Skip to content

changeOrigin:true changes Host header but not Origin header #1669

@johnhunter

Description

@johnhunter

This issue has come up in Vite, which delegates proxying to http-proxy.

When Proxying to a WebSocket service in local development the socket connection can rejected by the server as the Host and Origin header values do not match. This Origin-limiting behaviour is recommended in RFC 6455 and is implemented in the Java websockets API and the Gorillas websocket package.

I looked though http-proxy source code to understand the changeOrigin option. It seems that the option originally did set both Host and Origin headers to match the target, e.g. in version v0.10.4:

https://github.com/http-party/node-http-proxy/blob/v0.10.4/lib/node-http-proxy/http-proxy.js#L675-L678

A subsequent refactoring for v1.0.0 did not include the setting as far as I can tell. A PR then reinstated the changeOrigin option where it conditionally changes the Host header but not the Origin header. #723

So I'm interested to know if dropping the change to the Origin header was for a specific reason, or if it just got missed in the v1.0.0 refactor?

The name "changeOrigin" can be confusing but reflects the terms used the HTTP/1.1 RFC-7230 (Jun 2014). It is notable that the RFC does not mention an Origin header. The Origin header is typically only included in cross-origin requests or same-origin requests other than GET / OPTIONS as defined in RFC-6454 (Dec 2011). I believe the confusion arises from the use of “origin” to mean the originator of a request, and the specific request header. However, it seems reasonable to expect that an option changeOrigin:true would change the Origin header, even if it also changes the Host value.

I can raise a PR to reinstate the Origin change but I'm not clear if there are good reasons not to do so. Given there are quite a few open PRs I'm not confident that project maintainers are in a position to review, merge and release any changes.

Happy to hear any thoughts for or against reinstating the behaviour.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions