ffs_alloc.c

早期freebsd实现

	int cg;
	daddr_t bpref;
	int len;
{
	register struct fs *fs;
	register struct cg *cgp;
	struct buf *bp;
	int i, run, bno, bit, map;
	u_char *mapp;

	fs = ip->i_fs;
	if (fs->fs_cs(fs, cg).cs_nbfree < len)
		return (NULL);
	if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
	    NOCRED, &bp))
		goto fail;
	cgp = (struct cg *)bp->b_data;
	if (!cg_chkmagic(cgp))
		goto fail;
	/*
	 * Check to see if a cluster of the needed size (or bigger) is
	 * available in this cylinder group.
	 */
	for (i = len; i <= fs->fs_contigsumsize; i++)
		if (cg_clustersum(cgp)[i] > 0)
			break;
	if (i > fs->fs_contigsumsize)
		goto fail;
	/*
	 * Search the cluster map to find a big enough cluster.
	 * We take the first one that we find, even if it is larger
	 * than we need as we prefer to get one close to the previous
	 * block allocation. We do not search before the current
	 * preference point as we do not want to allocate a block
	 * that is allocated before the previous one (as we will
	 * then have to wait for another pass of the elevator
	 * algorithm before it will be read). We prefer to fail and
	 * be recalled to try an allocation in the next cylinder group.
	 */
	if (dtog(fs, bpref) != cg)
		bpref = 0;
	else
		bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
	mapp = &cg_clustersfree(cgp)[bpref / NBBY];
	map = *mapp++;
	bit = 1 << (bpref % NBBY);
	for (run = 0, i = bpref; i < cgp->cg_nclusterblks; i++) {
		if ((map & bit) == 0) {
			run = 0;
		} else {
			run++;
			if (run == len)
				break;
		}
		if ((i & (NBBY - 1)) != (NBBY - 1)) {
			bit <<= 1;
		} else {
			map = *mapp++;
			bit = 1;
		}
	}
	if (i == cgp->cg_nclusterblks)
		goto fail;
	/*
	 * Allocate the cluster that we have found.
	 */
	bno = cg * fs->fs_fpg + blkstofrags(fs, i - run + 1);
	len = blkstofrags(fs, len);
	for (i = 0; i < len; i += fs->fs_frag)
		if (ffs_alloccgblk(fs, cgp, bno + i) != bno + i)
			panic("ffs_clusteralloc: lost block");
	brelse(bp);
	return (bno);

fail:
	brelse(bp);
	return (0);
}

/*
 * Determine whether an inode can be allocated.
 *
 * Check to see if an inode is available, and if it is,
 * allocate it using the following policy:
 *   1) allocate the requested inode.
 *   2) allocate the next available inode after the requested
 *      inode in the specified cylinder group.
 */
static ino_t
ffs_nodealloccg(ip, cg, ipref, mode)
	struct inode *ip;
	int cg;
	daddr_t ipref;
	int mode;
{
	register struct fs *fs;
	register struct cg *cgp;
	struct buf *bp;
	int error, start, len, loc, map, i;

	fs = ip->i_fs;
	if (fs->fs_cs(fs, cg).cs_nifree == 0)
		return (NULL);
	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
		(int)fs->fs_cgsize, NOCRED, &bp);
	if (error) {
		brelse(bp);
		return (NULL);
	}
	cgp = (struct cg *)bp->b_data;
	if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
		brelse(bp);
		return (NULL);
	}
	cgp->cg_time = time.tv_sec;
	if (ipref) {
		ipref %= fs->fs_ipg;
		if (isclr(cg_inosused(cgp), ipref))
			goto gotit;
	}
	start = cgp->cg_irotor / NBBY;
	len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
	loc = skpc(0xff, len, &cg_inosused(cgp)[start]);
	if (loc == 0) {
		len = start + 1;
		start = 0;
		loc = skpc(0xff, len, &cg_inosused(cgp)[0]);
		if (loc == 0) {
			printf("cg = %d, irotor = %d, fs = %s\n",
			    cg, cgp->cg_irotor, fs->fs_fsmnt);
			panic("ffs_nodealloccg: map corrupted");
			/* NOTREACHED */
		}
	}
	i = start + len - loc;
	map = cg_inosused(cgp)[i];
	ipref = i * NBBY;
	for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
		if ((map & i) == 0) {
			cgp->cg_irotor = ipref;
			goto gotit;
		}
	}
	printf("fs = %s\n", fs->fs_fsmnt);
	panic("ffs_nodealloccg: block not in map");
	/* NOTREACHED */
gotit:
	setbit(cg_inosused(cgp), ipref);
	cgp->cg_cs.cs_nifree--;
	fs->fs_cstotal.cs_nifree--;
	fs->fs_cs(fs, cg).cs_nifree--;
	fs->fs_fmod = 1;
	if ((mode & IFMT) == IFDIR) {
		cgp->cg_cs.cs_ndir++;
		fs->fs_cstotal.cs_ndir++;
		fs->fs_cs(fs, cg).cs_ndir++;
	}
	bdwrite(bp);
	return (cg * fs->fs_ipg + ipref);
}

/*
 * Free a block or fragment.
 *
 * The specified block or fragment is placed back in the
 * free map. If a fragment is deallocated, a possible 
 * block reassembly is checked.
 */
ffs_blkfree(ip, bno, size)
	register struct inode *ip;
	daddr_t bno;
	long size;
{
	register struct fs *fs;
	register struct cg *cgp;
	struct buf *bp;
	daddr_t blkno;
	int i, error, cg, blk, frags, bbase;

	fs = ip->i_fs;
	if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
		printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n",
		    ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
		panic("blkfree: bad size");
	}
	cg = dtog(fs, bno);
	if ((u_int)bno >= fs->fs_size) {
		printf("bad block %d, ino %d\n", bno, ip->i_number);
		ffs_fserr(fs, ip->i_uid, "bad block");
		return;
	}
	error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
		(int)fs->fs_cgsize, NOCRED, &bp);
	if (error) {
		brelse(bp);
		return;
	}
	cgp = (struct cg *)bp->b_data;
	if (!cg_chkmagic(cgp)) {
		brelse(bp);
		return;
	}
	cgp->cg_time = time.tv_sec;
	bno = dtogd(fs, bno);
	if (size == fs->fs_bsize) {
		blkno = fragstoblks(fs, bno);
		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
			printf("dev = 0x%x, block = %d, fs = %s\n",
			    ip->i_dev, bno, fs->fs_fsmnt);
			panic("blkfree: freeing free block");
		}
		ffs_setblock(fs, cg_blksfree(cgp), blkno);
		ffs_clusteracct(fs, cgp, blkno, 1);
		cgp->cg_cs.cs_nbfree++;
		fs->fs_cstotal.cs_nbfree++;
		fs->fs_cs(fs, cg).cs_nbfree++;
		i = cbtocylno(fs, bno);
		cg_blks(fs, cgp, i)[cbtorpos(fs, bno)]++;
		cg_blktot(cgp)[i]++;
	} else {
		bbase = bno - fragnum(fs, bno);
		/*
		 * decrement the counts associated with the old frags
		 */
		blk = blkmap(fs, cg_blksfree(cgp), bbase);
		ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
		/*
		 * deallocate the fragment
		 */
		frags = numfrags(fs, size);
		for (i = 0; i < frags; i++) {
			if (isset(cg_blksfree(cgp), bno + i)) {
				printf("dev = 0x%x, block = %d, fs = %s\n",
				    ip->i_dev, bno + i, fs->fs_fsmnt);
				panic("blkfree: freeing free frag");
			}
			setbit(cg_blksfree(cgp), bno + i);
		}
		cgp->cg_cs.cs_nffree += i;
		fs->fs_cstotal.cs_nffree += i;
		fs->fs_cs(fs, cg).cs_nffree += i;
		/*
		 * add back in counts associated with the new frags
		 */
		blk = blkmap(fs, cg_blksfree(cgp), bbase);
		ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
		/*
		 * if a complete block has been reassembled, account for it
		 */
		blkno = fragstoblks(fs, bbase);
		if (ffs_isblock(fs, cg_blksfree(cgp), blkno)) {
			cgp->cg_cs.cs_nffree -= fs->fs_frag;
			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
			ffs_clusteracct(fs, cgp, blkno, 1);
			cgp->cg_cs.cs_nbfree++;
			fs->fs_cstotal.cs_nbfree++;
			fs->fs_cs(fs, cg).cs_nbfree++;
			i = cbtocylno(fs, bbase);
			cg_blks(fs, cgp, i)[cbtorpos(fs, bbase)]++;
			cg_blktot(cgp)[i]++;
		}
	}
	fs->fs_fmod = 1;
	bdwrite(bp);
}

/*
 * Free an inode.
 *
 * The specified inode is placed back in the free map.
 */
int
ffs_vfree(ap)
	struct vop_vfree_args /* {
		struct vnode *a_pvp;
		ino_t a_ino;
		int a_mode;
	} */ *ap;
{
	register struct fs *fs;
	register struct cg *cgp;
	register struct inode *pip;
	ino_t ino = ap->a_ino;
	struct buf *bp;
	int error, cg;

	pip = VTOI(ap->a_pvp);
	fs = pip->i_fs;
	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
		panic("ifree: range: dev = 0x%x, ino = %d, fs = %s\n",
		    pip->i_dev, ino, fs->fs_fsmnt);
	cg = ino_to_cg(fs, ino);
	error = bread(pip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
		(int)fs->fs_cgsize, NOCRED, &bp);
	if (error) {
		brelse(bp);
		return (0);
	}
	cgp = (struct cg *)bp->b_data;
	if (!cg_chkmagic(cgp)) {
		brelse(bp);
		return (0);
	}
	cgp->cg_time = time.tv_sec;
	ino %= fs->fs_ipg;
	if (isclr(cg_inosused(cgp), ino)) {
		printf("dev = 0x%x, ino = %d, fs = %s\n",
		    pip->i_dev, ino, fs->fs_fsmnt);
		if (fs->fs_ronly == 0)
			panic("ifree: freeing free inode");
	}
	clrbit(cg_inosused(cgp), ino);
	if (ino < cgp->cg_irotor)
		cgp->cg_irotor = ino;
	cgp->cg_cs.cs_nifree++;
	fs->fs_cstotal.cs_nifree++;
	fs->fs_cs(fs, cg).cs_nifree++;
	if ((ap->a_mode & IFMT) == IFDIR) {
		cgp->cg_cs.cs_ndir--;
		fs->fs_cstotal.cs_ndir--;
		fs->fs_cs(fs, cg).cs_ndir--;
	}
	fs->fs_fmod = 1;
	bdwrite(bp);
	return (0);
}

/*
 * Find a block of the specified size in the specified cylinder group.
 *
 * It is a panic if a request is made to find a block if none are
 * available.
 */
static daddr_t
ffs_mapsearch(fs, cgp, bpref, allocsiz)
	register struct fs *fs;
	register struct cg *cgp;
	daddr_t bpref;
	int allocsiz;
{
	daddr_t bno;
	int start, len, loc, i;
	int blk, field, subfield, pos;

	/*
	 * find the fragment by searching through the free block
	 * map for an appropriate bit pattern
	 */
	if (bpref)
		start = dtogd(fs, bpref) / NBBY;
	else
		start = cgp->cg_frotor / NBBY;
	len = howmany(fs->fs_fpg, NBBY) - start;
	loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[start],
		(u_char *)fragtbl[fs->fs_frag],
		(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
	if (loc == 0) {
		len = start + 1;
		start = 0;
		loc = scanc((u_int)len, (u_char *)&cg_blksfree(cgp)[0],
			(u_char *)fragtbl[fs->fs_frag],
			(u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
		if (loc == 0) {
			printf("start = %d, len = %d, fs = %s\n",
			    start, len, fs->fs_fsmnt);
			panic("ffs_alloccg: map corrupted");
			/* NOTREACHED */
		}
	}
	bno = (start + len - loc) * NBBY;
	cgp->cg_frotor = bno;
	/*
	 * found the byte in the map
	 * sift through the bits to find the selected frag
	 */
	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
		blk = blkmap(fs, cg_blksfree(cgp), bno);
		blk <<= 1;
		field = around[allocsiz];
		subfield = inside[allocsiz];
		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
			if ((blk & field) == subfield)
				return (bno + pos);
			field <<= 1;
			subfield <<= 1;
		}
	}
	printf("bno = %d, fs = %s\n", bno, fs->fs_fsmnt);
	panic("ffs_alloccg: block not in map");
	return (-1);
}

/*
 * Update the cluster map because of an allocation or free.
 *
 * Cnt == 1 means free; cnt == -1 means allocating.
 */
ffs_clusteracct(fs, cgp, blkno, cnt)
	struct fs *fs;
	struct cg *cgp;
	daddr_t blkno;
	int cnt;
{
	long *sump;
	u_char *freemapp, *mapp;
	int i, start, end, forw, back, map, bit;

	if (fs->fs_contigsumsize <= 0)
		return;
	freemapp = cg_clustersfree(cgp);
	sump = cg_clustersum(cgp);
	/*
	 * Allocate or clear the actual block.
	 */
	if (cnt > 0)
		setbit(freemapp, blkno);
	else
		clrbit(freemapp, blkno);
	/*
	 * Find the size of the cluster going forward.
	 */
	start = blkno + 1;
	end = start + fs->fs_contigsumsize;
	if (end >= cgp->cg_nclusterblks)
		end = cgp->cg_nclusterblks;
	mapp = &freemapp[start / NBBY];
	map = *mapp++;
	bit = 1 << (start % NBBY);
	for (i = start; i < end; i++) {
		if ((map & bit) == 0)
			break;
		if ((i & (NBBY - 1)) != (NBBY - 1)) {
			bit <<= 1;
		} else {
			map = *mapp++;
			bit = 1;
		}
	}
	forw = i - start;
	/*
	 * Find the size of the cluster going backward.
	 */
	start = blkno - 1;
	end = start - fs->fs_contigsumsize;
	if (end < 0)
		end = -1;
	mapp = &freemapp[start / NBBY];
	map = *mapp--;
	bit = 1 << (start % NBBY);
	for (i = start; i > end; i--) {
		if ((map & bit) == 0)
			break;
		if ((i & (NBBY - 1)) != 0) {
			bit >>= 1;
		} else {
			map = *mapp--;
			bit = 1 << (NBBY - 1);
		}
	}
	back = start - i;
	/*
	 * Account for old cluster and the possibly new forward and
	 * back clusters.
	 */
	i = back + forw + 1;
	if (i > fs->fs_contigsumsize)
		i = fs->fs_contigsumsize;
	sump[i] += cnt;
	if (back > 0)
		sump[back] -= cnt;
	if (forw > 0)
		sump[forw] -= cnt;
}

/*
 * Fserr prints the name of a file system with an error diagnostic.
 * 
 * The form of the error message is:
 *	fs: error message
 */
static void
ffs_fserr(fs, uid, cp)
	struct fs *fs;
	u_int uid;
	char *cp;
{

	log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp);
}