Xen devel - Jun 2006 - Xen bug 647: Creation of Mini-OS domain causes certain xm commands to hang

Dear Aravindh,

Did you or anyone else on the Xen distribution list make any progress
with bug 647: "Creation of Mini-OS domain causes certain xm commands to
hang"?

I am getting exactly the same fault with changeset 10508:1da8f53ce65b
when trying to run mini-OS.  I have tried to trace exactly what the
Python code in xend is doing, but I am not a Python expert, and I am not
convinced that what I am seeing makes sense.  

After starting mini-OS and running "xm list" it seems as if the
function
"domains" in XMLRPCServer.py is entered, but never returns.

Regards,

Melvin.

--

<Melvin.Anderson@hp.com>, HPLabs, Bristol.





_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Melvin,
> Did you or anyone else on the Xen distribution list make any progress
> with bug 647: "Creation of Mini-OS domain causes certain xm commands
to
> hang"?
Gregor said that he would be looking into this. I am not sure if he has
made any progress.
> After starting mini-OS and running "xm list" it seems as if the
function
> "domains" in XMLRPCServer.py is entered, but never returns.
Yup that is where the hang happens. BTW, if you don''t initialize xenbus
this problem does not happen the last time I tried. So I think it is a
locking issue in the mini-os xenbus code.

Cheers,
Aravindh

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Steven Smith, who originally wrote XenBus code has now updated it as a 
part of a larger patch. It''s possible that the problem will go away 
then. We are currently working on checking that in.

I''ll report back when that happens.

Cheers
Gregor
> Melvin,
> 
> 
>>Did you or anyone else on the Xen distribution list make any progress
>>with bug 647: "Creation of Mini-OS domain causes certain xm
commands
> 
> to
> 
>>hang"?
> 
> 
> Gregor said that he would be looking into this. I am not sure if he has
> made any progress.
> 
> 
>>After starting mini-OS and running "xm list" it seems as if
the
> 
> function
> 
>>"domains" in XMLRPCServer.py is entered, but never returns.
> 
> 
> Yup that is where the hang happens. BTW, if you don''t initialize
xenbus
> this problem does not happen the last time I tried. So I think it is a
> locking issue in the mini-os xenbus code.
> 
> Cheers,
> Aravindh
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 29 Jun 2006, at 18:40, Grzegorz Milos wrote:
> Steven Smith, who originally wrote XenBus code has now updated it as a 
> part of a larger patch. It''s possible that the problem will go
away
> then. We are currently working on checking that in.
>
> I''ll report back when that happens.
It''d be nice if broken guests didn''t hang the control server!

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hello all!

Xenbus patch seems to have solved the xend/xm problem. Could please  
apply it Keir?

Patch message:
Implements XenBus transactions in MiniOS.

Signed-of-by: Steven Smith <sos22@cam.ac.uk>



Cheers
Gregor

>> Steven Smith, who originally wrote XenBus code has now updated it  
>> as a part of a larger patch. It''s possible that the problem
will
>> go away then. We are currently working on checking that in.
>>
>> I''ll report back when that happens.
>
> It''d be nice if broken guests didn''t hang the control
server!
>
>  -- Keir

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Tue, Jul 04, 2006 at 06:35:18PM +0100, Grzegorz Milos
wrote:
> Hello all!
> 
> Xenbus patch seems to have solved the xend/xm problem. Could please  
> apply it Keir?
> 
> Patch message:
> Implements XenBus transactions in MiniOS.
It looks like this patch only updates MiniOS - does this mean that
broken (or malicious...) guests can still hang the control server, or
has that been fixed independently?

Cheers,
Muli

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>> Hello all!
>>
>> Xenbus patch seems to have solved the xend/xm problem. Could please
>> apply it Keir?
>>
>> Patch message:
>> Implements XenBus transactions in MiniOS.
>
> It looks like this patch only updates MiniOS - does this mean that
> broken (or malicious...) guests can still hang the control server, or
> has that been fixed independently?
>
Further tests revealed that the problem has not gone away completely  
(although it seems to be happening more sporadically).

If Mini-OS does not initialise XenBus (if you remove xenbus_init()  
call from kernel.c) the problem will not appear at all. So far I have  
not been able to pinpoint the root cause of the problem though. Maybe  
somebody more familiar with XenStore will be able to help.

The patch certainly does not break anything though (this is mostly  
for Keir''s attention).

I''ll get back to you if I find anything more.

Cheers
Gregor

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

>> It looks like this patch only updates MiniOS - does this mean that
>> broken (or malicious...) guests can still hang the control server, or
>> has that been fixed independently?
>>
>
> Further tests revealed that the problem has not gone away  
> completely (although it seems to be happening more sporadically).
>
> If Mini-OS does not initialise XenBus (if you remove xenbus_init()  
> call from kernel.c) the problem will not appear at all. So far I  
> have not been able to pinpoint the root cause of the problem  
> though. Maybe somebody more familiar with XenStore will be able to  
> help.
>
> The patch certainly does not break anything though (this is mostly  
> for Keir''s attention).
>
> I''ll get back to you if I find anything more.
>
Ok, found the cause:
MiniOS does test_xenbus soon after booting up. This in turn tries to  
write to some file under device/vif/0/. XenStrore wil create device/ 
vif/0 directory to make the write possible even if no network  
interfaces are defined. This confuses Xend as it expects to find a  
backend corresponding to the bogus frontend. Clearly MiniOS induces  
the incorrect behaviour but it''s more of Xend bug.

To answer your question Muli - yes, it''s possible for broken/ 
malicious guest to hang the control server.

I''m attaching a simple patch for MiniOS that disables the execution  
of test_xenbus (but removing the writes would do as well).

Keir could you apply please?

Cheers
Gregor


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel