Auditing Inherited Configurations Without Breaking Anything
Somewhere in the firewall you just inherited is a rule that permits something, from somewhere, to somewhere else. It has no comment. It has been there longer than anyone can remember. The uncomfortable part: it might be garbage from a long-dead project — or the only reason payroll works. From the outside, those two rules look identical.
Every inherited device is like this. A configuration is years of decisions compressed into a text file: projects, emergencies, vendor visits, and Tuesday-afternoon workarounds, stacked like layers of sediment. Nobody documented the decisions, the people who made them are gone, and some of those layers are load-bearing in ways no living person remembers. Your job is to figure out which — before you touch anything.
This guide gives you the method. The read-only audit starts with backups and walks every device class with the right questions. It sorts what you find into three honest categories and ends with a baseline you can defend. By the end you will know what you have, what scares you, and what to leave alone — having broken precisely nothing.
In this guide
- The audit ethos: look, don't touch
- Step zero: the config archive
- How to read each device class
- Interviewing the artifact
- The classification pass: GOOD, WEIRD, SCARY
- The change freeze — and its exceptions
- A worked mini-audit
- From archive to baseline
- Where to go next
The audit ethos: look, don't touch
New admins on inherited networks fail in one of two directions. Some never touch anything, and the network fossilizes around them. Others see an ugly config on day three, start "cleaning it up," and discover around lunchtime which ugly rule was holding up the warehouse scanners. The audit is the middle path: you earn the right to change things by understanding them first.
An old idea worth adopting as personal law is usually told as the story of a fence. A reformer finds a fence across a road and says, "I don't see the use of this; let us clear it away." The wise answer: if you don't see the use of it, you certainly may not remove it. Find out why it was built — then decide. Every uncommented firewall rule, odd static route, and strange VLAN is that fence. Someone built it to solve a problem. Until you know what problem, the fence stays.
So the audit phase has one rule: read everything, change nothing yet. Your output is not a cleaner network — not yet. It is paper: an archive, a findings ledger, a risk register, and a baseline. The cleanup comes later, and goes far faster and safer because you did this first.
Step zero: the config archive
Before you study a single rule, pull a copy of the configuration off every device you can reach. The iron rule of the exercise: backup before anything — before reading closely, before poking around web interfaces, absolutely before the first eventual change. Two reasons. First, aging hardware has a flair for drama. The device likeliest to die this month is the one nobody has logged into for years. Today's saved config is the difference between an afternoon of restoration and a week of reverse engineering. Second, you are human: a live session invites "just fixing" things, while a local copy lets you read all you like at zero risk.
What to pull, per device: the running configuration (what the device is doing right now). Where the platform has one, also pull the startup configuration (what it will do after a reboot). When they differ, that is your first finding: someone changed the device, never saved, and a power blip will silently undo their work. Also export anything the device offers as a file: rule sets, address books, VPN definitions, license details. Record firmware version and serial number while you are there.
Store it all in one boring, predictable structure — one folder per device, timestamped names, plain text wherever offered:
configs/
fw-main/
running-config_day01.txt
objects-export_day01.txt
sw-core/
running-config_day01.txt
sw-closet-2/
running-config_day01.txt
rtr-branch/
running-config_day01.txt
Then put the folder under version control from day one — Git is the usual choice, though disciplined dated copies beat nothing. It matters because this archive is about to become your baseline: the reference every future copy gets compared against.
Tip: pull configs from fragile, elderly gear during quiet-but-staffed hours — mid-morning, not midnight. Very old devices occasionally hiccup under an aggressive management session, and you want to be at your desk with coffee when one does. One at a time, gently.
Some devices will refuse you entry because nobody has the password. Do not fight that battle now — list them and move on. Recovering access safely is its own discipline with its own guide: credential archaeology.
How to read each device class
With the archive built, start reading — from your local copies, not live sessions. Each device class hides its secrets differently, so each gets its own questions. Work in risk order: firewalls first, where the security exposures live, then switches, routers, and the servers' own network settings.
Firewalls: rule by rule, no skipping
A firewall is read one rule at a time, top to bottom, and every rule gets the same three questions. What does it permit? Translate it into plain English: "anything on the internet may reach 192.0.2.10 on the remote-desktop port." Saying it aloud is often diagnosis enough. Is there a comment? A comment, even a cryptic one, is a message from the past; log it verbatim. Does traffic actually hit it? Most firewalls keep a hit counter per rule — a running tally of connections matched. Millions of hits means load-bearing whether you like it or not. Zero hits might mean dead — a clue, not a verdict.
Certain finds should raise the hair on your neck. Record them; do not fix them yet:
- Any-any rules — any source, any destination, any service. Inside-to-outside is sadly common; outside-to-inside is a five-alarm finding.
- Forgotten port forwards point at hosts nobody remembers exposing. A port forward is a NAT rule mapping a port on your public address to an inside machine. NAT, Network Address Translation, lets private addresses share a public one. Remote-desktop and camera-system forwards are the classics.
- Disabled logging — deny rules dropping traffic silently, or logging off globally. It costs nothing and hides everything; note the blind spots.
- Rules referencing ghosts — permits to addresses that answer nothing and appear in no inventory. Dead server, or a machine discovery missed?
Switches: does the topology match anyone's intent?
On switches, start with the VLAN inventory. A VLAN (Virtual LAN) is a way of slicing one physical switch into separate logical networks — and inherited switches accumulate them like attics accumulate boxes. List every VLAN on every switch and compare against what your scan found actually in use (your discovery sprint produced that list). VLANs defined but carrying nothing, VLANs in use but defined on only half the switches, VLAN 1 carrying everything — all findings. Next, read the trunk ports (switch-to-switch links carrying multiple VLANs): do they carry what they should, or every VLAN because someone gave up? Then find the spanning-tree root. Spanning tree is the protocol that prevents switch loops by electing one switch as the root of the tree. On a healthy network the root is your core switch on purpose. On inherited networks it is frequently a closet access switch that won the election by accident, quietly bending traffic through the wrong box. Finally, note whether port security (limits on what may plug into each port) exists at all. Its absence is normal — and still worth a line in the register.
Routers and gateways: routes to nowhere
Routing configs hide three species of mystery. Static routes to nowhere: a static route is a hand-entered instruction — "to reach network X, send traffic to address Y." Inherited routers collect routes whose next-hop no longer answers or whose destination appears nowhere else on earth. Each is a fossil of some old office, provider, or experiment; list them. NAT rules beyond the obvious — every translation is a promise the router is keeping to somebody; find out to whom. And the classic: the mystery tunnel. A VPN tunnel configured to a public address nobody recognizes, possibly still up, possibly exchanging traffic. Old vendor support access? A retired branch office? The predecessor's home lab? Write down the address, check the hit counters, and correlate with contracts before you dream of touching it. A live tunnel to an unknown party is a legitimate security concern — classify accordingly.
Servers: the network config nobody audits
Servers carry network configuration too, and it rots just as well. Check static addressing sanity: is each server's static address actually outside the DHCP pool? Or is it a collision waiting for the day DHCP hands the same address to a visitor's laptop? Check DNS settings: servers pointing first at a long-retired DNS server limp along on the second entry, adding a silent timeout to everything. That is a classic source of "this server has always been a bit slow." Check the host firewall: is it off entirely (finding), or does it contain a hand-made rule for one odd port? That port is an application talking; learn which before anyone "hardens" it away.
Interviewing the artifact
You cannot interview your predecessor, but you can interview what they left behind. Configs are full of testimony if you read like an archaeologist. Comments are the richest seam: "temp", "TEST", "do not remove", initials, ticket numbers — each one dates and explains a decision. A rule commented "TEMP for vendor migration" with two million hits is telling you a temporary fix became a permanent organ. Names talk too: an address object called LBL-FIX or a VLAN named CAMERAS-OLD is half a documentation page by itself. And hit counters are the traffic's own sworn statement about what is actually used — which frequently contradicts what everyone believes.
Then cross-examine. In your first week you interviewed the humans and collected the fear list — the things people warned you never to touch. (The first-week survival guide explains that ritual.) Now match fears to artifacts. The office manager said "whatever you do, don't break the label printers," and here is a rule named LBL-FIX permitting one weird port to one ancient host. You have just connected a fear to its fence. That pairing, human memory plus config evidence, is how undocumented networks slowly become documented ones.
The classification pass: GOOD, WEIRD, SCARY
Raw findings are anxiety in list form; the classification pass turns them into a plan. Every finding — every rule, route, VLAN, and oddity — gets exactly one of three labels:
| Label | Meaning | What you do with it |
|---|---|---|
| GOOD | You understand it, and it is sensible | Document it — this is how real documentation gets written, one verified fact at a time. |
| WEIRD | You do not understand it, but it is not obviously dangerous | Investigate before touching; it goes on the weird-but-working list with what you know so far. |
| SCARY | A security exposure or an outage waiting to happen | Into the risk register with a severity rating; the worst may earn a change-freeze exception. |
The label that needs defending is WEIRD. The weird-but-working principle says: a strange configuration that is demonstrably carrying traffic may be the fix for a problem you have not met yet. That deny rule for one specific device? Possibly the only thing stopping a chatty smart TV from flooding a segment — a problem you will rediscover about four minutes after deleting the rule. Weird earns investigation: comments, hit counters, long-tenured staff, old vendor tickets. Most weird findings eventually resolve in one of two ways. They become GOOD (there was a reason, and it still holds). Or they become removable junk (there was a reason, and it died with the old ERP system). But that resolution happens on your schedule, after the freeze — never as an impulse.
Expect the ratio to be lopsided: most findings turn out GOOD, a modest stack lands in WEIRD, and a small, memorable handful are SCARY. If everything looks scary, recalibrate — you may be grading normal small-business mess like a bank.
The change freeze — and its exceptions
While the audit runs, put yourself under a change freeze. This is a standing decision not to modify device configurations until the archive is complete, findings are classified, and the baseline exists. The engineering reason: you do not yet know what is load-bearing, so every change is a coin flip with someone else's workday. The political reason is just as real. The first outage on an inherited network gets blamed on the new admin no matter what caused it. An outage you personally caused in week two writes your reputation for a year. The freeze keeps your early record at "found problems," not "created them."
But a freeze with no exceptions is its own hazard. A remote-desktop port open to the whole internet does not care about your process. For genuine security exposures, use the emergency-change protocol — the freeze's one sanctioned door:
- Minimal fix. Close the hole and nothing else. Resist the adjacent cleanup begging to be done "while you're in there" — that urge has caused more outages than the holes ever did.
- Backup first. Fresh config pull immediately before the change, even though step zero gave you one.
- Rollback ready. The exact commands or clicks that undo the change, written down before you make it.
- Full notes. What you found, why it could not wait, what you changed, when, and how to reverse it — in the ledger, same day.
- Tell someone. Your manager hears about the hole and the fix in the same sentence. Found-plus-handled is a trust-building story; silent changes are not.
Warning: zero hits does not mean safe to delete. Counters on many platforms reset at reboot, so "zero" may only mean "zero since the last power cut." And some rules are supposed to sit at zero for months, like the disaster-recovery rule that matters one weekend a year. Zero-hit rules are candidates for investigation, never for casual deletion.
A worked mini-audit
Here is the method applied to a small office firewall — eight rules, lightly anonymized, exactly what you will inherit. The raw read from the archived config (public addresses from the documentation range):
# ACTION SOURCE DESTINATION SERVICE HITS COMMENT 1 allow outside 192.0.2.10:25 tcp/25 1482220 mail in 2 allow outside 192.168.1.40:3389 tcp/3389 9114 (none) 3 allow lan outside any 88914002 (none) 4 deny 192.168.40.31 outside any 310442 breakrm tv DO NOT REMOVE 5 allow outside 192.0.2.14:8443 tcp/8443 0 vendor portal TEMP 6 allow 192.168.1.0/24 192.168.20.0/24 any 771230 (none) 7 allow guest-vlan lan any 45113 printers for visitors 8 deny any any any 227901 (none, logging off)
Now the walk, each rule translated and labeled:
| # | Plain English | Label | Notes for the ledger |
|---|---|---|---|
| 1 | Internet may deliver mail to the mail server | GOOD | Heavy hits, matches a real server. Document and move on. |
| 2 | Anyone on earth may attempt remote desktop to one PC | SCARY | Uncommented, actively hit. Risk register, top severity — emergency-change candidate. |
| 3 | The LAN may reach the internet on any service | GOOD | Permissive but normal for a small office. Future tightening work, not a fire. |
| 4 | One device is blocked from the internet entirely | WEIRD | Huge hit count — blocking something constantly. The comment begs mercy. Learn what 192.168.40.31 is before ever touching. |
| 5 | Internet may reach a portal on 8443 — but nothing ever does | WEIRD | Zero hits, "TEMP" comment, target answers nothing. Probable fossil; confirm, then queue for post-freeze removal. |
| 6 | Office network may fully reach the 192.168.20.0 network | WEIRD | What lives there? Discovery says cameras. Why full access? Investigate. |
| 7 | Guest Wi-Fi may reach the entire internal LAN | SCARY | Comment says printers; rule says everything. Visitors can reach servers. Register; narrow it, expedited. |
| 8 | Everything else is dropped — silently | SCARY (low) | The deny is correct; the disabled logging blinds you. Enabling it is a fine first post-freeze change: low risk, high sight. |
Twenty minutes of reading turned a black box into a plan. It revealed two urgent risks (rules 2 and 7), a cheap visibility win (rule 8), and three mysteries with next steps. Two rules became documentation. Multiply by every device to see what the audit buys.
From archive to baseline
When the audit pass is done, your config archive quietly gets a promotion. It stops being "copies I took just in case" and becomes the baseline. It is the authoritative record of the reviewed state of every device, the reference all future change is measured against. That unlocks the most underrated habit in network administration: config drift monitoring.
The habit is simple. On a schedule — weekly is fine at first — re-pull each running config and compare it against the baseline copy. Any difference is one of three things. It may be a change you made (fine — commit it, and the baseline moves forward). It may be a change a colleague or vendor made (a conversation). Or it may be a change nobody admits to (a serious finding: misbehaving device, forgotten automation, or an intruder). Version control makes the comparison a one-command diff; the discipline costs minutes per week. Networks with this habit do not grow the next generation of mystery rules, because every change gets explained while its author still remembers why.
The freeze, meanwhile, does not end with a bang. It thaws. Once the baseline exists and the risk register is triaged, changes resume — through the groove the audit carved: change, note, re-pull, new baseline version. The fixes themselves, and the watching that should surround them, deserve a plan of their own. The guides under Monitoring Strategy cover how to build one.
Where to go next
The audit's four deliverables, one line each:
| Output | What it is | What it feeds |
|---|---|---|
| Config archive | Versioned copies of every reachable device's configuration | Disaster recovery today; the baseline tomorrow |
| Findings ledger | Every observation, labeled GOOD / WEIRD / SCARY, with evidence | Documentation (GOOD), investigations (WEIRD) |
| Risk register | The SCARY list, severity-ranked, with dates and owners | Your fix plan — and your credibility with management |
| Baseline | The reviewed archive, frozen as the reference state | Drift monitoring; every future diff |
You now know more about this network than anyone alive — on paper, with evidence. The next moves are turning that knowledge into visible progress and hunting the time bombs your predecessor left ticking; the rest of Taking Over an Undocumented Network walks both in order.
Put this into practice with SysPeep Monitor
A config audit is careful human work — no tool can read a firewall rule's intent for you. What software can do is watch the network while your head is buried in configs. SysPeep Monitor auto-discovers your devices and interfaces and collects health metrics over SNMP, WMI, and SSH. It alerts you when something starts misbehaving mid-audit, so surprises reach you from a dashboard, not from a hallway. The free trial installs on any Windows machine.
Download Free TrialLearn About SysPeep