Re: json_cmd_assoc and cmd

[Pablo Neira Ayuso <pablo@netfilter.org>]

On Thu, Jul 18, 2019 at 01:37:04PM +0100, Jeremy Sowden wrote:
> On 2019-07-16, at 21:39:03 +0200, Pablo Neira Ayuso wrote:
> > BTW, not directly related to this, but isn't this strange?
> >
> >         list_for_each_entry(cmd, cmds, list) {
> >                 memset(&ctx, 0, sizeof(ctx));
> >                 ctx.msgs = msgs;
> >                 ctx.seqnum = cmd->seqnum = mnl_seqnum_alloc(&seqnum);
> >                 ctx.batch = batch;
> >                 ctx.nft = nft;
> >                 init_list_head(&ctx.list);
> >                 ret = do_command(&ctx, cmd);
> >                 ...
> >
> > ctx is reset over and over again. Then, recycled here:
> >
> >                 ret = mnl_batch_talk(&ctx, &err_list, num_cmds);
> >
> > I wonder if we can get this better.
> 
> Something like this?

Yes, something like that would get things in better shape I think,
more comments below.

> 	struct netlink_ctx ctx = { .msgs = msgs, .nft = nft };
>         ...
> 
> 	ctx.batch = batch = mnl_batch_init();
> 	batch_seqnum = mnl_batch_begin(batch, mnl_seqnum_alloc(&seqnum));
> 	list_for_each_entry(cmd, cmds, list) {
> 		ctx.seqnum = cmd->seqnum = mnl_seqnum_alloc(&seqnum);
> 		init_list_head(&ctx.list);

I think we don't need to re-initialize this list over and over again
(from what I see when doing: git grep "ctx->list").

This always does list_splice_tail_init() to attach the object list
where they belong.

You can probably add something like:

        if (!list_empty(&ctx->list))
                BUG("command list is not empty\n");

I would make a patch and run tests/shell and tests/py to check if what
I'm suggesting this fine :-)

> 		ret = do_command(&ctx, cmd);
> 		...

Thanks!