Hackney has CRLF / header injection via unvalidated `domain` and `path` options

Summary

CRLF injection in hackney_cookie:setcookie/3 (src/hackney_cookie.erl). The function validates Name and Value against CR/LF and control characters but concatenates the domain and path options verbatim into the output binary. If either option carries attacker-controlled data, a Host header forwarded as the cookie domain, a request URI forwarded as the cookie path, a \r\n in the value splits the Set-Cookie header and lets the attacker inject additional headers into the HTTP response.

Details

1. Asymmetric validation

Lines 27–34 of hackney_cookie.erl run binary:match on Name and Value, rejecting =, ,, ;, whitespace, \r, \n, \013, and \014. The Domain and Path options (lines 47 and 51) skip this check entirely and land straight in the result iolist:

[<<"; Domain=">>, Domain]
[<<"; Path=">>,   Path]

iolist_to_binary(...) on line 63 flattens everything and returns it to the caller.

2. Injection

A Path of <<"/x\r\nSet-Cookie: admin=1; Path=/">> produces a binary with a literal \r\n. Written into a Set-Cookie response header, the receiving HTTP parser splits it into two headers — one legitimate, one attacker-controlled.

3. Realistic trigger

Common patterns: keying the cookie domain off Host, deriving the path from the request URI, or copying a Location path into a cookie. Any of these lets a remote attacker control the injected content.

PoC

1. Call hackney_cookie:setcookie(<<"sid">>, <<"abc">>, [{path, <<"/x\r\nSet-Cookie: admin=1; Path=/">>}]).
2. The returned binary contains a literal \r\n followed by a second Set-Cookie: line.
3. Write the result into a Set-Cookie response header — the client parses two headers, including admin=1.

Impact

Cookie injection / HTTP response splitting at the hackney_cookie API boundary. Affects hackney 0.9.0 through 4.0.0 wherever domain or path options are populated from request data. Exploitation can overwrite session/auth cookies, fix cookies, or strip Secure/HttpOnly flags. CVSS v4.0: 2.1 (LOW) — requires attacker-controlled input to reach the domain or path option.

Resources

• Introduction commit: https://github.com/benoitc/hackney/commit/602d5c7f2ea4acbc83ed75230655d935a0750ebc
• Patch commit: https://github.com/benoitc/hackney/commit/8e02b99c28aea1b3fa2ddc0e66f51fe5bb0ac540