On Thu, 17 Jun 1999, Matti Aarnio wrote:
> I finally noticed that this bit is STILL missing...
> Please Ingo, either scrap the current superblock entirely,
> or INCLUDE this patch in!
Matti, we cannot simply break existing installations! Yes, Sparc64 and
Alpha systems currently have a different RAID superblock layout than other
architectures, but fixing this will take a bit more than just changing the
structure.
> Now of course the RH 6.0 does contain this BUG, and all Alpha,
> and SPARC(64) systems (very least) create wrong format superblocks,
> which are not understandable by e.g. ia32 systems.
i very well know, but when we (you) noticed it it was already late. It is
_not_ acceptable to break working systems. The only effect of the bug is
that currently RAID arrays cannot be moved across 32-bit and 64-bit
systems. (and cannot be moved across different endianness)
> By the way, I would like to have these endianity independent, that way
> I could mount same disk also at e.g. ia32 system thru fiber-channel
> fabric... (No, I am not quite yet doing that...)
i'd welcome any patch that does that in a safe way. I have planned to add
all these things to RAID 1.0, which will be an incompatible layout anyway.
(of course no data will be lost, people will have to use a user-space tool
to upgrade in a more or less safe way. mkraid --upgrade will have to
detect endianness and wordsize. This still doesnt protect against someone
doing the --upgrade on the wrong system, but that should not be the case
anyway.) In all cases the problem can be 'fixed' by re-mkraid-ing the
array on the new architecture.
Ingo
