Linux Firewall Hardening

Safe Linux firewall hardening with backend detection, idempotent atomic rules, rollback protection, and AI-executable state-machine workflows. Covers ufw, firewalld, nftables, iptables, Docker, Kubernetes CNI awareness, and fail2ban with compliance mapping to CIS/PCI-DSS/SOC2.

When to Use

• Check if a Linux server has active firewall protection.
• Enable and configure a firewall without locking yourself out of SSH.
• Audit existing rules, troubleshoot connectivity, or apply a security profile.
• Automate firewall hardening via an AI agent or CI/CD pipeline.

When NOT to Use

Condition	Alternative
Kubernetes worker node	Use NetworkPolicies / CiliumNetworkPolicy
Firewall managed by Terraform/Ansible/Puppet/Chef	Update IaC source of truth
Cloud workload with Security Group / NSG only	Use cloud provider's firewall API
Inside a container	Escalate to host operator
WSL2, macOS, or shared/managed hosting	See references/special-environments.md

Support files: scripts/audit-firewall.sh (run first), scripts/firewall-plan.sh (dry-run), scripts/firewall-verify.sh (post-apply). Detailed backend guides, Docker/K8s policies, observability, compliance, and recovery are in references/.

---

Prerequisites

• Root or sudo access.
• An active SSH session (risk of lockout).
• Know which ports your services use.

---

NEVER DO (14 Rules)

1. Never flush iptables/nftables on Kubernetes nodes. CNI plugins manage netfilter.
2. Never run iptables -F or nft flush ruleset without a verified backup. Docker/K8s networking will break.
3. Never disable firewalld and use raw iptables simultaneously. Undefined behavior.
4. Never set DROP policy on INPUT before allowing your current SSH port. Immediate lockout.
5. Never disable Docker's iptables management without replacement NAT/routing rules.
6. Never restart networking.service or NetworkManager remotely without console access.
7. Never apply cloud SG and host firewall changes simultaneously without testing.
8. Never enable logging on high-traffic DROP rules without limit rate. Disk flood.
9. Never manage nftables/iptables directly when ufw or firewalld owns the policy. Split-brain state.
10. Never apply outbound default-deny without explicitly allowing DNS, NTP, package mirrors.
11. Never restore firewall backups from a different host, kernel version, or backend mode.
12. Never assume IPv4 rules protect IPv6. Verify both stacks separately.
13. Never change sysctl hardening values on K8s/CNI hosts without explicit CNI profile support.
14. Never enable verbose packet logging without rate limits and log rotation.

---

State Machine

Follow states in order. Do not skip.

DETECT → SELECT → PLAN → VALIDATE → APPLY → VERIFY

State: DETECT

Run the audit script:

bash scripts/audit-firewall.sh           # Human-readable
bash scripts/audit-firewall.sh --json    # Machine-readable

Key outputs: confidence, risk_tier, recommended_backend, halt_reasons, k8s_node, iac_owner.

Risk Tiers & Confidence Gating

Tier	Confidence	Agent Behavior
auto	≥ 90%	Proceed automatically to PLAN
confirmed	70–89%	Proceed but require human confirmation before APPLY
manual	50–69%	Audit-only mode. Generate recommendations, do not apply.
halt	< 50%	Stop immediately. Escalate findings to operator.

Additional halt triggers (regardless of confidence): containerized, K8s node, IaC managed, no rollback mechanism available.

Decision Tree

Condition	Path	Detail
Risk tier = halt	STOP	Resolve blockers first
Inside container	STOP	Escalate to host operator
K8s node detected	STOP	references/k8s-policy.md
Ubuntu/Debian + ufw active	Phase: UFW	references/backend-ufw.md
ufw + firewalld both active	STOP	Resolve conflict
RHEL/Rocky/Alma + firewalld active	Phase: firewalld	references/backend-firewalld.md
nftables active, no frontend	Phase: nftables	references/backend-nftables.md
iptables only	Phase: iptables	references/backend-iptables.md
Docker host	Apply Docker Hardening after phase above	references/docker-hardening.md

Ownership Boundary

Before modifying rules, verify no IaC tool manages the firewall. If Terraform/Ansible/Puppet/Chef/cloud-init is detected → do not mutate. Update the source of truth instead. Full detection logic is in scripts/audit-firewall.sh.

---

State: SELECT

Optionally load a pre-built security profile (references/security-profiles.md):

Profile	Use Case
public-web-server	Open 22, 80, 443. Rate-limit SSH.
internal-database	SSH from mgmt subnet only. DB port from app subnet only.
bastion-host	SSH only. Aggressive rate limiting.
zero-trust-node	Default deny all inbound and outbound.

Or use declarative YAML (references/declarative-policy.md):

Imperative (state machine) → Ad-hoc hardening, incident response
Declarative (YAML)        → GitOps, multi-host, reproducible
Mixed                     → YAML as source-of-truth, state machine for verification

---

State: PLAN

Generate a dry-run diff before applying:

bash scripts/firewall-plan.sh --profile public-web-server
bash scripts/firewall-plan.sh --ports 22,80,443
bash scripts/firewall-plan.sh --json     # Machine-readable diff

Review the output. If risk_tier is confirmed, present the plan and wait for human confirmation before APPLY.

Plan JSON schema (machine-readable output):

{
  "backend": "ufw",
  "current_state": "active",
  "diff": {
    "add":    [{"port": 80, "proto": "tcp", "source": "any"}],
    "skip":   [{"port": 22, "proto": "tcp", "reason": "already_exists"}],
    "remove": []
  },
  "risk_assessment": "low",
  "estimated_disruption": "none",
  "approval_token": "sha256:abc123..."
}

Approval gate: PLAN output includes an approval_token (hash of plan content). APPLY must be called with --approved-plan=<token>. Token mismatch → exit code 41. This forces explicit human confirmation before Apply.

Audit caching: firewall-plan.sh internally calls audit-firewall.sh --json and caches to /tmp/firewall-audit.json (TTL 5 min). Use --refresh-audit to force refresh.

---

State: VALIDATE

1. Create Backup (Mandatory)

BACKUP_DIR="$HOME/firewall-backup-$(date +%Y%m%d-%H%M%S)"
mkdir -p "$BACKUP_DIR"

sudo iptables-save > "$BACKUP_DIR/iptables-v4.rules" 2>/dev/null || true
sudo ip6tables-save > "$BACKUP_DIR/iptables-v6.rules" 2>/dev/null || true
sudo nft list ruleset > "$BACKUP_DIR/nftables.rules" 2>/dev/null || true
sudo ufw status verbose > "$BACKUP_DIR/ufw-status.txt" 2>/dev/null || true
sudo firewall-cmd --list-all --zone=$(sudo firewall-cmd --get-default-zone) > "$BACKUP_DIR/firewalld-default.txt" 2>/dev/null || true

echo "Backup saved to $BACKUP_DIR"

2. Schedule Rollback (Mandatory for Remote)

The rollback restores from backup — not just disables the firewall — so Docker NAT and pre-existing rules are preserved. Dual-backend: at preferred, systemd-run fallback. Full script in SKILL.md under Validate state (see previous version), also summarized in references/recovery.md.

3. Pre-Flight Checklist

• Backup created successfully
• Rollback scheduled (verify with atq or systemctl --user list-units)
• Second SSH session open and idle
• Real SSH port identified
• Confidence ≥ 70% and risk_tier is auto or confirmed
• Ownership verified — no IaC managing firewall
• Change window appropriate (maintenance window or low traffic)
• PLAN output reviewed and approved

---

State: APPLY

All commands use idempotent patterns: check-before-set. Safe to run repeatedly.

Backend	Pattern
ufw	sudo ufw status | awk '{print $1}' | grep -qx "22/tcp" || sudo ufw allow 22/tcp
firewalld	sudo firewall-cmd --query-service=ssh || sudo firewall-cmd --permanent --add-service=ssh
iptables	sudo iptables -C INPUT -p tcp --dport 22 -j ACCEPT 2>/dev/null || sudo iptables -A ...
nftables	Atomic ruleset: nft -c -f /etc/nftables.conf.new && nft -f /etc/nftables.conf.new

Full step-by-step commands per backend: references/backend-ufw.md, references/backend-firewalld.md, references/backend-nftables.md, references/backend-iptables.md.

Docker Hosts

Docker bypasses ufw by default. Use DOCKER-USER chain. Full guide: references/docker-hardening.md.

Kubernetes Nodes

Default: AUDIT-ONLY. Never modify host firewall. Full policy: references/k8s-policy.md.

---

State: VERIFY

Run post-hardening checks:

bash scripts/firewall-verify.sh

Success criteria (all must pass):

1. SSH remains reachable from current and second session
2. Only intended ports are externally reachable
3. Rules survive reboot (verified via service persistence)
4. IPv6 exposure matches IPv4 policy
5. Docker-published ports are intentional (no accidental 0.0.0.0)
6. fail2ban jails active (if installed) with correct backend
7. Rollback timer cancelled after successful verification

Verify behavior contract:

• Verify MUST complete within the rollback timer window (default 5 min)
• If verify times out before completion → timer auto-fires rollback (system-level protection)
• If verify FAILS but timer was already cancelled → manual rollback: bash scripts/manual-rollback.sh <backup_dir>
• The rollback is triggered by the timer (systemd-run/at), NOT by verify.sh itself — verify.sh exits with code 60 to signal failure, and the calling agent/scheduler handles the rollback decision

Exit Codes (Core Contract)

Code	Meaning	Agent Action
0	Success	Continue
10	Backend conflict	Halt; resolve manually
11	Backend detection failed	Halt; check firewall stack
12	Multiple backends active	Halt; resolve conflict
20	IaC-managed	Halt; update IaC source
21	Inside container	Halt; escalate to host operator
22	K8s node detected	Halt; audit-only mode
30	Low confidence (<70%)	Drop to audit-only mode
31	No rollback capability	Halt; ensure at or systemd-run
40	Preflight failed	Halt; check prerequisites
41	Plan approval mismatch	Halt; re-run PLAN with approval
42	Backup failed	Halt; resolve disk/permissions
50	Apply failed	Auto-rollback triggered
51	Apply partial	Auto-rollback triggered; verify backup
60	Verify failed	Auto-rollback triggered
61	State file conflict	Abort; resolve stale state

---

fail2ban Integration

If fail2ban is installed:

Host Firewall	Recommended backend
ufw	ufw or systemd
firewalld	firewalld
nftables	nftables
iptables	auto (default)

After changing backend: sudo fail2ban-client restart && sudo fail2ban-client status sshd.

---

Recovery

If you lose connectivity, priority order:

1. Wait for auto-rollback (scheduled during VALIDATE)
2. Use second SSH session
3. Cloud serial console / hypervisor console
4. Restore from backup
5. Emergency ACCEPT (last resort — exposes host completely)

Full procedures: references/recovery.md.

State Persistence & Interrupt-Resume

For agent interrupt-resume scenarios (e.g., Apply failed mid-run, agent restarted), the state machine writes a lightweight state file to enable recovery without starting from Detect:

STATE_DIR="$HOME/.firewall-hardening"
STATE_FILE="$STATE_DIR/state.json"

State file structure:

{
  "state": "validate",
  "started_at": "2026-05-11T16:00:00Z",
  "backend": "ufw",
  "risk_tier": "auto",
  "backup_dir": "/home/user/firewall-backup-20260511-160000",
  "rollback_timer_id": "firewall-rollback-12345",
  "plan_hash": "sha256:abc123..."
}

Resume logic:

• If state.json exists and started_at is within 1 hour → resume from that state
• If state.json is stale (>1 hour) → delete it and start fresh from Detect
• The file is advisory-only; agent can always restart from Detect

State persistence is optional. The skill defaults to restarting from Detect each run. Enable by creating $STATE_DIR before starting.

---

Cloud Security Group Reminder

The host firewall is your second layer. Verify cloud SGs are aligned:

Cloud	Outer Firewall
AWS	Security Groups
GCP	VPC Firewall Rules
Azure	Network Security Groups
DigitalOcean/Linode/Vultr	Cloud Firewall

Compatibility Matrix

Distro/Env	ufw	firewalld	nftables	iptables	Coverage
Ubuntu 22.04/24.04	Primary	—	Backend	Fallback	Full
Debian 12	Primary	—	Backend	Fallback	Full
RHEL 9	—	Primary	Native	Backend	Full
Rocky/Alma 9	—	Primary	Native	Backend	Full
Fedora 40+	—	Primary	Native	Backend	Partial
Alpine 3.18+	—	—	Native	Fallback	Partial
Arch	—	—	Native	Fallback	Community
Docker host	✅ DOCKER-USER chain	✅ docker-hardening.md	✅ docker-hardening.md	✅ docker-hardening.md	Full
LXC/LXD container	⚠️ Limited	⚠️ Limited	⚠️ Limited	⚠️ Limited	Partial
systemd-nspawn	⚠️ Limited	⚠️ Limited	⚠️ Limited	⚠️ Limited	Partial
WSL2	❌ Not supported	❌ Not supported	❌ Not supported	❌ Not supported	None

Container environments: Docker host is fully supported via DOCKER-USER chain. LXC/LXD/systemd-nspawn have limited support (kernel shares netfilter with host). WSL2 is explicitly unsupported. See references/special-environments.md.

---

Observability

Establish baselines after hardening: conntrack usage, dropped packet rates, fail2ban ban rate. Monitor for anomalies. Full guide: references/observability.md.

Compliance

Practices map to CIS, PCI-DSS, and SOC2 controls. Full mapping: references/compliance.md.

Quick Reference

Task	Command
Audit environment	bash scripts/audit-firewall.sh --json
Plan changes	bash scripts/firewall-plan.sh --profile web
Verify after apply	bash scripts/firewall-verify.sh
Allow port (ufw, idempotent)	sudo ufw status | awk '{print $1}' | grep -qx "80/tcp" || sudo ufw allow 80/tcp
View ufw rules	sudo ufw status numbered
View nft rules	sudo nft list ruleset
View iptables rules	sudo iptables -L -n -v
View ip6tables rules	sudo ip6tables -L -n -v
Atomic iptables replace	sudo iptables-restore < /tmp/rules.v4
Dry-run nftables	sudo nft -c -f /etc/nftables.conf
Backup rules	sudo iptables-save > ~/iptables.backup
fail2ban status	sudo fail2ban-client status sshd
Cancel rollback (at)	atrm <jobid>
Cancel rollback (systemd-run)	systemctl --user stop firewall-rollback-<pid>

See Also

• references/backend-ufw.md — Full UFW phase
• references/backend-firewalld.md — Full firewalld phase
• references/backend-nftables.md — Full nftables phase
• references/backend-iptables.md — Full iptables phase
• references/docker-hardening.md — Docker firewall hardening
• references/k8s-policy.md — Kubernetes node policy
• references/security-profiles.md — Pre-built configurations
• references/declarative-policy.md — YAML policy schema
• references/observability.md — Monitoring and baselines
• references/compliance.md — CIS/PCI-DSS/SOC2 mapping
• references/recovery.md — Recovery procedures
• references/special-environments.md — WSL2, containers, exit codes
• scripts/audit-firewall.sh — Environment detection
• scripts/firewall-plan.sh — Dry-run diff
• scripts/firewall-verify.sh — Post-apply verification