On Thu, Dec 13, 2012 at 5:39 AM, Jarno Rajahalme <jarno.rajaha...@nsn.com> wrote: > > On Dec 12, 2012, at 18:23 , ext Jesse Gross wrote: >> In any case, flags should be converted between userspace and kernel. >> The same set of flags happen to be defined now with the same values >> but it shouldn't be assumed that is the case. In particular, if there >> are different versions of userspace and kernel then new flags that the >> other side doesn't know about shouldn't just get carried over. This >> can result in not being able to setup flows properly. > > Is there a mechanism by which the userspace can detect which flags > are supported by the kernel module at runtime?
There isn't but all we need here is a mechanism to check whether userspace understands them. Here's the situation that I'm worried about: * Flow setup comes from the kernel with new flags. * Userspace doesn't understand them and doesn't store them properly. This could happen if the kernel has a flag in the high 16-bits, for example. * When userspace sets up a flow it thinks that it can replicate the original kernel flow from the information it has. * The kernel flow doesn't match because the flags aren't the same. So the only piece that we really need is a mechanism to detect whether we received flags the we don't understand and note that fact (i.e. return ODP_FIT_TOO_MUCH in odp_flow_key_to_flow()). The other part, which isn't strictly required, is that it seems nice to mask out any flags that we don't understand. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
