Re: [bpf PATCH v2 2/6] bpf: tls fix transition through disconnect with close

From: John Fastabend <john.fastabend@gmail.com>
To: Jakub Kicinski <jakub.kicinski@netronome.com>,
	John Fastabend <john.fastabend@gmail.com>
Cc: ast@kernel.org, daniel@iogearbox.net, netdev@vger.kernel.org,
	edumazet@google.com, bpf@vger.kernel.org
Subject: Re: [bpf PATCH v2 2/6] bpf: tls fix transition through disconnect with close
Date: Thu, 11 Jul 2019 09:35:36 -0700	[thread overview]
Message-ID: <5d2765584f043_698f2aaeaaf925bcb0@john-XPS-13-9370.notmuch> (raw)
In-Reply-To: <20190710123417.2157a459@cakuba.netronome.com>

Jakub Kicinski wrote:
> On Tue, 09 Jul 2019 20:39:24 -0700, John Fastabend wrote:
> > Jakub Kicinski wrote:
> > > On Mon, 08 Jul 2019 19:14:05 +0000, John Fastabend wrote:  
> > > > @@ -287,6 +313,27 @@ static void tls_sk_proto_cleanup(struct sock *sk,
> > > >  #endif
> > > >  }
> > > >  
> > > > +static void tls_sk_proto_unhash(struct sock *sk)
> > > > +{
> > > > +	struct inet_connection_sock *icsk = inet_csk(sk);
> > > > +	long timeo = sock_sndtimeo(sk, 0);
> > > > +	struct tls_context *ctx;
> > > > +
> > > > +	if (unlikely(!icsk->icsk_ulp_data)) {  
> > > 
> > > Is this for when sockmap is stacked on top of TLS and TLS got removed
> > > without letting sockmap know?  
> > 
> > Right its a pattern I used on the sockmap side and put here. But
> > I dropped the patch to let sockmap stack on top of TLS because
> > it was more than a fix IMO. We could probably drop this check on
> > the other hand its harmless.
> 
> I feel like this code is pretty complex I struggle to follow all the
> paths, so perhaps it'd be better to drop stuff that's not necessary 
> to have a clearer picture.
> 

Sure I can drop it and add it later when its necessary.

> > > > +		if (sk->sk_prot->unhash)
> > > > +			sk->sk_prot->unhash(sk);
> > > > +	}
> > > > +
> > > > +	ctx = tls_get_ctx(sk);
> > > > +	if (ctx->tx_conf == TLS_SW || ctx->rx_conf == TLS_SW)
> > > > +		tls_sk_proto_cleanup(sk, ctx, timeo);
> > > > +	icsk->icsk_ulp_data = NULL;  
> > > 
> > > I think close only starts checking if ctx is NULL in patch 6.
> > > Looks like some chunks of ctx checking/clearing got spread to
> > > patch 1 and some to patch 6.  
> > 
> > Yeah, I thought the patches were easier to read this way but
> > maybe not. Could add something in the commit log.
> 
> Ack! Let me try to get a full grip of patches 2 and 6 and come back 
> to this.
> 
> > > > +	tls_ctx_free_wq(ctx);
> > > > +
> > > > +	if (ctx->unhash)
> > > > +		ctx->unhash(sk);
> > > > +}
> > > > +
> > > >  static void tls_sk_proto_close(struct sock *sk, long timeout)
> > > >  {
> > > >  	struct tls_context *ctx = tls_get_ctx(sk);  