On Tue, May 05, 2026 at 01:11:42AM +0200, Stefano Brivio wrote:
> The new checks are actually sufficient but not enough for Coverity
> Scan. Now that fwd->sock_count and new->last are affected or supplied
> by clients, we need explicit (albeit redundant) checks on them.
> 
> Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

I'm assuming this does squash the warnings, but I think it does so in
a somewhat confusing way.

> ---
>  fwd_rule.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/fwd_rule.c b/fwd_rule.c
> index b55e4df..03e8e80 100644
> --- a/fwd_rule.c
> +++ b/fwd_rule.c
> @@ -271,13 +271,22 @@ int fwd_rule_add(struct fwd_table *fwd, const struct fwd_rule *new)
>  		warn("Too many rules (maximum %d)", ARRAY_SIZE(fwd->rules));
>  		return -ENOSPC;
>  	}
> +
>  	if ((fwd->sock_count + num) > ARRAY_SIZE(fwd->socks)) {
>  		warn("Rules require too many listening sockets (maximum %d)",
>  		     ARRAY_SIZE(fwd->socks));
>  		return -ENOSPC;
>  	}
> +	/* Redundant, to make static checkers happy */
> +	if (fwd->sock_count > ARRAY_SIZE(fwd->socks))
> +		return -ENOSPC;

So there's actually two conditions that this is kind of relevant to:

 1) (fwd->sock_count > ARRAY_SIZE(fwd->socks)) on entry

That means something is horribly wrong before we were even called.
So, I think that would be better as an assert().

  2) (fwd->sock_count + num) overflows

That's a closer-to-real concern.  I'm pretty sure we can't hit it for
real, because num is necessarily <= 65536, so as long as (1) is true
this can't overflow.  But that relies on the specific value of
ARRAY_SIZE(fwd->socks), so it's kind of fragile.

I think an explicit check for this is a good idea, but it should
actually check for this, not just side-effects of it, so:
	if (fwd->sock_count + num <= fwd->sock_count) {
		warn("Blah blah overflow");
		return -EFAULT; /* or whatever */
	}

>  	fwd->rulesocks[fwd->count] = &fwd->socks[fwd->sock_count];
> +
> +	/* Redundant ('num' checked above), but not for static checkers */
> +	if (new->last > ARRAY_SIZE(fwd->socks) + new->first)
> +		return -ENOSPC;

This way of organising the check is very confusing to me.  I'm not
really sure what it's trying to catch.  We've already checked that
last >= first, so using num is safer to deal with at this
point than ARRAY_SIZE() + first, which could in principle overflow
even if sock_count + num is perfectly ok.

>  	for (port = new->first; port <= new->last; port++)
>  		fwd->rulesocks[fwd->count][port - new->first] = -1;
>  
> -- 
> 2.43.0
> 

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson