On Fri, Mar 24, 2017 at 09:01:17PM +0800, Liping Zhang wrote: > Hi Pablo, > > 2017-03-24 20:17 GMT+08:00 Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx>: > >> --- a/net/netfilter/nfnetlink_cttimeout.c > >> +++ b/net/netfilter/nfnetlink_cttimeout.c > >> @@ -646,8 +646,8 @@ static void __exit cttimeout_exit(void) > >> #ifdef CONFIG_NF_CONNTRACK_TIMEOUT > >> RCU_INIT_POINTER(nf_ct_timeout_find_get_hook, NULL); > >> RCU_INIT_POINTER(nf_ct_timeout_put_hook, NULL); > >> + synchronize_rcu(); > >> #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */ > >> - rcu_barrier(); > > > > cttimeout relies on kfree_rcu(). > > > > Are you sure we don't need this? > > > > According to: > > > > https://www.kernel.org/doc/Documentation/RCU/rcubarrier.txt > > > > "We could try placing a synchronize_rcu() in the module-exit code path, > > but this is not sufficient." > > > > Then: > > > > "Please note that rcu_barrier() does -not- imply synchronize_rcu(), in > > particular, if there are no RCU callbacks queued anywhere, > > rcu_barrier() is within its rights to return immediately, without > > waiting for a grace period to elapse." > > This is because we use kfree_rcu to free the cttimeout objects. So I think > rcu_barrier() is not needed anymore. > > Quoted from https://lwn.net/Articles/433493/ : > "And kfree_rcu() is also help for unloadable modules, kfree_rcu() does not > queue any function which belong to the module, so a rcu_barrier() can > be avoid when module exit." > > Also from commit 9ab1544eb419 ("rcu: introduce kfree_rcu()"): > "Many rcu callbacks functions just call kfree() on the base structure. > These functions are trivial, but their size adds up, and furthermore > when they are used in a kernel module, that module must invoke the > high-latency rcu_barrier() function at module-unload time." Right, thanks for explaining. I think we can get this smaller: it should be possible to avoid this synchronize_rcu() call from nf_conntrack_{register,unregister}_notifier(). These two are called from ctnetlink netns path, this patch already adds a synchronize_rcu() spot to ctnetlink module removal which is where the event callback can vanish. You can just add a comment there so we don't forget about this, eg. @@ -290,6 +290,7 @@ void nf_conntrack_unregister_notifier(struct net *net, BUG_ON(notify != new); RCU_INIT_POINTER(net->ct.nf_conntrack_event_cb, NULL); mutex_unlock(&nf_ct_ecache_mutex); + /* synchronize_rcu() is called from ctnetlink. */ } What do you think? -- To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
