[OpenSIPS-Users] Failed INVITE tcp_send to UDP UACs

[opensipslist at encambio.com]

Hello Bogdan,

An mar., déc 22, 2009, Bogdan-Andrei Iancu schrieb:
>opensipslist at encambio.com wrote:
>> PROBLEM
>> -------
>>
>> Registrations work well, but when sending IVITEs I see this in log:
>>
>>   Dec 10 15:55:18 name.host.tld <error> opensips[10848]: ERROR:core:tls_blocking_write: too many retries with no operation
>>   Dec 10 15:55:18 name.host.tld <debug> opensips[10848]: DBG:core:tcp_send: after write: c= ceb903f0 n=-1 fd=18
>>   Dec 10 15:55:18 name.host.tld <debug> opensips[10848]: DBG:core:tcp_send: buf=
>>   INVITE sips:person at voip.host.tld SIP/2.0^M
>>   Record-Route: <sip:211.123.22.12:5061;transport=tls;lr=on;ftag=grophp7yc3>^M
>>   Via: SIP/2.0/TLS 211.123.22.12:5061;branch=z9hG4bK4d6.913.0;i=2^M
>>   Via: SIP/2.0/TLS 192.168.1.12:3352;received=125.81.6.152;branch=z9hG4bK-pt7eil3u8qci;rport=3352^M
>>   [...]
>>   Content-Type: application/sdp^M
>>   ^@Dec 10 15:55:18 name.host.tld <error> opensips[10848]: ERROR:core:tcp_send: failed to send
>>   Dec 10 15:55:18 name.host.tld <error> opensips[10848]: ERROR:tm:msg_send: tcp_send failed
>>   Dec 10 15:55:18 name.host.tld <debug> opensips[10850]: DBG:core:handle_ser_child: read response= ceb903f0, -2, fd -1 from 6 (10848)
>>   Dec 10 15:55:18 name.host.tld <error> opensips[10848]: ERROR:tm:t_forward_nonack: sending request failed
>>   Dec 10 15:55:18 name.host.tld <debug> opensips[10850]: DBG:core:io_watch_del: io_watch_del (82d10c0, 21, -1, 0x10) fd_no=11 called
>>   Dec 10 15:55:18 name.host.tld <debug> opensips[10848]: DBG:tm:t_relay_to: t_forward_nonack returned error
>>
>> The forward (I assume by t_relay()) is failing. Is it because a SRV
>> lookup is being done and finding that raw SIP over TCP is being sent
>> to the OpenSIPS TLS listener? Should I be using something other than
>> t_relay() in the config?
>>
>the t_relay() function does server location discovery via DNS (for
>domain voip.host.tld ).
>
>Based on NAPTR and SRV records, a TLS destination is chosen (this
>selection also depends on the RURI - like if a specific SIP schema
>is required, etc).
>
>So, your problem is why TLS was chosen or why TLS write failed ?
>
Your explanation sounds reasonable. The problem is that TLS writes
and sometimes even reads are failing. I wrote about the NAPTR and
SRV records, because I suspected that OpenSIPS was sending TCP
(unencrypted) commands to a TLS connection or vise versa.

You mentioned in another email that this could be NAT related and
I agree. It seems that you don't think there is some mismatch
between TCP and TLS, maybe involving the NAPTR and SRV records.

>> The same config worked well with OpenSER 1.3.X. Only when migrating
>> to 1.6.0 do we see these errors. What could have changed?
>>
But this is maybe a clue. It would seem that something in TLS
writing has changed between these two versions, maybe fundementally?

Thanks,
Brian