Re: [PATCH nft,v4 7/7] intervals: support to partial deletion with automerge

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

On Wed, Apr 13, 2022 at 04:38:02PM +0200, Phil Sutter wrote:
> On Wed, Apr 13, 2022 at 04:27:07PM +0200, Pablo Neira Ayuso wrote:
> [...]
> > > The one with EXPR_F_REMOVE comes *after* the one to be removed, right?
> > 
> > Right, the other way around actually.
> > 
> > > My question again: Is it possible for 'prev' to have EXPR_F_REMOVE set?
> > > Maybe I miss something, but to me it looks like not although the code
> > > expects it.
> > 
> > prev never has EXPR_F_REMOVE, so it points to an existing element.
> 
> So below change should be fine?

Wait.

> diff --git a/src/intervals.c b/src/intervals.c
> index 451bc4dd4dd45..c0077c06880ff 100644
> --- a/src/intervals.c
> +++ b/src/intervals.c
[...]
> @@ -360,18 +358,15 @@ static int setelem_adjust(struct set *set, struct expr *add, struct expr *purge,
>  {
>  	if (mpz_cmp(prev_range->low, range->low) == 0 &&
>  	    mpz_cmp(prev_range->high, range->high) > 0) {
> -		if (!(prev->flags & EXPR_F_REMOVE) &&
> -		    i->flags & EXPR_F_REMOVE)
> +		if (i->flags & EXPR_F_REMOVE)

This chunk is not correct.

User might ask to delete an element which does not exist.

Then, you might find two consecutive EXPR_F_REMOVE.

Only the initial chunk in this patch is fine.