From: Herbert Xu <herbert@gondor.apana.org.au> Date: Sun, 1 Jun 2003 17:07:17 +1000 And here is a patch to fix the leak in the flow cache. They used to be pol_put's before the move. I understand the bug, but I'd prefer to fix it in some other way. The idea is to make flow cache completely stateless and to know as little about objects as possible. In this view the flow cache does not kill objects, it merely get/put references to them. So what we might need to do instead is to create a policy garbage collection list, xfrm_policy_kill() adds entries to here and a periodic timer checks the entries for zero refcount. This points out a flaw in the flow cache (regardless of your scheme or mine), namely that until something causes flow cache to shrink the policies can hang around forever. (Ie. is no further packets go through the system, the freeing engine hangs) Therefore this is my bug and my problem. I will take care of this. I need to add a forceful flush mechanism to the flow cache that will destroy all such references. Let us name this interface: void flow_cache_flush(struct flowi *key); It is defined to delete all flow cache entries matching given key. It must be called after the tables that the "flow_resolve_t" callback for this object will search. This means we'd invoke it from xfrm_policy_kill() or similar. - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
