* we only need one piece of code to calculate rsl_ie_chan_mode from
our run-time data structures (gsm_lchan)
* add some more channel modes for TCH/H and data
* use enum's to make the compiler warn us about unhandled enum values
* make sure the caller determines the (signalling,speech,data) mode
only after the LCHAN_MODIFY we know the final mode of the channel,
so we have to postpone our IPAC_BIND until then to make sure we set
the correct speech codec.
Up until now, we only supported direct RTP streams between ip.access BTS.
With this commit, the user can specify '-P' to the command line to enable
a RTP/RTCP proxy inside OpenBSC. The nanoBTS will then send all their voice
data to OpenBSC, which will relay it to the respective destination BTS (which
can be the same BTS).
The default behaviour remains unchanged. Without '-P' on the command line,
RTP/RTCP is exchanged directly.
* explicitly set the "ip speech mode" IE during BIND and CONNECT messages,
depending on the speech codec used by the voice call
* more verbose debug messages regarding IPAC_BIND and IPAC_CONNECT
* do not always blindly specify RTP payload type, but use the value
returned by BIND_ACK, _if_ it is present.
The rtp_proxy.[ch] code is intended to be used as a transparent
RTP/RTCP proxy, relaying the media streams from one ip.access BTS
to another. In an 'ideal' network, this is obviously not needed,
since the BTS's can send those streams directly between each other.
However, for debugging, 'lawful interception', transcoding or interfacing
a TRAU/E1 based BTS, we actually need to process those RTP streams
ourselves.
There were many places in the code where we had to explicitly
reference the transaction_id and put it into a packet. By introducing
and optional gsm_trans parameter to gsm48_sendmsg(), we can implement
this code once rather than dozens of time.
since a subscriber is an element of the gsm_network, we have to ensure
subscr->net is always set correctly. We do this by using gsm_network
as an argument to all functions that resolve or create a subscriber.
Since a transaction is associated to a gsm_subscriber, and the subsciber
is part of a network, we don't need to have a dedicated transaction->network
pointer.
This changeset factors out gsm_transaction as something independent
of call control in preparation to re-use the code from SMS. A
transaction is uniquely identified by either its callref, or by
a tuple of (transaction_id, protocol, subscriber).
Currently we send the attribute changes in a send and forget
fashion. But sometimes the nanoBTS is sending us a NACK, e.g
with a invalid unit id. Start handling the NACK and provide
an error message to the user. The error message is not yet
describing the cause of the error but this is a slight progress
to the previous silent failure.
For further evaluation/analysis, this patch stores the classmark 1, 2 and 3
values of every equipment in the SQL database. We can use this non-volatile
data to determine the supported features for each handset that we've ever
seen on our network.
As Dieter has pointed out, we currently send incorrect information
in the rest octets, particularly about our GPRS capability. Since
the format of the rest octets is highly complex, and we don't
actually need any of those features yet, we might just fill them
with padding.
If we receive one of those strange BS-11 "Cause 22" errors, we don't need
to check if the lchan use counter is > 0. If it was 0, the lchan gets
released anyway.
This was proposed by Andreas Eversberg. I made it conditional on the T200
timer expired cause, as I'm not sure if we really should give up that quickly
on other errors such as just simply receiving an unsolicited response.
Harald Welte
Holger Hans Peter Freyther