Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.6
      1  1.6  lukem /*	$NetBSD: ffs_alloc.c,v 1.6 2002/01/07 16:56:27 lukem Exp $	*/
      2  1.1  lukem /* From: NetBSD: ffs_alloc.c,v 1.50 2001/09/06 02:16:01 lukem Exp */
      3  1.1  lukem 
      4  1.1  lukem /*
      5  1.1  lukem  * Copyright (c) 1982, 1986, 1989, 1993
      6  1.1  lukem  *	The Regents of the University of California.  All rights reserved.
      7  1.1  lukem  *
      8  1.1  lukem  * Redistribution and use in source and binary forms, with or without
      9  1.1  lukem  * modification, are permitted provided that the following conditions
     10  1.1  lukem  * are met:
     11  1.1  lukem  * 1. Redistributions of source code must retain the above copyright
     12  1.1  lukem  *    notice, this list of conditions and the following disclaimer.
     13  1.1  lukem  * 2. Redistributions in binary form must reproduce the above copyright
     14  1.1  lukem  *    notice, this list of conditions and the following disclaimer in the
     15  1.1  lukem  *    documentation and/or other materials provided with the distribution.
     16  1.1  lukem  * 3. All advertising materials mentioning features or use of this software
     17  1.1  lukem  *    must display the following acknowledgement:
     18  1.1  lukem  *	This product includes software developed by the University of
     19  1.1  lukem  *	California, Berkeley and its contributors.
     20  1.1  lukem  * 4. Neither the name of the University nor the names of its contributors
     21  1.1  lukem  *    may be used to endorse or promote products derived from this software
     22  1.1  lukem  *    without specific prior written permission.
     23  1.1  lukem  *
     24  1.1  lukem  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  1.1  lukem  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  1.1  lukem  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  1.1  lukem  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  1.1  lukem  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  1.1  lukem  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  1.1  lukem  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  1.1  lukem  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  1.1  lukem  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  1.1  lukem  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  1.1  lukem  * SUCH DAMAGE.
     35  1.1  lukem  *
     36  1.1  lukem  *	@(#)ffs_alloc.c	8.19 (Berkeley) 7/13/95
     37  1.1  lukem  */
     38  1.2  lukem 
     39  1.2  lukem #include <sys/cdefs.h>
     40  1.2  lukem #ifndef __lint
     41  1.6  lukem __RCSID("$NetBSD: ffs_alloc.c,v 1.6 2002/01/07 16:56:27 lukem Exp $");
     42  1.2  lukem #endif	/* !__lint */
     43  1.1  lukem 
     44  1.1  lukem #include <sys/param.h>
     45  1.1  lukem #include <sys/time.h>
     46  1.1  lukem 
     47  1.1  lukem #include <err.h>
     48  1.1  lukem #include <errno.h>
     49  1.4  lukem 
     50  1.4  lukem #include "makefs.h"
     51  1.1  lukem 
     52  1.5  lukem #include <ufs/ufs/ufs_bswap.h>
     53  1.5  lukem #include <ufs/ffs/fs.h>
     54  1.1  lukem 
     55  1.1  lukem #include "ffs/buf.h"
     56  1.6  lukem #include "ffs/ufs_inode.h"
     57  1.1  lukem #include "ffs/ffs_extern.h"
     58  1.1  lukem 
     59  1.1  lukem 
     60  1.1  lukem static int scanc(u_int, const u_char *, const u_char *, int);
     61  1.1  lukem 
     62  1.1  lukem static ufs_daddr_t ffs_alloccg(struct inode *, int, ufs_daddr_t, int);
     63  1.1  lukem static ufs_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs_daddr_t);
     64  1.1  lukem static u_long ffs_hashalloc(struct inode *, int, long, int,
     65  1.1  lukem 		     ufs_daddr_t (*)(struct inode *, int, ufs_daddr_t, int));
     66  1.1  lukem static ufs_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs_daddr_t, int);
     67  1.1  lukem 
     68  1.1  lukem /* in ffs_tables.c */
     69  1.1  lukem extern const int inside[], around[];
     70  1.1  lukem extern const u_char * const fragtbl[];
     71  1.1  lukem 
     72  1.1  lukem /*
     73  1.1  lukem  * Allocate a block in the file system.
     74  1.1  lukem  *
     75  1.1  lukem  * The size of the requested block is given, which must be some
     76  1.1  lukem  * multiple of fs_fsize and <= fs_bsize.
     77  1.1  lukem  * A preference may be optionally specified. If a preference is given
     78  1.1  lukem  * the following hierarchy is used to allocate a block:
     79  1.1  lukem  *   1) allocate the requested block.
     80  1.1  lukem  *   2) allocate a rotationally optimal block in the same cylinder.
     81  1.1  lukem  *   3) allocate a block in the same cylinder group.
     82  1.1  lukem  *   4) quadradically rehash into other cylinder groups, until an
     83  1.1  lukem  *      available block is located.
     84  1.1  lukem  * If no block preference is given the following hierarchy is used
     85  1.1  lukem  * to allocate a block:
     86  1.1  lukem  *   1) allocate a block in the cylinder group that contains the
     87  1.1  lukem  *      inode for the file.
     88  1.1  lukem  *   2) quadradically rehash into other cylinder groups, until an
     89  1.1  lukem  *      available block is located.
     90  1.1  lukem  */
     91  1.1  lukem int
     92  1.1  lukem ffs_alloc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t bpref, int size,
     93  1.1  lukem     ufs_daddr_t *bnp)
     94  1.1  lukem {
     95  1.1  lukem 	struct fs *fs = ip->i_fs;
     96  1.1  lukem 	ufs_daddr_t bno;
     97  1.1  lukem 	int cg;
     98  1.1  lukem 
     99  1.1  lukem 	*bnp = 0;
    100  1.1  lukem 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
    101  1.1  lukem 		errx(1, "ffs_alloc: bad size: bsize %d size %d",
    102  1.1  lukem 		    fs->fs_bsize, size);
    103  1.1  lukem 	}
    104  1.1  lukem 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
    105  1.1  lukem 		goto nospace;
    106  1.1  lukem 	if (bpref >= fs->fs_size)
    107  1.1  lukem 		bpref = 0;
    108  1.1  lukem 	if (bpref == 0)
    109  1.1  lukem 		cg = ino_to_cg(fs, ip->i_number);
    110  1.1  lukem 	else
    111  1.1  lukem 		cg = dtog(fs, bpref);
    112  1.1  lukem 	bno = (ufs_daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
    113  1.1  lukem 	    			     ffs_alloccg);
    114  1.1  lukem 	if (bno > 0) {
    115  1.1  lukem 		ip->i_ffs_blocks += btodb(size);
    116  1.1  lukem 		*bnp = bno;
    117  1.1  lukem 		return (0);
    118  1.1  lukem 	}
    119  1.1  lukem nospace:
    120  1.1  lukem 	return (ENOSPC);
    121  1.1  lukem }
    122  1.1  lukem 
    123  1.1  lukem /*
    124  1.1  lukem  * Select the desired position for the next block in a file.  The file is
    125  1.1  lukem  * logically divided into sections. The first section is composed of the
    126  1.1  lukem  * direct blocks. Each additional section contains fs_maxbpg blocks.
    127  1.1  lukem  *
    128  1.1  lukem  * If no blocks have been allocated in the first section, the policy is to
    129  1.1  lukem  * request a block in the same cylinder group as the inode that describes
    130  1.1  lukem  * the file. If no blocks have been allocated in any other section, the
    131  1.1  lukem  * policy is to place the section in a cylinder group with a greater than
    132  1.1  lukem  * average number of free blocks.  An appropriate cylinder group is found
    133  1.1  lukem  * by using a rotor that sweeps the cylinder groups. When a new group of
    134  1.1  lukem  * blocks is needed, the sweep begins in the cylinder group following the
    135  1.1  lukem  * cylinder group from which the previous allocation was made. The sweep
    136  1.1  lukem  * continues until a cylinder group with greater than the average number
    137  1.1  lukem  * of free blocks is found. If the allocation is for the first block in an
    138  1.1  lukem  * indirect block, the information on the previous allocation is unavailable;
    139  1.1  lukem  * here a best guess is made based upon the logical block number being
    140  1.1  lukem  * allocated.
    141  1.1  lukem  *
    142  1.1  lukem  * If a section is already partially allocated, the policy is to
    143  1.1  lukem  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    144  1.1  lukem  * contiguous blocks and the beginning of the next is physically separated
    145  1.1  lukem  * so that the disk head will be in transit between them for at least
    146  1.1  lukem  * fs_rotdelay milliseconds.  This is to allow time for the processor to
    147  1.1  lukem  * schedule another I/O transfer.
    148  1.1  lukem  */
    149  1.1  lukem ufs_daddr_t
    150  1.1  lukem ffs_blkpref(struct inode *ip, ufs_daddr_t lbn, int indx, ufs_daddr_t *bap)
    151  1.1  lukem {
    152  1.1  lukem 	struct fs *fs;
    153  1.1  lukem 	int cg;
    154  1.1  lukem 	int avgbfree, startcg;
    155  1.1  lukem 	ufs_daddr_t nextblk;
    156  1.1  lukem 
    157  1.1  lukem 	fs = ip->i_fs;
    158  1.1  lukem 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    159  1.1  lukem 		if (lbn < NDADDR + NINDIR(fs)) {
    160  1.1  lukem 			cg = ino_to_cg(fs, ip->i_number);
    161  1.1  lukem 			return (fs->fs_fpg * cg + fs->fs_frag);
    162  1.1  lukem 		}
    163  1.1  lukem 		/*
    164  1.1  lukem 		 * Find a cylinder with greater than average number of
    165  1.1  lukem 		 * unused data blocks.
    166  1.1  lukem 		 */
    167  1.1  lukem 		if (indx == 0 || bap[indx - 1] == 0)
    168  1.1  lukem 			startcg =
    169  1.1  lukem 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    170  1.1  lukem 		else
    171  1.1  lukem 			startcg = dtog(fs,
    172  1.1  lukem 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    173  1.1  lukem 		startcg %= fs->fs_ncg;
    174  1.1  lukem 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    175  1.1  lukem 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    176  1.1  lukem 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    177  1.1  lukem 				fs->fs_cgrotor = cg;
    178  1.1  lukem 				return (fs->fs_fpg * cg + fs->fs_frag);
    179  1.1  lukem 			}
    180  1.1  lukem 		for (cg = 0; cg <= startcg; cg++)
    181  1.1  lukem 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    182  1.1  lukem 				fs->fs_cgrotor = cg;
    183  1.1  lukem 				return (fs->fs_fpg * cg + fs->fs_frag);
    184  1.1  lukem 			}
    185  1.1  lukem 		return (0);
    186  1.1  lukem 	}
    187  1.1  lukem 	/*
    188  1.1  lukem 	 * One or more previous blocks have been laid out. If less
    189  1.1  lukem 	 * than fs_maxcontig previous blocks are contiguous, the
    190  1.1  lukem 	 * next block is requested contiguously, otherwise it is
    191  1.1  lukem 	 * requested rotationally delayed by fs_rotdelay milliseconds.
    192  1.1  lukem 	 */
    193  1.1  lukem 	nextblk = ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    194  1.1  lukem 	if (indx < fs->fs_maxcontig ||
    195  1.1  lukem 		ufs_rw32(bap[indx - fs->fs_maxcontig], UFS_FSNEEDSWAP(fs)) +
    196  1.1  lukem 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
    197  1.1  lukem 		return (nextblk);
    198  1.1  lukem 	if (fs->fs_rotdelay != 0)
    199  1.1  lukem 		/*
    200  1.1  lukem 		 * Here we convert ms of delay to frags as:
    201  1.1  lukem 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
    202  1.1  lukem 		 *	((sect/frag) * (ms/sec))
    203  1.1  lukem 		 * then round up to the next block.
    204  1.1  lukem 		 */
    205  1.1  lukem 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
    206  1.1  lukem 		    (NSPF(fs) * 1000), fs->fs_frag);
    207  1.1  lukem 	return (nextblk);
    208  1.1  lukem }
    209  1.1  lukem 
    210  1.1  lukem /*
    211  1.1  lukem  * Implement the cylinder overflow algorithm.
    212  1.1  lukem  *
    213  1.1  lukem  * The policy implemented by this algorithm is:
    214  1.1  lukem  *   1) allocate the block in its requested cylinder group.
    215  1.1  lukem  *   2) quadradically rehash on the cylinder group number.
    216  1.1  lukem  *   3) brute force search for a free block.
    217  1.1  lukem  *
    218  1.1  lukem  * `size':	size for data blocks, mode for inodes
    219  1.1  lukem  */
    220  1.1  lukem /*VARARGS5*/
    221  1.1  lukem static u_long
    222  1.1  lukem ffs_hashalloc(struct inode *ip, int cg, long pref, int size,
    223  1.1  lukem     ufs_daddr_t (*allocator)(struct inode *, int, ufs_daddr_t, int))
    224  1.1  lukem {
    225  1.1  lukem 	struct fs *fs;
    226  1.1  lukem 	long result;
    227  1.1  lukem 	int i, icg = cg;
    228  1.1  lukem 
    229  1.1  lukem 	fs = ip->i_fs;
    230  1.1  lukem 	/*
    231  1.1  lukem 	 * 1: preferred cylinder group
    232  1.1  lukem 	 */
    233  1.1  lukem 	result = (*allocator)(ip, cg, pref, size);
    234  1.1  lukem 	if (result)
    235  1.1  lukem 		return (result);
    236  1.1  lukem 	/*
    237  1.1  lukem 	 * 2: quadratic rehash
    238  1.1  lukem 	 */
    239  1.1  lukem 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    240  1.1  lukem 		cg += i;
    241  1.1  lukem 		if (cg >= fs->fs_ncg)
    242  1.1  lukem 			cg -= fs->fs_ncg;
    243  1.1  lukem 		result = (*allocator)(ip, cg, 0, size);
    244  1.1  lukem 		if (result)
    245  1.1  lukem 			return (result);
    246  1.1  lukem 	}
    247  1.1  lukem 	/*
    248  1.1  lukem 	 * 3: brute force search
    249  1.1  lukem 	 * Note that we start at i == 2, since 0 was checked initially,
    250  1.1  lukem 	 * and 1 is always checked in the quadratic rehash.
    251  1.1  lukem 	 */
    252  1.1  lukem 	cg = (icg + 2) % fs->fs_ncg;
    253  1.1  lukem 	for (i = 2; i < fs->fs_ncg; i++) {
    254  1.1  lukem 		result = (*allocator)(ip, cg, 0, size);
    255  1.1  lukem 		if (result)
    256  1.1  lukem 			return (result);
    257  1.1  lukem 		cg++;
    258  1.1  lukem 		if (cg == fs->fs_ncg)
    259  1.1  lukem 			cg = 0;
    260  1.1  lukem 	}
    261  1.1  lukem 	return (0);
    262  1.1  lukem }
    263  1.1  lukem 
    264  1.1  lukem /*
    265  1.1  lukem  * Determine whether a block can be allocated.
    266  1.1  lukem  *
    267  1.1  lukem  * Check to see if a block of the appropriate size is available,
    268  1.1  lukem  * and if it is, allocate it.
    269  1.1  lukem  */
    270  1.1  lukem static ufs_daddr_t
    271  1.1  lukem ffs_alloccg(struct inode *ip, int cg, ufs_daddr_t bpref, int size)
    272  1.1  lukem {
    273  1.1  lukem 	struct cg *cgp;
    274  1.1  lukem 	struct buf *bp;
    275  1.1  lukem 	ufs_daddr_t bno, blkno;
    276  1.1  lukem 	int error, frags, allocsiz, i;
    277  1.1  lukem 	struct fs *fs = ip->i_fs;
    278  1.1  lukem 	const int needswap = UFS_FSNEEDSWAP(fs);
    279  1.1  lukem 
    280  1.1  lukem 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
    281  1.1  lukem 		return (0);
    282  1.1  lukem 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    283  1.1  lukem 		(int)fs->fs_cgsize, &bp);
    284  1.1  lukem 	if (error) {
    285  1.1  lukem 		brelse(bp);
    286  1.1  lukem 		return (0);
    287  1.1  lukem 	}
    288  1.1  lukem 	cgp = (struct cg *)bp->b_data;
    289  1.1  lukem 	if (!cg_chkmagic(cgp, needswap) ||
    290  1.1  lukem 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
    291  1.1  lukem 		brelse(bp);
    292  1.1  lukem 		return (0);
    293  1.1  lukem 	}
    294  1.1  lukem 	if (size == fs->fs_bsize) {
    295  1.1  lukem 		bno = ffs_alloccgblk(ip, bp, bpref);
    296  1.1  lukem 		bdwrite(bp);
    297  1.1  lukem 		return (bno);
    298  1.1  lukem 	}
    299  1.1  lukem 	/*
    300  1.1  lukem 	 * check to see if any fragments are already available
    301  1.1  lukem 	 * allocsiz is the size which will be allocated, hacking
    302  1.1  lukem 	 * it down to a smaller size if necessary
    303  1.1  lukem 	 */
    304  1.1  lukem 	frags = numfrags(fs, size);
    305  1.1  lukem 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
    306  1.1  lukem 		if (cgp->cg_frsum[allocsiz] != 0)
    307  1.1  lukem 			break;
    308  1.1  lukem 	if (allocsiz == fs->fs_frag) {
    309  1.1  lukem 		/*
    310  1.1  lukem 		 * no fragments were available, so a block will be
    311  1.1  lukem 		 * allocated, and hacked up
    312  1.1  lukem 		 */
    313  1.1  lukem 		if (cgp->cg_cs.cs_nbfree == 0) {
    314  1.1  lukem 			brelse(bp);
    315  1.1  lukem 			return (0);
    316  1.1  lukem 		}
    317  1.1  lukem 		bno = ffs_alloccgblk(ip, bp, bpref);
    318  1.1  lukem 		bpref = dtogd(fs, bno);
    319  1.1  lukem 		for (i = frags; i < fs->fs_frag; i++)
    320  1.1  lukem 			setbit(cg_blksfree(cgp, needswap), bpref + i);
    321  1.1  lukem 		i = fs->fs_frag - frags;
    322  1.1  lukem 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    323  1.1  lukem 		fs->fs_cstotal.cs_nffree += i;
    324  1.1  lukem 		fs->fs_cs(fs, cg).cs_nffree += i;
    325  1.1  lukem 		fs->fs_fmod = 1;
    326  1.1  lukem 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
    327  1.1  lukem 		bdwrite(bp);
    328  1.1  lukem 		return (bno);
    329  1.1  lukem 	}
    330  1.1  lukem 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
    331  1.1  lukem 	for (i = 0; i < frags; i++)
    332  1.1  lukem 		clrbit(cg_blksfree(cgp, needswap), bno + i);
    333  1.1  lukem 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
    334  1.1  lukem 	fs->fs_cstotal.cs_nffree -= frags;
    335  1.1  lukem 	fs->fs_cs(fs, cg).cs_nffree -= frags;
    336  1.1  lukem 	fs->fs_fmod = 1;
    337  1.1  lukem 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
    338  1.1  lukem 	if (frags != allocsiz)
    339  1.1  lukem 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
    340  1.1  lukem 	blkno = cg * fs->fs_fpg + bno;
    341  1.1  lukem 	bdwrite(bp);
    342  1.1  lukem 	return blkno;
    343  1.1  lukem }
    344  1.1  lukem 
    345  1.1  lukem /*
    346  1.1  lukem  * Allocate a block in a cylinder group.
    347  1.1  lukem  *
    348  1.1  lukem  * This algorithm implements the following policy:
    349  1.1  lukem  *   1) allocate the requested block.
    350  1.1  lukem  *   2) allocate a rotationally optimal block in the same cylinder.
    351  1.1  lukem  *   3) allocate the next available block on the block rotor for the
    352  1.1  lukem  *      specified cylinder group.
    353  1.1  lukem  * Note that this routine only allocates fs_bsize blocks; these
    354  1.1  lukem  * blocks may be fragmented by the routine that allocates them.
    355  1.1  lukem  */
    356  1.1  lukem static ufs_daddr_t
    357  1.1  lukem ffs_alloccgblk(struct inode *ip, struct buf *bp, ufs_daddr_t bpref)
    358  1.1  lukem {
    359  1.1  lukem 	struct cg *cgp;
    360  1.1  lukem 	ufs_daddr_t bno, blkno;
    361  1.1  lukem 	int cylno, pos, delta;
    362  1.1  lukem 	short *cylbp;
    363  1.1  lukem 	int i;
    364  1.1  lukem 	struct fs *fs = ip->i_fs;
    365  1.1  lukem 	const int needswap = UFS_FSNEEDSWAP(fs);
    366  1.1  lukem 
    367  1.1  lukem 	cgp = (struct cg *)bp->b_data;
    368  1.1  lukem 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
    369  1.1  lukem 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
    370  1.1  lukem 		goto norot;
    371  1.1  lukem 	}
    372  1.1  lukem 	bpref = blknum(fs, bpref);
    373  1.1  lukem 	bpref = dtogd(fs, bpref);
    374  1.1  lukem 	/*
    375  1.1  lukem 	 * if the requested block is available, use it
    376  1.1  lukem 	 */
    377  1.1  lukem 	if (ffs_isblock(fs, cg_blksfree(cgp, needswap),
    378  1.1  lukem 		fragstoblks(fs, bpref))) {
    379  1.1  lukem 		bno = bpref;
    380  1.1  lukem 		goto gotit;
    381  1.1  lukem 	}
    382  1.1  lukem 	if (fs->fs_nrpos <= 1 || fs->fs_cpc == 0) {
    383  1.1  lukem 		/*
    384  1.1  lukem 		 * Block layout information is not available.
    385  1.1  lukem 		 * Leaving bpref unchanged means we take the
    386  1.1  lukem 		 * next available free block following the one
    387  1.1  lukem 		 * we just allocated. Hopefully this will at
    388  1.1  lukem 		 * least hit a track cache on drives of unknown
    389  1.1  lukem 		 * geometry (e.g. SCSI).
    390  1.1  lukem 		 */
    391  1.1  lukem 		goto norot;
    392  1.1  lukem 	}
    393  1.1  lukem 	/*
    394  1.1  lukem 	 * check for a block available on the same cylinder
    395  1.1  lukem 	 */
    396  1.1  lukem 	cylno = cbtocylno(fs, bpref);
    397  1.1  lukem 	if (cg_blktot(cgp, needswap)[cylno] == 0)
    398  1.1  lukem 		goto norot;
    399  1.1  lukem 	/*
    400  1.1  lukem 	 * check the summary information to see if a block is
    401  1.1  lukem 	 * available in the requested cylinder starting at the
    402  1.1  lukem 	 * requested rotational position and proceeding around.
    403  1.1  lukem 	 */
    404  1.1  lukem 	cylbp = cg_blks(fs, cgp, cylno, needswap);
    405  1.1  lukem 	pos = cbtorpos(fs, bpref);
    406  1.1  lukem 	for (i = pos; i < fs->fs_nrpos; i++)
    407  1.1  lukem 		if (ufs_rw16(cylbp[i], needswap) > 0)
    408  1.1  lukem 			break;
    409  1.1  lukem 	if (i == fs->fs_nrpos)
    410  1.1  lukem 		for (i = 0; i < pos; i++)
    411  1.1  lukem 			if (ufs_rw16(cylbp[i], needswap) > 0)
    412  1.1  lukem 				break;
    413  1.1  lukem 	if (ufs_rw16(cylbp[i], needswap) > 0) {
    414  1.1  lukem 		/*
    415  1.1  lukem 		 * found a rotational position, now find the actual
    416  1.1  lukem 		 * block. A panic if none is actually there.
    417  1.1  lukem 		 */
    418  1.1  lukem 		pos = cylno % fs->fs_cpc;
    419  1.1  lukem 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
    420  1.1  lukem 		if (fs_postbl(fs, pos)[i] == -1) {
    421  1.1  lukem 			errx(1,
    422  1.1  lukem 			    "ffs_alloccgblk: cyl groups corrupted: pos %d i %d",
    423  1.1  lukem 			    pos, i);
    424  1.1  lukem 		}
    425  1.1  lukem 		for (i = fs_postbl(fs, pos)[i];; ) {
    426  1.1  lukem 			if (ffs_isblock(fs, cg_blksfree(cgp, needswap), bno + i)) {
    427  1.1  lukem 				bno = blkstofrags(fs, (bno + i));
    428  1.1  lukem 				goto gotit;
    429  1.1  lukem 			}
    430  1.1  lukem 			delta = fs_rotbl(fs)[i];
    431  1.1  lukem 			if (delta <= 0 ||
    432  1.1  lukem 			    delta + i > fragstoblks(fs, fs->fs_fpg))
    433  1.1  lukem 				break;
    434  1.1  lukem 			i += delta;
    435  1.1  lukem 		}
    436  1.1  lukem 		errx(1, "ffs_alloccgblk: can't find blk in cyl: pos %d i %d",
    437  1.1  lukem 		    pos, i);
    438  1.1  lukem 	}
    439  1.1  lukem norot:
    440  1.1  lukem 	/*
    441  1.1  lukem 	 * no blocks in the requested cylinder, so take next
    442  1.1  lukem 	 * available one in this cylinder group.
    443  1.1  lukem 	 */
    444  1.1  lukem 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
    445  1.1  lukem 	if (bno < 0)
    446  1.1  lukem 		return (0);
    447  1.1  lukem 	cgp->cg_rotor = ufs_rw32(bno, needswap);
    448  1.1  lukem gotit:
    449  1.1  lukem 	blkno = fragstoblks(fs, bno);
    450  1.1  lukem 	ffs_clrblock(fs, cg_blksfree(cgp, needswap), (long)blkno);
    451  1.1  lukem 	ffs_clusteracct(fs, cgp, blkno, -1);
    452  1.1  lukem 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
    453  1.1  lukem 	fs->fs_cstotal.cs_nbfree--;
    454  1.1  lukem 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
    455  1.1  lukem 	cylno = cbtocylno(fs, bno);
    456  1.1  lukem 	ufs_add16(cg_blks(fs, cgp, cylno, needswap)[cbtorpos(fs, bno)], -1,
    457  1.1  lukem 	    needswap);
    458  1.1  lukem 	ufs_add32(cg_blktot(cgp, needswap)[cylno], -1, needswap);
    459  1.1  lukem 	fs->fs_fmod = 1;
    460  1.1  lukem 	blkno = ufs_rw32(cgp->cg_cgx, needswap) * fs->fs_fpg + bno;
    461  1.1  lukem 	return (blkno);
    462  1.1  lukem }
    463  1.1  lukem 
    464  1.1  lukem /*
    465  1.1  lukem  * Free a block or fragment.
    466  1.1  lukem  *
    467  1.1  lukem  * The specified block or fragment is placed back in the
    468  1.1  lukem  * free map. If a fragment is deallocated, a possible
    469  1.1  lukem  * block reassembly is checked.
    470  1.1  lukem  */
    471  1.1  lukem void
    472  1.1  lukem ffs_blkfree(struct inode *ip, ufs_daddr_t bno, long size)
    473  1.1  lukem {
    474  1.1  lukem 	struct cg *cgp;
    475  1.1  lukem 	struct buf *bp;
    476  1.1  lukem 	ufs_daddr_t blkno;
    477  1.1  lukem 	int i, error, cg, blk, frags, bbase;
    478  1.1  lukem 	struct fs *fs = ip->i_fs;
    479  1.1  lukem 	const int needswap = UFS_FSNEEDSWAP(fs);
    480  1.1  lukem 
    481  1.1  lukem 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
    482  1.1  lukem 	    fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
    483  1.1  lukem 		errx(1, "blkfree: bad size: bno %u bsize %d size %ld",
    484  1.1  lukem 		    bno, fs->fs_bsize, size);
    485  1.1  lukem 	}
    486  1.1  lukem 	cg = dtog(fs, bno);
    487  1.1  lukem 	if ((u_int)bno >= fs->fs_size) {
    488  1.1  lukem 		warnx("bad block %d, ino %d\n", bno, ip->i_number);
    489  1.1  lukem 		return;
    490  1.1  lukem 	}
    491  1.1  lukem 	error = bread(ip->i_fd, ip->i_fs, fsbtodb(fs, cgtod(fs, cg)),
    492  1.1  lukem 		(int)fs->fs_cgsize, &bp);
    493  1.1  lukem 	if (error) {
    494  1.1  lukem 		brelse(bp);
    495  1.1  lukem 		return;
    496  1.1  lukem 	}
    497  1.1  lukem 	cgp = (struct cg *)bp->b_data;
    498  1.1  lukem 	if (!cg_chkmagic(cgp, needswap)) {
    499  1.1  lukem 		brelse(bp);
    500  1.1  lukem 		return;
    501  1.1  lukem 	}
    502  1.1  lukem 	bno = dtogd(fs, bno);
    503  1.1  lukem 	if (size == fs->fs_bsize) {
    504  1.1  lukem 		blkno = fragstoblks(fs, bno);
    505  1.1  lukem 		if (!ffs_isfreeblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    506  1.1  lukem 			errx(1, "blkfree: freeing free block %d", bno);
    507  1.1  lukem 		}
    508  1.1  lukem 		ffs_setblock(fs, cg_blksfree(cgp, needswap), blkno);
    509  1.1  lukem 		ffs_clusteracct(fs, cgp, blkno, 1);
    510  1.1  lukem 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    511  1.1  lukem 		fs->fs_cstotal.cs_nbfree++;
    512  1.1  lukem 		fs->fs_cs(fs, cg).cs_nbfree++;
    513  1.1  lukem 		i = cbtocylno(fs, bno);
    514  1.1  lukem 		ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs, bno)], 1,
    515  1.1  lukem 		    needswap);
    516  1.1  lukem 		ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    517  1.1  lukem 	} else {
    518  1.1  lukem 		bbase = bno - fragnum(fs, bno);
    519  1.1  lukem 		/*
    520  1.1  lukem 		 * decrement the counts associated with the old frags
    521  1.1  lukem 		 */
    522  1.1  lukem 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    523  1.1  lukem 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
    524  1.1  lukem 		/*
    525  1.1  lukem 		 * deallocate the fragment
    526  1.1  lukem 		 */
    527  1.1  lukem 		frags = numfrags(fs, size);
    528  1.1  lukem 		for (i = 0; i < frags; i++) {
    529  1.1  lukem 			if (isset(cg_blksfree(cgp, needswap), bno + i)) {
    530  1.1  lukem 				errx(1, "blkfree: freeing free frag: block %d",
    531  1.1  lukem 				    bno + i);
    532  1.1  lukem 			}
    533  1.1  lukem 			setbit(cg_blksfree(cgp, needswap), bno + i);
    534  1.1  lukem 		}
    535  1.1  lukem 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
    536  1.1  lukem 		fs->fs_cstotal.cs_nffree += i;
    537  1.1  lukem 		fs->fs_cs(fs, cg).cs_nffree += i;
    538  1.1  lukem 		/*
    539  1.1  lukem 		 * add back in counts associated with the new frags
    540  1.1  lukem 		 */
    541  1.1  lukem 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bbase);
    542  1.1  lukem 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
    543  1.1  lukem 		/*
    544  1.1  lukem 		 * if a complete block has been reassembled, account for it
    545  1.1  lukem 		 */
    546  1.1  lukem 		blkno = fragstoblks(fs, bbase);
    547  1.1  lukem 		if (ffs_isblock(fs, cg_blksfree(cgp, needswap), blkno)) {
    548  1.1  lukem 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
    549  1.1  lukem 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
    550  1.1  lukem 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
    551  1.1  lukem 			ffs_clusteracct(fs, cgp, blkno, 1);
    552  1.1  lukem 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
    553  1.1  lukem 			fs->fs_cstotal.cs_nbfree++;
    554  1.1  lukem 			fs->fs_cs(fs, cg).cs_nbfree++;
    555  1.1  lukem 			i = cbtocylno(fs, bbase);
    556  1.1  lukem 			ufs_add16(cg_blks(fs, cgp, i, needswap)[cbtorpos(fs,
    557  1.1  lukem 								bbase)], 1,
    558  1.1  lukem 			    needswap);
    559  1.1  lukem 			ufs_add32(cg_blktot(cgp, needswap)[i], 1, needswap);
    560  1.1  lukem 		}
    561  1.1  lukem 	}
    562  1.1  lukem 	fs->fs_fmod = 1;
    563  1.1  lukem 	bdwrite(bp);
    564  1.1  lukem }
    565  1.1  lukem 
    566  1.1  lukem 
    567  1.1  lukem static int
    568  1.1  lukem scanc(u_int size, const u_char *cp, const u_char table[], int mask)
    569  1.1  lukem {
    570  1.1  lukem 	const u_char *end = &cp[size];
    571  1.1  lukem 
    572  1.1  lukem 	while (cp < end && (table[*cp] & mask) == 0)
    573  1.1  lukem 		cp++;
    574  1.1  lukem 	return (end - cp);
    575  1.1  lukem }
    576  1.1  lukem 
    577  1.1  lukem /*
    578  1.1  lukem  * Find a block of the specified size in the specified cylinder group.
    579  1.1  lukem  *
    580  1.1  lukem  * It is a panic if a request is made to find a block if none are
    581  1.1  lukem  * available.
    582  1.1  lukem  */
    583  1.1  lukem static ufs_daddr_t
    584  1.1  lukem ffs_mapsearch(struct fs *fs, struct cg *cgp, ufs_daddr_t bpref, int allocsiz)
    585  1.1  lukem {
    586  1.1  lukem 	ufs_daddr_t bno;
    587  1.1  lukem 	int start, len, loc, i;
    588  1.1  lukem 	int blk, field, subfield, pos;
    589  1.1  lukem 	int ostart, olen;
    590  1.1  lukem 	const int needswap = UFS_FSNEEDSWAP(fs);
    591  1.1  lukem 
    592  1.1  lukem 	/*
    593  1.1  lukem 	 * find the fragment by searching through the free block
    594  1.1  lukem 	 * map for an appropriate bit pattern
    595  1.1  lukem 	 */
    596  1.1  lukem 	if (bpref)
    597  1.1  lukem 		start = dtogd(fs, bpref) / NBBY;
    598  1.1  lukem 	else
    599  1.1  lukem 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
    600  1.1  lukem 	len = howmany(fs->fs_fpg, NBBY) - start;
    601  1.1  lukem 	ostart = start;
    602  1.1  lukem 	olen = len;
    603  1.1  lukem 	loc = scanc((u_int)len,
    604  1.1  lukem 		(const u_char *)&cg_blksfree(cgp, needswap)[start],
    605  1.1  lukem 		(const u_char *)fragtbl[fs->fs_frag],
    606  1.1  lukem 		(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    607  1.1  lukem 	if (loc == 0) {
    608  1.1  lukem 		len = start + 1;
    609  1.1  lukem 		start = 0;
    610  1.1  lukem 		loc = scanc((u_int)len,
    611  1.1  lukem 			(const u_char *)&cg_blksfree(cgp, needswap)[0],
    612  1.1  lukem 			(const u_char *)fragtbl[fs->fs_frag],
    613  1.1  lukem 			(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
    614  1.1  lukem 		if (loc == 0) {
    615  1.1  lukem 			errx(1,
    616  1.1  lukem     "ffs_alloccg: map corrupted: start %d len %d offset %d %ld",
    617  1.1  lukem 				ostart, olen,
    618  1.1  lukem 				ufs_rw32(cgp->cg_freeoff, needswap),
    619  1.1  lukem 				(long)cg_blksfree(cgp, needswap) - (long)cgp);
    620  1.1  lukem 			/* NOTREACHED */
    621  1.1  lukem 		}
    622  1.1  lukem 	}
    623  1.1  lukem 	bno = (start + len - loc) * NBBY;
    624  1.1  lukem 	cgp->cg_frotor = ufs_rw32(bno, needswap);
    625  1.1  lukem 	/*
    626  1.1  lukem 	 * found the byte in the map
    627  1.1  lukem 	 * sift through the bits to find the selected frag
    628  1.1  lukem 	 */
    629  1.1  lukem 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
    630  1.1  lukem 		blk = blkmap(fs, cg_blksfree(cgp, needswap), bno);
    631  1.1  lukem 		blk <<= 1;
    632  1.1  lukem 		field = around[allocsiz];
    633  1.1  lukem 		subfield = inside[allocsiz];
    634  1.1  lukem 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
    635  1.1  lukem 			if ((blk & field) == subfield)
    636  1.1  lukem 				return (bno + pos);
    637  1.1  lukem 			field <<= 1;
    638  1.1  lukem 			subfield <<= 1;
    639  1.1  lukem 		}
    640  1.1  lukem 	}
    641  1.1  lukem 	errx(1, "ffs_alloccg: block not in map: bno %d", bno);
    642  1.1  lukem 	return (-1);
    643  1.1  lukem }
    644  1.1  lukem 
    645  1.1  lukem /*
    646  1.1  lukem  * Update the cluster map because of an allocation or free.
    647  1.1  lukem  *
    648  1.1  lukem  * Cnt == 1 means free; cnt == -1 means allocating.
    649  1.1  lukem  */
    650  1.1  lukem void
    651  1.1  lukem ffs_clusteracct(struct fs *fs, struct cg *cgp, ufs_daddr_t blkno, int cnt)
    652  1.1  lukem {
    653  1.1  lukem 	int32_t *sump;
    654  1.1  lukem 	int32_t *lp;
    655  1.1  lukem 	u_char *freemapp, *mapp;
    656  1.1  lukem 	int i, start, end, forw, back, map, bit;
    657  1.1  lukem 	const int needswap = UFS_FSNEEDSWAP(fs);
    658  1.1  lukem 
    659  1.1  lukem 	if (fs->fs_contigsumsize <= 0)
    660  1.1  lukem 		return;
    661  1.1  lukem 	freemapp = cg_clustersfree(cgp, needswap);
    662  1.1  lukem 	sump = cg_clustersum(cgp, needswap);
    663  1.1  lukem 	/*
    664  1.1  lukem 	 * Allocate or clear the actual block.
    665  1.1  lukem 	 */
    666  1.1  lukem 	if (cnt > 0)
    667  1.1  lukem 		setbit(freemapp, blkno);
    668  1.1  lukem 	else
    669  1.1  lukem 		clrbit(freemapp, blkno);
    670  1.1  lukem 	/*
    671  1.1  lukem 	 * Find the size of the cluster going forward.
    672  1.1  lukem 	 */
    673  1.1  lukem 	start = blkno + 1;
    674  1.1  lukem 	end = start + fs->fs_contigsumsize;
    675  1.1  lukem 	if (end >= ufs_rw32(cgp->cg_nclusterblks, needswap))
    676  1.1  lukem 		end = ufs_rw32(cgp->cg_nclusterblks, needswap);
    677  1.1  lukem 	mapp = &freemapp[start / NBBY];
    678  1.1  lukem 	map = *mapp++;
    679  1.1  lukem 	bit = 1 << (start % NBBY);
    680  1.1  lukem 	for (i = start; i < end; i++) {
    681  1.1  lukem 		if ((map & bit) == 0)
    682  1.1  lukem 			break;
    683  1.1  lukem 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
    684  1.1  lukem 			bit <<= 1;
    685  1.1  lukem 		} else {
    686  1.1  lukem 			map = *mapp++;
    687  1.1  lukem 			bit = 1;
    688  1.1  lukem 		}
    689  1.1  lukem 	}
    690  1.1  lukem 	forw = i - start;
    691  1.1  lukem 	/*
    692  1.1  lukem 	 * Find the size of the cluster going backward.
    693  1.1  lukem 	 */
    694  1.1  lukem 	start = blkno - 1;
    695  1.1  lukem 	end = start - fs->fs_contigsumsize;
    696  1.1  lukem 	if (end < 0)
    697  1.1  lukem 		end = -1;
    698  1.1  lukem 	mapp = &freemapp[start / NBBY];
    699  1.1  lukem 	map = *mapp--;
    700  1.1  lukem 	bit = 1 << (start % NBBY);
    701  1.1  lukem 	for (i = start; i > end; i--) {
    702  1.1  lukem 		if ((map & bit) == 0)
    703  1.1  lukem 			break;
    704  1.1  lukem 		if ((i & (NBBY - 1)) != 0) {
    705  1.1  lukem 			bit >>= 1;
    706  1.1  lukem 		} else {
    707  1.1  lukem 			map = *mapp--;
    708  1.1  lukem 			bit = 1 << (NBBY - 1);
    709  1.1  lukem 		}
    710  1.1  lukem 	}
    711  1.1  lukem 	back = start - i;
    712  1.1  lukem 	/*
    713  1.1  lukem 	 * Account for old cluster and the possibly new forward and
    714  1.1  lukem 	 * back clusters.
    715  1.1  lukem 	 */
    716  1.1  lukem 	i = back + forw + 1;
    717  1.1  lukem 	if (i > fs->fs_contigsumsize)
    718  1.1  lukem 		i = fs->fs_contigsumsize;
    719  1.1  lukem 	ufs_add32(sump[i], cnt, needswap);
    720  1.1  lukem 	if (back > 0)
    721  1.1  lukem 		ufs_add32(sump[back], -cnt, needswap);
    722  1.1  lukem 	if (forw > 0)
    723  1.1  lukem 		ufs_add32(sump[forw], -cnt, needswap);
    724  1.1  lukem 
    725  1.1  lukem 	/*
    726  1.1  lukem 	 * Update cluster summary information.
    727  1.1  lukem 	 */
    728  1.1  lukem 	lp = &sump[fs->fs_contigsumsize];
    729  1.1  lukem 	for (i = fs->fs_contigsumsize; i > 0; i--)
    730  1.1  lukem 		if (ufs_rw32(*lp--, needswap) > 0)
    731  1.1  lukem 			break;
    732  1.1  lukem 	fs->fs_maxcluster[ufs_rw32(cgp->cg_cgx, needswap)] = i;
    733  1.1  lukem }
    734