Hi Ben,
This short series attempts to resolve two problems related to changing
flow_push_mpls() to use the slow path if there are two many MPLS LSEs
present.
1) In this case it is not safe to call flow_push_mpls() as
it can't do anything useful: there is no way to correctly
update the MPLS label stack. Instead it aborts which seems
undesirable :)
My proposal is that if the slow path is chosen then
compose_mpls_push_action should return before updating flow.
I'm unsure of the correctness of this.
An alternate approach is to go back to dropping the packet
in this situation.
2) Update the test suite to expect a userspace action instead of a drop.
This test only passes with a fix for 1) in place.
Simon Horman (2):
ofproto-dpif-xlate: Do not push LSE to flow if its label stack is full
ofproto-dpif: Test for A MPLS pop on a packet whose label stack going
to userspace
ofproto/ofproto-dpif-xlate.c | 1 +
tests/ofproto-dpif.at | 30 +++++++++++++++++++++++++-----
2 files changed, 26 insertions(+), 5 deletions(-)
--
1.8.5.2
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev
