Fix a bug introduced in commit
implement CM Re-Establish for voice calls
ae98b97382
Neels Hofmeyr <neels@hofmeyr.de>
Thu Jul 29 22:40:59 2021 +0200
I6fa37d6ca9fcb1637742b40e37b68d67664c9b60
We should only succeed when conn_accepted == true!
Related: SYS#5130
Change-Id: I3679162143e8d7d8c0878de2102faa11eadfccfc
When a subscriber first attaches by TMSI only, and later tells the IMSI
via ID Response, it may turn out that this IMSI already exists in the
VLR database. If this happens, the TMSI that the subscriber issued was
not known in the existing VLR entry, indicating that the subscriber has
in the meantime camped on a different core. Which means we can assume
that there cannot be any active connections, and the old subscriber can
be discarded, for the benefit of the new one.
(We could also discard the new one, but it is more complex to reparent
the ongoing FSMs for Compl L3 than to copy some dormant VLR state.)
In vlr_subscr_set_imsi(), check for an existing IMSI entry in the VLR.
If such exists, copy any pending Paging and auth tuple state to the new
subscriber, and discard the old one from the VLR.
In order to safely discard a vlr subscriber by force, add a new vlr_ops
function: subscr_inval(), to tell the MSC that a vlr_subscr is no longer
valid.
Upcoming patch I583682d1a35a70b008d7bb2d89ba7c3109a60b21 better clears
TMSI state from the VLR, making it more likely to hit the evil twin
situation this patch fixes; hence this is, sort of, preparation.
Related: SYS#6860 OS#4721
Change-Id: Ifdabe0b65bffafbf7b8e5cc10e2d225d1ed1cecd
We have an msc_conn_ref pointer from vlr_subscr to an active msc_a
instance. So far, we just keep it pointing at discarded memory. Instead,
make sure it goes back to NULL when the msc_a instance deallocates.
This way the VLR can reliably tell whether a given VLR entry still has
an active connection or is just inactively caching the subscriber.
Related: SYS#6860 OS#4721
Change-Id: Ic63d01d220b63453976fe06a7c6b606f97172c99
This commit simply fixes a -Wenum-conversion thrown by clang.
No idea why are we using the SM (GPRS Session Management) cause values.
msc_a_release_mo() does not even use the given SM cause value.
Change-Id: Iade6bf97466ab2b3b39e9ea123fc90d06c0f6a9b
I'm not sure why so many files (particularly written by Neels)
did contain a GPLv2+ header, instead of the AGPLv3+ which is the
actual overall project license. I consider it a mistake.
In any case, any copyrightable contribution to those files was done by
sysmocom employees, so I as managing directory can legally make a
license change, whther or not it was a mistake early on or not.
The only GPLv2-or-later file remaining is mncc_internal.c, as it has
more contributors and a longer history.
Change-Id: I8650697592b3160c4d0a7c61ae9c46d4aacb3bef
Better match the pattern of sdp_audio_codecs_* instead of having
foreach_ in the front. Prepare for prepending osmo_ some day, because I
plan to move the SDP API to a separate library.
Change-Id: Ia96190e0bdb513886663be1c8c12be3b403b71c9
When we get the codec filter result logged, it is most interesting to
know the caller. So wrap a file-line macro around trans_cc_filter_run().
Change-Id: I243404487c1871e921b08098086ef2fc78a5561d
Currently the CSD check is in the middle of figuring out the voice codec
for normal voice calls. Rather do the CSD check first, and then do voice
in one coherent section.
(prep for upcoming change in this code, to support AMR rate selection.)
Change-Id: Ibd21f0bb46c66a406904105564ce961a8760cbe7
Before the codec filter, it would have been the CN side codec, but now
it is only the codec that the RAN reports as assigned, fed into the
codecs filter.
(prep for upcoming change in this code, to support AMR rate selection.)
Change-Id: Ie7966099c5565013018734b0c2028484c24341a7
low/high layer compatibility are used for capability checking between
caller and called entitiy. The transcoding is performed by libosmogsm.
Related: OS#6152
Depends: libosmocore.git Ia6a2159ecf810a02f85b558026edf20b934567de
Change-Id: I760980a7e17e2fa81615adc69ef85797eb0c07f1
We do include this IE in result and error messages, but somehow
not in the request messages. For the sake of consistency, let's
ensure that the Source Name IE is present in all SMS related PDUs.
This additionally brings osmo-msc in sync with ttcn3-msc-test, which
was modified to expect the Source Name IE in all receive templates,
and makes the following testcases pass [again]:
* MSC_Tests.TC_gsup_mo_sms
* MSC_Tests.TC_gsup_mo_smma
* MSC_Tests.TC_gsup_mo_mt_sms_rp_mr
Change-Id: I65f5e3b7a0688e258979bb2679598659881a4321
Related: osmo-ttcn3-hacks.git Ic24d3082fe3dce08e43e8f3ecb6d6132503c55c6
Related: OS#6135
... so that it's clear which MNCC handler is used by looking
at the output of `show running-config`.
Change-Id: Id1fe7aecc1c8445db48ff5fddcf6df0f05ba5e2e
Prior to this change, if there was no explicit ipa-name configuration
in OsmoMSC, OsmoHLR would see the GSUP connection as MSC-00-00-00-00-00-00.
However, this default is constructed somewhere deep in IPA libraries
and is not visible to the GSUP client application, in this case OsmoMSC.
This situation creates a problem for SMS-over-GSUP routing: when we get
MT-forwardSM.req from an SMSC, we have to send a GSUP response, and this
response needs to get back to the MT-sending SMSC. Because OsmoHLR
applies only passive routing for these responses, we have to set
source_name when generating MT-forwardSM.res in OsmoMSC - but we cannot
do so if don't know our own IPA name.
Change the default OsmoMSC ipa-name from MSC-00-00-00-00-00-00 to
unnamed-MSC, mirroring OsmoHLR default of unnamed-HLR, and set it
at our application level rather than deep in the libraries.
Related: OS#6135
Change-Id: I7bacd001b81326c32bc262c7d0c0491ded822fa8
This is the last missing piece that allows osmo-msc to make good TFO
codecs choices.
Since the codec_filter, osmo-msc properly gathers codec options and
limitations. But the MO call leg still assigns a voice channel before
getting a response from the MT call leg, and is then stuck with that.
Add the capability to adjust the MO call leg's codec in case the MT side
needs a different codec for TFO.
This is only relevant for 2G; on 3G we always have AMR/IuUP.
For inter-MSC handover, keep the behavior unchanged: offer only the
currently assigned codec to the remote side. Codec-changing HO should be
equally trivial to implement, but that is for another day.
msc_vlr_test_call's codec tests are adjusted to test the new feature in
Ib933554f826c1b4347dfa3f6c4f6fe086be8b133. For now, avoid change in
these tests by validating the first codec in SDP lists only.
Related: OS#6258
Related: osmo-ttcn3-hacks I402ed0523a2a87b83f29c5577b2c828102005d53
Change-Id: I8760feaa8598047369ef8c3ab2673013bac8ac8a
Used by I8760feaa8598047369ef8c3ab2673013bac8ac8a to add just a single
codec to a speech codec list, instead of a list.
Change-Id: I6ac23c54bc26939e048ff2df06eb987421cfb1c5
To parse and handle SDP included in incoming MNCC, use rx_mncc_sdp()
everywhere. So now rx_mncc_sdp() is the single implementation for
parsing the SDP string and taking action for codecs if needed.
One current dup of this code has a fall-back to use legacy bearer cap --
absorb that into rx_mncc_sdp(), so that we now also do that fall-back
for all of the incoming MNCC that contains bcap.
This is a cosmetic preparation for implementing MO Re-Assignment to
match MT's codec limitations.
Change-Id: I94ae11654e1f88fbd64361b639a4c583836dc13e
We're checking the result of trans_alloc() 6 out of 7 times, so check
it in gsm_silent_call_start() too, for the sake of consistency.
Change-Id: Ie989cd8146d66d9531cf3f3d84f46a2c6fcc2e5c
Fixes: CID#322140
"[ESTABLISHED] transition to state ESTABLISHED not permitted"
i.e. don't complain when we already are in the established state.
Change-Id: I9b1fd63ed1ee7ed2877a4b2059386354598f4ea4
The cfg bits are for AMR-HR, not GSM-HR. The function
gsm0808_enc_speech_codec_list2() will return -EINVAL when it encounters
GSM-HR with non-zero cfg bits.
It appears this mapping was never used before, and my testing of call
re-assignment to match MT's codecs (it allows more than just the
assigned codec, because it can re-assign) has uncovered this bug
via MSC_Tests.TC_ho_inter_msc_out. I don't fully understand all the
details why we didn't see this before; anyway, the fix is obvious.
Change-Id: I19cff847a0f618ad000d12c1df54c55ef2f79699
Only the originator may terminate the VGCS/VBS call. This will not
happen in real life, because the UI of the MS should not allow
termination of a recevied VGCS call.
Change-Id: Ibe289920fa3ea50dd3e7d5c1371456dca9b72604
Related: OS#4854
Certain calls (seen on very old Nokias) won't have the rate adaptation flag
set on "analog" CSD calls. The field for the intermediate rate (after RA) is
still filled correctly.
Workaround this by setting the RA to V.110 whenever the RA is unset but an
intermediate rate is specified.
Change-Id: I5b3e5649fe071636f1becddfbfee06f9175a5f17
Bearer capability 3k1_AUDIO and FAX_G3 are only important
for the interworking function, the MSC should handle
these calls the same as CSD calls with unrestricted digital
bearer capability.
Change-Id: I198aa867a8f236b8ddd05d3b2356f64b876fd4c1
For MO-forwardSM and MT-forwardSM request messages, OsmoHLR applies
routing based on the SMSC address for MO or based on the IMSI for MT.
However, reply messages following these requests are routed passively
based on the destination_name IE. This passive message routing path
requires the source_name IE to be set as well - implement this
source_name setting.
Related: OS#6135
Change-Id: I0b7f4760bdce8a38d43d3860086c6dfb7b390701
When OsmoMSC is used with OsmoHLR rather than a GSUP-to-MAP gateway,
MT-forwardSM.req GSUP messages delivering MT SMS will be coming from
a separate SMSC relayed via OsmoHLR, rather than from OsmoHLR itself.
When we reply to these messages, in order for these replies to reach
the MT-sending SMSC via OsmoHLR, we need to save source_name from
the request and regurgitate it into destination_name in our response
messages. Implement this logic.
Related: OS#6135
Change-Id: I436e333035b8f6e27f86a49fe293ea48ea07a013
If the GSUP request message to which we are replying is an MT SMS
delivery from an SMSC relayed via OsmoHLR, we must set destination_name
in our reply - otherwise our reply won't make it back to the SMSC.
Related: OS#6135
Change-Id: I892fe87a733a78ed9d5761a8ce238caa135dea1e
The intent of the guard timer is to clear hung or stuck states
during call setup or teardown. However, there are some MNCC
messages that will be exchanged between OsmoMSC (passing CC
messages to and from the MS) and the external MNCC agent during
the active call state, not related to setup or teardown: DTMF
start and stop, plus call hold and retrieve operations for call
waiting. Unpatched OsmoMSC restarts the guard timer on every
received MNCC message, even those that pass through to CC without
affecting any state, and the result is breakage for users.
Consider the case of an IVR where you have to press some DTMF keys
before you can be transferred to a human operator. You press the
needed keys, get the human operator, and start talking. Then
3 minutes into your conversion (default guard timer duration)
your call unceremoniously disconnects without any warning.
Fix: look at the MNCC message type, and skip the call to start
the guard timer for known-benign MNCC messages.
Change-Id: Ibe2dd53f8e9e06d175b64df67d2a2e3e2d4155aa
This is a fixup for the patch
'3G: decapsulate IuUP to AMR at the MGW; allow 3G<-AMR->2G'
I386a6a426c318040b019ab5541689c67e94672a1
After above patch, osmo-msc intelligently decides which codecs to run on
which legs of the RTP streams. In the meantime, it seems the necessary
matching changes to call_leg_local_bridge() had been lost somehow.
Testing 3G to 3G voice now, I noticed that call_leg_local_bridge()
overwrites the intelligent choices made earlier.
The history of an MGW endpoint that should convert from IUFP to plain
AMR, extracted from a pcap, looks like this:
<- CRCX None None
-> CRCX-OK audio 4050 RTP/AVP 112 None
<- MDCX audio 4056 RTP/AVP 112 AMR
-> MDCX-OK audio 4050 RTP/AVP 112 AMR
<- MDCX audio 4056 RTP/AVP 96 VND.3GPP.IUFP
-> MDCX-OK audio 4050 RTP/AVP 96 VND.3GPP.IUFP
So after call_leg_local_bridge(), there is an extra MDCX + MDCX-OK that
switches the codec from 112 AMR back to 96 IUFP.
That is because call_leg_local_bridge() copies the *RAN* side's codec to
both CN sides, which used to be ok when RAN and CN codecs were always
identical.
Instead, adjust only the CN sides of the MGW endpoints, and adjust them
so that both CN sides are identical. osmo-mgw should then be able to
trivially translate the codecs appropriately.
Change-Id: I130bcd77ec57e332370c487a11b0b973b6e1089d
Fail if MNCC tries to switch the Information Transfer Capability from
CSD to speech, so it is obvious that something is wrong here. I ran into
this while writing a test.
Related: OS#4394
Change-Id: Ibb76d08cad1ac3bc3320391c89766150a2e605c3
Reject any other codec than GSM0808_SCT_CSD in Assignment Complete from
RAN, if OsmoMSC is preparing a CSD call.
Related: OS#4394
Change-Id: I94de84df41bcd050d0e7b4e4fea1c6a6551ef7d3
Instead of asserting on an empty list of bearer services, return
-EINVAL. This makes the function more similar to
sdp_audio_codecs_to_gsm0808_channel_type which also doesn't assert if
an empty list of codecs is passed.
Related: OS#4394
Change-Id: I15a389e1f7a9d3d17b6531c9836d3d5f9d148267
The MS in general provides the Selected PLMN ID (IE) in the Complete
Layer 3 Information message. osmo-msc handles that message in
msc_a_ran_dec_from_msc_i() and stores the information of the PLMN in
msc_a->via_cell. If no PLMN information is provided in the message, then
at that same place the PLMN configured in the VTY is taken as an implicit
default.
This patch changes trans_lcls_compose() to use the PLMN stored in
msc_a->via_cell instead of the VTY configured one, meaning the PLMN
provided by the MS (through the RAN in use) is used if available
(otherwise the VTY-configure one is still used, as before).
With this patch the PLMN VTY config option use is relegated to a single
point of use in msc_a_ran_dec_from_msc_i() where the Complete Layer 3
Information is used. As a result, it becomes clear now that the VTY
config is only applied in the scenario where no PLMN is provided at that
time.
Related: SYS#6360
Change-Id: Ibad0005a1d7cef64dd8fefa3e554ba99a06c3666
The MS in general provides the Selected PLMN ID (IE) in the Complete
Layer 3 Information message. osmo-msc handles that message in
msc_a_ran_dec_from_msc_i() and stores the information of the PLMN in
msc_a->via_cell. If no PLMN information is provided in the message, then
at that same place the PLMN configured in the VTY is taken as an implicit
default.
The PLMN information stored in msc_a->via_cell is then finally stored
into vsub->cgi in evaluate_acceptance_outcome().
This patch changes gsm0408_loc_upd_acc() to avoid re-applying the PLMN
configured at the VTY again, and instead use whatever is already in
vsub->cgi. This is more correct since the PLMN provided by the MS takes
precedence over the implicitly configured one, meaning several PLMNs can
be handled. Otherwise, the code is always overwriting the PLMN announced
by the network on a specific RAN with the one in the MSC, which may end
up with unexpected results.
Related: SYS#6360
Change-Id: I421bd63a264db2bf6e1c4a4eea976f389e87b332
Currently this function fails to initialize all bcap fields properly,
so the resulting CC Setup message generated by osmo-msc has some
fields set to reserved/invalid values.
With these changes I am able to establish a data call on TCH/F9.6:
* cap->{mode,coding}: assign default values explicitly;
* cap->radio: value 0 is reserved, set GSM48_BCAP_RRQ_FR_ONLY;
* cap->data.sig_access: value 0 is reserved, set GSM48_BCAP_SA_I440_I450;
* cap->data.transp: this is not a bool, set GSM48_BCAP_TR_{TRANSP,RLP};
* cap->data.{nr_{data,stop}_bits,parity}: set 8N1 by default;
* cap->data.modem_type: explicitly assign default value;
* cap->data.interm_rate: value 0 is reserved, set GSM48_BCAP_IR_{8k,16k}.
The related libosmocore.git patch additionally fixes encoding of the
"Connection element (octet 6c)", so that bcap->data.transp is used.
Change-Id: If49c89e4f867bac92ad062c062b9f36bab2b4531
Related: libosmocore.git I7339908864e8a2aef6f2b48a108650167e413c7f
Related: OS#6110, OS#4394
Without the gsm0808_speech_codec functions:
* codec_mapping_by_gsm0808_speech_codec_type(), and
* codec_mapping_by_gsm0808_speech_codec()
fail to find the codec mapping for CLEARMODE.
Change-Id: I87b3aedaf7ff7bbbcb381e94158566dc765e3ae6
Related: OS#6110, OS#4394
As per 3GPP TS 48.008, section 3.2.2.103, the Codec Type is valid if
at least one of FI, PI or PT is set to '1'. Otherwise the Speech
Codec Element is considered invalid and shall be ignored.
Change-Id: Ibc452d37d4215c961a7946eef3ba2e7efdba078b
Related: OS#6110, OS#4394
Neels Hofmeyr
Vadim Yanitskiy
Harald Welte
Andreas Eversberg
Manawyrm
Oliver Smith
Pau Espin