-remove NHLFE lookup from ipv4 slow path-lsp ping-RA label-PF_MPLS-move instructions to modules-move SPEC_NH to modules reference by name (not ETH ID)-multicast fwd-load balance forward with algos as modules-LER add hash under label stack-clean up code-make instructions be RTAs in netlink-add shift to NF, TC and DS instructions-add VRF instruction-add priority instruction-verify locking for nhlfe and ilm radix (RCU?)-clean up source file documentation-verify MTU change-expose flush options to netlink userland-port l2cc code to mpls-linux-add l2cc proto-test ppp support-add ATM / FR support-PHP support-examples mapping EXP to queuing discipline-per ILM or NHLFE locking?-external ILM and NHLFE references?-get MPLS to load unload as a module-SMP safe?-debug toggle via netlink* PRIO1 20040207  RCAS :Allocations in control plane use GPF_ATOMIC because we are working  under a spinlock. This is suboptimal. We should define a better locking  scheme (most MII/MOI modifications are performed in user context, after  a IOCTL).  We could define a better locking scheme and revert to GPF_KERNEL