TUN-2490: respect original representation of HTTP request path

This commit is contained in:
Nick Vollmar
2019-11-13 09:11:35 -06:00
parent 1d96bccc04
commit 6ea9b5c3ff
4 changed files with 531 additions and 92 deletions

View File

@@ -35,26 +35,49 @@ func createRequest(stream *h2mux.MuxedStream, url *url.URL) (*http.Request, erro
return req, nil
}
// H2RequestHeadersToH1Request converts the HTTP/2 headers to an HTTP/1 Request
// object. This includes conversion of the pseudo-headers into their closest
// HTTP/1 equivalents. See https://tools.ietf.org/html/rfc7540#section-8.1.2.3
func H2RequestHeadersToH1Request(h2 []h2mux.Header, h1 *http.Request) error {
for _, header := range h2 {
switch header.Name {
case ":method":
h1.Method = header.Value
case ":scheme":
// noop - use the preexisting scheme from h1.URL
case ":authority":
// Otherwise the host header will be based on the origin URL
h1.Host = header.Value
case ":path":
u, err := url.Parse(header.Value)
// We don't want to be an "opinionated" proxy, so ideally we would use :path as-is.
// However, this HTTP/1 Request object belongs to the Go standard library,
// whose URL package makes some opinionated decisions about the encoding of
// URL characters: see the docs of https://godoc.org/net/url#URL,
// in particular the EscapedPath method https://godoc.org/net/url#URL.EscapedPath,
// which is always used when computing url.URL.String(), whether we'd like it or not.
//
// Well, not *always*. We could circumvent this by using url.URL.Opaque. But
// that would present unusual difficulties when using an HTTP proxy: url.URL.Opaque
// is treated differently when HTTP_PROXY is set!
// See https://github.com/golang/go/issues/5684#issuecomment-66080888
//
// This means we are subject to the behavior of net/url's function `shouldEscape`
// (as invoked with mode=encodePath): https://github.com/golang/go/blob/go1.12.7/src/net/url/url.go#L101
if header.Value == "*" {
h1.URL.Path = "*"
continue
}
// Due to the behavior of validation.ValidateUrl, h1.URL may
// already have a partial value, with or without a trailing slash.
base := h1.URL.String()
base = strings.TrimRight(base, "/")
// But we know :path begins with '/', because we handled '*' above - see RFC7540
url, err := url.Parse(base + header.Value)
if err != nil {
return fmt.Errorf("unparseable path")
return errors.Wrap(err, fmt.Sprintf("invalid path '%v'", header.Value))
}
resolved := h1.URL.ResolveReference(u)
// prevent escaping base URL
if !strings.HasPrefix(resolved.String(), h1.URL.String()) {
return fmt.Errorf("invalid path %s", header.Value)
}
h1.URL = resolved
h1.URL = url
case "content-length":
contentLength, err := strconv.ParseInt(header.Value, 10, 64)
if err != nil {