Include sent headers with CONNECT request to upstream proxy

[Yopi]

Description

Includes the request headers from the client to the upstream proxy in the CONNECT request. Looking at #5802, and the relevant events, I don't believe that there is a way today to access the headers that are sent from the client and override the empty headers sent to the upstream proxy.

I don't know if this is the best way to tackle it, or if there is a reason to strip all the headers in the CONNECT request, but for my specific use case of the day, it would be nice to be able to handle the traffic differently based on the user agent of the caller.

Checklist

• I have updated tests where applicable.
• I have added an entry to the CHANGELOG.

[mhils]

Thanks for the PR! I see how this can be useful, but forwarding headers by default can also lead to unexpected leaks. For example, you may not want to forward your proxy authentication headers to an upstream proxy, or you do not want to forward the true user agent.

I'd be happy to have this as an off-by-default option (something like forward_connect_headers?). Ideally this is implemented in an isolated addon that uses a WeakKeyDictionary[Client, Headers] (see https://github.com/mitmproxy/mitmproxy/blob/10.0.0/mitmproxy/addons/proxyauth.py#L28-L30) to store the headers in http_connect and then set them in http_connect_upstream if the option is set. Let me know if that doesn't make sense!

I don't believe that there is a way today to access the headers that are sent from the client and override the empty headers sent to the upstream proxy.

The upstream_auth option is already doing that I believe, see https://github.com/mitmproxy/mitmproxy/blob/main/mitmproxy/addons/upstream_auth.py.

[Yopi]

Thank you for the feedback. That makes sense, that it might accidentally leak content.

I'll have a look if we can make it as an addon, but when I was playing with the different events it didn't seem like we could actually access the headers sent from the client to mitmproxy on the connect call. What upstream_auth is doing, if I'm not mistaken, is adding headers from mitmproxy to the upstream proxy on the request.

Since I had to make the changes that I did in this PR for the connection between mitmproxy to the upstream proxy for any headers to be available in the flow in this event:

    def http_connect_upstream(self, f: http.HTTPFlow):

[mhils]

I'll have a look if we can make it as an addon, but when I was playing with the different events it didn't seem like we could actually access the headers sent from the client to mitmproxy on the connect call.

Take a look at the proxyauth addons - that one accesses them. http_connect to access, http_connect_upstream to set, and a weak key dict in the meantime to remember them.

[Yopi]

Yes, you were completely right. Looks like it wasn't too difficult to achieve!

I'll add some tests as well, but I'm not sure what your general policy is on addons. Do you merge them if they are general enough, or are they better left to each user to have on their side?

Similar question to whether or not it should be a default addon or not?

[mhils]

An forward_connect_headers option (even if off by default) with a corresponding addon seems to be generic enough to be in mitmproxy by default. We don't have non-default addons. For things that cater very specific use cases or incur significant complexity, examples/contrib is a great place to share them. :)

[mhils]

Any update on this? :)