Re: [PATCH RFC v7 4/5] net: resolve uncommon partial checksums at ingress

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

Michael S. Tsirkin wrote:
> As reported first by Paulos Yibelo <habte.yibelo@gmail.com> (after which
> others, credited below, found more ways the bug can affect users) a
> virtio net header can place CHECKSUM_PARTIAL at any in-packet offset.
> If that state is retained, later checksum completion can modify the
> packet after protocol processing has made decisions from it.
> 
> Let's fix that: let's validate csum_start and csum_offset and only keep
> CHECKSUM_PARTIAL when csum_start and csum_offset match a parsed TCP or
> UDP header, while handling VLAN headers, IPv4 options and IPv6 extension
> headers.  Complete other non-GSO checksums immediately, before any
> protocol processing runs.
> 
> We can't easily "complete the checksum" for GSO we don't recognize - GSO
> assumes knowledge of the protocol. We could segment ignoring the
> checksum then do whatever csum_start and csum_offset say, but that
> seems like an overkill for something that shouldn't happen.
> 
> There's some risk here:
> - non-gso packets which can't be parsed get checksum calculated in
>   software, adding performance overhead
> - gso packets which can't be parsed get dropped
> 
> Note that for zero copy, userspace can change the data after the
> checksum has been calculated.  This is fundamental to zero copy by
> design, and the only outcome is a bad checksum and a packet drop.
> 
> To mitigate the risks, the following patch will add a sysctl knob to
> bypass the resolution. This is reasonable if the setup is such that
> there is no protocol processing at all.
> 
> UDP tunnel headers get the same handling on the inner header.
> 
> Packets without VIRTIO_NET_HDR_F_NEEDS_CSUM are unaffected.
> 
> Note a fundamental limitation of this approach: an nf/BPF program
> that tweaks the payload of a packet in flight may rely on the
> checksum being fixed up later, assuming the packet carries
> CHECKSUM_PARTIAL.
> 
> For metadata that fails validation the checksum
> is now completed at input, so such edits leave a stale checksum
> and the packet is dropped at the receiver.  Protocols whose
> transport header we cannot parse (anything but TCP and UDP) are
> affected even when the metadata is self-consistent.
> 
> The sysctl added in the next patch is the escape hatch for these setups,
> but it is a very blunt tool.
> 
> Reported-by: Paulos Yibelo <habte.yibelo@gmail.com>
> Reported-by: Chris Arges <carges@cloudflare.com>
> Tested-by: Chris Arges <carges@cloudflare.com>
> Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
> Suggested-by: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

> +static inline bool
> +virtio_net_l4_start(struct sk_buff *skb, unsigned int nhoff, __be16 proto,
> +		    unsigned int *thoff, int *ip_proto)
> +{
> +	if (proto == htons(ETH_P_IP)) {
> +		const struct iphdr *iph = virtio_net_ip4_parse(skb, nhoff, thoff);
> +
> +		if (!iph || ip_is_fragment(iph))
> +			return false;
> +		*ip_proto = iph->protocol;
> +		return true;
> +	}
> +
> +	if (proto == htons(ETH_P_IPV6)) {
> +		const struct ipv6hdr *iph;
> +		int flags = 0;
> +
> +		if (!pskb_may_pull(skb, nhoff + sizeof(*iph)))
> +			return false;
> +		iph = (const void *)(skb->data + nhoff);
> +		if (iph->version != 6)
> +			return false;
> +		*thoff = nhoff + sizeof(*iph);
> +		*ip_proto = iph->nexthdr;
> +		if (*ip_proto != IPPROTO_TCP && *ip_proto != IPPROTO_UDP) {
> +			*thoff = nhoff;
> +			/* Find the transport header behind any
> +			 * extension headers, AH included.
> +			 */
> +			*ip_proto = ipv6_find_hdr(skb, thoff, -1, NULL,
> +						  &flags);
> +			if (*ip_proto < 0 || flags & IP6_FH_F_FRAG)
> +				return false;

Minor, not cause for respinning on its own.

    `flags` should include `IP6_FH_F_AUTH` so `ipv6_find_hdr()` stops at
    `NEXTHDR_AUTH` instead of skipping it:

    - `CHECKSUM_PARTIAL` and GSO behind IPsec AH are invalid: AH authenticates
      the inner L4 header including its checksum field

> +		}
> +		return true;
> +	}
> +
> +	return false;
> +}
> +

> @@ -368,6 +660,21 @@ virtio_net_hdr_tnl_to_skb(struct sk_buff *skb,
>  	skb->inner_mac_header = inner_nh + skb_headroom(skb);
>  	skb->transport_header = outer_th + skb_headroom(skb);
>  	skb->encapsulation = 1;
> +
> +	if (!skb_is_gso(skb)) {
> +		/* Too small to segment: complete the checksum in place
> +		 * instead of dropping - senders do tag small packets
> +		 * with GSO headers, e.g. packetdrill sets gso_type and
> +		 * gso_size with --mss, whatever the packet size.
> +		 * The outer UDP checksum is not calculated, but a zero
> +		 * one is fine: UDP over IPv4 allows a disabled
> +		 * checksum, and so do tunnels over IPv6 where the
> +		 * receiver opts in (RFC 6935, RFC 6936).
> +		 */
> +		skb_shinfo(skb)->gso_type &= ~(SKB_GSO_UDP_TUNNEL_CSUM |
> +					       SKB_GSO_UDP_TUNNEL);

Maybe this carve-out for non-gso packet is best moved to its own
patch, as that is where the remaining issues (to merging the rest)
seem to be.

    1. Moving `!tnl_csum_negotiated` below `virtio_net_csum_partial_ok_at_offset()`
       accepts packets with an unnegotiated `VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM`
       flag whenever inner validation falls back to
       `__virtio_net_resolve_csum_partial_mismatch()`.
    2. `__virtio_net_resolve_csum_partial_mismatch()` calls `skb_checksum_help()`,
       which completes only the inner checksum at `hdr->csum_start` and sets
       `skb->ip_summed = CHECKSUM_NONE`, leaving the outer UDP `uh->check` with
       an uncompleted pseudo-header checksum seed.

    Despite the comment, this block neither completes any checksum in place nor
    is the outer UDP checksum zero when `VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM` is set:
  
    - With `VIRTIO_NET_HDR_F_UDP_TUNNEL_CSUM`, the outer `uh->check` carries the
      non-zero pseudo-header checksum seed (as expected by
      `__skb_udp_tunnel_segment()`), not `0`.
    - Clearing `SKB_GSO_UDP_TUNNEL_CSUM` on a `!skb_is_gso(skb)` packet leaves
      `skb->ip_summed == CHECKSUM_PARTIAL` at `inner_th` with nothing completing
      the outer `uh->check` on transmit/forwarding (`validate_xmit_skb()` only
      completes/offloads the single inner checksum, emitting a corrupt outer UDP
      checksum). Even zeroing `uh->check` would be invalid over IPv6 unless the
      receiver explicitly opted into RFC 6936 (`UDP_NO_CHECK6_RX`).

> +	}
> +
>  	return 0;
>  }
>  
> -- 
> MST
>