Re: [PATCH v5 net-next 08/10] geneve: extract hint option at GRO stage

[Paolo Abeni <pabeni@redhat.com>]

On 1/21/26 5:11 PM, Paolo Abeni wrote:
> @@ -566,6 +726,25 @@ static struct sk_buff *geneve_gro_receive(struct sock *sk,
>  			goto out;
>  	}
>  
> +	/* The GRO hint/nested hdr could use a different ethernet type. */
> +	hint_off = geneve_sk_gro_hint_off(sk, gh, &type, &gh_len);
> +	if (hint_off) {
> +		const struct geneve_opt_gro_hint *gro_hint;
> +
> +		/*
> +		 * If the hint is present, and nested hdr validation fails, do
> +		 * not attempt plain GRO: it will ignore inner hdrs and cause
> +		 * OoO.
> +		 */
> +		gh = skb_gro_header(skb, off_gnv + gh_len, off_gnv);
> +		if (unlikely(!gh))
> +			goto out;
> +
> +		gro_hint = geneve_opt_gro_hint(gh, hint_off);
> +		if (!geneve_opt_gro_hint_validate_csum(skb, gh, gro_hint))
> +			goto out;
> +	}
> +
>  	list_for_each_entry(p, head, list) {
>  		if (!NAPI_GRO_CB(p)->same_flow)
>  			continue;

AI review reports a possible OoB access later on in:

	if (gh->opt_len != gh2->opt_len ||
		    memcmp(gh, gh2, gh_len)) {
			NAPI_GRO_CB(p)->same_flow = 0;
			continue;
		}

specifically in the memcmp() accessing bytes not pulled yet from gh2. I
think such OoB access is not possible: before reaching the buffer bounds
the other packet hdr should match the current one geneve hdr, including
the hint option. Thus, while processing the such packet, the GRO stage
should have already pulled all the relevant data.

I think this is simply too much to get for the LLM, at least ATM.

Side note: I was sure I did the shellcheck test before posting, but nipa
says I at very least forgot about it before some latest update (warn in
patch 10). @Jakub: please LMK if you prefer another revision.

/P