gfs2 endianness bug: be16 assigned to be32 field
PATCH 1/1 gfs2 endianness bug: be16 assigned to be32 field
--- fs/gfs2/dir.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
fs/gfs2/dir.c +1 -1
@@ -815,7 +815,7 @@ static struct gfs2_leaf *new_leaf(struct leaf = (struct gfs2_leaf *)bh->b_data; leaf->lf_depth =cpu_to_be16 (depth); leaf->lf_entries =0 ; - leaf->lf_dirent_format =cpu_to_be16 (GFS2_FORMAT_DE ); + leaf->lf_dirent_format =cpu_to_be32 (GFS2_FORMAT_DE ); leaf->lf_next =0 ;memset (leaf->lf_reserved,0 ,sizeof (leaf->lf_reserved)); dent = (struct gfs2_dirent *)(leaf+1 );
Signature · 1 line
1.4.2.GIT
fs/gfs2/dir.c static struct gfs2_leaf *new_leaf(struct · line 818
Quoted 3 lines · Al Viro
fs/gfs2/dir.c +1
-1- leaf->lf_dirent_format =cpu_to_be16 (GFS2_FORMAT_DE ); + leaf->lf_dirent_format =cpu_to_be32 (GFS2_FORMAT_DE );
Hmm. Doesn't this change the on-disk format on a LE machine (eg x86)?
In other words, this change makes me nervous. A quick grep seems to indicate that nothing actually _uses_ this field, so maybe we don't really care, but I think we should double-check that this is what the GFS2 people really want.
If we don't want to change the format on a LE machine, then maybe the gfs2_leaf structure should be changed to be
.. __be16 lf_dirent_format; __be16 lf_unused; ..
which should keep the bits in the same position on LE.
Regardless, the old code was clearly wrong, since it gives different on-disk format for a big-endian and a little-endian machine. Al's fix is proper, but perhaps people would prefer something that breaks the BE format rather than the LE format. Hmm?
Steven?
Linus
Hi,
Quoted 16 lines · Linus Torvalds
fs/gfs2/dir.c static struct gfs2_leaf *new_leaf(struct · line 818
Quoted 3 lines · Al Viro
fs/gfs2/dir.c +1
-1- leaf->lf_dirent_format =cpu_to_be16 (GFS2_FORMAT_DE ); + leaf->lf_dirent_format =cpu_to_be32 (GFS2_FORMAT_DE );Hmm. Doesn't this change the on-disk format on a LE machine (eg x86)?
In other words, this change makes me nervous. A quick grep seems to indicate that nothing actually _uses_ this field, so maybe we don't really care, but I think we should double-check that this is what the GFS2 people really want.
If we don't want to change the format on a LE machine, then maybe the
gfs2_leafstructure should be changed to be
Quoted 7 lines
.. __be16 lf_dirent_format; __be16 lf_unused; ..which should keep the bits in the same position on LE.
Its a tricky question, but I think on balance that Al's proposed fix is the right one...
Quoted 8 lines
Regardless, the old code was clearly wrong, since it gives different on-disk format for a big-endian and a little-endian machine. Al's fix is proper, but perhaps people would prefer something that breaks the BE format rather than the LE format. Hmm?
Steven?
Linus
As you say, this field isn't actually used by anything. Its whole reason for existing is to be forwards compatible with GFS1. So I'd be inclined to accept the patch as proposed in order that as few filesystems as possible will have the incorrect format. We'll make a note of this problem and ensure that fsck can fix up the error and if we need to use this field in future, then we'll take into account that there may be some filesystems with the incorrect format number.
I think its fairly unlikely that we will need to use it in the near future though... Its original purpose was to allow future changes in the metadata format, but this is also allowed for by the mh_format field in the common metadata header, so I'm not entirely sure why a second format field was included in this structure (the directory leaf header).
Steve.
Hi,
I the light of my earlier reply to Linus, I've applied this to the GFS2 git tree. Thanks,
Steve.
Quoted 18 lines · Al Viro
Signed-off-by Al Viro <[email protected]>--- fs/gfs2/dir.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)fs/gfs2/dir.c +1
-1@@ -815,7 +815,7 @@ static struct gfs2_leaf *new_leaf(struct leaf = (struct gfs2_leaf *)bh->b_data; leaf->lf_depth =cpu_to_be16 (depth); leaf->lf_entries =0 ; - leaf->lf_dirent_format =cpu_to_be16 (GFS2_FORMAT_DE ); + leaf->lf_dirent_format =cpu_to_be32 (GFS2_FORMAT_DE ); leaf->lf_next =0 ;memset (leaf->lf_reserved,0 ,sizeof (leaf->lf_reserved)); dent = (struct gfs2_dirent *)(leaf+1 );