[ENBD] Client freeze with Reiserfs over enbd

[Peter T. Breuer]

"A month of sundays ago Brian Keefer wrote:"
> I'm using enbd-2.4.31 and linux-2.4.20 on both client and server. I was in
> X when my system froze, so I only have the system log of the failure. The

Oh, I see. The usual technique is to have a serial line hooked up and
to be scrolling the logs to it in realtime via syslog :-). Useful for
rebooting too.

If you have the kernel debugger compiled in, you can generally get out
of jail (blind). Maybe with magic-sysreq too. 

> next time I decide to crash my Client, what could I do to facilitate
> debugging (besides doing it in console, with a Polaroid).

The latter, remotely, :-). Serial console. Kernel log level zero, etc.

> Here's the syslog from the crash 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) found device /dev/nda2 ok 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) opened socket 5 to 192.168.1.9:2010 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) read passwd ok from 192.168.1.9:2010 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) got cliserv magic ok from 192.168.1.9:2010 
> > May 11 22:51:07 moya enbd-client: enbd-client  5367: <#2933> negotiate client (1) got signature [lkbA8C], had [lkbA8C] 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) got a signature ok from 192.168.1.9:2010 
> > May 11 22:51:07 moya enbd-client: enbd-client: client (1) begins main loop 
> > May 11 23:07:33 moya kernel: ENBD enbd.c #1691[2]: nbd_rollback rollback req d9413d20, type 1, sector 8581968, 4 blks, 8  sectors from slot 0! 

well, the driver was alive. The indication is that the client daemon
didn't respond in time for a request. Maybe the network didn't. How
about daemon.log?


> > May 11 23:07:46 moya kernel: ENBD enbd.c #1691[3]: nbd_rollback rollback req d9413d20, type 1, sector 8581968, 4 blks, 8  sectors from slot 1! 
> > May 11 23:07:53 moya kernel: ENBD enbd.c #2410[0]: nbd_ack ignoring ack of req d9413d20 which slot does not have 

Isn't that interesting. The reply came in, but 13s late!

What could do that?

Anyway, the client daemon was also OK. It treated the reply.

> > May 11 23:11:17 moya syslogd 1.3-3: restart.

Yerrs. Maybe I should look at what happens in that pathway. At least I
should maybe flag the request so that positive news is logged when
it does finally get treated.

It's quite possible it's a bug.

Peter