Re: [RFCv0 1/5] Bluetooth: Use locks in RCU updater code

[Emeltchenko Andrei <Andrei.Emeltchenko.news@gmail.com>]

Hi Ulisses,

On Mon, Jan 30, 2012 at 03:17:15PM -0200, Ulisses Furquim wrote:
> Hi Andrei,
> 
> On Mon, Jan 30, 2012 at 1:09 PM, Emeltchenko Andrei
> <Andrei.Emeltchenko.news@gmail.com> wrote:
> > From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> >
> > Code which makes changes to RCU list shall be locked.
> >
> > Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> > ---
> >  net/bluetooth/l2cap_core.c |   13 +++++++++----
> >  1 files changed, 9 insertions(+), 4 deletions(-)
> >
> > diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > index 6991821..f54768e 100644
> > --- a/net/bluetooth/l2cap_core.c
> > +++ b/net/bluetooth/l2cap_core.c
> > @@ -743,13 +743,13 @@ static void l2cap_send_disconn_req(struct l2cap_conn *conn, struct l2cap_chan *c
> >  /* ---- L2CAP connections ---- */
> >  static void l2cap_conn_start(struct l2cap_conn *conn)
> >  {
> > -       struct l2cap_chan *chan;
> > +       struct l2cap_chan *chan, *tmp;
> >
> >        BT_DBG("conn %p", conn);
> >
> > -       rcu_read_lock();
> > +       mutex_lock(&conn->chan_lock);
> >
> > -       list_for_each_entry_rcu(chan, &conn->chan_l, list) {
> > +       list_for_each_entry_safe(chan, tmp, &conn->chan_l, list) {
> >                struct sock *sk = chan->sk;
> >
> >                bh_lock_sock(sk);
> > @@ -829,7 +829,7 @@ static void l2cap_conn_start(struct l2cap_conn *conn)
> >                bh_unlock_sock(sk);
> >        }
> >
> > -       rcu_read_unlock();
> > +       mutex_unlock(&conn->chan_lock);
> >  }
> >
> >  /* Find socket with cid and source bdaddr.
> > @@ -1009,6 +1009,8 @@ static void l2cap_conn_del(struct hci_conn *hcon, int err)
> >
> >        kfree_skb(conn->rx_skb);
> >
> > +       mutex_lock(&conn->chan_lock);
> > +
> >        /* Kill channels */
> >        list_for_each_entry_safe(chan, l, &conn->chan_l, list) {
> >                sk = chan->sk;
> > @@ -1018,6 +1020,8 @@ static void l2cap_conn_del(struct hci_conn *hcon, int err)
> >                chan->ops->close(chan->data);
> >        }
> >
> > +       mutex_unlock(&conn->chan_lock);
> > +
> >        hci_chan_del(conn->hchan);
> >
> >        if (conn->info_state & L2CAP_INFO_FEAT_MASK_REQ_SENT)
> > @@ -1075,6 +1079,7 @@ static struct l2cap_conn *l2cap_conn_add(struct hci_conn *hcon, u8 status)
> >        conn->feat_mask = 0;
> >
> >        spin_lock_init(&conn->lock);
> > +       mutex_init(&conn->chan_lock);
> >
> >        INIT_LIST_HEAD(&conn->chan_l);
> >
> 
> I was under the impression you'd remove RCU for conn->chan_l
> completely. You're adding a lock only in the updaters? If so, please
> take a look at commit 3d57dc680 which shows all changes from mutex to
> RCU. I don't think just adding a lock/unlock in l2cap_conn_start and
> l2cap_conn_del will be enough. l2cap_chan_add seems to be called from
> other contexts and it does a list_add_rcu(). Have you thought of that?

I am adding lock to updaters and to the places we need to sleep and
rcu_read_lock cannot be used. This patch adds locks to updaters and
following patches cover other places. Maybe I need to split them better.

Best regards 
Andrei Emeltchenko