The destroy() method sets the IKE_SA on the bus to NULL, we reset it to
the current IKE_SA so any events and log messages that follow happen in
the correct context.
A practical example where this is problematic is a DH group mismatch,
which causes the first CREATE_CHILD_SA exchange to fail. Because the SA
was not reset previously, the message() hook for the CREATE_CHILD_SA
response, for instance, was triggered outside the context of an IKE_SA,
that is, the ike_sa parameter was NULL, which is definitely not expected
by several plugins.
Fixes#862.
If additional tasks get queued before/while rekeying an IKE_SA, these get
migrated to the new IKE_SA. We previously did not trigger initiation of these
tasks, though, leaving the task unexecuted until a new task gets queued.
Under some conditions it can happen that the CREATE_CHILD_SA exchange for
rekeying the IKE_SA initiated by the peer is successful, but the delete message
does not follow. For example if processing takes just too long locally, the
peer might consider us dead, but we won't notice that.
As this leaves the old IKE_SA in IKE_REKEYING state, we currently avoid actively
initiating any tasks, such as rekeying or scheduled DPD. This leaves the IKE_SA
in a dead and unusable state. To avoid that situation, we schedule a timeout
to wait for the DELETE message to follow the CREATE_CHILD_SA, before we
actively start to delete the IKE_SA.
Alternatively we could start a liveness check on the SA after a timeout to see
if the peer still has that state and we can expect the delete to follow. But
it is unclear if all peers can handle such messages in this very special state,
so we currently don't go for that approach.
While we could calculate the timeout based on the local retransmission timeout,
the peer might use a different scheme, so a fixed timeout works as well.
Fixes#742.
Andreas Steffen