[kernel 5.14] remove workaround for skb->mark scrub during veth traversal / bpf_redirect() #36329
Description
Background:
The BPF datapath code currently contains a workaround for a kernel limitation. When traversing between a veth pair in the same netns, the skb->mark is scrubbed. So we need to use the skb->cb instead. This has been resolved in newer kernels (5.14+) and backported to older kernels, but we're still carrying the workaround.
Proposal:
Remove the workaround once we require a new-enough kernel. Until then we can implement a probe, but it's not a priority.