bfdd: fix BFD_GETCBIT

[idryzhov]

The macro should check C bit, not F bit.

[ton31337]

Wondering how it worked stuff before this 😄

But, seems the tests are failing now.

[idryzhov]

Apparently it didn't work :)
Unfortunately I don't have time right now to look into the test failures. Maybe someone else could take a look...

[rwestphal]

My analysis is that your fix exposed a problem in the topotest logic.

This is the test that is failing:

def test_bfd_loss_intermediate():
    """
    Assert that BFD notices the bfd link down failure.
    but BGP entries should still be present
    """

The problem is that the premise of this test is wrong.

Since the BGP peers are configured with the bfd check-control-plane-failure option, we're declaring that their corresponding BFD session doesn't share fate with the control plane. In practice, this means the BFD packets are sent with the "Control Plane Independent" bit (aka C-bit) set.

When a BFD down event is detected and C-bit is set, that represents an actual forwarding failure that can not be ignored. As such, bgpd is correct in bringing the session down when that happens.

Prior to this PR, the topotest was passing because of this:

BGP: [Q4BCV-6FHZ5] zclient_bfd_session_update: 2001:db8:1::1/128 -> 2001:db8:4::1/128 VRF default(0) (CPI bit no): Down
BGP: [MKVHZ-7MS3V] bfd_session_status_update: neighbor 2001:db8:4::1 vrf default(0) bfd state Up -> Down 
BGP: [JG709-75BG6] 2001:db8:4::1 BFD DOWN message ignored in the process of graceful restart when C bit is cleared

bgpd was ignoring the BFD down event because it thought the C-bit was clear, so the failure was being dismissed as a side effect of the restart taking place. This is the correct behavior as specified here.

The topotest test descriptions and reference data need to be fixed accordingly.

[donaldsharp]

@rwestphal -> Do we need to worry about backwards comparability issues with these changes? And can we get some help fixing up the tests?

[rwestphal]

@donaldsharp I think this PR will merely fix protocol behavior in the face of BFD link down events. I can't think of any backward compatibility issues.

Also, I believe very few people should be using the neighbor A.B.C.D bfd check-control-plane-failure command, as it only makes sense when BFD is implemented in hardware.

I've pushed a topotest update to Igor's branch.