On Mon, Jul 17, 2017 at 06:12:34PM +0200, Pablo Neira Ayuso wrote:
> On Thu, Jul 13, 2017 at 08:22:02PM +0200, Phil Sutter wrote:
> > Just a quick status update: It's a mess. ;)
>
> OK, let's address problems one by one.
>
> > There are so many different cases, I actually started drawing flow
> > diagrams (can't remember when I did that last time). In addition to what
> > we discussed already, I realized that via 'nft -f', I can make multiple
> > changes to even different sets within a single transaction - this
> > requires dealing with cached half-open ranges everywhere, not just in
> > NEWGEN callback.
>
> half-open ranges always start by a NFT_SET_ELEM_INTERVAL_END flag set
> on, eg.
No, they don't. See the end of segtree_linearize() src/segtree.c in
nftables code: EI_F_INTERVAL_END is set for intervals which don't match,
so if the matching interval extends to the end, no element with that
flag set will be inserted.
> # nft --debug=netlink add element x y { 5-65535 }
> element 00000000 : 1 [end] element 00000500 : 0 [end]
Here, the first element is the "null" element indicating a non-matching
segment from 0 to 4, the second one marks a matching segment from 5 till
the end. That '[end]' marker is printed unconditionally for all
elements.
> > Another trap is 'nft flush set': The elements are reported in
> > reverse order.
>
> Could you have a look at the function to order elements using the
> mergesort function? It's currently only called for non-intervals by
> now, so it would be good to converge to use it in all cases.
You mean the call to list_expr_sort() in netlink_get_setelems()? It is
not called because interval_map_decompose() (which is called later in
the same function does it's own sorting.
Cheers, Phil
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
