On Fri, Oct 27, 2017 at 04:18:11PM +0200, Christophe Ronco wrote:

Extent index blocks are read many times when reading completely a big file. This is time costly if underlying driver is slow. Caching the first index blocks cost a bit of RAM but speed up file reading a lot.

---

fs/ext4/ext4_common.c | 93 ++++++++++++++++++++++++++++++++++++++++++++++++--- fs/ext4/ext4_common.h | 16 +++++++++ fs/ext4/ext4fs.c | 16 +++++++-- 3 files changed, 117 insertions(+), 8 deletions(-)

diff --git a/fs/ext4/ext4_common.c b/fs/ext4/ext4_common.c index cab5465..68ab2e9 100644 --- a/fs/ext4/ext4_common.c +++ b/fs/ext4/ext4_common.c @@ -44,6 +44,7 @@ int ext4fs_indir3_size; int ext4fs_indir3_blkno = -1; struct ext2_inode *g_parent_inode; static int symlinknest; +struct ext4_extent_blocs ext4fs_extents_blocs;

#if defined(CONFIG_EXT4_WRITE) uint32_t ext4fs_div_roundup(uint32_t size, uint32_t n) @@ -1413,6 +1414,7 @@ static struct ext4_extent_header *ext4fs_get_extent_block unsigned long long block; int blksz = EXT2_BLOCK_SIZE(data); int i;

• int found;

  while (1) { index = (struct ext4_extent_idx *)(ext_block + 1);

@@ -1435,11 +1437,24 @@ 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
  

• 	return 0;
  

• /* look in saved extent blocks */
  

• found = 0;
  

• for (i = 0; i < ext4fs_extents_blocs.bloc_nb; i++) {
  

• 	if (ext4fs_extents_blocs.ext_blocs[i].block == block) {
  

• 		memcpy(buf, ext4fs_extents_blocs.ext_blocs[i].ext_header,
  

• 		       blksz);
  

• 		ext_block = (struct ext4_extent_header *)buf;
  

• 		found = 1;
  

• 	}
  

• }
  

• if (found == 0) {
  

• 	if (ext4fs_devread((lbaint_t)block << log2_blksz, 0,
  

• 			   blksz, buf))
  

• 		ext_block = (struct ext4_extent_header *)buf;
  

• 	else
  

• 		return 0;
  

• }
  

  }

}

@@ -1465,6 +1480,74 @@ static int ext4fs_blockgroup (char *)blkgrp); }

+void ext4fs_init_extents_blocks_global(void) +{

• memset(&ext4fs_extents_blocs, 0, sizeof(struct ext4_extent_blocs));

+}

+int ext4fs_read_extents_blocks(struct ext2_inode *inode) +{

• int blksz = EXT2_BLOCK_SIZE(ext4fs_root);
• int log2_blksz;
• struct ext4_extent_header *ext_block;
• struct ext4_extent_idx *index;
• unsigned long long block;
• int entries;
• char *buf;
• int i;
• ext4fs_extents_blocs.bloc_nb = 0;
• log2_blksz = LOG2_BLOCK_SIZE(ext4fs_root)

• - get_fs()->dev_desc->log2blksz;
  

• if ((le32_to_cpu(inode->flags) & EXT4_EXTENTS_FL) == 0)

• return 0;
  

• ext_block = (struct ext4_extent_header *)inode->b.blocks.dir_blocks;
• if (le16_to_cpu(ext_block->eh_magic) != EXT4_EXT_MAGIC)

• return 0;
  

• if (ext_block->eh_depth == 0)

• return 0;
  

• entries = le16_to_cpu(ext_block->eh_entries);
• index = (struct ext4_extent_idx *)(ext_block + 1);
• for (i = 0; (i < entries) && (i < MAX_EXTENT_BLOCS); i++) {

• /* bloc number */
  

• block = le16_to_cpu(index[i].ei_leaf_hi);
  

• block = (block << 32) + le32_to_cpu(index[i].ei_leaf_lo);
  

• /* bloc data */
  

• buf = zalloc(blksz);
  

• if (!buf)
  

• 	break;
  

• if (ext4fs_devread((lbaint_t)block << log2_blksz, 0, blksz,
  

• 		   buf) == 0) {
  

• 	free(buf);
  

• 	break;
  

• }
  

• ext4fs_extents_blocs.ext_blocs[i].block = block;
  

• ext4fs_extents_blocs.ext_blocs[i].ext_header =
  

• 			(struct ext4_extent_header *)buf;
  

• }
• ext4fs_extents_blocs.bloc_nb = i;
• return i;

+}

+void ext4fs_free_extents_blocks(void) +{

• int i;
• for (i = 0; i < ext4fs_extents_blocs.bloc_nb; i++) {

• if (ext4fs_extents_blocs.ext_blocs[i].ext_header) {
  

• 	free(ext4fs_extents_blocs.ext_blocs[i].ext_header);
  

• 	ext4fs_extents_blocs.ext_blocs[i].ext_header = NULL;
  

• 	ext4fs_extents_blocs.ext_blocs[i].block = 0;
  

• }
  

• }
• ext4fs_extents_blocs.bloc_nb = 0;

+}

int ext4fs_read_inode(struct ext2_data *data, int ino, struct ext2_inode *inode) { struct ext2_block_group blkgrp; diff --git a/fs/ext4/ext4_common.h b/fs/ext4/ext4_common.h index 48fd2ac..0f80856 100644 --- a/fs/ext4/ext4_common.h +++ b/fs/ext4/ext4_common.h @@ -41,6 +41,18 @@ #define SUPERBLOCK_SIZE 1024 #define F_FILE 1

+#define MAX_EXTENT_BLOCS 10

+struct ext4_extent_bloc {

• unsigned long long block;
• struct ext4_extent_header *ext_header;

+};

+struct ext4_extent_blocs {

• int bloc_nb;
• struct ext4_extent_bloc ext_blocs[MAX_EXTENT_BLOCS];

+};

static inline void *zalloc(size_t size) { void *p = memalign(ARCH_DMA_MINALIGN, size); @@ -57,6 +69,10 @@ int ext4fs_find_file(const char *path, struct ext2fs_node *rootnode, int ext4fs_iterate_dir(struct ext2fs_node *dir, char *name, struct ext2fs_node **fnode, int *ftype);

+void ext4fs_init_extents_blocks_global(void); +int ext4fs_read_extents_blocks(struct ext2_inode *inode); +void ext4fs_free_extents_blocks(void);

#if defined(CONFIG_EXT4_WRITE) uint32_t ext4fs_div_roundup(uint32_t size, uint32_t n); int ext4fs_checksum_update(unsigned int i); diff --git a/fs/ext4/ext4fs.c b/fs/ext4/ext4fs.c index 258b937..88db1fb 100644 --- a/fs/ext4/ext4fs.c +++ b/fs/ext4/ext4fs.c @@ -70,6 +70,12 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos,

blockcnt = lldiv(((len + pos) + blocksize - 1), blocksize);

• ext4fs_init_extents_blocks_global();
• if (blockcnt > 100) {

• /* Big reading: read extent blocks tree if any */
  

• ext4fs_read_extents_blocks(&(node->inode));
  

• }
• for (i = lldiv(pos, blocksize); i < blockcnt; i++) { lbaint_t blknr; int blockoff = pos - (blocksize * i);

@@ -77,7 +83,7 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos, int skipfirst = 0; blknr = read_allocated_block(&(node->inode), i); if (blknr < 0)

• 	return -1;
  

• 	goto error;
  
  

  blknr = blknr << log2_fs_blocksize;

@@ -134,7 +140,7 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos, delayed_extent, delayed_buf); if (status == 0)

• 			return -1;
  

• 			goto error;
  	previous_block_number = -1;
  }
  memset(buf, 0, blocksize - skipfirst);
  

@@ -147,12 +153,16 @@ int ext4fs_read_file(struct ext2fs_node *node, loff_t pos, delayed_skipfirst, delayed_extent, delayed_buf); if (status == 0)

• 	return -1;
  

• 	goto error;
  

  previous_block_number = -1; }

  *actread = len;

• ext4fs_free_extents_blocks(); return 0;

+error:

• ext4fs_free_extents_blocks();
• return -1;

}

int ext4fs_ls(const char *dirname)

Is this still needed with: commit ecdfb4195b20eb2dcde3c4083170016c13c69e8b Author: Ian Ray ian.ray@ge.com Date: Wed Nov 8 15:35:10 2017 +0000

ext4: recover from filesystem corruption when reading

Some fixes when reading EXT files and directory entries were identified after using e2fuzz to corrupt an EXT3 filesystem:

- Stop reading directory entries if the offset becomes badly aligned.

- Avoid overwriting memory by clamping the length used to zero the buffer in ext4fs_read_file. Also sanity check blocksize.

Signed-off-by: Ian Ray ian.ray@ge.com Signed-off-by: Martyn Welch martyn.welch@collabora.co.uk Reviewed-by: Stefano Babic sbabic@denx.de

Applied? Thanks!