Re: [PATCH net-next v2 2/6] net: Introduce read_sock_rectype proto_ops for control record delivery

Sabrina Dubroca <[email protected]> Fri, 31 Jul 2026 01:40:41 +0200
Newsgroups dev.linux.lists.kernel-tls-handshake,org.kernel.vger.linux-kselftest,org.kernel.vger.linux-nfs,org.kernel.vger.netdev
Message-ID <amvg-dZzDB6GzzSH@krikkit>
2026-07-30, 16:34:57 -0700, Jakub Kicinski wrote:
> On Fri, 31 Jul 2026 01:09:33 +0200 Sabrina Dubroca wrote:
> > To avoid that, you'd have to run recvmsg-nocmsg/read_sock to get all
> > the DATA out, then once you know the next chunk of payload is a
> > control record (and hopefully nothing else is reading from your
> > socket), one recvmsg+cmsg into a separate buffer.
> 
> I guess the difference in perspective stems from a fact that
> I'd look at a socket as a stream one reads and dispatches work
> based on the input. Not an abstraction layer that by itself
> dispatches different handlers for different message types.

I agree. But TCP is a magical world where users don't have to worry
about message types, and TLS isn't. It seems the users of the existing
(TCP-oriented) API are unhappy with the mismatch, but I'm not
convinced we can (and maybe "should") make that work.

-- 
Sabrina