Hi again,
Having solved the previous topic using Bogdan's hint I stumbled straight into a new problem. Is there a workaround for not being able to use the avp-ops from onreply_route?
Some details on the reason: We implement the Service- Route extension according to RFC 3608. The registrar returns in the 200OK a ServiceRoute header where all proxies on the (return) path register themselves.
The last proxy (that forwards the 200 OK to the UA) is supposed to store the ServiceRoute locally. Our plan was straight-forward: to store the header field's content via avp-ops from within the onreply_route.
Unfortunately avp-ops are not available in onreply_route - Bogdan explains the reason why in http://www.archivum.info/serdev%40iptel.org/2005-02/msg00274.html
Is there a specific reason why the access to the avp list is not synchronized? Performance? Basically, provided a good description of the implementation and concept, imho it could be left to users to restrict themselves to have avpops writes in only in their onreply_routes for a specific avp pair. Or, provide synchronized avps?
So, does anyone have an idea how we can: - Store an AVP containing header fields extracted from a 200 OK message - Retrieve this AVP on incoming invites in order to place these values into the message (security check and/or Route enforcement).
Thanks in advance, best regards --Joachim
Hi Joachim,
the reason haven't changed: avps doesn't work in on_reply route since the avp belong to a transaction and the on_reply route execution is not synchronized and may be done in parallel for same transaction.
For the moment there is no solution to this - synchronizing the on_reply routes will be quite ugly :-/
But what will be done (there was a discussion with Juha on this topic) is to add global avps which not being bound to a transaction can be used inside on_reply route. I guess this will solve your problem....
regards, bogdan
Joachim Fabini wrote:
Hi again,
Having solved the previous topic using Bogdan's hint I stumbled straight into a new problem. Is there a workaround for not being able to use the avp-ops from onreply_route?
Some details on the reason: We implement the Service- Route extension according to RFC 3608. The registrar returns in the 200OK a ServiceRoute header where all proxies on the (return) path register themselves.
The last proxy (that forwards the 200 OK to the UA) is supposed to store the ServiceRoute locally. Our plan was straight-forward: to store the header field's content via avp-ops from within the onreply_route.
Unfortunately avp-ops are not available in onreply_route - Bogdan explains the reason why in http://www.archivum.info/serdev%40iptel.org/2005-02/msg00274.html
Is there a specific reason why the access to the avp list is not synchronized? Performance? Basically, provided a good description of the implementation and concept, imho it could be left to users to restrict themselves to have avpops writes in only in their onreply_routes for a specific avp pair. Or, provide synchronized avps?
So, does anyone have an idea how we can:
- Store an AVP containing header fields extracted from a
200 OK message
- Retrieve this AVP on incoming invites in order to place
these values into the message (security check and/or Route enforcement).
Thanks in advance, best regards --Joachim
Users mailing list Users@openser.org http://openser.org/cgi-bin/mailman/listinfo/users
Hi Bogdan,
the reason haven't changed: avps doesn't work in on_reply route since the avp belong to a transaction and the on_reply route execution is not synchronized and may be done in parallel for same transaction.
For the moment there is no solution to this - synchronizing the on_reply routes will be quite ugly :-/
Do you see another - even proprietary - very-short-term-workaround that can provide this functionality (storing avps within onreply_route in the database that can be later on accessed from the route block using avp-ops) without extra module coding? Ugly is not important at all if it works. After all it's just about prototyping, the final code should be clean... ;))
But what will be done (there was a discussion with Juha on this topic) is to add global avps which not being bound to a transaction can be used inside on_reply route. I guess this will solve your problem....
Definitely. That's also conceptually closer to what Service-Route is supposed to do - it's valid for the lifetime of a registration (although it might be changed by re-registrations) and not just for one transaction. Any estimate when this will be available?
Thanks for your help, regards --Joachim
Hi Joachim,
apply the attached tm.patch - it's trivial change that will synchronize the execution of on_reply route. And it will be safe to use avpops functions in on_reply routes. Note that you have to change the module interface of AVPOPS and add the ONREPLY_ROUTE flag.
regards, bogdan
PS: if you get a coredump, I'm free of any responsibilities :D
Joachim Fabini wrote:
Hi Bogdan,
the reason haven't changed: avps doesn't work in on_reply route since the avp belong to a transaction and the on_reply route execution is not synchronized and may be done in parallel for same transaction.
For the moment there is no solution to this - synchronizing the on_reply routes will be quite ugly :-/
Do you see another - even proprietary - very-short-term-workaround that can provide this functionality (storing avps within onreply_route in the database that can be later on accessed from the route block using avp-ops) without extra module coding? Ugly is not important at all if it works. After all it's just about prototyping, the final code should be clean... ;))
But what will be done (there was a discussion with Juha on this topic) is to add global avps which not being bound to a transaction can be used inside on_reply route. I guess this will solve your problem....
Definitely. That's also conceptually closer to what Service-Route is supposed to do - it's valid for the lifetime of a registration (although it might be changed by re-registrations) and not just for one transaction. Any estimate when this will be available?
Thanks for your help, regards --Joachim
Index: modules/tm/t_reply.c =================================================================== RCS file: /cvsroot/openser/sip-server/modules/tm/t_reply.c,v retrieving revision 1.16 diff -u -r1.16 t_reply.c --- modules/tm/t_reply.c 18 Oct 2005 17:42:16 -0000 1.16 +++ modules/tm/t_reply.c 19 Oct 2005 14:56:34 -0000 @@ -1228,6 +1228,9 @@ is_local(t)?"yes":"no"); }
+ /* lock the reply*/ + LOCK_REPLIES( t ); + /* processing of on_reply block */ if (t->on_reply) { /* transfer transaction flag to branch context */ @@ -1249,8 +1252,8 @@ set_avp_list( backup_list ); }
- /* lock the reply*/ - LOCK_REPLIES( t ); + /* lock the reply + LOCK_REPLIES( t );*/ /* mark that the UAC received replies */ uac->flags |= T_UAC_HAS_RECV_REPLY;
Hi Bogdan,
apply the attached tm.patch - it's trivial change that will synchronize the execution of on_reply route. And it will be safe to use avpops functions in on_reply routes. Note that you have to change the module interface of AVPOPS and add the ONREPLY_ROUTE flag.
Many, many thanks. Will try it tonight.
PS: if you get a coredump, I'm free of any responsibilities :D
No, no, we insist on ownership of code. Quality management rulez. ;)))
Thanks again, Best regards --Joachim
Hi Bogdan,
I tried your patch, works perfectly, absolutely no problem so far. It's exactly what we needed. Just one more question (for now ;): Is the same procedure safe to be used with 0.9.5 and 0.9.0 devel? Obviously the patch must be applied manually for 0.9.x; Looking at the sources and comparing the affected 0.9.x and 0.10.x files I am pretty sure that moving the lock in the 0.9.x t_reply.c is supposed to do exactly the same thing like in 0.10 devel. And, as written above, anything works as expected - also for 0.9.x.
So, is there something we miss?
Thanks again, best regards --Joachim
-----Original Message----- From: Bogdan-Andrei Iancu [mailto:bogdan@voice-system.ro] Sent: Mittwoch, 19. Oktober 2005 17:07 To: Joachim Fabini Cc: users@openser.org; 'Joachim Fabini' Subject: Re: [Users] Persistently storing headers from onreply_route()?
Hi Joachim,
apply the attached tm.patch - it's trivial change that will synchronize the execution of on_reply route. And it will be safe to use avpops functions in on_reply routes. Note that you have to change the module interface of AVPOPS and add the ONREPLY_ROUTE flag.
regards, bogdan
PS: if you get a coredump, I'm free of any responsibilities :D
Joachim Fabini wrote:
Hi Bogdan,
the reason haven't changed: avps doesn't work in on_reply
route since
the avp belong to a transaction and the on_reply route execution is not synchronized and may be done in parallel for same transaction.
For the moment there is no solution to this - synchronizing the on_reply routes will be quite ugly :-/
Do you see another - even proprietary - very-short-term-workaround that can provide this functionality (storing avps within
onreply_route
in the database that can be later on accessed from the route block using avp-ops) without extra module coding? Ugly is not important at all if it works. After all it's just about prototyping, the final code should be clean... ;))
But what will be done (there was a discussion with Juha on this topic) is to add global avps which not being bound to a transaction can be used inside on_reply route. I guess this will solve your problem....
Definitely. That's also conceptually closer to what Service-Route is supposed to do - it's valid for the lifetime of a registration (although it might be changed by re-registrations) and not just for one transaction. Any estimate when this will be available?
Thanks for your help, regards --Joachim
Hi Joachim,
you are perfectly right - do the same trick for devel also, if needed.
regards, bogdan
Joachim Fabini wrote:
Hi Bogdan,
I tried your patch, works perfectly, absolutely no problem so far. It's exactly what we needed. Just one more question (for now ;): Is the same procedure safe to be used with 0.9.5 and 0.9.0 devel? Obviously the patch must be applied manually for 0.9.x; Looking at the sources and comparing the affected 0.9.x and 0.10.x files I am pretty sure that moving the lock in the 0.9.x t_reply.c is supposed to do exactly the same thing like in 0.10 devel. And, as written above, anything works as expected - also for 0.9.x.
So, is there something we miss?
Thanks again, best regards --Joachim
-----Original Message----- From: Bogdan-Andrei Iancu [mailto:bogdan@voice-system.ro] Sent: Mittwoch, 19. Oktober 2005 17:07 To: Joachim Fabini Cc: users@openser.org; 'Joachim Fabini' Subject: Re: [Users] Persistently storing headers from onreply_route()?
Hi Joachim,
apply the attached tm.patch - it's trivial change that will synchronize the execution of on_reply route. And it will be safe to use avpops functions in on_reply routes. Note that you have to change the module interface of AVPOPS and add the ONREPLY_ROUTE flag.
regards, bogdan
PS: if you get a coredump, I'm free of any responsibilities :D
Joachim Fabini wrote:
Hi Bogdan,
the reason haven't changed: avps doesn't work in on_reply
route since
the avp belong to a transaction and the on_reply route execution is not synchronized and may be done in parallel for same transaction.
For the moment there is no solution to this - synchronizing the on_reply routes will be quite ugly :-/
Do you see another - even proprietary - very-short-term-workaround that can provide this functionality (storing avps within
onreply_route
in the database that can be later on accessed from the route block using avp-ops) without extra module coding? Ugly is not important at all if it works. After all it's just about prototyping, the final code should be clean... ;))
But what will be done (there was a discussion with Juha on this topic) is to add global avps which not being bound to a transaction can be used inside on_reply route. I guess this will solve your problem....
Definitely. That's also conceptually closer to what Service-Route is supposed to do - it's valid for the lifetime of a registration (although it might be changed by re-registrations) and not just for one transaction. Any estimate when this will be available?
Thanks for your help, regards --Joachim
-
Bogdan-Andrei Iancu -
Joachim Fabini