Nexus-Guard-Suite/.sisyphus/plans/archived/firewall-device-status-fix.md

Plan: Firewall INPUT Fix + Device Status via WireGuard Handshake

TL;DR

Fix two critical production issues: (1) firewall INPUT chain doesn't isolate WireGuard peers, (2) device status shows offline because it depends on Redis heartbeats instead of WireGuard handshake data.

Deliverables:

• WireGuard per-peer handshake status (replace Redis heartbeat dependency)
• Firewall INPUT chain rules in code (persist across restarts)
• ICMP + established/related accept rules for WG traffic

Estimated Effort: Medium Parallel Execution: YES - 2 waves Critical Path: T1 (wgctrl) → T2 (SyncToDB) → T3 (INPUT rules) → F1-F4

---

Context

Original Request

1. Device gogo2 connected via WireGuard (peer shows handshake in wg show) but dashboard shows offline (IsActive=false)
2. Firewall INPUT chain rules don't work — ping from WG peer still works even with drop rules
3. User wants a proper plan, not ad-hoc fixes

Interview Summary

Key Discussions:

• Device was created manually via API, not via device-agent → no heartbeats sent
• Heartbeat system: device-agent → POST /api/v1/heartbeat → Redis TTL 90s → SyncToDB sets is_active
• Without device-agent, no heartbeats → is_active stays false
• Solution: use WireGuard handshake time from wgctrl instead of Redis heartbeats
• Firewall: table ip nexusguard INPUT chain has rules but ping still works
• Root cause: rules not in code (manually added, lost on restart), ICMP not explicitly allowed, possible table inet vs table ip priority conflict

Research Findings:

• wgmanager_linux.go:50 iterates dev.Peers and reads peer.LastHandshakeTime — but only stores MAX across all peers, losing per-peer granularity
• WgStatus struct has no per-peer handshake map
• SyncToDB() in heartbeat/redis.go checks Redis key existence → sets is_active
• Dashboard reads device.IsActive from DB — no frontend changes needed if SyncToDB is fixed
• InitNetwork() creates INPUT chain with zero rules — all rules were manual
• AddInputRule() only adds udp dport rules — no ICMP, no established/related

Metis Review

Identified Gaps (addressed):

• INPUT chain rules must be added to InitNetwork() code for persistence
• ICMP accept rule missing — only TCP/UDP explicitly allowed
• Per-peer handshake data needed (current GetStatus() loses per-peer granularity)
• Remote node devices won't have local WG data — scoped to local-only
• Docker's inet filter chains must not be modified

---

Work Objectives

Core Objective

Device status reflects actual WireGuard connectivity (not Redis heartbeats), and firewall INPUT chain properly isolates WG peers.

Concrete Deliverables

• wgmanager interface: new GetPeerHandshakes() method
• heartbeat/redis.go: SyncToDB() uses WG handshake data
• nftables_linux.go: INPUT chain rules in InitNetwork()
• Production deployment with verification

Definition of Done

• Device with active WG peer shows is_active=true in API
• Device without WG handshake shows is_active=false
• ping 10.172.21.1 from WG peer → RTO (dropped)
• TCP 8080 from WG peer → works (API access)
• UDP 51820 from WG peer → works (WG tunnel)
• Non-WG traffic → unaffected (Docker, SSH, etc.)
• Rules persist across server restart

Must Have

• GetPeerHandshakes() map[string]time.Time in wgmanager
• SyncToDB() matches device public keys to WG peer public keys
• INPUT chain rules in InitNetwork() code (not manual)
• ICMP accept rule for WG traffic
• established/related accept rule for WG traffic

Must NOT Have (Guardrails)

• NEVER nft flush table — project anti-pattern
• NEVER change policy accept on any chain
• NEVER modify Docker's inet filter chains
• NEVER add IPv6 rules
• NEVER remove heartbeat API endpoint (keep as fallback)
• NEVER add new API endpoints — modify existing only
• NEVER touch shared/crypto/encryptor.go
• NEVER add per-device firewall INPUT rules (INPUT is server-level, not peer isolation)

---

Verification Strategy

ZERO HUMAN INTERVENTION — ALL verification is agent-executed.

Test Decision

• Infrastructure exists: YES (Go build, nft CLI, curl)
• Automated tests: Tests-after (verify existing tests still pass)
• Framework: go build -tags dev ./..., nft list chain, curl API

QA Policy

Every task includes agent-executed QA scenarios. Evidence saved to .sisyphus/evidence/task-{N}-{scenario-slug}.{ext}.

• Backend: Use Bash (curl) — API requests, assert status + response fields
• Firewall: Use Bash (ssh + nft) — rule verification, counter checks
• WireGuard: Use Bash (ssh + wg) — handshake verification

---

Execution Strategy

Parallel Execution Waves

Wave 1 (Start Immediately - backend core):
├── Task 1: wgmanager GetPeerHandshakes() [quick]
├── Task 2: SyncToDB uses WG handshake [quick]
└── Task 3: InitNetwork INPUT chain rules [quick]

Wave 2 (After Wave 1 - deployment + verification):
├── Task 4: Build + deploy to production [quick]
└── Task 5: Verify all acceptance criteria [quick]

Wave FINAL (After ALL tasks — 4 parallel reviews, then user okay):
├── Task F1: Plan compliance audit (oracle)
├── Task F2: Code quality review (unspecified-high)
├── Task F3: Real manual QA (unspecified-high)
└── Task F4: Scope fidelity check (deep)
-> Present results -> Get explicit user okay

Dependency Matrix

Task	Depends On	Blocks
T1	None	T2
T2	T1	T4
T3	None	T4
T4	T2, T3	T5, F1-F4
T5	T4	F1-F4

Agent Dispatch Summary

• Wave 1: 3 tasks — T1 quick, T2 quick, T3 quick
• Wave 2: 2 tasks — T4 quick, T5 quick
• FINAL: 4 tasks — F1 oracle, F2 unspecified-high, F3 unspecified-high, F4 deep

---

TODOs

• 1. wgmanager: add GetPeerHandshakes() method

  What to do:

  • Add GetPeerHandshakes() map[string]time.Time to WgManager interface in internal/wgmanager/manager.go
  • Implement in internal/wgmanager/wgmanager_linux.go: iterate dev.Peers, return map of peer.PublicKey.String() → peer.LastHandshakeTime
  • Implement stub in internal/wgmanager/wgmanager_stub.go (returns empty map)
  • Add test in internal/wgmanager/ if test file exists

  Must NOT do:

  • Don't modify existing GetStatus() method
  • Don't change WgStatus struct

  Recommended Agent Profile:

  • Category: quick
  • Skills: []

  Parallelization:

  • Can Run In Parallel: YES (with T3)
  • Parallel Group: Wave 1 (with T3)
  • Blocks: T2
  • Blocked By: None

  References:

  • apps/server-core/internal/wgmanager/manager.go:27-39 — WgManager interface + UpConfig struct
  • apps/server-core/internal/wgmanager/wgmanager_linux.go:27-58 — GetStatus() implementation, iterates dev.Peers at line 50
  • apps/server-core/internal/wgmanager/wgmanager_stub.go — Stub implementation

  Acceptance Criteria:

  • go build -tags dev ./... passes
  • GetPeerHandshakes() returns non-empty map when peers are connected
  • Method is on both interface and implementations

  QA Scenarios:

  Scenario: GetPeerHandshakes returns peer handshake data
    Tool: Bash (go test or manual verification)
    Preconditions: Server running with connected WG peer
    Steps:
      1. SSH to server
      2. Run: docker exec nexus-guard-suite-server-core-1 ./server-core -tags dev -test.run TestGetPeerHandshakes 2>&1 || echo "verify via API"
      3. Check wg show wg0 latest-handshakes for reference
    Expected Result: Map contains public key → timestamp for connected peer
    Evidence: .sisyphus/evidence/task1-peer-handshakes.json
  

  Commit: YES

  • Message: feat(wgmanager): add GetPeerHandshakes() for per-peer handshake data
  • Files: internal/wgmanager/manager.go, internal/wgmanager/wgmanager_linux.go, internal/wgmanager/wgmanager_stub.go

• 2. heartbeat: SyncToDB uses WireGuard handshake instead of Redis

  What to do:

  • Modify SyncToDB() in internal/heartbeat/redis.go to:
    1. Query wgMgr.GetPeerHandshakes() (inject wgMgr into Manager or pass as parameter)
    2. For each device, match device.PublicKey against WG peer public keys
    3. If match found AND handshake within 120s → set is_active=true, last_handshake=handshake_time
    4. If no match OR handshake > 120s → set is_active=false
  • Keep Redis heartbeat recording active (don't break existing device-agents)
  • Update NewHeartbeatManager to accept wgmanager.WgManager parameter
  • Update main.go to pass wgMgr to heartbeat manager

  Must NOT do:

  • Don't remove Redis heartbeat recording (keep as fallback)
  • Don't change the heartbeat API endpoint
  • Don't change RecordPing() method

  Recommended Agent Profile:

  • Category: quick
  • Skills: []

  Parallelization:

  • Can Run In Parallel: NO
  • Parallel Group: Wave 1 (after T1)
  • Blocks: T4
  • Blocked By: T1

  References:

  • apps/server-core/internal/heartbeat/redis.go:57-84 — Current SyncToDB implementation
  • apps/server-core/internal/heartbeat/redis.go:14-17 — Manager struct (needs wgMgr field)
  • apps/server-core/main.go:216-219 — HeartbeatManager creation (needs wgMgr param)
  • apps/server-core/api/devices.go:35-60 — Device List handler (reads is_active from DB)
  • apps/dashboard-ui/src/views/Devices.vue — Dashboard reads IsActive from API response

  Acceptance Criteria:

  • go build -tags dev ./... passes
  • Device with active WG peer → is_active=true in DB after SyncToDB cycle
  • Device without WG handshake → is_active=false
  • Redis heartbeat recording still works (existing device-agents unaffected)

  QA Scenarios:

  Scenario: Device with active WG peer shows online
    Tool: Bash (curl + ssh)
    Preconditions: Server running, device gogo2 connected via WG
    Steps:
      1. SSH to server, get token: TOKEN=$(curl -s http://localhost:8080/api/v1/auth/login -X POST -H "Content-Type: application/json" -d '{"username":"admin","password":"admin123"}' | python3 -c "import sys,json; print(json.load(sys.stdin)['token'])")
      2. Wait 35s for SyncToDB cycle
      3. curl -s http://localhost:8080/api/v1/devices -H "Authorization: Bearer $TOKEN" | python3 -c "import sys,json; devices=json.load(sys.stdin); [print(d['Name'], d.get('is_active')) for d in devices]"
    Expected Result: gogo2 shows True (is_active=true)
    Evidence: .sisyphus/evidence/task2-device-online.json
  
  Scenario: WG peer handshake data present
    Tool: Bash (ssh)
    Steps:
      1. SSH to server
      2. wg show wg0 latest-handshakes
      3. Compare timestamp with device LastHandshake in DB
    Expected Result: Timestamps match (within 30s)
    Evidence: .sisyphus/evidence/task2-handshake-match.json
  

  Commit: YES (groups with T1)

  • Message: feat(heartbeat): use WireGuard handshake for device status instead of Redis
  • Files: internal/heartbeat/redis.go, main.go

• 3. nftables: add INPUT chain rules to InitNetwork()

  What to do:

  • In InitNetwork() (after forward chain rules), add INPUT chain rules via nft CLI:

    nft add rule ip nexusguard input ct state established,related accept comment "estab"
    nft add rule ip nexusguard input icmp type echo-request accept comment "icmp_ping"
    nft add rule ip nexusguard input ip saddr 10.172.21.0/24 tcp dport 8080 accept comment "wg_api"
    nft add rule ip nexusguard input ip saddr 10.172.21.0/24 udp dport 51820 accept comment "wg_tunnel"
    nft add rule ip nexusguard input ip saddr 10.172.21.0/24 drop comment "wg_isolation"
    

  • Each rule must be idempotent (check comment before adding)
  • Use detectWGSubnet() (already exists) for WG subnet detection
  • Remove old manual rules on production server during deployment

  Must NOT do:

  • Don't change policy accept on INPUT chain
  • Don't add rules to Docker's inet filter chains
  • Don't add IPv6 rules

  Recommended Agent Profile:

  • Category: quick
  • Skills: []

  Parallelization:

  • Can Run In Parallel: YES (with T1)
  • Parallel Group: Wave 1 (with T1)
  • Blocks: T4
  • Blocked By: None

  References:

  • apps/server-core/internal/firewall/nftables_linux.go:25-43 — Current InitNetwork (INPUT chain created but empty)
  • apps/server-core/internal/firewall/nftables_linux.go:46-62 — detectWGSubnet() function
  • apps/server-core/internal/firewall/nftables_linux.go:156-174 — AddInputRule/RemoveInputRule (existing pattern)

  Acceptance Criteria:

  • go build -tags dev ./... passes
  • nft list chain ip nexusguard input shows all 5 rules after server start
  • Rules are idempotent (restart doesn't duplicate rules)

  QA Scenarios:

  Scenario: INPUT chain rules exist after restart
    Tool: Bash (ssh + nft)
    Preconditions: Server rebuilt with new code
    Steps:
      1. SSH to server
      2. docker compose down && docker compose up -d --build
      3. sleep 5
      4. nft list chain ip nexusguard input
    Expected Result: 5 rules visible (estab, icmp, wg_api, wg_tunnel, wg_isolation)
    Evidence: .sisyphus/evidence/task3-input-rules.txt
  
  Scenario: Ping from WG peer is dropped
    Tool: Bash (ssh)
    Steps:
      1. SSH to server
      2. nft list chain ip nexusguard input | grep wg_isolation
      3. Count packets: nft -a list chain ip nexusguard input | grep wg_isolation
    Expected Result: Drop rule exists with counter > 0
    Evidence: .sisyphus/evidence/task3-ping-dropped.txt
  
  Scenario: Docker networking unaffected
    Tool: Bash (ssh + curl)
    Steps:
      1. curl -s -o /dev/null -w '%{http_code}' http://172.20.8.191:80
      2. docker ps
    Expected Result: HTTP 200, containers running
    Evidence: .sisyphus/evidence/task3-docker-ok.txt
  

  Commit: YES

  • Message: fix(nftables): add INPUT chain rules for WG peer isolation in InitNetwork
  • Files: internal/firewall/nftables_linux.go

• 4. Build + deploy to production

  What to do:

  • Build server-core: cd apps/server-core && go build -tags dev ./...
  • Build dashboard-ui: cd apps/dashboard-ui && npm run build
  • Commit all changes
  • Push to remote
  • SSH to production: git pull --recurse-submodules && docker compose down && docker compose up -d --build
  • Remove old manual nftables rules on production

  Must NOT do:

  • Don't skip build verification
  • Don't push broken code

  Recommended Agent Profile:

  • Category: quick
  • Skills: []

  Parallelization:

  • Can Run In Parallel: NO
  • Parallel Group: Wave 2
  • Blocks: T5, F1-F4
  • Blocked By: T2, T3

  References:

  • connect_remote.txt — Production server SSH details
  • docker-compose.yml — Build configuration

  Acceptance Criteria:

  • go build -tags dev ./... passes locally
  • npm run build passes locally
  • Production server running with new code
  • curl http://172.20.8.191:80 returns 200

  QA Scenarios:

  Scenario: Production deployment successful
    Tool: Bash (ssh + curl)
    Steps:
      1. SSH to server
      2. docker logs nexus-guard-suite-server-core-1 2>&1 | tail -5
      3. curl -s -o /dev/null -w '%{http_code}' http://172.20.8.191:80
      4. curl -s -o /dev/null -w '%{http_code}' http://172.20.8.191:8080/api/v1/health
    Expected Result: Server started, HTTP 200 for both endpoints
    Evidence: .sisyphus/evidence/task4-deployment.txt
  

  Commit: YES

  • Message: feat: firewall INPUT fix + WG handshake device status
  • Files: all changed files

• 5. Verify all acceptance criteria on production

  What to do:

  • Run all acceptance tests from Metis review
  • Verify device status (gogo2 shows online)
  • Verify firewall (ping dropped, API works, Docker unaffected)
  • Verify rules persist (restart server, check rules)
  • Save all evidence

  Must NOT do:

  • Don't skip any acceptance test
  • Don't assume — verify each criterion

  Recommended Agent Profile:

  • Category: quick
  • Skills: []

  Parallelization:

  • Can Run In Parallel: NO
  • Parallel Group: Wave 2 (after T4)
  • Blocks: F1-F4
  • Blocked By: T4

  References:

  • connect_remote.txt — Production server SSH details
  • Plan "Definition of Done" section

  Acceptance Criteria:

  • All 7 acceptance tests pass
  • Evidence files saved

  QA Scenarios:

  Scenario: Full acceptance test suite
    Tool: Bash (ssh + curl + nft + wg)
    Steps:
      1. Test 1: Device gogo2 is_active=true via API
      2. Test 2: ping 10.172.21.1 from WG peer → RTO
      3. Test 3: curl http://10.172.21.1:8080/api/v1/health → 200
      4. Test 4: curl http://172.20.8.191:80 → 200 (Docker)
      5. Test 5: nft list chain ip nexusguard input → 5 rules
      6. Test 6: Restart server, verify rules persist
      7. Test 7: Dashboard shows device Online
    Expected Result: All 7 tests pass
    Evidence: .sisyphus/evidence/task5-full-qa.txt
  

  Commit: NO

---

Final Verification Wave

4 review agents run in PARALLEL. ALL must APPROVE. Present consolidated results to user and get explicit "okay" before completing.

• F1. Plan Compliance Audit — oracle Read the plan end-to-end. For each "Must Have": verify implementation exists. For each "Must NOT Have": search codebase for forbidden patterns. Check evidence files exist. Compare deliverables against plan. Output: Must Have [N/N] | Must NOT Have [N/N] | Tasks [N/N] | VERDICT: APPROVE/REJECT

• F2. Code Quality Review — unspecified-high Run go build -tags dev ./... + go vet ./.... Review all changed files for: empty catches, console.log in prod, commented-out code, unused imports. Check AI slop: excessive comments, over-abstraction. Output: Build [PASS/FAIL] | Vet [PASS/FAIL] | Files [N clean/N issues] | VERDICT

• F3. Real Manual QA — unspecified-high Start from clean state. Execute EVERY QA scenario from EVERY task. Test cross-task integration. Test edge cases. Save to .sisyphus/evidence/final-qa/. Output: Scenarios [N/N pass] | Integration [N/N] | Edge Cases [N tested] | VERDICT

• F4. Scope Fidelity Check — deep For each task: read "What to do", read actual diff. Verify 1:1. Check "Must NOT do" compliance. Flag unaccounted changes. Output: Tasks [N/N compliant] | Unaccounted [CLEAN/N files] | VERDICT

---

Commit Strategy

• Commit #1: Backend — wgmanager + heartbeat SyncToDB
• Commit #2: Backend — nftables INPUT chain rules
• Commit #3: Root — submodule refs

Success Criteria

Verification Commands

go build -tags dev ./...  # Expected: no errors
go vet ./...              # Expected: no warnings
wg show wg0 latest-handshakes  # Expected: Unix timestamps
curl -s http://localhost:8080/api/v1/devices | jq '.[].is_active'  # Expected: true for connected devices
nft list chain ip nexusguard input  # Expected: rules with icmp, established, drop

Final Checklist

• All "Must Have" present
• All "Must NOT Have" absent
• Device with active WG peer shows Online
• Firewall INPUT chain drops non-allowed WG traffic
• Rules persist across restart