What It Actually Means
The mental model most people carry is a wall around the building, and it is a reasonable one as far as it goes. The limitation is that a wall with a door the public must be able to walk through is a wall with a door, and for anything on the internet that door has to be open.
What It Genuinely Stops
Plenty, and it is worth being fair about this.
It stops the constant automated background noise of the internet: scanners probing every address for exposed databases, remote desktop, administrative interfaces and old services. That traffic is relentless, and a firewall refusing it is doing real work continuously.
It contains mistakes. A database accidentally listening on a public interface is a serious exposure, and a firewall that refuses external connections to it turns the mistake into a non-event. A great many of the largest data exposures on record are exactly this: a system that should never have been reachable, reachable.
It limits lateral movement. If someone gets into one machine, internal firewall rules decide how much else they can reach from there.
What Walks Straight Past It
Your website is on port 443 and it must be open to everybody. That means every attack delivered through the website arrives through a door the firewall is required to hold open.
SQL injection arrives as a normal web request. So does cross-site scripting, and so does someone logging in with a password taken from an unrelated breach. To a network firewall all of that is ordinary, correctly formed, permitted traffic, because it is.
Nor does it help with phishing, a malicious attachment, a compromised supplier, or an employee with more access than they need. Those are the routes that account for most real incidents at this size.
The Two Different Products
The word covers two things that are frequently confused when quoted.
A network firewall filters by address, port and protocol. It does not understand what a request means. This is what your hosting and cloud provider give you, usually free, and it is the one that stops the background noise.
A web application firewall inspects the content of web requests and blocks patterns that look like attacks. It works at the layer where real attacks now arrive, and it is a useful additional layer. It is also pattern matching: it blocks known shapes of attack, produces false positives that block real customers, and is not a substitute for the underlying flaw being absent.
A WAF buys time to patch. It does not remove the need to patch, and any supplier presenting it as a reason not to is selling comfort.
What To Ask
- What is reachable from the internet, and does it need to be? The most valuable question here. Administrative interfaces, database ports and staging environments frequently answer badly.
- Is the staging site public? Often it is, with a copy of production data on it and weaker protection than the live site.
- Do we have a WAF, and what has it blocked? If nobody has looked, nobody would notice it failing.
- What can reach what internally? Flat internal networks mean one compromised machine reaches everything.
More terms are in the glossary.