On 12/27/12 06:28, Graham Murray wrote: > Michael Orlitzky <mich...@orlitzky.com> writes: > >> The 'conntrack' module is supposed to be a superset of 'state', so most >> things should be compatible. You really have two warnings there; the >> first is for the state -> conntrack switch, and the second is because >> you're missing the --state flag in your rules. >> >> In your example, you turn on the state matching, >> >> iptables -A TCP_IN -p tcp -m state -m tcp -j UNSOLICITED >> >> but you don't specify *which* state(s) you want to match. It wants you >> to specify --state SOMETHING. I'd guess that it used to interpret "no >> state" as "any state." > > The problem is not really the OP's fault. The problem is that if you > have tables with the form "-m state --state XXX" at the point you > upgrade, iptables-save (quite possibly called automatically by > /etc/init.d/iptables stop) will save it as "-m state --state" - ie > 'forgetting' which state(s) the rule applies to. >
Youch, thanks, I'll keep an eye out for this when iptables wants a bump. I already keep the rules in a script, but it sounds like this will clobber the running rules after e.g. a reboot. My first -m state rule is, iptables -A INPUT -p ALL -m state \ --state ESTABLISHED,RELATED -j ACCEPT And if what you say is true, I'd be in deep shit if it reset to, iptables -A INPUT -p ALL -m state -j ACCEPT without a warning. > > I think that this is a serious enough change in behaviour that an elog > warning should have been issued. It's not stable yet, right? File a bug (and CC me, please).