Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote: > On Fri, Oct 01, 2021 at 03:21:28PM +0200, Florian Westphal wrote: > > Alternate fix idea: > > > > 1. store skb->skb_iif in nf_conn. > > > > This means locally vs. remote-generated nf_conn can be identified > > via ct->skb_iff != 0. > > > > 2. For "remote" case, force following behaviour: > > check that sport > dport and sport > 1024. > > > > OTOH, this isn't transparent to users and might cause issues > > with very very old "credential passing" applications that insist > > on using privileged port range (< 1024) :-/ > > Can't this be just expressed through ruleset? I mean, conditionally > masquerade depending on whether the packet is locally generated or > not, for remove for sport > 1024 range. Yes, see patch #1, it demos a couple of ruleset based fixes/mitigations for this problem.
