[PATCH 1/6][64-bit] Eliminate erroneous blk_t casts in ext2fs_get_free_blocks2().

[PATCH 1/6][64-bit] Eliminate erroneous blk_t casts in ext2fs_get_free_blocks2().

[Nick Dokos]

Running e2fsck -n -f <device> on a brand-new, never-mounted 32Tib fs
was producing checksum errors on half the groups (the lower half)
before it got to pass1 and block or inode bitmap conflicts on some
groups in the upper half. The checksum errors were actually caused by
the conflicts: e2fsck does a sanity check on block allocations, finds
a conflict, and uses the backup superblock at 32768. But when it does
that, it assumes the worst: it clears the UNINIT FLAGS and zeroes the
free inode count for each group, assuming that it will fix them up
during the run. That triggers the checksum errors (although it's not
clear to me why only the lower half of the groups get checksum
errors.)

dumpe2fs showed that the conflicts were not artifacts of e2fsck
processing: they existed on disk. In fact, the block bitmap of a group
conflicted with the inode bitmap of the group fifteen steps further
along. That smelled like flex_bg, so I remade the fs with flex_bg
off and then *every* group in the upper half (more precisely, with
blocks above the 32-bit boundary) had a conflict with itself: the
block bitmap and inode bitmap were allocated on top of each other.
That led to ext2fs_get_free_blocks2() which was truncating 64-bit block
numbers by casting them to blk_t.

This patch eliminates the casts. The resulting file system is free of
the conflicts (with or without flex_bg.) However, it does not fsck
cleanly yet: there are block bitmap differences  in the very last group.

Signed-off-by: Nick Dokos <nicholas.dokos@hp.com>
---
 lib/ext2fs/alloc.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/lib/ext2fs/alloc.c b/lib/ext2fs/alloc.c
index b1d1f9c..9f11673 100644
--- a/lib/ext2fs/alloc.c
+++ b/lib/ext2fs/alloc.c
@@ -272,8 +272,7 @@ errcode_t ext2fs_get_free_blocks2(ext2_filsys fs, blk64_t start, blk64_t finish,
 	do {
 		if (b+num-1 > ext2fs_blocks_count(fs->super))
 			b = fs->super->s_first_data_block;
-		if (ext2fs_fast_test_block_bitmap_range2(map, (blk_t) b,
-							(blk_t) num)) {
+		if (ext2fs_fast_test_block_bitmap_range2(map, b, num)) {
 			*ret = b;
 			return 0;
 		}
-- 
1.6.0.6

Re: [PATCH 1/6][64-bit] Eliminate erroneous blk_t casts in ext2fs_get_free_blocks2().

[Valerie Aurora]

On Fri, May 01, 2009 at 04:46:13AM -0400, Nick Dokos wrote:
> Running e2fsck -n -f <device> on a brand-new, never-mounted 32Tib fs
> was producing checksum errors on half the groups (the lower half)
> before it got to pass1 and block or inode bitmap conflicts on some
> groups in the upper half. The checksum errors were actually caused by
> the conflicts: e2fsck does a sanity check on block allocations, finds
> a conflict, and uses the backup superblock at 32768. But when it does
> that, it assumes the worst: it clears the UNINIT FLAGS and zeroes the
> free inode count for each group, assuming that it will fix them up
> during the run. That triggers the checksum errors (although it's not
> clear to me why only the lower half of the groups get checksum
> errors.)
> 
> dumpe2fs showed that the conflicts were not artifacts of e2fsck
> processing: they existed on disk. In fact, the block bitmap of a group
> conflicted with the inode bitmap of the group fifteen steps further
> along. That smelled like flex_bg, so I remade the fs with flex_bg
> off and then *every* group in the upper half (more precisely, with
> blocks above the 32-bit boundary) had a conflict with itself: the
> block bitmap and inode bitmap were allocated on top of each other.
> That led to ext2fs_get_free_blocks2() which was truncating 64-bit block
> numbers by casting them to blk_t.
> 
> This patch eliminates the casts. The resulting file system is free of
> the conflicts (with or without flex_bg.) However, it does not fsck
> cleanly yet: there are block bitmap differences  in the very last group.
> 
> Signed-off-by: Nick Dokos <nicholas.dokos@hp.com>
> ---
>  lib/ext2fs/alloc.c |    3 +--
>  1 files changed, 1 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/ext2fs/alloc.c b/lib/ext2fs/alloc.c
> index b1d1f9c..9f11673 100644
> --- a/lib/ext2fs/alloc.c
> +++ b/lib/ext2fs/alloc.c
> @@ -272,8 +272,7 @@ errcode_t ext2fs_get_free_blocks2(ext2_filsys fs, blk64_t start, blk64_t finish,
>  	do {
>  		if (b+num-1 > ext2fs_blocks_count(fs->super))
>  			b = fs->super->s_first_data_block;
> -		if (ext2fs_fast_test_block_bitmap_range2(map, (blk_t) b,
> -							(blk_t) num)) {
> +		if (ext2fs_fast_test_block_bitmap_range2(map, b, num)) {
>  			*ret = b;
>  			return 0;
>  		}
> -- 
> 1.6.0.6

Signed-off-by: Valerie Aurora (Henson) <vaurora@redhat.com>

-VAL