mirror of
https://github.com/jmcorgan/fips.git
synced 2026-10-06 03:28:24 +00:00
Johnathan Corgan
0ed841b4a5
The data frame carries a two-byte length field so the receiver can trim Ethernet minimum-frame padding, which would otherwise reach AEAD as ciphertext. The unit test for that trim rebuilt the parse inline instead of calling the receive loop, so a regression in the loop's slice could not fail it: changing the loop to slice buf[3..len] left every Ethernet test green. The parse is now one function, data_payload, which the receive loop calls for every data frame and the tests drive directly; the loop keeps no other slice of the buffer for data frames. The padding test now runs against it, and new tests cover an unpadded frame, a zero-length padded frame, a frame shorter than the header and a length field larger than the received bytes. Dropping the length trim now fails the padding and zero-length tests, and dropping the bounds check fails the over-long length test. Behaviour is unchanged: the loop forwards the same bytes and drops the same frames with the same trace messages. The trim is covered by a unit test with a synthetic padded frame because no traffic reaches it on the veth links the chaos harness uses. Measured on a live ethernet-mesh run: an empty ICMPv6 echo is a 226-byte frame and a 1200-byte echo a 1342-byte frame; the smallest data frame seen was a 54-byte control frame, and short frames (54-byte data, 48-byte beacons) arrived unpadded; none of 250 received data frames carried bytes beyond its length field.