Skip to content

PVF Dispatch Implementation Playbook

rayswaynl edited this page Jul 1, 2026 · 25 revisions

PVF Dispatch Implementation Playbook

This page turns DR-1 and DR-38 into a patch-ready guide for hardening the generic PVF dispatch layer.

Scope: Chernarus source mission first, then LoadoutManager propagation. All source paths are relative to Missions/[55-2hc]warfarev2_073v48co.chernarus/ unless noted. Arma 2 Operation Arrowhead 1.64 only.

Use this with Networking and public variables, Hardening roadmap, Server authority map, Testing workflow and agent-hardening-backlog.jsonl.

2026-07-01 PR #125 update: draft release branch codex/release-command-center-20260630@c441d6f38df103e32694791f0bca5c5a70428c12 carries the registered-handler allowlist layer described here in both maintained terrains. Common/Init/Init_PublicVariables.sqf exports WFBE_CL_PVF_ALLOWED and WFBE_SE_PVF_ALLOWED, both dispatchers reject unregistered handler names before missionNamespace getVariable, and the manually registered GuerVbiedBounty client receiver is added to the client allowlist at its existing Init_Common.sqf registration point. Static smoke now gates those strings. This closes the PR #125 generic handler-name allowlist gap only; DR-55 authenticated sender context, legitimate-handler payload validation, direct channels such as ATTACK_WAVE_INIT, side-supply temp PVs and SEND_MESSAGE direct compile remain separate authority work. Runtime RPT evidence is still pending.

Status

Item State
Finding Confirmed high live-server hardening gap: DR-1.
Performance note Confirmed low/medium perf win: DR-38.
Current code Branch-split: current stable/B74.1 origin/master@f8a76de34 / origin/claude/b74.1-aicom@f8a76de34, current B69 origin/claude/b69@8d465fce and adjacent B74 origin/claude/b74-aicom-spend@b23f557f Chernarus plus maintained Vanilla use missionNamespace getVariable _script plus a typeName == "CODE" guard in both PVF dispatchers; old docs/Miksuu/perf/historical refs still use dispatch-time Call Compile.
Remaining dispatcher fix Add registered SRVFNC* / CLTFNC* allowlists and rejection logging before calling the current lookup fully hardened; backport the namespace lookup to old branches that still use Call Compile.
What current stable fixes Dispatch-time arbitrary SQF-text compilation and avoidable per-message recompilation in the generic PVF dispatch.
What remains in this layer Forged payloads can still name any existing global CODE value unless a registered-handler allowlist is added; bad names silently no-op instead of logging.
What it does not fix Payload forgery inside legitimate handlers, missing authenticated sender context (DR-55), direct publicVariable channels outside WFBE_PVF_*, or missing BattlEye defense-in-depth.

Current Branch Matrix

Branch routes pvf-dispatcher-current-stable-closeout-2026-06-21, pvf-dispatch-current-stable-partial-closeout, pvf-sender-auth-current-b69-head-refresh-2026-06-22 and pvf-dispatch-current-b74-refresh-2026-06-22 rechecked the generic PVF dispatcher across maintained roots and active candidate branches after origin/master advanced to 0139a346, B69 advanced to 8d465fce and adjacent B74 appeared at b23f557f; current origin exposes no release/*, feat/*pvf*, feat/*network*, feat/*auth* or feat/*public* heads on 2026-06-22:

Scope checked Server dispatcher Client dispatcher PVF init / registry Practical meaning
Docs checkout HEAD@925d56e01a1e Chernarus and maintained Vanilla Takistan Both roots still read _script from payload index 0 and run _parameters Spawn (Call Compile _script) at Server/Functions/Server_HandlePVF.sqf:14. Both roots still run _parameters Spawn (Call Compile _script) at Client/Functions/Client_HandlePVF.sqf:22. Both roots keep the older registry shape; no missionNamespace getVariable _script / PVF_ALLOWED guard is present in the checked dispatcher files. The checked PVF paths are unchanged from a0721301d4f5. Patch/port is still required if this docs/source branch becomes the code target.
Current origin/master@f8a76de34 / origin/claude/b74.1-aicom@f8a76de34 Chernarus and maintained Vanilla Takistan Both roots read _script from payload index 0, resolve _code = missionNamespace getVariable _script at Server/Functions/Server_HandlePVF.sqf:14, and spawn only when typeName _code == "CODE" at :15. Chernarus blame is 7d60b02b4; maintained Vanilla propagation blame is 9b49883cb. Both roots keep the adjacent headless-client destination filter, then resolve _code = missionNamespace getVariable _script at Client/Functions/Client_HandlePVF.sqf:32 and spawn only CODE at :33. Both roots precompile 20 client handlers and 19 server handlers, then add PVEHs at Common/Init/Init_PublicVariables.sqf:55-61. No WFBE_CL_PVF_ALLOWED, WFBE_SE_PVF_ALLOWED, PVF_ALLOWED, allowlist, rejection-log, sender or requester guard exists in checked current dispatcher/init files. DR-38 and the raw dispatch-time Call Compile part of DR-1 are source-present/fixed on current stable/B74.1; registered-handler allowlist, warning logs, Arma smoke and DR-55 sender authentication remain open.
Current B69 origin/claude/b69@8d465fce and adjacent B74 origin/claude/b74-aicom-spend@b23f557f Chernarus and maintained Vanilla Takistan Same namespace/CODE server dispatcher as current stable/B74.1 at Server_HandlePVF.sqf:14-15 in both maintained roots. Same namespace/CODE client dispatcher as current stable/B74.1 at Client_HandlePVF.sqf:32-33 in both maintained roots. B69 adds a default-off aicom-team-merge case to Client/PVFunctions/HandleSpecial.sqf:57 with WFBE_C_AICOM_HC_MERGE_ENABLE gate at :59; B74 has no checked dispatcher/init delta from B69. Same value-only PVEH registration at Init_PublicVariables.sqf:56,61; no sender context, registered-handler allowlist or warning log was found. git diff --name-status 0a1ccb4d..origin/claude/b69 changes only Client/PVFunctions/HandleSpecial.sqf among the checked PVF/PVEH paths; git diff --name-status origin/claude/b69..origin/claude/b74-aicom-spend and origin/claude/b74-aicom-spend..origin/master are empty for the checked generic PVF dispatcher/init paths. Treat B69/B74 as branch-only evidence that dispatch lookup matches current stable/B74.1 but DR-55 remains open. The client special tag is not a sender-authentication fix.
Miksuu upstream miksuu/master b8389e74 Same server dispatcher compile at Server_HandlePVF.sqf:14 in both maintained roots. Same client dispatcher compile at Client_HandlePVF.sqf:22 in both maintained roots. Same precompile-and-PVEH registry shape at Init_PublicVariables.sqf:45,50; no allowlist or missionNamespace getVariable dispatch guard. No upstream rescue exists.
origin/perf/quick-wins 0076040f Same server dispatcher compile at :14 in Chernarus and maintained Vanilla. Same client dispatcher compile at :22 in Chernarus and maintained Vanilla. Same registry shape; Chernarus PVEH lines are Init_PublicVariables.sqf:46,51, Vanilla remains :45,50. Performance branch does not cover DR-1/DR-38 despite this being a small perf/security patch.
Current origin release/PVF/network/auth/public heads Remote branch scan returned no release/*, *pvf*, *network*, *auth* or *public* heads on 2026-06-23. Same. Same. Older origin/release/2026-06-feature-bundle@a96fdda2 evidence is historical target-commit evidence only, not a current remote branch head. Recheck that commit explicitly if it becomes a target again.

The current stable partial fix is split across two source commits: 7d60b02b4 changes the Chernarus dispatchers from Spawn (Call Compile _script) to missionNamespace getVariable _script plus a CODE guard, and 9b49883cb propagates the same dispatcher change into maintained Vanilla Takistan. Neither commit adds a registered-handler allowlist or a rejected-handler warning log.

Bohemia's Community Wiki lists missionNamespace as introduced with Arma 2 1.00 and shows missionNamespace getVariable as the supported namespace lookup shape, so the recommended lookup is Arma 2 OA-compatible rather than an Arma 3 import.

What I Read

  • Common/Init/Init_PublicVariables.sqf:9-29, :31-52, :54-62
  • Server/Functions/Server_HandlePVF.sqf:7-15
  • Client/Functions/Client_HandlePVF.sqf:7-33
  • Common/Functions/Common_SendToServer.sqf:12-18
  • Common/Functions/Common_SendToServerOptimized.sqf:12-18
  • Common/Functions/Common_SendToClient.sqf:13-21
  • Common/Functions/Common_SendToClients.sqf:12-19
  • Client/Module/supplyMission/supplyMissionCompletedMessage.sqf:22
  • Current stable/B74.1 origin/master@f8a76de34 / origin/claude/b74.1-aicom@f8a76de34 maintained-root dispatcher proof: Chernarus Server_HandlePVF.sqf:14-15, Client_HandlePVF.sqf:32-33; Vanilla Server_HandlePVF.sqf:14-15, Client_HandlePVF.sqf:32-33; no PVF_ALLOWED/allowlist/sender/requester hits in the checked maintained-root dispatcher/init files.
  • Current B69 origin/claude/b69@8d465fce and adjacent B74 origin/claude/b74-aicom-spend@b23f557f maintained-root dispatcher proof: same PVF init/dispatcher line shape as current stable at Init_PublicVariables.sqf:56,61, Server_HandlePVF.sqf:14-15 and Client_HandlePVF.sqf:32-33; checked B69 movement since 0a1ccb4d is Client/PVFunctions/HandleSpecial.sqf:57,59 for the default-off aicom-team-merge tag, and the checked B69..B74 generic dispatcher/init delta is empty.
  • Deep-review findings DR-1 and DR-38
  • Networking and public variables, especially direct-PV and residual authority sections
  • Server authority map PVF dispatch row
  • Hardening roadmap P0 PVF section
  • Bohemia Interactive missionNamespace page, which lists it as an Arma 2 OA scripting command and demonstrates missionNamespace getVariable: https://community.bistudio.com/wiki/missionNamespace

What The Code Actually Does

Init_PublicVariables.sqf builds two command arrays on current stable. The server command list has 20 active entries at :9-28, from RequestVehicleLock through HCStat (line 28), with RequestFOBStructure at line 15 added in b75.7). The client command list has 20 active entries at :31-50, from AllCampsCaptured through DashboardAnnounce; DatabaseDebug is still commented at :36 and is not active.

The same file already compiles every registered command into global code variables:

Call Compile Format["CLTFNC%1 = compile preprocessFileLineNumbers 'Client\PVFunctions\%1.sqf'", _x];
Call Compile Format["SRVFNC%1 = compile preprocessFileLineNumbers 'Server\PVFunctions\%1.sqf'", _x];

It also registers one WFBE_PVF_<Command> event handler per command. Client PVF variables call WFBE_CL_FNC_HandlePVF; server PVF variables call WFBE_SE_FNC_HandlePVF.

The send helpers rewrite a logical command name into one of those compiled function variable names:

  • server-bound helpers set payload index 0 to SRVFNC<Command>;
  • client-bound helpers set payload index 1 to CLTFNC<Command>;
  • dedicated/multiplayer branches publish WFBE_PVF_<Command>;
  • hosted branches call the same handler locally with Spawn WFBE_*_FNC_HandlePVF.

On current stable, the dispatchers no longer compile the string from the payload. They look up the named missionNamespace value and only spawn it if it resolves to CODE. Docs checkout, origin/perf/quick-wins and the older Miksuu/release evidence rows above still use the old dispatch-time compile shape.

// Server_HandlePVF.sqf:11-15
_script = _publicVar select 0;
_parameters = if (count _publicVar > 1) then {_publicVar select 1} else {[]};
_code = missionNamespace getVariable _script;
if (!(isNil "_code") && {typeName _code == "CODE"}) then {_parameters Spawn _code};

// Client_HandlePVF.sqf:13-33, after destination filtering
_script = _publicVar select 1;
_parameters = if (count _this > 2) then {_publicVar select 2} else {[]};
_code = missionNamespace getVariable _script;
if (!(isNil "_code") && {typeName _code == "CODE"}) then {_parameters Spawn _code};

This closes the dispatch-time arbitrary text compilation class on current stable/B74.1 origin/master@f8a76de34, but it is not the full playbook patch yet because the lookup accepts any existing global CODE value and silently ignores bad names. Add a registered-handler allowlist before treating dispatch as fully hardened.

Why It Matters

DR-1 is the security boundary: the receiver runs a handler name chosen by the sender. Current stable no longer compiles that string at dispatch time, but it still does not prove the name came from the registered SRVFNC* / CLTFNC* set before spawning the resolved CODE value.

DR-38 is the performance angle on old branches: registered handlers were already compiled at init, so dispatch-time Call Compile _script was just doing a variable lookup in the slowest and riskiest way. Current stable has removed that recompile class; the remaining current-stable work is allowlisting and logging.

The source-backed opportunity is now smaller on current stable: complete the existing namespace-lookup patch with a registered-handler allowlist and logging, then smoke valid traffic and forged unregistered names. Older target branches such as Miksuu and perf/quick-wins still need the full Call Compile replacement.

Implementation Shape

For current origin/master, patch the registry and both dispatchers together to add explicit allowlists and rejection logs. For older branches that still use Call Compile, apply the lookup replacement and the allowlist in the same pass.

1. Export allowlists at PVF init

In Common/Init/Init_PublicVariables.sqf, create mission globals for the compiled handler names. The exact names are owner choice; keep them clear and side-specific.

WFBE_CL_PVF_ALLOWED = [];
WFBE_SE_PVF_ALLOWED = [];

{
    Call Compile Format["CLTFNC%1 = compile preprocessFileLineNumbers 'Client\PVFunctions\%1.sqf'", _x];
    WFBE_CL_PVF_ALLOWED = WFBE_CL_PVF_ALLOWED + [Format["CLTFNC%1", _x]];
    if (!isServer || local player) then {Format['WFBE_PVF_%1',_x] addPublicVariableEventHandler {(_this select 1) Spawn WFBE_CL_FNC_HandlePVF}};
} forEach _clientCommandPV;

{
    Call Compile Format["SRVFNC%1 = compile preprocessFileLineNumbers 'Server\PVFunctions\%1.sqf'", _x];
    WFBE_SE_PVF_ALLOWED = WFBE_SE_PVF_ALLOWED + [Format["SRVFNC%1", _x]];
    if (isServer) then {Format['WFBE_PVF_%1',_x] addPublicVariableEventHandler {(_this select 1) Spawn WFBE_SE_FNC_HandlePVF}};
} forEach _serverCommandPV;

Why not only getVariable? A plain namespace lookup prevents arbitrary SQF text from being compiled, but it could still resolve another global CODE variable if a forged payload names it. The allowlist limits dispatch to exactly the registered PVF handlers.

2. Harden the server dispatcher

Replace the final Call Compile line in Server/Functions/Server_HandlePVF.sqf with allowlist membership and namespace lookup.

Private ["_handler","_parameters","_publicVar","_script"];

_publicVar = _this;

_script = _publicVar select 0;
_parameters = if (count _publicVar > 1) then {_publicVar select 1} else {[]};

if !(_script in WFBE_SE_PVF_ALLOWED) exitWith {
    ["WARNING", Format ["Server_HandlePVF: rejected unregistered PVF handler [%1].", _script]] Call WFBE_CO_FNC_LogContent;
};

_handler = missionNamespace getVariable _script;
if (typeName _handler != "CODE") exitWith {
    ["WARNING", Format ["Server_HandlePVF: registered PVF handler [%1] is not CODE.", _script]] Call WFBE_CO_FNC_LogContent;
};

_parameters Spawn _handler;

3. Harden the client dispatcher

Keep the destination filter intact. Replace only the final dispatch in Client/Functions/Client_HandlePVF.sqf.

_script = _publicVar select 1;
_parameters = if (count _this > 2) then {_publicVar select 2} else {[]};

if !(_script in WFBE_CL_PVF_ALLOWED) exitWith {
    ["WARNING", Format ["Client_HandlePVF: rejected unregistered PVF handler [%1].", _script]] Call WFBE_CO_FNC_LogContent;
};

_handler = missionNamespace getVariable _script;
if (typeName _handler != "CODE") exitWith {
    ["WARNING", Format ["Client_HandlePVF: registered PVF handler [%1] is not CODE.", _script]] Call WFBE_CO_FNC_LogContent;
};

_parameters Spawn _handler;

4. Keep Spawn

Do not convert dispatch to Call in this patch. Existing PVF functions are scheduled through Spawn; some handlers or downstream calls may sleep, wait or spawn their own work. Changing scheduling semantics belongs to a separate handler-by-handler audit.

Boundary: Dispatcher Hardening Vs Authority Validation

This patch is the PVF foundation, not the whole server-authority migration.

Layer Fixed by this playbook? Example
Sender-chosen arbitrary handler string Yes Forged payload trying to compile arbitrary SQF text instead of SRVFNCRequestJoin.
Avoidable per-message compile Already resolved on master Dispatchers use missionNamespace getVariable + CODE guard; no Call Compile at dispatch time.
Legitimate handler with forged payload No RequestSpecial ICBM branch, RequestStructure, RequestUpgrade, RequestChangeScore.
Missing authenticated sender context No DR-55: docs/source HEAD@925d56e01a1e is unchanged from a0721301d4f5 for checked PVF init/dispatcher paths, while Miksuu b8389e74 and perf 0076040f forward only the public-variable value into WFBE_SE_FNC_HandlePVF and keep old dispatcher Call Compile; current stable origin/master@f8a76de34 / B74.1, B69 8d465fce and B74 b23f557f use namespace/CODE dispatch but still pass only (_this select 1) at Init_PublicVariables.sqf:56,61, so handlers receive parameters without publisher context.
Direct publicVariable channel outside PVF No ATTACK_WAVE_INIT, side-supply temp PVs, supply mission PVs, MASH marker relay, HQ state channels.
BattlEye defense-in-depth No Repo BattlEyeFilter/publicvariable.txt still only ships the kickAFK feature rule.

After this patch, a forged command name should be rejected. A forged payload sent to a real registered handler still reaches that handler and must be validated there.

Sender Authentication Boundary

Do not treat handler-name allowlisting as full PVF hardening. DR-55 found that the server PVEH registration calls (_this select 1) Spawn WFBE_SE_FNC_HandlePVF, so the dispatcher receives the value tuple without the publisher identity. The 2026-06-23 current-B74.1 refresh keeps that split: current stable origin/master@f8a76de34 equals origin/claude/b74.1-aicom@f8a76de34, B69 8d465fce and B74 b23f557f removed dispatch-time Call Compile but still register value-only PVEHs at Init_PublicVariables.sqf:56,61, while docs/source HEAD@925d56e01a1e (PVF paths unchanged from a0721301d4f5), Miksuu b8389e74 and perf 0076040f keep both value-only PVEHs and old compile dispatchers. The checked B69..B74 and B74..B74.1 generic PVF dispatcher/init deltas are empty, and no allowlist/sender/requester symbol was found in checked current-stable/B69/B74 dispatcher/init files. The server dispatcher then selects _script and _parameters and spawns the handler, which leaves legitimate handlers unable to distinguish a real requester from a forged client payload unless each handler re-derives authority from trusted server state.

The DR-55 lane belongs in Server authority migration map and agent-hardening-backlog.jsonl#pvf-handler-sender-authentication: carry authenticated requester context to the server handler layer, then validate side, commander/team role, funds, target objects and ownership per handler.

Direct PublicVariable Boundary

Do not claim this patch hardens all public variables.

The generic PVF dispatcher only handles variables registered as WFBE_PVF_<Command> in Init_PublicVariables.sqf. DR-41 proved a separate surface: ATTACK_WAVE_INIT is a direct publicVariableServer channel and is not routed through Server_HandlePVF.sqf at all. Its fix is Attack-wave authority playbook: re-derive side supply server-side, validate requester/side, deduct cost server-side and clamp the resulting modifier/duration.

Other direct-channel families remain separate review targets in Networking and public variables: side supply temps, supply mission PVs, MASH marker relay, HQ state, AntiStack compensation, day/night, server FPS, AFK and marker/message channels.

Validation Plan

Source-only checks:

  1. Confirm WFBE_SE_PVF_ALLOWED contains one SRVFNC* name for every _serverCommandPV entry.
  2. Confirm WFBE_CL_PVF_ALLOWED contains one CLTFNC* name for every _clientCommandPV entry.
  3. Confirm every allowlisted name resolves to typeName == "CODE" after Init_PublicVariables.sqf.
  4. Confirm Server_HandlePVF.sqf and Client_HandlePVF.sqf no longer contain Spawn (Call Compile _script).
  5. Confirm no new Arma 3-only syntax such as isEqualTo, params, remoteExec, BIS_fnc_MP or CBA helpers was introduced.

Hosted/local smoke:

  1. Start a hosted mission.
  2. Verify RequestJoin still completes and the join-answer path reaches the player.
  3. Toggle a vehicle or MHQ lock, or perform another small RequestVehicleLock flow.
  4. Trigger one client-bound message such as LocalizeMessage or HandleSpecial.
  5. Inject or temporarily call a bogus handler name in a test-only path and confirm it logs one WARNING and no-ops.

Dedicated smoke:

  1. Repeat one server-bound PVF (RequestJoin or RequestVehicleLock).
  2. Repeat one server-to-client PVF (LocalizeMessage, SetVehicleLock or HandleSpecial).
  3. Watch the RPT for rejected-handler spam. Legitimate traffic should not hit the warning path.
  4. JIP sanity: late joiner still passes the existing RequestJoin and post-join state sync chain. DR-37 already marks PVF event replay as separate from durable state sync.

Security/authority negative checks:

  1. Forged unregistered handler name should no-op and log.
  2. Forged registered handler with bad payload may still execute the handler; record it under the relevant per-handler authority item, not as a PVF dispatch regression.
  3. Forged ATTACK_WAVE_INIT is unchanged by this patch; validate that under Attack-wave authority playbook.

Handoff

Future code owner:

  1. Implement this as branch hardening/pvf-dispatch.
  2. Edit the Chernarus source mission first.
  3. Run Tools/LoadoutManager after mission edits so vanilla Takistan receives the source change.
  4. Record validation in Testing workflow terms. Source-only review is useful but does not prove hosted/dedicated/JIP behavior.
  5. After merge, update Feature status, Hardening roadmap, Server authority map, Codebase coverage ledger, agent-hardening-backlog.jsonl, agent-context.json and Agent worklog.

Codex or Claude follow-up:

  • Review the allowlist patch before merge for Arma 2 OA syntax and hosted/dedicated path preservation.
  • Then move to P0 ICBM RequestSpecial or P1 attack-wave authority; those are per-handler/direct-channel authority fixes and should not be bundled into the dispatcher patch.

Continue Reading

Previous: Server authority map | Next: Attack-wave authority playbook

Main map: Home | Fast path: Quickstart | Agent file: agent-context.json

Sidebar

Clone this wiki locally