Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.6
      1 /*	$NetBSD: ffs_alloc.c,v 1.6 2002/01/07 16:56:27 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 1.6 2002/01/07 16:56:27 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 "makefs.h"
     51 
     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 ufs_daddr_t ffs_alloccg(struct inode *, int, ufs_daddr_t, int);
     63 static ufs_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs_daddr_t);
     64 static u_long ffs_hashalloc(struct inode *, int, long, int,
     65 		     ufs_daddr_t (*)(struct inode *, int, ufs_daddr_t, int));
     66 static ufs_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs_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, ufs_daddr_t lbn, ufs_daddr_t bpref, int size,
     93     ufs_daddr_t *bnp)
     94 {
     95 	struct fs *fs = ip->i_fs;
     96 	ufs_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 = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
    113 	    			     ffs_alloccg);
    114 	if (bno > 0) {
    115 		ip->i_ffs_blocks += btodb(size);
    116 		*bnp = bno;
    117 		return (0);
    118 	}
    119 nospace:
    120 	return (ENOSPC);
    121 }
    122 
    123 /*
    124  * Select the desired position for the next block in a file.  The file is
    125  * logically divided into sections. The first section is composed of the
    126  * direct blocks. Each additional section contains fs_maxbpg blocks.
    127  *
    128  * If no blocks have been allocated in the first section, the policy is to
    129  * request a block in the same cylinder group as the inode that describes
    130  * the file. If no blocks have been allocated in any other section, the
    131  * policy is to place the section in a cylinder group with a greater than
    132  * average number of free blocks.  An appropriate cylinder group is found
    133  * by using a rotor that sweeps the cylinder groups. When a new group of
    134  * blocks is needed, the sweep begins in the cylinder group following the
    135  * cylinder group from which the previous allocation was made. The sweep
    136  * continues until a cylinder group with greater than the average number
    137  * of free blocks is found. If the allocation is for the first block in an
    138  * indirect block, the information on the previous allocation is unavailable;
    139  * here a best guess is made based upon the logical block number being
    140  * allocated.
    141  *
    142  * If a section is already partially allocated, the policy is to
    143  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    144  * contiguous blocks and the beginning of the next is physically separated
    145  * so that the disk head will be in transit between them for at least
    146  * fs_rotdelay milliseconds.  This is to allow time for the processor to
    147  * schedule another I/O transfer.
    148  */
    149 ufs_daddr_t
    150 ffs_blkpref(struct inode *ip, ufs_daddr_t lbn, int indx, ufs_daddr_t *bap)
    151 {
    152 	struct fs *fs;
    153 	int cg;
    154 	int avgbfree, startcg;
    155 	ufs_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 	nextblk = ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    194 	if (indx < fs->fs_maxcontig ||
    195 		ufs_rw32(bap[indx - fs->fs_maxcontig], UFS_FSNEEDSWAP(fs)) +
    196 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
    197 		return (nextblk);
    198 	if (fs->fs_rotdelay != 0)
    199 		/*
    200 		 * Here we convert ms of delay to frags as:
    201 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
    202 		 *	((sect/frag) * (ms/sec))
    203 		 * then round up to the next block.
    204 		 */
    205 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
    206 		    (NSPF(fs) * 1000), fs->fs_frag);
    207 	return (nextblk);
    208 }
    209 
    210 /*
    211  * Implement the cylinder overflow algorithm.
    212  *
    213  * The policy implemented by this algorithm is:
    214  *   1) allocate the block in its requested cylinder group.
    215  *   2) quadradically rehash on the cylinder group number.
    216  *   3) brute force search for a free block.
    217  *
    218  * `size':	size for data blocks, mode for inodes
    219  */
    220 /*VARARGS5*/
    221 static u_long
    222 ffs_hashalloc(struct inode *ip, int cg, long pref, int size,
    223     ufs_daddr_t (*allocator)(struct inode *, int, ufs_daddr_t, int))
    224 {
    225 	struct fs *fs;
    226 	long result;
    227 	int i, icg = cg;
    228 
    229 	fs = ip->i_fs;
    230 	/*
    231 	 * 1: preferred cylinder group
    232 	 */
    233 	result = (*allocator)(ip, cg, pref, size);
    234 	if (result)
    235 		return (result);
    236 	/*
    237 	 * 2: quadratic rehash
    238 	 */
    239 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    240 		cg += i;
    241 		if (cg >= fs->fs_ncg)
    242 			cg -= fs->fs_ncg;
    243 		result = (*allocator)(ip, cg, 0, size);
    244 		if (result)
    245 			return (result);
    246 	}
    247 	/*
    248 	 * 3: brute force search
    249 	 * Note that we start at i == 2, since 0 was checked initially,
    250 	 * and 1 is always checked in the quadratic rehash.
    251 	 */
    252 	cg = (icg + 2) % fs->fs_ncg;
    253 	for (i = 2; i < fs->fs_ncg; i++) {
    254 		result = (*allocator)(ip, cg, 0, size);
    255 		if (result)
    256 			return (result);
    257 		cg++;
    258 		if (cg == fs->fs_ncg)
    259 			cg = 0;
    260 	}
    261 	return (0);
    262 }
    263 
    264 /*
    265  * Determine whether a block can be allocated.
    266  *
    267  * Check to see if a block of the appropriate size is available,
    268  * and if it is, allocate it.
    269  */
    270 static ufs_daddr_t
    271 ffs_alloccg(struct inode *ip, int cg, ufs_daddr_t bpref, int size)
    272 {
    273 	struct cg *cgp;
    274 	struct buf *bp;
    275 	ufs_daddr_t bno, blkno;
    276 	int error, frags, allocsiz, i;
    277 	struct fs *fs = ip->i_fs;
    278 	const int needswap = UFS_FSNEEDSWAP(fs);
    279 
    280 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
    281 		return (0);
    282 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    283 		(int)fs->fs_cgsize, &bp);
    284 	if (error) {
    285 		brelse(bp);
    286 		return (0);
    287 	}
    288 	cgp = (struct cg *)bp->b_data;
    289 	if (!cg_chkmagic(cgp, needswap) ||
    290 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
    291 		brelse(bp);
    292 		return (0);
    293 	}
    294 	if (size == fs->fs_bsize) {
    295 		bno = ffs_alloccgblk(ip, bp, bpref);
    296 		bdwrite(bp);
    297 		return (bno);
    298 	}
    299 	/*
    300 	 * check to see if any fragments are already available
    301 	 * allocsiz is the size which will be allocated, hacking
    302 	 * it down to a smaller size if necessary
    303 	 */
    304 	frags = numfrags(fs, size);
    305 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
    306 		if (cgp->cg_frsum[allocsiz] != 0)
    307 			break;
    308 	if (allocsiz == fs->fs_frag) {
    309 		/*
    310 		 * no fragments were available, so a block will be
    311 		 * allocated, and hacked up
    312 		 */
    313 		if (cgp->cg_cs.cs_nbfree == 0) {
    314 			brelse(bp);
    315 			return (0);
    316 		}
    317 		bno = ffs_alloccgblk(ip, bp, bpref);
    318 		bpref = dtogd(fs, bno);
    319 		for (i = frags; i < fs->fs_frag; i++)
    320 			setbit(cg_blksfree(cgp, needswap), bpref + i);
    321 		i = fs->fs_frag - frags;
    322 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    323 		fs->fs_cstotal.cs_nffree += i;
    324 		fs->fs_cs(fs, cg).cs_nffree += i;
    325 		fs->fs_fmod = 1;
    326 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
    327 		bdwrite(bp);
    328 		return (bno);
    329 	}
    330 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
    331 	for (i = 0; i < frags; i++)
    332 		clrbit(cg_blksfree(cgp, needswap), bno + i);
    333 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
    334 	fs->fs_cstotal.cs_nffree -= frags;
    335 	fs->fs_cs(fs, cg).cs_nffree -= frags;
    336 	fs->fs_fmod = 1;
    337 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
    338 	if (frags != allocsiz)
    339 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
    340 	blkno = cg * fs->fs_fpg + bno;
    341 	bdwrite(bp);
    342 	return blkno;
    343 }
    344 
    345 /*
    346  * Allocate a block in a cylinder group.
    347  *
    348  * This algorithm implements the following policy:
    349  *   1) allocate the requested block.
    350  *   2) allocate a rotationally optimal block in the same cylinder.
    351  *   3) allocate the next available block on the block rotor for the
    352  *      specified cylinder group.
    353  * Note that this routine only allocates fs_bsize blocks; these
    354  * blocks may be fragmented by the routine that allocates them.
    355  */
    356 static ufs_daddr_t
    357 ffs_alloccgblk(struct inode *ip, struct buf *bp, ufs_daddr_t bpref)
    358 {
    359 	struct cg *cgp;
    360 	ufs_daddr_t bno, blkno;
    361 	int cylno, pos, delta;
    362 	short *cylbp;
    363 	int i;
    364 	struct fs *fs = ip->i_fs;
    365 	const int needswap = UFS_FSNEEDSWAP(fs);
    366 
    367 	cgp = (struct cg *)bp->b_data;
    368 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
    369 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
    370 		goto norot;
    371 	}
    372 	bpref = blknum(fs, bpref);
    373 	bpref = dtogd(fs, bpref);
    374 	/*
    375 	 * if the requested block is available, use it
    376 	 */
    377 	if (ffs_isblock(fs, cg_blksfree(cgp, needswap),
    378 		fragstoblks(fs, bpref))) {
    379 		bno = bpref;
    380 		goto gotit;
    381 	}
    382 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
    383 		/*
    384 		 * Block layout information is not available.
    385 		 * Leaving bpref unchanged means we take the
    386 		 * next available free block following the one
    387 		 * we just allocated. Hopefully this will at
    388 		 * least hit a track cache on drives of unknown
    389 		 * geometry (e.g. SCSI).
    390 		 */
    391 		goto norot;
    392 	}
    393 	/*
    394 	 * check for a block available on the same cylinder
    395 	 */
    396 	cylno = cbtocylno(fs, bpref);
    397 	if (cg_blktot(cgp, needswap)[cylno] == 0)
    398 		goto norot;
    399 	/*
    400 	 * check the summary information to see if a block is
    401 	 * available in the requested cylinder starting at the
    402 	 * requested rotational position and proceeding around.
    403 	 */
    404 	cylbp = cg_blks(fs, cgp, cylno, needswap);
    405 	pos = cbtorpos(fs, bpref);
    406 	for (i = pos; i < fs->fs_nrpos; i++)
    407 		if (ufs_rw16(cylbp[i], needswap) > 0)
    408 			break;
    409 	if (i == fs->fs_nrpos)
    410 		for (i = 0; i < pos; i++)
    411 			if (ufs_rw16(cylbp[i], needswap) > 0)
    412 				break;
    413 	if (ufs_rw16(cylbp[i], needswap) > 0) {
    414 		/*
    415 		 * found a rotational position, now find the actual
    416 		 * block. A panic if none is actually there.
    417 		 */
    418 		pos = cylno % fs->fs_cpc;
    419 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
    420 		if (fs_postbl(fs, pos)[i] == -1) {
    421 			errx(1,
    422 			    "ffs_alloccgblk: cyl groups corrupted: pos %d i %d",
    423 			    pos, i);
    424 		}
    425 		for (i = fs_postbl(fs, pos)[i];; ) {
    426 			if (ffs_isblock(fs, cg_blksfree(cgp, needswap), bno + i)) {
    427 				bno = blkstofrags(fs, (bno + i));
    428 				goto gotit;
    429 			}
    430 			delta = fs_rotbl(fs)[i];
    431 			if (delta <= 0 ||
    432 			    delta + i > fragstoblks(fs, fs->fs_fpg))
    433 				break;
    434 			i += delta;
    435 		}
    436 		errx(1, "ffs_alloccgblk: can't find blk in cyl: pos %d i %d",
    437 		    pos, i);
    438 	}
    439 norot:
    440 	/*
    441 	 * no blocks in the requested cylinder, so take next
    442 	 * available one in this cylinder group.
    443 	 */
    444 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
    445 	if (bno < 0)
    446 		return (0);
    447 	cgp->cg_rotor = ufs_rw32(bno, needswap);
    448 gotit:
    449 	blkno = fragstoblks(fs, bno);
    450 	ffs_clrblock(fs, cg_blksfree(cgp, needswap), (long)blkno);
    451 	ffs_clusteracct(fs, cgp, blkno, -1);
    452 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
    453 	fs->fs_cstotal.cs_nbfree--;
    454 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
    455 	cylno = cbtocylno(fs, bno);
    456 	ufs_add16(cg_blks(fs, cgp, cylno, needswap)[cbtorpos(fs, bno)], -1,
    457 	    needswap);
    458 	ufs_add32(cg_blktot(cgp, needswap)[cylno], -1, needswap);
    459 	fs->fs_fmod = 1;
    460 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
    461 	return (blkno);
    462 }
    463 
    464 /*
    465  * Free a block or fragment.
    466  *
    467  * The specified block or fragment is placed back in the
    468  * free map. If a fragment is deallocated, a possible
    469  * block reassembly is checked.
    470  */
    471 void
    472 ffs_blkfree(struct inode *ip, ufs_daddr_t bno, long size)
    473 {
    474 	struct cg *cgp;
    475 	struct buf *bp;
    476 	ufs_daddr_t blkno;
    477 	int i, error, cg, blk, frags, bbase;
    478 	struct fs *fs = ip->i_fs;
    479 	const int needswap = UFS_FSNEEDSWAP(fs);
    480 
    481 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
    482 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
    483 		errx(1, "blkfree: bad size: bno %u bsize %d size %ld",
    484 		    bno, fs->fs_bsize, size);
    485 	}
    486 	cg = dtog(fs, bno);
    487 	if ((u_int)bno >= fs->fs_size) {
    488 		warnx("bad block %d, ino %d\n", bno, ip->i_number);
    489 		return;
    490 	}
    491 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    492 		(int)fs->fs_cgsize, &bp);
    493 	if (error) {
    494 		brelse(bp);
    495 		return;
    496 	}
    497 	cgp = (struct cg *)bp->b_data;
    498 	if (!cg_chkmagic(cgp, needswap)) {
    499 		brelse(bp);
    500 		return;
    501 	}
    502 	bno = dtogd(fs, bno);
    503 	if (size == fs->fs_bsize) {
    504 		blkno = fragstoblks(fs, bno);
    505 		if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    506 			errx(1, "blkfree: freeing free block %d", bno);
    507 		}
    508 		ffs_setblock(fs, cg_blksfree(cgp, needswap), blkno);
    509 		ffs_clusteracct(fs, cgp, blkno, 1);
    510 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    511 		fs->fs_cstotal.cs_nbfree++;
    512 		fs->fs_cs(fs, cg).cs_nbfree++;
    513 		i = cbtocylno(fs, bno);
    514 		ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs, bno)], 1,
    515 		    needswap);
    516 		ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    517 	} else {
    518 		bbase = bno - fragnum(fs, bno);
    519 		/*
    520 		 * decrement the counts associated with the old frags
    521 		 */
    522 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    523 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
    524 		/*
    525 		 * deallocate the fragment
    526 		 */
    527 		frags = numfrags(fs, size);
    528 		for (i = 0; i < frags; i++) {
    529 			if (isset(cg_blksfree(cgp, needswap), bno + i)) {
    530 				errx(1, "blkfree: freeing free frag: block %d",
    531 				    bno + i);
    532 			}
    533 			setbit(cg_blksfree(cgp, needswap), bno + i);
    534 		}
    535 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    536 		fs->fs_cstotal.cs_nffree += i;
    537 		fs->fs_cs(fs, cg).cs_nffree += i;
    538 		/*
    539 		 * add back in counts associated with the new frags
    540 		 */
    541 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    542 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
    543 		/*
    544 		 * if a complete block has been reassembled, account for it
    545 		 */
    546 		blkno = fragstoblks(fs, bbase);
    547 		if (ffs_isblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    548 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
    549 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
    550 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
    551 			ffs_clusteracct(fs, cgp, blkno, 1);
    552 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    553 			fs->fs_cstotal.cs_nbfree++;
    554 			fs->fs_cs(fs, cg).cs_nbfree++;
    555 			i = cbtocylno(fs, bbase);
    556 			ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs,
    557 								bbase)], 1,
    558 			    needswap);
    559 			ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    560 		}
    561 	}
    562 	fs->fs_fmod = 1;
    563 	bdwrite(bp);
    564 }
    565 
    566 
    567 static int
    568 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
    569 {
    570 	const u_char *end = &cp[size];
    571 
    572 	while (cp < end && (table[*cp] & mask) == 0)
    573 		cp++;
    574 	return (end - cp);
    575 }
    576 
    577 /*
    578  * Find a block of the specified size in the specified cylinder group.
    579  *
    580  * It is a panic if a request is made to find a block if none are
    581  * available.
    582  */
    583 static ufs_daddr_t
    584 ffs_mapsearch(struct fs *fs, struct cg *cgp, ufs_daddr_t bpref, int allocsiz)
    585 {
    586 	ufs_daddr_t bno;
    587 	int start, len, loc, i;
    588 	int blk, field, subfield, pos;
    589 	int ostart, olen;
    590 	const int needswap = UFS_FSNEEDSWAP(fs);
    591 
    592 	/*
    593 	 * find the fragment by searching through the free block
    594 	 * map for an appropriate bit pattern
    595 	 */
    596 	if (bpref)
    597 		start = dtogd(fs, bpref) / NBBY;
    598 	else
    599 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
    600 	len = howmany(fs->fs_fpg, NBBY) - start;
    601 	ostart = start;
    602 	olen = len;
    603 	loc = scanc((u_int)len,
    604 		(const u_char *)&cg_blksfree(cgp, needswap)[start],
    605 		(const u_char *)fragtbl[fs->fs_frag],
    606 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    607 	if (loc == 0) {
    608 		len = start + 1;
    609 		start = 0;
    610 		loc = scanc((u_int)len,
    611 			(const u_char *)&cg_blksfree(cgp, needswap)[0],
    612 			(const u_char *)fragtbl[fs->fs_frag],
    613 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    614 		if (loc == 0) {
    615 			errx(1,
    616     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
    617 				ostart, olen,
    618 				ufs_rw32(cgp->cg_freeoff, needswap),
    619 				(long)cg_blksfree(cgp, needswap) - (long)cgp);
    620 			/* NOTREACHED */
    621 		}
    622 	}
    623 	bno = (start + len - loc) * NBBY;
    624 	cgp->cg_frotor = ufs_rw32(bno, needswap);
    625 	/*
    626 	 * found the byte in the map
    627 	 * sift through the bits to find the selected frag
    628 	 */
    629 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
    630 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
    631 		blk <<= 1;
    632 		field = around[allocsiz];
    633 		subfield = inside[allocsiz];
    634 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
    635 			if ((blk & field) == subfield)
    636 				return (bno + pos);
    637 			field <<= 1;
    638 			subfield <<= 1;
    639 		}
    640 	}
    641 	errx(1, "ffs_alloccg: block not in map: bno %d", bno);
    642 	return (-1);
    643 }
    644 
    645 /*
    646  * Update the cluster map because of an allocation or free.
    647  *
    648  * Cnt == 1 means free; cnt == -1 means allocating.
    649  */
    650 void
    651 ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs_daddr_t blkno, int cnt)
    652 {
    653 	int32_t *sump;
    654 	int32_t *lp;
    655 	u_char *freemapp, *mapp;
    656 	int i, start, end, forw, back, map, bit;
    657 	const int needswap = UFS_FSNEEDSWAP(fs);
    658 
    659 	if (fs->fs_contigsumsize <= 0)
    660 		return;
    661 	freemapp = cg_clustersfree(cgp, needswap);
    662 	sump = cg_clustersum(cgp, needswap);
    663 	/*
    664 	 * Allocate or clear the actual block.
    665 	 */
    666 	if (cnt > 0)
    667 		setbit(freemapp, blkno);
    668 	else
    669 		clrbit(freemapp, blkno);
    670 	/*
    671 	 * Find the size of the cluster going forward.
    672 	 */
    673 	start = blkno + 1;
    674 	end = start + fs->fs_contigsumsize;
    675 	if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
    676 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
    677 	mapp = &freemapp[start / NBBY];
    678 	map = *mapp++;
    679 	bit = 1 << (start % NBBY);
    680 	for (i = start; i < end; i++) {
    681 		if ((map & bit) == 0)
    682 			break;
    683 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
    684 			bit <<= 1;
    685 		} else {
    686 			map = *mapp++;
    687 			bit = 1;
    688 		}
    689 	}
    690 	forw = i - start;
    691 	/*
    692 	 * Find the size of the cluster going backward.
    693 	 */
    694 	start = blkno - 1;
    695 	end = start - fs->fs_contigsumsize;
    696 	if (end < 0)
    697 		end = -1;
    698 	mapp = &freemapp[start / NBBY];
    699 	map = *mapp--;
    700 	bit = 1 << (start % NBBY);
    701 	for (i = start; i > end; i--) {
    702 		if ((map & bit) == 0)
    703 			break;
    704 		if ((i & (NBBY - 1)) != 0) {
    705 			bit >>= 1;
    706 		} else {
    707 			map = *mapp--;
    708 			bit = 1 << (NBBY - 1);
    709 		}
    710 	}
    711 	back = start - i;
    712 	/*
    713 	 * Account for old cluster and the possibly new forward and
    714 	 * back clusters.
    715 	 */
    716 	i = back + forw + 1;
    717 	if (i > fs->fs_contigsumsize)
    718 		i = fs->fs_contigsumsize;
    719 	ufs_add32(sump[i], cnt, needswap);
    720 	if (back > 0)
    721 		ufs_add32(sump[back], -cnt, needswap);
    722 	if (forw > 0)
    723 		ufs_add32(sump[forw], -cnt, needswap);
    724 
    725 	/*
    726 	 * Update cluster summary information.
    727 	 */
    728 	lp = &sump[fs->fs_contigsumsize];
    729 	for (i = fs->fs_contigsumsize; i > 0; i--)
    730 		if (ufs_rw32(*lp--, needswap) > 0)
    731 			break;
    732 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
    733 }
    734