NAT implementation in OVN uses connection tracker to replace
source and dest ips. This implementation works fine and
is the right approach for cases where external ips are shared
(i.e. SNAT) or where we replace ip only when relevant flow is there
(i.e. DNAT).

However, it opens the possibility of Dos Attack, where attacker
can easily simluate multiple 5 tuples, to consume the connection
tracker entry in an OVN chassis. This way they can easily attain
the CT limit, there by impacting the usage of it by other features
like valid NAT, ACL etc.

This attack is even worse, when external ip is a public ip,
i.e internet routable ip.

In this patch we are introducing an option with NAT table entry.
Option "stateless=true" indicates that NAT implmentation
should not be using CT, i.e it should not use ct_snat/dnat actions.

Instead of ct_* actions, we will use ip4.src/dst OVN actions, which
will replace source and destination ips, while recalculating the
checksums.

This option is applicable only for the NAT rules which can be
1:1 mapped between inner and external ips, i.e dnat_and_snat rule.

Signed-off-by: Ankur Sharma <ankur.sha...@nutanix.com>

Ankur Sharma (2):
  OVN: ADD nbctl cli to mark a dnat_and_snat rule as stateless
  OVN: Use ip4.src and ip4.dst actions for NAT rules

 northd/ovn-northd.8.xml   |  34 +++--
 northd/ovn-northd.c       |  86 +++++++++++--
 ovn-nb.ovsschema          |   6 +-
 ovn-nb.xml                |   5 +
 tests/ovn-nbctl.at        |  29 +++++
 tests/ovn-northd.at       |  57 +++++++++
 tests/ovn.at              | 311 ++++++++++++++++++++++++++++++++++++++++++++++
 utilities/ovn-nbctl.8.xml |  12 +-
 utilities/ovn-nbctl.c     |  30 ++++-
 9 files changed, 550 insertions(+), 20 deletions(-)

-- 
1.8.3.1

_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to