On 02.03, Pablo Neira Ayuso wrote:
> On Mon, Mar 02, 2015 at 02:36:47PM +0000, Patrick McHardy wrote:
> > > > > This just makes sure that SYNPROXY6 is not called for non-tcp traffic
> > > > > in the rule loading path, which should be OK.
> > > >
> > > > Yeah, but for ip6_tables we actually need the check the way it is,
> > > > without IP6T_F_PROTO we will not perform the protocol match.
> > >
> > > if (!(e->ipv6.flags & IP6T_F_PROTO) ||
> > > e->ipv6.proto != IPPROTO_TCP ||
> > > e->ipv6.invflags & XT_INV_PROTO)
> > > return -EINVAL;
> > >
> > > e->ipv6.flags & IP6T_F_PROTO seems redundant to me, it just says
> > > e->ipv6.proto is set.
> >
> > No, it also instructs ip6_tables to actually match on that value.
> >
> > > If that flag is not set, then e->ipv6.proto is left unset. But the
> > > effect should be the same since 0 != IPPROTO_TCP.
> > >
> > > This is just relaxing the validation in SYNPROXY6 to only check
> > > e->ipv6.proto which is what nft_compat sets.
> > >
> > > Am I missing anything?
> >
> > Yes, ip6_tables only performs the protocol match when that flag is set.
> > Look:
> >
> > /* look for the desired protocol header */
> > if((ip6info->flags & IP6T_F_PROTO)) {
> > ...
> > if (ip6info->proto == protohdr) {
> >
> > This is where ip6_tables and ip_tables are different, ip_tables
> > takes ipinfo->proto as an indicator, ip6_tables the flag.
>
> I'm not altering the ip6_tables core logic.
>
> ip6tables from userspace sets IP6T_F_PROTO, the core handles things as
> you described, but SYNPROXY6 only checks if e->ipv6.proto != IPPROTO_TCP.
It will allow an ip6tables rule containing e->ipv6.proto == IPPROTO_TCP
but not IP6T_F_PROTO, meaning that the protocol match is not performed,
ergo any protocol can hit the target at runtime if such a rule is set.
I'm not saying that this will necessarily break something since
ip6tables does set that flag, but we usually do perform full validation
in ip*tables in the kernel, this patch changes that logic.
> REJECT6 does exactly the same that I need:
>
> if (e->ipv6.proto != IPPROTO_TCP ||
> (e->ipv6.invflags & XT_INV_PROTO)) {
> pr_info("TCP_RESET illegal for non-tcp\n");
> return -EINVAL;
> }
That contains the same problem. I'd say most people have copy and
pasted this from iptables without realizing the difference.
> I just want to relax the check in SYNPROXY6, so it works fine with
> nft_compat. I think nothing will break.
Not with our current userspace, I agree. But its a change in policy
for ip*tables to rely on that.
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
