Every FFS site has a guard in front of it. It runs in two places: in the front-end, before a request is passed to a site at all, and inside each site's worker, where it sees the request as the application will. Nothing it needs is a separate product — no web application firewall to buy, no reverse proxy to configure.
What the guard looks at
- Who is asking. A whitelist for addresses you trust, a permanent blacklist you keep by hand, and an automatic blacklist the guard keeps itself. Repeat offenders are banned for longer each time; a ban expires on its own.
- What is being asked. Known attack patterns — probes for admin panels, configuration files, injection strings — and paths that should never be requested. Methods a site does not use are refused.
- How it is being asked. User agents that belong to known attack tools, and request rates no person produces.
- Where the requests come from. When several addresses of one network range misbehave, the guard can ban the range rather than playing whack-a-mole with single addresses.
Minor offences are counted first and expire; only repeated ones lead to a ban. A whitelisted address is never counted.
Tripwires
A guard that only reacts to known attacks is always one step behind. FFS also sets traps:
- Honeypot paths — addresses no legitimate visitor or crawler has any reason to request. Touching one is treated as proof of hostile intent.
- Bait links — the same addresses can be offered to scrapers that follow every link they find, invisible to people.
Designed not to get in the way
- Applications that serve files can exempt their own paths from the attack-pattern check — in a code repository
.envorconfig.jsonis a file name, not a probe. Every other check still applies. - The rules reload while the server runs: a new pattern or an extra whitelist entry takes effect without a restart.
- Everything the guard does is logged, and command-line tools show what it blocked and why, and lift a ban that was a mistake.
- Sites can share their pattern lists, honeypots and permanent bans, so a lesson learned on one site protects all of them.
And then the bots
Modern hostile automation does not announce itself with an attack string. It looks like a browser, from many addresses at once. For that, FFS has a second line in the front-end: the bot filter.
