBSD: ffs_alloc.c,v 1.11 2003/01/24 21:55:32 fvdl 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 #if defined(__RCSID) && !defined(__lint)
     41 __RCSID("$NetBSD: ffs_alloc.c,v 1.11 2003/01/24 21:55:32 fvdl Exp $");
     42 #endif	/* !__lint */
     43 
     44 #include <sys/param.h>
     45 #include <sys/time.h>
     46 
     47 #include <errno.h>
     48 
     49 #include "makefs.h"
     50 
     51 #include <ufs/ufs/dinode.h>
     52 #include <ufs/ufs/ufs_bswap.h>
     53 #include <ufs/ffs/fs.h>
     54 
     55 #include "ffs/buf.h"
     56 #include "ffs/ufs_inode.h"
     57 #include "ffs/ffs_extern.h"
     58 
     59 
     60 static int scanc(u_int, const u_char *, const u_char *, int);
     61 
     62 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int);
     63 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t);
     64 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int,
     65 		     daddr_t (*)(struct inode *, int, daddr_t, int));
     66 static daddr_t ffs_mapsearch(struct fs *, struct cg *, daddr_t, int);
     67 
     68 /* in ffs_tables.c */
     69 extern const int inside[], around[];
     70 extern const u_char * const fragtbl[];
     71 
     72 /*
     73  * Allocate a block in the file system.
     74  *
     75  * The size of the requested block is given, which must be some
     76  * multiple of fs_fsize and <= fs_bsize.
     77  * A preference may be optionally specified. If a preference is given
     78  * the following hierarchy is used to allocate a block:
     79  *   1) allocate the requested block.
     80  *   2) allocate a rotationally optimal block in the same cylinder.
     81  *   3) allocate a block in the same cylinder group.
     82  *   4) quadradically rehash into other cylinder groups, until an
     83  *      available block is located.
     84  * If no block preference is given the following hierarchy is used
     85  * to allocate a block:
     86  *   1) allocate a block in the cylinder group that contains the
     87  *      inode for the file.
     88  *   2) quadradically rehash into other cylinder groups, until an
     89  *      available block is located.
     90  */
     91 int
     92 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
     93     daddr_t *bnp)
     94 {
     95 	struct fs *fs = ip->i_fs;
     96 	daddr_t bno;
     97 	int cg;
     98 
     99 	*bnp = 0;
    100 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
    101 		errx(1, "ffs_alloc: bad size: bsize %d size %d",
    102 		    fs->fs_bsize, size);
    103 	}
    104 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
    105 		goto nospace;
    106 	if (bpref >= fs->fs_size)
    107 		bpref = 0;
    108 	if (bpref == 0)
    109 		cg = ino_to_cg(fs, ip->i_number);
    110 	else
    111 		cg = dtog(fs, bpref);
    112 	bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
    113 	if (bno > 0) {
    114 		ip->i_ffs_blocks += size / DEV_BSIZE;
    115 		*bnp = bno;
    116 		return (0);
    117 	}
    118 nospace:
    119 	return (ENOSPC);
    120 }
    121 
    122 /*
    123  * Select the desired position for the next block in a file.  The file is
    124  * logically divided into sections. The first section is composed of the
    125  * direct blocks. Each additional section contains fs_maxbpg blocks.
    126  *
    127  * If no blocks have been allocated in the first section, the policy is to
    128  * request a block in the same cylinder group as the inode that describes
    129  * the file. If no blocks have been allocated in any other section, the
    130  * policy is to place the section in a cylinder group with a greater than
    131  * average number of free blocks.  An appropriate cylinder group is found
    132  * by using a rotor that sweeps the cylinder groups. When a new group of
    133  * blocks is needed, the sweep begins in the cylinder group following the
    134  * cylinder group from which the previous allocation was made. The sweep
    135  * continues until a cylinder group with greater than the average number
    136  * of free blocks is found. If the allocation is for the first block in an
    137  * indirect block, the information on the previous allocation is unavailable;
    138  * here a best guess is made based upon the logical block number being
    139  * allocated.
    140  *
    141  * If a section is already partially allocated, the policy is to
    142  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    143  * contiguous blocks and the beginning of the next is physically separated
    144  * so that the disk head will be in transit between them for at least
    145  * fs_rotdelay milliseconds.  This is to allow time for the processor to
    146  * schedule another I/O transfer.
    147  */
    148 /* XXX ondisk32 */
    149 daddr_t
    150 ffs_blkpref(struct inode *ip, daddr_t lbn, int indx, int32_t *bap)
    151 {
    152 	struct fs *fs;
    153 	int cg;
    154 	int avgbfree, startcg;
    155 	daddr_t nextblk;
    156 
    157 	fs = ip->i_fs;
    158 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    159 		if (lbn < NDADDR + NINDIR(fs)) {
    160 			cg = ino_to_cg(fs, ip->i_number);
    161 			return (fs->fs_fpg * cg + fs->fs_frag);
    162 		}
    163 		/*
    164 		 * Find a cylinder with greater than average number of
    165 		 * unused data blocks.
    166 		 */
    167 		if (indx == 0 || bap[indx - 1] == 0)
    168 			startcg =
    169 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    170 		else
    171 			startcg = dtog(fs,
    172 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    173 		startcg %= fs->fs_ncg;
    174 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    175 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    176 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    177 				fs->fs_cgrotor = cg;
    178 				return (fs->fs_fpg * cg + fs->fs_frag);
    179 			}
    180 		for (cg = 0; cg <= startcg; cg++)
    181 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    182 				fs->fs_cgrotor = cg;
    183 				return (fs->fs_fpg * cg + fs->fs_frag);
    184 			}
    185 		return (0);
    186 	}
    187 	/*
    188 	 * One or more previous blocks have been laid out. If less
    189 	 * than fs_maxcontig previous blocks are contiguous, the
    190 	 * next block is requested contiguously, otherwise it is
    191 	 * requested rotationally delayed by fs_rotdelay milliseconds.
    192 	 */
    193 	/* XXX ondisk32 */
    194 	nextblk = ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    195 	if (indx < fs->fs_maxcontig ||
    196 		ufs_rw32(bap[indx - fs->fs_maxcontig], UFS_FSNEEDSWAP(fs)) +
    197 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
    198 		return (nextblk);
    199 	if (fs->fs_rotdelay != 0)
    200 		/*
    201 		 * Here we convert ms of delay to frags as:
    202 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
    203 		 *	((sect/frag) * (ms/sec))
    204 		 * then round up to the next block.
    205 		 */
    206 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
    207 		    (NSPF(fs) * 1000), fs->fs_frag);
    208 	return (nextblk);
    209 }
    210 
    211 /*
    212  * Implement the cylinder overflow algorithm.
    213  *
    214  * The policy implemented by this algorithm is:
    215  *   1) allocate the block in its requested cylinder group.
    216  *   2) quadradically rehash on the cylinder group number.
    217  *   3) brute force search for a free block.
    218  *
    219  * `size':	size for data blocks, mode for inodes
    220  */
    221 /*VARARGS5*/
    222 static daddr_t
    223 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref, int size,
    224     daddr_t (*allocator)(struct inode *, int, daddr_t, int))
    225 {
    226 	struct fs *fs;
    227 	daddr_t result;
    228 	int i, icg = cg;
    229 
    230 	fs = ip->i_fs;
    231 	/*
    232 	 * 1: preferred cylinder group
    233 	 */
    234 	result = (*allocator)(ip, cg, pref, size);
    235 	if (result)
    236 		return (result);
    237 	/*
    238 	 * 2: quadratic rehash
    239 	 */
    240 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    241 		cg += i;
    242 		if (cg >= fs->fs_ncg)
    243 			cg -= fs->fs_ncg;
    244 		result = (*allocator)(ip, cg, 0, size);
    245 		if (result)
    246 			return (result);
    247 	}
    248 	/*
    249 	 * 3: brute force search
    250 	 * Note that we start at i == 2, since 0 was checked initially,
    251 	 * and 1 is always checked in the quadratic rehash.
    252 	 */
    253 	cg = (icg + 2) % fs->fs_ncg;
    254 	for (i = 2; i < fs->fs_ncg; i++) {
    255 		result = (*allocator)(ip, cg, 0, size);
    256 		if (result)
    257 			return (result);
    258 		cg++;
    259 		if (cg == fs->fs_ncg)
    260 			cg = 0;
    261 	}
    262 	return (0);
    263 }
    264 
    265 /*
    266  * Determine whether a block can be allocated.
    267  *
    268  * Check to see if a block of the appropriate size is available,
    269  * and if it is, allocate it.
    270  */
    271 static daddr_t
    272 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
    273 {
    274 	struct cg *cgp;
    275 	struct buf *bp;
    276 	daddr_t bno, blkno;
    277 	int error, frags, allocsiz, i;
    278 	struct fs *fs = ip->i_fs;
    279 	const int needswap = UFS_FSNEEDSWAP(fs);
    280 
    281 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
    282 		return (0);
    283 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    284 		(int)fs->fs_cgsize, &bp);
    285 	if (error) {
    286 		brelse(bp);
    287 		return (0);
    288 	}
    289 	cgp = (struct cg *)bp->b_data;
    290 	if (!cg_chkmagic(cgp, needswap) ||
    291 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
    292 		brelse(bp);
    293 		return (0);
    294 	}
    295 	if (size == fs->fs_bsize) {
    296 		bno = ffs_alloccgblk(ip, bp, bpref);
    297 		bdwrite(bp);
    298 		return (bno);
    299 	}
    300 	/*
    301 	 * check to see if any fragments are already available
    302 	 * allocsiz is the size which will be allocated, hacking
    303 	 * it down to a smaller size if necessary
    304 	 */
    305 	frags = numfrags(fs, size);
    306 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
    307 		if (cgp->cg_frsum[allocsiz] != 0)
    308 			break;
    309 	if (allocsiz == fs->fs_frag) {
    310 		/*
    311 		 * no fragments were available, so a block will be
    312 		 * allocated, and hacked up
    313 		 */
    314 		if (cgp->cg_cs.cs_nbfree == 0) {
    315 			brelse(bp);
    316 			return (0);
    317 		}
    318 		bno = ffs_alloccgblk(ip, bp, bpref);
    319 		bpref = dtogd(fs, bno);
    320 		for (i = frags; i < fs->fs_frag; i++)
    321 			setbit(cg_blksfree(cgp, needswap), bpref + i);
    322 		i = fs->fs_frag - frags;
    323 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    324 		fs->fs_cstotal.cs_nffree += i;
    325 		fs->fs_cs(fs, cg).cs_nffree += i;
    326 		fs->fs_fmod = 1;
    327 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
    328 		bdwrite(bp);
    329 		return (bno);
    330 	}
    331 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
    332 	for (i = 0; i < frags; i++)
    333 		clrbit(cg_blksfree(cgp, needswap), bno + i);
    334 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
    335 	fs->fs_cstotal.cs_nffree -= frags;
    336 	fs->fs_cs(fs, cg).cs_nffree -= frags;
    337 	fs->fs_fmod = 1;
    338 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
    339 	if (frags != allocsiz)
    340 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
    341 	blkno = cg * fs->fs_fpg + bno;
    342 	bdwrite(bp);
    343 	return blkno;
    344 }
    345 
    346 /*
    347  * Allocate a block in a cylinder group.
    348  *
    349  * This algorithm implements the following policy:
    350  *   1) allocate the requested block.
    351  *   2) allocate a rotationally optimal block in the same cylinder.
    352  *   3) allocate the next available block on the block rotor for the
    353  *      specified cylinder group.
    354  * Note that this routine only allocates fs_bsize blocks; these
    355  * blocks may be fragmented by the routine that allocates them.
    356  */
    357 static daddr_t
    358 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref)
    359 {
    360 	struct cg *cgp;
    361 	daddr_t bno, blkno;
    362 	int cylno, pos, delta;
    363 	short *cylbp;
    364 	int i;
    365 	struct fs *fs = ip->i_fs;
    366 	const int needswap = UFS_FSNEEDSWAP(fs);
    367 
    368 	cgp = (struct cg *)bp->b_data;
    369 	/* XXX ondisk32 */
    370 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
    371 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
    372 		goto norot;
    373 	}
    374 	bpref = blknum(fs, bpref);
    375 	bpref = dtogd(fs, bpref);
    376 	/*
    377 	 * if the requested block is available, use it
    378 	 */
    379 	if (ffs_isblock(fs, cg_blksfree(cgp, needswap),
    380 		fragstoblks(fs, bpref))) {
    381 		bno = bpref;
    382 		goto gotit;
    383 	}
    384 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
    385 		/*
    386 		 * Block layout information is not available.
    387 		 * Leaving bpref unchanged means we take the
    388 		 * next available free block following the one
    389 		 * we just allocated. Hopefully this will at
    390 		 * least hit a track cache on drives of unknown
    391 		 * geometry (e.g. SCSI).
    392 		 */
    393 		goto norot;
    394 	}
    395 	/*
    396 	 * check for a block available on the same cylinder
    397 	 */
    398 	cylno = cbtocylno(fs, bpref);
    399 	if (cg_blktot(cgp, needswap)[cylno] == 0)
    400 		goto norot;
    401 	/*
    402 	 * check the summary information to see if a block is
    403 	 * available in the requested cylinder starting at the
    404 	 * requested rotational position and proceeding around.
    405 	 */
    406 	cylbp = cg_blks(fs, cgp, cylno, needswap);
    407 	pos = cbtorpos(fs, bpref);
    408 	for (i = pos; i < fs->fs_nrpos; i++)
    409 		if (ufs_rw16(cylbp[i], needswap) > 0)
    410 			break;
    411 	if (i == fs->fs_nrpos)
    412 		for (i = 0; i < pos; i++)
    413 			if (ufs_rw16(cylbp[i], needswap) > 0)
    414 				break;
    415 	if (ufs_rw16(cylbp[i], needswap) > 0) {
    416 		/*
    417 		 * found a rotational position, now find the actual
    418 		 * block. A panic if none is actually there.
    419 		 */
    420 		pos = cylno % fs->fs_cpc;
    421 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
    422 		if (fs_postbl(fs, pos)[i] == -1) {
    423 			errx(1,
    424 			    "ffs_alloccgblk: cyl groups corrupted: pos %d i %d",
    425 			    pos, i);
    426 		}
    427 		for (i = fs_postbl(fs, pos)[i];; ) {
    428 			if (ffs_isblock(fs, cg_blksfree(cgp, needswap), bno + i)) {
    429 				bno = blkstofrags(fs, (bno + i));
    430 				goto gotit;
    431 			}
    432 			delta = fs_rotbl(fs)[i];
    433 			if (delta <= 0 ||
    434 			    delta + i > fragstoblks(fs, fs->fs_fpg))
    435 				break;
    436 			i += delta;
    437 		}
    438 		errx(1, "ffs_alloccgblk: can't find blk in cyl: pos %d i %d",
    439 		    pos, i);
    440 	}
    441 norot:
    442 	/*
    443 	 * no blocks in the requested cylinder, so take next
    444 	 * available one in this cylinder group.
    445 	 */
    446 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
    447 	if (bno < 0)
    448 		return (0);
    449 	/* XXX ondisk32 */
    450 	cgp->cg_rotor = ufs_rw64(bno, needswap);
    451 gotit:
    452 	blkno = fragstoblks(fs, bno);
    453 	ffs_clrblock(fs, cg_blksfree(cgp, needswap), (long)blkno);
    454 	ffs_clusteracct(fs, cgp, blkno, -1);
    455 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
    456 	fs->fs_cstotal.cs_nbfree--;
    457 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
    458 	cylno = cbtocylno(fs, bno);
    459 	ufs_add16(cg_blks(fs, cgp, cylno, needswap)[cbtorpos(fs, bno)], -1,
    460 	    needswap);
    461 	ufs_add32(cg_blktot(cgp, needswap)[cylno], -1, needswap);
    462 	fs->fs_fmod = 1;
    463 	/* XXX ondisk32 */
    464 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
    465 	return (blkno);
    466 }
    467 
    468 /*
    469  * Free a block or fragment.
    470  *
    471  * The specified block or fragment is placed back in the
    472  * free map. If a fragment is deallocated, a possible
    473  * block reassembly is checked.
    474  */
    475 void
    476 ffs_blkfree(struct inode *ip, daddr_t bno, long size)
    477 {
    478 	struct cg *cgp;
    479 	struct buf *bp;
    480 	daddr_t blkno;
    481 	int i, error, cg, blk, frags, bbase;
    482 	struct fs *fs = ip->i_fs;
    483 	const int needswap = UFS_FSNEEDSWAP(fs);
    484 
    485 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
    486 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
    487 		errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
    488 		    (long long)bno, fs->fs_bsize, size);
    489 	}
    490 	cg = dtog(fs, bno);
    491 	if ((u_int)bno >= fs->fs_size) {
    492 		warnx("bad block %lld, ino %d", (long long)bno, ip->i_number);
    493 		return;
    494 	}
    495 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    496 		(int)fs->fs_cgsize, &bp);
    497 	if (error) {
    498 		brelse(bp);
    499 		return;
    500 	}
    501 	cgp = (struct cg *)bp->b_data;
    502 	if (!cg_chkmagic(cgp, needswap)) {
    503 		brelse(bp);
    504 		return;
    505 	}
    506 	bno = dtogd(fs, bno);
    507 	if (size == fs->fs_bsize) {
    508 		blkno = fragstoblks(fs, bno);
    509 		if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    510 			errx(1, "blkfree: freeing free block %lld",
    511 			    (long long)bno);
    512 		}
    513 		ffs_setblock(fs, cg_blksfree(cgp, needswap), blkno);
    514 		ffs_clusteracct(fs, cgp, blkno, 1);
    515 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    516 		fs->fs_cstotal.cs_nbfree++;
    517 		fs->fs_cs(fs, cg).cs_nbfree++;
    518 		i = cbtocylno(fs, bno);
    519 		ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs, bno)], 1,
    520 		    needswap);
    521 		ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    522 	} else {
    523 		bbase = bno - fragnum(fs, bno);
    524 		/*
    525 		 * decrement the counts associated with the old frags
    526 		 */
    527 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    528 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
    529 		/*
    530 		 * deallocate the fragment
    531 		 */
    532 		frags = numfrags(fs, size);
    533 		for (i = 0; i < frags; i++) {
    534 			if (isset(cg_blksfree(cgp, needswap), bno + i)) {
    535 				errx(1, "blkfree: freeing free frag: block %lld",
    536 				    (long long)(bno + i));
    537 			}
    538 			setbit(cg_blksfree(cgp, needswap), bno + i);
    539 		}
    540 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    541 		fs->fs_cstotal.cs_nffree += i;
    542 		fs->fs_cs(fs, cg).cs_nffree += i;
    543 		/*
    544 		 * add back in counts associated with the new frags
    545 		 */
    546 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    547 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
    548 		/*
    549 		 * if a complete block has been reassembled, account for it
    550 		 */
    551 		blkno = fragstoblks(fs, bbase);
    552 		if (ffs_isblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    553 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
    554 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
    555 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
    556 			ffs_clusteracct(fs, cgp, blkno, 1);
    557 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    558 			fs->fs_cstotal.cs_nbfree++;
    559 			fs->fs_cs(fs, cg).cs_nbfree++;
    560 			i = cbtocylno(fs, bbase);
    561 			ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs,
    562 								bbase)], 1,
    563 			    needswap);
    564 			ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    565 		}
    566 	}
    567 	fs->fs_fmod = 1;
    568 	bdwrite(bp);
    569 }
    570 
    571 
    572 static int
    573 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
    574 {
    575 	const u_char *end = &cp[size];
    576 
    577 	while (cp < end && (table[*cp] & mask) == 0)
    578 		cp++;
    579 	return (end - cp);
    580 }
    581 
    582 /*
    583  * Find a block of the specified size in the specified cylinder group.
    584  *
    585  * It is a panic if a request is made to find a block if none are
    586  * available.
    587  */
    588 static daddr_t
    589 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
    590 {
    591 	daddr_t bno;
    592 	int start, len, loc, i;
    593 	int blk, field, subfield, pos;
    594 	int ostart, olen;
    595 	const int needswap = UFS_FSNEEDSWAP(fs);
    596 
    597 	/*
    598 	 * find the fragment by searching through the free block
    599 	 * map for an appropriate bit pattern
    600 	 */
    601 	if (bpref)
    602 		start = dtogd(fs, bpref) / NBBY;
    603 	else
    604 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
    605 	len = howmany(fs->fs_fpg, NBBY) - start;
    606 	ostart = start;
    607 	olen = len;
    608 	loc = scanc((u_int)len,
    609 		(const u_char *)&cg_blksfree(cgp, needswap)[start],
    610 		(const u_char *)fragtbl[fs->fs_frag],
    611 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    612 	if (loc == 0) {
    613 		len = start + 1;
    614 		start = 0;
    615 		loc = scanc((u_int)len,
    616 			(const u_char *)&cg_blksfree(cgp, needswap)[0],
    617 			(const u_char *)fragtbl[fs->fs_frag],
    618 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    619 		if (loc == 0) {
    620 			errx(1,
    621     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
    622 				ostart, olen,
    623 				ufs_rw32(cgp->cg_freeoff, needswap),
    624 				(long)cg_blksfree(cgp, needswap) - (long)cgp);
    625 			/* NOTREACHED */
    626 		}
    627 	}
    628 	bno = (start + len - loc) * NBBY;
    629 	cgp->cg_frotor = ufs_rw32(bno, needswap);
    630 	/*
    631 	 * found the byte in the map
    632 	 * sift through the bits to find the selected frag
    633 	 */
    634 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
    635 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
    636 		blk <<= 1;
    637 		field = around[allocsiz];
    638 		subfield = inside[allocsiz];
    639 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
    640 			if ((blk & field) == subfield)
    641 				return (bno + pos);
    642 			field <<= 1;
    643 			subfield <<= 1;
    644 		}
    645 	}
    646 	errx(1, "ffs_alloccg: block not in map: bno %lld", (long long)bno);
    647 	return (-1);
    648 }
    649 
    650 /*
    651  * Update the cluster map because of an allocation or free.
    652  *
    653  * Cnt == 1 means free; cnt == -1 means allocating.
    654  */
    655 void
    656 ffs_clusteracct(struct fs *fs, struct cg *cgp, daddr_t blkno, int cnt)
    657 {
    658 	int32_t *sump;
    659 	int32_t *lp;
    660 	u_char *freemapp, *mapp;
    661 	int i, start, end, forw, back, map, bit;
    662 	const int needswap = UFS_FSNEEDSWAP(fs);
    663 
    664 	if (fs->fs_contigsumsize <= 0)
    665 		return;
    666 	freemapp = cg_clustersfree(cgp, needswap);
    667 	sump = cg_clustersum(cgp, needswap);
    668 	/*
    669 	 * Allocate or clear the actual block.
    670 	 */
    671 	if (cnt > 0)
    672 		setbit(freemapp, blkno);
    673 	else
    674 		clrbit(freemapp, blkno);
    675 	/*
    676 	 * Find the size of the cluster going forward.
    677 	 */
    678 	start = blkno + 1;
    679 	end = start + fs->fs_contigsumsize;
    680 	if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
    681 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
    682 	mapp = &freemapp[start / NBBY];
    683 	map = *mapp++;
    684 	bit = 1 << (start % NBBY);
    685 	for (i = start; i < end; i++) {
    686 		if ((map & bit) == 0)
    687 			break;
    688 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
    689 			bit <<= 1;
    690 		} else {
    691 			map = *mapp++;
    692 			bit = 1;
    693 		}
    694 	}
    695 	forw = i - start;
    696 	/*
    697 	 * Find the size of the cluster going backward.
    698 	 */
    699 	start = blkno - 1;
    700 	end = start - fs->fs_contigsumsize;
    701 	if (end < 0)
    702 		end = -1;
    703 	mapp = &freemapp[start / NBBY];
    704 	map = *mapp--;
    705 	bit = 1 << (start % NBBY);
    706 	for (i = start; i > end; i--) {
    707 		if ((map & bit) == 0)
    708 			break;
    709 		if ((i & (NBBY - 1)) != 0) {
    710 			bit >>= 1;
    711 		} else {
    712 			map = *mapp--;
    713 			bit = 1 << (NBBY - 1);
    714 		}
    715 	}
    716 	back = start - i;
    717 	/*
    718 	 * Account for old cluster and the possibly new forward and
    719 	 * back clusters.
    720 	 */
    721 	i = back + forw + 1;
    722 	if (i > fs->fs_contigsumsize)
    723 		i = fs->fs_contigsumsize;
    724 	ufs_add32(sump[i], cnt, needswap);
    725 	if (back > 0)
    726 		ufs_add32(sump[back], -cnt, needswap);
    727 	if (forw > 0)
    728 		ufs_add32(sump[forw], -cnt, needswap);
    729 
    730 	/*
    731 	 * Update cluster summary information.
    732 	 */
    733 	lp = &sump[fs->fs_contigsumsize];
    734 	for (i = fs->fs_contigsumsize; i > 0; i--)
    735 		if (ufs_rw32(*lp--, needswap) > 0)
    736 			break;
    737 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
    738 }
    739