On Wed, 5 Oct 2022 at 22:53, Cong Wang <xiyou.wangcong@xxxxxxxxx> wrote: > > From: Cong Wang <cong.wang@xxxxxxxxxxxxx> > > Insert: > bpf_map_update(&map, &key, &val, flag); > > Delete a specific key-val pair: > bpf_map_delete_elem(&map, &key); > > Pop the minimum one: > bpf_map_pop(&map, &val); > > Lookup: > val = bpf_map_lookup_elem(&map, &key); > > Iterator: > bpf_for_each_map_elem(&map, callback, key, val); > > Signed-off-by: Cong Wang <cong.wang@xxxxxxxxxxxxx> > --- Instead of a dedicated BPF map and using kptr inside the map value, we should probably lift Dave's series [0] adding the rbtree, and allow linking sk_buff ctx directly into it. It would require recognising the rb_node in sk_buff (or __sk_buff shadow struct) as a valid bpf_rb_node similar to those in user allocated types. Overall it would be a much better approach IMO and avoid having different rbtree implementations. We would probably follow a similar approach for xdp_frame as well. It can also be a union of bpf_rb_node, bpf_list_node, etc. Since the type can only be in only one collection at once it would allow it to be linked into different types of structures without wasting any space. [0]: https://lore.kernel.org/bpf/20220830172759.4069786-1-davemarchevsky@xxxxxx
