[ENBD] (no subject)

[Peter T. Breuer]

"A month of sundays ago HUDZIA Bertrand wrote:"
> On Wed, 10 Apr 2002 15:53:07 +0200 (MET DST)
> "Peter T. Breuer" <ptb@it.uc3m.es> wrote:
> > No it isn't. Try "echo 0 > /proc/nbdinfo".

(reformatted to 72 chars/line)

> Well , when I say the client is out, i mean the pc where the client
> was running is OUT.  i can't get a hand into it, all tty are broken, ssh
> too , ect , i * really * have no other solution than a hard-reboot.

Then the kernel has oopsed (if i/o were deadlocked, that would explain
it, but i/o can't deadlock in the driver itself - just the device can
deadlock if anything deadlocks, so you must be outside normal operating
parameters, and hence the kernel has oopsed).

> > > Both ran 2.4.18 kernel under a RedHat-7.2 , the shared device was
> > > 35Go in reiserfs (problem also on ext2).

> > Shared device? That makes me suspicious. Nothing should be shared!
> > It's borrowed :-).

> hda3 is not mounted on serv, only through enbd and only once.

OK

> Both pc run 2.4.18 kernel under a RedHat-7.2 on a pIII 950Mhz.
> serv got 256 Mo RAM and cli have 128 Mo.

So these are uniprocessors? On a uniprocessor kernel?

> HDD are 40Go.
> serv have an eth0: Intel Corp. 82557 [Ethernet Pro 100]
> cli have eth0: Intel Corp. 82820 (ICH2) Chipset Ethernet Controller
> Both card use the same drv : eepro100.c
> They are pluged onto a Titan T4 (switch 100).
> Motherboard ?, hum , i didn't look at them.

It sounds all OK. Would you mind running nbd-test on the device, with
some size limit like "-s 4G""?  The only thing untested about your
setup is the very large size it has. I run pretty much the same setup
but at 4GB.

> They use the same binarys of enbd which were compiled on the serv.
> > Let me know and I'll comment further. I also need to see /proc/nbdinfo.

> Arf , unfortunately I can't get the /proc/nbdinfo * after * the crash ... but i'll try to get one ..

The more info you can get the better.  What are your
/proc/sys/vm/bdflush settings? Memory overcommit?

Peter