概要
Response splitting is the network protocol equivalent of finding a typewriter and discovering the carriage-return key works. HTTP headers are delimited by `\r\n`. Any place user input gets concatenated into a header without filtering those bytes, the attacker can end one header line and start another. Modern frameworks strip CR/LF before they ever reach the wire, so most apps are immune by accident — but the bug still appears in custom header-handling code, in proxy layers that forward user-controlled values, and in language features that lag behind framework norms (raw FastCGI bridges, mod_perl, certain Lua patterns, anywhere someone wrote `header('Location: ' + req.query.next)` without the framework's redirect helper).
仕組み
CRLF injection appears when user-controlled text reaches response headers without proper normalization. Attackers may be able to split headers, poison caches, or alter downstream browser behavior.
被害範囲
Cookie injection (session fixation, role escalation if the app trusts cookies for authorization). Cache poisoning across CDN edges, where one attacker request taints responses for every later visitor. XSS via injected Content-Type or HTML body. CRLF in `Location:` headers can also enable open-redirect-style phishing through legitimate-looking flows.
// what fixvibe checks
What FixVibe checks
FixVibe checks this class with verified-domain active testing that is bounded, non-destructive, and evidence-driven. Public reports describe the affected surface and remediation. For check-specific questions about exact detection heuristics, active payload details, or source-code rule patterns, contact support@fixvibe.app.
鉄壁の防御
Strip or reject CR (`\r`, `%0d`) and LF (`\n`, `%0a`) in any user input that reaches a response header. Most modern frameworks do this automatically: Express's `res.redirect` rejects line breaks, Django's `HttpResponseRedirect` URL-encodes them, Rails sanitizes header values. The bugs cluster in custom code — proxy handlers, signed-URL generators, OAuth callback constructors, anywhere someone reaches into `res.setHeader` directly. Audit those code paths and pass user input through a strict header-value sanitizer. As a defense-in-depth measure, ensure your CDN normalizes or rejects ambiguous responses (most modern CDNs do this by default; older nginx-as-cache configurations may not). Use HTTP/2 between proxy and origin where possible — HTTP/2 frames eliminate the CR/LF parsing ambiguity entirely.