zfs code - Oct 2007 - zio_checksum_SHA256 bug

http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/fs/zfs/sha256.c

The block padding calculation in ZFS''s sha256.c (lines 114-115) appears
to be wrong:
       for (i = 0; i < padsize; i++)
                pad[i] = ((uint8_t *)buf)[i];

The padding there is taking bytes from the start of buf, when in fact it should
be from the end (the part not already transformed). As such, the
zio_checksum_SHA256 function is not actually calculating the correct sha256
digest of the passed buf. It could be corrected something like this:

        for (i = 0; i < padsize; i++)
                pad[i] = ((uint8_t *)buf)[i + size - padsize];
--
This messages posted from opensolaris.org

Good catch!  You are quite right, and I just putback the fix:

6640582 zio_checksum_sha256() incorrect if buffer size not a multiple of 64

Fortunately the old code generates the right result whenever the buffer size
*is* a multiple of 64.  All buffers are at least sector-aligned (512 bytes),
so everything''s OK on disk.

Thanks again,

Jeff

On Sun, Oct 07, 2007 at 10:36:08AM -0700, douglas steinwand
wrote:
>
http://cvs.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/uts/common/fs/zfs/sha256.c
> 
> The block padding calculation in ZFS''s sha256.c (lines 114-115)
appears to be wrong:
>        for (i = 0; i < padsize; i++)
>                 pad[i] = ((uint8_t *)buf)[i];
> 
> The padding there is taking bytes from the start of buf, when in fact it
should be from the end (the part not already transformed). As such, the
zio_checksum_SHA256 function is not actually calculating the correct sha256
digest of the passed buf. It could be corrected something like this:
> 
>         for (i = 0; i < padsize; i++)
>                 pad[i] = ((uint8_t *)buf)[i + size - padsize];
> --
> This messages posted from opensolaris.org
> _______________________________________________
> zfs-code mailing list
> zfs-code at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-code