PCI DSS gives you six months between firewall rule reviews, and the standard is specific about the point of the exercise: confirming that every rule is still relevant and still effective. Most teams cannot answer that question for a large share of their rulebase, and the reason is not laziness. The firewall stores what a rule does. It does not store why the rule exists, who asked for it, or when it was supposed to go away. A review that starts from the running config alone is being asked to judge rules with half the evidence missing.

This post covers what the six-month review actually requires under PCI DSS 4.0.1, what to check on each rule, why hit counts are weaker evidence than they look, and how to set up your change process so the next review takes hours instead of weeks.

Key takeaway: A firewall rule review is a re-approval exercise, not a cleanup exercise. The unit you review is not the rule. It is the rule plus the change record that created it. If the record is missing, the review cannot be completed honestly, no matter how good your traffic data is.

Firewall rule review (also called firewall rule recertification) is the periodic process of checking every rule on a firewall against its documented business justification, confirming it is still needed and still as narrow as it can be, and removing or tightening the rules that are not. PCI DSS requires it at least once every six months for firewalls protecting cardholder data.

What does PCI DSS 4.0.1 require for firewall rule reviews?

Requirement 1.2.7 of PCI DSS 4.0.1 requires that the configurations of network security controls are reviewed at least once every six months to confirm they are “relevant and effective.” Network security controls is the v4 term that replaced “firewalls and routers,” so cloud security groups, ACLs, and host firewalls in scope count too. The assessor tests three things: that you have a documented review procedure, that reviews actually happened on schedule, and that rules with no remaining business justification were removed or updated as a result.

That last test is the one people miss. A review that produces a signed spreadsheet and zero changes to the rulebase, cycle after cycle, is a red flag to a competent QSA. 1.2.7 also does not stand alone. It depends on three neighbouring requirements that decide whether your review is even possible:

RequirementWhat it asks forWhy the review depends on it
1.2.2Every change to network connections and firewall configs is approved and managed through change controlThis is where the “why” for each rule gets recorded
1.2.5Every allowed service, protocol, and port is identified, approved, and has a defined business needThe review compares each rule against this list
1.2.7Configurations reviewed at least every six monthsThe review itself
1.2.8Config files are secured and consistent with the active configurationProves you reviewed what is actually running

PCI is the most explicit, but it is not the only framework asking. NIST SP 800-41 Rev. 1 recommends periodic review of firewall rulesets, and SOC 2 and ISO 27001 auditors look for the same thing under change management and network security controls. We covered the wider set of v4 changes in our PCI DSS v4.0 firewall requirements breakdown.

Why do most firewall rule reviews stall?

They stall because the reviewer is handed a config export and asked to judge intent from syntax. A policy that allows a DMZ host to reach an external address on TCP 22 tells you nothing about whether that is a live vendor integration or a migration leftover from two years ago. The person who knew may have left. The ticket, if one existed, lives in a system nobody linked to the rule.

So the reviewer does the rational thing: marks it “keep” because removing an unknown rule might break production, and nobody gets blamed for a rule that stays. Repeat that for six or eight cycles and you have a rulebase where a large fraction of entries have been “reviewed” many times and understood zero times. The review happened. The control did not.

What should you check on each firewall rule?

Check each rule against one question: would this be approved today, exactly as written? That breaks down into six checks, in this order, because the early ones are cheap and eliminate rules before you spend time on the expensive ones.

  1. Is there a change record? A ticket or change request ID, an approver, and a date. No record means the rule goes on the exceptions list straight away.
  2. Is there a living owner? A team or named role that will confirm the need. “Network team” is not an owner. The owner is whoever’s application breaks if the rule goes.
  3. Does the business need still exist? The vendor contract is active, the server is still in service, the project did not end.
  4. Is it as narrow as it can be? Any “all” in source, destination, or service needs a written reason. Check whether a /16 could be three hosts, or whether ALL could be TCP 443.
  5. Was it meant to be temporary? Anything created for a migration, a test, a vendor troubleshooting session, or an incident should have had an end date. If it did and the date has passed, remove it.
  6. Is it shadowed or redundant? A rule that can never match because a broader rule sits above it is dead weight, and dead weight hides real problems.

Put rules that touch the firewall’s own management plane and anything internet-facing at the top of the queue. That is where the exposure is.

Edge devices and VPNs made up 22% of vulnerability exploitation targets in the Verizon 2025 DBIR, up from 3% the year before.

That figure comes from the Verizon 2025 Data Breach Investigations Report, and the cases behind it are familiar. CVE-2024-55591, the FortiOS authentication bypass disclosed in January 2025, was exploited against firewalls whose management interfaces were reachable from the internet. An allow rule or access setting that exposed management to the world was almost never approved as such. It drifted there, and a real rule review is one of the few routine controls that would have caught it.

Can you use hit counts to decide which rules to remove?

No, not on their own. Hit counts tell you which rules to look at first. They do not tell you which rules are justified. On a FortiGate you can pull the counters for a policy from the CLI:

diagnose firewall iprope show 100004 118

The output includes the hit count plus first hit and last hit timestamps, the same data behind the Hit Count and Last Used columns in the policy list. Useful, but it fails as review evidence in both directions:

  • Zero hits does not mean unused. Counters do not survive a reboot, so a firmware upgrade last month wipes the history. Anyone with admin rights can also clear them. Disaster recovery paths, year-end batch jobs, and quarterly vendor uploads can all sit at zero for most of a review window and still be needed.
  • A stale Last Used does not mean idle. Fortinet documents that hit count and Last Used may not advance while existing long-lived sessions are still being handled by the policy. A database replication link that has been up for weeks can look quiet.
  • Lots of hits does not mean legitimate. This is the one that matters. An overly broad rule gets plenty of traffic precisely because it is overly broad. A rule added outside the change process and actively used by something it should not serve will have the healthiest counters on the box.

Use counters to sort the queue. Use the change record to make the decision. We went deeper on counter behaviour in FortiGate Security: 7 Things Most Admins Miss.

Review the changes, not just the rules

Here is the part most review procedures leave out. A point-in-time review of the current rulebase can only see what exists on review day. It cannot see a rule that was widened to “all” for three weeks in the middle of the cycle and narrowed back before anyone looked. It cannot see a rule that was added, used, and deleted. From a risk standpoint those are the most interesting events of the whole six months, and a snapshot review is blind to every one of them.

The stronger approach is to review the diff. Take the configuration as it stood at the last review, list every change since, and reconcile each one against an approved change request. That gives you three piles:

What you findWhat it meansWhat to do
Change with a matching approved requestProcess workedConfirm the rule matches what was approved, no wider
Change with no matching requestUnauthorized changeInvestigate first, then ratify or remove. This is a finding, not a paperwork gap
Approved request with no matching changeNever implemented, or implemented differentlyClose the request or find where the change actually landed

Rules that have not changed since the last review and were justified then only need the “does the need still exist” check. On a mature rulebase that turns a full recertification into a review of a few dozen changes plus a quick owner confirmation on the rest. It is also a better answer to the auditor, because it shows 1.2.2 and 1.2.7 working together instead of as two separate binders.

How do you make a rule reviewable from the day it is created?

Attach the justification to the rule at creation, in a fixed format, so the next reviewer does not have to go looking. The minimum is four fields: change request ID, owner, approval date, and review-by or expiry date. On FortiOS the comments field is the natural place:

config firewall policy
    edit 118
        set name "VendorX-SFTP-out"
        set srcintf "dmz"
        set dstintf "wan1"
        set srcaddr "app-sftp-01"
        set dstaddr "vendorx-sftp"
        set service "SSH"
        set action accept
        set schedule "always"
        set logtraffic all
        set comments "CR-2291 | owner: payments-ops | approved 2026-05-12 | review-by 2026-11-12"
    next
end

A fixed delimiter matters more than it looks. It lets you export the policy table and split the comment into columns, which turns “find every rule with no change record” into a one-line filter instead of a manual read-through.

For rules that are temporary by design, do not rely on someone remembering. Recent FortiOS releases support a per-policy expiry date:

config firewall policy
    edit 131
        set policy-expiry enable
        set policy-expiry-date 2026-12-15 00:00:00
    next
end

Watch for this: An expired policy stops being applied, but it stays in the configuration. It still needs to be deleted at the next review, or your rulebase fills up with dead entries that every future reviewer has to read past. Expiry reduces risk. It does not reduce clutter.

The weakness of the comment convention is that it is voluntary. Nothing on the firewall stops an admin from creating a policy with an empty comment or pasting in a ticket number that belongs to something else. The convention only holds if something outside the firewall checks each change against the request that authorised it.

What evidence will the auditor ask for?

Expect to hand over five things for each review cycle:

  • The written review procedure, including who performs it and who signs off
  • The dated record of each review, covering every in-scope device, with no gap longer than six months between them
  • The configuration that was reviewed, with proof it matches what was running (this is where 1.2.8 comes in)
  • The list of rules flagged, with the decision on each: kept with justification, tightened, or removed
  • The change requests that carried out those removals and updates, showing the review led to action

The assessor will usually sample. They pick a handful of rules and ask you to walk each one back to its approval and business need. If you can do that for any rule they point at, in a couple of minutes, the rest of the conversation is short.

Firewall rule review FAQ

How often should firewall rules be reviewed?

At least every six months if the firewall is in PCI DSS scope. That is a floor, not a target. Environments with a high volume of changes, including most MSP-managed estates, should review more often, and reviewing changes continuously as they happen removes most of the work from the formal six-month cycle.

Who should perform a firewall rule review?

Someone other than the person who made the changes. The firewall team can prepare the data, but the decision to keep a rule belongs to the rule’s business owner, and sign-off should sit with security or compliance. A review where admins certify their own rules is weak evidence of anything.

What is the difference between a firewall rule review and a firewall audit?

A rule review checks each rule against its business justification. A firewall audit is broader: it also covers firmware and patch level, admin accounts and access, logging, backups, and the change process itself. The rule review is one input to the audit.

What should you do with a rule that has no owner and no change record?

Do not delete it blind and do not rubber-stamp it. Turn on logging for the rule, watch what matches it for a fixed period, and use the source and destination to trace an owner. If nobody claims it, disable it through a normal change request with a date to delete it later. Disabling first gives you a fast way back if something breaks.

Where to start before your next review

  1. Save a baseline today. Export the running config from every in-scope firewall and store it off the device. That is your “last review” reference point, even if the last review was thin.
  2. Adopt the comment convention for every new rule from now on. Do not try to backfill the whole rulebase first. Stop the problem growing.
  3. Triage the existing rules by exposure. Management access and internet-facing rules first, then anything with “all” in a field, then everything else.
  4. Build the exceptions list. Every rule with no record gets logged, traced, and either justified or scheduled for removal. An honest exceptions list with dates on it is far better audit evidence than a clean-looking sign-off.
  5. Start reconciling changes as they happen. Every config change matched to a request within days, not at month five.

The bottom line

The six-month review is hard because of what happened in the six months before it, not because of the review itself. If every change was tied to a request when it was made, the review is a confirmation exercise. If not, it is archaeology, and archaeology under audit pressure ends in rubber stamps. Fix the record at the point of change and the review mostly takes care of itself.

That point-of-change record is what CyberX is built around. Changes go through a request and approval workflow, every configuration change on the firewall is detected and checked against those requests, and anything with no matching approval is flagged as unauthorized when it happens instead of at month five. The audit trail is stored off the device, so the diff since your last review is already there when you need it, across every tenant if you are an MSP. CyberX is available on a six-month free trial with no obligation, which happens to be exactly one PCI review cycle: long enough to run your next review with the change record already in place. See how firewall change management works in CyberX or look through the feature list.