Trillium is a composable toolkit for building internet applications with async rust. In trillium-http prior to 0.3.12 and trillium-client prior to 0.5.4, insufficient validation of outbound header values may lead to request splitting or response splitting attacks in scenarios where attackers have sufficient control over headers. This only affects use cases where attackers have control of request headers, and can insert rn sequences. Specifically, if untrusted and unvalidated input is inserted into header names or values.
Outbound trillium_http::HeaderValue and trillium_http::HeaderName can be constructed infallibly and were not checked for illegal bytes when sending requests from the client or responses from the server. Thus, if an attacker has sufficient control over header values (or names) in a request or response that they could inject rn sequences, they could get the client and server out of sync, and then pivot to gain control over other parts of requests or responses. (i.e. exfiltrating data from other requests, SSRF, etc.)
In trillium-http versions 0.3.12 and later, if a header name is invalid in server response headers, the specific header and any associated values are omitted from network transmission. Additionally, if a header value is invalid in server response headers, the individual header value is omitted from network transmission. Other headers values with the same header name will still be sent. In trillium-client versions 0.5.4 and later, if any header name or header value is invalid in the client request headers, awaiting the client Conn returns an Error::MalformedHeader prior to any network access. As a workaround, Trillium services and client applications should sanitize or validate untrusted input that is included in header values and header names. Carriage return, newline, and null characters are not allowed.
The product receives data from an HTTP agent/component (e.g., web server, proxy, browser, etc.), but it does not neutralize or incorrectly neutralizes CR and LF characters before the data is included in outgoing HTTP headers.
| Name | Vendor | Start Version | End Version |
|---|---|---|---|
| Trillium | Trillium | * | 0.5.4 (excluding) |
| Trillium-http | Trillium | * | 0.3.12 (excluding) |
HTTP agents or components may include a web server, load balancer, reverse proxy, web caching proxy, application firewall, web browser, etc. Regardless of the role, they are expected to maintain coherent, consistent HTTP communication state across all components. However, including unexpected data in an HTTP header allows an attacker to specify the entirety of the HTTP message that is rendered by the client HTTP agent (e.g., web browser) or back-end HTTP agent (e.g., web server), whether the message is part of a request or a response.
When an HTTP request contains unexpected CR and LF characters, the server may respond with an output stream that is interpreted as “splitting” the stream into two different HTTP messages instead of one. CR is carriage return, also given by %0d or \r, and LF is line feed, also given by %0a or \n. In addition to CR and LF characters, other valid/RFC compliant special characters and unique character encodings can be utilized, such as HT (horizontal tab, also given by %09 or \t) and SP (space, also given as + sign or %20). These types of unvalidated and unexpected data in HTTP message headers allow an attacker to control the second “split” message to mount attacks such as server-side request forgery, cross-site scripting, and cache poisoning attacks. HTTP response splitting weaknesses may be present when: