Re: [PATCH net] net: gso: fix panic on frag_list with mixed head alloc types

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 27 Oct 2022 10:20:56 +0200 Jiri Benc wrote:
> Since commit 3dcbdb134f32 ("net: gso: Fix skb_segment splat when
> splitting gso_size mangled skb having linear-headed frag_list"), it is
> allowed to change gso_size of a GRO packet. However, that commit assumes
> that "checking the first list_skb member suffices; i.e if either of the
> list_skb members have non head_frag head, then the first one has too".
> 
> It turns out this assumption does not hold. We've seen BUG_ON being hit
> in skb_segment when skbs on the frag_list had differing head_frag. That
> particular case was with vmxnet3; looking at the driver, it indeed uses
> different skb allocation strategies based on the packet size.

Where are you looking? I'm not seeing it TBH.

I don't think the driver is that important, tho, __napi_alloc_skb() 
will select page backing or kmalloc, all by itself.

The patch LGTM, adding more CCs in case I'm missing something.

> The last packet in frag_list can thus be kmalloced if it is
> sufficiently small. And there's nothing preventing drivers from
> mixing things even more freely.
> 
> There are three different locations where this can be fixed:
> 
> (1) We could check head_frag in GRO and not allow GROing skbs with
>     different head_frag. However, that would lead to performance
>     regression (at least on vmxnet3) on normal forward paths with
>     unmodified gso_size, where mixed head_frag is not a problem.
> 
> (2) Set a flag in bpf_skb_net_grow and bpf_skb_net_shrink indicating
>     that NETIF_F_SG is undesirable. That would need to eat a bit in
>     sk_buff. Furthermore, that flag can be unset when all skbs on the
>     frag_list are page backed. To retain good performance,
>     bpf_skb_net_grow/shrink would have to walk the frag_list.
> 
> (3) Walk the frag_list in skb_segment when determining whether
>     NETIF_F_SG should be cleared. This of course slows things down.
> 
> This patch implements (3). To limit the performance impact in
> skb_segment, the list is walked only for skbs with SKB_GSO_DODGY set
> that have gso_size changed. Normal paths thus will not hit it.
> 
> Fixes: 3dcbdb134f32 ("net: gso: Fix skb_segment splat when splitting
> gso_size mangled skb having linear-headed frag_list") Signed-off-by:
> Jiri Benc <jbenc@redhat.com>

> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 1d9719e72f9d..bbf3acff44c6 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -4134,23 +4134,25 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
>  	int i = 0;
>  	int pos;
>  
> -	if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
> -	    (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
> -		/* gso_size is untrusted, and we have a frag_list with a linear
> -		 * non head_frag head.
> -		 *
> -		 * (we assume checking the first list_skb member suffices;
> -		 * i.e if either of the list_skb members have non head_frag
> -		 * head, then the first one has too).
> -		 *
> -		 * If head_skb's headlen does not fit requested gso_size, it
> -		 * means that the frag_list members do NOT terminate on exact
> -		 * gso_size boundaries. Hence we cannot perform skb_frag_t page
> -		 * sharing. Therefore we must fallback to copying the frag_list
> -		 * skbs; we do so by disabling SG.
> -		 */
> -		if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
> -			features &= ~NETIF_F_SG;
> +	if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
> +	    mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
> +		struct sk_buff *check_skb;
> +
> +		for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
> +			if (skb_headlen(check_skb) && !check_skb->head_frag) {
> +				/* gso_size is untrusted, and we have a frag_list with
> +				 * a linear non head_frag item.
> +				 *
> +				 * If head_skb's headlen does not fit requested gso_size,
> +				 * it means that the frag_list members do NOT terminate
> +				 * on exact gso_size boundaries. Hence we cannot perform
> +				 * skb_frag_t page sharing. Therefore we must fallback to
> +				 * copying the frag_list skbs; we do so by disabling SG.
> +				 */
> +				features &= ~NETIF_F_SG;
> +				break;
> +			}
> +		}
>  	}
>  
>  	__skb_push(head_skb, doffset);