[Proposal] The bft module's gateway should check block request and block response from peers to avoid spam

[elderhammer]

💥 Proposal

Issue #3315 points out that malicious validator can DDoS other validator by sending block response spam.
To solve this problem, in addition to putting the deserialization logic into a separate rayon thread #3316, we should also:

1. Check the request frequency of block request
2. Check whether the block response has a corresponding block request

In fact, node/router has already implemented the above two checks:

snarkOS/node/router/src/inbound.rs

Lines 87 to 92 in 5d4de02

	// Insert the block request for the peer, and fetch the recent frequency.
	let frequency = self.router().cache.insert_inbound_block_request(peer_ip);
	// Check if the number of block requests is within the limit.
	if frequency > Self::MAXIMUM_BLOCK_REQUESTS_PER_INTERVAL {
	bail!("Peer '{peer_ip}' is not following the protocol (excessive block requests)")
	}

snarkOS/node/router/src/inbound.rs

Lines 111 to 114 in 5d4de02

	// Remove the block request, checking if this node previously sent a block request to this peer.
	if !self.router().cache.remove_outbound_block_request(peer_ip, &request) {
	bail!("Peer '{peer_ip}' is not following the protocol (unexpected block response)")
	}

[raychu86]

Thanks for bringing this up @elderhammer, I believe that this is something we should address, and should in theory be fairly straightforward to do so. The changes should resolve #3315.

@vicsn It may be easier for you to tackle the rayon thread for deserialization in the block deserialization since you already did it for the Router - #3304.

I can tackle the cache checks here, which hopefully should also be pretty simple.

[elderhammer]

@vicsn It may be easier for you to tackle the rayon thread for deserialization in the block deserialization since you already did it for the Router - #3304.

There is already a pull request to tackle this part of the issue: #3316

[raychu86]

@ljedrz @niklaslong May be better if you guys tackle the cache updates. We may also want to consider if we can unify the caches down the line.