[NCP] Pinging non-RLOC IPs of OTBR (using NCP) fails

[hemanth-silabs]

I have a OTBR on Raspi which is connected to a NCP. When I try to ping this OTBR from a FTD in the network, the ping fails.

> ping fe80:0:0:0:546c:e537:d819:10bd
ping fe80:0:0:0:546c:e537:d819:10bd
1 packets transmitted, 0 packets received. Packet loss = 100.0%.
Done
> ping fdef:a0ff:9149:1:69b7:afde:8967:9dbf
ping fdef:a0ff:9149:1:69b7:afde:8967:9dbf
1 packets transmitted, 0 packets received. Packet loss = 100.0%.
Done

This works fine when pinging a RLOC IP.

Looking at the logs, the ping fails because of a read/write error on the mTunFd.

2025-07-02T13:38:17.292247-04:00 raspi-169-224 otbr-agent[27642]: [INFO]-NETIF---: Packet from NCP (56 bytes)
2025-07-02T13:38:17.292932-04:00 raspi-169-224 otbr-agent[27642]: [WARN]-NETIF---: Failed to receive, error:Input/output error
2025-07-02T13:38:17.293282-04:00 raspi-169-224 otbr-agent[27642]: [INFO]-NcpSpin-: NcpSpinel: HandleValueIs: OK
2025-07-02T13:38:17.293589-04:00 raspi-169-224 otbr-agent[27642]: [INFO]-NcpSpin-: HandleNotification: OK

To Reproduce Information to reproduce the behavior, including:

1. Git commit id: 622f5ec
2. IEEE 802.15.4 hardware platform: EFR32MG21
3. This is the build command used to build OTBR:

sudo INFRA_IF_NAME=eth0 NAT64=0 DNS64=0 WEB_GUI=0 REST_API=0 FIREWALL=0 OTBR_OPTIONS="-DCMAKE_BUILD_TYPE=Debug -DOT_THREAD_VERSION=1.4 -DOTBR_COVERAGE=ON -DOTBR_DBUS=ON -DOTBR_FEATURE_FLAGS=ON -DOTBR_TELEMETRY_DATA_API=ON -DOTBR_UNSECURE_JOIN=ON -DOTBR_TREL=ON -DOTBR_SRP_ADVERTISING_PROXY=ON -DBUILD_TESTING=OFF" ./script/setup

4. Network topology
   There are only two devices in the network, OTBR and a router/child.

Expected behavior Pings should work for any OTBR IPs

[hemanth-silabs]

Cc: @Irving-cl

[Irving-cl]

Hi @hemanth-silabs,

I just tried it and I can ping all these addresses successfully from a FTD.

But actually, in our case where we ping from an FTD to the OTBR with NCP, the ping packet won't reach the host of OTBR. It will be directly handled by the NCP: Ip6::HandleDatagram -> Ip6::Receive -> Icmp::HandleMessage. So I suggest adding some logs around here and see how the incoming Icmp packet goes.

That requires platform logging facilities on your platform. Ideally, your NCP uses USB to communicate with the host and has another interface to output logs (by JLinkRTT probably). Not sure if your platform supports it.

[hemanth-silabs]

@Irving-cl I debugged a bit more and found that mEchoMode is set to OT_ICMP6_ECHO_HANDLER_RLOC_ALOC_ONLY in icmp6.cpp:

While digging through the code I see this in ncp_base.cpp:

otIcmp6SetEchoMode(mInstance, OT_ICMP6_ECHO_HANDLER_RLOC_ALOC_ONLY);
#if OPENTHREAD_FTD
otThreadRegisterNeighborTableCallback(mInstance, &NcpBase::HandleNeighborTableChanged);
#endif

We seem to be overwriting this in ncp_base_mtd.cpp, but not in ncp_base_ftd.cpp. I feel we need to add a config item for this for NCPs.

What do you think?

[Irving-cl]

Hi @hemanth-silabs,

Sorry I just gave some wrong information in my previous comment. I checked the code and it's intentional to handle RLOC/ALOC ping request inside the OT core while passing the link-local ping request (has other source addresses) to the host and let the host handle it. And on my raspberry-pi, the link-local ping was indeed replied by the host.

So the issue on your side is the I/O error. But I don't any idea on why writing the tunFd failed. Are there any access policy on your device? Have you run otbr-agent with the root user? Could you try ping from your raspberry pi to another FTD on the host using system ping and specify -I wpan0. Let's first check whether otbr-agent can receive packets from the system through the tunFd.

[hemanth-silabs]

@Irving-cl I tried to ping raspi from another ftd (running through Raspi) using ping6. I don't see this work as well. I see this error:

2025-07-10T13:24:26.373443-04:00 raspi-169-224 otbr-agent[830]: [WARN]-NcpSpin-: Received uncognized key: 115
2025-07-10T13:24:26.373968-04:00 raspi-169-224 otbr-agent[830]: [INFO]-NcpSpin-: NcpSpinel: HandleValueIs: OK
2025-07-10T13:24:26.374246-04:00 raspi-169-224 otbr-agent[830]: [INFO]-NcpSpin-: HandleNotification: OK
2025-07-10T13:24:40.490668-04:00 raspi-169-224 otbr-agent[830]: [WARN]-NcpSpin-: Received uncognized key: 115
2025-07-10T13:24:40.492417-04:00 raspi-169-224 otbr-agent[830]: [INFO]-NcpSpin-: NcpSpinel: HandleValueIs: OK
2025-07-10T13:24:40.492724-04:00 raspi-169-224 otbr-agent[830]: [INFO]-NcpSpin-: HandleNotification: OK

[Irving-cl]

@hemanth-silabs

Sorry somehow I wasn't notified and didn't see your reply.

2025-07-10T13:24:40.490668-04:00 raspi-169-224 otbr-agent[830]: [WARN]-NcpSpin-: Received uncognized key: 115

This is not an error. 115 is SPINEL_PROP_STREAM_NET_INSECURE. Sometimes we can receive these from the NCP.

I tried to ping raspi from another ftd (running through Raspi) using ping6.

Then it seems that the packet doesn't successfully go from the system network interface (wpan0) to the tun fd in otbr-agent. Could you use tcpdump and see if there are any packets received on wpan interface when you did pinging? On the other hand, could you write a simple program that simply creates a tun interface on the network and receives on that. And then run this program on your raspberry pi and debug what's wrong with that.