Re: [PATCH 1/2] src: store expr, not dtype to track data in sets

[Florian Westphal <fw@strlen.de>]

Pablo Neira Ayuso <pablo@netfilter.org> wrote:
> > +static struct expr *concat_expr_alloc_by_type_FIXME(uint32_t type)
> > +{
> > +	struct expr *concat_expr = concat_expr_alloc(&netlink_location);
> > +	unsigned int n;
> > +	int size = 0;
> > +
> > +	n = div_round_up(fls(type), TYPE_BITS);
> > +	while (n > 0 && concat_subtype_id(type, --n)) {
> > +		const struct datatype *i;
> > +		struct expr *expr;
> > +
> > +		i = concat_subtype_lookup(type, n);
> > +		if (i == NULL)
> > +			return NULL;
> > +
> > +		if (i->size == 0)
> > +			size = -1;
> > +		else if (size >= 0)
> > +			size += i->size;
> > +
> > +		expr = constant_expr_alloc(&netlink_location, i, i->byteorder,
> > +					   i->size, NULL);
> > +
> > +		compound_expr_add(concat_expr, expr);
> > +	}
> > +
> > +	/* can be incorrect in case of i->size being 0 (variable length). */
> > +	concat_expr->len = size > 0 ? size : 0;
> > +
> > +	return concat_expr;
> > +}
> > +
> > +static struct expr *
> > +data_expr_alloc_by_type_FIXME(enum nft_data_types type, enum byteorder keybyteorder)
> 
> There is no support for concatenations from the right hand side of the
> mapping, so I would just calloc a constant expression itself.

Excellent.  This is what I concluded when I was working on this, but at
that point i was already backed into a corner, hence the function name
:-)

> will be more simple. Same comment applies to dtype_map_from_kernel().

Oh, right.

> In general, I agree in the direction where this is going, that is,
> turn the datatype field in the set object into an expression.

Perfect.