HTTP/1.1 security is back in the spotlight after researchers reported new classes of HTTP desync attacks that exposed tens of millions of websites through flaws in several content delivery networks. Their argument is blunt: the protocol can’t be patched into safety, and sites should move upstream traffic to HTTP/2.
Vendors have spent six years deploying mitigations since desync attacks were first disclosed in 2019, the researchers say, and attackers have repeatedly found ways around them.
What a desync attack does
HTTP/1.1 leaves room for ambiguity about where one request ends and the next begins. An attacker can exploit that gap to confuse a reverse proxy and the origin server behind it. Both sides then disagree about how requests are split, and responses can land with the wrong user.
The consequences are severe. A single crafted request can make a site lose track of which response belongs to whom, which can log users into other people’s live accounts. Attackers can also poison a site’s cache with malicious JavaScript, which gives persistent control over every page and a way to steal passwords and card details. The researchers say they previously compromised PayPal’s login page this way.
Why patching hasn’t worked
The vulnerabilities in the new research have been fixed. The researchers say that doesn’t make HTTP/1.1 safe, because the design keeps producing new variants. Their view is that more flaws are on the way.
HTTPS offers no protection either. Wrapping HTTP/1.1 in TLS has no effect on this class of bug. Nor does it help to serve clients over HTTP/2 while keeping HTTP/1.1 behind the proxy, since the severe attacks happen on the upstream link.
That upstream link matters because reverse proxies route requests from many users down a shared pool of connections. A desync on a shared connection hits other people. Connections between a client and the edge server are rarely shared, so the risk there is low, and sites needn’t drop HTTP/1.1 for visitors whose browsers still need it.
The proposed fix: HTTP/2 upstream
HTTP/2 has very little room for doubt about message length, which makes desync attacks far rarer. Most flaws found in HTTP/2 implementations so far are denial-of-service bugs, which matter less on an upstream connection.
The advice is specific. Check that the origin server supports HTTP/2, then turn on upstream HTTP/2 in the front-end servers. Enabling HTTP/2 only at the edge doesn’t count.
What to do if you can’t switch yet
Some major vendors don’t support upstream HTTP/2 yet. The researchers name nginx, Akamai, CloudFront and Fastly. For teams stuck on HTTP/1.1, they recommend a short list of steps:
- Scan sites with HTTP Request Smuggler 3.0, an open-source tool, to find imminent threats, and repeat the scans regularly.
- Turn on every request validation and normalisation feature available on the front end.
- Consider disabling upstream connection reuse, which cuts the risk sharply but may slow performance.
- Ask your vendor when upstream HTTP/2 support is coming.
The team also released a second open-source tool, HTTP Hacker, and a free lab on 0.CL request smuggling in its Web Security Academy for people who want to practise the technique safely. A whitepaper titled “HTTP/1.1 Must Die: The Desync Endgame” lays out the full case, and the work was presented at DEF CON.
Why the risk is not theoretical
The researchers frame this as a pattern, not a one-off. Each round of fixes closes known tricks, and then a new way to hide the boundary between two requests turns up. They say that cycle continues for as long as HTTP/1.1 stays in use between proxies and origin servers.
Teams that want to see their own exposure can run the scanning tools above against staging systems first, then production. The sandbox lab is a quick way to learn how 0.CL smuggling behaves before testing live traffic.
What happens next
The push puts pressure on infrastructure vendors. Until nginx, Akamai, CloudFront and Fastly support upstream HTTP/2, many sites can’t follow the advice even if they want to. Expect customers to press those providers for timelines, and expect researchers to keep probing the mitigations that remain in place.
Source: http1mustdie.com