xref: /csrg-svn/sys/ufs/ffs/ffs_alloc.c (revision 67869)
123394Smckusick /*
263371Sbostic  * Copyright (c) 1982, 1986, 1989, 1993
363371Sbostic  *	The Regents of the University of California.  All rights reserved.
423394Smckusick  *
544536Sbostic  * %sccs.include.redist.c%
634432Sbostic  *
7*67869Smckusick  *	@(#)ffs_alloc.c	8.10 (Berkeley) 10/27/94
823394Smckusick  */
94359Smckusick 
1051469Sbostic #include <sys/param.h>
1151469Sbostic #include <sys/systm.h>
1251469Sbostic #include <sys/buf.h>
1351469Sbostic #include <sys/proc.h>
1451469Sbostic #include <sys/vnode.h>
1554653Smckusick #include <sys/mount.h>
1651469Sbostic #include <sys/kernel.h>
1751469Sbostic #include <sys/syslog.h>
184359Smckusick 
1953317Smckusick #include <vm/vm.h>
2053317Smckusick 
2151469Sbostic #include <ufs/ufs/quota.h>
2251469Sbostic #include <ufs/ufs/inode.h>
2347571Skarels 
2451469Sbostic #include <ufs/ffs/fs.h>
2551469Sbostic #include <ufs/ffs/ffs_extern.h>
264359Smckusick 
2751469Sbostic extern u_long nextgennumber;
2851469Sbostic 
2951469Sbostic static daddr_t	ffs_alloccg __P((struct inode *, int, daddr_t, int));
3051469Sbostic static daddr_t	ffs_alloccgblk __P((struct fs *, struct cg *, daddr_t));
3165999Smckusick static daddr_t	ffs_clusteralloc __P((struct inode *, int, daddr_t, int));
3251469Sbostic static ino_t	ffs_dirpref __P((struct fs *));
3351469Sbostic static daddr_t	ffs_fragextend __P((struct inode *, int, long, int, int));
3451469Sbostic static void	ffs_fserr __P((struct fs *, u_int, char *));
3551469Sbostic static u_long	ffs_hashalloc
3651469Sbostic 		    __P((struct inode *, int, long, int, u_long (*)()));
37*67869Smckusick static u_long	ffs_nodealloccg __P((struct inode *, int, daddr_t, int));
3851469Sbostic static daddr_t	ffs_mapsearch __P((struct fs *, struct cg *, daddr_t, int));
3951469Sbostic 
405375Smckusic /*
415375Smckusic  * Allocate a block in the file system.
425375Smckusic  *
435375Smckusic  * The size of the requested block is given, which must be some
445375Smckusic  * multiple of fs_fsize and <= fs_bsize.
455375Smckusic  * A preference may be optionally specified. If a preference is given
465375Smckusic  * the following hierarchy is used to allocate a block:
475375Smckusic  *   1) allocate the requested block.
485375Smckusic  *   2) allocate a rotationally optimal block in the same cylinder.
495375Smckusic  *   3) allocate a block in the same cylinder group.
505375Smckusic  *   4) quadradically rehash into other cylinder groups, until an
515375Smckusic  *      available block is located.
525375Smckusic  * If no block preference is given the following heirarchy is used
535375Smckusic  * to allocate a block:
545375Smckusic  *   1) allocate a block in the cylinder group that contains the
555375Smckusic  *      inode for the file.
565375Smckusic  *   2) quadradically rehash into other cylinder groups, until an
575375Smckusic  *      available block is located.
585375Smckusic  */
5953244Smckusick ffs_alloc(ip, lbn, bpref, size, cred, bnp)
604463Smckusic 	register struct inode *ip;
6139678Smckusick 	daddr_t lbn, bpref;
624359Smckusick 	int size;
6353244Smckusick 	struct ucred *cred;
6439678Smckusick 	daddr_t *bnp;
654359Smckusick {
6665468Sbostic 	register struct fs *fs;
674359Smckusick 	daddr_t bno;
6837735Smckusick 	int cg, error;
694359Smckusick 
7039678Smckusick 	*bnp = 0;
715965Smckusic 	fs = ip->i_fs;
7253244Smckusick #ifdef DIAGNOSTIC
7364508Sbostic 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
746716Smckusick 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
756716Smckusick 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
7651469Sbostic 		panic("ffs_alloc: bad size");
776716Smckusick 	}
7853244Smckusick 	if (cred == NOCRED)
7953244Smckusick 		panic("ffs_alloc: missing credential\n");
8053244Smckusick #endif /* DIAGNOSTIC */
815322Smckusic 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
824359Smckusick 		goto nospace;
8341309Smckusick 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
844792Smckusic 		goto nospace;
857650Ssam #ifdef QUOTA
8641309Smckusick 	if (error = chkdq(ip, (long)btodb(size), cred, 0))
8737735Smckusick 		return (error);
887483Skre #endif
894948Smckusic 	if (bpref >= fs->fs_size)
904948Smckusic 		bpref = 0;
914359Smckusick 	if (bpref == 0)
9264603Sbostic 		cg = ino_to_cg(fs, ip->i_number);
934359Smckusick 	else
945377Smckusic 		cg = dtog(fs, bpref);
9551469Sbostic 	bno = (daddr_t)ffs_hashalloc(ip, cg, (long)bpref, size,
9651469Sbostic 	    (u_long (*)())ffs_alloccg);
9739678Smckusick 	if (bno > 0) {
9839678Smckusick 		ip->i_blocks += btodb(size);
9964603Sbostic 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
10039678Smckusick 		*bnp = bno;
10139678Smckusick 		return (0);
10239678Smckusick 	}
10345173Smckusick #ifdef QUOTA
10445173Smckusick 	/*
10545173Smckusick 	 * Restore user's disk quota because allocation failed.
10645173Smckusick 	 */
10745173Smckusick 	(void) chkdq(ip, (long)-btodb(size), cred, FORCE);
10845173Smckusick #endif
1094359Smckusick nospace:
11051469Sbostic 	ffs_fserr(fs, cred->cr_uid, "file system full");
1114359Smckusick 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
11237735Smckusick 	return (ENOSPC);
1134359Smckusick }
1144359Smckusick 
1155375Smckusic /*
1165375Smckusic  * Reallocate a fragment to a bigger size
1175375Smckusic  *
1185375Smckusic  * The number and size of the old block is given, and a preference
1195375Smckusic  * and new size is also specified. The allocator attempts to extend
1205375Smckusic  * the original block. Failing that, the regular block allocator is
1215375Smckusic  * invoked to get an appropriate block.
1225375Smckusic  */
12353244Smckusick ffs_realloccg(ip, lbprev, bpref, osize, nsize, cred, bpp)
1245965Smckusic 	register struct inode *ip;
12553059Smckusick 	daddr_t lbprev;
12639678Smckusick 	daddr_t bpref;
1274426Smckusic 	int osize, nsize;
12853244Smckusick 	struct ucred *cred;
12937735Smckusick 	struct buf **bpp;
1304426Smckusic {
1314426Smckusic 	register struct fs *fs;
13265468Sbostic 	struct buf *bp;
13345719Smckusick 	int cg, request, error;
13445719Smckusick 	daddr_t bprev, bno;
1354426Smckusic 
13637735Smckusick 	*bpp = 0;
1375965Smckusic 	fs = ip->i_fs;
13853244Smckusick #ifdef DIAGNOSTIC
13964508Sbostic 	if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
14064508Sbostic 	    (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
14151469Sbostic 		printf(
14251469Sbostic 		    "dev = 0x%x, bsize = %d, osize = %d, nsize = %d, fs = %s\n",
1436716Smckusick 		    ip->i_dev, fs->fs_bsize, osize, nsize, fs->fs_fsmnt);
14451469Sbostic 		panic("ffs_realloccg: bad size");
1456716Smckusick 	}
14653244Smckusick 	if (cred == NOCRED)
14753244Smckusick 		panic("ffs_realloccg: missing credential\n");
14853244Smckusick #endif /* DIAGNOSTIC */
14941309Smckusick 	if (cred->cr_uid != 0 && freespace(fs, fs->fs_minfree) <= 0)
1504792Smckusic 		goto nospace;
15139678Smckusick 	if ((bprev = ip->i_db[lbprev]) == 0) {
1526716Smckusick 		printf("dev = 0x%x, bsize = %d, bprev = %d, fs = %s\n",
1536716Smckusick 		    ip->i_dev, fs->fs_bsize, bprev, fs->fs_fsmnt);
15451469Sbostic 		panic("ffs_realloccg: bad bprev");
1556716Smckusick 	}
15639678Smckusick 	/*
15739678Smckusick 	 * Allocate the extra space in the buffer.
15839678Smckusick 	 */
15939678Smckusick 	if (error = bread(ITOV(ip), lbprev, osize, NOCRED, &bp)) {
16039678Smckusick 		brelse(bp);
16139678Smckusick 		return (error);
16239678Smckusick 	}
16345173Smckusick #ifdef QUOTA
16445173Smckusick 	if (error = chkdq(ip, (long)btodb(nsize - osize), cred, 0)) {
16545173Smckusick 		brelse(bp);
16645173Smckusick 		return (error);
16745173Smckusick 	}
16845173Smckusick #endif
16939678Smckusick 	/*
17039678Smckusick 	 * Check for extension in the existing location.
17139678Smckusick 	 */
1726294Smckusick 	cg = dtog(fs, bprev);
17351469Sbostic 	if (bno = ffs_fragextend(ip, cg, (long)bprev, osize, nsize)) {
17439887Smckusick 		if (bp->b_blkno != fsbtodb(fs, bno))
17539678Smckusick 			panic("bad blockno");
17639762Smckusick 		ip->i_blocks += btodb(nsize - osize);
17764603Sbostic 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
17848948Smckusick 		allocbuf(bp, nsize);
17948948Smckusick 		bp->b_flags |= B_DONE;
18064508Sbostic 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
18137735Smckusick 		*bpp = bp;
18237735Smckusick 		return (0);
1834463Smckusic 	}
18439678Smckusick 	/*
18539678Smckusick 	 * Allocate a new disk location.
18639678Smckusick 	 */
1874948Smckusic 	if (bpref >= fs->fs_size)
1884948Smckusic 		bpref = 0;
18926253Skarels 	switch ((int)fs->fs_optim) {
19025256Smckusick 	case FS_OPTSPACE:
19125256Smckusick 		/*
19225256Smckusick 		 * Allocate an exact sized fragment. Although this makes
19325256Smckusick 		 * best use of space, we will waste time relocating it if
19425256Smckusick 		 * the file continues to grow. If the fragmentation is
19525256Smckusick 		 * less than half of the minimum free reserve, we choose
19625256Smckusick 		 * to begin optimizing for time.
19725256Smckusick 		 */
19824698Smckusick 		request = nsize;
19925256Smckusick 		if (fs->fs_minfree < 5 ||
20025256Smckusick 		    fs->fs_cstotal.cs_nffree >
20125256Smckusick 		    fs->fs_dsize * fs->fs_minfree / (2 * 100))
20225256Smckusick 			break;
20325256Smckusick 		log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
20425256Smckusick 			fs->fs_fsmnt);
20525256Smckusick 		fs->fs_optim = FS_OPTTIME;
20625256Smckusick 		break;
20725256Smckusick 	case FS_OPTTIME:
20825256Smckusick 		/*
20951469Sbostic 		 * At this point we have discovered a file that is trying to
21051469Sbostic 		 * grow a small fragment to a larger fragment. To save time,
21151469Sbostic 		 * we allocate a full sized block, then free the unused portion.
21251469Sbostic 		 * If the file continues to grow, the `ffs_fragextend' call
21351469Sbostic 		 * above will be able to grow it in place without further
21451469Sbostic 		 * copying. If aberrant programs cause disk fragmentation to
21551469Sbostic 		 * grow within 2% of the free reserve, we choose to begin
21651469Sbostic 		 * optimizing for space.
21725256Smckusick 		 */
21824698Smckusick 		request = fs->fs_bsize;
21925256Smckusick 		if (fs->fs_cstotal.cs_nffree <
22025256Smckusick 		    fs->fs_dsize * (fs->fs_minfree - 2) / 100)
22125256Smckusick 			break;
22225256Smckusick 		log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
22325256Smckusick 			fs->fs_fsmnt);
22425256Smckusick 		fs->fs_optim = FS_OPTSPACE;
22525256Smckusick 		break;
22625256Smckusick 	default:
22725256Smckusick 		printf("dev = 0x%x, optim = %d, fs = %s\n",
22825256Smckusick 		    ip->i_dev, fs->fs_optim, fs->fs_fsmnt);
22951469Sbostic 		panic("ffs_realloccg: bad optim");
23025256Smckusick 		/* NOTREACHED */
23125256Smckusick 	}
23251469Sbostic 	bno = (daddr_t)ffs_hashalloc(ip, cg, (long)bpref, request,
23351469Sbostic 	    (u_long (*)())ffs_alloccg);
2346567Smckusic 	if (bno > 0) {
23545719Smckusick 		bp->b_blkno = fsbtodb(fs, bno);
23645719Smckusick 		(void) vnode_pager_uncache(ITOV(ip));
23753059Smckusick 		ffs_blkfree(ip, bprev, (long)osize);
23824698Smckusick 		if (nsize < request)
23951469Sbostic 			ffs_blkfree(ip, bno + numfrags(fs, nsize),
24053059Smckusick 			    (long)(request - nsize));
24112643Ssam 		ip->i_blocks += btodb(nsize - osize);
24264603Sbostic 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
24348948Smckusick 		allocbuf(bp, nsize);
24448948Smckusick 		bp->b_flags |= B_DONE;
24564508Sbostic 		bzero((char *)bp->b_data + osize, (u_int)nsize - osize);
24637735Smckusick 		*bpp = bp;
24737735Smckusick 		return (0);
2484463Smckusic 	}
24945173Smckusick #ifdef QUOTA
25045173Smckusick 	/*
25145173Smckusick 	 * Restore user's disk quota because allocation failed.
25245173Smckusick 	 */
25345173Smckusick 	(void) chkdq(ip, (long)-btodb(nsize - osize), cred, FORCE);
25445173Smckusick #endif
25539678Smckusick 	brelse(bp);
2564792Smckusic nospace:
2574463Smckusic 	/*
2584463Smckusic 	 * no space available
2594463Smckusic 	 */
26051469Sbostic 	ffs_fserr(fs, cred->cr_uid, "file system full");
2614426Smckusic 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
26237735Smckusick 	return (ENOSPC);
2634426Smckusic }
2644426Smckusic 
2655375Smckusic /*
26665999Smckusick  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
26765999Smckusick  *
26865999Smckusick  * The vnode and an array of buffer pointers for a range of sequential
26965999Smckusick  * logical blocks to be made contiguous is given. The allocator attempts
27065999Smckusick  * to find a range of sequential blocks starting as close as possible to
27165999Smckusick  * an fs_rotdelay offset from the end of the allocation for the logical
27265999Smckusick  * block immediately preceeding the current range. If successful, the
27365999Smckusick  * physical block numbers in the buffer pointers and in the inode are
27465999Smckusick  * changed to reflect the new allocation. If unsuccessful, the allocation
27565999Smckusick  * is left unchanged. The success in doing the reallocation is returned.
27665999Smckusick  * Note that the error return is not reflected back to the user. Rather
27765999Smckusick  * the previous block allocation will be used.
27865999Smckusick  */
27966176Smckusick #include <sys/sysctl.h>
28066176Smckusick int doasyncfree = 1;
28167392Smkm #ifdef DEBUG
28266176Smckusick struct ctldebug debug14 = { "doasyncfree", &doasyncfree };
28367392Smkm #endif
28467392Smkm 
28565999Smckusick int
28665999Smckusick ffs_reallocblks(ap)
28765999Smckusick 	struct vop_reallocblks_args /* {
28865999Smckusick 		struct vnode *a_vp;
28965999Smckusick 		struct cluster_save *a_buflist;
29065999Smckusick 	} */ *ap;
29165999Smckusick {
29265999Smckusick 	struct fs *fs;
29365999Smckusick 	struct inode *ip;
29465999Smckusick 	struct vnode *vp;
29565999Smckusick 	struct buf *sbp, *ebp;
29665999Smckusick 	daddr_t *bap, *sbap, *ebap;
29765999Smckusick 	struct cluster_save *buflist;
29865999Smckusick 	daddr_t start_lbn, end_lbn, soff, eoff, newblk, blkno;
29965999Smckusick 	struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
30065999Smckusick 	int i, len, start_lvl, end_lvl, pref, ssize;
30165999Smckusick 
30265999Smckusick 	vp = ap->a_vp;
30365999Smckusick 	ip = VTOI(vp);
30465999Smckusick 	fs = ip->i_fs;
30565999Smckusick 	if (fs->fs_contigsumsize <= 0)
30665999Smckusick 		return (ENOSPC);
30765999Smckusick 	buflist = ap->a_buflist;
30865999Smckusick 	len = buflist->bs_nchildren;
30965999Smckusick 	start_lbn = buflist->bs_children[0]->b_lblkno;
31065999Smckusick 	end_lbn = start_lbn + len - 1;
31165999Smckusick #ifdef DIAGNOSTIC
31265999Smckusick 	for (i = 1; i < len; i++)
31365999Smckusick 		if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
31465999Smckusick 			panic("ffs_reallocblks: non-cluster");
31565999Smckusick #endif
31665999Smckusick 	/*
31765999Smckusick 	 * If the latest allocation is in a new cylinder group, assume that
31865999Smckusick 	 * the filesystem has decided to move and do not force it back to
31965999Smckusick 	 * the previous cylinder group.
32065999Smckusick 	 */
32165999Smckusick 	if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
32265999Smckusick 	    dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
32365999Smckusick 		return (ENOSPC);
32465999Smckusick 	if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
32565999Smckusick 	    ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
32665999Smckusick 		return (ENOSPC);
32765999Smckusick 	/*
32865999Smckusick 	 * Get the starting offset and block map for the first block.
32965999Smckusick 	 */
33065999Smckusick 	if (start_lvl == 0) {
33165999Smckusick 		sbap = &ip->i_db[0];
33265999Smckusick 		soff = start_lbn;
33365999Smckusick 	} else {
33465999Smckusick 		idp = &start_ap[start_lvl - 1];
33565999Smckusick 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
33665999Smckusick 			brelse(sbp);
33765999Smckusick 			return (ENOSPC);
33865999Smckusick 		}
33965999Smckusick 		sbap = (daddr_t *)sbp->b_data;
34065999Smckusick 		soff = idp->in_off;
34165999Smckusick 	}
34265999Smckusick 	/*
34365999Smckusick 	 * Find the preferred location for the cluster.
34465999Smckusick 	 */
34565999Smckusick 	pref = ffs_blkpref(ip, start_lbn, soff, sbap);
34665999Smckusick 	/*
34765999Smckusick 	 * If the block range spans two block maps, get the second map.
34865999Smckusick 	 */
34966086Smckusick 	if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
35065999Smckusick 		ssize = len;
35165999Smckusick 	} else {
35266086Smckusick #ifdef DIAGNOSTIC
35366086Smckusick 		if (start_ap[start_lvl-1].in_lbn == idp->in_lbn)
35466086Smckusick 			panic("ffs_reallocblk: start == end");
35566086Smckusick #endif
35665999Smckusick 		ssize = len - (idp->in_off + 1);
35765999Smckusick 		if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
35865999Smckusick 			goto fail;
35965999Smckusick 		ebap = (daddr_t *)ebp->b_data;
36065999Smckusick 	}
36165999Smckusick 	/*
36265999Smckusick 	 * Search the block map looking for an allocation of the desired size.
36365999Smckusick 	 */
36465999Smckusick 	if ((newblk = (daddr_t)ffs_hashalloc(ip, dtog(fs, pref), (long)pref,
36565999Smckusick 	    len, (u_long (*)())ffs_clusteralloc)) == 0)
36665999Smckusick 		goto fail;
36765999Smckusick 	/*
36865999Smckusick 	 * We have found a new contiguous block.
36965999Smckusick 	 *
37065999Smckusick 	 * First we have to replace the old block pointers with the new
37165999Smckusick 	 * block pointers in the inode and indirect blocks associated
37265999Smckusick 	 * with the file.
37365999Smckusick 	 */
37465999Smckusick 	blkno = newblk;
37565999Smckusick 	for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
37665999Smckusick 		if (i == ssize)
37765999Smckusick 			bap = ebap;
37865999Smckusick #ifdef DIAGNOSTIC
379*67869Smckusick 		if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
38065999Smckusick 			panic("ffs_reallocblks: alloc mismatch");
38165999Smckusick #endif
38265999Smckusick 		*bap++ = blkno;
38365999Smckusick 	}
38465999Smckusick 	/*
38565999Smckusick 	 * Next we must write out the modified inode and indirect blocks.
38666176Smckusick 	 * For strict correctness, the writes should be synchronous since
38766176Smckusick 	 * the old block values may have been written to disk. In practise
38866176Smckusick 	 * they are almost never written, but if we are concerned about
38966176Smckusick 	 * strict correctness, the `doasyncfree' flag should be set to zero.
39065999Smckusick 	 *
39166176Smckusick 	 * The test on `doasyncfree' should be changed to test a flag
39266176Smckusick 	 * that shows whether the associated buffers and inodes have
39366176Smckusick 	 * been written. The flag should be set when the cluster is
39466176Smckusick 	 * started and cleared whenever the buffer or inode is flushed.
39566176Smckusick 	 * We can then check below to see if it is set, and do the
39666176Smckusick 	 * synchronous write only when it has been cleared.
39765999Smckusick 	 */
39865999Smckusick 	if (sbap != &ip->i_db[0]) {
39966176Smckusick 		if (doasyncfree)
40066176Smckusick 			bdwrite(sbp);
40166176Smckusick 		else
40266176Smckusick 			bwrite(sbp);
40365999Smckusick 	} else {
40465999Smckusick 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
40566176Smckusick 		if (!doasyncfree)
40666176Smckusick 			VOP_UPDATE(vp, &time, &time, MNT_WAIT);
40765999Smckusick 	}
40865999Smckusick 	if (ssize < len)
40966176Smckusick 		if (doasyncfree)
41066176Smckusick 			bdwrite(ebp);
41166176Smckusick 		else
41266176Smckusick 			bwrite(ebp);
41365999Smckusick 	/*
41465999Smckusick 	 * Last, free the old blocks and assign the new blocks to the buffers.
41565999Smckusick 	 */
41665999Smckusick 	for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
41765999Smckusick 		ffs_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno),
41865999Smckusick 		    fs->fs_bsize);
41965999Smckusick 		buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
42065999Smckusick 	}
42165999Smckusick 	return (0);
42265999Smckusick 
42365999Smckusick fail:
42465999Smckusick 	if (ssize < len)
42565999Smckusick 		brelse(ebp);
42665999Smckusick 	if (sbap != &ip->i_db[0])
42765999Smckusick 		brelse(sbp);
42865999Smckusick 	return (ENOSPC);
42965999Smckusick }
43065999Smckusick 
43165999Smckusick /*
4325375Smckusic  * Allocate an inode in the file system.
4335375Smckusic  *
43451469Sbostic  * If allocating a directory, use ffs_dirpref to select the inode.
43551469Sbostic  * If allocating in a directory, the following hierarchy is followed:
43651469Sbostic  *   1) allocate the preferred inode.
4375375Smckusic  *   2) allocate an inode in the same cylinder group.
4385375Smckusic  *   3) quadradically rehash into other cylinder groups, until an
4395375Smckusic  *      available inode is located.
4405375Smckusic  * If no inode preference is given the following heirarchy is used
4415375Smckusic  * to allocate an inode:
4425375Smckusic  *   1) allocate an inode in cylinder group 0.
4435375Smckusic  *   2) quadradically rehash into other cylinder groups, until an
4445375Smckusic  *      available inode is located.
4455375Smckusic  */
44654653Smckusick ffs_valloc(ap)
44754653Smckusick 	struct vop_valloc_args /* {
44854653Smckusick 		struct vnode *a_pvp;
44954653Smckusick 		int a_mode;
45054653Smckusick 		struct ucred *a_cred;
45154653Smckusick 		struct vnode **a_vpp;
45254653Smckusick 	} */ *ap;
4534359Smckusick {
45453865Sheideman 	register struct vnode *pvp = ap->a_pvp;
45551541Smckusick 	register struct inode *pip;
4564359Smckusick 	register struct fs *fs;
4574359Smckusick 	register struct inode *ip;
45854653Smckusick 	mode_t mode = ap->a_mode;
45951469Sbostic 	ino_t ino, ipref;
46037735Smckusick 	int cg, error;
4614359Smckusick 
46253583Sheideman 	*ap->a_vpp = NULL;
46353865Sheideman 	pip = VTOI(pvp);
4645965Smckusic 	fs = pip->i_fs;
4654792Smckusic 	if (fs->fs_cstotal.cs_nifree == 0)
4664359Smckusick 		goto noinodes;
46751469Sbostic 
46854653Smckusick 	if ((mode & IFMT) == IFDIR)
46951541Smckusick 		ipref = ffs_dirpref(fs);
47051469Sbostic 	else
47151469Sbostic 		ipref = pip->i_number;
4724948Smckusic 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
4734948Smckusic 		ipref = 0;
47464603Sbostic 	cg = ino_to_cg(fs, ipref);
47564603Sbostic 	ino = (ino_t)ffs_hashalloc(pip, cg, (long)ipref, mode, ffs_nodealloccg);
4764359Smckusick 	if (ino == 0)
4774359Smckusick 		goto noinodes;
47854653Smckusick 	error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp);
47937735Smckusick 	if (error) {
48054653Smckusick 		VOP_VFREE(pvp, ino, mode);
48137735Smckusick 		return (error);
4824359Smckusick 	}
48353583Sheideman 	ip = VTOI(*ap->a_vpp);
4846716Smckusick 	if (ip->i_mode) {
48554653Smckusick 		printf("mode = 0%o, inum = %d, fs = %s\n",
4866716Smckusick 		    ip->i_mode, ip->i_number, fs->fs_fsmnt);
48751541Smckusick 		panic("ffs_valloc: dup alloc");
4886716Smckusick 	}
48912643Ssam 	if (ip->i_blocks) {				/* XXX */
49012643Ssam 		printf("free inode %s/%d had %d blocks\n",
49112643Ssam 		    fs->fs_fsmnt, ino, ip->i_blocks);
49212643Ssam 		ip->i_blocks = 0;
49312643Ssam 	}
49439516Smckusick 	ip->i_flags = 0;
49538255Smckusick 	/*
49638255Smckusick 	 * Set up a new generation number for this inode.
49738255Smckusick 	 */
49838255Smckusick 	if (++nextgennumber < (u_long)time.tv_sec)
49938255Smckusick 		nextgennumber = time.tv_sec;
50038255Smckusick 	ip->i_gen = nextgennumber;
50137735Smckusick 	return (0);
5024359Smckusick noinodes:
50353583Sheideman 	ffs_fserr(fs, ap->a_cred->cr_uid, "out of inodes");
5046294Smckusick 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
50537735Smckusick 	return (ENOSPC);
5064359Smckusick }
5074359Smckusick 
5084651Smckusic /*
5095375Smckusic  * Find a cylinder to place a directory.
5105375Smckusic  *
5115375Smckusic  * The policy implemented by this algorithm is to select from
5125375Smckusic  * among those cylinder groups with above the average number of
5135375Smckusic  * free inodes, the one with the smallest number of directories.
5144651Smckusic  */
51551469Sbostic static ino_t
51651469Sbostic ffs_dirpref(fs)
5175965Smckusic 	register struct fs *fs;
5184359Smckusick {
5194651Smckusic 	int cg, minndir, mincg, avgifree;
5204359Smckusick 
5214792Smckusic 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
5224651Smckusic 	minndir = fs->fs_ipg;
5234359Smckusick 	mincg = 0;
5244651Smckusic 	for (cg = 0; cg < fs->fs_ncg; cg++)
5255322Smckusic 		if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
5265322Smckusic 		    fs->fs_cs(fs, cg).cs_nifree >= avgifree) {
5274359Smckusick 			mincg = cg;
5285322Smckusic 			minndir = fs->fs_cs(fs, cg).cs_ndir;
5294359Smckusick 		}
5309163Ssam 	return ((ino_t)(fs->fs_ipg * mincg));
5314359Smckusick }
5324359Smckusick 
5334651Smckusic /*
5349163Ssam  * Select the desired position for the next block in a file.  The file is
5359163Ssam  * logically divided into sections. The first section is composed of the
5369163Ssam  * direct blocks. Each additional section contains fs_maxbpg blocks.
5379163Ssam  *
5389163Ssam  * If no blocks have been allocated in the first section, the policy is to
5399163Ssam  * request a block in the same cylinder group as the inode that describes
5409163Ssam  * the file. If no blocks have been allocated in any other section, the
5419163Ssam  * policy is to place the section in a cylinder group with a greater than
5429163Ssam  * average number of free blocks.  An appropriate cylinder group is found
54317696Smckusick  * by using a rotor that sweeps the cylinder groups. When a new group of
54417696Smckusick  * blocks is needed, the sweep begins in the cylinder group following the
54517696Smckusick  * cylinder group from which the previous allocation was made. The sweep
54617696Smckusick  * continues until a cylinder group with greater than the average number
54717696Smckusick  * of free blocks is found. If the allocation is for the first block in an
54817696Smckusick  * indirect block, the information on the previous allocation is unavailable;
54917696Smckusick  * here a best guess is made based upon the logical block number being
55017696Smckusick  * allocated.
5519163Ssam  *
5529163Ssam  * If a section is already partially allocated, the policy is to
5539163Ssam  * contiguously allocate fs_maxcontig blocks.  The end of one of these
5549163Ssam  * contiguous blocks and the beginning of the next is physically separated
5559163Ssam  * so that the disk head will be in transit between them for at least
5569163Ssam  * fs_rotdelay milliseconds.  This is to allow time for the processor to
5579163Ssam  * schedule another I/O transfer.
5584651Smckusic  */
5595212Smckusic daddr_t
56051469Sbostic ffs_blkpref(ip, lbn, indx, bap)
5619163Ssam 	struct inode *ip;
5629163Ssam 	daddr_t lbn;
5639163Ssam 	int indx;
5649163Ssam 	daddr_t *bap;
5659163Ssam {
5665965Smckusic 	register struct fs *fs;
56717696Smckusick 	register int cg;
56817696Smckusick 	int avgbfree, startcg;
5699163Ssam 	daddr_t nextblk;
5704651Smckusic 
5719163Ssam 	fs = ip->i_fs;
5729163Ssam 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
5739163Ssam 		if (lbn < NDADDR) {
57464603Sbostic 			cg = ino_to_cg(fs, ip->i_number);
5755322Smckusic 			return (fs->fs_fpg * cg + fs->fs_frag);
5764651Smckusic 		}
5779163Ssam 		/*
5789163Ssam 		 * Find a cylinder with greater than average number of
5799163Ssam 		 * unused data blocks.
5809163Ssam 		 */
58117696Smckusick 		if (indx == 0 || bap[indx - 1] == 0)
58264603Sbostic 			startcg =
58364603Sbostic 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
58417696Smckusick 		else
58517696Smckusick 			startcg = dtog(fs, bap[indx - 1]) + 1;
58617696Smckusick 		startcg %= fs->fs_ncg;
5879163Ssam 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
58817696Smckusick 		for (cg = startcg; cg < fs->fs_ncg; cg++)
5899163Ssam 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
5909163Ssam 				fs->fs_cgrotor = cg;
5919163Ssam 				return (fs->fs_fpg * cg + fs->fs_frag);
5929163Ssam 			}
59317696Smckusick 		for (cg = 0; cg <= startcg; cg++)
5949163Ssam 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
5959163Ssam 				fs->fs_cgrotor = cg;
5969163Ssam 				return (fs->fs_fpg * cg + fs->fs_frag);
5979163Ssam 			}
5989163Ssam 		return (NULL);
5999163Ssam 	}
6009163Ssam 	/*
6019163Ssam 	 * One or more previous blocks have been laid out. If less
6029163Ssam 	 * than fs_maxcontig previous blocks are contiguous, the
6039163Ssam 	 * next block is requested contiguously, otherwise it is
6049163Ssam 	 * requested rotationally delayed by fs_rotdelay milliseconds.
6059163Ssam 	 */
6069163Ssam 	nextblk = bap[indx - 1] + fs->fs_frag;
60756665Smargo 	if (indx < fs->fs_maxcontig || bap[indx - fs->fs_maxcontig] +
60856665Smargo 	    blkstofrags(fs, fs->fs_maxcontig) != nextblk)
6099163Ssam 		return (nextblk);
6109163Ssam 	if (fs->fs_rotdelay != 0)
6119163Ssam 		/*
6129163Ssam 		 * Here we convert ms of delay to frags as:
6139163Ssam 		 * (frags) = (ms) * (rev/sec) * (sect/rev) /
6149163Ssam 		 *	((sect/frag) * (ms/sec))
6159163Ssam 		 * then round up to the next block.
6169163Ssam 		 */
6179163Ssam 		nextblk += roundup(fs->fs_rotdelay * fs->fs_rps * fs->fs_nsect /
6189163Ssam 		    (NSPF(fs) * 1000), fs->fs_frag);
6199163Ssam 	return (nextblk);
6204651Smckusic }
6214651Smckusic 
6225375Smckusic /*
6235375Smckusic  * Implement the cylinder overflow algorithm.
6245375Smckusic  *
6255375Smckusic  * The policy implemented by this algorithm is:
6265375Smckusic  *   1) allocate the block in its requested cylinder group.
6275375Smckusic  *   2) quadradically rehash on the cylinder group number.
6285375Smckusic  *   3) brute force search for a free block.
6295375Smckusic  */
6305212Smckusic /*VARARGS5*/
63151469Sbostic static u_long
63251469Sbostic ffs_hashalloc(ip, cg, pref, size, allocator)
6335965Smckusic 	struct inode *ip;
6344359Smckusick 	int cg;
6354359Smckusick 	long pref;
6364359Smckusick 	int size;	/* size for data blocks, mode for inodes */
6375212Smckusic 	u_long (*allocator)();
6384359Smckusick {
6395965Smckusic 	register struct fs *fs;
6404359Smckusick 	long result;
6414359Smckusick 	int i, icg = cg;
6424359Smckusick 
6435965Smckusic 	fs = ip->i_fs;
6444359Smckusick 	/*
6454359Smckusick 	 * 1: preferred cylinder group
6464359Smckusick 	 */
6475965Smckusic 	result = (*allocator)(ip, cg, pref, size);
6484359Smckusick 	if (result)
6494359Smckusick 		return (result);
6504359Smckusick 	/*
6514359Smckusick 	 * 2: quadratic rehash
6524359Smckusick 	 */
6534359Smckusick 	for (i = 1; i < fs->fs_ncg; i *= 2) {
6544359Smckusick 		cg += i;
6554359Smckusick 		if (cg >= fs->fs_ncg)
6564359Smckusick 			cg -= fs->fs_ncg;
6575965Smckusic 		result = (*allocator)(ip, cg, 0, size);
6584359Smckusick 		if (result)
6594359Smckusick 			return (result);
6604359Smckusick 	}
6614359Smckusick 	/*
6624359Smckusick 	 * 3: brute force search
66310847Ssam 	 * Note that we start at i == 2, since 0 was checked initially,
66410847Ssam 	 * and 1 is always checked in the quadratic rehash.
6654359Smckusick 	 */
66610848Smckusick 	cg = (icg + 2) % fs->fs_ncg;
66710847Ssam 	for (i = 2; i < fs->fs_ncg; i++) {
6685965Smckusic 		result = (*allocator)(ip, cg, 0, size);
6694359Smckusick 		if (result)
6704359Smckusick 			return (result);
6714359Smckusick 		cg++;
6724359Smckusick 		if (cg == fs->fs_ncg)
6734359Smckusick 			cg = 0;
6744359Smckusick 	}
6756294Smckusick 	return (NULL);
6764359Smckusick }
6774359Smckusick 
6785375Smckusic /*
6795375Smckusic  * Determine whether a fragment can be extended.
6805375Smckusic  *
6815375Smckusic  * Check to see if the necessary fragments are available, and
6825375Smckusic  * if they are, allocate them.
6835375Smckusic  */
68451469Sbostic static daddr_t
68551469Sbostic ffs_fragextend(ip, cg, bprev, osize, nsize)
6865965Smckusic 	struct inode *ip;
6874426Smckusic 	int cg;
6884463Smckusic 	long bprev;
6894426Smckusic 	int osize, nsize;
6904426Smckusic {
6915965Smckusic 	register struct fs *fs;
6924463Smckusic 	register struct cg *cgp;
69337735Smckusick 	struct buf *bp;
6944463Smckusic 	long bno;
6954463Smckusic 	int frags, bbase;
69637735Smckusick 	int i, error;
6974426Smckusic 
6985965Smckusic 	fs = ip->i_fs;
69917224Smckusick 	if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
7006531Smckusick 		return (NULL);
7015960Smckusic 	frags = numfrags(fs, nsize);
70217224Smckusick 	bbase = fragnum(fs, bprev);
70317224Smckusick 	if (bbase > fragnum(fs, (bprev + frags - 1))) {
70430749Skarels 		/* cannot extend across a block boundary */
7056294Smckusick 		return (NULL);
7064463Smckusic 	}
70737735Smckusick 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
70838776Smckusick 		(int)fs->fs_cgsize, NOCRED, &bp);
70937735Smckusick 	if (error) {
71037735Smckusick 		brelse(bp);
71137735Smckusick 		return (NULL);
71237735Smckusick 	}
71364508Sbostic 	cgp = (struct cg *)bp->b_data;
71437735Smckusick 	if (!cg_chkmagic(cgp)) {
7155960Smckusic 		brelse(bp);
7166294Smckusick 		return (NULL);
7175960Smckusic 	}
7188105Sroot 	cgp->cg_time = time.tv_sec;
7195377Smckusic 	bno = dtogd(fs, bprev);
7205960Smckusic 	for (i = numfrags(fs, osize); i < frags; i++)
72134143Smckusick 		if (isclr(cg_blksfree(cgp), bno + i)) {
7225361Smckusic 			brelse(bp);
7236294Smckusick 			return (NULL);
7245361Smckusic 		}
7255361Smckusic 	/*
7265361Smckusic 	 * the current fragment can be extended
7275361Smckusic 	 * deduct the count on fragment being extended into
7285361Smckusic 	 * increase the count on the remaining fragment (if any)
7295361Smckusic 	 * allocate the extended piece
7305361Smckusic 	 */
7315361Smckusic 	for (i = frags; i < fs->fs_frag - bbase; i++)
73234143Smckusick 		if (isclr(cg_blksfree(cgp), bno + i))
7334463Smckusic 			break;
7345960Smckusic 	cgp->cg_frsum[i - numfrags(fs, osize)]--;
7355361Smckusic 	if (i != frags)
7365361Smckusic 		cgp->cg_frsum[i - frags]++;
7375960Smckusic 	for (i = numfrags(fs, osize); i < frags; i++) {
73834143Smckusick 		clrbit(cg_blksfree(cgp), bno + i);
7395361Smckusic 		cgp->cg_cs.cs_nffree--;
7405361Smckusic 		fs->fs_cstotal.cs_nffree--;
7415361Smckusic 		fs->fs_cs(fs, cg).cs_nffree--;
7424463Smckusic 	}
74350893Smckusick 	fs->fs_fmod = 1;
7445361Smckusic 	bdwrite(bp);
7455361Smckusic 	return (bprev);
7464426Smckusic }
7474426Smckusic 
7485375Smckusic /*
7495375Smckusic  * Determine whether a block can be allocated.
7505375Smckusic  *
75164603Sbostic  * Check to see if a block of the appropriate size is available,
7525375Smckusic  * and if it is, allocate it.
7535375Smckusic  */
75451779Smarc static daddr_t
75551469Sbostic ffs_alloccg(ip, cg, bpref, size)
7565965Smckusic 	struct inode *ip;
7574359Smckusick 	int cg;
7584359Smckusick 	daddr_t bpref;
7594359Smckusick 	int size;
7604359Smckusick {
7615965Smckusic 	register struct fs *fs;
7624463Smckusic 	register struct cg *cgp;
76337735Smckusick 	struct buf *bp;
7644463Smckusic 	register int i;
76537735Smckusick 	int error, bno, frags, allocsiz;
7664359Smckusick 
7675965Smckusic 	fs = ip->i_fs;
7685322Smckusic 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
7696294Smckusick 		return (NULL);
77037735Smckusick 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
77138776Smckusick 		(int)fs->fs_cgsize, NOCRED, &bp);
77237735Smckusick 	if (error) {
77337735Smckusick 		brelse(bp);
77437735Smckusick 		return (NULL);
77537735Smckusick 	}
77664508Sbostic 	cgp = (struct cg *)bp->b_data;
77737735Smckusick 	if (!cg_chkmagic(cgp) ||
77815950Smckusick 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) {
7795960Smckusic 		brelse(bp);
7806294Smckusick 		return (NULL);
7815960Smckusic 	}
7828105Sroot 	cgp->cg_time = time.tv_sec;
7835322Smckusic 	if (size == fs->fs_bsize) {
78451469Sbostic 		bno = ffs_alloccgblk(fs, cgp, bpref);
7854463Smckusic 		bdwrite(bp);
7864463Smckusic 		return (bno);
7874463Smckusic 	}
7884463Smckusic 	/*
7894463Smckusic 	 * check to see if any fragments are already available
7904463Smckusic 	 * allocsiz is the size which will be allocated, hacking
7914463Smckusic 	 * it down to a smaller size if necessary
7924463Smckusic 	 */
7935960Smckusic 	frags = numfrags(fs, size);
7945322Smckusic 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
7954463Smckusic 		if (cgp->cg_frsum[allocsiz] != 0)
7964463Smckusic 			break;
7975322Smckusic 	if (allocsiz == fs->fs_frag) {
7984463Smckusic 		/*
7994463Smckusic 		 * no fragments were available, so a block will be
8004463Smckusic 		 * allocated, and hacked up
8014463Smckusic 		 */
8024792Smckusic 		if (cgp->cg_cs.cs_nbfree == 0) {
8034463Smckusic 			brelse(bp);
8046294Smckusick 			return (NULL);
8054463Smckusic 		}
80651469Sbostic 		bno = ffs_alloccgblk(fs, cgp, bpref);
8075377Smckusic 		bpref = dtogd(fs, bno);
8085322Smckusic 		for (i = frags; i < fs->fs_frag; i++)
80934143Smckusick 			setbit(cg_blksfree(cgp), bpref + i);
8105322Smckusic 		i = fs->fs_frag - frags;
8114792Smckusic 		cgp->cg_cs.cs_nffree += i;
8124792Smckusic 		fs->fs_cstotal.cs_nffree += i;
8135322Smckusic 		fs->fs_cs(fs, cg).cs_nffree += i;
81450893Smckusick 		fs->fs_fmod = 1;
8154463Smckusic 		cgp->cg_frsum[i]++;
8164463Smckusic 		bdwrite(bp);
8174463Smckusic 		return (bno);
8184463Smckusic 	}
81951469Sbostic 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
82015950Smckusick 	if (bno < 0) {
82115950Smckusick 		brelse(bp);
8226294Smckusick 		return (NULL);
82315950Smckusick 	}
8244463Smckusic 	for (i = 0; i < frags; i++)
82534143Smckusick 		clrbit(cg_blksfree(cgp), bno + i);
8264792Smckusic 	cgp->cg_cs.cs_nffree -= frags;
8274792Smckusic 	fs->fs_cstotal.cs_nffree -= frags;
8285322Smckusic 	fs->fs_cs(fs, cg).cs_nffree -= frags;
82950893Smckusick 	fs->fs_fmod = 1;
8304463Smckusic 	cgp->cg_frsum[allocsiz]--;
8314463Smckusic 	if (frags != allocsiz)
8324463Smckusic 		cgp->cg_frsum[allocsiz - frags]++;
8334463Smckusic 	bdwrite(bp);
8344463Smckusic 	return (cg * fs->fs_fpg + bno);
8354463Smckusic }
8364463Smckusic 
8375375Smckusic /*
8385375Smckusic  * Allocate a block in a cylinder group.
8395375Smckusic  *
8405375Smckusic  * This algorithm implements the following policy:
8415375Smckusic  *   1) allocate the requested block.
8425375Smckusic  *   2) allocate a rotationally optimal block in the same cylinder.
8435375Smckusic  *   3) allocate the next available block on the block rotor for the
8445375Smckusic  *      specified cylinder group.
8455375Smckusic  * Note that this routine only allocates fs_bsize blocks; these
8465375Smckusic  * blocks may be fragmented by the routine that allocates them.
8475375Smckusic  */
84851469Sbostic static daddr_t
84951469Sbostic ffs_alloccgblk(fs, cgp, bpref)
8505965Smckusic 	register struct fs *fs;
8514463Smckusic 	register struct cg *cgp;
8524463Smckusic 	daddr_t bpref;
8534463Smckusic {
85465999Smckusick 	daddr_t bno, blkno;
8556294Smckusick 	int cylno, pos, delta;
8564651Smckusic 	short *cylbp;
8575361Smckusic 	register int i;
8584463Smckusic 
85965999Smckusick 	if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
8604651Smckusic 		bpref = cgp->cg_rotor;
8615361Smckusic 		goto norot;
8625361Smckusic 	}
86317224Smckusick 	bpref = blknum(fs, bpref);
8645377Smckusic 	bpref = dtogd(fs, bpref);
8655361Smckusic 	/*
8665361Smckusic 	 * if the requested block is available, use it
8675361Smckusic 	 */
86851469Sbostic 	if (ffs_isblock(fs, cg_blksfree(cgp), fragstoblks(fs, bpref))) {
8695361Smckusic 		bno = bpref;
8705361Smckusic 		goto gotit;
8715361Smckusic 	}
8725361Smckusic 	/*
8735361Smckusic 	 * check for a block available on the same cylinder
8745361Smckusic 	 */
8755361Smckusic 	cylno = cbtocylno(fs, bpref);
87634143Smckusick 	if (cg_blktot(cgp)[cylno] == 0)
8775375Smckusic 		goto norot;
8785375Smckusic 	if (fs->fs_cpc == 0) {
8795375Smckusic 		/*
88057000Smckusick 		 * Block layout information is not available.
88157000Smckusick 		 * Leaving bpref unchanged means we take the
88257000Smckusick 		 * next available free block following the one
88357000Smckusick 		 * we just allocated. Hopefully this will at
88457000Smckusick 		 * least hit a track cache on drives of unknown
88557000Smckusick 		 * geometry (e.g. SCSI).
8865375Smckusic 		 */
8875375Smckusic 		goto norot;
8885375Smckusic 	}
8895375Smckusic 	/*
8905361Smckusic 	 * check the summary information to see if a block is
8915361Smckusic 	 * available in the requested cylinder starting at the
8929163Ssam 	 * requested rotational position and proceeding around.
8935361Smckusic 	 */
89434143Smckusick 	cylbp = cg_blks(fs, cgp, cylno);
8959163Ssam 	pos = cbtorpos(fs, bpref);
89634143Smckusick 	for (i = pos; i < fs->fs_nrpos; i++)
8975361Smckusic 		if (cylbp[i] > 0)
8985361Smckusic 			break;
89934143Smckusick 	if (i == fs->fs_nrpos)
9005361Smckusic 		for (i = 0; i < pos; i++)
9015361Smckusic 			if (cylbp[i] > 0)
9025361Smckusic 				break;
9035361Smckusic 	if (cylbp[i] > 0) {
9044651Smckusic 		/*
9055361Smckusic 		 * found a rotational position, now find the actual
9065361Smckusic 		 * block. A panic if none is actually there.
9074651Smckusic 		 */
9085361Smckusic 		pos = cylno % fs->fs_cpc;
9095361Smckusic 		bno = (cylno - pos) * fs->fs_spc / NSPB(fs);
91034143Smckusick 		if (fs_postbl(fs, pos)[i] == -1) {
9116716Smckusick 			printf("pos = %d, i = %d, fs = %s\n",
9126716Smckusick 			    pos, i, fs->fs_fsmnt);
91351469Sbostic 			panic("ffs_alloccgblk: cyl groups corrupted");
9146716Smckusick 		}
91534143Smckusick 		for (i = fs_postbl(fs, pos)[i];; ) {
91651469Sbostic 			if (ffs_isblock(fs, cg_blksfree(cgp), bno + i)) {
91711638Ssam 				bno = blkstofrags(fs, (bno + i));
9185361Smckusic 				goto gotit;
9195361Smckusic 			}
92034143Smckusick 			delta = fs_rotbl(fs)[i];
92134143Smckusick 			if (delta <= 0 ||
92234143Smckusick 			    delta + i > fragstoblks(fs, fs->fs_fpg))
9234651Smckusic 				break;
9246294Smckusick 			i += delta;
9254651Smckusic 		}
9266716Smckusick 		printf("pos = %d, i = %d, fs = %s\n", pos, i, fs->fs_fsmnt);
92751469Sbostic 		panic("ffs_alloccgblk: can't find blk in cyl");
9284359Smckusick 	}
9295361Smckusic norot:
9305361Smckusic 	/*
9315361Smckusic 	 * no blocks in the requested cylinder, so take next
9325361Smckusic 	 * available one in this cylinder group.
9335361Smckusic 	 */
93451469Sbostic 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
9356567Smckusic 	if (bno < 0)
9366294Smckusick 		return (NULL);
9374651Smckusic 	cgp->cg_rotor = bno;
9384359Smckusick gotit:
93965999Smckusick 	blkno = fragstoblks(fs, bno);
94065999Smckusick 	ffs_clrblock(fs, cg_blksfree(cgp), (long)blkno);
94165999Smckusick 	ffs_clusteracct(fs, cgp, blkno, -1);
9424792Smckusic 	cgp->cg_cs.cs_nbfree--;
9434792Smckusic 	fs->fs_cstotal.cs_nbfree--;
9445322Smckusic 	fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
9455375Smckusic 	cylno = cbtocylno(fs, bno);
94634143Smckusick 	cg_blks(fs, cgp, cylno)[cbtorpos(fs, bno)]--;
94734143Smckusick 	cg_blktot(cgp)[cylno]--;
94850893Smckusick 	fs->fs_fmod = 1;
9494651Smckusic 	return (cgp->cg_cgx * fs->fs_fpg + bno);
9504359Smckusick }
95134143Smckusick 
9525375Smckusic /*
95365999Smckusick  * Determine whether a cluster can be allocated.
95465999Smckusick  *
95565999Smckusick  * We do not currently check for optimal rotational layout if there
95665999Smckusick  * are multiple choices in the same cylinder group. Instead we just
95765999Smckusick  * take the first one that we find following bpref.
95865999Smckusick  */
95965999Smckusick static daddr_t
96065999Smckusick ffs_clusteralloc(ip, cg, bpref, len)
96165999Smckusick 	struct inode *ip;
96265999Smckusick 	int cg;
96365999Smckusick 	daddr_t bpref;
96465999Smckusick 	int len;
96565999Smckusick {
96665999Smckusick 	register struct fs *fs;
96765999Smckusick 	register struct cg *cgp;
96865999Smckusick 	struct buf *bp;
96965999Smckusick 	int i, run, bno, bit, map;
97065999Smckusick 	u_char *mapp;
971*67869Smckusick 	int32_t *lp;
97265999Smckusick 
97365999Smckusick 	fs = ip->i_fs;
974*67869Smckusick 	if (fs->fs_maxcluster[cg] < len)
97565999Smckusick 		return (NULL);
97665999Smckusick 	if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
97765999Smckusick 	    NOCRED, &bp))
97865999Smckusick 		goto fail;
97965999Smckusick 	cgp = (struct cg *)bp->b_data;
98065999Smckusick 	if (!cg_chkmagic(cgp))
98165999Smckusick 		goto fail;
98265999Smckusick 	/*
98365999Smckusick 	 * Check to see if a cluster of the needed size (or bigger) is
98465999Smckusick 	 * available in this cylinder group.
98565999Smckusick 	 */
986*67869Smckusick 	lp = &cg_clustersum(cgp)[len];
98765999Smckusick 	for (i = len; i <= fs->fs_contigsumsize; i++)
988*67869Smckusick 		if (*lp++ > 0)
98965999Smckusick 			break;
990*67869Smckusick 	if (i > fs->fs_contigsumsize) {
991*67869Smckusick 		/*
992*67869Smckusick 		 * This is the first time looking for a cluster in this
993*67869Smckusick 		 * cylinder group. Update the cluster summary information
994*67869Smckusick 		 * to reflect the true maximum sized cluster so that
995*67869Smckusick 		 * future cluster allocation requests can avoid reading
996*67869Smckusick 		 * the cylinder group map only to find no clusters.
997*67869Smckusick 		 */
998*67869Smckusick 		lp = &cg_clustersum(cgp)[len - 1];
999*67869Smckusick 		for (i = len - 1; i > 0; i--)
1000*67869Smckusick 			if (*lp-- > 0)
1001*67869Smckusick 				break;
1002*67869Smckusick 		fs->fs_maxcluster[cg] = i;
100365999Smckusick 		goto fail;
1004*67869Smckusick 	}
100565999Smckusick 	/*
100665999Smckusick 	 * Search the cluster map to find a big enough cluster.
100765999Smckusick 	 * We take the first one that we find, even if it is larger
100865999Smckusick 	 * than we need as we prefer to get one close to the previous
100965999Smckusick 	 * block allocation. We do not search before the current
101065999Smckusick 	 * preference point as we do not want to allocate a block
101165999Smckusick 	 * that is allocated before the previous one (as we will
101265999Smckusick 	 * then have to wait for another pass of the elevator
101365999Smckusick 	 * algorithm before it will be read). We prefer to fail and
101465999Smckusick 	 * be recalled to try an allocation in the next cylinder group.
101565999Smckusick 	 */
101665999Smckusick 	if (dtog(fs, bpref) != cg)
101765999Smckusick 		bpref = 0;
101865999Smckusick 	else
101965999Smckusick 		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
102065999Smckusick 	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
102165999Smckusick 	map = *mapp++;
102265999Smckusick 	bit = 1 << (bpref % NBBY);
102365999Smckusick 	for (run = 0, i = bpref; i < cgp->cg_nclusterblks; i++) {
102465999Smckusick 		if ((map & bit) == 0) {
102565999Smckusick 			run = 0;
102665999Smckusick 		} else {
102765999Smckusick 			run++;
102865999Smckusick 			if (run == len)
102965999Smckusick 				break;
103065999Smckusick 		}
103165999Smckusick 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
103265999Smckusick 			bit <<= 1;
103365999Smckusick 		} else {
103465999Smckusick 			map = *mapp++;
103565999Smckusick 			bit = 1;
103665999Smckusick 		}
103765999Smckusick 	}
103865999Smckusick 	if (i == cgp->cg_nclusterblks)
103965999Smckusick 		goto fail;
104065999Smckusick 	/*
104165999Smckusick 	 * Allocate the cluster that we have found.
104265999Smckusick 	 */
104365999Smckusick 	bno = cg * fs->fs_fpg + blkstofrags(fs, i - run + 1);
104465999Smckusick 	len = blkstofrags(fs, len);
104565999Smckusick 	for (i = 0; i < len; i += fs->fs_frag)
104665999Smckusick 		if (ffs_alloccgblk(fs, cgp, bno + i) != bno + i)
104765999Smckusick 			panic("ffs_clusteralloc: lost block");
104865999Smckusick 	brelse(bp);
104965999Smckusick 	return (bno);
105065999Smckusick 
105165999Smckusick fail:
105265999Smckusick 	brelse(bp);
105365999Smckusick 	return (0);
105465999Smckusick }
105565999Smckusick 
105665999Smckusick /*
10575375Smckusic  * Determine whether an inode can be allocated.
10585375Smckusic  *
10595375Smckusic  * Check to see if an inode is available, and if it is,
10605375Smckusic  * allocate it using the following policy:
10615375Smckusic  *   1) allocate the requested inode.
10625375Smckusic  *   2) allocate the next available inode after the requested
10635375Smckusic  *      inode in the specified cylinder group.
10645375Smckusic  */
1065*67869Smckusick static u_long
106664603Sbostic ffs_nodealloccg(ip, cg, ipref, mode)
10675965Smckusic 	struct inode *ip;
10684359Smckusick 	int cg;
10694359Smckusick 	daddr_t ipref;
10704359Smckusick 	int mode;
10714359Smckusick {
10725965Smckusic 	register struct fs *fs;
10734463Smckusic 	register struct cg *cgp;
107416784Smckusick 	struct buf *bp;
107537735Smckusick 	int error, start, len, loc, map, i;
10764359Smckusick 
10775965Smckusic 	fs = ip->i_fs;
10785322Smckusic 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
10796294Smckusick 		return (NULL);
108037735Smckusick 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
108138776Smckusick 		(int)fs->fs_cgsize, NOCRED, &bp);
108237735Smckusick 	if (error) {
108337735Smckusick 		brelse(bp);
108437735Smckusick 		return (NULL);
108537735Smckusick 	}
108664508Sbostic 	cgp = (struct cg *)bp->b_data;
108737735Smckusick 	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
10885960Smckusic 		brelse(bp);
10896294Smckusick 		return (NULL);
10905960Smckusic 	}
10918105Sroot 	cgp->cg_time = time.tv_sec;
10924359Smckusick 	if (ipref) {
10934359Smckusick 		ipref %= fs->fs_ipg;
109434143Smckusick 		if (isclr(cg_inosused(cgp), ipref))
10954359Smckusick 			goto gotit;
109616784Smckusick 	}
109716784Smckusick 	start = cgp->cg_irotor / NBBY;
109816784Smckusick 	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
109934143Smckusick 	loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
110016784Smckusick 	if (loc == 0) {
110117697Smckusick 		len = start + 1;
110217697Smckusick 		start = 0;
110334143Smckusick 		loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
110417697Smckusick 		if (loc == 0) {
110565718Sbostic 			printf("cg = %d, irotor = %d, fs = %s\n",
110617697Smckusick 			    cg, cgp->cg_irotor, fs->fs_fsmnt);
110764603Sbostic 			panic("ffs_nodealloccg: map corrupted");
110817697Smckusick 			/* NOTREACHED */
110917697Smckusick 		}
111016784Smckusick 	}
111116784Smckusick 	i = start + len - loc;
111234143Smckusick 	map = cg_inosused(cgp)[i];
111316784Smckusick 	ipref = i * NBBY;
111416784Smckusick 	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
111516784Smckusick 		if ((map & i) == 0) {
11164359Smckusick 			cgp->cg_irotor = ipref;
11174359Smckusick 			goto gotit;
11184359Smckusick 		}
11194359Smckusick 	}
112016784Smckusick 	printf("fs = %s\n", fs->fs_fsmnt);
112164603Sbostic 	panic("ffs_nodealloccg: block not in map");
112216784Smckusick 	/* NOTREACHED */
11234359Smckusick gotit:
112434143Smckusick 	setbit(cg_inosused(cgp), ipref);
11254792Smckusic 	cgp->cg_cs.cs_nifree--;
11264792Smckusic 	fs->fs_cstotal.cs_nifree--;
11275322Smckusic 	fs->fs_cs(fs, cg).cs_nifree--;
112850893Smckusick 	fs->fs_fmod = 1;
11294359Smckusick 	if ((mode & IFMT) == IFDIR) {
11304792Smckusic 		cgp->cg_cs.cs_ndir++;
11314792Smckusic 		fs->fs_cstotal.cs_ndir++;
11325322Smckusic 		fs->fs_cs(fs, cg).cs_ndir++;
11334359Smckusick 	}
11344359Smckusick 	bdwrite(bp);
11354359Smckusick 	return (cg * fs->fs_ipg + ipref);
11364359Smckusick }
11374359Smckusick 
11385375Smckusic /*
11395375Smckusic  * Free a block or fragment.
11405375Smckusic  *
11415375Smckusic  * The specified block or fragment is placed back in the
11425375Smckusic  * free map. If a fragment is deallocated, a possible
11435375Smckusic  * block reassembly is checked.
11445375Smckusic  */
114551469Sbostic ffs_blkfree(ip, bno, size)
11465965Smckusic 	register struct inode *ip;
11474359Smckusick 	daddr_t bno;
114853059Smckusick 	long size;
11494359Smckusick {
11504359Smckusick 	register struct fs *fs;
11514359Smckusick 	register struct cg *cgp;
115237735Smckusick 	struct buf *bp;
115365999Smckusick 	daddr_t blkno;
115465999Smckusick 	int i, error, cg, blk, frags, bbase;
11554359Smckusick 
11565965Smckusic 	fs = ip->i_fs;
115764508Sbostic 	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
11586716Smckusick 		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
11596716Smckusick 		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
116031402Smckusick 		panic("blkfree: bad size");
11616716Smckusick 	}
11625377Smckusic 	cg = dtog(fs, bno);
116364508Sbostic 	if ((u_int)bno >= fs->fs_size) {
11646567Smckusic 		printf("bad block %d, ino %d\n", bno, ip->i_number);
116553244Smckusick 		ffs_fserr(fs, ip->i_uid, "bad block");
11664359Smckusick 		return;
11676567Smckusic 	}
116837735Smckusick 	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
116938776Smckusick 		(int)fs->fs_cgsize, NOCRED, &bp);
117037735Smckusick 	if (error) {
117137735Smckusick 		brelse(bp);
117237735Smckusick 		return;
117337735Smckusick 	}
117464508Sbostic 	cgp = (struct cg *)bp->b_data;
117537735Smckusick 	if (!cg_chkmagic(cgp)) {
11765960Smckusic 		brelse(bp);
11774359Smckusick 		return;
11785960Smckusic 	}
11798105Sroot 	cgp->cg_time = time.tv_sec;
11805377Smckusic 	bno = dtogd(fs, bno);
11815322Smckusic 	if (size == fs->fs_bsize) {
118265999Smckusick 		blkno = fragstoblks(fs, bno);
118365999Smckusick 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
11846716Smckusick 			printf("dev = 0x%x, block = %d, fs = %s\n",
11856716Smckusick 			    ip->i_dev, bno, fs->fs_fsmnt);
118631402Smckusick 			panic("blkfree: freeing free block");
11876567Smckusic 		}
118865999Smckusick 		ffs_setblock(fs, cg_blksfree(cgp), blkno);
118965999Smckusick 		ffs_clusteracct(fs, cgp, blkno, 1);
11904792Smckusic 		cgp->cg_cs.cs_nbfree++;
11914792Smckusic 		fs->fs_cstotal.cs_nbfree++;
11925322Smckusic 		fs->fs_cs(fs, cg).cs_nbfree++;
11935375Smckusic 		i = cbtocylno(fs, bno);
119434143Smckusick 		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
119534143Smckusick 		cg_blktot(cgp)[i]++;
11964426Smckusic 	} else {
119717224Smckusick 		bbase = bno - fragnum(fs, bno);
11984463Smckusic 		/*
11994463Smckusic 		 * decrement the counts associated with the old frags
12004463Smckusic 		 */
120134143Smckusick 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
120251469Sbostic 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
12034463Smckusic 		/*
12044463Smckusic 		 * deallocate the fragment
12054463Smckusic 		 */
12065960Smckusic 		frags = numfrags(fs, size);
12074463Smckusic 		for (i = 0; i < frags; i++) {
120834143Smckusick 			if (isset(cg_blksfree(cgp), bno + i)) {
12096716Smckusick 				printf("dev = 0x%x, block = %d, fs = %s\n",
12106716Smckusick 				    ip->i_dev, bno + i, fs->fs_fsmnt);
121131402Smckusick 				panic("blkfree: freeing free frag");
12126716Smckusick 			}
121334143Smckusick 			setbit(cg_blksfree(cgp), bno + i);
12144426Smckusic 		}
12156294Smckusick 		cgp->cg_cs.cs_nffree += i;
12166294Smckusick 		fs->fs_cstotal.cs_nffree += i;
12176294Smckusick 		fs->fs_cs(fs, cg).cs_nffree += i;
12184463Smckusic 		/*
12194463Smckusic 		 * add back in counts associated with the new frags
12204463Smckusic 		 */
122134143Smckusick 		blk = blkmap(fs, cg_blksfree(cgp), bbase);
122251469Sbostic 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
12234463Smckusic 		/*
12244463Smckusic 		 * if a complete block has been reassembled, account for it
12254463Smckusic 		 */
122665999Smckusick 		blkno = fragstoblks(fs, bbase);
122765999Smckusick 		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
12285322Smckusic 			cgp->cg_cs.cs_nffree -= fs->fs_frag;
12295322Smckusic 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
12305322Smckusic 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
123165999Smckusick 			ffs_clusteracct(fs, cgp, blkno, 1);
12324792Smckusic 			cgp->cg_cs.cs_nbfree++;
12334792Smckusic 			fs->fs_cstotal.cs_nbfree++;
12345322Smckusic 			fs->fs_cs(fs, cg).cs_nbfree++;
12355375Smckusic 			i = cbtocylno(fs, bbase);
123634143Smckusick 			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
123734143Smckusick 			cg_blktot(cgp)[i]++;
12384426Smckusic 		}
12394426Smckusic 	}
124050893Smckusick 	fs->fs_fmod = 1;
12414359Smckusick 	bdwrite(bp);
12424359Smckusick }
12434359Smckusick 
12445375Smckusic /*
12455375Smckusic  * Free an inode.
12465375Smckusic  *
12475375Smckusic  * The specified inode is placed back in the free map.
12485375Smckusic  */
124953577Sheideman int
125054653Smckusick ffs_vfree(ap)
125154653Smckusick 	struct vop_vfree_args /* {
125254653Smckusick 		struct vnode *a_pvp;
125354653Smckusick 		ino_t a_ino;
125454653Smckusick 		int a_mode;
125554653Smckusick 	} */ *ap;
12564359Smckusick {
12574359Smckusick 	register struct fs *fs;
12584359Smckusick 	register struct cg *cgp;
125951541Smckusick 	register struct inode *pip;
126054653Smckusick 	ino_t ino = ap->a_ino;
126137735Smckusick 	struct buf *bp;
126237735Smckusick 	int error, cg;
12634359Smckusick 
126453583Sheideman 	pip = VTOI(ap->a_pvp);
126551469Sbostic 	fs = pip->i_fs;
126654653Smckusick 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
126754653Smckusick 		panic("ifree: range: dev = 0x%x, ino = %d, fs = %s\n",
126854653Smckusick 		    pip->i_dev, ino, fs->fs_fsmnt);
126964603Sbostic 	cg = ino_to_cg(fs, ino);
127051469Sbostic 	error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
127138776Smckusick 		(int)fs->fs_cgsize, NOCRED, &bp);
127237735Smckusick 	if (error) {
127337735Smckusick 		brelse(bp);
127453577Sheideman 		return (0);
127537735Smckusick 	}
127664508Sbostic 	cgp = (struct cg *)bp->b_data;
127737735Smckusick 	if (!cg_chkmagic(cgp)) {
12785960Smckusic 		brelse(bp);
127953577Sheideman 		return (0);
12805960Smckusic 	}
12818105Sroot 	cgp->cg_time = time.tv_sec;
128254653Smckusick 	ino %= fs->fs_ipg;
128354653Smckusick 	if (isclr(cg_inosused(cgp), ino)) {
128454653Smckusick 		printf("dev = 0x%x, ino = %d, fs = %s\n",
128554653Smckusick 		    pip->i_dev, ino, fs->fs_fsmnt);
128648057Smckusick 		if (fs->fs_ronly == 0)
128748057Smckusick 			panic("ifree: freeing free inode");
12886716Smckusick 	}
128954653Smckusick 	clrbit(cg_inosused(cgp), ino);
129054653Smckusick 	if (ino < cgp->cg_irotor)
129154653Smckusick 		cgp->cg_irotor = ino;
12924792Smckusic 	cgp->cg_cs.cs_nifree++;
12934792Smckusic 	fs->fs_cstotal.cs_nifree++;
12945322Smckusic 	fs->fs_cs(fs, cg).cs_nifree++;
129553583Sheideman 	if ((ap->a_mode & IFMT) == IFDIR) {
12964792Smckusic 		cgp->cg_cs.cs_ndir--;
12974792Smckusic 		fs->fs_cstotal.cs_ndir--;
12985322Smckusic 		fs->fs_cs(fs, cg).cs_ndir--;
12994359Smckusick 	}
130050893Smckusick 	fs->fs_fmod = 1;
13014359Smckusick 	bdwrite(bp);
130253577Sheideman 	return (0);
13034359Smckusick }
13044359Smckusick 
13054463Smckusic /*
13065375Smckusic  * Find a block of the specified size in the specified cylinder group.
13075375Smckusic  *
13084651Smckusic  * It is a panic if a request is made to find a block if none are
13094651Smckusic  * available.
13104651Smckusic  */
131151469Sbostic static daddr_t
131251469Sbostic ffs_mapsearch(fs, cgp, bpref, allocsiz)
13134651Smckusic 	register struct fs *fs;
13144651Smckusic 	register struct cg *cgp;
13154651Smckusic 	daddr_t bpref;
13164651Smckusic 	int allocsiz;
13174651Smckusic {
13184651Smckusic 	daddr_t bno;
13194651Smckusic 	int start, len, loc, i;
13204651Smckusic 	int blk, field, subfield, pos;
13214651Smckusic 
13224651Smckusic 	/*
13234651Smckusic 	 * find the fragment by searching through the free block
13244651Smckusic 	 * map for an appropriate bit pattern
13254651Smckusic 	 */
13264651Smckusic 	if (bpref)
13275377Smckusic 		start = dtogd(fs, bpref) / NBBY;
13284651Smckusic 	else
13294651Smckusic 		start = cgp->cg_frotor / NBBY;
13305398Smckusic 	len = howmany(fs->fs_fpg, NBBY) - start;
133164508Sbostic 	loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
133234476Smckusick 		(u_char *)fragtbl[fs->fs_frag],
133334476Smckusick 		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
13344651Smckusic 	if (loc == 0) {
13356531Smckusick 		len = start + 1;
13366531Smckusick 		start = 0;
133764508Sbostic 		loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
133834476Smckusick 			(u_char *)fragtbl[fs->fs_frag],
133934476Smckusick 			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
134016784Smckusick 		if (loc == 0) {
134116784Smckusick 			printf("start = %d, len = %d, fs = %s\n",
134216784Smckusick 			    start, len, fs->fs_fsmnt);
134351469Sbostic 			panic("ffs_alloccg: map corrupted");
134417697Smckusick 			/* NOTREACHED */
134516784Smckusick 		}
13464651Smckusic 	}
13474651Smckusic 	bno = (start + len - loc) * NBBY;
13484651Smckusic 	cgp->cg_frotor = bno;
13494651Smckusic 	/*
13504651Smckusic 	 * found the byte in the map
13514651Smckusic 	 * sift through the bits to find the selected frag
13524651Smckusic 	 */
13536294Smckusick 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
135434143Smckusick 		blk = blkmap(fs, cg_blksfree(cgp), bno);
13554651Smckusic 		blk <<= 1;
13564651Smckusic 		field = around[allocsiz];
13574651Smckusic 		subfield = inside[allocsiz];
13585322Smckusic 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
13596294Smckusick 			if ((blk & field) == subfield)
13606294Smckusick 				return (bno + pos);
13614651Smckusic 			field <<= 1;
13624651Smckusic 			subfield <<= 1;
13634651Smckusic 		}
13644651Smckusic 	}
13656716Smckusick 	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
136651469Sbostic 	panic("ffs_alloccg: block not in map");
13676531Smckusick 	return (-1);
13684651Smckusic }
13694651Smckusic 
13704651Smckusic /*
137165999Smckusick  * Update the cluster map because of an allocation or free.
137265999Smckusick  *
137365999Smckusick  * Cnt == 1 means free; cnt == -1 means allocating.
137465999Smckusick  */
137565999Smckusick ffs_clusteracct(fs, cgp, blkno, cnt)
137665999Smckusick 	struct fs *fs;
137765999Smckusick 	struct cg *cgp;
137865999Smckusick 	daddr_t blkno;
137965999Smckusick 	int cnt;
138065999Smckusick {
138165999Smckusick 	long *sump;
1382*67869Smckusick 	int32_t *lp;
138365999Smckusick 	u_char *freemapp, *mapp;
138465999Smckusick 	int i, start, end, forw, back, map, bit;
138565999Smckusick 
138665999Smckusick 	if (fs->fs_contigsumsize <= 0)
138765999Smckusick 		return;
138865999Smckusick 	freemapp = cg_clustersfree(cgp);
138965999Smckusick 	sump = cg_clustersum(cgp);
139065999Smckusick 	/*
139165999Smckusick 	 * Allocate or clear the actual block.
139265999Smckusick 	 */
139365999Smckusick 	if (cnt > 0)
139465999Smckusick 		setbit(freemapp, blkno);
139565999Smckusick 	else
139665999Smckusick 		clrbit(freemapp, blkno);
139765999Smckusick 	/*
139865999Smckusick 	 * Find the size of the cluster going forward.
139965999Smckusick 	 */
140065999Smckusick 	start = blkno + 1;
140165999Smckusick 	end = start + fs->fs_contigsumsize;
140265999Smckusick 	if (end >= cgp->cg_nclusterblks)
140365999Smckusick 		end = cgp->cg_nclusterblks;
140465999Smckusick 	mapp = &freemapp[start / NBBY];
140565999Smckusick 	map = *mapp++;
140665999Smckusick 	bit = 1 << (start % NBBY);
140765999Smckusick 	for (i = start; i < end; i++) {
140865999Smckusick 		if ((map & bit) == 0)
140965999Smckusick 			break;
141065999Smckusick 		if ((i & (NBBY - 1)) != (NBBY - 1)) {
141165999Smckusick 			bit <<= 1;
141265999Smckusick 		} else {
141365999Smckusick 			map = *mapp++;
141465999Smckusick 			bit = 1;
141565999Smckusick 		}
141665999Smckusick 	}
141765999Smckusick 	forw = i - start;
141865999Smckusick 	/*
141965999Smckusick 	 * Find the size of the cluster going backward.
142065999Smckusick 	 */
142165999Smckusick 	start = blkno - 1;
142265999Smckusick 	end = start - fs->fs_contigsumsize;
142365999Smckusick 	if (end < 0)
142465999Smckusick 		end = -1;
142565999Smckusick 	mapp = &freemapp[start / NBBY];
142665999Smckusick 	map = *mapp--;
142765999Smckusick 	bit = 1 << (start % NBBY);
142865999Smckusick 	for (i = start; i > end; i--) {
142965999Smckusick 		if ((map & bit) == 0)
143065999Smckusick 			break;
143165999Smckusick 		if ((i & (NBBY - 1)) != 0) {
143265999Smckusick 			bit >>= 1;
143365999Smckusick 		} else {
143465999Smckusick 			map = *mapp--;
143565999Smckusick 			bit = 1 << (NBBY - 1);
143665999Smckusick 		}
143765999Smckusick 	}
143865999Smckusick 	back = start - i;
143965999Smckusick 	/*
144065999Smckusick 	 * Account for old cluster and the possibly new forward and
144165999Smckusick 	 * back clusters.
144265999Smckusick 	 */
144365999Smckusick 	i = back + forw + 1;
144465999Smckusick 	if (i > fs->fs_contigsumsize)
144565999Smckusick 		i = fs->fs_contigsumsize;
144665999Smckusick 	sump[i] += cnt;
144765999Smckusick 	if (back > 0)
144865999Smckusick 		sump[back] -= cnt;
144965999Smckusick 	if (forw > 0)
145065999Smckusick 		sump[forw] -= cnt;
1451*67869Smckusick 	/*
1452*67869Smckusick 	 * Update cluster summary information.
1453*67869Smckusick 	 */
1454*67869Smckusick 	lp = &sump[fs->fs_contigsumsize];
1455*67869Smckusick 	for (i = fs->fs_contigsumsize; i > 0; i--)
1456*67869Smckusick 		if (*lp-- > 0)
1457*67869Smckusick 			break;
1458*67869Smckusick 	fs->fs_maxcluster[cgp->cg_cgx] = i;
145965999Smckusick }
146065999Smckusick 
146165999Smckusick /*
14625375Smckusic  * Fserr prints the name of a file system with an error diagnostic.
14635375Smckusic  *
14645375Smckusic  * The form of the error message is:
14654359Smckusick  *	fs: error message
14664359Smckusick  */
146751469Sbostic static void
146851469Sbostic ffs_fserr(fs, uid, cp)
14694359Smckusick 	struct fs *fs;
147051469Sbostic 	u_int uid;
14714359Smckusick 	char *cp;
14724359Smckusick {
14734359Smckusick 
147442318Smckusick 	log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp);
14754359Smckusick }
1476