On Tue, Jan 13, 2009 at 11:12 AM, Marcel Holtmann
<marcel@holtmann.org> wrote:
Hi Nick,
> >> Here is set of unclean patches for discussion. These are still untested
> >> and contain issues, but I would like to know if we should continue with
> >> these or change something radically. Also commit messages are somewhat
> >> misleading in some places.
> >>
> >> Basically idea is to implement connection rejection support for l2cap
> >> and rfcomm sockets. IOW possibility to check initiator bd-address and
> >> ask for authorization before accepting connection.
> >>
> >> Other goal is to fix encryption and authentication levels to fulfill 2.1
> >> requirements.
> >
> > so I pushed my re-worked and pending patches to bluetooth-testing.git
> > now. This adds support for BT_DEFER_SETUP and BT_SECURITY (only the
> > logic and not the actual socket option).
>
> Ville: thanks for the patches, pausing RFCOMM while encryption is
> disabled is a nice fix.
>
> I imagine that if encryption is not quickly re-established we need to
> drop the RFCOMM link rather than leaving it paused though.
>
> We will do some testing of the current rfcomm-pause patches.
so I pushed another set of patches to the bluetooth-testing.git tree and
it should include the full BT_SECURITY implemention, BT_DEFER_SETUP for
RFCOMM and SCO rejection if no listen socket is present.
It currently only drops the connection is encryption is disabled when
using BT_SECURITY_HIGH. That is only used for SAP anyway. For all other
profiles we use BT_SECURITY_MEDIUM.
I updated our build with the patches and after picking up 6e26576c (Pause RFCOMM TX when encryption drops) and fixes upto f32ef1836 (Enforce authentication before encryption) I see a couple of problems::
a) I see that in rfcomm/core.c in function rfcomm_security_cfm: when the remote side has dropped the encryption for the role change, we set the RFCOMM_SEC_PENDING bit but we don't set RFCOMM_AUTH_PENDING as suggested in Ville's original patch and so when encryption is re-established we dont get past the - test_and_clear_bit (RFCOMM_AUTH_PENDING) check in the function.
b) I also see that we are not clearing the timer and hence after RFCOMM_AUTH_TIMEOUT period expires we bring down the RFCOMM connection even though the encryption has been established.
Sorry, I don't have a patch ready as yet and hence the description above to check if you have encountered the same.
Thanks
Jaikumar
Regards
Marcel