[ENBD] 2.5.64 port?

[Peter T. Breuer]

"A month of sundays ago Ciprian Paunescu wrote:"
> 2.5.64 had a problem with NFS mounts, fixed in 2.5.65.

Hmm.  There are many things wrong with 2.5.64 (it segfaults on nearly
every module remove for a start).  I didn't know nfs was working yet!

> I am writing this message without direct access to the
> machine I've worked on, so most of this is from memory.
> Files and other names are only aproximative, sorry for
> the lack of precision.
> I tried to compile agains the 2.5.65, and there are
> some errors generated from the inclusion of linux/fs.h

Not on my 2.5.64.  If you get an error, it must be some change in fs.h.

> As far as I can tell, all is needed from that file are
> the ioctl defines.

Probably.

> Once I got those, I got most of the enbd-2.4.31 to
> compile, but I gave up on the raid utilities, it wants

Raid utilities? 

> some defines that aren't anywhere.
> I also copied the "linux-2.5.65" directory over the

This is NOT A GOOD IDEA!  Why not apply the patch instead?  The
directory only contains my snapshot of code, not the means to integrate
it into the kernel compile tree.  yes, the changes needed should be
within your compass, but it took me about a day to do it!

> 2.5.65 source tree, and compiled the enbd in the kernel,
> with NBD turned off, ENBD on and ENBD-IOCTL off.
> The enbd-ioctl and enbd-main seem to have a linux_version

I think your procedure is somehow flawed.

> variable duplication problem, so the enbd-ioctl doesn't 
> link.

It most certainly does.  But on the other hand it shouldn't link!  It's
a standalone module.  What exactly are you trying to do?

Maybe your problem arises from trying to link two things together that
nobody said to link ... how are you doing all this? Why not just apply
the provided patch?

> Another problem was a register_devblk call, that had an 
> extra argument.

Not in 2.5.64 at least. But I get the impression that what you are
doing is somehow not what is intended. All you are supposed to do is
apply the patch and compile the kernel. If the patch needs adjusting to
run against 2.5.65 instead of 2.5.64, so be it, but it is to be
expected! 

> I also have devfs on, after rebooting with the new kernel,
> the /dev/nb/a .... appeared.
> So far so good! What I am trying to do is export a 3ware
> RAID (/dev/sda), 1.725TB in size from raschal2 to raschal1.
> I've executed
> raschal2:enbd-server 4567 /dev/sda
> 
> raschal1:enbd-client raschal2:4567 /dev/nb/a
> 
> This seems to work for a little while, the client dumps a 
> few messages, including the fact that the imported disk
> is the proper size.
> 
> Once I try a "fsdisk /dev/nb/a/0", the client crashes.

I have never heard of "fsdisk". What is it?

> So, what did I do wrong? I didn't apply the 2.5.64 patch,

Please do! That;s what you are supposed to do!

> since the kernel I had was 2.5.65, I just copied the
> "nbd/linux-2.5.64" subtree over /usr/src/linux, an unpatched
> 2.5.65 kernel tree.

Well, this is a completely wrong thing to do.

Peter