Re: [patch net-next V2] net: introduce ethernet teaming device

[Eric Dumazet <eric.dumazet@gmail.com>]

Le dimanche 23 octobre 2011 à 10:25 +0200, Jiri Pirko a écrit :

> Please forgive me, it's possible I'm missing something. But I see no way how
> team->mode_ops.receive can change during team_handle_frame (holding
> rcu_read_lock) for the reason I stated earlier.
> 
> team_port_del() calls netdev_rx_handler_unregister() and after that it
> calls synchronize_rcu(). That ensures that by the finish of team_port_del()
> run, team_handle_frame() is not called for this port anymore.
> 
> And this combined with "if (!list_empty(&team->port_list))" check in
> team_change_mode() ensures safety.
> 
> Of course team_port_del() and team_change_mode() are both protected by
> team->lock so they are mutually excluded.

Oh well.

Jirka, I believe I do understand how RCU is working ;)

There is an obvious problem I pointed to you, but you persist leaving
this potential bug.

After netdev_rx_handler_unregister(), you can still have other cpus
calling your handler and reading/using previous memory content. Only
after synchronize_rcu() you can be safe. But in your patch the bug
window is _exactly_ _before_ synchronize_rcu() returns.

Your spinlock wont help you at all, since readers dont take it.
Spinlock only protects writers.


So a reader, even holding rcu lock, can really see two different
mode_ops.receive values for the :

if (team->mode_ops.receive)
	res = team->mode_ops.receive(team, port, skb);


rcu_lock() doesnt mean the reader can see an unique .receive value,

I am afraid you misunderstood the point.

Real point of RCU here is that the _writer_ wont returns from
synchronize_rcu() if at least one reader is still running the handler.

No problem with me, I'll just post a patch later, I just cant Ack your
work as is.