Cellular Network Infrastructure » OsmoGSMTester

Added bits required to test this part in https://gerrit.osmocom.org/#/c/3582/ and https://gerrit.osmocom.org/#/c/3583/

I also have some tests locally to test it but I see some buggy behavior still to investigate.

One more test added for the infra to test this: https://gerrit.osmocom.org/#/c/3584/

And I still lack a "subscriber_remove" API in hlr to have encryption tests working in AoIP. I pushed the tests I created in branch pespin/encryption of osmo-gsm-tester.

Issues I saw so far:
- On NITB, there seems to be some rance condition which makes it sometimes accept location updates instead of rejecting or asking for a challenge.
- Even if NITB/MSC rejects the registering after failed challenge (due to using wrong KI), ofono signals status "registered" for up to a few seconds before going back to unregisterd. Sometimes it signals denied. This seems to be related the issue in qmi/ofono we already saw, in which QMI sends "registered" signal too soon (lynxis this may be interesting for you). IMHO this is wrong and it should never be shown as "registered".

I implemented generic operations required for the tests like deleting subscribers (to re-register them with correct KI after registering first with wrong KI).

I also added support for the "authentication" parameter.

As soon as I use "authentication required" in osmo-msc, the modems cannot register, they are rejected. I can see the HLR generating the challenges and expected responses, with correct KI and IMSI. I can see the request with the challenge being sent to the modem. However, the Response to the challenge from the modem seems different than the "sres" I see in the HLR log, which probably explain why it receives a Reject afterwards.

This happens even with encryption a5 0.

All updated suite tests can be found in osmo-gsm-tester pespin/encryption branch.

It seems the sim cards we are using in the setup are using XOR algorithm, not COMP128v1. Once I change database specifying to use XOR for those auth works fine.

Let's add a "algo" param for each object of type modem in osmo-gsm-tester. Also support setting a default value. This can then be obtained by using modem.algo(), and we can pass that to subscriber_add algo parameter.

I added more code required (set encryption attr in bsc, auth_algo attr in modem). After that, I have 3 working tests which have been pushed to gerrit. These 3 test check if a modem can register when the core is configured with:

1. authentication optional and encryption a5 0 (off).
2. authentication required and encryption a5 0 (off).
3. authentication required and encryption a5 1 (on).

Those 3 tests are to be used with AoIP setup (osmo-msc). I still see issues when using osmo-nitb: sometimes the modem is accepted even if the KI is incorrect, ie no auth is done.

I have extra tests for NITB and for AoIP testing a52 and a53 (which are not working) in branch pespin/encryption and I won't submit those for now.

I think we can close this one, what we have is enough for a first round.

Tests in NITB seem to have some issue, but as far as I was told, auth code has changed from nitb->msc, and nitb is going to be deprecated soon anyway, so let's support testing this in AoIP builds for now.