Received-SPF: pass (google.com: domain of linux-bluetooth-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <e17b2044-62cb-cbf7-64b8-c647aa7f6ee7@pabigot.com> <A6782BCD-0C05-40AD-B567-7A1E163184E0@holtmann.org>
In-Reply-To: <A6782BCD-0C05-40AD-B567-7A1E163184E0@holtmann.org>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Thu, 19 Aug 2021 10:51:21 -0700
Message-ID: <CABBYNZLg=zujCZbbhrZq4mGpNLTSuKK6BrN8b_8Sf-NrhmDbbw@mail.gmail.com>
Subject: Re: how to cleanly shutdown an application using HCI_CHANNEL_USER
To:     Marcel Holtmann <marcel@holtmann.org>
Cc:     "Peter A. Bigot" <pab@pabigot.com>,
        "linux-bluetooth@vger.kernel.org" <linux-bluetooth@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Marcel, Peter,

On Thu, Aug 19, 2021 at 8:04 AM Marcel Holtmann <marcel@holtmann.org> wrote:
>
> Hi Peter,
>
> > I'm using an AF_BLUETOOTH socket bound with HCI_CHANNEL_USER from a
> > user-mode application with cap_net_admin=ep.  As expected this
> > requires the device be down, and brings the device up automatically.
> >
> > When I close that socket and exit the application, the device appears
> > to remain up forever.  Which prevents me from re-starting the
> > application.
> >
> > I tried to issue HCIDEVDOWN before closing, but that produces EBADFD
> > because ioctls cannot be performed with HCI_CHANNEL_RAW.
> >
> > I can bring the interface down from within the application if, after
> > closing the socket, I wait a second or so, then create a new bound
> > HCI_CHANNEL_RAW socket and issue HCIDEVDOWN on it.
> >
> > Is there some other way to cleanly shut down an application that used
> > HCI_CHANNEL_USER so that the device is returned to down state on exit?
> > Or is this supposed to happen automatically (I see code that suggests
> > it should)?
> >
> > Kernel version is 5.11.0-7620-generic (System76), and I'm using go
> > 1.16, if that's relevant.
>
> I think you found a regression. Calling close() on the HCI User Channel should bring the device back down. This used to work correctly, can you please bisect which kernel patch broke this.
>
> Back in the days I added tools/userchan-tester, but it seems I never included enough test cases to catch this regression.
>
> Regards
>
> Marcel

It is worth checking if this is affected by the following fix:

https://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next.git/commit/?id=58ce6d5b271ab25fb2056f84a8e5546945eb5fc9

-- 
Luiz Augusto von Dentz