Mesh LPN RX race condition

[isundaylee]

Suppose node L is in a friendship with node F. Consider the following: 

1. Some other node C sends a message to L. 

2. F hears the message, and saves the message into L's friend queue.

3. L also happens to be scanning (e.g. L is currently in a Friend Poll receive window), and hears the message itself.

4. F sends the recorded message to L as the Friend Poll reply.

5. However, when L receives the recorded message from F, it will discard it on the Network layer due to message cache hit (it has already seen the message itself in step 2). This happens in the 3rd if statement in `deobfuscated_header_is_valid`.

6. Therefore, the LPN stack on L never receives L's Friend Poll reply. It thus never updates fsn, and will keep requesting the same message from L. 

7. After 5 Friend Polls, the LPN stack on L will terminate the friendship. 

I think in this case the LPN stack should still somehow increase its fsn despite of the message cache hit. 

To reproduce this, I had C continuously sending messages to L (3 messages every second). It usually reproduces within the first 20 seconds or so. 

Thanks! 

[Hung Bui]

Hi Isundaylee, 

Thanks for reporting this. However, by design, our LPN should not receive packet to itself when it's in friendship. You can have a look at the LPN life cycle figure here.

But we suspect it might be due to another bug we found in SDK v3.0 and fixed in SDK v3.1. Could you reproduce the issue on SDK v3.1 ? 

In addition, could you explain a little bit on how you figured out that the LPN discard packet from friend, do you have some log ? 

[isundaylee]

BTW the log above is for the similar scenario I described in my comment: "A similar thing can happen when L sends to a group address G which L also subscribes to itself. In this case the recorded Friend Poll reply will get discarded in the 4th if statement in `deobfuscated_header_is_valid`.", which repros deterministically every time.