Fix kernel BUG at fs/ext4/mballoc.c:910!
From: Valerie Clement <[email protected]>
With the flex_bg feature enabled, a large file creation oopses the
kernel.
The BUG_ON is:
BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
As the allocation of the bitmaps and the inode table can be done
outside the block group with flex_bg, this allows to allocate up to
EXT4_BLOCKS_PER_GROUP blocks in a group.
Depending on the group size and the block size, extents might be
larger than BLOCKS_PER_GROUP(); use EXT_INIT_MAX_LEN instead of
BLOCKS_PER_GROUP().
This patch fixes the oops.
Signed-off-by: Valerie Clement <[email protected]>
---
fs/ext4/mballoc.c | 5 ++++-
1 files changed, 4 insertions(+), 1 deletions(-)
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index b0f84b4..ccc33e9 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -34,6 +34,7 @@
#include <linux/pagemap.h>
#include <linux/seq_file.h>
#include <linux/version.h>
+#include <linux/ext4_fs_extents.h>
#include "group.h"
/*
@@ -907,7 +908,9 @@ static void ext4_mb_mark_free_simple(struct super_block *sb,
unsigned short chunk;
unsigned short border;
- BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
+ BUG_ON(len > EXT4_HAS_INCOMPAT_FEATURE(sb,
+ EXT4_FEATURE_INCOMPAT_FLEX_BG) ? EXT_INIT_MAX_LEN :
+ EXT4_BLOCKS_PER_GROUP(sb));
border = 2 << sb->s_blocksize_bits;
Valerie Clement wrote:
> Fix kernel BUG at fs/ext4/mballoc.c:910!
>
> From: Valerie Clement <[email protected]>
>
> With the flex_bg feature enabled, a large file creation oopses the
> kernel.
Valerie, what's the specific testcase for this?
Thanks,
-Eric
On Thu, 2008-02-14 at 17:35 +0100, Valerie Clement wrote:
> Fix kernel BUG at fs/ext4/mballoc.c:910!
>
> From: Valerie Clement <[email protected]>
>
> With the flex_bg feature enabled, a large file creation oopses the
> kernel.
> The BUG_ON is:
> BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
>
> As the allocation of the bitmaps and the inode table can be done
> outside the block group with flex_bg, this allows to allocate up to
> EXT4_BLOCKS_PER_GROUP blocks in a group.
> Depending on the group size and the block size, extents might be
> larger than BLOCKS_PER_GROUP(); use EXT_INIT_MAX_LEN instead of
> BLOCKS_PER_GROUP().
>
Yep, I think using EXT_INIT_MAX_LEN is better.
Will add to the patch queue shortly.
Thanks,
Mingming
> This patch fixes the oops.
>
> Signed-off-by: Valerie Clement <[email protected]>
> ---
>
> fs/ext4/mballoc.c | 5 ++++-
> 1 files changed, 4 insertions(+), 1 deletions(-)
>
> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
> index b0f84b4..ccc33e9 100644
> --- a/fs/ext4/mballoc.c
> +++ b/fs/ext4/mballoc.c
> @@ -34,6 +34,7 @@
> #include <linux/pagemap.h>
> #include <linux/seq_file.h>
> #include <linux/version.h>
> +#include <linux/ext4_fs_extents.h>
> #include "group.h"
>
> /*
> @@ -907,7 +908,9 @@ static void ext4_mb_mark_free_simple(struct super_block *sb,
> unsigned short chunk;
> unsigned short border;
>
> - BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
> + BUG_ON(len > EXT4_HAS_INCOMPAT_FEATURE(sb,
> + EXT4_FEATURE_INCOMPAT_FLEX_BG) ? EXT_INIT_MAX_LEN :
> + EXT4_BLOCKS_PER_GROUP(sb));
>
> border = 2 << sb->s_blocksize_bits;
>
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Feb 14, 2008 17:35 +0100, Valerie Clement wrote:
> From: Valerie Clement <[email protected]>
>
> With the flex_bg feature enabled, a large file creation oopses the
> kernel.
> The BUG_ON is:
> BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
>
> As the allocation of the bitmaps and the inode table can be done
> outside the block group with flex_bg, this allows to allocate up to
> EXT4_BLOCKS_PER_GROUP blocks in a group.
> Depending on the group size and the block size, extents might be
> larger than BLOCKS_PER_GROUP(); use EXT_INIT_MAX_LEN instead of
> BLOCKS_PER_GROUP().
In fact, my earlier review of this patch was incorrect, and Aneesh pointed
out the correct answer. The ext4_mb_mark_free_simple() function is only
called from ext4_mb_generate_buddy() to generate the buddy bitmap from the
on-disk block bitmap, and in that case the @len parameter should always
be <= EXT4_BLOCKS_PER_GROUP(). I think the original patch was correct.
Sorry about the confusion. I thought at first glance this was for
freeing the blocks from releasing an extent, but that is incorrect.
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
On Thu, 2008-02-14 at 15:13 -0700, Andreas Dilger wrote:
> On Feb 14, 2008 17:35 +0100, Valerie Clement wrote:
> > From: Valerie Clement <[email protected]>
> >
> > With the flex_bg feature enabled, a large file creation oopses the
> > kernel.
> > The BUG_ON is:
> > BUG_ON(len >= EXT4_BLOCKS_PER_GROUP(sb));
> >
> > As the allocation of the bitmaps and the inode table can be done
> > outside the block group with flex_bg, this allows to allocate up to
> > EXT4_BLOCKS_PER_GROUP blocks in a group.
> > Depending on the group size and the block size, extents might be
> > larger than BLOCKS_PER_GROUP(); use EXT_INIT_MAX_LEN instead of
> > BLOCKS_PER_GROUP().
>
> In fact, my earlier review of this patch was incorrect, and Aneesh pointed
> out the correct answer. The ext4_mb_mark_free_simple() function is only
> called from ext4_mb_generate_buddy() to generate the buddy bitmap from the
> on-disk block bitmap, and in that case the @len parameter should always
> be <= EXT4_BLOCKS_PER_GROUP(). I think the original patch was correct.
>
> Sorry about the confusion. I thought at first glance this was for
> freeing the blocks from releasing an extent, but that is incorrect.
>
Ok, I added the revised first patch
http://repo.or.cz/w/ext4-patch-queue.git?a=blob;f=ext4_fix_BUGON_at_mballoc.patch;h=c2d11bd2ec46c8103d4fad7c62cd792b7daec202;hb=067598f1a058a05342afee9720a0538b871c61f4
Mingming
> Cheers, Andreas
> --
> Andreas Dilger
> Sr. Staff Engineer, Lustre Group
> Sun Microsystems of Canada, Inc.
>
Eric Sandeen wrote:
> Valerie Clement wrote:
>> Fix kernel BUG at fs/ext4/mballoc.c:910!
>>
>> From: Valerie Clement <[email protected]>
>>
>> With the flex_bg feature enabled, a large file creation oopses the
>> kernel.
>
> Valerie, what's the specific testcase for this?
>
> Thanks,
> -Eric
I just did a simple dd,
dd if=/dev/zero of=/mnt/test/foo bs=1M count=8k
on an ext4 filesystem with the uninit_groups, lazy_bg and flex_bg
features enabled and the default mount options.
The difficulty is to build e2fsprogs programs that integrate all
these new features together.
Valerie
Valerie Clement wrote:
> Eric Sandeen wrote:
>> Valerie Clement wrote:
>>> Fix kernel BUG at fs/ext4/mballoc.c:910!
>>>
>>> From: Valerie Clement <[email protected]>
>>>
>>> With the flex_bg feature enabled, a large file creation oopses the
>>> kernel.
>> Valerie, what's the specific testcase for this?
>>
>> Thanks,
>> -Eric
> I just did a simple dd,
> dd if=/dev/zero of=/mnt/test/foo bs=1M count=8k
> on an ext4 filesystem with the uninit_groups, lazy_bg and flex_bg
> features enabled and the default mount options.
Ok, thanks Valerie. It's just always nice to know exactly how to
reproduce a problem that a patch fixes, IMHO.
-Eric