[Serdev] Re: [Devel] tm bug in t_hooks.c

[Federico Giannici]

Bogdan-Andrei Iancu wrote:
> please try the following patch and let me know if solves the problem. 
> ...just a fast guess...

No, the problem remain.
When a retrasmission arrives, both t_lookup_request() and 
t_check_trans() are TRUE, and the following error appears:

Dec  8 16:44:15 eowyn OpenSER[7103]: ERROR: t_newtran: transaction 
already in process 0x512a22e0
Dec  8 16:44:15 eowyn OpenSER[7103]: ERROR: sl_reply_error used: I'm 
terribly sorry, server error occurred (1/SL)


Bye.



> Federico Giannici wrote:
> 
>> Bogdan-Andrei Iancu wrote:
>>
>>> Federico Giannici wrote:
>>>
>>>> Bogdan-Andrei Iancu wrote:
>>>>
>>>>> Hi Federico,
>>>>>
>>>>> I'm afraid you have a script error - you called twice some 
>>>>> functions which create transactions...The error is generate most 
>>>>> probably by t_relay() which found the transaction already created 
>>>>> in the script (maybe via t_newtran).
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Yes, I know it.
>>>> It is the UAC module that create a transaction (I don't know if in 
>>>> any case or only in some cases).
>>>>
>>> the uac module does not create a transaction - uac_replace_from() 
>>> performs just text mangling and installs some hooks.
>>
>>
>>
>> You are right, this problem doesn't seem to be related to the UAC 
>> module (I was confused by the previous AVPs vanishing bug).
>>
>> Anyway, when a retrasmission of an INVITE reaches the t_relay() 
>> function, the following error appear:
>>
>> Dec  8 14:56:18 eowyn OpenSER[21824]: ERROR: t_newtran: transaction 
>> already in process 0x50132480
>> Dec  8 14:56:18 eowyn OpenSER[21824]: ERROR: sl_reply_error used: I'm 
>> terribly sorry, server error occurred (1/SL)
>>
>> And the UAC closes the call (because of the 500 error)!
>>
>> Please note that there is no t_newtran() in the entire script, and the 
>> problem occours only with the retrasmissions!
>>
>> What is the supposed way to avoid this?
>>
>>
>> Thanks.
>>
>>
>>
>>>> So when the message arrives at the classic t_relay() part of the 
>>>> script, I cannot know if a transactionis is already established.
>>>>
>>>>
>>>>> To be able to help, please send the script (privately, if 
>>>>> necessary) to take a look.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> If you still think that there is problem somewhere, I will send the 
>>>> script to you...
>>>>
>>>>
>>>> Bye.
>>>>
>>>>
>>>> P.S.
>>>> I sent a similar question at "serusers at iptel.org", but still no 
>>>> usefull reply...
>>>>
>>>>
>>>>
>>>>
>>>>> Federico Giannici wrote:
>>>>>
>>>>>> Bogdan-Andrei Iancu wrote:
>>>>>>
>>>>>>> Hi Frederico,
>>>>>>>
>>>>>>> that's good..one more down....
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> Well.... the AVPs are no more disappearing, but I still have some 
>>>>>> problem with the reinvites...
>>>>>>
>>>>>> It seems that uac_replace_from() creates a transaction when an 
>>>>>> INVITE retrasmission is received. Infact in that case the 
>>>>>> following errors are logged for the retrasmitted INVITEs:
>>>>>>
>>>>>> Dec  5 18:49:25 eowyn OpenSER[5654]: ERROR: t_newtran: transaction 
>>>>>> already in process 0x502a56f8
>>>>>> Dec  5 18:49:25 eowyn OpenSER[5654]: ERROR: sl_reply_error used: 
>>>>>> I'm terribly sorry, server error occurred (1/SL)
>>>>>>
>>>>>> This is generated by this standard code:
>>>>>>
>>>>>> if ( !t_relay() ) {
>>>>>>     sl_reply_error();
>>>>>> }
>>>>>>
>>>>>> I thought that I could find the retrasmissions (with the 
>>>>>> t_lookup_request() function) and use the t_forward_nonack_uri() 
>>>>>> function in that case, but I got the following error:
>>>>>>
>>>>>> Dec  5 18:47:29 eowyn OpenSER[26647]: ERROR:tm:t_forward_nonack: 
>>>>>> no branch for forwarding
>>>>>>
>>>>>> Obviously, I have not enought undestanding of TM functions and 
>>>>>> transactions in general.
>>>>>>
>>>>>> So, the question is: can anybody post a correct fragment of script 
>>>>>> code that relay a message, correctly handling retrasmissions?
>>>>>>
>>>>>> Otherwise, could I simply discard retrasmissions (with a simple 
>>>>>> exit)?
>>>>>> And how I can recognize retrasmissions? Is t_lookup_request() the 
>>>>>> correct function for a test?
>>>>>>
>>>>>>
>>>>>> Thanks.
>>>>>>
>>>>>>
>>>>>>> Federico Giannici wrote:
>>>>>>>
>>>>>>>> Bogdan-Andrei Iancu wrote:
>>>>>>>>
>>>>>>>>> Hi Cesc,
>>>>>>>>>
>>>>>>>>> right!! as the transaction list is set all the time 
>>>>>>>>> (disregarding the presence of callbacks), it should be also 
>>>>>>>>> unset all the time....
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> This solves the bug of the AVPs disappearing with retrasmitted 
>>>>>>>> INVITES too!!!
>>>>>>>>
>>>>>>>> Bogdan, now you can remove that bug from the ones to search for...
>>>>>>>>
>>>>>>>> Thank you Cesc.
>>>>>>>
>>>>>>>
>>
>>
>>
> 
> 
> ------------------------------------------------------------------------
> 
> Index: modules/tm/tm.c
> ===================================================================
> RCS file: /cvsroot/openser/sip-server/modules/tm/tm.c,v
> retrieving revision 1.15
> diff -u -r1.15 tm.c
> --- modules/tm/tm.c	18 Oct 2005 17:03:15 -0000	1.15
> +++ modules/tm/tm.c	8 Dec 2005 15:29:41 -0000
> @@ -490,6 +490,7 @@
>  	t_on_negative( 0 );
>  	t_on_reply(0);
>  	t_on_branch(0);
> +	set_t(T_UNDEFINED);
>  	/* reset the kr status */
>  	set_kr(0);
>  	return 1;


-- 
___________________________________________________
     __
    |-                      giannici at neomedia.it
    |ederico Giannici      http://www.neomedia.it
___________________________________________________