Here is our pjsip_endpoint.conf:
outgoing_call_offer_pref = local_merge
incoming_call_offer_pref = remote_first

For outgoing calls we use local_merge (UNION) which means "Merge the lists with the preferred
codecs first"
For incomming calls we use remote_first (INTERSECT) which means "Only common codecs with the
preferred codecs first"

When UPDATE was received for incoming call, device responded with
488 Not Acceptable Here with Reason:Q.850;cause=16
When UPDATE was received for outgoing call, device responded with
488 Not Acceptable Here with Reason:Q.850;cause=58 
BYE
 
After this commit we have the same behaviour for UPDATE during both incoming and outgoing call.

Incoming calls were fixed in similar way before in fb501964.

The workaround is to send one CNG packet to the negotiated remote address and port. Why
asterisk refuses to receive/accept incoming RTP packets until at least one RTP packet
is sent remains unknown.

The commit contains 2 fixes making the authorization header is cached during initial registration
and then reused for further re-REGISTER requests.
- Enable flags in pjproject so the authorization header is cached and then automatically included
  in next re-REGISTER requests.
- Set initial registration outbound authentication credentials properly so the following error in
  pjproject is fixed and authorization header is cached properly:
  pjproject: <?>: sip_auth_clien ...Unable to set auth for tdta0x: can not find credential for
  X.X.X.X/Digest

- Add Require=replaces, user=phone into Refer-To header field
- Remove to-tag, from-tag, from Refer-To:replaces

Note that the changes only make the test move forward a bit. But external transfer
as a feature still does NOT work after this commit. The reason to accept this partial
change is that it has been pushed into the customer's git repo and we have to keep
the code synchronised.

Before this fix, the emergency registration (e.g. UDP) tries falling back to the configured
transport (e.g. TLS) even if there is a subsequent emergency call ongoing.
After this fix, the emergency registration doesn't fallback to the configured transport until
the current registration expires and there is no emergency call ongoing.

* do not de-register while hangup an emergency-call through emergency-register, and
  reset to primary under next registration instead.
* accept incoming call through emergency-transport during emergency registration period

Check for NTP synchronization for a maximum duration of ~90 seconds.
The reason for around 90 seconds is that in a case where TLS fails for some other reason, (not
due to lack of NTP synchronization) such as that none of the 3 servers can support TLS, it
will take up to 32+32+32 seconds to fail over to UDP following DT normal fallback algorithm.
If it takes the same amount of time this in this case, it should be ok, and easy to argue for.
Currently it takes 10 x 60 seconds which is way to long.

This commit fixes the below issue.
When there are no resolved addresses (tdata->dest_info.addr.count == 0), no registration attempts
occur (client_state->failures == 0), then transport2 is never tried.

registration_resend() simply does nothing and returns 0 when client_state->
registration_resend_timer_id is -1.
This commit creates a 1s timer and causes registration_resend() to do the actual
work.

This is used to validate the source address of a SIP request from LAN side.

- Update the flag endpoint->failover_reg_addr with 3 states, 0/1/2 ->
  none/504-register-ongoing/504-registerfailed
- Add the 504_flag into the pjsip show registrations which will be used by
  chan-voicemngr to decide if an INVITE should be sent after 504.
- Restore the original client_state->registration_expires if 504-register failed.

A record which has the unspecified transport type is considered as invalid.

* 408 response received from the server should be handled the same as other codes like 503
* Introduce response code 444 from pjsip while no response been received to differentiate
  it with 408 which is from the server
* there should have no retry during recovery flow if no retry-after, retry should only
  happened during the first round if no retry-after.

* Add actual registration address:port into debug log
* Add latest registration address:port into `pjsip show registrations` CLI to avoid
  confusion

The CLI shows below before the changes, while TLS(failed) falls back to UDP(registered).

<Registration/ServerURI>             <Auth>  <Status>
=======================================================================================
 sip0/sip:tel.t-online.de:5061       sip0    Registered     (exp. 376s)

After the changes.

<Registration/ServerURI(Primarily)>  <Auth>  <Latest Registration>  <Status>
=======================================================================================
 sip0/sip:tel.t-online.de:5061       sip0    192.168.2.150:5060     Registered

* fix the invite sent to wrong address other than the registered one
* fix the un-registration issue after failover during emergency registration

Rebuild the structure of E-reg flags and un-reg states, skipped unwanted request
message while hangup before registered.

If the 'endpoint->fromdomain' config is an IPv4 or IPv6 address then each
incoming request is checked if it is from this address.
If the 'endpoint->fromdomain' is a host name then we are checking the
DNS cache list and allow only incoming requests from the address in this
list to be accepted.

* registration retry to the same address in 15s when response with no re-try, and fail
  over to next address after 2nd attempt.(#15581)
* register request failure count to each single request that been sent out.(401/407
  resend not count)
* update a bit with the upper bound wait time log and update conditions to avoid
  redundant calculation.
* fix request uri to make it consistency.(due to 7349c566
  from #15901)

An emergency call might cause a SIP client fall back to the emergency transport if it is not
registered. After the emergency call ends, the client tries to register via the "normal"
transport. But the Expires header is missing in the first REGISTER requests and this causes
the default expiration value from the registrar side is being used.

This commit adds an Expires header with the configured expiration value in an outgoing
REGISTER request if the header is absent.

Fallback to the transport for emergency only when the current transport is different.

The available transports were not updated, which caused error when
resolving for new enabled transport.

Add validation check to prevent NULL pointers from being accessed.

Grzegorz Sluja authored 2 months ago and Yalu Zhang committed 2 months ago

It occured that changing the condition for AST_CONTROL_RINGING
in a fix for REF 15850 introduced a degrade so the '180 RINGING'
SIP event is sent many times for the incoming call processed to DUT.
Restore the previous condition and fix the issue of early-media with
reset the flag 'pa->sentringing' in other place makes the issue of
early-media REF 15850 is fixed and there is no longer multiple
'180 Ringing' SIP messages sent to the Caller.

The attributes of the rejected media stream were missing in SDP answer before. With this
commit, all attributes from the SDP offer for the rejected media stream are kept in the
SDP answer. 

Add and support max_sessions to the pjsip endpoint config, which reflect as `VoiceService.
{i}.SIP.Client.{i}.MaxSessions`.

No actual behavior changes have been done, just followed the same behavior and add an extra
condition check after max session per line as the line_calls count was calculated per
endpoint.

Retrieve XML content for INVITE response 504 with application/3gpp-ims+xml,
and trigger registration recovery flow to the next address when XML content
includes <action>initial-registration</action>.

- Not trigger re-registration when Status 500 received for outgoing Invite

- The emergency call flow is kept, which may trigger re-registration to
  recover from faulty status.

'use_callerid_contact' is set to 'yes' by default in our configuration. However,
the Contact header did not match the Caller ID because it was replaced with
`session->endpoint->contact_user`.

Change the value from sendonly to supported.