is_token implementation is more permissive than standard

[declanvk]

RFC 9110 Section 5.6.2 defines the grammar for field value tokens as

  token          = 1*tchar

  tchar          = "!" / "#" / "$" / "%" / "&" / "'" / "*"
                 / "+" / "-" / "." / "^" / "_" / "`" / "|" / "~"
                 / DIGIT / ALPHA
                 ; any VCHAR, except delimiters

---

However in practice it seems that this specific grammar is relaxed to a larger set of allowed characters. I made a small test which checks this behavior:

#[test]
fn test_header_with_invalid_rfc91100_token_in_value() {
    const RESPONSE: &[u8] =
        b"HTTP/1.1 200 OK\r\nAccess-Control-Allow-Credentials: hell[o\r\nBread: baguette\r\n\r\n";

    let mut headers = [EMPTY_HEADER; 2];
    let mut response = Response::new(&mut headers[..]);

    let result = crate::ParserConfig::default().parse_response(&mut response, RESPONSE);
    let status = result.unwrap();

    assert_eq!(status, Status::Complete(78));
    assert_eq!(response.version.unwrap(), 1);
    assert_eq!(response.code.unwrap(), 200);
    assert_eq!(response.reason.unwrap(), "OK");
    assert_eq!(response.headers.len(), 2);
    assert_eq!(response.headers[0].name, "Access-Control-Allow-Credentials");
    assert_eq!(response.headers[0].value, &b"hell[o"[..]);
    assert_eq!(response.headers[1].name, "Bread");
    assert_eq!(response.headers[1].value, &b"baguette"[..]);
}

I also checked how Firefox handles this header with a small program

printf "HTTP/1.1 200 OK\r\nAccess-Control-Allow-Credentials: hell[o\r\nBread: baguette\r\n\r\n" | nc -l 4040

Then visited localhost:4040 in Firefox. The recorded execution also showed the delimiter in the token value:

---

Based on the test assertions and the firefox screenshot it seems consistent that both programs allow characters outside the grammar specified in RFC91100.

• Was this intentional at the time of writing the tokenization code?
• Would it possible to have a strict mode which would reject more field values for invalid tokens?

[seanmonstar]

It's true that the is_token method differs from the spec, which was motivated by real world traffic. However, the parts you're showing, header values, are not defined to be tokens even in the spec. Header values are defined as:

  field-value    = *field-content
  field-content  = field-vchar
                   [ 1*( SP / HTAB / field-vchar ) field-vchar ]
  field-vchar    = VCHAR / obs-text
  obs-text       = %x80-FF

Which allows [ and plenty more in header values.

[declanvk]

Thanks for clarifying! When I re-read, I think then only field-names are defined as tokens: https://www.rfc-editor.org/rfc/rfc9110#section-5.1-2? I see it used in a couple other places like methods, parameter names, connection options, etc.

However, looking in

httparse/src/lib.rs

Line 58 in 0f5e6fb

fn is_token(b: u8) -> bool {

, I only see the is_token used to parse tokens and something with parsing URIs.