Re: [PATCH net-next 1/1] net/smc: add SMC rendezvous protocol

[Florian Westphal <fw@strlen.de>]

Ursula Braun <ubraun@linux.vnet.ibm.com> wrote:
> On 10/11/2017 11:06 PM, David Miller wrote:
> > From: Ursula Braun <ubraun@linux.vnet.ibm.com>
> > Date: Tue, 10 Oct 2017 16:14:19 +0200
> > 
> >> The goal of this patch is to leave common TCP code unmodified. Thus,
> >> it uses netfilter hooks to intercept TCP SYN and SYN/ACK
> >> packets. For outgoing packets originating from SMC sockets, the
> >> experimental option is added. For inbound packets destined for SMC
> >> sockets, the experimental option is checked.
> > 
> > I think this really isn't going to pass.
> > 
> > It's a user experience nightmare when the kernel inserts and
> > deletes filtering rules outside of what the user configures
> > on their system.

It depends if the hook is passive or not (i.e. mangles
payload/metadata or returns verdict other than NF_ACCEPT).

OUTPUT hook added here is not passive as it mangles tcp options.

> > This approach was also considerd for ipv6 ILA, and the same
> > pushback was given.

ahem.
net/ipv6/ila/ila_xlat.c:   err = nf_register_net_hooks(net, ila_nf_hook_ops,

FWIW at least the input hook seems ok to me provided it would use
skb_header_pointer for tcp header access (there is no guarantee
tcp_hdr() works or that the tcp header has been sanity checked in any
way).

Perhaps its time to consider moving net/netfilter/core.c into net/core
and rename NF_HOOK to NET_HOOK?