* Include stdlib.h before the snmp headers to have a free declaration
* Use sprintf(dest, "%s", str) to avoid format string attacks
* Avoid bogus assignment. This pattern was used for marking something
as unused in the past.
This is a hot fix to make CIC reading (and later status) work on
big endian machines. There might be a more elegant way to do it
and I will explore this later.
This way one can check the logs if this code path is hit. The
logging normally only occurs for the two messages we are interested
in. It doesn't make sense to reset with a range of 0 so the code
has been re-ordered.
GCC 3.x on PowerPC correctly highlights that the code is fishy.
Re-reading the RFC 3550 shows that we should subtract it and then
we are in the 16bit range. The probation and re-sync code is still
missing.
GCC:
mgcp/mgcp_network.c:200: warning: comparison is always true due to limited range of data type
For the ISUP/MGCP handling we will need the same code, extract it
from the msc_connection. For the reading code callback is introduced
that will pass the MGCP message to the higher layer.
The RSIP has morphed from a global reset, to a per trunk reset and now
it is possible to reset specific ranges on a trunk. This will be used by
the ISUP filter code in the STP.
For legacy range == -1 will be used. This will reset all endpoints
on the trunk. Use OSMO_MAX on endpoint and number of endpoints in case
number_endpoints is 0.
This code will now free everything from the endpoint to endpoint + range
including endpoint+range.
In case the endpoint is allocated and the VTY is used to force it
to be freed use the realloc_cb to make sure that any resources
allocated to it will be freed. The mgcp_ss7 binary will then call
mgcp_ss7_endp_free to free audio encoder resources.
There were several changes in the upstream code. These include
statistics, DTMF/RQNT, changes in the parsing code and re-transmission
handling. The last item is the main reason to do the merge now.
Create a simple queue for pending DTMF tones, play them using the
MTN API, and then send the next tones once the playback is complete.
The callback and scheduling is done from the same context so no locking
needs to be done.
Introduce a callback for the request and forward the signalrequest
to the callback. This is not a full implementation of MGCP RQNT.
Manual merge and backport from OpenBSC.
In case MtnSaAllocate failed the code did not restore the audio routing
when destroying the endpoint. Move the hw_maybe_loop_endp out of the if
statement.
This got broken in the previous commit. Use the hw_dsp_port we assigned
to the endpoint instead of the dsp_resource that is one ahead. For the
non loop-on-idle configuration audio routing appears to be broken.
* Change the assigning dsp_resource to start with one and increase
after usage and not before usage (so it becomes the next free
port).
* Subtract one from dsp_port after we have selected the path. In the
longer run I should make dsp_resources zero based all the way.
For a linkset define where SCCP/ISUP should be send. This config
should probably move up to the application part when real work on
the routing is done. Right now the sccp_opc/sccp_dpc need to stay
inside the mtp_layer3.c to be able to send a TFA for the reachable
OPC and it is easier to keep both (dpc/opc) in the same file.
* This changes bss_patch_filter_msg to return -1 or BSS_FILTER_DTAP
for DTAP messages. This way app_forward_sccp should continue to behave
the same besides now looking into DTAP messages.
* Introduce a direction in case we want to advertize FR into the BSS
side and HR into the other direction.
* Patch AMR HR3 and Fullrate/Halfrate capabilities in the Bearer
Capabilities. Add a test case that is patching the bearer capabilities
The current ping/pong timeout is 0 which means the MSC connection
will be taken down almost immediately. Set it to -1 to disable sending
pings and waiting for the pong.
MGCP RFC 3435 does not specify that the Connection Id must be
generated with any kind of random. It must uniquely identify
the connection of an endpoint. So we can make it per trunk group
or could even have it per endpoint.
The code does not support multiple connections on the same endpoint
right now but the spec allows it.
It can be difficult to find the Timeslot/Multiplex for a higher
number virtual trunk. This would be used by default, but normally
the endpoint would be blocked on the switch already.
Instead of assuming which endpoints are blocked there is now a VTY
command to block those. Clean up the init of the trunks, the only difference
between Virtual and E1 is in the way to calculate the start port.
Reduce the number of endpoints to 32, 31 is the last one that can be
used on the E1 trunk, otherwise we move into TS 0 of the following trunk.
The endpoint offset is needed for two reasons, first the API is 0
based here while we are normally 1 based, second because of the trunks
the first usable endpoint would be '2' (0 is CRC, 1 is signalling), but
this endpoint offset falls apart when we would block timeslots inside
this range.
Remove the endpoint offset, in each endpoint we will store the HW DSP
Port (1 based API) and then subtract one to get to the 0 based API for
the Simple API. Print a warning when someone is using the endpoint offset.
The MSC workaround was added in 5960ba387a
but it has never worked as in 8fd28dbbe6 (earlier)
we were checking for link->conn != conn in the dispatch method. Move the
code over to the generic dispatch and check for NULL.
It appears that it is possible to have a stale SCTP connection
and this added LOGL_NOTICE and the VTY interface might help to
identify this situation in the future (the mean time of failure
is about five month).
Instead of hardcoding which timeslot is blocked we will just
use the blocked flag in an endpoint. This should fix call
handling for CIC on the trunk config.
We don't let CGUAs pass when handling circuit blocking and
unblocking locally. But we did let a CGU go through and then
we never sent the response back to the sender. Respond to a
CGU with the same content.
We need some way to forward the failure of one link to another but
they are not normally routed so we can not send a TFP. Right now we
will simply stop responding until both links are up. This should make
the SLTM fail and trigger a re-alignment on both sides. The key here
is that the 2 * SLTM timeout needs to be higher than it takes to re-align
the link. I'm not sure this code will work.
After the expiry of T18 we should have collected the routing data
from the adjacent links and should be able to send SCCP packages
to remote endpoints.
The MSC we test is not sending an ASP Active when the
link is unblocked. If the m2ua_link has no connection
associated we will forgive the MSC and active it.
Right now for the virtual trunk 0x0 and 0x1F is blocked, for the
E1 like interface we have 0x0 and 0x1 blocked. This should start
to be configurable in the future.
Configure the routing of audio ports if mgcp_mgw is configured
to do this. This allows to have multiple trunks, make virtual
ports go to a specific trunk as well.
With this commit we can have more than 30 endpoints that will work. We
ignore the blocked endpoints 0x1 and 0x1f for each trunk and calculate
everything from the right start point.
* Upstream has a separation of BTS and NET side for RTP ports and
can allocate them dynamically.
* Upstream has gained the concept of trunks. We will now have various
trunks to connect audio things.
* We will now be able to utilize multiple trunks and have the endpoints
used properly.
Holger Hans Peter Freyther
Harald Welte