Skip to content

LAPB conformance — C# LapbLayer vs the authoritative ND LAPB spec

This document records how the C# LAPB implementation satisfies the authoritative ND LAPB Layer-2 spec (canonical copy at WSL /home/ronny/repos/os/x25emu/docs/lapb-nd-spec.md). It walks the spec section 8.1 validator checklist (12 items) and maps each to the code and the unit test that proves it.

Implementation: SINTRAN/XMSG/SRC/Xmsg.Live/LapbLayer.cs (state machine), SINTRAN/XMSG/SRC/Xmsg.Live/LapbOptions.cs (timers/window), SINTRAN/XMSG/SRC/Xmsg.Hdlc/LapbFrame.cs (control-byte decode), SINTRAN/XMSG/SRC/Xmsg.Live/Seam/LapbLayerAdapter.cs (live clock wiring).

Tests: all under SINTRAN/XMSG/SRC/Xmsg.Live.Tests/ — LapbFrameDecodeTests, LapbLayerTests, LapbFrmrTests, LapbTimerTests, LapbConformanceVectorsTests (byte vectors S1-S8).

Section 8.1 validator checklist

Each item answers YES with the code location and the proving test.

  1. S-frame subtype via ctrl & 0x0F; RR/RNR/REJ distinguished; REJ 0x9 not conflated with data address 0x09. LapbFrame.DecodeSupervisory decodes the low nibble by field position. Test: LapbFrameDecodeTests.SupervisoryFrame_DecodesSubtype_NrAndPollFinal, .RejControlByte_NotConfusedWithDataAddress.

  2. Retransmit queue covers the whole window (not one buffer); go-back-N triggerable by REJ. LapbLayer._txInfo[8] stores a payload per N(S); GoBackN replays them. Test: LapbLayerTests.Rej_GoBackN_RetransmitsAllOutstanding, LapbConformanceVectorsTests.S2.

  3. T1 ticked in the LIVE runner; T1/N2 values. LapbLayerAdapter injects a monotonic millisecond clock (CurrentMillis) into Tick from the pump loop; defaults T1=3000, N2=10 (LapbOptions). Test: LapbTimerTests (all).

  4. RNR handled: busy blocks new I-frames; RR(P=1) probe. HandleSupervisory sets/clears _peerBusy; TransmitPending gates on !_peerBusy; OnT1Expiry polls RR(P=1). Test: LapbLayerTests.Rnr_HoldsSends_ThenRrResumes, LapbConformanceVectorsTests.S3.

  5. P=1 answered with F=1 and current N(R). HandleSupervisory / HandleInformation emit RR(F=1) when frame.PollFinal. Test: LapbLayerTests.ReceivedPoll_AnsweredWithFinal, LapbConformanceVectorsTests.S6.

  6. On SABM (sent/received/mid-session) V(S)/V(R)/V(A) reset to 0 and queue cleared; NO adopt-peer-sequence path. Reset is called on UA-accepted and on every received SABM; sequence adoption was removed. Test: LapbLayerTests.MidSessionSabm_ResetsButStaysConnected, .RxUa_Connects_ResetsVariables_EmitsRr_LearnsPeer, LapbConformanceVectorsTests.S7.

  7. Constructed with the 16-bit node number; stamped on SABM/UA/RR; peer node kept OUT of routing. LapbLayer(ushort ownNode, ...); SendUnnumbered / EmitSupervisory stamp the node; PeerNode is read-only and never a routing key. Test: LapbLayerTests.MultipleTrunks_KeepIndependentSequenceAndIdentity.

  8. Two I-frames outstanding and the FIRST retransmitted on REJ(0) (depth-1-buffer regression). Test: LapbConformanceVectorsTests.S2, LapbLayerTests.Rej_GoBackN_RetransmitsAllOutstanding.

  9. No code path sends an I-frame while peer-busy. TransmitPending requires !_peerBusy. Test: LapbLayerTests.Rnr_HoldsSends_ThenRrResumes (held send asserts empty output).

  10. LAPB module free of X.25/XMSG types (payload independence). LapbLayer references only System, System.Collections.Generic and NDInsight.Sintran.Xmsg.Hdlc (LapbFrame); the information field is opaque ReadOnlyMemory<byte> / ReadOnlySpan<byte>.

  11. Bad N(R) -> FRMR reason Z; over-long I-field -> FRMR reason Y. AcknowledgeThrough sends FRMR(Z); HandleInformation sends FRMR(Y) for info > 312 bytes. Test: LapbFrmrTests.BadReceiveSequence_SendsFrmrReasonZ_EntersFrmrSent, .OverlongInformation_SendsFrmrReasonY_NotDelivered.

  12. Mid-session SABM -> UA + reset (NOT DM + disconnect). HandleUnnumbered SABM case sends UA, resets, stays CONNECTED. Test: LapbConformanceVectorsTests.S7, LapbLayerTests.MidSessionSabm_ResetsButStaysConnected.

Byte-level vectors (section 8.2)

S1-S7 are asserted frame-for-frame in LapbConformanceVectorsTests (local node 100, peer 102). S8 is a framing-layer vector (byte-stuffing / FCS) covered by the HDLC tests (HdlcEncoderTests), not the LAPB layer.

Our implementation emits one spec-permitted extra RR keepalive on entering CONNECTED (spec 3.4, live-necessary so the ND peer advances to RUN); the vectors account for it and otherwise assert the required frames exactly.

Deviation from prior behaviour (live re-verification pending)

The previous LapbLayer ADOPTED the peer's sequence from its first I-frame — a live-verified workaround for the epoch-1 reconnect (the ND kernel appearing to keep its sequence across our TCP reconnect). The spec forbids adoption (section 3.2) and this implementation removes it, hard-zeroing on every SABM. The offline gates (S1-S7, SeamParityTests) pass, but they cannot prove the LIVE reconnect. Live re-verification against machine 100 — bring-up, reachability/list-route, connect-to, and a reconnect after a simulated machine restart — remains the operator's gate. If the live reconnect regresses, the documented fallback is a default-OFF LapbOptions.AdoptPeerSequenceOnReconnect flag, not yet implemented.