GFS2: Add hints to directory leaf blocks

This patch adds four new fields to directory leaf blocks. The intent is not to use them in the kernel itself, although perhaps we may be able to use them as hints at some later date, but instead to provide more information for debug/fsck use. One new field adds a pointer to the inode to which the leaf belongs. This can be useful if the pointer to the leaf block has become corrupt, as it will allow us to know which inode this block should be associated with. This field is set when the leaf is created and never changed over its lifetime. The second field is a "distance from the hash table" field. The meaning is as follows: 0 = An old leaf in which this value has not been set 1 = This leaf is pointed to directly from the hash table 2+ = This leaf is part of a chain, pointed to by another leaf block, the value gives the position in the chain. The third and fourth fields combine to give a time stamp of the most recent directory insertion or deletion from this leaf block. The time stamp is not updated when a new leaf block is chained from the current one. The code is currently written such that the timestamp on the dir inode will match that of the leaf block for the most recent insertion/deletion. For backwards compatibility, any of these new fields which is zero should be considered to be "unknown".

Steven Whitehouse committed 12 years ago

Commit 01bcb0d1 parent 22b5a6c2 files · +47 −4Browse files

Unified · Split

2 files

fs/gfs2/dir.c +37 −3
include/uapi/linux/gfs2_ondisk.h +10 −1

fs/gfs2/dir.c+37−3

@@ -834,6 +834,7 @@ static struct gfs2_leaf new_leaf(struct inode inode, struct buffer_head **pbh, · 833 unmodified lines
834	834	struct gfs2_leaf *leaf;
835	835	struct gfs2_dirent *dent;
836	836	struct qstr name = { .name = "" };
	837	+ struct timespec tv = CURRENT_TIME;
837	838
838	839	error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL);
839	840	if (error)
@@ -850,7 +851,11 @@ static struct gfs2_leaf new_leaf(struct inode inode, struct buffer_head **pbh, · 10 unmodified lines
850	851	leaf->lf_entries = 0;
851	852	leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
852	853	leaf->lf_next = 0;
853		− memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved));
	854	+ leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
	855	+ leaf->lf_dist = cpu_to_be32(1);
	856	+ leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
	857	+ leaf->lf_sec = cpu_to_be64(tv.tv_sec);
	858	+ memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
854	859	dent = (struct gfs2_dirent *)(leaf+1);
855	860	gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent);
856	861	*pbh = bh;
@@ -1612,11 +1617,31 @@ out: · 755 unmodified lines
1612	1617	return ret;
1613	1618	}
1614	1619
	1620	+/**
	1621	+ * dir_new_leaf - Add a new leaf onto hash chain
	1622	+ * @inode: The directory
	1623	+ * @name: The name we are adding
	1624	+ *
	1625	+ * This adds a new dir leaf onto an existing leaf when there is not
	1626	+ * enough space to add a new dir entry. This is a last resort after
	1627	+ * we've expanded the hash table to max size and also split existing
	1628	+ * leaf blocks, so it will only occur for very large directories.
	1629	+ *
	1630	+ * The dist parameter is set to 1 for leaf blocks directly attached
	1631	+ * to the hash table, 2 for one layer of indirection, 3 for two layers
	1632	+ * etc. We are thus able to tell the difference between an old leaf
	1633	+ * with dist set to zero (i.e. "don't know") and a new one where we
	1634	+ * set this information for debug/fsck purposes.
	1635	+ *
	1636	+ * Returns: 0 on success, or -ve on error
	1637	+ */
	1638	+
1615	1639	static int dir_new_leaf(struct inode inode, const struct qstr name)
1616	1640	{
1617	1641	struct buffer_head bh, obh;
1618	1642	struct gfs2_inode *ip = GFS2_I(inode);
1619	1643	struct gfs2_leaf leaf, oleaf;
	1644	+ u32 dist = 1;
1620	1645	int error;
1621	1646	u32 index;
1622	1647	u64 bn;
@@ -1626,6 +1651,7 @@ static int dir_new_leaf(struct inode inode, const struct qstr name) · 3 unmodified lines
1626	1651	if (error)
1627	1652	return error;
1628	1653	do {
	1654	+ dist++;
1629	1655	oleaf = (struct gfs2_leaf *)obh->b_data;
1630	1656	bn = be64_to_cpu(oleaf->lf_next);
1631	1657	if (!bn)
@@ -1643,6 +1669,7 @@ static int dir_new_leaf(struct inode inode, const struct qstr name) · 11 unmodified lines
1643	1669	brelse(obh);
1644	1670	return -ENOSPC;
1645	1671	}
	1672	+ leaf->lf_dist = cpu_to_be32(dist);
1646	1673	oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
1647	1674	brelse(bh);
1648	1675	brelse(obh);
@@ -1679,6 +1706,7 @@ int gfs2_dir_add(struct inode inode, const struct qstr name, · 30 unmodified lines
1679	1706	struct gfs2_inode *ip = GFS2_I(inode);
1680	1707	struct buffer_head *bh = da->bh;
1681	1708	struct gfs2_dirent *dent = da->dent;
	1709	+ struct timespec tv;
1682	1710	struct gfs2_leaf *leaf;
1683	1711	int error;
1684	1712
@@ -1693,15 +1721,18 @@ int gfs2_dir_add(struct inode inode, const struct qstr name, · 8 unmodified lines
1693	1721	dent = gfs2_init_dirent(inode, dent, name, bh);
1694	1722	gfs2_inum_out(nip, dent);
1695	1723	dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
	1724	+ tv = CURRENT_TIME;
1696	1725	if (ip->i_diskflags & GFS2_DIF_EXHASH) {
1697	1726	leaf = (struct gfs2_leaf *)bh->b_data;
1698	1727	be16_add_cpu(&leaf->lf_entries, 1);
	1728	+ leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
	1729	+ leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1699	1730	}
1700	1731	da->dent = NULL;
1701	1732	da->bh = NULL;
1702	1733	brelse(bh);
1703	1734	ip->i_entries++;
1704		− ip->i_inode.i_mtime = ip->i_inode.i_ctime = CURRENT_TIME;
	1735	+ ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv;
1705	1736	if (S_ISDIR(nip->i_inode.i_mode))
1706	1737	inc_nlink(&ip->i_inode);
1707	1738	mark_inode_dirty(inode);
@@ -1752,6 +1783,7 @@ int gfs2_dir_del(struct gfs2_inode dip, const struct dentry dentry) · 44 unmodified lines
1752	1783	const struct qstr *name = &dentry->d_name;
1753	1784	struct gfs2_dirent dent, prev = NULL;
1754	1785	struct buffer_head *bh;
	1786	+ struct timespec tv = CURRENT_TIME;
1755	1787
1756	1788	/* Returns _either_ the entry (if its first in block) or the
1757	1789	previous entry otherwise */
@@ -1777,13 +1809,15 @@ int gfs2_dir_del(struct gfs2_inode dip, const struct dentry dentry) · 19 unmodified lines
1777	1809	if (!entries)
1778	1810	gfs2_consist_inode(dip);
1779	1811	leaf->lf_entries = cpu_to_be16(--entries);
	1812	+ leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
	1813	+ leaf->lf_sec = cpu_to_be64(tv.tv_sec);
1780	1814	}
1781	1815	brelse(bh);
1782	1816
1783	1817	if (!dip->i_entries)
1784	1818	gfs2_consist_inode(dip);
1785	1819	dip->i_entries--;
1786		− dip->i_inode.i_mtime = dip->i_inode.i_ctime = CURRENT_TIME;
	1820	+ dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv;
1787	1821	if (S_ISDIR(dentry->d_inode->i_mode))
1788	1822	drop_nlink(&dip->i_inode);
1789	1823	mark_inode_dirty(&dip->i_inode);

include/uapi/linux/gfs2_ondisk.h+10−1

@@ -319,7 +319,16 @@ struct gfs2_leaf { · 318 unmodified lines
319	319	__be32 lf_dirent_format; /* Format of the dirents */
320	320	__be64 lf_next; /* Next leaf, if overflow */
321	321
322		− __u8 lf_reserved[64];
	322	+ union {
	323	+ __u8 lf_reserved[64];
	324	+ struct {
	325	+ __be64 lf_inode; /* Dir inode number */
	326	+ __be32 lf_dist; /* Dist from inode on chain */
	327	+ __be32 lf_nsec; /* Last ins/del usecs */
	328	+ __be64 lf_sec; /* Last ins/del in secs */
	329	+ __u8 lf_reserved2[40];
	330	+ };
	331	+ };
323	332	};
324	333
325	334	/*