And to add the relevant code:

modules/tsilo/ts_append.c line 85:

    orig_msg = t->uas.request;

    ret = _regapi.lookup_to_dset(orig_msg, table, NULL);
    if(ret != 1) {
        LM_DBG("transaction %u:%u: error updating dset (%d)\n", tindex, tlabel, ret);
        return -1;
    }

I want to pass the uri parameter given to t_append() to:

ret = _regapi.lookup_to_dset(orig_msg, table, NULL);

instead of NULL.

Cheers,
Daniel

On 22/09/15 15:34, Daniel-Constantin Mierla wrote:

Hello,

it was no longer about what tsilo stores internally, but what lookup_to_dset() is given to search for new contact.

To explain with an example:

Call to sip:user@domain.

After lookup("location"), the r-uri of INVITE is changed to contact address found in location table (let's say it was only one contact there, which is sip:randomstring@deviceip:deviceport).

t_relay() is execute and transaction is created, with uas field holding the INVITE with r-uri sip:randomstring@deviceip:deviceport.

After a while, a device for same user registers, if I call ts_append(..., "sip:user@domain"), it finds the INVITE transaction to be associated with sip:user@domain, but lookup_to_dset() is executed internally by tsilo with the argument being the request from Transaction->uas (and no explicit URI parameter). So the lookup is done for sip:randomstring@deviceip:deviceport), because that is the r-uri in T-uas reques. Obviously that will fail to find new contacts.

Hopefully I could expose the issue better now.

Cheers,
Daniel

On 22/09/15 15:14, Federico Cabiddu wrote:

Hi,
I'm not sure I fully understand the problem. See my comments inline

On Tue, Sep 22, 2015 at 2:24 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:

Hello,

tsilo is not using the uri parameter for looking up location for new
destinations. It passes the uas request from transaction module to
registrar lookup_to_dset() function, so the r-uri from that request is used.

However, there is at least one scenario that makes tsilo not working
properly:

- invite comes in
- lookup location finds a contact
- request is forwarded with a new r-uri (the contact of the location
record) -- t_relay() is used and trasaction created, r-uri stored for
uas is contact header
- a new registration comes and ts_append() is executed, failing to find
contacts for r-uri in uas because the r-uri there can be anything
pointing to the first device found in location

This happens because since commit 1e5bad019c450a003e812ee051d84134aad6c5f0, we are using the current uri to store the transaction.

That's why I was wondering (probably I have not been clear in my other mail :)) if, since now ts_store accepts uris as parameter, we still need this.

In this case, the scenario you describe would work flawlessly and, if one need to store the transaction under a new ruri after some kind of message manipulation (like using alias etc) he can always call ts_store with the specific uri he wants to use.

Am I missing something?

Some solutions can be found by creating transaction earlier, but not

easy to tackle always.

I think it would be better to just pass the uri parameter from
ts_append() to lookup_to_dset(). But I noticed that the uri can be just
username if use_domain is 0, which makes things a little more complex,
because registrar/usrloc expect full sip uri.

The solution would be to 'force' always that the uri parameters for
tsilo functions are full SIP uri, and let the code extract only username
when use_domain is 0. If someone has only the username in config, it can
simply put the uri parameter as "sip:username@localhost" -- domain will
be ignored anyhow.

Agree that this could avoid future headaches and provide more flexibility for other use cases. In this case, should we totally remove the dependency on usrloc "use_domain" parameter?

Alternative, if the parameter is just username, built a temporary uri
for it before passing to lookup_to_dset().

I find first solution a bit better because the parameter for tsilo
functions is called uri and building an internal one can end up in some
limitations if it is going to be something that requires tel: or sips:
or parameters such as instance.

Agree.

Any comments/suggestions?

Cheers,
Daniel

Cheers,

Federico
-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat

-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat