On Tue, May 04, 2021 at 04:24:05AM IST, Andrii Nakryiko wrote:

On Fri, Apr 30, 2021 at 11:32 PM Kumar Kartikeya Dwivedi
[off-list ref] wrote:

quoted

On Sat, May 01, 2021 at 01:05:40AM IST, Andrii Nakryiko wrote:

[...]

quoted

quoted

why didn't you put flags into bpf_tc_opts? they are clearly optional
and fit into "opts" paradigm...

I can move this into opts, but during previous discussion it was kept outside
opts by Daniel, so I kept that unchanged.

for bpf_tc_attach() I see no reason to keep flags separate. For
bpf_tc_hook_create()... for extensibility it would need it's own opts
for hook creation. But if flags is 99% the only thing we'll need, then
we can always add extra bpf_tc_hook_create_opts() later.

I'll put flags in the respective opts struct for both.

The hook creation path was kept generic enough so that this can be extended to
complex qdisc setup in the future than just clsact (even classful qdiscs should
be possible). So it is quite possible for bpf_tc_hook to take more parameters
than just flags by mapping different attach_point to different qdiscs.

Given some parameters are already optional depending on attach_point, it is
probably better to put flags in opts than dropping opts for now.

--
Kartikeya