mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-05 19:18:25 +00:00
Johnathan Corgan
cfee2ccde5
The received frame header declares a payload length that nothing read. It now gets compared against the frame the transport actually delivered, at the one dispatch point every transport converges on, before that field can be used as a parsing input. This changes behaviour on a deployed line, so it is worth being exact about what it drops. The stream transports read their frame boundary out of this same field, so the comparison holds by construction and never fires for them. The datagram transports deliver one whole frame per packet, where the length is known exactly and nothing checked it before. A short read there is a truncated frame, which already failed the tag or the exact-size parse; this changes which reason it is dropped for, not whether it is dropped. An unrecognised phase carries no fixed relationship, so it is left alone and reaches the dispatch as before, rather than being rejected on a guess. The drop gets its own rejection reason and its own counter. Reusing the admission reason would have mis-attributed it: this is a framing rejection decided before the phase dispatch, and it applies to data frames as well as handshakes. The dispatch function is now visible to the rest of the node module so a test can drive one packet through it, which is the reach the handshake handlers beside it already had.