librbd: prevent self-blacklisting during break lock

[dillaman]

Fixes: http://tracker.ceph.com/issues/18666
Signed-off-by: Jason Dillaman dillaman@redhat.com

[trociny]

@dillaman just wondering, it looks it would be safer to check address. Do you check entity because there is no rados API for get_myaddr?

Also, if we are breaking without blacklisting, is it ok to self-break?

[dillaman]

@trociny The client id is a globally unique identifier (and the only one available to us) and the lock itself is attached to the client global id, not its IP address. Do you have a case in mind where this would fail?

We aren't introducing any additional worries if you disabled blacklisting and you attempted to self-break. In that case, it would at least be able to recover the lock (unlike if it blacklisted itself).

[trociny]

@dillaman

Do you have a case in mind where this would fail?

Never mind, I forgot that the address also includes nonce. I was thinking that if an application had two clients, their addresses would match (which is not the case, and actually then our BreakLock tests would not work).

[trociny]

LGTM