SGWC unable to connect to SMF gtpc port != default

[pespin]

For network setup convinience, in osmo-gsm-tester I am using 1 IP address per object when setting up tests, which means all nodes of an EPC object get 1 unique IP address, and I can configure all open5gs processes to work under the same IP address by configuring different ports to work fine.

I saw most config files can set the "port" together with the "addr", and it's working mostly fine, I have all processes running and connecting each other.

For instance in GTP, I have:
MME = 172.18.50.100:2123
SGWC = 172.18.50.100:2125
SMF = 172.18.50.100:2124

Hence, following config files:
mme.yaml:

mme:
    gtpc:
      - addr: ${epc.run_addr} #default port 2123 is used here
sgwc:
    gtpc:
      - addr: ${epc.run_addr}
        port: 2125
smf:
    gtpc:
      - addr: ${epc.run_addr}
        port: 2124

sgwc.yaml:

sgwc:
    gtpc:
      - addr: ${epc.run_addr}
        port: 2125

smf.yaml:

smf:
    gtpc:
      - addr: ${epc.run_addr}
        port: 2124

However, upon UE attachment, I noticed the following set of messages being sent:

MME  ->    SGWC:    Create Session Request
SGWC -> SGWU:    Session Establishment Request
SGWU -> SGWC:    Session Establishment Response
SGWC ->    MME:    Create Session Request

And MME then warning about unsupported rx message "Create Session Request". Looking at SGWC code (open5gs.git/src/sgwc/sxa-handler.c sgwc_sxa_handle_session_establishment_response()), that message is aimed at the SMF (interface S5), so clearly something is wrong, since it being sent to the MME (172.18.50.100:2123).

It seems that happens because the port being used in sgwc_sxa_handle_session_establishment_response() is taken from somewhere which is never set, and hence is always 2123. As a result, since the IP address is the same for all nodes, and 2123 was the port MME was listening to, the message is sent to the MME.

After finding that out, I simply did the following changes, and I can have the attachment complete fine with srsUE+srsENB (signalling, didn't get to the user plane setup yet):

diff --git a/src/osmo_gsm_tester/templates/open5gs-mmed.yaml.tmpl b/src/osmo_gsm_tester/templates/open5gs-mmed.yaml.tmpl
index 9aec578..d8e23de 100644
--- a/src/osmo_gsm_tester/templates/open5gs-mmed.yaml.tmpl
+++ b/src/osmo_gsm_tester/templates/open5gs-mmed.yaml.tmpl
@@ -146,8 +146,8 @@ sgwc:
 #        apn: volte
 smf:
     gtpc:
-      - addr: ${epc.run_addr}
-        port: 2124
+      - addr: 127.0.0.1
+        port: 2123

 #
 # parameter:
diff --git a/src/osmo_gsm_tester/templates/open5gs-smfd.yaml.tmpl b/src/osmo_gsm_tester/templates/open5gs-smfd.yaml.tmpl
index 0e01337..dfd1092 100644
--- a/src/osmo_gsm_tester/templates/open5gs-smfd.yaml.tmpl
+++ b/src/osmo_gsm_tester/templates/open5gs-smfd.yaml.tmpl
@@ -313,8 +313,8 @@ smf:
       - addr: ${epc.run_addr}
         port: 8808
     gtpc:
-      - addr: ${epc.run_addr}
-        port: 2124
+      - addr: 127.0.0.1
+        port: 2123
     gtpu:
       - addr: ${epc.run_addr}
         port: 2153

So as you see, I did temporarily change the IP address, and I can say the IP address is passed correctly. However, the port used by SGWC to connect to SMF is always 2123, despite being able to configure it fine in all nodes.
We do maybe need a new "smf: gtpc: port:" section in sgwc.yaml? That would be really useful to be able to run everything using 1 IP address, or simply to be able to pick any random port number by automated tests.

[acetcom]

@pespin

That's just a testing purposes for me. You should not use this. According to the 3GPP standard, the port number cannot be delivered to the other entity using S1AP/GTPv2/PFCP.

Thanks a lot!
Sukchan

[pespin]

While I agree that in a usual setup the default port should be used, I see it totally confusing that one can actually configure those ports in some places except in one. So that's why I proposed adding the extra configuration node to set the port in SGWC. Otherwise, what's the point in being able to configure the other GTPC ports?

Would you accept such a patch?

[acetcom]

@pespin I have no plan to be able to set the port in SGWC. Sorry for not help this!

[pespin]

Thanks for your feedback. However, my question was more in the line of knowing if you would accept such a patch if someone else (for example: me) submitted a PR for it.

[acetcom]

No. I don't like to accept such a patch.

[laf0rge]

@acetcom would you then prefer a patch to remove the configuration option from all the other places?

As @pespin states, it is inconsistent that you can configure non-standard ports in all except one place in open5gs. This just makes it easy for users to shoot themselves into the foot. So IMHO, it should be possible everywhere (so it can create a working config), or nowhere.

[acetcom]

@laf0rge Yes! I'll remove an unused port in the configuration files tomorrow!

[acetcom]

Today I reviewed the configuration files related to port number. Port settings in GTP/PFCP have already been removed in configuration files. Only S1AP/NGAP/SGsAP/SBI port is configurable.

I think we can just leave it as it is.

Let me know if you have any further questions.
Sukchan

[pespin]

Hi,

so if I understand correctly, you simply checked that the example config files don't come with the "port:" attribute. Still, by looking at the code, one can infer that those ports can be set by config files (see the yaml parser related code), hence creating the false illusion that different ports than standard one can be used. That's precisely what happened to me.

Hence, the solution would be also to drop the code parsing the "port" attribute from yaml in the related nodes.

[github-actions]

This issue has been closed automatically due to lack of activity. This has been done to try and reduce the amount of noise. Please do not comment any further. The Open5GS Team may choose to re-open this issue if necessary.