rxrpc: Fix net namespace cleanup

In rxrpc_destroy_all_calls(), there are two phases: (1) make sure the
->calls list is empty, emitting error messages if not, and (2) wait for the
RCU cleanup to happen on outstanding calls (ie. ->nr_calls becomes 0).

To avoid taking the call_lock, the function prechecks ->calls and if empty,
it returns to avoid taking the lock - this is wrong, however: it still
needs to go and do the second phase and wait for ->nr_calls to become 0.

Without this, the rxrpc_net struct may get deallocated before we get to the
RCU cleanup for the last calls.  This can lead to:

  Slab corruption (Not tainted): kmalloc-16k start=ffff88802b178000, len=16384
  050: 6b 6b 6b 6b 6b 6b 6b 6b 61 6b 6b 6b 6b 6b 6b 6b  kkkkkkkkakkkkkkk

Note the "61" at offset 0x58.  This corresponds to the ->nr_calls member of
struct rxrpc_net (which is >9k in size, and thus allocated out of the 16k
slab).

Fix this by flipping the condition on the if-statement, putting the locked
section inside the if-body and dropping the return from there.  The
function will then always go on to wait for the RCU cleanup on outstanding
calls.

Fixes: 2baec2c3 ("rxrpc: Support network namespacing")
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/rxrpc/call_object.c

@@ -604,13 +604,12 @@ void rxrpc_destroy_all_calls(struct rxrpc_net *rxnet)
 
 	_enter("");
 
-	if (list_empty(&rxnet->calls))
-		return;
-
+	if (!list_empty(&rxnet->calls)) {
 		write_lock(&rxnet->call_lock);
 
 		while (!list_empty(&rxnet->calls)) {
-		call = list_entry(rxnet->calls.next, struct rxrpc_call, link);
+			call = list_entry(rxnet->calls.next,
+					  struct rxrpc_call, link);
 			_debug("Zapping call %p", call);
 
 			rxrpc_see_call(call);
@@ -627,6 +626,7 @@ void rxrpc_destroy_all_calls(struct rxrpc_net *rxnet)
 		}
 
 		write_unlock(&rxnet->call_lock);
+	}
 
 	atomic_dec(&rxnet->nr_calls);
 	wait_var_event(&rxnet->nr_calls, !atomic_read(&rxnet->nr_calls));