[ENBD] 2.5.46 update

[Peter T. Breuer]

"Tad Kollar wrote:"
> Peter T. Breuer wrote:
> > I got the impression that the sector count mught be involved.
> 
> Well, the errors start appearing at logical block 2. I tried it on two 
> physical disks, 160GB each. It's okay I guess, I didn't really want to use the 
> linear driver, just test with it.
> 
> Luckily, LVM2 does work, and there's even a Debian package for it. I just used 
> it to combine three 400GB enbd-exported volumes, format with XFS, and mount 
> successfully. Only problem so far is that it thinks a single 'physical' disk 
> of a little greater than 1TB is actually 2TB+. It doesn't want to work with 

That would be 2^31 sectors of 512 bytes each. It has a sign problem.

> disks over 2TB. Hopefully it can *create* a volume larger than that... that's 
> the next test.

I think one needs "large block devices" compiled in the kernel. 

There was a move to count in blocks, not sectors, which would have taken
the 32 bit code up to about 16TB max, but I don't recall the current 
state of the counters. I imagine that the large compile option simply
moves to 64 bit counters.

OTOH there are already 64bit ioctl calls for size and so on ... hmmm.

> I think my last stumbling block is that the current 2.5 md driver can't 

There was something on the kernel list today. Apparently one must
remake the device at every start. No permanent state kept in the
device.

> restart a previously created device, which I need to do on the servers. If I 
> have to, I can go back to a 2.4 kernel there, though.

Peter