Hi,
I tried to check through the code and identify problems not yet mentioned by
Paul E. McKenney. They were found by reading through
Documentation/RCU/checklist.txt and related documents in the folder.
It doesn't address the reference counting problem for gw_nodes and interfaces.
Those leaks are happen in gw_election, get_batman_if_by_netdev and
get_active_batman_if. We must increase the refcnt (using atomic_inc) inside the
rcu_read_lock()..rcu_read_unlock() before we attach to the structure it
"leaks". When another function now removed it from its usage context
(primary_if, usage on stack, ...) then atomic_dec_and_test the refcnt. If it is
decremented to zero then we can issue the call_rcu to the freeing function. So
"put" of those functions is not allowed inside an rcu_read_lock. As said before
the hold must always be called inside a rcu_read_lock.
Best regards,
Sven
