Return-Path: MIME-Version: 1.0 In-Reply-To: References: <1331214279-27521-1-git-send-email-arik@wizery.com> <1331214279-27521-2-git-send-email-arik@wizery.com> From: Arik Nemtsov Date: Wed, 14 Mar 2012 14:11:20 +0200 Message-ID: Subject: Re: [PATCH 2/2] Revert "core: Add provision for terminating an ATT connection" To: Anderson Lizardo Cc: linux-bluetooth@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-bluetooth-owner@vger.kernel.org List-ID: On Wed, Mar 14, 2012 at 14:08, Anderson Lizardo wrote: > Hi Arik, > > On Thu, Mar 8, 2012 at 9:44 AM, Arik Nemtsov wrote: >> This reverts commit f89a77478af78d41c80ab7605662382b9e4e1c36. >> >> This is not needed and actually introduces a bug. When the "Disconnect" >> API of device is called device->attrib is unref-ed via a watch set on >> G_IO_HUP. The channel is shutdown when the last reference is removed. >> >> The code introduced here shuts down the channel and prevents the watch >> from getting called. This means we leak a reference to device->attrib. >> This can cause a number of bad things. For example, if the device is >> temporary, it will never be freed, and we won't be able to pair to it >> again. >> --- >> ?src/device.c | ? ?8 -------- >> ?1 files changed, 0 insertions(+), 8 deletions(-) > > While I agree with your explanation, we still need a way to for link > disconnection. There is a Disconnect command on mgmt API that may be > used for this (I suppose there is a similar one for hciops). This way > the HUP watches will be called because disconnection will be initiated > by the kernel. > The current disconnect command (exposed via D-Bus as Device.Disconnect()) works just fine for me. Arik