2009-04-18 06:05:00

by Nick Dokos

[permalink] [raw]
Subject: [PATCH 2/2][64-BIT] ext2fs_group_first_block2(): 32-bit truncation.

ext2fs_group_first_block2() returns the product of the group number
and the number of blocks per group (from the superblock). Unfortunately,
both of these are 32-bit quantities, so the multiplication result is
also 32 bits wide. It is then returned as a 64-bit quantity, but by then,
it's too late.

Cast one of the operands to blk64_t, so the multiplication will be done
in 64 bits.

e2fsck was complaining about a group that was marked BLOCK_UNINIT, but
had blocks in use (it turned out that a different group had blocks in
use, but the block numbers of the two different groups differed by
2^32, so this bug conflated them). With this change, this complaint
goes away. In addition, dumpe2fs produces the right blocks for all the
groups, whereas it was wrapping them at the 32-bit boundary previously.

Signed-off-by: Nick Dokos <[email protected]>
---
lib/ext2fs/blknum.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/lib/ext2fs/blknum.c b/lib/ext2fs/blknum.c
index b9666fb..fd56d53 100644
--- a/lib/ext2fs/blknum.c
+++ b/lib/ext2fs/blknum.c
@@ -28,7 +28,7 @@ dgrp_t ext2fs_group_of_blk2(ext2_filsys fs, blk64_t blk)
blk64_t ext2fs_group_first_block2(ext2_filsys fs, dgrp_t group)
{
return fs->super->s_first_data_block +
- (group * fs->super->s_blocks_per_group);
+ ((blk64_t)group * fs->super->s_blocks_per_group);
}

/*
--
1.6.0.6



2009-04-18 13:09:48

by Eric Sandeen

[permalink] [raw]
Subject: Re: [PATCH 2/2][64-BIT] ext2fs_group_first_block2(): 32-bit truncation.

Nick Dokos wrote:
> ext2fs_group_first_block2() returns the product of the group number
> and the number of blocks per group (from the superblock). Unfortunately,
> both of these are 32-bit quantities, so the multiplication result is
> also 32 bits wide. It is then returned as a 64-bit quantity, but by then,
> it's too late.
>
> Cast one of the operands to blk64_t, so the multiplication will be done
> in 64 bits.
>
> e2fsck was complaining about a group that was marked BLOCK_UNINIT, but
> had blocks in use (it turned out that a different group had blocks in
> use, but the block numbers of the two different groups differed by
> 2^32, so this bug conflated them). With this change, this complaint
> goes away. In addition, dumpe2fs produces the right blocks for all the
> groups, whereas it was wrapping them at the 32-bit boundary previously.
>
> Signed-off-by: Nick Dokos <[email protected]>

This one looks obviously correct, thanks.

Reviewed-by: Eric Sandeen <[email protected]>

> ---
> lib/ext2fs/blknum.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/lib/ext2fs/blknum.c b/lib/ext2fs/blknum.c
> index b9666fb..fd56d53 100644
> --- a/lib/ext2fs/blknum.c
> +++ b/lib/ext2fs/blknum.c
> @@ -28,7 +28,7 @@ dgrp_t ext2fs_group_of_blk2(ext2_filsys fs, blk64_t blk)
> blk64_t ext2fs_group_first_block2(ext2_filsys fs, dgrp_t group)
> {
> return fs->super->s_first_data_block +
> - (group * fs->super->s_blocks_per_group);
> + ((blk64_t)group * fs->super->s_blocks_per_group);
> }
>
> /*


2009-04-21 00:50:02

by Valerie Aurora

[permalink] [raw]
Subject: Re: [PATCH 2/2][64-BIT] ext2fs_group_first_block2(): 32-bit truncation.

On Sat, Apr 18, 2009 at 02:05:13AM -0400, Nick Dokos wrote:
> ext2fs_group_first_block2() returns the product of the group number
> and the number of blocks per group (from the superblock). Unfortunately,
> both of these are 32-bit quantities, so the multiplication result is
> also 32 bits wide. It is then returned as a 64-bit quantity, but by then,
> it's too late.
>
> Cast one of the operands to blk64_t, so the multiplication will be done
> in 64 bits.
>
> e2fsck was complaining about a group that was marked BLOCK_UNINIT, but
> had blocks in use (it turned out that a different group had blocks in
> use, but the block numbers of the two different groups differed by
> 2^32, so this bug conflated them). With this change, this complaint
> goes away. In addition, dumpe2fs produces the right blocks for all the
> groups, whereas it was wrapping them at the 32-bit boundary previously.
>
> Signed-off-by: Nick Dokos <[email protected]>

Signed-off-by: Valerie Aurora (Henson) <[email protected]>

Pulled and pushed.

-VAL

> ---
> lib/ext2fs/blknum.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/lib/ext2fs/blknum.c b/lib/ext2fs/blknum.c
> index b9666fb..fd56d53 100644
> --- a/lib/ext2fs/blknum.c
> +++ b/lib/ext2fs/blknum.c
> @@ -28,7 +28,7 @@ dgrp_t ext2fs_group_of_blk2(ext2_filsys fs, blk64_t blk)
> blk64_t ext2fs_group_first_block2(ext2_filsys fs, dgrp_t group)
> {
> return fs->super->s_first_data_block +
> - (group * fs->super->s_blocks_per_group);
> + ((blk64_t)group * fs->super->s_blocks_per_group);
> }
>
> /*
> --
> 1.6.0.6
>