Is it correct to reset the connection in websocket::stream_base::decorator callback?

Question

I have a websocket server where I call websocket::stream::async_accept to perform a handshake with the client.

I also set a decorator to modify the headers like this

conn->get_websocket_stream().set_option(boost::beast::websocket::stream_base::decorator(
[this, self, conn](boost::beast::websocket::response_type& response)
{
    // modify response
    
    if (some_condition)
    {
        conn->close(); // is it OK?
    }
});

conn->get_websocket_stream().async_accept(request, ...);

Inside the decorator, if a given condition is met, I need to abort the execution of async_accept by closing the connection to the client.

Is it allowed to close the connection to the client inside the decorator? I didn't find an answer to this question in the documentation https://www.boost.org/doc/libs/1_81_0/libs/beast/doc/html/beast/using_websocket/decorator.html

UPDATE conn is a connection object, a normal asio session based on shared_ptr.

get_websocket_stream() returns websocket::stream

conn->close() simply closes the socket (socket used by websocket::stream)

Please show the relevant code. We don't know what conn is or what conn->close() ends up doing with anything we can reason about. I suspect you're going to say get_websocket_stream() gives a beast::websocket::stream<>& and close() calls close() on that stream. That's would obviously NOT be valid for two reasons: 1. close() isn't valid until the ws handshake completes and 2. close() constitutes a write, and the handshake constitutes writes. You cannot overlap the writes. Instead cancel operations, the lower layer and/or raise an exception — sehe
– sehe, Commented Aug 8 at 12:50
Oh, and it's also not ok to mix async and sync operations on beast::websocket::stream<> — sehe
– sehe, Commented Aug 8 at 12:51
@sehe "I suspect you're going to say get_websocket_stream() gives a beast::websocket::stream<>&": yes, "and close() calls close() on that stream.": No, close just closes socket used by websocket::stream — Joe J
– Joe J, Commented Aug 8 at 12:57

sehe · Accepted Answer · 2025-08-08 14:12:07Z

5

Decorators lack a way to report error. Raising exceptions will break the flow, but will likely cause more side-effects than just invalidating the current session. Probably the best you can do is replacing the entire response with an appropriate kind of HTTP error code:

// decorate upgrade response
ws.set_option(boost::beast::websocket::stream_base::decorator(
    [](boost::beast::http::response<boost::beast::http::string_body>& res) {
        res.result(boost::beast::http::status::i_am_a_teapot);
        res.reason("The proxy doesn't like your request");
    }));

You may be tempted to just overwrite the entire request:

    res = {boost::beast::http::status::i_am_a_teapot, 11/*, "Some body text"*/};

Sadly, this is against the documented contract, which specifically warns of UB:

Undefined behavior results if the decorator modifies the fields specific to perform the WebSocket Upgrade, such as the Upgrade or Connection fields.

Side Notes

I'm a little confused. If conn is "a normal Asio session based on shared_ptr" why on earth is async_accept being triggered outside that "normal session" violating LoD in the process.

edited Aug 8 at 14:12

answered Aug 8 at 13:26

sehe

400k49 gold badges475 silver badges671 bronze badges

Sign up to request clarification or add additional context in comments.

9 Comments

Joe J Aug 8 at 13:40

Thank you so much! I understood your proposed solution, it looks interesting. However, I did not understand whether it is allowed to close the socket in the decorator immediately or not?

Joe J Aug 8 at 13:47

Also, if I touch websocket-specific headers (for example, remove this headers) I think UB is came according to documentation: "Undefined behavior results if the decorator modifies the fields specific to perform the WebSocket Upgrade, such as the Upgrade or Connection fields." boost.org/doc/libs/1_81_0/libs/beast/doc/html/beast/…

sehe Aug 8 at 14:09

Closing the underlying socket should probably be fine (assuming thread safety) but not the best idea. I'd probably cancel operations instead.

sehe Aug 8 at 14:10

Regarding the documented UB: that's a very good note. I happen to think the documentation is currently overly restrictive, but I could be wrong. Also, Beast clearly reserves the right to depend on these headers to stay valid and unchanged, so that's that. I'll amend the answer text.

Joe J Aug 8 at 14:11

I got you, thank you so much!

|

Collectives™ on Stack Overflow

Is it correct to reset the connection in websocket::stream_base::decorator callback?

1 Answer 1

Side Notes

9 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Side Notes

9 Comments

Your Answer

Sign up or log in

Post as a guest

Related