What do you need help with?

We are here and ready to help.
Email: servicedesk@socfortress.co

SIEM Stack Host Firewall: Required Inbound and Outbound Ports

Category: SIEM / Infrastructure / Hardening
Applies to: SOCFortress SIEM stack deployments (CoPilot, Wazuh 4.14.x, Wazuh Indexer, Graylog, Grafana, Velociraptor, Portainer, HAProxy, Talon, SOCFortress UBA) on Debian/Ubuntu VMs
Audience: Customer IT / Security teams hardening the stack VMs


Summary

SOCFortress deploys and configures the SIEM software. OS-level hardening of the VMs (host firewalls, sudo policy, CIS baselines) is yours to apply. You are welcome to apply it, and this article is here to keep that from taking the stack down.

Most firewall outages on the stack come from one of three things:

  1. A firewall reload that wipes Docker's own rules, which drops container networking.
  2. A missing port between two stack components. Agents, alerts, or dashboards then go quiet, usually with no error you would notice.
  3. Default-deny outbound rules that block updates, licensing, threat intel, or cloud integrations that poll vendor APIs.

The port numbers below are the defaults we deploy. Your environment may differ, so always confirm with the listening-port check in step 1 before writing rules.


Before you start

  • Take a snapshot of each VM before you change it.
  • Roll out one host at a time, starting with the least critical (for example Grafana or Talon) and finishing with the Wazuh master and the indexer nodes.
  • Keep a timed rollback armed while you test (see "Safe rollout" below).
  • Tell SOCFortress support before you start (case or portal). We can review your ruleset first, and we will know what changed if something breaks.

Step 1: See what each host actually listens on

Run this on every stack VM and keep the output. It is the source of truth for your inbound rules.

sudo ss -tlnpu

Anything listening on 127.0.0.1 or ::1 is local only and needs no firewall rule. Focus on services bound to 0.0.0.0, ::, or the VM's LAN IP.


Step 2: Inbound rules

Applies to every stack VM

Source Port Purpose
Internal proxy All service ports on the VM The internal proxy fronts the stack services and must reach every VM. Allow it first.
Admin / SOCFortress support jump host 22/tcp SSH for administration and support
Monitoring (if used) As configured Your own monitoring agents

Per component

Component Port(s) Allow from
CoPilot 443/tcp Analyst / admin networks, internal proxy
Grafana 3000/tcp CoPilot, internal proxy, analyst networks
Portainer 9000/tcp (9443/tcp if HTTPS UI is used) Admin networks, SOCFortress support
Wazuh manager (master) 1514/tcp (agent events), 1515/tcp (enrollment) Agents (directly or via external proxy)
Wazuh manager (master) 1516/tcp (cluster) All Wazuh worker nodes / worker containers
Wazuh manager (master) 55000/tcp (API) CoPilot
Per-customer Wazuh workers 20000-21000/tcp External proxy (HAProxy)
Wazuh worker provisioning app (worker host) 5003/tcp CoPilot (used during customer provisioning)
Wazuh Indexer 9200/tcp (REST) CoPilot, Graylog, Grafana, Wazuh manager
Wazuh Indexer 9300/tcp (transport) Other indexer nodes only
Graylog Web/API port (9000/tcp by default) CoPilot, internal proxy, admin networks
Graylog Input ports (Fluent Bit / syslog / integrations) Wazuh manager (Fluent Bit), syslog sources, collectors. Check your configured inputs under System > Inputs.
Velociraptor 8000/tcp (client frontend) Endpoints (directly or via external proxy)
Velociraptor 8001/tcp (API) CoPilot
Velociraptor 8889/tcp (GUI) Admin / analyst networks
External proxy (HAProxy) 1514, 1515, 8000, 20000-21000/tcp Internet / agent networks
HAProxy provisioning app (external proxy) 5003/tcp CoPilot (used during customer provisioning)
Talon analyst (if deployed) 3100/tcp CoPilot only
SOCFortress UBA (if deployed) 8010/tcp CoPilot only

MongoDB (27017) on the Graylog host should only listen locally. If it is bound to a LAN IP, do not open it to anything beyond Graylog.


Step 3: Outbound rules

If you only filter inbound, skip this section. If you plan default-deny egress, every destination below needs an allow rule. Missing ones fail quietly: updates stop, feeds go stale, integrations stop ingesting.

Every VM

Destination Port Purpose
Your DNS resolvers 53/udp+tcp Name resolution
Your NTP servers 123/udp Time sync (cluster, certificates, and log timestamps depend on it)
OS package mirrors (deb.debian.org, security.debian.org, or your Ubuntu / local mirror) 443/tcp, 80/tcp OS updates

Stack updates and images

Destination Port Used by
registry-1.docker.io, auth.docker.io, production.cloudflare.docker.com 443/tcp Docker image pulls (CoPilot, workers, Portainer)
ghcr.io, pkg-containers.githubusercontent.com 443/tcp SOCFortress container images
github.com, raw.githubusercontent.com, objects.githubusercontent.com 443/tcp Install scripts, rules, and config repos
packages.wazuh.com 443/tcp Wazuh packages
cti.wazuh.com 443/tcp Wazuh vulnerability detection feed
packages.graylog2.org, artifacts.opensearch.org 443/tcp Graylog / indexer packages
apt.grafana.com 443/tcp Grafana packages
packages.fluentbit.io 443/tcp Fluent Bit packages
download.maxmind.com 443/tcp GeoLite2 database updates (Graylog GeoIP)

CoPilot and Talon

Destination Port Used by
SOCFortress licensing and add-on services (Threat Intel, exclusion rule generator) 443/tcp CoPilot license validation and licensed features. Contact support for the current hostnames.
api.anthropic.com (or your configured model provider) 443/tcp Talon, when using a hosted model

Cloud integrations (only the ones you use)

These are polled from the CoPilot / Graylog / Wazuh hosts, depending on the integration. Examples:

Integration Destinations
Microsoft 365 / Entra / Defender / Intune login.microsoftonline.com, graph.microsoft.com, manage.office.com, api.security.microsoft.com
SentinelOne Your tenant URL (for example .sentinelone.net)
CrowdStrike Your cloud's API host (for example api.crowdstrike.com, api.us-2.crowdstrike.com)
AWS (CloudTrail, GuardDuty via S3) s3..amazonaws.com, sts.amazonaws.com
Google Cloud (Pub/Sub) pubsub.googleapis.com, oauth2.googleapis.com
Email alerts (Exchange Online) smtp.office365.com:587
Teams webhooks Your workflow / webhook URL host

If an integration is not listed, check the vendor's API documentation or ask support.


Docker: the part that breaks most often

Most stack VMs run containers (CoPilot, per-customer Wazuh workers, Portainer). Docker writes its own iptables/nftables rules for container networking and published ports. Two things to know:

  1. Never flush the whole ruleset. A typical Debian /etc/nftables.conf starts with flush ruleset. Reloading it deletes Docker's chains, and every container loses networking until Docker restarts. Use your own table and flush only that:

``` #!/usr/sbin/nft -f table inet filter flush table inet filter

table inet filter { chain input { type filter hook input priority 0; policy drop; ct state established,related accept iif lo accept # your allow rules here } } ```

The same caution applies to firewall-cmd --reload on firewalld systems.

  1. Published container ports bypass your INPUT chain. Traffic to a port that Docker publishes is DNAT'd and passes through the FORWARD path, so INPUT rules (and UFW) do not filter it. To restrict who can reach a published container port, add rules to the DOCKER-USER chain, which Docker leaves alone:

bash # Example: only allow the external proxy (172.21.0.10) to reach worker ports sudo iptables -I DOCKER-USER -p tcp -m conntrack --ctorigdstport 20000:21000 ! -s 172.21.0.10 -j DROP

Replace the IP with your proxy's address, and persist the rule with your normal tooling.

If container networking drops after a firewall change, the quickest recovery is:

sudo systemctl restart docker

Safe rollout

Arm an automatic rollback before you apply rules on each host. If you lose access or the stack goes quiet, the rules remove themselves:

# Requires the 'at' package
echo "nft delete table inet filter" | sudo at now + 10 minutes
sudo nft -f /etc/nftables.conf

Run the self-test below. If everything passes, cancel the rollback:

sudo atq # find the job number
sudo atrm <job>

Self-test (run after each host)

On the host you just changed:

# Services still listening
sudo ss -tlnpu

# Containers healthy (Docker hosts)
sudo docker ps --format 'table {{.Names}}\t{{.Status}}'

From the other components, test reachability to the changed host:

nc -zv <host> <port>

Component checks:

# Wazuh master: cluster nodes connected
sudo /var/ossec/bin/cluster_control -l

# Wazuh master: agents active
sudo /var/ossec/bin/agent_control -l | grep -c Active

# Indexer: cluster health green/yellow, all nodes present
curl -sk -u admin https://localhost:9200/_cluster/health?pretty

# Outbound (if egress is filtered)
curl -sI https://packages.wazuh.com | head -1
curl -sI https://ghcr.io | head -1

Then confirm in the UIs:

  • CoPilot loads and the agent list shows recent check-ins.
  • Graylog > System > Inputs shows messages arriving on every input.
  • Grafana dashboards show data in the last 5 minutes.
  • Velociraptor shows clients online.

Provisioning is easy to miss because it only runs when you onboard a new customer. Blocking 5003 won't affect anything day to day, but the next customer you provision in CoPilot will fail. Test it from the CoPilot host:

nc -zv <wazuh-worker-host> 5003
nc -zv <external-proxy> 5003

If any check fails, let the timed rollback fire (or remove the table yourself), then open a support case with your ss -tlnpu output and the ruleset you applied.


A note on sudo hardening

Requiring a password for sudo does not affect the stack services. Make sure the SOCFortress support account has a password shared with us through OneHub so support access is not lost.

Facebook Share Tweet

Was this article helpfu?

Yes No

Thank you for voting

×
Select company

You are related to multiple companies. Please select the company you wish to login as.