ext3-extents-multiblock-directio-2.6.5-suse.patch

非常经典的一个分布式系统

Index: linux-2.6.5-7.283-full/include/linux/ext3_fs.h
===================================================================
--- linux-2.6.5-7.283-full.orig/include/linux/ext3_fs.h	2007-03-28 02:13:37.000000000 +0400
+++ linux-2.6.5-7.283-full/include/linux/ext3_fs.h	2007-03-28 02:21:37.000000000 +0400
@@ -815,7 +815,7 @@ extern struct inode_operations ext3_fast
 
 /* extents.c */
 extern int ext3_ext_writepage_trans_blocks(struct inode *, int);
-extern int ext3_ext_get_block(handle_t *, struct inode *, long,
+extern int ext3_ext_get_block(handle_t *, struct inode *, long, int,
 			      struct buffer_head *, int, int);
 extern void ext3_ext_truncate(struct inode *, struct page *);
 extern void ext3_ext_init(struct super_block *);
Index: linux-2.6.5-7.283-full/fs/ext3/extents.c
===================================================================
--- linux-2.6.5-7.283-full.orig/fs/ext3/extents.c	2007-03-28 02:14:25.000000000 +0400
+++ linux-2.6.5-7.283-full/fs/ext3/extents.c	2007-03-28 02:21:37.000000000 +0400
@@ -2024,7 +2024,8 @@ void ext3_init_tree_desc(struct ext3_ext
 }
 
 int ext3_ext_get_block(handle_t *handle, struct inode *inode,
-		       long iblock, struct buffer_head *bh_result,
+		       long iblock, int max_blocks, 
+		       struct buffer_head *bh_result,
 		       int create, int extend_disksize)
 {
 	struct ext3_ext_path *path = NULL;
@@ -2032,6 +2033,11 @@ int ext3_ext_get_block(handle_t *handle,
 	struct ext3_extent *ex;
 	int goal, newblock, err = 0, depth;
 	struct ext3_extents_tree tree;
+	unsigned long next;
+	int allocated = 0;
+
+	/* until we have multiblock allocation */
+	max_blocks = 1;
 
 	__clear_bit(BH_New, &bh_result->b_state);
 	ext3_init_tree_desc(&tree, inode);
@@ -2051,6 +2057,9 @@ int ext3_ext_get_block(handle_t *handle,
 		} else if (goal == EXT3_EXT_CACHE_EXTENT) {
 			/* block is already allocated */
 			newblock = iblock - newex.ee_block + newex.ee_start;
+			/* number of remaining blocks in the extent */
+			EXT_ASSERT(iblock >= newex.ee_block);
+			allocated = newex.ee_len - (iblock - newex.ee_block);
 			goto out;
 		} else {
 			EXT_ASSERT(0);
@@ -2078,6 +2087,8 @@ int ext3_ext_get_block(handle_t *handle,
 		/* if found exent covers block, simple return it */
 		if (iblock >= ex->ee_block && iblock < ex->ee_block + ex->ee_len) {
 			newblock = iblock - ex->ee_block + ex->ee_start;
+			/* number of remaining blocks in the extent */
+			allocated = ex->ee_len - (iblock - ex->ee_block);
 			ext_debug(&tree, "%d fit into %d:%d -> %d\n",
 				  (int) iblock, ex->ee_block, ex->ee_len,
 				  newblock);
@@ -2098,6 +2109,15 @@ int ext3_ext_get_block(handle_t *handle,
 		goto out2;
 	}
 
+	/* find next allocated block so that we know how many
+	 * blocks we can allocate without ovelapping next extent */
+	EXT_ASSERT(iblock >= ex->ee_block + ex->ee_len);
+	next = ext3_ext_next_allocated_block(path);
+	EXT_ASSERT(next > iblock);
+	allocated = next - iblock;
+	if (allocated > max_blocks)
+		allocated = max_blocks;
+
 	/* allocate new block */
 	goal = ext3_ext_find_goal(inode, path, iblock);
 	newblock = ext3_new_block(handle, inode, goal, &err);
@@ -2112,8 +2132,11 @@ int ext3_ext_get_block(handle_t *handle,
 	newex.ee_start_hi = 0;
 	newex.ee_len = 1;
 	err = ext3_ext_insert_extent(handle, &tree, path, &newex);
-	if (err)
+	if (err) {
+		/* free data blocks we just allocated */
+		ext3_free_blocks(handle, inode, newex.ee_start, newex.ee_len);
 		goto out2;
+	}
 	
 	if (extend_disksize && inode->i_size > EXT3_I(inode)->i_disksize)
 		EXT3_I(inode)->i_disksize = inode->i_size;
@@ -2125,10 +2148,13 @@ int ext3_ext_get_block(handle_t *handle,
 	ext3_ext_put_in_cache(&tree, newex.ee_block, newex.ee_len,
 			      newex.ee_start, EXT3_EXT_CACHE_EXTENT);
 out:
+	if (allocated > max_blocks)
+		allocated = max_blocks;
 	ext3_ext_show_leaf(&tree, path);
 	__set_bit(BH_Mapped, &bh_result->b_state);
 	bh_result->b_bdev = inode->i_sb->s_bdev;
 	bh_result->b_blocknr = newblock;
+	bh_result->b_size = (allocated << inode->i_blkbits);
 out2:
 	if (path) {
 		ext3_ext_drop_refs(path);
Index: linux-2.6.5-7.283-full/fs/ext3/inode.c
===================================================================
--- linux-2.6.5-7.283-full.orig/fs/ext3/inode.c	2007-03-28 02:13:37.000000000 +0400
+++ linux-2.6.5-7.283-full/fs/ext3/inode.c	2007-03-28 02:50:19.000000000 +0400
@@ -800,13 +800,17 @@ changed:
 
 static inline int
 ext3_get_block_wrap(handle_t *handle, struct inode *inode, long block,
-		    struct buffer_head *bh, int create, int extend_disksize)
+			int max_blocks, struct buffer_head *bh, int create,
+			int extend_disksize)
 {
+	int ret;
 	if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL)
-		return ext3_ext_get_block(handle, inode, block, bh, create,
-					  extend_disksize);
-	return ext3_get_block_handle(handle, inode, block, bh, create,
+		return ext3_ext_get_block(handle, inode, block, max_blocks,
+					bh, create, extend_disksize);
+	ret = ext3_get_block_handle(handle, inode, block, bh, create,
 				     extend_disksize);
+	bh->b_size = (1 << inode->i_blkbits);
+	return ret;
 }
 
 static int ext3_get_block(struct inode *inode, sector_t iblock,
@@ -819,7 +823,7 @@ static int ext3_get_block(struct inode *
 		handle = ext3_journal_current_handle();
 		J_ASSERT(handle != 0);
 	}
-	ret = ext3_get_block_wrap(handle, inode, iblock,
+	ret = ext3_get_block_wrap(handle, inode, iblock, 1,
 				  bh_result, create, 1);
 	return ret;
 }
@@ -847,10 +851,8 @@ ext3_direct_io_get_blocks(struct inode *
 		}
 	}
 	if (ret == 0)
-		ret = ext3_get_block_wrap(handle, inode, iblock,
+		ret = ext3_get_block_wrap(handle, inode, iblock, max_blocks,
 					  bh_result, create, 0);
-	if (ret == 0)
-		bh_result->b_size = (1 << inode->i_blkbits);
 	return ret;
 }
 
@@ -869,7 +871,7 @@ struct buffer_head *ext3_getblk(handle_t
 	dummy.b_state = 0;
 	dummy.b_blocknr = -1000;
 	buffer_trace_init(&dummy.b_history);
-	*errp = ext3_get_block_wrap(handle, inode, block, &dummy, create, 1);
+	*errp = ext3_get_block_wrap(handle, inode, block, 1, &dummy, create, 1);
 	if (!*errp && buffer_mapped(&dummy)) {
 		struct buffer_head *bh;
 		bh = sb_getblk(inode->i_sb, dummy.b_blocknr);