design: faithful sessions & network subsystems (LOGINOUT consolidation + DECnet NETACP) [rd vms-515] - #1068
design: faithful sessions & network subsystems (LOGINOUT consolidation + DECnet NETACP) [rd vms-515]#1068baron-3dl wants to merge 1 commit into
Conversation
…n + DECnet NETACP) The ratified design record (rd vms-515; roadmap epics vms-43b login-consolidation + vms-ec9 DECnet-NETACP), from a 4-disposition adversarial-design forum grounded in code. Course-corrects two faithfulness findings an operator-caught login LARP exposed: - ONE authenticator: LOGINOUT.EXE, reached by every path (console/SSH/SET HOST) via $CREPRC — no Linux fork/execl/pty above the VMS layer; LOGINOUT installed with privilege so it re-personas post-auth (dissolving the "network daemon must forge identity" attack). Kills vmssshd's C reimplementation + the CTERM no-auth hole. Net code reduction. - DECnet reframed to NETACP (a privileged DETACHED process, NOT kernel-resident) + an executive device/dispatch face (_NET:/EWA0:/RTAn:/object-table, $GETDVI- visible), the userspace wire engine (PR #1057 codecs) demoted to its datalink per vms-a1c. Faithful line drawn at the device face, not the wire. - RTAn:/TNAn:/FTAn: as real-but-cheap executive terminal devices (OPA0:-shape), not a new QIO subsystem. Carries the 9 adversary attacks → disposition, the 5 executive-substance ratification gates (incl. the anti-LARP tell: $GETDVI RTA0: cross-process or it is a LARP), the P0 clean-room oracle-capture prerequisite, and the P0-P5 phased plan for /swarm-plan decomposition. Doc only — no product code. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01T4csyFSMUsS8k1D2MgMxk2
|
Superseded by #1248 (work/vms-515-design-ratified). The DECnet lane — subsystem authority — reconciled this design record against the live NETACP implementation and ruled it NOT obsolete but GOVERNING: the architecture (NETACP as privileged RUN/DETACHED $CREPRC, network-object-table dispatch, PR#1057 codecs demoted to NETACP datalink, A2/A8 wire-isolation dispositions) matches how NETACP was actually built and is now largely implemented on main. Landing #1068 verbatim was wrong only because its 2026-09-02 'hold all code / PR#1057 held' framing is now false (the code landed). #1248 preserves the original 213 lines verbatim + a corrected status + a §8 implementation map tying each design element to its landed code. Closing this in favor of #1248. |
…reality (#1248) Freshens PR #1068's clean-room design record (Faithful Sessions & Network Subsystems — LOGINOUT consolidation + DECnet NETACP) onto current main and reconciles it: the 2026-09-02 draft was written as a ratify-before-coding target with all code held, but the DECnet lane has since BUILT the subsystem to this design and it is on main. Landing it verbatim would ship a stale "hold all code / PR #1057 held" framing that is now false. Preserves the original 213-line record verbatim (the governing architecture, the §4 A2/A8 dispositions, the §7 ratification gates) and adds: - a corrected Status header (RATIFIED BY IMPLEMENTATION, reconciled 2026-09-15); - a note on the superseded draft blockquote; - §8, an implementation map tying each design element to the landed code: $CREPRC-LOGINOUT session primitive, RTAn: executive device (P2), the _NET: device face + object table (§3.3), wire-parse isolation A2/A8 (decnet$wire-isolation), SSH->$CREPRC(LOGINOUT) (#1229), the CTERM credential decoder + object-42 auth (P4), the oracle-pinned CTERM semantics (A4/A9), and the a1-0/a1-1 + a1-2 broker (rd vms-22c) as the ACP-served _NET: $QIO transport the record left for later (extends, does not contradict, §3.2). Supersedes PR #1068 (its branch is 179 behind and carries the stale framing). rd vms-515. Claude-Session: https://claude.ai/code/session_01M6uDPQagdhE9HJNM8mNctk Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The ratified design record for the "faithful sessions & network subsystems" reframe (rd vms-515; roadmap epics vms-43b login-consolidation + vms-ec9 DECnet-NETACP). Doc only — no product code. Produced by a 4-disposition adversarial-design forum (adversary/creative/systems/purist), grounded in the actual code.
Origin: an operator-caught login LARP (a broken
vmsdcl --logindefault) exposed that OVMX has no single VMS-authentic login subsystem, and that "decnetd" drifted into a Linux daemon.Core of the design:
$CREPRC. Nofork/execl/ptyabove the VMS layer (the operator's ruling); LOGINOUT installed with privilege so it re-personas post-auth, which also dissolves the adversary's worst attack (a network daemon needing to forge identity). Kills vmssshd's C reimplementation + the CTERM no-auth hole. Net code reduction._NET:/EWA0:/RTAn:/object-table,$GETDVI-visible), the userspace wire engine (PR DECnet Phase IV lane: live NSP link + NCP config + CTERM $SET HOST + first real-VAX adjacency #1057 codecs) demoted to its datalink per vms-a1c. Faithful line at the device face, not the wire.Includes the 9 adversary attacks → dispositions, the 5 executive-substance ratification gates (incl. the anti-LARP tell:
$GETDVI RTA0:cross-process or it's a LARP), the P0 clean-room oracle-capture prerequisite (real VAX↔VAX SET HOST), and the P0–P5 phased plan for/swarm-plandecomposition — to be reconciled with the existing login/loginout branches, not duplicated.🤖 Generated with Claude Code