Re: [PATCH v2 net-next 01/14] rtnetlink: prepare nla_put_iflink() to run under RCU

[Donald Hunter <donald.hunter@gmail.com>]

Eric Dumazet <edumazet@google.com> writes:
>>
>> And use of them here:
>>
>> > diff --git a/drivers/net/can/vxcan.c b/drivers/net/can/vxcan.c
>> > index 98c669ad5141479b509ee924ddba3da6bca554cd..f7fabba707ea640cab8863e63bb19294e333ba2c 100644
>> > --- a/drivers/net/can/vxcan.c
>> > +++ b/drivers/net/can/vxcan.c
>> > @@ -119,7 +119,7 @@ static int vxcan_get_iflink(const struct net_device *dev)
>> >
>> >       rcu_read_lock();
>> >       peer = rcu_dereference(priv->peer);
>> > -     iflink = peer ? peer->ifindex : 0;
>> > +     iflink = peer ? READ_ONCE(peer->ifindex) : 0;
>> >       rcu_read_unlock();
>> >
>> >       return iflink;
>>
>>
>> > We do not need an rcu_read_lock() only to fetch dev->ifindex, if this
>> > is what concerns you.
>>
>> In which case, it seems that no .ndo_get_iflink implementations should
>> need the rcu_read_* calls?
>
> rcu_read_lock() is needed in all cases a dereference is performed,
> expecting RCU protection of the pointer.
>
> In vxcan_get_iflink(), we access priv->peer, then peer->ifindex.
>
> rcu_read_lock() is needed because of the second dereference, peer->ifindex.
>
> Without rcu_read_lock(), peer could be freed before we get a chance to
> read peer->ifindex.

Thanks for the detailed explanation.