[GFS2] Reverse block order in build_height
The original code ordered the blocks allocated in the build_height
routine backwards causing excessive disk seeks during a read of the
metadata. This patch reverses the order to try and reduce disk seeks.
Example: A five level metadata tree, I = Inode, P = Pointers, D = Data
You need to read the blocks in the order:
I P5 P4 P3 P2 P1 D
in order to read a single data block. The new code now orders the blocks
in this way. The old code used to order them as:
I P1 P2 P3 P4 P5 D
requiring two extra seeks on average. Note that for files which are
grown by gradual extension rather than by truncate or by llseek/write
at a large offset, this doesn't apply. In the case of writing to a
file linearly, this routine will only be called upon to extend the
height of the tree by one block at a time, so the ordering is
determined by when its called rather than by the internals of the
routine itself. Optimising that part of the ordering is a much
harder problem.
Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 474b9a1..31c3e92 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -164,72 +164,62 @@
* @ip: The GFS2 inode
* @height: The height to build to
*
- * This routine makes sure that the metadata tree is tall enough to hold
- * "size" bytes of data.
*
* Returns: errno
*/
-static int build_height(struct gfs2_inode *ip, int height)
+static int build_height(struct inode *inode, unsigned height)
{
- struct gfs2_sbd *sdp = ip->i_sbd;
- struct buffer_head *bh, *dibh;
- uint64_t block = 0, *bp;
- unsigned int x;
- int new_block;
+ struct gfs2_inode *ip = inode->u.generic_ip;
+ unsigned new_height = height - ip->i_di.di_height;
+ struct buffer_head *dibh;
+ struct buffer_head *blocks[GFS2_MAX_META_HEIGHT];
int error;
+ u64 *bp;
+ u64 bn;
+ unsigned n;
- while (ip->i_di.di_height < height) {
- error = gfs2_meta_inode_buffer(ip, &dibh);
- if (error)
- return error;
+ if (height <= ip->i_di.di_height)
+ return 0;
- new_block = 0;
- bp = (uint64_t *)(dibh->b_data + sizeof(struct gfs2_dinode));
- for (x = 0; x < sdp->sd_diptrs; x++, bp++)
- if (*bp) {
- new_block = 1;
- break;
- }
+ error = gfs2_meta_inode_buffer(ip, &dibh);
+ if (error)
+ return error;
- if (new_block) {
- /* Get a new block, fill it with the old direct
- pointers, and write it out */
-
- block = gfs2_alloc_meta(ip);
-
- bh = gfs2_meta_new(ip->i_gl, block);
- gfs2_trans_add_bh(ip->i_gl, bh, 1);
- gfs2_metatype_set(bh,
- GFS2_METATYPE_IN,
- GFS2_FORMAT_IN);
- gfs2_buffer_copy_tail(bh,
- sizeof(struct gfs2_meta_header),
- dibh, sizeof(struct gfs2_dinode));
-
- brelse(bh);
- }
-
- /* Set up the new direct pointer and write it out to disk */
-
- gfs2_trans_add_bh(ip->i_gl, dibh, 1);
-
- gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
-
- if (new_block) {
- *(uint64_t *)(dibh->b_data +
- sizeof(struct gfs2_dinode)) =
- cpu_to_be64(block);
- ip->i_di.di_blocks++;
- }
-
- ip->i_di.di_height++;
-
- gfs2_dinode_out(&ip->i_di, dibh->b_data);
- brelse(dibh);
+ for(n = 0; n < new_height; n++) {
+ bn = gfs2_alloc_meta(ip);
+ blocks[n] = gfs2_meta_new(ip->i_gl, bn);
+ gfs2_trans_add_bh(ip->i_gl, blocks[n], 1);
}
-
- return 0;
+
+ n = 0;
+ bn = blocks[0]->b_blocknr;
+ if (new_height > 1) {
+ for(; n < new_height-1; n++) {
+ gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN,
+ GFS2_FORMAT_IN);
+ gfs2_buffer_clear_tail(blocks[n],
+ sizeof(struct gfs2_meta_header));
+ bp = (u64 *)(blocks[n]->b_data +
+ sizeof(struct gfs2_meta_header));
+ *bp = cpu_to_be64(blocks[n+1]->b_blocknr);
+ brelse(blocks[n]);
+ blocks[n] = NULL;
+ }
+ }
+ gfs2_metatype_set(blocks[n], GFS2_METATYPE_IN, GFS2_FORMAT_IN);
+ gfs2_buffer_copy_tail(blocks[n], sizeof(struct gfs2_meta_header),
+ dibh, sizeof(struct gfs2_dinode));
+ brelse(blocks[n]);
+ gfs2_trans_add_bh(ip->i_gl, dibh, 1);
+ gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
+ bp = (u64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
+ *bp = cpu_to_be64(bn);
+ ip->i_di.di_height += new_height;
+ ip->i_di.di_blocks += new_height;
+ gfs2_dinode_out(&ip->i_di, dibh->b_data);
+ brelse(dibh);
+ return error;
}
/**
@@ -416,7 +406,7 @@
if (!create)
goto out;
- error = build_height(ip, height);
+ error = build_height(inode, height);
if (error)
goto out;
}
@@ -806,7 +796,7 @@
h = calc_tree_height(ip, size);
if (ip->i_di.di_height < h) {
down_write(&ip->i_rw_mutex);
- error = build_height(ip, h);
+ error = build_height(ip->i_vnode, h);
up_write(&ip->i_rw_mutex);
if (error)
goto out_end_trans;