[PATCH] Bluetooth: Use non locking get_chan and explicit lock

[PATCH] Bluetooth: Use non locking get_chan and explicit lock

[Emeltchenko Andrei]

From: Andrei Emeltchenko <andrei.emeltchenko@intel.com>

For receiving ACL packets use __l2cap_get_chan_by_scid which is not
locking sk and explicitly lock sk after checking that it is exist.
Code looks nicer since now it is surrounded by lock/release.

Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
---
 net/bluetooth/l2cap_core.c |    7 +++++--
 1 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
index d16ad49..53dbfd3 100644
--- a/net/bluetooth/l2cap_core.c
+++ b/net/bluetooth/l2cap_core.c
@@ -4243,7 +4243,7 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
 	u16 tx_seq;
 	int len;
 
-	chan = l2cap_get_chan_by_scid(conn, cid);
+	chan = __l2cap_get_chan_by_scid(conn, cid);
 	if (!chan) {
 		if (cid == L2CAP_CID_A2MP) {
 			chan = a2mp_channel_create(conn, skb);
@@ -4255,6 +4255,8 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
 	}
 
 	sk = chan->sk;
+	if (sk)
+		lock_sock(sk);
 
 	BT_DBG("chan %p, len %d", chan, skb->len);
 
@@ -4681,10 +4683,11 @@ int l2cap_recv_acldata(struct hci_conn *hcon, struct sk_buff *skb, u16 flags)
 			goto drop;
 		}
 
-		chan = l2cap_get_chan_by_scid(conn, cid);
+		chan = __l2cap_get_chan_by_scid(conn, cid);
 
 		if (chan && chan->sk) {
 			struct sock *sk = chan->sk;
+			lock_sock(sk);
 
 			if (chan->imtu < len - L2CAP_HDR_SIZE) {
 				BT_ERR("Frame exceeding recv MTU (len %d, "
-- 
1.7.4.1

Re: [PATCH] Bluetooth: Use non locking get_chan and explicit lock

[Marcel Holtmann]

Hi Andrei,

> For receiving ACL packets use __l2cap_get_chan_by_scid which is not
> locking sk and explicitly lock sk after checking that it is exist.
> Code looks nicer since now it is surrounded by lock/release.
> 
> Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> ---
>  net/bluetooth/l2cap_core.c |    7 +++++--
>  1 files changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> index d16ad49..53dbfd3 100644
> --- a/net/bluetooth/l2cap_core.c
> +++ b/net/bluetooth/l2cap_core.c
> @@ -4243,7 +4243,7 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
>  	u16 tx_seq;
>  	int len;
>  
> -	chan = l2cap_get_chan_by_scid(conn, cid);
> +	chan = __l2cap_get_chan_by_scid(conn, cid);
>  	if (!chan) {
>  		if (cid == L2CAP_CID_A2MP) {
>  			chan = a2mp_channel_create(conn, skb);
> @@ -4255,6 +4255,8 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
>  	}
>  
>  	sk = chan->sk;
> +	if (sk)
> +		lock_sock(sk);

this is the part I clearly do not like. This is pretty nasty conditional
locking. We need to figure out something better.

Regards

Marcel

Re: [PATCH] Bluetooth: Use non locking get_chan and explicit lock

[Emeltchenko Andrei]

Hi Marcel,

On Sat, Jan 21, 2012 at 05:41:34PM +0100, Marcel Holtmann wrote:
> Hi Andrei,
> 
> > For receiving ACL packets use __l2cap_get_chan_by_scid which is not
> > locking sk and explicitly lock sk after checking that it is exist.
> > Code looks nicer since now it is surrounded by lock/release.
> > 
> > Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> > ---
> >  net/bluetooth/l2cap_core.c |    7 +++++--
> >  1 files changed, 5 insertions(+), 2 deletions(-)
> > 
> > diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > index d16ad49..53dbfd3 100644
> > --- a/net/bluetooth/l2cap_core.c
> > +++ b/net/bluetooth/l2cap_core.c
> > @@ -4243,7 +4243,7 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
> >  	u16 tx_seq;
> >  	int len;
> >  
> > -	chan = l2cap_get_chan_by_scid(conn, cid);
> > +	chan = __l2cap_get_chan_by_scid(conn, cid);
> >  	if (!chan) {
> >  		if (cid == L2CAP_CID_A2MP) {
> >  			chan = a2mp_channel_create(conn, skb);
> > @@ -4255,6 +4255,8 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
> >  	}
> >  
> >  	sk = chan->sk;
> > +	if (sk)
> > +		lock_sock(sk);
> 
> this is the part I clearly do not like. This is pretty nasty conditional
> locking. We need to figure out something better.

Actually l2cap_data_channel function ends with:

        <------8<--------------------
        |  if (sk)
        |          release_sock(sk);
        |
        <------8<--------------------

So in _this_ case it looks consistent. But I agree that generally it does
not looks nice. The reason for that is l2cap_chan is locked with socket
lock. Maybe we could create mutex l2cap_lock and functions like
lock_chan(chan) / unlock_chan(chan) and change socket locks in l2cap_core
with these new locks?

Best regards 
Andrei Emeltchenko 

Re: [PATCH] Bluetooth: Use non locking get_chan and explicit lock

[Marcel Holtmann]

Hi Andrei,

> > > For receiving ACL packets use __l2cap_get_chan_by_scid which is not
> > > locking sk and explicitly lock sk after checking that it is exist.
> > > Code looks nicer since now it is surrounded by lock/release.
> > > 
> > > Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
> > > ---
> > >  net/bluetooth/l2cap_core.c |    7 +++++--
> > >  1 files changed, 5 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> > > index d16ad49..53dbfd3 100644
> > > --- a/net/bluetooth/l2cap_core.c
> > > +++ b/net/bluetooth/l2cap_core.c
> > > @@ -4243,7 +4243,7 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
> > >  	u16 tx_seq;
> > >  	int len;
> > >  
> > > -	chan = l2cap_get_chan_by_scid(conn, cid);
> > > +	chan = __l2cap_get_chan_by_scid(conn, cid);
> > >  	if (!chan) {
> > >  		if (cid == L2CAP_CID_A2MP) {
> > >  			chan = a2mp_channel_create(conn, skb);
> > > @@ -4255,6 +4255,8 @@ static inline int l2cap_data_channel(struct l2cap_conn *conn, u16 cid, struct sk
> > >  	}
> > >  
> > >  	sk = chan->sk;
> > > +	if (sk)
> > > +		lock_sock(sk);
> > 
> > this is the part I clearly do not like. This is pretty nasty conditional
> > locking. We need to figure out something better.
> 
> Actually l2cap_data_channel function ends with:
> 
>         <------8<--------------------
>         |  if (sk)
>         |          release_sock(sk);
>         |
>         <------8<--------------------

that one slipped past me :(

> So in _this_ case it looks consistent. But I agree that generally it does
> not looks nice. The reason for that is l2cap_chan is locked with socket
> lock. Maybe we could create mutex l2cap_lock and functions like
> lock_chan(chan) / unlock_chan(chan) and change socket locks in l2cap_core
> with these new locks?

I agree. Lets see what happens if we lock l2cap_chan with its own lock.

Regards

Marcel