fs: ext4: cache extent data

When a file contains extents, U-Boot currently reads extent-related data
for each block in the file, even if that data is located in the same
block each time. This significantly slows down loading of files that use
extents. Implement a very dumb cache to prevent repeatedly reading the
same block. Files with extents now load as fast as files without.

Note: There are many cases where read_allocated_block() is called. This
patch only addresses one of those places; all others still read redundant
data in any case they did before. This is a minimal patch to fix the
load command; other cases aren't fixed.

Signed-off-by: Stephen Warren <swarren@nvidia.com>

4 files changed, 88 insertions, 36 deletions

diff --git a/fs/ext4/ext4_common.c b/fs/ext4/ext4_common.c
index 67e2471bd3..29308e3b44 100644
--- a/fs/ext4/ext4_common.c
+++ b/fs/ext4/ext4_common.c
@@ -510,7 +510,8 @@ restart:
 
 restart_read:
 	/* read the block no allocated to a file */
-	first_block_no_of_root = read_allocated_block(g_parent_inode, blk_idx);
+	first_block_no_of_root = read_allocated_block(g_parent_inode, blk_idx,
+						      NULL);
 	if (first_block_no_of_root <= 0)
 		goto fail;
 
@@ -646,7 +647,7 @@ static int search_dir(struct ext2_inode *parent_inode, char *dirname)
 
 	/* get the block no allocated to a file */
 	for (blk_idx = 0; blk_idx < directory_blocks; blk_idx++) {
-		blknr = read_allocated_block(parent_inode, blk_idx);
+		blknr = read_allocated_block(parent_inode, blk_idx, NULL);
 		if (blknr <= 0)
 			goto fail;
 
@@ -943,7 +944,7 @@ int ext4fs_filename_unlink(char *filename)
 
 	/* read the block no allocated to a file */
 	for (blk_idx = 0; blk_idx < directory_blocks; blk_idx++) {
-		blknr = read_allocated_block(g_parent_inode, blk_idx);
+		blknr = read_allocated_block(g_parent_inode, blk_idx, NULL);
 		if (blknr <= 0)
 			break;
 		inodeno = unlink_filename(filename, blknr);
@@ -1522,7 +1523,7 @@ void ext4fs_allocate_blocks(struct ext2_inode *file_inode,
 #endif
 
 static struct ext4_extent_header *ext4fs_get_extent_block
-	(struct ext2_data *data, char *buf,
+	(struct ext2_data *data, struct ext_block_cache *cache,
 		struct ext4_extent_header *ext_block,
 		uint32_t fileblock, int log2_blksz)
 {
@@ -1551,12 +1552,10 @@ static struct ext4_extent_header *ext4fs_get_extent_block
 
 		block = le16_to_cpu(index[i].ei_leaf_hi);
 		block = (block << 32) + le32_to_cpu(index[i].ei_leaf_lo);
-
-		if (ext4fs_devread((lbaint_t)block << log2_blksz, 0, blksz,
-				   buf))
-			ext_block = (struct ext4_extent_header *)buf;
-		else
+		block <<= log2_blksz;
+		if (!ext_cache_read(cache, (lbaint_t)block, blksz))
 			return NULL;
+		ext_block = (struct ext4_extent_header *)cache->buf;
 	}
 }
 
@@ -1613,7 +1612,8 @@ int ext4fs_read_inode(struct ext2_data *data, int ino, struct ext2_inode *inode)
 	return 1;
 }
 
-long int read_allocated_block(struct ext2_inode *inode, int fileblock)
+long int read_allocated_block(struct ext2_inode *inode, int fileblock,
+			      struct ext_block_cache *cache)
 {
 	long int blknr;
 	int blksz;
@@ -1630,20 +1630,26 @@ long int read_allocated_block(struct ext2_inode *inode, int fileblock)
 
 	if (le32_to_cpu(inode->flags) & EXT4_EXTENTS_FL) {
 		long int startblock, endblock;
-		char *buf = zalloc(blksz);
-		if (!buf)
-			return -ENOMEM;
+		struct ext_block_cache *c, cd;
 		struct ext4_extent_header *ext_block;
 		struct ext4_extent *extent;
 		int i;
+
+		if (cache) {
+			c = cache;
+		} else {
+			c = &cd;
+			ext_cache_init(c);
+		}
 		ext_block =
-			ext4fs_get_extent_block(ext4fs_root, buf,
+			ext4fs_get_extent_block(ext4fs_root, c,
 						(struct ext4_extent_header *)
 						inode->b.blocks.dir_blocks,
 						fileblock, log2_blksz);
 		if (!ext_block) {
 			printf("invalid extent block\n");
-			free(buf);
+			if (!cache)
+				ext_cache_fini(c);
 			return -EINVAL;
 		}
 
@@ -1655,19 +1661,22 @@ long int read_allocated_block(struct ext2_inode *inode, int fileblock)
 
 			if (startblock > fileblock) {
 				/* Sparse file */
-				free(buf);
+				if (!cache)
+					ext_cache_fini(c);
 				return 0;
 
 			} else if (fileblock < endblock) {
 				start = le16_to_cpu(extent[i].ee_start_hi);
 				start = (start << 32) +
 					le32_to_cpu(extent[i].ee_start_lo);
-				free(buf);
+				if (!cache)
+					ext_cache_fini(c);
 				return (fileblock - startblock) + start;
 			}
 		}
 
-		free(buf);
+		if (!cache)
+			ext_cache_fini(c);
 		return 0;
 	}
 
diff --git a/fs/ext4/ext4_journal.c b/fs/ext4/ext4_journal.c
index 148593da7f..6adbab93a6 100644
--- a/fs/ext4/ext4_journal.c
+++ b/fs/ext4/ext4_journal.c
@@ -347,7 +347,7 @@ void recover_transaction(int prev_desc_logical_no)
 	ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO,
 			  (struct ext2_inode *)&inode_journal);
 	blknr = read_allocated_block((struct ext2_inode *)
-				     &inode_journal, i);
+				     &inode_journal, i, NULL);
 	ext4fs_devread((lbaint_t)blknr * fs->sect_perblk, 0, fs->blksz,
 		       temp_buff);
 	p_jdb = (char *)temp_buff;
@@ -372,7 +372,7 @@ void recover_transaction(int prev_desc_logical_no)
 				be32_to_cpu(jdb->h_sequence)) == 0)
 				continue;
 		}
-		blknr = read_allocated_block(&inode_journal, i);
+		blknr = read_allocated_block(&inode_journal, i, NULL);
 		ext4fs_devread((lbaint_t)blknr * fs->sect_perblk, 0,
 			       fs->blksz, metadata_buff);
 		put_ext4((uint64_t)((uint64_t)be32_to_cpu(tag->block) * (uint64_t)fs->blksz),
@@ -419,7 +419,8 @@ int ext4fs_check_journal_state(int recovery_flag)
 	}
 
 	ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO, &inode_journal);
-	blknr = read_allocated_block(&inode_journal, EXT2_JOURNAL_SUPERBLOCK);
+	blknr = read_allocated_block(&inode_journal, EXT2_JOURNAL_SUPERBLOCK,
+				     NULL);
 	ext4fs_devread((lbaint_t)blknr * fs->sect_perblk, 0, fs->blksz,
 		       temp_buff);
 	jsb = (struct journal_superblock_t *) temp_buff;
@@ -443,7 +444,7 @@ int ext4fs_check_journal_state(int recovery_flag)
 
 	i = be32_to_cpu(jsb->s_first);
 	while (1) {
-		blknr = read_allocated_block(&inode_journal, i);
+		blknr = read_allocated_block(&inode_journal, i, NULL);
 		memset(temp_buff1, '\0', fs->blksz);
 		ext4fs_devread((lbaint_t)blknr * fs->sect_perblk,
 			       0, fs->blksz, temp_buff1);
@@ -537,7 +538,7 @@ end:
 		ext4_read_superblock((char *)fs->sb);
 
 		blknr = read_allocated_block(&inode_journal,
-					 EXT2_JOURNAL_SUPERBLOCK);
+					 EXT2_JOURNAL_SUPERBLOCK, NULL);
 		put_ext4((uint64_t) ((uint64_t)blknr * (uint64_t)fs->blksz),
 			 (struct journal_superblock_t *)temp_buff,
 			 (uint32_t) fs->blksz);
@@ -566,7 +567,7 @@ static void update_descriptor_block(long int blknr)
 
 	ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO, &inode_journal);
 	jsb_blknr = read_allocated_block(&inode_journal,
-					 EXT2_JOURNAL_SUPERBLOCK);
+					 EXT2_JOURNAL_SUPERBLOCK, NULL);
 	ext4fs_devread((lbaint_t)jsb_blknr * fs->sect_perblk, 0, fs->blksz,
 		       temp_buff);
 	jsb = (struct journal_superblock_t *) temp_buff;
@@ -618,7 +619,7 @@ static void update_commit_block(long int blknr)
 	ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO,
 			  &inode_journal);
 	jsb_blknr = read_allocated_block(&inode_journal,
-					 EXT2_JOURNAL_SUPERBLOCK);
+					 EXT2_JOURNAL_SUPERBLOCK, NULL);
 	ext4fs_devread((lbaint_t)jsb_blknr * fs->sect_perblk, 0, fs->blksz,
 		       temp_buff);
 	jsb = (struct journal_superblock_t *) temp_buff;
@@ -645,16 +646,17 @@ void ext4fs_update_journal(void)
 	long int blknr;
 	int i;
 	ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO, &inode_journal);
-	blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++);
+	blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++, NULL);
 	update_descriptor_block(blknr);
 	for (i = 0; i < MAX_JOURNAL_ENTRIES; i++) {
 		if (journal_ptr[i]->blknr == -1)
 			break;
-		blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++);
+		blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++,
+					     NULL);
 		put_ext4((uint64_t) ((uint64_t)blknr * (uint64_t)fs->blksz),
 			 journal_ptr[i]->buf, fs->blksz);
 	}
-	blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++);
+	blknr = read_allocated_block(&inode_journal, jrnl_blk_idx++, NULL);
 	update_commit_block(blknr);
 	printf("update journal finished\n");
 }
diff --git a/fs/ext4/ext4_write.c b/fs/ext4/ext4_write.c
index 4eb77c327e..95ffa3dfad 100644
--- a/fs/ext4/ext4_write.c
+++ b/fs/ext4/ext4_write.c
@@ -479,7 +479,7 @@ static int ext4fs_delete_file(int inodeno)
 
 	/* release data blocks */
 	for (i = 0; i < no_blocks; i++) {
-		blknr = read_allocated_block(&inode, i);
+		blknr = read_allocated_block(&inode, i, NULL);
 		if (blknr == 0)
 			continue;
 		if (blknr < 0)
@@ -695,7 +695,7 @@ void ext4fs_deinit(void)
 		ext4fs_read_inode(ext4fs_root, EXT2_JOURNAL_INO,
 				  &inode_journal);
 		blknr = read_allocated_block(&inode_journal,
-					EXT2_JOURNAL_SUPERBLOCK);
+					EXT2_JOURNAL_SUPERBLOCK, NULL);
 		ext4fs_devread((lbaint_t)blknr * fs->sect_perblk, 0, fs->blksz,
 			       temp_buff);
 		jsb = (struct journal_superblock_t *)temp_buff;
@@ -776,7 +776,7 @@ static int ext4fs_write_file(struct ext2_inode *file_inode,
 		long int blknr;
 		int blockend = fs->blksz;
 		int skipfirst = 0;
-		blknr = read_allocated_block(file_inode, i);
+		blknr = read_allocated_block(file_inode, i, NULL);
 		if (blknr <= 0)
 			return -1;
 
diff --git a/fs/ext4/ext4fs.c b/fs/ext4/ext4fs.c
index 2a28031d14..26db677a1f 100644
--- a/fs/ext4/ext4fs.c
+++ b/fs/ext4/ext4fs.c
@@ -62,6 +62,9 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,
 	lbaint_t delayed_next = 0;
 	char *delayed_buf = NULL;
 	short status;
+	struct ext_block_cache cache;
+
+	ext_cache_init(&cache);
 
 	if (blocksize <= 0)
 		return -1;
@@ -77,9 +80,11 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,
 		int blockoff = pos - (blocksize * i);
 		int blockend = blocksize;
 		int skipfirst = 0;
-		blknr = read_allocated_block(&(node->inode), i);
-		if (blknr < 0)
+		blknr = read_allocated_block(&node->inode, i, &cache);
+		if (blknr < 0) {
+			ext_cache_fini(&cache);
 			return -1;
+		}
 
 		blknr = blknr << log2_fs_blocksize;
 
@@ -109,8 +114,10 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,
 							delayed_skipfirst,
 							delayed_extent,
 							delayed_buf);
-					if (status == 0)
+					if (status == 0) {
+						ext_cache_fini(&cache);
 						return -1;
+					}
 					previous_block_number = blknr;
 					delayed_start = blknr;
 					delayed_extent = blockend;
@@ -136,8 +143,10 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,
 							delayed_skipfirst,
 							delayed_extent,
 							delayed_buf);
-				if (status == 0)
+				if (status == 0) {
+					ext_cache_fini(&cache);
 					return -1;
+				}
 				previous_block_number = -1;
 			}
 			/* Zero no more than `len' bytes. */
@@ -153,12 +162,15 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,
 		status = ext4fs_devread(delayed_start,
 					delayed_skipfirst, delayed_extent,
 					delayed_buf);
-		if (status == 0)
+		if (status == 0) {
+			ext_cache_fini(&cache);
 			return -1;
+		}
 		previous_block_number = -1;
 	}
 
 	*actread  = len;
+	ext_cache_fini(&cache);
 	return 0;
 }
 
@@ -252,3 +264,32 @@ int ext4fs_uuid(char *uuid_str)
 	return -ENOSYS;
 #endif
 }
+
+void ext_cache_init(struct ext_block_cache *cache)
+{
+	memset(cache, 0, sizeof(*cache));
+}
+
+void ext_cache_fini(struct ext_block_cache *cache)
+{
+	free(cache->buf);
+	ext_cache_init(cache);
+}
+
+int ext_cache_read(struct ext_block_cache *cache, lbaint_t block, int size)
+{
+	/* This could be more lenient, but this is simple and enough for now */
+	if (cache->buf && cache->block == block && cache->size == size)
+		return 1;
+	ext_cache_fini(cache);
+	cache->buf = malloc(size);
+	if (!cache->buf)
+		return 0;
+	if (!ext4fs_devread(block, 0, size, cache->buf)) {
+		free(cache->buf);
+		return 0;
+	}
+	cache->block = block;
+	cache->size = size;
+	return 1;
+}