Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.13
      1 /*	$NetBSD: ffs_alloc.c,v 1.13 2003/08/07 11:25:33 agc Exp $	*/
      2 /* From: NetBSD: ffs_alloc.c,v 1.50 2001/09/06 02:16:01 lukem Exp */
      3 
      4 /*
      5  * Copyright (c) 2002 Networks Associates Technology, Inc.
      6  * All rights reserved.
      7  *
      8  * This software was developed for the FreeBSD Project by Marshall
      9  * Kirk McKusick and Network Associates Laboratories, the Security
     10  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
     11  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
     12  * research program
     13  *
     14  * Copyright (c) 1982, 1986, 1989, 1993
     15  *	The Regents of the University of California.  All rights reserved.
     16  *
     17  * Redistribution and use in source and binary forms, with or without
     18  * modification, are permitted provided that the following conditions
     19  * are met:
     20  * 1. Redistributions of source code must retain the above copyright
     21  *    notice, this list of conditions and the following disclaimer.
     22  * 2. Redistributions in binary form must reproduce the above copyright
     23  *    notice, this list of conditions and the following disclaimer in the
     24  *    documentation and/or other materials provided with the distribution.
     25  * 3. Neither the name of the University nor the names of its contributors
     26  *    may be used to endorse or promote products derived from this software
     27  *    without specific prior written permission.
     28  *
     29  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     30  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     31  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     32  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     33  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     34  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     35  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     36  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     37  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     38  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     39  * SUCH DAMAGE.
     40  *
     41  *	@(#)ffs_alloc.c	8.19 (Berkeley) 7/13/95
     42  */
     43 
     44 #include <sys/cdefs.h>
     45 #if defined(__RCSID) && !defined(__lint)
     46 __RCSID("$NetBSD: ffs_alloc.c,v 1.13 2003/08/07 11:25:33 agc Exp $");
     47 #endif	/* !__lint */
     48 
     49 #include <sys/param.h>
     50 #include <sys/time.h>
     51 
     52 #include <errno.h>
     53 
     54 #include "makefs.h"
     55 
     56 #include <ufs/ufs/dinode.h>
     57 #include <ufs/ufs/ufs_bswap.h>
     58 #include <ufs/ffs/fs.h>
     59 
     60 #include "ffs/buf.h"
     61 #include "ffs/ufs_inode.h"
     62 #include "ffs/ffs_extern.h"
     63 
     64 
     65 static int scanc(u_int, const u_char *, const u_char *, int);
     66 
     67 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int);
     68 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t);
     69 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int,
     70 		     daddr_t (*)(struct inode *, int, daddr_t, int));
     71 static int32_t ffs_mapsearch(struct fs *, struct cg *, daddr_t, int);
     72 
     73 /* in ffs_tables.c */
     74 extern const int inside[], around[];
     75 extern const u_char * const fragtbl[];
     76 
     77 /*
     78  * Allocate a block in the file system.
     79  *
     80  * The size of the requested block is given, which must be some
     81  * multiple of fs_fsize and <= fs_bsize.
     82  * A preference may be optionally specified. If a preference is given
     83  * the following hierarchy is used to allocate a block:
     84  *   1) allocate the requested block.
     85  *   2) allocate a rotationally optimal block in the same cylinder.
     86  *   3) allocate a block in the same cylinder group.
     87  *   4) quadradically rehash into other cylinder groups, until an
     88  *      available block is located.
     89  * If no block preference is given the following hierarchy is used
     90  * to allocate a block:
     91  *   1) allocate a block in the cylinder group that contains the
     92  *      inode for the file.
     93  *   2) quadradically rehash into other cylinder groups, until an
     94  *      available block is located.
     95  */
     96 int
     97 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
     98     daddr_t *bnp)
     99 {
    100 	struct fs *fs = ip->i_fs;
    101 	daddr_t bno;
    102 	int cg;
    103 
    104 	*bnp = 0;
    105 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
    106 		errx(1, "ffs_alloc: bad size: bsize %d size %d",
    107 		    fs->fs_bsize, size);
    108 	}
    109 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
    110 		goto nospace;
    111 	if (bpref >= fs->fs_size)
    112 		bpref = 0;
    113 	if (bpref == 0)
    114 		cg = ino_to_cg(fs, ip->i_number);
    115 	else
    116 		cg = dtog(fs, bpref);
    117 	bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
    118 	if (bno > 0) {
    119 		DIP(ip, blocks) += size / DEV_BSIZE;
    120 		*bnp = bno;
    121 		return (0);
    122 	}
    123 nospace:
    124 	return (ENOSPC);
    125 }
    126 
    127 /*
    128  * Select the desired position for the next block in a file.  The file is
    129  * logically divided into sections. The first section is composed of the
    130  * direct blocks. Each additional section contains fs_maxbpg blocks.
    131  *
    132  * If no blocks have been allocated in the first section, the policy is to
    133  * request a block in the same cylinder group as the inode that describes
    134  * the file. If no blocks have been allocated in any other section, the
    135  * policy is to place the section in a cylinder group with a greater than
    136  * average number of free blocks.  An appropriate cylinder group is found
    137  * by using a rotor that sweeps the cylinder groups. When a new group of
    138  * blocks is needed, the sweep begins in the cylinder group following the
    139  * cylinder group from which the previous allocation was made. The sweep
    140  * continues until a cylinder group with greater than the average number
    141  * of free blocks is found. If the allocation is for the first block in an
    142  * indirect block, the information on the previous allocation is unavailable;
    143  * here a best guess is made based upon the logical block number being
    144  * allocated.
    145  *
    146  * If a section is already partially allocated, the policy is to
    147  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    148  * contiguous blocks and the beginning of the next is physically separated
    149  * so that the disk head will be in transit between them for at least
    150  * fs_rotdelay milliseconds.  This is to allow time for the processor to
    151  * schedule another I/O transfer.
    152  */
    153 /* XXX ondisk32 */
    154 daddr_t
    155 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int32_t *bap)
    156 {
    157 	struct fs *fs;
    158 	int cg;
    159 	int avgbfree, startcg;
    160 
    161 	fs = ip->i_fs;
    162 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    163 		if (lbn < NDADDR + NINDIR(fs)) {
    164 			cg = ino_to_cg(fs, ip->i_number);
    165 			return (fs->fs_fpg * cg + fs->fs_frag);
    166 		}
    167 		/*
    168 		 * Find a cylinder with greater than average number of
    169 		 * unused data blocks.
    170 		 */
    171 		if (indx == 0 || bap[indx - 1] == 0)
    172 			startcg =
    173 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    174 		else
    175 			startcg = dtog(fs,
    176 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    177 		startcg %= fs->fs_ncg;
    178 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    179 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    180 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
    181 				return (fs->fs_fpg * cg + fs->fs_frag);
    182 		for (cg = 0; cg <= startcg; cg++)
    183 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree)
    184 				return (fs->fs_fpg * cg + fs->fs_frag);
    185 		return (0);
    186 	}
    187 	/*
    188 	 * We just always try to lay things out contiguously.
    189 	 */
    190 	return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    191 }
    192 
    193 daddr_t
    194 ffs_blkpref_ufs2(ip, lbn, indx, bap)
    195 	struct inode *ip;
    196 	daddr_t lbn;
    197 	int indx;
    198 	int64_t *bap;
    199 {
    200 	struct fs *fs;
    201 	int cg;
    202 	int avgbfree, startcg;
    203 
    204 	fs = ip->i_fs;
    205 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    206 		if (lbn < NDADDR + NINDIR(fs)) {
    207 			cg = ino_to_cg(fs, ip->i_number);
    208 			return (fs->fs_fpg * cg + fs->fs_frag);
    209 		}
    210 		/*
    211 		 * Find a cylinder with greater than average number of
    212 		 * unused data blocks.
    213 		 */
    214 		if (indx == 0 || bap[indx - 1] == 0)
    215 			startcg =
    216 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    217 		else
    218 			startcg = dtog(fs,
    219 				ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    220 		startcg %= fs->fs_ncg;
    221 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    222 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    223 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    224 				return (fs->fs_fpg * cg + fs->fs_frag);
    225 			}
    226 		for (cg = 0; cg < startcg; cg++)
    227 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    228 				return (fs->fs_fpg * cg + fs->fs_frag);
    229 			}
    230 		return (0);
    231 	}
    232 	/*
    233 	 * We just always try to lay things out contiguously.
    234 	 */
    235 	return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    236 }
    237 
    238 /*
    239  * Implement the cylinder overflow algorithm.
    240  *
    241  * The policy implemented by this algorithm is:
    242  *   1) allocate the block in its requested cylinder group.
    243  *   2) quadradically rehash on the cylinder group number.
    244  *   3) brute force search for a free block.
    245  *
    246  * `size':	size for data blocks, mode for inodes
    247  */
    248 /*VARARGS5*/
    249 static daddr_t
    250 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref, int size,
    251     daddr_t (*allocator)(struct inode *, int, daddr_t, int))
    252 {
    253 	struct fs *fs;
    254 	daddr_t result;
    255 	int i, icg = cg;
    256 
    257 	fs = ip->i_fs;
    258 	/*
    259 	 * 1: preferred cylinder group
    260 	 */
    261 	result = (*allocator)(ip, cg, pref, size);
    262 	if (result)
    263 		return (result);
    264 	/*
    265 	 * 2: quadratic rehash
    266 	 */
    267 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    268 		cg += i;
    269 		if (cg >= fs->fs_ncg)
    270 			cg -= fs->fs_ncg;
    271 		result = (*allocator)(ip, cg, 0, size);
    272 		if (result)
    273 			return (result);
    274 	}
    275 	/*
    276 	 * 3: brute force search
    277 	 * Note that we start at i == 2, since 0 was checked initially,
    278 	 * and 1 is always checked in the quadratic rehash.
    279 	 */
    280 	cg = (icg + 2) % fs->fs_ncg;
    281 	for (i = 2; i < fs->fs_ncg; i++) {
    282 		result = (*allocator)(ip, cg, 0, size);
    283 		if (result)
    284 			return (result);
    285 		cg++;
    286 		if (cg == fs->fs_ncg)
    287 			cg = 0;
    288 	}
    289 	return (0);
    290 }
    291 
    292 /*
    293  * Determine whether a block can be allocated.
    294  *
    295  * Check to see if a block of the appropriate size is available,
    296  * and if it is, allocate it.
    297  */
    298 static daddr_t
    299 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size)
    300 {
    301 	struct cg *cgp;
    302 	struct buf *bp;
    303 	daddr_t bno, blkno;
    304 	int error, frags, allocsiz, i;
    305 	struct fs *fs = ip->i_fs;
    306 	const int needswap = UFS_FSNEEDSWAP(fs);
    307 
    308 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
    309 		return (0);
    310 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    311 		(int)fs->fs_cgsize, &bp);
    312 	if (error) {
    313 		brelse(bp);
    314 		return (0);
    315 	}
    316 	cgp = (struct cg *)bp->b_data;
    317 	if (!cg_chkmagic(cgp, needswap) ||
    318 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
    319 		brelse(bp);
    320 		return (0);
    321 	}
    322 	if (size == fs->fs_bsize) {
    323 		bno = ffs_alloccgblk(ip, bp, bpref);
    324 		bdwrite(bp);
    325 		return (bno);
    326 	}
    327 	/*
    328 	 * check to see if any fragments are already available
    329 	 * allocsiz is the size which will be allocated, hacking
    330 	 * it down to a smaller size if necessary
    331 	 */
    332 	frags = numfrags(fs, size);
    333 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
    334 		if (cgp->cg_frsum[allocsiz] != 0)
    335 			break;
    336 	if (allocsiz == fs->fs_frag) {
    337 		/*
    338 		 * no fragments were available, so a block will be
    339 		 * allocated, and hacked up
    340 		 */
    341 		if (cgp->cg_cs.cs_nbfree == 0) {
    342 			brelse(bp);
    343 			return (0);
    344 		}
    345 		bno = ffs_alloccgblk(ip, bp, bpref);
    346 		bpref = dtogd(fs, bno);
    347 		for (i = frags; i < fs->fs_frag; i++)
    348 			setbit(cg_blksfree(cgp, needswap), bpref + i);
    349 		i = fs->fs_frag - frags;
    350 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    351 		fs->fs_cstotal.cs_nffree += i;
    352 		fs->fs_cs(fs, cg).cs_nffree += i;
    353 		fs->fs_fmod = 1;
    354 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
    355 		bdwrite(bp);
    356 		return (bno);
    357 	}
    358 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
    359 	for (i = 0; i < frags; i++)
    360 		clrbit(cg_blksfree(cgp, needswap), bno + i);
    361 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
    362 	fs->fs_cstotal.cs_nffree -= frags;
    363 	fs->fs_cs(fs, cg).cs_nffree -= frags;
    364 	fs->fs_fmod = 1;
    365 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
    366 	if (frags != allocsiz)
    367 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
    368 	blkno = cg * fs->fs_fpg + bno;
    369 	bdwrite(bp);
    370 	return blkno;
    371 }
    372 
    373 /*
    374  * Allocate a block in a cylinder group.
    375  *
    376  * This algorithm implements the following policy:
    377  *   1) allocate the requested block.
    378  *   2) allocate a rotationally optimal block in the same cylinder.
    379  *   3) allocate the next available block on the block rotor for the
    380  *      specified cylinder group.
    381  * Note that this routine only allocates fs_bsize blocks; these
    382  * blocks may be fragmented by the routine that allocates them.
    383  */
    384 static daddr_t
    385 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref)
    386 {
    387 	struct cg *cgp;
    388 	daddr_t blkno;
    389 	int32_t bno;
    390 	struct fs *fs = ip->i_fs;
    391 	const int needswap = UFS_FSNEEDSWAP(fs);
    392 	u_int8_t *blksfree;
    393 
    394 	cgp = (struct cg *)bp->b_data;
    395 	blksfree = cg_blksfree(cgp, needswap);
    396 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
    397 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
    398 	} else {
    399 		bpref = blknum(fs, bpref);
    400 		bno = dtogd(fs, bpref);
    401 		/*
    402 		 * if the requested block is available, use it
    403 		 */
    404 		if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
    405 			goto gotit;
    406 	}
    407 	/*
    408 	 * Take the next available one in this cylinder group.
    409 	 */
    410 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
    411 	if (bno < 0)
    412 		return (0);
    413 	cgp->cg_rotor = ufs_rw32(bno, needswap);
    414 gotit:
    415 	blkno = fragstoblks(fs, bno);
    416 	ffs_clrblock(fs, blksfree, (long)blkno);
    417 	ffs_clusteracct(fs, cgp, blkno, -1);
    418 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
    419 	fs->fs_cstotal.cs_nbfree--;
    420 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
    421 	fs->fs_fmod = 1;
    422 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
    423 	return (blkno);
    424 }
    425 
    426 /*
    427  * Free a block or fragment.
    428  *
    429  * The specified block or fragment is placed back in the
    430  * free map. If a fragment is deallocated, a possible
    431  * block reassembly is checked.
    432  */
    433 void
    434 ffs_blkfree(struct inode *ip, daddr_t bno, long size)
    435 {
    436 	struct cg *cgp;
    437 	struct buf *bp;
    438 	int32_t fragno, cgbno;
    439 	int i, error, cg, blk, frags, bbase;
    440 	struct fs *fs = ip->i_fs;
    441 	const int needswap = UFS_FSNEEDSWAP(fs);
    442 
    443 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
    444 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
    445 		errx(1, "blkfree: bad size: bno %lld bsize %d size %ld",
    446 		    (long long)bno, fs->fs_bsize, size);
    447 	}
    448 	cg = dtog(fs, bno);
    449 	if (bno >= fs->fs_size) {
    450 		warnx("bad block %lld, ino %d", (long long)bno, ip->i_number);
    451 		return;
    452 	}
    453 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    454 		(int)fs->fs_cgsize, &bp);
    455 	if (error) {
    456 		brelse(bp);
    457 		return;
    458 	}
    459 	cgp = (struct cg *)bp->b_data;
    460 	if (!cg_chkmagic(cgp, needswap)) {
    461 		brelse(bp);
    462 		return;
    463 	}
    464 	cgbno = dtogd(fs, bno);
    465 	if (size == fs->fs_bsize) {
    466 		fragno = fragstoblks(fs, cgbno);
    467 		if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), fragno)) {
    468 			errx(1, "blkfree: freeing free block %lld",
    469 			    (long long)bno);
    470 		}
    471 		ffs_setblock(fs, cg_blksfree(cgp, needswap), fragno);
    472 		ffs_clusteracct(fs, cgp, fragno, 1);
    473 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    474 		fs->fs_cstotal.cs_nbfree++;
    475 		fs->fs_cs(fs, cg).cs_nbfree++;
    476 	} else {
    477 		bbase = cgbno - fragnum(fs, cgbno);
    478 		/*
    479 		 * decrement the counts associated with the old frags
    480 		 */
    481 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    482 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
    483 		/*
    484 		 * deallocate the fragment
    485 		 */
    486 		frags = numfrags(fs, size);
    487 		for (i = 0; i < frags; i++) {
    488 			if (isset(cg_blksfree(cgp, needswap), cgbno + i)) {
    489 				errx(1, "blkfree: freeing free frag: block %lld",
    490 				    (long long)(cgbno + i));
    491 			}
    492 			setbit(cg_blksfree(cgp, needswap), cgbno + i);
    493 		}
    494 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    495 		fs->fs_cstotal.cs_nffree += i;
    496 		fs->fs_cs(fs, cg).cs_nffree += i;
    497 		/*
    498 		 * add back in counts associated with the new frags
    499 		 */
    500 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    501 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
    502 		/*
    503 		 * if a complete block has been reassembled, account for it
    504 		 */
    505 		fragno = fragstoblks(fs, bbase);
    506 		if (ffs_isblock(fs, cg_blksfree(cgp, needswap), fragno)) {
    507 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
    508 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
    509 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
    510 			ffs_clusteracct(fs, cgp, fragno, 1);
    511 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    512 			fs->fs_cstotal.cs_nbfree++;
    513 			fs->fs_cs(fs, cg).cs_nbfree++;
    514 		}
    515 	}
    516 	fs->fs_fmod = 1;
    517 	bdwrite(bp);
    518 }
    519 
    520 
    521 static int
    522 scanc(u_int size, const u_char *cp, const u_char table[], int mask)
    523 {
    524 	const u_char *end = &cp[size];
    525 
    526 	while (cp < end && (table[*cp] & mask) == 0)
    527 		cp++;
    528 	return (end - cp);
    529 }
    530 
    531 /*
    532  * Find a block of the specified size in the specified cylinder group.
    533  *
    534  * It is a panic if a request is made to find a block if none are
    535  * available.
    536  */
    537 static int32_t
    538 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
    539 {
    540 	int32_t bno;
    541 	int start, len, loc, i;
    542 	int blk, field, subfield, pos;
    543 	int ostart, olen;
    544 	const int needswap = UFS_FSNEEDSWAP(fs);
    545 
    546 	/*
    547 	 * find the fragment by searching through the free block
    548 	 * map for an appropriate bit pattern
    549 	 */
    550 	if (bpref)
    551 		start = dtogd(fs, bpref) / NBBY;
    552 	else
    553 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
    554 	len = howmany(fs->fs_fpg, NBBY) - start;
    555 	ostart = start;
    556 	olen = len;
    557 	loc = scanc((u_int)len,
    558 		(const u_char *)&cg_blksfree(cgp, needswap)[start],
    559 		(const u_char *)fragtbl[fs->fs_frag],
    560 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    561 	if (loc == 0) {
    562 		len = start + 1;
    563 		start = 0;
    564 		loc = scanc((u_int)len,
    565 			(const u_char *)&cg_blksfree(cgp, needswap)[0],
    566 			(const u_char *)fragtbl[fs->fs_frag],
    567 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    568 		if (loc == 0) {
    569 			errx(1,
    570     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
    571 				ostart, olen,
    572 				ufs_rw32(cgp->cg_freeoff, needswap),
    573 				(long)cg_blksfree(cgp, needswap) - (long)cgp);
    574 			/* NOTREACHED */
    575 		}
    576 	}
    577 	bno = (start + len - loc) * NBBY;
    578 	cgp->cg_frotor = ufs_rw32(bno, needswap);
    579 	/*
    580 	 * found the byte in the map
    581 	 * sift through the bits to find the selected frag
    582 	 */
    583 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
    584 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
    585 		blk <<= 1;
    586 		field = around[allocsiz];
    587 		subfield = inside[allocsiz];
    588 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
    589 			if ((blk & field) == subfield)
    590 				return (bno + pos);
    591 			field <<= 1;
    592 			subfield <<= 1;
    593 		}
    594 	}
    595 	errx(1, "ffs_alloccg: block not in map: bno %lld", (long long)bno);
    596 	return (-1);
    597 }
    598 
    599 /*
    600  * Update the cluster map because of an allocation or free.
    601  *
    602  * Cnt == 1 means free; cnt == -1 means allocating.
    603  */
    604 void
    605 ffs_clusteracct(struct fs *fs, struct cg *cgp, int32_t blkno, int cnt)
    606 {
    607 	int32_t *sump;
    608 	int32_t *lp;
    609 	u_char *freemapp, *mapp;
    610 	int i, start, end, forw, back, map, bit;
    611 	const int needswap = UFS_FSNEEDSWAP(fs);
    612 
    613 	if (fs->fs_contigsumsize <= 0)
    614 		return;
    615 	freemapp = cg_clustersfree(cgp, needswap);
    616 	sump = cg_clustersum(cgp, needswap);
    617 	/*
    618 	 * Allocate or clear the actual block.
    619 	 */
    620 	if (cnt > 0)
    621 		setbit(freemapp, blkno);
    622 	else
    623 		clrbit(freemapp, blkno);
    624 	/*
    625 	 * Find the size of the cluster going forward.
    626 	 */
    627 	start = blkno + 1;
    628 	end = start + fs->fs_contigsumsize;
    629 	if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
    630 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
    631 	mapp = &freemapp[start / NBBY];
    632 	map = *mapp++;
    633 	bit = 1 << (start % NBBY);
    634 	for (i = start; i < end; i++) {
    635 		if ((map & bit) == 0)
    636 			break;
    637 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
    638 			bit <<= 1;
    639 		} else {
    640 			map = *mapp++;
    641 			bit = 1;
    642 		}
    643 	}
    644 	forw = i - start;
    645 	/*
    646 	 * Find the size of the cluster going backward.
    647 	 */
    648 	start = blkno - 1;
    649 	end = start - fs->fs_contigsumsize;
    650 	if (end < 0)
    651 		end = -1;
    652 	mapp = &freemapp[start / NBBY];
    653 	map = *mapp--;
    654 	bit = 1 << (start % NBBY);
    655 	for (i = start; i > end; i--) {
    656 		if ((map & bit) == 0)
    657 			break;
    658 		if ((i & (NBBY - 1)) != 0) {
    659 			bit >>= 1;
    660 		} else {
    661 			map = *mapp--;
    662 			bit = 1 << (NBBY - 1);
    663 		}
    664 	}
    665 	back = start - i;
    666 	/*
    667 	 * Account for old cluster and the possibly new forward and
    668 	 * back clusters.
    669 	 */
    670 	i = back + forw + 1;
    671 	if (i > fs->fs_contigsumsize)
    672 		i = fs->fs_contigsumsize;
    673 	ufs_add32(sump[i], cnt, needswap);
    674 	if (back > 0)
    675 		ufs_add32(sump[back], -cnt, needswap);
    676 	if (forw > 0)
    677 		ufs_add32(sump[forw], -cnt, needswap);
    678 
    679 	/*
    680 	 * Update cluster summary information.
    681 	 */
    682 	lp = &sump[fs->fs_contigsumsize];
    683 	for (i = fs->fs_contigsumsize; i > 0; i--)
    684 		if (ufs_rw32(*lp--, needswap) > 0)
    685 			break;
    686 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
    687 }
    688