e2image (after enabling 64-bitness) was reporting a corrupt extent header
in a 16TiB file created on a 32TiB filesystem. Checking the on-disk extents
did not uncover any problem. Debugging e2image showed truncation in the
ext2fs_extent_get() routine. Code examination showed an inconsistency
on line 428:
blk = ext2fs_le32_to_cpu(ix->ei_leaf) +
((__u64) ext2fs_le16_to_cpu(ix->ei_leaf_hi) << 32);
blk is treated as a 64 bit quantity but it is declared blk_t. This
changes it to blk64_t. With this change, e2image has been running to
completion.
The second change is not based on debugging, just code examination:
ext2fs_alloc_block2() has been modified in the 64-bit patch series (in
the patch called add_64-bit_alloc_interface) so that its second
argument (the goal block) is a blk64_t. However, extent_node_split()
calls it with its second argument explicitly cast to blk_t. That looks
wrong.
There are some more blk_t's used in this file (extent.c), but they are
in #ifdef DEBUG code, so I have not examined them any closer and I
have not changed them (yet?).
Signed-off-by: Nick Dokos <[email protected]>
---
lib/ext2fs/extent.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/lib/ext2fs/extent.c b/lib/ext2fs/extent.c
index a6edd0c..1778955 100644
--- a/lib/ext2fs/extent.c
+++ b/lib/ext2fs/extent.c
@@ -273,7 +273,7 @@ errcode_t ext2fs_extent_get(ext2_extent_handle_t handle,
struct ext3_extent_idx *ix = 0;
struct ext3_extent *ex;
errcode_t retval;
- blk_t blk;
+ blk64_t blk;
blk64_t end_blk;
int orig_op, op;
@@ -904,7 +904,7 @@ static errcode_t extent_node_split(ext2_extent_handle_t handle)
goal_blk = (group * handle->fs->super->s_blocks_per_group) +
handle->fs->super->s_first_data_block;
}
- retval = ext2fs_alloc_block2(handle->fs, (blk_t) goal_blk, block_buf,
+ retval = ext2fs_alloc_block2(handle->fs, goal_blk, block_buf,
&new_node_pblk);
if (retval)
goto done;
--
1.6.0.6
On Wed, Apr 08, 2009 at 05:46:56PM -0400, Nick Dokos wrote:
> e2image (after enabling 64-bitness) was reporting a corrupt extent header
> in a 16TiB file created on a 32TiB filesystem. Checking the on-disk extents
> did not uncover any problem. Debugging e2image showed truncation in the
> ext2fs_extent_get() routine. Code examination showed an inconsistency
> on line 428:
>
> blk = ext2fs_le32_to_cpu(ix->ei_leaf) +
> ((__u64) ext2fs_le16_to_cpu(ix->ei_leaf_hi) << 32);
>
> blk is treated as a 64 bit quantity but it is declared blk_t. This
> changes it to blk64_t. With this change, e2image has been running to
> completion.
>
> The second change is not based on debugging, just code examination:
> ext2fs_alloc_block2() has been modified in the 64-bit patch series (in
> the patch called add_64-bit_alloc_interface) so that its second
> argument (the goal block) is a blk64_t. However, extent_node_split()
> calls it with its second argument explicitly cast to blk_t. That looks
> wrong.
>
> There are some more blk_t's used in this file (extent.c), but they are
> in #ifdef DEBUG code, so I have not examined them any closer and I
> have not changed them (yet?).
>
> Signed-off-by: Nick Dokos <[email protected]>
Signed-off-by: Valerie Aurora (Henson) <[email protected]>
> ---
> lib/ext2fs/extent.c | 4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/lib/ext2fs/extent.c b/lib/ext2fs/extent.c
> index a6edd0c..1778955 100644
> --- a/lib/ext2fs/extent.c
> +++ b/lib/ext2fs/extent.c
> @@ -273,7 +273,7 @@ errcode_t ext2fs_extent_get(ext2_extent_handle_t handle,
> struct ext3_extent_idx *ix = 0;
> struct ext3_extent *ex;
> errcode_t retval;
> - blk_t blk;
> + blk64_t blk;
> blk64_t end_blk;
> int orig_op, op;
>
> @@ -904,7 +904,7 @@ static errcode_t extent_node_split(ext2_extent_handle_t handle)
> goal_blk = (group * handle->fs->super->s_blocks_per_group) +
> handle->fs->super->s_first_data_block;
> }
> - retval = ext2fs_alloc_block2(handle->fs, (blk_t) goal_blk, block_buf,
> + retval = ext2fs_alloc_block2(handle->fs, goal_blk, block_buf,
> &new_node_pblk);
> if (retval)
> goto done;
> --
> 1.6.0.6
>