Fortinet NSE4_FGT_AD-7.6: FortiGate Central NAT
Central NAT changes the way a FortiGate administrator separates address translation from security policy. In the default policy-NAT model, source NAT can be enabled directly on the IPv4 firewall policy. When central NAT is enabled, that relationship changes: source translation is defined in the central SNAT table, and destination NAT is handled through central DNAT and VIP objects rather than being embedded in the firewall policy in the same way. The operational benefit is clearer translation control; the risk is troubleshooting with the wrong mental model.
Central NAT is a practical FortiGate Administrator topic because routing, firewall policy, and address translation are independent decisions that can each be correct or wrong without proving the others. A packet can have the correct route and match the correct security policy yet still leave with the wrong source address. Conversely, a NAT rule can be correct while the firewall policy denies the session. Treating all three decisions as one problem leads to unnecessary configuration changes.
Within Fortinet Security Operations, Central NAT should be operated as a controlled translation policy. Teams need to know which address should be presented externally, which rule owns that decision, how ordering affects matches, and how changes interact with logging, HA, routing, and upstream allowlists.
Enabling Central NAT changes where source translation is decided
In FortiOS 7.6.0 profile-based NGFW mode, Central NAT is not enabled by default. Policy-based NGFW mode is different: Central SNAT is always enabled, and an administrator must define an appropriate central SNAT policy for traffic that needs source translation. When Central NAT is enabled, the NAT option under individual IPv4 firewall policies is skipped and SNAT is handled through the central-snat-map. That detail is fundamental during troubleshooting because an engineer inspecting only the firewall policy can incorrectly conclude that no translation should occur. Translation debugging should capture the pre-NAT tuple, post-NAT tuple, matched security rule, matched Central NAT rule, and return traffic. That evidence prevents engineers from changing security policy to compensate for a translation decision that occurred in a separate rulebase.
Central SNAT rules are processed from the top down until a matching rule is found. Matching can consider source, destination, and source port, and the rule can select the IP pool used for translation. Because ordering matters, a broad rule placed above a more specific rule can produce valid but unintended translation. The packet may still pass security policy, which makes the incident appear application-specific until the translated source is inspected.
NAT and PAT change identity at the network boundary, which affects how upstream systems and logs interpret the session. That identity can affect external allowlists, logging attribution, return routing, rate limits, and service-side access control. Translation is not just a convenience for conserving addresses.
Security policy and NAT policy must be debugged separately
FortiOS applies central SNAT after the security policy decision. That means the firewall policy evaluates the pre-translation source and destination context according to the relevant processing logic, while central SNAT determines how the permitted session is represented as it leaves. A troubleshooting sequence should therefore first prove routing and policy acceptance, then prove the expected translation rule and IP pool.
This separation prevents a common operational mistake: modifying the firewall rule because a remote service sees an unexpected source address. If the session is already allowed, widening the security policy cannot correct the identity presented after SNAT. The right evidence is the session table, NAT decision, matching central SNAT entry, and packet capture on the egress side.
FortiGate policy logic has to be verified separately from routing and NAT because each stage can fail independently. A policy hit counter proves a rule is seeing traffic; it does not prove that routing, NAT, inspection, or return traffic behaves as intended. Each stage needs its own verification.
Central DNAT changes how VIPs participate in policy
Destination translation also looks different in central NAT mode. Fortinet documents that VIPs are configured as separate objects and are not configured directly in the firewall policy in the same way as the non-central mode. Administrators moving between the two models must account for that change before assuming an old runbook still applies.
This is particularly important during migrations. Existing VIP objects can carry over between modes, but an object assigned to a firewall policy in non-central NAT mode may need to be unassigned before switching. A change that alters NAT mode without reviewing these relationships can create partial reachability failures: the object still exists, the policy still exists, but their relationship no longer matches the processing model.
Inbound troubleshooting should identify the public address, determine the DNAT/VIP object responsible for mapping it, confirm the translated destination, verify routing toward the internal server, and then inspect the security policy and return translation. Starting with the internal server address alone skips the public-side identity that the client actually used.
For a profile-based DNAT publication, create a virtual IP under Policy & Objects > Virtual IPs, defining the public address, mapped internal server and port-forwarding behavior only when the actual service requires it. In the applicable non-central mode, select the VIP as the destination of an appropriately scoped inbound firewall policy. Central DNAT processes those objects differently, so a non-central VIP runbook must not be copied unchanged. Validate the public destination, translated address, application port, matching policy, return route and a negative test that must remain blocked.
DNAT troubleshooting should include a positive and a negative VIP example. In a profile-based, noncentral policy configuration, an administrator defines the external-to-mapped-address VIP and applies it to the destination of the appropriate inbound firewall rule. In central-DNAT mode, review the separate central DNAT rule and policy relation instead of assuming the same VIP-policy editor behavior. Check whether an external address and port actually map to the intended internal server, whether an earlier rule captures the connection, and whether the server reply returns through the expected stateful boundary. A successful connection to one service does not prove that every port is deliberately published, so test another service that should remain denied.
Rule ordering should express intent explicitly
Central SNAT becomes easier to operate when rules are ordered from specific exceptions toward broader defaults. A rule for a partner application, dedicated egress IP, or regulated workload should be clearly separated from general internet translation. If the configuration depends on subtle overlaps between broad address groups, engineers have to reconstruct intent every time they troubleshoot.
Change reviews should therefore include overlap analysis. Ask whether the new rule can match traffic already handled by an earlier entry, whether its IP pool is valid on the egress interface, whether the destination condition is necessary, and whether the external address is referenced by upstream firewalls or SaaS allowlists. A technically valid rule can still be operationally dangerous if it changes a source identity used elsewhere.
NAT shapes connectivity by changing the identities seen by endpoints and intermediate controls, so FortiGate rule order has operational consequences. Translation affects the conversation seen by both endpoints, so a Central NAT change should be treated as an application-path change even when no security policy is modified.
Routing and return traffic determine whether translation succeeds
A correct NAT rule cannot compensate for a wrong route. FortiGate still needs an egress path toward the translated destination or upstream network, and the far side needs a return path to the translated source. When traffic leaves with a different source address, external routing and filtering must be prepared for that identity.
Asymmetric return paths can complicate the picture further, especially around stateful inspection and HA. If the outbound session is created on one FortiGate path but return traffic reaches another enforcement point that lacks the corresponding state, the symptom may look like a NAT failure even though the translation itself is correct. Packet captures on both directions help separate state, route, and translation problems.
FortiGate routing relationships belong in NAT troubleshooting because translation cannot complete a session when the forward or return path is wrong. The session table tells you what the firewall decided; the routing table and captures tell you whether the rest of the network can complete the conversation.
Logging should preserve both original and translated context
Operations teams need enough logging detail to correlate a client’s original identity with the address seen by the remote service. Without both sides of the translation, an incident can become a manual search through pools and timestamps. This is particularly important when many internal systems share one public address or when different applications use different egress pools.
The translation record also supports security investigations. If a third party reports abusive traffic from a public IP, analysts must map that address and time back to the internal session. NAT therefore creates an attribution requirement. Retention, timestamp accuracy, and session logging should be designed with that investigation in mind rather than only with day-to-day troubleshooting.
FortiGate troubleshooting evidence is especially valuable when a NAT change is suspected because pre-change and post-change sessions can prove whether the translated source, destination, or selected pool actually changed. That record prevents a later routing or policy adjustment from obscuring the translation decision that caused the original symptom.
Migration to Central NAT needs a controlled verification plan
Organizations often enable Central NAT because they want translation rules managed independently from firewall policy. The migration should still be treated as a production redesign. Inventory existing policy-based NAT, IP pools, VIPs, central rules if any, and external dependencies. Then define the intended equivalent state before enabling the new mode.
Verification should include representative outbound applications, inbound published services, exception pools, VPN traffic where relevant, return routing, HA behavior, and logging. A simple internet-browsing test proves only that a broad outbound path works. It does not prove that partner allowlists, published services, or application-specific translations survived the change.
For Fortinet administrators, the strongest Central NAT practice is to keep the translation model explainable. An engineer should be able to identify the matching security policy, the matching NAT rule, the selected pool or VIP, and the resulting session without guessing. That clarity is what makes the feature scalable.
The safest operational convention is to name translation rules for their business purpose rather than only their address objects. “Partner API egress” or “shared services outbound” gives the reviewer a reason for the rule, while a name built from pool numbers forces interpretation from raw configuration. Clear naming does not change packet processing, but it reduces the chance that an emergency edit modifies a rule whose wider scope was not understood.
Central NAT is not inherently safer or more complex than policy-based NAT. It is a different separation of responsibilities. When rule order, routing, logging, and external dependencies are understood, the model can make translation policy easier to govern. When those relationships are hidden, the same centralization can make a small rule-order mistake affect far more traffic than expected.