Re: [PATCH tip/core/rcu 05/13] net: add checking to rcu_dereference() primitives

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

Le dimanche 14 février 2010 à 09:34 +0100, Michał Mirosław a écrit :
> 2010/2/12 Eric Dumazet <eric.dumazet@gmail.com>:
> > Le jeudi 11 février 2010 à 16:00 -0800, Paul E. McKenney a écrit :
> [...]
> >> @@ -1654,12 +1654,12 @@ static struct dn_route *dn_rt_cache_get_first(struct seq_file *seq)
> >>
> >>       for(s->bucket = dn_rt_hash_mask; s->bucket >= 0; --s->bucket) {
> >>               rcu_read_lock_bh();
> >> -             rt = dn_rt_hash_table[s->bucket].chain;
> >> +             rt = rcu_dereference_bh(dn_rt_hash_table[s->bucket].chain);
> >>               if (rt)
> >>                       break;
> >>               rcu_read_unlock_bh();
> >>       }
> >> -     return rcu_dereference(rt);
> >> +     return rt;
> >>  }
> 
> Isn't there a bug? Looks like data pointed to by rt should be
> protected by RCU, but the rcu_read_lock is withdrawn before access.
> 

Not really a bug, since we exit from dn_rt_cache_get_first() with
rcu_read_lock_bh()

We call the unlock only if NULL is returned, and rcu_dereference(NULL)
can be done in any context.

Paul had to move the rcu_dereference() so that no lockdep warning
triggers for rcu_dereference(NULL), its more a cleanup than a bug fix.