From: Theo Buehler <tb@theobuehler.org>
Subject: Re: bgpd: fix for aspa_table_equal
To: tech@openbsd.org
Date: Mon, 14 Sep 2026 10:10:07 +0200

On Mon, Sep 14, 2026 at 09:58:38AM +0200, Claudio Jeker wrote:
> In some edge cases the aspa_table_equal function may return true even
> though the tables are not the same.
> 
> The trivial case is:
> 	aspa-set {
> 		customer-as 1 provider-as { 4 5 6 }
> 		customer-as 2 provider-as { 10 11 }
> 	}
> vs.
> 	aspa-set {
> 		customer-as 1 provider-as { 5 6 }
> 		customer-as 2 provider-as { 10 11 4 }
> 	}
> 
> Right now the code checks the customer-as list comparing just the AS and
> then looking at the flat array of all provider-as numbers. Since in the
> above case both are the same (AS 4 moved sets but remains in the same
> location in the flat array) the function reports equality even though
> the tables are not equal.
> 
> Checking the number of provider-as per customer-as is enough to fix this.

I think your description is a bit too short.

ok tb

> 
> OK?
> -- 
> :wq Claudio
> 
> Index: rde_aspa.c
> ===================================================================
> RCS file: /cvs/src/usr.sbin/bgpd/rde_aspa.c,v
> diff -u -p -r1.8 rde_aspa.c
> --- rde_aspa.c	30 Aug 2026 23:43:22 -0000	1.8
> +++ rde_aspa.c	14 Sep 2026 07:29:02 -0000
> @@ -459,9 +459,12 @@ aspa_table_equal(const struct rde_aspa *
>  	if (ra->maxset != rb->maxset ||
>  	    ra->maxdata != rb->maxdata)
>  		return 0;
> -	for (i = 0; i < ra->maxset; i++)
> +	for (i = 0; i < ra->maxset; i++) {
>  		if (ra->sets[i].as != rb->sets[i].as)
>  			return 0;
> +		if (ra->sets[i].num != rb->sets[i].num)
> +			return 0;
> +	}
>  	if (memcmp(ra->data, rb->data, ra->maxdata * sizeof(ra->data[0])) != 0)
>  		return 0;
>  
>