[net] net: validate untrusted gso packets without csum offload

From: Willem de Bruijn <willemb@google.com>

Syzkaller again found a path to a kernel crash through bad gso input.
By building an excessively large packet to cause an skb field to wrap.

If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
skb_partial_csum_set.

GSO packets that do not set checksum offload are suspicious and rare.
Most callers of virtio_net_hdr_to_skb already pass them to
skb_probe_transport_header.

Move that test forward, change it to detect parse failure and drop
packets on failure as those cleary are not one of the legitimate
VIRTIO_NET_HDR_GSO types.

Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>

---

This captures a variety of bad gso packets, but to tighten further:

- drop SKB_GSO_DODGY packets with ipip/sit/.. , which cannot be legal.
    by ipip_gso_segment wrappers around inet_gso_segment
    expands on 121d57af308d ("gso: validate gso_type in GSO handlers")

- limit the number of ipv6 exthdrs allowed from dodgy sources.
    not sure where to draw the line. but not at 64K ;)

- validate the network and transport protocol returned in
    skb_probe_transport_header against the VIRTIO_NET_HDR_GSO type

- probe all dodgy GSO packets, also those that set checksum offload.
    this will have a performance impact, discussed previously in
    http://patchwork.ozlabs.org/patch/861874/
    but it would have blocked this latest bug as well

All but the last one seem pretty uncontroversial to me. If no one
objects I plan to send those to net-next.

---
 include/linux/skbuff.h     | 2 +-
 include/linux/virtio_net.h | 9 +++++++++
 2 files changed, 10 insertions(+), 1 deletion(-)

On 02/15/2019 09:15 AM, Willem de Bruijn wrote:
> From: Willem de Bruijn <willemb@google.com>
> 
> Syzkaller again found a path to a kernel crash through bad gso input.
> By building an excessively large packet to cause an skb field to wrap.
> 
> If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
> skb_partial_csum_set.
> 
> GSO packets that do not set checksum offload are suspicious and rare.
> Most callers of virtio_net_hdr_to_skb already pass them to
> skb_probe_transport_header.
> 
> Move that test forward, change it to detect parse failure and drop
> packets on failure as those cleary are not one of the legitimate
> VIRTIO_NET_HDR_GSO types.
> 
> Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
> Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
> Reported-by: syzbot <syzkaller@googlegroups.com>
> Signed-off-by: Willem de Bruijn <willemb@google.com>

Reviewed-by: Eric Dumazet <edumazet@google.com>

From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date: Fri, 15 Feb 2019 12:15:47 -0500

> From: Willem de Bruijn <willemb@google.com>
> 
> Syzkaller again found a path to a kernel crash through bad gso input.
> By building an excessively large packet to cause an skb field to wrap.
> 
> If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
> skb_partial_csum_set.
> 
> GSO packets that do not set checksum offload are suspicious and rare.
> Most callers of virtio_net_hdr_to_skb already pass them to
> skb_probe_transport_header.
> 
> Move that test forward, change it to detect parse failure and drop
> packets on failure as those cleary are not one of the legitimate
> VIRTIO_NET_HDR_GSO types.
> 
> Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
> Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
> Reported-by: syzbot <syzkaller@googlegroups.com>
> Signed-off-by: Willem de Bruijn <willemb@google.com>

Applied and queued up for -stable, thanks Willem.

> This captures a variety of bad gso packets, but to tighten further:
> 
> - drop SKB_GSO_DODGY packets with ipip/sit/.. , which cannot be legal.
>     by ipip_gso_segment wrappers around inet_gso_segment
>     expands on 121d57af308d ("gso: validate gso_type in GSO handlers")
> 
> - limit the number of ipv6 exthdrs allowed from dodgy sources.
>     not sure where to draw the line. but not at 64K ;)
> 
> - validate the network and transport protocol returned in
>     skb_probe_transport_header against the VIRTIO_NET_HDR_GSO type
> 
> - probe all dodgy GSO packets, also those that set checksum offload.
>     this will have a performance impact, discussed previously in
>     http://patchwork.ozlabs.org/patch/861874/
>     but it would have blocked this latest bug as well
> 
> All but the last one seem pretty uncontroversial to me. If no one
> objects I plan to send those to net-next.

No objections from me.

On Fri, Feb 15, 2019 at 12:15 PM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> From: Willem de Bruijn <willemb@google.com>
>
> Syzkaller again found a path to a kernel crash through bad gso input.
> By building an excessively large packet to cause an skb field to wrap.
>
> If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
> skb_partial_csum_set.
>
> GSO packets that do not set checksum offload are suspicious and rare.
> Most callers of virtio_net_hdr_to_skb already pass them to
> skb_probe_transport_header.
>
> Move that test forward, change it to detect parse failure and drop
> packets on failure as those cleary are not one of the legitimate
> VIRTIO_NET_HDR_GSO types.
>
> Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
> Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
> Reported-by: syzbot <syzkaller@googlegroups.com>
> Signed-off-by: Willem de Bruijn <willemb@google.com>

This causes false positive drops on virtio-net and tun for these
packets with gso without csum_off. And on pf_packet with proto 0.

This happens because skb->protocol is set in these callers after
the call to virtio_net_hdr_to_skb. And the flow dissector relies on
this to start dissection, not the link layer header (if present).
Moving this logic forward is too much churn for net, especially since
eth_type_header pulls the header, requiring additional changes to
adjust csum_start.

virtio_net_hdr_set_proto() aims to fix this by deriving skb->protocol
from the gso_type. But unfortunately for UDP it unconditionally
selects ipv4, which will cause drops for UDP over ipv6.

For net I plan to just ignore the error for these callers that do not
set skb->protocol.

-                       if (!skb_transport_header_was_set(skb))
+                       if (!skb_transport_header_was_set(skb) && skb->protocol)

Possibly with an extension of tpacket_set_protocol to also cover
packet_snd, so that that cannot evade it on purpose. Other callers can
wait till net-next.



>
> ---
>
> This captures a variety of bad gso packets, but to tighten further:
>
> - drop SKB_GSO_DODGY packets with ipip/sit/.. , which cannot be legal.
>     by ipip_gso_segment wrappers around inet_gso_segment
>     expands on 121d57af308d ("gso: validate gso_type in GSO handlers")
>
> - limit the number of ipv6 exthdrs allowed from dodgy sources.
>     not sure where to draw the line. but not at 64K ;)

This already exists, in the form of skb_flow_dissect_allowed






> - validate the network and transport protocol returned in
>     skb_probe_transport_header against the VIRTIO_NET_HDR_GSO type
>
> - probe all dodgy GSO packets, also those that set checksum offload.
>     this will have a performance impact, discussed previously in
>     http://patchwork.ozlabs.org/patch/861874/
>     but it would have blocked this latest bug as well
>
> All but the last one seem pretty uncontroversial to me. If no one
> objects I plan to send those to net-next.
>
> ---
>  include/linux/skbuff.h     | 2 +-
>  include/linux/virtio_net.h | 9 +++++++++
>  2 files changed, 10 insertions(+), 1 deletion(-)
>
> diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> index 95d25b010a25..4c1c82a5678c 100644
> --- a/include/linux/skbuff.h
> +++ b/include/linux/skbuff.h
> @@ -2434,7 +2434,7 @@ static inline void skb_probe_transport_header(struct sk_buff *skb,
>
>         if (skb_flow_dissect_flow_keys_basic(skb, &keys, NULL, 0, 0, 0, 0))
>                 skb_set_transport_header(skb, keys.control.thoff);
> -       else
> +       else if (offset_hint >= 0)
>                 skb_set_transport_header(skb, offset_hint);
>  }
>
> diff --git a/include/linux/virtio_net.h b/include/linux/virtio_net.h
> index cb462f9ab7dd..71f2394abbf7 100644
> --- a/include/linux/virtio_net.h
> +++ b/include/linux/virtio_net.h
> @@ -57,6 +57,15 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,
>
>                 if (!skb_partial_csum_set(skb, start, off))
>                         return -EINVAL;
> +       } else {
> +               /* gso packets without NEEDS_CSUM do not set transport_offset.
> +                * probe and drop if does not match one of the above types.
> +                */
> +               if (gso_type) {
> +                       skb_probe_transport_header(skb, -1);
> +                       if (!skb_transport_header_was_set(skb))
> +                               return -EINVAL;
> +               }
>         }
>
>         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
> --
> 2.21.0.rc0.258.g878e2cd30e-goog
>

On Mon, Feb 18, 2019 at 2:12 PM Willem de Bruijn
<willemdebruijn.kernel@gmail.com> wrote:
>
> On Fri, Feb 15, 2019 at 12:15 PM Willem de Bruijn
> <willemdebruijn.kernel@gmail.com> wrote:
> >
> > From: Willem de Bruijn <willemb@google.com>
> >
> > Syzkaller again found a path to a kernel crash through bad gso input.
> > By building an excessively large packet to cause an skb field to wrap.
> >
> > If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
> > skb_partial_csum_set.
> >
> > GSO packets that do not set checksum offload are suspicious and rare.
> > Most callers of virtio_net_hdr_to_skb already pass them to
> > skb_probe_transport_header.
> >
> > Move that test forward, change it to detect parse failure and drop
> > packets on failure as those cleary are not one of the legitimate
> > VIRTIO_NET_HDR_GSO types.
> >
> > Fixes: bfd5f4a3d605 ("packet: Add GSO/csum offload support.")
> > Fixes: f43798c27684 ("tun: Allow GSO using virtio_net_hdr")
> > Reported-by: syzbot <syzkaller@googlegroups.com>
> > Signed-off-by: Willem de Bruijn <willemb@google.com>
>
> This causes false positive drops on virtio-net and tun for these
> packets with gso without csum_off. And on pf_packet with proto 0.
>
> This happens because skb->protocol is set in these callers after
> the call to virtio_net_hdr_to_skb. And the flow dissector relies on
> this to start dissection, not the link layer header (if present).
> Moving this logic forward is too much churn for net, especially since
> eth_type_header pulls the header, requiring additional changes to
> adjust csum_start.
>
> virtio_net_hdr_set_proto() aims to fix this by deriving skb->protocol
> from the gso_type. But unfortunately for UDP it unconditionally
> selects ipv4, which will cause drops for UDP over ipv6.

Suggested fix at http://patchwork.ozlabs.org/patch/1044429/