Re: [Nbd] Random NBD connection shutdowns...

[Michal Belczyk <belczyk@...1274...>]

On Sun, Mar 03, 2013 at 03:53:02PM +0100, Michal Belczyk wrote:
> Hi there,
> 
> a few months ago I noticed that nbd-client (kmod code actually) randomly
> shuts down the connection during heavy parallel xdd benchmarks in my
> local setup.  The following change to nbd.c fixed that for me:
> 
> @@ -303,9 +303,13 @@
>         struct request *req, *tmp;
>         int err;
>  
> +wait:
>         err = wait_event_interruptible(nbd->active_wq, nbd->active_req != xreq);
> -       if (unlikely(err))
> +       if (unlikely(err)) {
> +               if (err == -ERESTARTSYS)
> +                       goto wait;
>                 goto out;
> +       }
> 
> Please consider pushing it upstream.
> Thanks!

I was able to hit this bug with the stock nbd-server, it is NOT
bnbd-specific.  If this mailing list is the wrong place to submit NBD
driver bugs, then please point me to the proper location, although I
believe there are kernel people signed up here as well...
I remember that it was fairly easy to trigger with a single volume NBD
connection and large queuedepth passed to xdd -- can't remember if I did
reduce the stack size before running xdd, perhaps the memory was running
low... yes, it was over localhost, not remote.
Thanks!

-- 
Michal Belczyk Sr.