[PATCH-possible bugfix] xfs: ensure extents are read as be64

[PATCH-possible bugfix] xfs: ensure extents are read as be64

[Harvey Harrison]

The debug code reads the extents in cpu-order rather than BE.  Make the
debug code match.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
---
Possibly 2.6.25 material....if it actually is a bug.  See elsewhere
in the same file that reads as be64. (look for ep->l0).

 fs/xfs/xfs_inode.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
index f43a6e0..eacf2ca 100644
--- a/fs/xfs/xfs_inode.c
+++ b/fs/xfs/xfs_inode.c
@@ -85,8 +85,8 @@ xfs_validate_extents(
 
 	for (i = 0; i < nrecs; i++) {
 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
-		rec.l0 = get_unaligned(&ep->l0);
-		rec.l1 = get_unaligned(&ep->l1);
+		rec.l0 = be64_to_cpu(get_unaligned(&ep->l0));
+		rec.l1 = be64_to_cpu(get_unaligned(&ep->l1));
 		xfs_bmbt_get_all(&rec, &irec);
 		if (fmt == XFS_EXTFMT_NOSTATE)
 			ASSERT(irec.br_state == XFS_EXT_NORM);
-- 
1.5.5.144.g3e42

Re: [PATCH-possible bugfix] xfs: ensure extents are read as be64

[David Chinner]

On Mon, Apr 14, 2008 at 08:28:51PM -0700, Harvey Harrison wrote:
> The debug code reads the extents in cpu-order rather than BE.  Make the
> debug code match.

xfs_validate_extents() is reading in-core extents which have already
been byte-swapped when they were read into memory.

FWIW, when changing anything to do with endian conversion, please
ensure you run:

$ make C=2 CHECKFLAGS="-D__CHECK_ENDIAN__"

over the change. This change would introduce errors because:

typedef struct xfs_bmbt_rec_host {
        __uint64_t              l0, l1;
} xfs_bmbt_rec_host_t;

is not defined with __beXX types like the disk format version of this
structure which is:

typedef struct xfs_bmbt_rec_64 {
        __be64                  l0, l1;
} xfs_bmbt_rec_64_t;

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: [PATCH-possible bugfix] xfs: ensure extents are read as be64

[Harvey Harrison]

On Tue, 2008-04-15 at 14:58 +1000, David Chinner wrote:
> On Mon, Apr 14, 2008 at 08:28:51PM -0700, Harvey Harrison wrote:
> > The debug code reads the extents in cpu-order rather than BE.  Make the
> > debug code match.
> 
> xfs_validate_extents() is reading in-core extents which have already
> been byte-swapped when they were read into memory.
> 
> FWIW, when changing anything to do with endian conversion, please
> ensure you run:
> 
> $ make C=2 CHECKFLAGS="-D__CHECK_ENDIAN__"
> 
> over the change. This change would introduce errors because:
> 
> typedef struct xfs_bmbt_rec_host {
>         __uint64_t              l0, l1;
> } xfs_bmbt_rec_host_t;
> 
> is not defined with __beXX types like the disk format version of this
> structure which is:
> 
> typedef struct xfs_bmbt_rec_64 {
>         __be64                  l0, l1;
> } xfs_bmbt_rec_64_t;
> 

Sorry for the noise, I didn't notice the (slightly) different struct
types.  Just the very similar looking blocks in both functions.

My Bad,

Harvey