Re: [RFC] Bluetooth: Adds unlink to chan ops

[Gustavo Padovan <gustavo@padovan.org>]

Hi Andrei,

* Andrei Emeltchenko <Andrei.Emeltchenko.news@gmail.com> [2012-05-24 11:18:23 +0300]:

> Hi Gustavo and Ulisses,
> 
> On Wed, May 23, 2012 at 07:03:21PM -0300, Ulisses Furquim wrote:
> > Hi Gustavo,
> > 
> > On Wed, May 23, 2012 at 5:31 PM, Gustavo Padovan <gustavo@padovan.org> wrote:
> > > Hi Andrei,
> > >
> > > * Andrei Emeltchenko <Andrei.Emeltchenko.news@gmail.com> [2012-05-23 11:05:49 +0300]:
> > >
> > >> From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> > >>
> > >> This helps to separate socket and socketless code. Now socket related
> > >> operations moved to l2cap_sock in unlink callback for channels with sk.
> > >>
> > >> Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> > >> Reported-by: Mat Martineau <mathewm@codeaurora.org>
> > >> ---
> > >>  include/net/bluetooth/l2cap.h |    1 +
> > >>  net/bluetooth/l2cap_core.c    |   20 ++------------------
> > >>  net/bluetooth/l2cap_sock.c    |   25 +++++++++++++++++++++++++
> > >>  3 files changed, 28 insertions(+), 18 deletions(-)
> > >>
> > >> diff --git a/include/net/bluetooth/l2cap.h b/include/net/bluetooth/l2cap.h
> > >> index 69ef077..7ed13e7 100644
> > >> --- a/include/net/bluetooth/l2cap.h
> > >> +++ b/include/net/bluetooth/l2cap.h
> > >> @@ -532,6 +532,7 @@ struct l2cap_ops {
> > >>       void                    (*state_change) (void *data, int state);
> > >>       struct sk_buff          *(*alloc_skb) (struct l2cap_chan *chan,
> > >>                                              unsigned long len, int nb);
> > >> +     void                    (*unlink) (struct l2cap_chan *chan, int err);
> > >>  };
> > >>
> > >>  struct l2cap_conn {
> > >> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > >> index c64da38..d2b16b5 100644
> > >> --- a/net/bluetooth/l2cap_core.c
> > >> +++ b/net/bluetooth/l2cap_core.c
> > >> @@ -520,9 +520,7 @@ void l2cap_chan_add(struct l2cap_conn *conn, struct l2cap_chan *chan)
> > >>
> > >>  static void l2cap_chan_del(struct l2cap_chan *chan, int err)
> > >>  {
> > >> -     struct sock *sk = chan->sk;
> > >>       struct l2cap_conn *conn = chan->conn;
> > >> -     struct sock *parent;
> > >>
> > >>       __clear_chan_timer(chan);
> > >>
> > >> @@ -541,22 +539,8 @@ static void l2cap_chan_del(struct l2cap_chan *chan, int err)
> > >>       if (chan->chan_type == L2CAP_CHAN_CONN_FIX_A2MP)
> > >>               goto clean;
> > >>
> > >> -     lock_sock(sk);
> > >> -
> > >> -     __l2cap_state_change(chan, BT_CLOSED);
> > >> -     sock_set_flag(sk, SOCK_ZAPPED);
> > >> -
> > >> -     if (err)
> > >> -             __l2cap_chan_set_err(chan, err);
> > >> -
> > >> -     parent = bt_sk(sk)->parent;
> > >> -     if (parent) {
> > >> -             bt_accept_unlink(sk);
> > >> -             parent->sk_data_ready(parent, 0);
> > >> -     } else
> > >> -             sk->sk_state_change(sk);
> > >> -
> > >> -     release_sock(sk);
> > >> +     if (chan->ops->unlink)
> > >> +             chan->ops->unlink(chan, err);
> > >>
> > >>       if (test_bit(CONF_NOT_COMPLETE, &chan->conf_state))
> > >>               return;
> > >> diff --git a/net/bluetooth/l2cap_sock.c b/net/bluetooth/l2cap_sock.c
> > >> index 3bb1611..d6f481b 100644
> > >> --- a/net/bluetooth/l2cap_sock.c
> > >> +++ b/net/bluetooth/l2cap_sock.c
> > >> @@ -955,6 +955,30 @@ static struct sk_buff *l2cap_sock_alloc_skb_cb(struct l2cap_chan *chan,
> > >>       return skb;
> > >>  }
> > >>
> > >> +static void l2cap_sock_unlink_cb(struct l2cap_chan *chan, int err)
> > >> +{
> > >> +     struct sock *sk = chan->sk;
> > >> +     struct sock *parent = bt_sk(sk)->parent;
> > >> +
> > >> +     lock_sock(sk);
> > >> +
> > >> +     chan->state = BT_CLOSED;
> > >> +     sk->sk_state = BT_CLOSED;
> > >> +     sock_set_flag(sk, SOCK_ZAPPED);
> > >
> > > I think we can improve this code and call it everytime we need to set
> > > SOCK_ZAPPED. We just need a bit more of logic here.
> > 
> > We might, if it's not a big of a change. Otherwise I think this kind
> > of change needs to be incremental.
> 
> I agree with Ulisses that it shall be incremental since it is not
> straightforward. I am thinking that  sock_set_flag(sk, SOCK_ZAPPED) might
> be put before switch in l2cap_chan_close and maybe further. Then we need
> to think about how to make it working for socketless chans...

Check my patch inside the RFC series, it can be done easily.

	Gustavo