netdev - Re: [RFC PATCH net-next 2/2] net: tls: export protocol version and cipher to socket diag

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] [day] [month] [year] [list]

Message-ID: <20190617110706.7c269328@cakuba.netronome.com>
Date:   Mon, 17 Jun 2019 11:07:06 -0700
From:   Jakub Kicinski <jakub.kicinski@...ronome.com>
To:     Davide Caratti <dcaratti@...hat.com>
Cc:     "David S. Miller" <davem@...emloft.net>,
        Dave Watson <davejwatson@...com>,
        Boris Pismenny <borisp@...lanox.com>,
        Aviad Yehezkel <aviadye@...lanox.com>,
        John Fastabend <john.fastabend@...il.com>,
        Daniel Borkmann <daniel@...earbox.net>, netdev@...r.kernel.org
Subject: Re: [RFC PATCH net-next 2/2] net: tls: export protocol version and
 cipher to socket diag

On Mon, 17 Jun 2019 18:04:06 +0200, Davide Caratti wrote:
> On Wed, 2019-06-05 at 16:25 -0700, Jakub Kicinski wrote:
> > On Wed,  5 Jun 2019 17:39:23 +0200, Davide Caratti wrote:  
> > We need some indication of the directions in which kTLS is active
> > (none, rx, tx, rx/tx).
> > 
> > Also perhaps could you add TLS_SW vs TLS_HW etc. ? :)  
> 
> I can add a couple of u16 (or larger?) bitmasks to dump txconf and rxconf.
> do you think this is sufficient?

SGTM!

> > > +	int err = 0;  
> > 
> > There should be no need to init this.
> >   
> > > +	if (sk->sk_state != TCP_ESTABLISHED)  
> > 
> > Hmm.. why this check?  We never clean up the state once installed until
> > the socket dies completely (currently, pending John's unhash work).  
> 
> the goal was to ensure that we don't read ctx anymore after
> tls_sk_proto_close() has freed ctx, and I thought that a test on the value
> of sk_state was sufficient.
> 
> If it's not, then we might invent something else. For example, we might
> defer freeing kTLS ctx, so that it's called as the very last thing with
> tcp_cleanup_ulp().

Mm.. I was hoping the user space can no longer access a socket once
it reaches sk_prot->close :S  Perhaps I got this wrong.  If it can 
we need to make sure we don't free context before calling tcp_close()
otherwise the state may still be established, no?

In particular:

#ifdef CONFIG_TLS_DEVICE
	if (ctx->rx_conf == TLS_HW)
		tls_device_offload_cleanup_rx(sk);

	if (ctx->tx_conf != TLS_HW && ctx->rx_conf != TLS_HW) {
#else
	{
#endif
		tls_ctx_free(ctx);        <<<  <<<   <<<   <<< kfree()
		ctx = NULL;
	}

skip_tx_cleanup:
	release_sock(sk);
	sk_proto_close(sk, timeout);      <<<  <<<   <<<   <<< tcp_close()
	/* free ctx for TLS_HW_RECORD, used by tcp_set_state
	 * for sk->sk_prot->unhash [tls_hw_unhash]
	 */
	if (free_ctx)
		tls_ctx_free(ctx);