I agree with (3). To achieve backward compatibility, we need to add an egress rule to allow ALL on existing ACLs which do not contain any egress rule.
For now almost every one uses integrated network offerings for VPC networks: DefaultIsolatedNetworkOfferingForVpcNetworks DefaultIsolatedNetworkOfferingForVpcNetworksNoLB DefaultIsolatedNetworkOfferingForVpcNetworksWithInternalLB The default egress policy are "Deny" in all network offerings above. -Wei 2017-11-13 18:47 GMT+01:00 Rene Moser <m...@renemoser.net>: > Hi Devs > > The last days I fought with the ACL egress rule behaviour and I would > like to make a poll in which direction the fix should go. > > Short Version: > > We need to define a better default behaviour for acl default egress > rule. I see 3 different options: > > 1. always add a default deny all egress rule. > > This would be super easy to do (should probably also the intermediate > fix for 4.9, see https://github.com/apache/cloudstack/pull/2323) > > > 2. add a deny all egress rule in case if have at least one egress allow > rule. > > A bit intransparent to the user, but doable. This seems to be the > behaviour how it was designed and should have been implemented. > > > 3. use the default setting in the network offering "egressdefaultpolicy" > to specify the default behavior. > > There is already a setting which specifies this behaviour but is not > used in VPC. Why not use it? > > As a consequence when using this setting, the user should get more infos > about the policy of the network offering while choosing it for the tier. > > > Poll: > > 1. [] > 2. [] > 3. [] > 4. [] Other? What? > > > Long Version: > > First, let's have a look of the issue: > > In version 4.5, creating a new acl with no egress (ACL_OUTBOUND) rule > would result in a "accept egress all": > > -A PREROUTING -s 10.10.0.0/24 ! -d 10.10.0.1/32 -i eth2 -m state --state > NEW -j ACL_OUTBOUND_eth2 > -A ACL_OUTBOUND_eth2 -j ACCEPT > > When an egress (here deny 25 egress) rule (no mather if deny or allow) > gets added the result is a "deny all" appended: > > -A PREROUTING -s 10.10.0.0/24 ! -d 10.10.0.1/32 -i eth2 -m state --state > NEW -j ACL_OUTBOUND_eth2 > -A ACL_OUTBOUND_eth2 -p tcp -m tcp --dport 25 -j DROP > -A ACL_OUTBOUND_eth2 -j DROP > > This does not make any sense and is a bug IMHO. > > > In 4.9 the behaviour is different: > > (note there is a bug in the ordering of egress rules which is fixed by > https://github.com/apache/cloudstack/pull/2313) > > The default policy is kept accept egress all. > > -A PREROUTING -s 10.11.1.0/24 ! -d 10.11.1.1/32 -i eth2 -m state --state > NEW -j ACL_OUTBOUND_eth2 > -A ACL_OUTBOUND_eth2 -d 224.0.0.18/32 -j ACCEPT > -A ACL_OUTBOUND_eth2 -d 225.0.0.50/32 -j ACCEPT > -A ACL_OUTBOUND_eth2 -p tcp -m tcp --dport 80 -j ACCEPT > > > To me it looks like the wanted behavior was "egress all as default. If > we have allow rules, append deny all". This would make sense but is > quite instransparent. > > But let's poll > > >
