mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-06 03:28:24 +00:00
Johnathan Corgan
bde5d797ac
A session rekey this node initiated could end only on a SessionAck, a lost tie-break, cutover or session removal. The only expiry for an armed rekey handshake covered one the peer armed, timed from the peer's setup message, and no rekey SessionSetup is ever resent. One lost SessionSetup or SessionAck therefore left the rekey in flight for good, and a rekey in flight vetoes the trigger that would start another: the session kept running on its current keys and stopped rotating them. Where this node has the smaller address it also drops the peer's own setup on the tie-break, leaving the peer stuck the same way. The handshake now carries its own deadline, taken when this node sends its setup, and expires on the handshake timeout as one the peer armed does. Only the handshake is dropped, so the next tick starts a fresh rekey. Putting the handshake back after an unreadable SessionAck does not restart the deadline, so forged acks cannot hold the rekey open. The expiry for a handshake the peer armed is unchanged, and neither expiry reads the other side's clock. The retirement is logged and counted in a new rekey_unanswered session counter. The unit test that expected an expired rekey of our own to be left alone is replaced. Its reasoning, that the msg3 retransmission budget bounds the cycle, holds only once msg2 has been read. No wire format change.