Hi Patrik Sorry for the late response. Lot's going on :)
On Fri, Jul 15, 2011 at 03:13:14PM +0300, [email protected] wrote: > From: Patrik Flykt <[email protected]> > > Hi, > > Here is a second attempt on session VPN support. It's not that much about > VPNs, it's almost all about selecting a better service if one is available > for the session. So if a session requests [wlan cellular] as bearer types > where initially only cellular is available, the idea is to switch over to > wlan when one becomes availble. The current code is happily camping in > cellular, it seems. Yes, switching to the best available service wan't yet implement :) > If the code is updated so that VPN has the highest priority of the bearers, > the VPN should be selected when it becomes available or at the start when > there is another service with a real network connection. Okay, sounds reasonable. > The patches do the following things: > - 1/4 tries to update the session to a "better" service if the service > selected was not first in the session sequence > - 2/4 Avoids VPNs until the session becomes online. Requiring sessions to > declare AllowedBearers to have non-VPN services in order to become connected > was the easiest way out of the selection process. Thus a session that wants > a VPN would need to specify AllowedBearers as [vpn, *]. > - 3/4 Sets VPNs to be sorted first > - 4/4 Sets session info to FALSE if the associated service got deleted. This > is a bug and can be send/applied independently > > This patch set will still send unnecessary updates about the '*' bearers, a > next step would be to eliminate those. > > Instead of selecting a generic "better" bearer, patch 1/4 can be changed to > react on VPNs only. > > I had some difficulty testing the update of AllowedBearers, something is not > happening or then it works ok but isn't verbal about it. I have started to add unit tests to get a more stable API there. You might want also to add some tests (if this is possible). > Last but not least: I don't see any evidence of services being shut down even > if the session selects another service. This might lead to a > session->vpn_service dependign on the intended behavior. Hmm, maybe we should soon merge the ref counting patches :) > As a next step I'm trying to get a better understanding how the Session API > behaves from a client point of viev. Great and I would really love to get unit tests. cheers, daniel _______________________________________________ connman mailing list [email protected] http://lists.connman.net/listinfo/connman
