netdev - Re: [PATCH 1/2] tls: remove close callback sock unlock/lock and flush

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190628175925.79a763f5@cakuba.netronome.com>
Date:   Fri, 28 Jun 2019 17:59:25 -0700
From:   Jakub Kicinski <jakub.kicinski@...ronome.com>
To:     John Fastabend <john.fastabend@...il.com>
Cc:     daniel@...earbox.net, ast@...nel.org, netdev@...r.kernel.org,
        edumazet@...gle.com, bpf@...r.kernel.org
Subject: Re: [PATCH 1/2] tls: remove close callback sock unlock/lock and
 flush_sync

On Fri, 28 Jun 2019 17:20:23 -0700, John Fastabend wrote:
> > Why can't tls sockets exist outside of established state?  If shutdown
> > doesn't call close, perhaps we can add a shutdown callback?  It doesn't
> > seem to be called from BH?
> >  
> 
> Because the ulp would be shared in this case,
> 
> 	/* The TLS ulp is currently supported only for TCP sockets
> 	 * in ESTABLISHED state.
> 	 * Supporting sockets in LISTEN state will require us
> 	 * to modify the accept implementation to clone rather then
> 	 * share the ulp context.
> 	 */
> 	if (sk->sk_state != TCP_ESTABLISHED)
> 		return -ENOTSUPP;
> 
> In general I was trying to avoid modifying core TCP layer to fix
> this corner case in tls.

I see, thanks for clarifying! I was wondering if there's anything wrong
in being in CLOSE/SYN/FIN states.

> > Sorry for all the questions, I'm not really able to fully wrap my head
> > around this. I also feel like I'm missing the sockmap piece that may
> > be why you prefer unhash over disconnect.  
> 
> Yep, if we try to support listening sockets we need a some more
> core infrastructure to push around ulp and user_data portions of
> sockets. Its not going to be nice for stable. Also at least in TLS
> and sockmap case its not really needed for any use case I know
> of.

IIUC we can't go from ESTABLISHED to LISTEN without calling close() 
or disconnect() so I'm not clear on why are we hooking into unhash() 😕