Jakub Kicinski [off-list ref] writes:

On Fri, 29 Mar 2024 13:37:31 +0000 Donald Hunter wrote:

quoted

quoted

We'd only support multiple "do" requests, I wonder if we should somehow
call this out. Is --multi-do unnecessary extra typing?  

I prefer --multi but will update the help text to say "DO-OPERATIION"
and "... several do operations".

Alright, technically doing multi-dump should also work, but maybe
there's less of a benefit there, so we can keep the multi focused
on do for now.

Looking at the code again, are you sure we'll process all the responses
not just the first one?

Shouldn't this:

+                    del reqs_by_seq[nl_msg.nl_seq]
                     done = True

be something like:

		del reqs_by_seq[nl_msg.nl_seq]
		done = len(reqs_by_seq) == 0

Hmm yes, that's a good catch. I need to check the DONE semantics for
these nftables batch operations.

Would be good to add an example of multi executing some get operations.

I think this was a blind spot on my part because nftables doesn't
support batch for get operations:

https://elixir.bootlin.com/linux/latest/source/net/netfilter/nf_tables_api.c#L9092

I'll need to try using multi for gets without any batch messages and see how
everything behaves.

My other concern is the formatting of the response. For mutli we should
probably retain the indexes, e.g. 3 dos should produce an array with a
length of 3, some of the entries may be None if the command only acked.
Would that make sense?

As I said, a blind spot on my part - I didn't really think there was a
need to do anything for None responses but if get can work then an array
of responses will be needed.