On Mon, Jan 20, 2020 at 2:11 PM Daniel Borkmann daniel@iogearbox.net wrote:
On 1/20/20 9:10 PM, Matt Cover wrote:
On Mon, Jan 20, 2020 at 11:11 AM Matt Cover werekraken@gmail.com wrote:
On Sat, Jan 18, 2020 at 8:05 PM John Fastabend john.fastabend@gmail.com wrote:
Matthew Cover wrote:
Allow looking up an nf_conn. This allows eBPF programs to leverage nf_conntrack state for similar purposes to socket state use cases, as provided by the socket lookup helpers. This is particularly useful when nf_conntrack state is locally available, but socket state is not.
Signed-off-by: Matthew Cover matthew.cover@stackpath.com
Couple coding comments below. Also looks like a couple build errors so fix those up. I'm still thinking over this though.
Thank you for taking the time to look this over. I will be looking into the build issues.
Looks like I missed static inline on a couple functions when nf_conntrack isn't builtin. I'll include the fix in v2.
One of the big issues I'd see with this integration is that literally no-one will be able to use it unless they manually recompile their distro kernel with ct as builtin instead of module .. Have you considered writing a tcp/udp ct in plain bpf? Perhaps would make sense to have some sort of tools/lib/bpf/util/ with bpf prog library code that can be included.
I don't believe the builtin requirement is permanent. Currently, that requirement comes from an undefined reference to nf_conntrack_find_get() during linking. As a future improvement, I am planning to propose a function pointer which ct_lookup() uses. The kernel proper would point this to an always NULL stub. nf_conntrack would populate to the real function when builtin or when loaded as a module.
If there is a better way to solve the kernel proper using an exported symbol provided by a module, I'd be happy to hear of it.