Redo wire protocol race condition fix. #2164
Conversation
In commit a76acd4 we attempted to fix issues on service nodes that related to spamming service nodes with traffc, but this caused a regression where clients could not connect to the network. However, we also did something idiotic with the protocol handshake. The handshake logic was modified and caused client connections to fail. This was due to a change in how we decided what a timed out session is. We redo the close bug and revert the change to the protocol handshake.
|
@dr7ana - can you |
Merging cherry-picks back to testnet branch: - oxen-io#2164 - oxen-io#2134
| @@ -352,7 +351,7 @@ namespace llarp | |||
| bool | |||
| Session::TimedOut(llarp_time_t now) const | |||
| { | |||
| if (m_State == State::Ready) | |||
There was a problem hiding this comment.
this line here is the most important part of the change.
State::LinkIntro means we are finalizing the protocol handshake, we are in a kind of half established state were we are an encrypted but unauthenticated session. this is where we are sending the other end a "link intro" which identifies who we are (by snode address), including the ephemeral pubkey that we were challenged with for establishing the encryption. it's stupid af and i hate it.
State::Ready means we are authenticated and ready to exchange frames for inter node communication.
the state we hit here in this regression that one side thinks the other side is timed out when they really aren't, and are trying to finish the handshake. they can still send/recv but they just have no idea who it is they are talking to in terms of their .snode addr but the other end disagrees about where we are in that process and it gets stuck. this is because snode and client timeouts are different (because of course they are). someone wanted to add that requirement onto the pile and so it was added, even though it was something i did object to for complexity reasons.
as usual i gave up fighting on that because people have strong opinions here and i just want stuff to get done, rocking the boat is a waste of time.
In commit a76acd4 we attempted to fix issues on service nodes that related to spamming service nodes with traffc, but this caused a regression where clients could not connect to the network.
However, we also did something idiotic with the protocol handshake. The handshake logic was modified and caused client connections to fail. This was due to a change in how we decided what a timed out session is. We redo the close bug and revert the change to the protocol handshake.