On Wed, Feb 20, 2019 at 08:14:01PM +0530, Numan Siddique wrote: > On Tue, Feb 12, 2019 at 7:57 AM Ben Pfaff <b...@ovn.org> wrote: > > There is a special issue with this action, which is a lot like a special > > issue with the "sample". That is that the action has flow control that > > might change the flow in different ways, and this can make an important > > difference for actions that follow this one. If one fork of the action > > pops off a header, and the other one does not, then that makes > > validating actions that come after it complicated. I do not see > > anything here that takes that into account. I recommend reading the > > validation code for the sample action for inspiration. > > > > > From what I see from the sample code, it checks if the actions inside the > sample > modify the flow key or not and set the 'sample_arg->exec' accordingly if the > 'sample' action is not the last action. > > In the ovs-vswitchd patch (patch 2 of this series) when it translates the > 'check_pkt_len' > action, 'check_pkt_len' will be the last action of the flow. > > In the datapath implementation I am thinking to mandate 'check_pkt_len' to > be last > action. If any actions follow 'check_pkt_len', then entire flow should be > rejected. > Any thoughts, comments on this. I really don't see a scenario where another > action > would follow 'check_pkt_len'.
That would also solve the problem, and it would not reduce the generality, since any actions that would otherwise have followed the check_pkt_len action could be duplicated into both of the branches. It might have difficult implications for flow translation in userspace, but I won't be sure until I see the corresponding userspace patches. _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev