Hi Ihor,
Attached is a patch for doublechecking the AF of the forced socket - if
not matching, it will be discarded and a new socket will be selected
based on proto and AF family. It is not ideal, but let's see.
For doing DNS at the script level, see the IP transformation (it is what
you worked on previously):
https://www.opensips.org/Documentation/Script-Tran-3-6#ip.resolve
I agree, it might be useful to have it extended to (a) control the AF
and (b) to return maybe a list with all the IPs.
On a much broader view, talking with Razvan on this topic, we were
brainstorming on a more generic and flexible approach for this kind of
problem (trying to force from script some kind of outbound socket, but
without knowing (yet) all the details on the destination (like proto,
port, IP). The idea here is to come up with a kind of phony socket
definition (phony as it will not be used for listening on) to act as a
routing table, to allow the selection of a real socket based on a later
determined port, proto, etc.
Like (in a json format):
socket = my_socket { AF=4:udp:1.2.3.4:5060, AF=6: udp:[2001:db8::1:0]:5060}
or something more complex, with proto filtering too:
socket = my_socket { AF=4,proto=udp:udp:1.2.3.4:5060, AF=6,proto=tcp:
tcp:[2001:db8::1:0]:5060}
The idea is to still have at the code level a `struct socket_info`, but
to be interpreted in a more generic way.
Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
https://www.opensips-solutions.com
https://www.siphub.com
On 08.01.2026 18:33, Ihor Olkhovskyi wrote:
Bogdan,
Thanks for this, I'm for sure will be happy to test a patch.
For the DNS/socket/AF, as I've said, I'm using now explicit resolution
of address with AF (resolve4 and resolve6) and acting accordingly.
Still, sort of solution I see here -
https://www.kamailio.org/docs/modules/devel/modules/ipops.html#ipops.f.dns_query
, it also allows to deal with multiple A and AAAA records.
Le 08/01/2026 à 16:55, Bogdan-Andrei Iancu a écrit :
Hi Ihor,
First, without the `dns_try_ipv6` options, the DNS lookup will never
return an IPv6 address (as gethostbyname() will be used, which is
AF_INET only).
Now, going back to the original report: having the `dns_try_ipv6`
option on, you get back an IPv6 address (from the DNS resolver). Now,
checking the code, if I'm not missing something obvious, I would say
the code allows sending to an IPv6 destination via an IPv4 socket :-/
...which seems broken, IMHO. The code verifies only the compatibility
(between destination and sending socket) when comes to the protocol,
but not AF.
For this case, will you be willing to test a small patch? this will
discard the forced socket if not compatible with the destination
(this is also done for the proto mismatch) - and a new matching
socket will be looked up.
Now, about the multi IPs from DNS - if the DNS based fallback is not
turned off (as flag to t_relay), opensips will try to do dns based
failover (if the sending fails). But this will not change whatever
outbound socket was set...
Basically, to wrap this up, your dilemma is like this: you want to
force a socket, but the DNS may give IPv4 or IPV6, so you have no
idea which socket to force...
Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
https://www.opensips-solutions.com
https://www.siphub.com
On 02.01.2026 13:27, Ihor Olkhovskyi wrote:
Bohdan,
I've looked through code (resolve.c) and got an impression, that
only first resolved address is used along the way. Means if
destination address resolves to IPv6 with dns_try_ipv6=yes, it uses
only IPv6 in packet processing (namely for $du) and you cannot force
IPv4 outbound socket later, cause no info about IPv4 is present.
I've made a small hack to have a possibility to choose, which
address family I want to use
(https://github.com/igorolhovskiy/opensips-v6/commit/527d8d3d010d3a5a1e4b68475df45cb4e9f44604)
but I don't like the solution.
Maybe I'm wrong, but my impression, that when DNS resolution is
happens, only 1 address (which includes the family as well) is being
used later with no possibility to change it.
I can provide more logs soon (for relaying part), but not sure it's
the case here.
Regards,
Ihor
Le lun. 29 déc. 2025 à 10:33, Bogdan-Andrei Iancu
<[email protected]> a écrit :
Hi Ihor,
So, you say OpenSIPS picks the IPv6 address as destination
(confirmed by the err log), but you are not sure what interface
is to be used (even if you forced the ipv4 one, it may be
discarded by the internal fwd logic).
Could you provide debug level logs for the relaying part ?
Regards,
Bogdan-Andrei Iancu
OpenSIPS Founder and Developer
https://www.opensips-solutions.com
https://www.siphub.com
On 10.12.2025 17:58, Ihor Olkhovskyi wrote:
Hello!
I'm trying to build ipv4/ipv6 bridge with OpenSIPS 3.6.2.
I've declared the sockets like this:
# v4 addresses
socket=tls:OPENSIPS_IPV4_ADDR:TLS_PORT
socket=wss:OPENSIPS_IPV4_ADDR:WSS_PORT
# v6 addresses
socket=tls:[OPENSIPS_IPV6_ADDR]:TLS_PORT
socket=wss:[OPENSIPS_IPV6_ADDR]:WSS_PORT
dns_try_ipv6=yes
# Always listen on localhost v4
socket=tls:127.0.0.1:TLS_PORT
Means explicit to start on v4 and v6 addresses for lately to
change address family explicitly
OpenSIPS starts without any issues, but when I try to do socket
forcing with
$socket_out = "wss:OPENSIPS_IPV4_ADDR:WSS_PORT";
t_relay();
I'm getting
ERROR:proto_wss:ws_sync_connect: bind failed (22) Invalid argument
ERROR:proto_wss:ws_connect: connect failed
ERROR:proto_wss:proto_wss_send: connect failed
tm:msg_send: send() to EXTERNAL_IPV6(sic!) for proto wss/6 failed
tm:t_forward_nonack: sending request failed
So OpenSIPS is trying to send over IPv6. I guess that is cause
network address of RURI is v6 already, but DNS answers in
dual-stack. If I remove `dns_try_ipv6` it works, but how than I
should work with v6?
What am I missing?
Thanks in advance!
--
Best regards,
Ihor (Igor)
_______________________________________________
Users mailing list
[email protected]
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
--
Best regards,
Ihor (Igor)
--
Regards,
Ihor
diff --git a/forward.c b/forward.c
index c123cdf8c..2704cb6c6 100644
--- a/forward.c
+++ b/forward.c
@@ -154,11 +154,16 @@ const struct socket_info* get_send_socket(struct sip_msg *msg,
msg->force_send_socket=find_si(&(msg->force_send_socket->address),
msg->force_send_socket->port_no,
proto);
- }
+ } else
+ if (msg->force_send_socket->address.af!=to->s.sa_family){
+ LM_DBG("force_send_socket of different AF (sock=%d, dst=%d)!\n",
+ msg->force_send_socket->address.af, to->s.sa_family);
+ msg->force_send_socket=NULL;
+ } else
if (msg->force_send_socket)
return msg->force_send_socket;
else
- LM_WARN("protocol/port mismatch\n");
+ LM_WARN("protocol/port/af mismatch\n");
};
if (mhomed && proto==PROTO_UDP)
_______________________________________________
Users mailing list
[email protected]
http://lists.opensips.org/cgi-bin/mailman/listinfo/users
