Re: [PATCH 1/2 net] ip_gre: add validation for csum_start

On Wed, Sep 1, 2021 at 11:53 AM Ido Schimmel [off-list ref] wrote:

Thanks for the quick reply, Willem.

On Wed, Sep 01, 2021 at 09:46:48AM -0400, Willem de Bruijn wrote:

quoted

Thanks for the detailed report, Ido.

This is a gre tunnel device with csum/ocsum enabled, correct?

Correct.

quoted

How was this packet generated: does it come from the local stack or is
it a custom packet injected from userspace, e.g., with a packet socket
with vnet_hdr?

The packet is received by a physical port and injected to the kernel's
Rx path by mlxsw (which does not support checksumming). The IPv4 routing
code then forwards the packet to the GRE tunnel.

I was able to reproduce the issue using veth pairs and a packet socket
[1]. Running the reproducer with the debug patch from before, I get the
following output [2].

Thanks for that device independent repro.

As expected, the following fixes it for these packets:

-       if (csum && skb_checksum_start(skb) < skb->data)
+       if (csum && skb->ip_summed == CHECKSUM_PARTIAL &&
+           skb_checksum_start(skb) < skb->data)

The question is whether we're doing the right thing when CHECKSUM_PARTIAL
is set.

Local checksum offload allows for cheap calculation of outer checksums, by
relying on the fact that the inner packet with the checksum field filled in will
sum to zero. It relies on checksum offload to compute this inner checksum,
so expects csum_start and csum_off to point after the GRE header.

If so, then the existing fix won't break correctly configured skbs as it only
drops packets for which this does not hold.