Explore our new Cybersecurity Terms Glossary, explained by our experts.

View the Glossary View the Glossary
PRIVIA

Cybersecurity Terms Glossary

What Is a Slowloris Attack?

Slowloris is a low-bandwidth denial-of-service technique that exhausts a web server by opening many connections and sending HTTP request headers as slowly as the server will tolerate.

Illustration for the cybersecurity term Slowloris Attack
Home / Blog / Cybersecurity Terms / What Is a Slowloris Attack?
17 November 2026

What Is a Slowloris Attack?

What Is a Slowloris Attack?

Slowloris is a denial-of-service technique that starves a web server of connection slots rather than of bandwidth. The attacker opens hundreds or thousands of TCP connections, begins an HTTP request on each one, and then sends the remaining header bytes at a trickle, just often enough to keep the server from timing the connection out. The request is never completed, so the worker handling it never becomes free.

Robert Hansen released the original Perl tool in June 2009, and the technique was used soon afterwards against Iranian government websites during the post-election protests. It remains relevant because a single laptop on a domestic connection can take down an unprotected server, which makes it attractive to attackers who have no botnet at all.

The Mechanics of a Half-Finished Request

An HTTP request header block ends with a blank line. Until the server sees that blank line, it has to assume more headers are coming and keep the connection open. Slowloris exploits that contract:

  1. Open a TCP connection and send GET / HTTP/1.1 followed by a Host header, but no terminating blank line.
  2. Every few seconds, send one more innocuous header, for example X-a: b, to reset the server’s idle timer.
  3. Repeat across as many parallel connections as the attacker can hold open.
  4. Replace any connection the server closes, keeping the pool permanently full.

On a thread-per-connection or process-per-connection server, each stalled request occupies a worker. Once every worker is held, legitimate visitors queue and then time out. Traffic volume is tiny, often a few kilobits per second, which is exactly why volumetric defences and traffic graphs tend to miss it.

Which Servers Are Exposed

Architecture decides the outcome. Apache HTTP Server with the prefork or worker MPM was the classic victim, as were dhttpd, older Tomcat builds, and a long tail of embedded web interfaces on routers, printers, and cameras. Event-driven servers such as nginx and Microsoft IIS allocate a small amount of state per connection instead of a whole worker, so they absorb the same attack with far less damage, though they are not immune if connection limits are set carelessly.

Variants reuse the same idea in different places. R-U-Dead-Yet (RUDY) sends a POST body one byte at a time rather than stalling the headers. Slow Read inverts it by advertising a tiny TCP receive window so the server cannot finish writing the response. All three are classed as low and slow attacks, a subset of application-layer denial of service.

Spotting It in the Logs

Slowloris leaves a distinctive fingerprint once you know where to look:

Because the source addresses are real, not spoofed, the attack cannot be launched blind; the attacker must be able to complete a TCP handshake. That also makes blocking by source address genuinely effective, unlike with a SYN flood.

Mitigation That Works

Several layers help, and most organisations need more than one:

The NCSC’s denial-of-service guidance makes the same point in broader terms: know your capacity limits before an incident, and rehearse how you would shed load. For an internet-facing service covered by the NIS Regulations, that rehearsal is part of demonstrating availability measures to your competent authority.

Our DoS and DDoS Testing Services include controlled low-and-slow testing so you learn your real connection ceiling in a maintenance window rather than during an attack.

FAQ – Frequently Asked Questions

Can One Computer Really Take Down a Web Server With Slowloris?

Against an unprotected thread-per-connection server, yes. The limiting factor is the number of concurrent connections the server will hold, which is often a few hundred, and a single host can easily open that many sockets. Bandwidth is irrelevant, so a domestic connection is enough. Servers behind a reverse proxy, CDN, or properly tuned timeout settings are far more resilient.

Is nginx Vulnerable to Slowloris?

Not in the same way. nginx handles connections with an event loop rather than dedicating a worker process or thread to each one, so thousands of stalled connections consume memory but do not block new requests. It still has limits, so set client_header_timeout, client_body_timeout, and worker_connections deliberately rather than leaving the defaults in place on a busy site.

How Is Slowloris Different From a Volumetric DDoS Attack?

A volumetric attack tries to saturate your bandwidth or packet-processing capacity with overwhelming traffic from many sources. Slowloris does the opposite: it uses very little traffic from very few sources and targets a software limit, the pool of available connection handlers. Detection tools tuned to traffic spikes often miss it entirely, which is part of its appeal to attackers.

Does Rate Limiting Stop Slowloris?

Request-rate limiting on its own does not, because the attacker sends very few requests. What works is limiting concurrent connections per source address and enforcing a minimum data rate for incoming headers. Combining connection limits with a header timeout such as Apache mod_reqtimeout addresses both the volume of sockets and the speed at which they are fed.

Related Terms

You May Be Interested In These

What Is a Timestamping Authority?

Cybersecurity Terms

17 November 2026

What Is a Timestamping Authority?

Read More
What Is the TCP/IP Model?

Cybersecurity Terms

17 November 2026

What Is the TCP/IP Model?

Read More
What Are HTTP Security Headers?

Cybersecurity Terms

17 November 2026

What Are HTTP Security Headers?

Read More