Re: SCSI disk devices

Martin von Loewis (martin@mira.isdn.cs.tu-berlin.de)
Sun, 4 May 1997 19:47:12 +0200


> One other thing - I was just looking through the latest kernel
> sources, and on i386, the dev_t is *still* 16 bits :-(. It isn't just
> i386 either - only the Alpha and PPC get this right. Someone needs to
> eventually take the initiative and fix the kernel so that it has new
> versions syscalls for *stat and mknod that support a larger dev_t.
> This does need to be coordinated with *lots* of other people to make
> sure that there are no other things that need to get changed at the
> same time, and to make sure that in fact the change is done correctly
> for the architecture in question.

I agree this has to be done in the 2.2 time frame. We've been through this
over and over again, and there probably no point in re-iterating all the
arguments. Somebody just needs to provide a set of patches, and we'll see
what will break. glibc already made a decision: dev_t is going to be 64 bits.
(Of course, the Linux kernel could still use something else, but it looks
like an acceptable decision - especially as libc 6 clients don't know
what the in-kernel dev_t is, anyways).
The remaining question is how to divide the 64 bits between major and minor
number. Linus claims this is not an issue. I don't agree, because file
systems have to know how to convert the old dev_t to the new dev_t, or
better, there has to be a conversion somewhere.
Finally, we need a file system that can actually store the new dev_t. I
understand the latest ext2 has patches for this, and it uses 64 bits for
dev_t as well. This should probably be incorporated around the same time
when the other patches are available.

Regards,
Martin