Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Azure network security groups (NSGs) filter traffic to an Azure Linux 4 virtual machine (VM) at the subnet or network interface. Kernel parameters harden how the guest handles network traffic. A host firewall inside the guest is optional, and you add it when your workload needs one.
Default network filtering and hardening settings
The general-purpose image doesn't include an active host firewall. firewalld
isn't installed. The nftables and iptables-legacy tools are installed, but
the nftables service isn't enabled and no inbound filtering rules are loaded.
Inbound filtering depends on the NSG rules for the VM. If your workload needs a
host firewall, install firewalld from the Azure Linux repository or manage
nftables rules directly.
IP forwarding, ICMP redirects, and source routing are disabled by the network
hardening defaults. At first boot, cloud-init generates the primary interface
configuration for systemd-networkd from Azure metadata.
Prerequisites
- An Azure Linux virtual machine (VM).
- A user account with
sudoprivileges. - Access to the NSG rules for the VM's subnet and network interface.
- A running application to test connectivity after adding a firewall rule.
Verify host firewall, listeners, and network kernel parameters
Connection handling depends on the applicable Azure rules, the host firewall if you installed one, and the application's listener.
How the filtering layers interact
- Azure NSGs filter traffic in the virtual network before it reaches the VM's network interface. Inspect their rules through Azure; guest firewall commands don't show or change them.
- A host firewall, such as
firewalld, filters traffic that already reached the VM.firewalldis a zone-based front end fornftables. Named services such ashttpsdefine traffic rules; they don't authenticate or identify the application process.
For a failed connection, check the listener, the NSG rules, and the host firewall if one is running. Routing and application configuration also affect the result.
Check for a host firewall
Check whether firewalld is installed and whether any nftables rules are
loaded:
rpm -q firewalld
systemctl is-enabled nftables
sudo nft list ruleset
Expected result for the default configuration:
package firewalld is not installed
disabled
An empty nft list ruleset result means that no nftables rules are loaded.
The Azure guest agent adds iptables rules to the security table that limit
outbound connections to the Azure host endpoint, 168.63.129.16. Those rules
protect the agent's communication with the platform. They don't filter inbound
traffic to your workload. To see them, run sudo iptables -t security -S.
See what the host exposes
Firewall rules matter only in relation to what's listening behind them. List listening sockets and the processes that own them:
sudo ss -tulnp
Read the local address column:
127.0.0.1or::1means the service accepts local connections only, and no ordinary port-opening rule makes it listen on another interface.0.0.0.0or::means the service accepts connections on every interface, for the corresponding address family. Review whether that exposure is necessary.
Audit network kernel parameters
Compare the vendor network settings with the effective runtime values:
cat /usr/lib/sysctl.d/70-azurelinux-hardening.conf
sysctl \
net.ipv4.ip_forward \
net.ipv6.conf.all.forwarding \
net.ipv4.conf.all.rp_filter \
net.ipv4.conf.default.rp_filter \
net.ipv4.conf.all.accept_redirects \
net.ipv6.conf.all.accept_redirects \
net.ipv4.conf.all.accept_source_route \
net.ipv4.conf.all.send_redirects \
net.ipv4.conf.all.log_martians \
net.ipv4.tcp_syncookies
Expected result for the default configuration:
net.ipv4.ip_forward = 0
net.ipv6.conf.all.forwarding = 0
net.ipv4.conf.all.rp_filter = 1
net.ipv4.conf.default.rp_filter = 1
net.ipv4.conf.all.accept_redirects = 0
net.ipv6.conf.all.accept_redirects = 0
net.ipv4.conf.all.accept_source_route = 0
net.ipv4.conf.all.send_redirects = 0
net.ipv4.conf.all.log_martians = 1
net.ipv4.tcp_syncookies = 1
| Parameter | Why the value matters |
|---|---|
ip_forward, forwarding |
The VM doesn't route traffic between interfaces, so it can't be used as an unintended path into another subnet. Routing software, NAT, and container networking turn this on deliberately. |
rp_filter |
Strict reverse path filtering drops a packet when the reply would leave through a different interface, which discards the simplest form of source address spoofing. |
accept_redirects |
The host ignores ICMP redirects, so a device on the same segment can't rewrite its routing table. |
accept_source_route |
Source-routed packets are dropped, so a sender can't dictate the path its traffic takes through your network. |
send_redirects |
The host doesn't advertise better routes to other systems. |
log_martians |
Packets with impossible source addresses are logged, which turns spoofing attempts into evidence you can query. |
tcp_syncookies |
The host keeps completing legitimate handshakes when the SYN backlog fills, which limits the effect of a SYN flood. |
The kernel applies the higher of the all and per-interface values for
rp_filter. Check the interface values as well as the global setting:
sudo sysctl -a | grep 'rp_filter'
Add a host firewall and change network settings
Firewall allowances should match application requirements. Changes to network hardening defaults require a reviewed network-design exception.
Add a host firewall
Add a host firewall when NSG rules alone don't meet your requirements, for example when you need per-VM rules managed inside the guest.
Warning
When firewalld starts, its default public zone allows only the ssh and
dhcpv6-client services. Other services that listen on the network become
unreachable until you allow them. Before you start the firewall, confirm you
can reach the VM through the
serial console
or Run command.
List the services the VM exposes, so you can allow the ones the workload needs:
sudo ss -tulnpInstall
firewalld:sudo dnf install -y firewalldThe package preset enables the service, so it starts at the next restart even if you don't start it now.
Start the firewall:
sudo systemctl enable --now firewalldConfirm the state, backend, default zone, and allowed services:
sudo firewall-cmd --state sudo grep -E '^FirewallBackend=' /etc/firewalld/firewalld.conf sudo firewall-cmd --get-default-zone sudo firewall-cmd --list-servicesrunning FirewallBackend=nftables public dhcpv6-client sshAdd a rule for each workload service, as described in the next section.
Check active zones before changing a rule. A zone can be selected by an interface or source binding; otherwise, the default zone applies:
sudo firewall-cmd --get-active-zones
Add a rule without locking yourself out
firewalld keeps runtime and permanent configurations separately. This example
adds HTTPS to one zone without saving unrelated runtime changes. Use the zone
that handles the service's traffic; public is the firewalld default.
Warning
Removing the SSH allowance can prevent new management connections. Before you change SSH rules, confirm you can reach the VM through the serial console or Run command.
Select the zone and add the runtime rule:
ZONE=public sudo firewall-cmd --zone="$ZONE" --add-service=httpssuccessConfirm the rule is live:
sudo firewall-cmd --zone="$ZONE" --query-service=httpsyesTest the application from a client. Once it works, persist only this rule and confirm the permanent state:
sudo firewall-cmd --permanent --zone="$ZONE" --add-service=https sudo firewall-cmd --permanent --zone="$ZONE" --query-service=httpsThe query returns
yes. A firewall query confirms the rule, not that the application is listening or that Azure permits the connection.If you added the rule for this example and no longer need it, remove it from both configurations:
sudo firewall-cmd --zone="$ZONE" --remove-service=https sudo firewall-cmd --permanent --zone="$ZONE" --remove-service=https
Avoid --runtime-to-permanent for a single-rule change unless you intend to
save all active runtime configuration. It can persist changes unrelated to
your task.
If you edit permanent files directly, reload to apply them. A reload replaces runtime configuration, including unsaved runtime changes:
sudo firewall-cmd --reload
Tip
For a temporary service allowance, use
sudo firewall-cmd --zone="$ZONE" --add-service=https --timeout=5m.
The allowance expires automatically. This isn't a rollback mechanism for
other changes, such as removing SSH access.
Change a network parameter
To change a parameter permanently, write it to a file under /etc/sysctl.d/
with a name that sorts after 70-azurelinux-hardening.conf, such as
90-network-local.conf. For the precedence rules and how to apply the change,
see
Inspect kernel hardening.
Override the Azure-generated interface configuration
At first boot, cloud-init reads the network configuration from Azure metadata
and writes a systemd-networkd file for the primary interface. It rewrites the
file at each boot, so edits to that file don't persist.
| File | Created by | Purpose |
|---|---|---|
/etc/systemd/network/10-cloud-init-eth0.network |
cloud-init | Configures eth0 with DHCP for IPv4. Rewritten at each boot. |
/usr/lib/systemd/network/01-azure-unmanaged-sriov.network |
azure-vm-utils package |
Leaves the Accelerated Networking virtual function unmanaged, because it's bonded to the synthetic interface. |
Identify the file that configures the primary interface:
networkctl status eth0 | grep 'Network File'
Network File: /etc/systemd/network/10-cloud-init-eth0.network
With Accelerated Networking, networkctl list also shows an interface such as
ens1 as enslaved and unmanaged. Don't add configuration for it.
To change a setting for eth0, add a drop-in file instead of editing the
generated file. Drop-in files persist across restarts and apply on top of the
generated settings. This example adds a DNS server:
Create the drop-in directory and file:
sudo mkdir -p /etc/systemd/network/10-cloud-init-eth0.network.d printf '[Network]\nDNS=<dns-server-ip>\n' | sudo tee \ /etc/systemd/network/10-cloud-init-eth0.network.d/50-local.confReload
systemd-networkdand confirm that the drop-in is applied:sudo networkctl reload networkctl status eth0 | grep -A1 'Network File' resolvectl dns eth0The status output lists the drop-in under the generated file:
Network File: /etc/systemd/network/10-cloud-init-eth0.network └─/etc/systemd/network/10-cloud-init-eth0.network.d/50-local.confThe
resolvectloutput lists your DNS server and the Azure resolver,168.63.129.16.
To remove the override, delete the drop-in file and run
sudo networkctl reload.
Caution
A separate .network file that sorts before 10-cloud-init-eth0.network and
matches eth0 replaces the generated configuration instead of extending it,
because systemd-networkd uses only the first matching file. Use a drop-in
unless you intend to own the entire interface configuration. Keep serial
console access available when you test interface changes.
Troubleshoot host firewall and network connectivity
A connection times out instead of being refused
A refusal can indicate no listener or an explicit rejection. A timeout can result from filtering, packet loss, or a routing problem. Start with the listener and the zone handling the traffic:
sudo ss -tulnp
systemctl is-active firewalld
If firewalld is active, list the zone that handles the traffic:
sudo firewall-cmd --get-active-zones
sudo firewall-cmd --zone=<zone-name> --list-all
If a process is listening and no host firewall blocks the port, review the NSG rules on the subnet and network interface in Azure.
A service became unreachable after installing firewalld
The public zone allows only SSH and DHCPv6 client traffic. List the zone and
add a rule for the service, as described in
Add a rule without locking yourself out:
sudo firewall-cmd --list-all
A change to the generated network file disappeared
cloud-init rewrites /etc/systemd/network/10-cloud-init-eth0.network at each
boot. Move the setting into a drop-in file, as described in
Override the Azure-generated interface configuration.
A firewall rule disappeared after a restart
Compare the runtime and permanent rules in the same zone. If the rule exists
only at runtime, add that specific rule with --permanent. Don't overwrite
the full permanent configuration just to save one service allowance.
Traffic is dropped after adding a second network interface
Asymmetric routing and strict reverse path filtering don't coexist. Look for the evidence in the kernel log:
sudo journalctl -k --grep='martian' --since '1 hour ago'
Fix the routing so replies leave through the interface the traffic arrived on.
Loosening rp_filter removes a spoofing control, so change it only when the
network design genuinely requires asymmetric paths.
IP forwarding is enabled and you didn't enable it
Container runtimes and Kubernetes node components enable forwarding by design. Confirm what set it before you change it back:
grep -rnsE 'ip_forward|conf\.(all|default)\.forwarding' \
/etc/sysctl.conf /etc/sysctl.d /usr/lib/sysctl.d
Turning forwarding off on a node that runs containers breaks pod networking.
firewall-cmd output doesn't match the live ruleset
Software that writes nftables rules directly bypasses firewalld. Compare
the generated ruleset with what firewalld reports. The nft command ships in
the nftables package:
sudo nft list ruleset