Re: [PATCH] Bluetooth: silence lockdep warning

On Sat, Jan 21, 2012 at 8:29 PM, Marcel Holtmann [off-list ref] wrot=
e:

Hi Octavian,

quoted

quoted

quoted

+void bt_sock_reclassify_lock(struct sock *sk, int proto)
=A0{
- =A0 =A0 struct sock *sk =3D sock->sk;
-
=A0 =A0 =A0 if (!sk)
=A0 =A0 =A0 =A0 =A0 =A0 =A0 return;

Why are we keeping the !sk check here if we already hand in the sk. It
is most likely checked by the caller already.

In rfcomm_sock_create (called from bt_sock_create) sock->sk can be set
to NULL so I think we should keep the check.

it has been too long since I looked at this part of the code. You need
to walk me through it why this is still true.

Hi Marcel,

In bt_sock_create we have:

        if (bt_proto[proto] && try_module_get(bt_proto[proto]->owner)) {
        	err =3D bt_proto[proto]->create(net, sock, proto, kern);
                bt_sock_reclassify_lock(sock->sk, proto);

and create can be rfcomm_sock_create where we have

        sk =3D rfcomm_sock_alloc(net, sock, protocol, GFP_ATOMIC);
        if (!sk)
                return -ENOMEM;

So after calling ->create() sock->sk can be NULL and thus we can call
bt_sock_reclassify with a NULL parameter.

Does this make sense?

Thanks,
Tavi