feat: add UDP in HTTP

[imgk]

1. What does this change do, exactly?

Add UDP in HTTP support

2. Please link to the relevant issues.

klzgrad/naiveproxy#617

3. Which documentation changes (if any) need to be made because of this PR?

4. Checklist

• I have written tests and verified that they fail without my change
• I have squashed any insignificant commits
• This change has comments for package types, values, functions, and non-obvious lines of code

[klzgrad]

Is it necessary to introduce a library for this? Can you use simpler functions?

[imgk]

Removed

[klzgrad]

Seems to have duplicated forwardproxy.go.

I think much of this should be embedded along side the logic of tcp/http.

[imgk]

Done

[naiba]

any update?

[klzgrad]

Review comments are not addressed yet.

[naiba]

How is it now? This PR looks very interesting, does the naiveproxy client need to be modified?

[klzgrad]

I asked for not duplicating forwardproxy.go into a udp version, and there was no answer for that.

[imgk]

I asked for not duplicating forwardproxy.go into a udp version, and there was no answer for that.

Hi, I misunderstood your request there. I believe this is done now.

[naiba]

Hi @imgk, Does this feature require any changes to the client? Or does it already support the client that complies with the RFC definition?

[imgk]

Hi @imgk, Does this feature require any changes to the client? Or does it already support the client that complies with the RFC definition?

Hi, this modified forwardproxy should be compatible with any client which supports RFC 9298. However, this module is not well tested yet. The function is not guaranteed. Feedbacks are welcomed.

[aaravrav]

How is the status of the PR now?

[naiba]

Code Review (AI-assisted review by Claude)

Thanks for the implementation! Reviewed against RFC 9298/9297 and the existing codebase. A few issues found:

Blocking

1. HTTP/2 path uses Hijack() — breaks HTTP/2 multiplexing

case 2:
    conn, _, err := rc.Hijack()

Hijack() drops down to the raw TCP connection, bypassing HTTP/2 framing entirely. This destroys all other streams on the same connection. The existing TCP CONNECT handler uses dualStream() for HTTP/2 (reading from r.Body + writing to ResponseWriter). The UDP handler should follow the same pattern.

2. ReceiveBuffer — remote crash via malicious Length

data.Length, err = quicvarint.Read(rr)  // up to 2^62-1
bb := b[:data.Length]                    // panic if Length > len(b)

A malicious client sending a large Length value will panic the server. Needs bounds check:

if data.Length > uint64(len(b)) {
    return fmt.Errorf("datagram too large: %d > %d", data.Length, len(b))
}

3. HTTP/1.1 Upgrade header typo

w.Header().Set("Upgrade:", RequestProtocol)  // extra colon in header name

Should be "Upgrade", not "Upgrade:".

Should Fix

4. ACL logic duplicated — The anonymous function in tryUDPoverHTTP duplicates dialContextCheckACL. Should extract a shared helper to avoid divergence.

5. target_port range not validated — strconv.Atoi succeeds for 0, negative, or >65535 values. RFC 9298 requires 1-65535.

6. ReadPacket lacks bounds checking — Parsing bb[1], bb[2]...bb[19] without checking len(bb) first. Short payloads will panic.

7. Non-standard extensions not documented — CompressionAssignValue (0x1C0FE323) / CompressionCloseValue (0x1C0FE324) don't match current draft-ietf-masque-connect-udp-listen values (0x11/0x12/0x13), and COMPRESSION_ACK is missing. Connect-UDP-Bind is also draft-only. These should be clearly documented as experimental.

8. HandlePacketBind returns error at runtime — HTTP/3 + bind mode should be rejected at request parsing time with a proper HTTP status code, not deep in the handler.

Minor

• All types (Payload, Datagram, BytePayload, etc.) and vars (CapsuleProtocolHeaderValue, etc.) are exported but are internal implementation details — should be unexported.
• HandlePacket relies on b[0] being zero-initialized for Context ID 0 — should explicitly set b[0] = 0.
• RequestMatcher uses (.+) which matches across path segments — consider ([^/]+).