[ENBD] 2.4.27 will go out this week

[Peter T. Breuer]

I spent last week chasing bugs under 2.4.17 smp kernels.  In the end, it
turned out there weren't any - just size mismatches between different
gcc compilations and different versions.  If you find that "mke2fs
panics kernel" or other mysterious things like that, then you are a
candidate for the "I have mismatched clients and servers and drivers"
prize. I'm sorry about that. It's my fault for not having padded the
structs with extra bytes in the past.

I only have to try and get the "buffer writes" option back into the
driver for 2.4 kernels, and then its out  the door. I am satisfied
there is no bug.

What's new:

    support for remote ioctls (at least if they don't need more than
    16 bytes data).

    Not a lot else - that was hard enough. Oh yes, there will be a
    new driver for the 2.5.1 kernel in the package.

    Oh yes, the "async mode" (-a) seems to work. Here the client
    acks the kernel before it receives an ack from the server,
    which should make memory/tcp resource deadlocks very hard
    to achieve, even against localhost.

    In principle, the client can be adapted to work entirely against
    localhost, which may be useful for some people (substitute
    "fileserver" for "netserver" in the nbd-client.c code) trying
    to do weird things locally.

Ioctls need a translation table.  It's in an obviously-titled file.
I've included all the floppy and cd ioctls now, but a few of them may
not work (i.e cause bad things) due to my overoptimism.  I'll // them
out of the list as I notice them. Please try them out.

As soon as I get rid of 2.4.27, I'll add support for arbitrary sized
ioctls. 

   ftp://oboe.it.uc3m.es/pub/Programs/nbd-2.4.27pre1.tgz

Those of you with positions in distros, please remind marcelo
to consider the inclusion in new 2.4 issues (I sent him code last
year), and I'll also send code to Linus.

Peter