Re: [Nbd] gznbd endian fix

[Wouter Verhelst <wouter@...38...>]

Op za, 22-01-2005 te 14:26 -0800, schreef Stephanie J. Lockwood-Childs:
> Hello all,
> 
> after many years of seeing a mysterious "Network Block Device" option
> under kernel config I saw mention of it in an LWN comment and got curious
> enough to investigate. I remember seeing that gznbd is considered
> "experimental" but it's a nice simple example of a userspace program
> talking to the nbd module. PPC being my platform of choice, I noticed an
> endian problem because WORDS_BIGENDIAN wasn't getting defined like it was
> in the main nbd server & client.

That's just bit rot, I guess. I haven't much looked at gznbd; my plan is
to integrate it into nbd-server proper at some time in the future.

> My simple fix is just to use the common
> headers that were set up for the right endianness by the configure script;
> patch with this fix follows.

Yeah, that's sensible. Applied (to HEAD, not the 2.7 branch).

> With this patch applied, gznbd seems to work ok, though I can definitely
> see why it is deemed experimental... the gznbd.c comment "ro is important,
> since writes will fail horribly" is no exaggeration; forgetting the ro
> option in the mount results in failure & a kernel oops when attempting to
> unmount the device. On the other hand, it's likely a problem in kernel
> error recovery rather than a user-space bug, since gznbd returned a
> read-only-filesystem error as seems appropriate.
> 
> BTW, let me know if patches are not welcome on this list. 

Oh, they are. Most certainly. I actually prefer it this way; it's just
that there is (AFAIK) no way to switch off that damn patch thingy.

-- 
Wouter Verhelst
NixSys BVBA
Louizastraat 14, 2800 Mechelen
T:+32 15 27 69 50 / F:+32 15 27 60 51 / M:+32 486 836 198