MIME-Version: 1.0
In-Reply-To: <CAG6hwVPvRQw64A=-+J0J8ioru8-fze3NuEEA1z2+c_skk5tAdA@mail.gmail.com>
References: <1367548442-8229-1-git-send-email-thomas@cozybit.com>
 <1367548442-8229-7-git-send-email-thomas@cozybit.com> <1367934155.8328.21.camel@jlt4.sipsolutions.net>
 <CAG6hwVPah-8FjH=ntFPzxMzXi5ePcmG-UEhdsvEgHkJVE9khYQ@mail.gmail.com>
 <1369427550.13623.3.camel@johannes> <CAG6hwVPvRQw64A=-+J0J8ioru8-fze3NuEEA1z2+c_skk5tAdA@mail.gmail.com>
From: Thomas Pedersen <thomas@cozybit.com>
Date: Wed, 29 May 2013 09:11:39 -0700
Message-ID: <CAG6hwVNNBOJDUCNzmocfN+SsySp4+U8Q3aAET=_7NRj8X6nGsw@mail.gmail.com> (sfid-20130529_181206_774436_F14197CB)
Subject: Re: [RFC 06/12] mac80211: notify bridge when leaving mesh
To: Johannes Berg <johannes@sipsolutions.net>
Cc: linux-wirelss <linux-wireless@vger.kernel.org>,
	Bob Copeland <me@bobcopeland.com>,
	open80211s <devel@lists.open80211s.org>
Content-Type: text/plain; charset=ISO-8859-1
Sender: linux-wireless-owner@vger.kernel.org

On Tue, May 28, 2013 at 10:01 AM, Thomas Pedersen <thomas@cozybit.com> wrote:
> On Fri, May 24, 2013 at 1:32 PM, Johannes Berg
> <johannes@sipsolutions.net> wrote:
>> On Tue, 2013-05-21 at 18:09 -0700, Thomas Pedersen wrote:
>>> Hey, Sorry for the late response.
>>
>> Me too :)
>>
>>> >>       netif_carrier_off(sdata->dev);
>>> >> +     if (sdata->dev->priv_flags & IFF_BRIDGE_PORT)
>>> >> +             /* stop bridge transmissions */
>>> >> +             call_netdevice_notifiers(NETDEV_CHANGE, sdata->dev);
>>> >
>>> > Err, this seems like a really bad hack? I don't really think drivers
>>> > should call that?
>>>
>>> Why not? We're just notifying the bridge interface that this port has
>>> gone down, to avoid dereferencing a null pointer (on bridge flood
>>> traffic) after the mesh_bss has been removed.
>>>
>>> Is cfg80211_leave_mesh() an acceptable location for this fix?
>>
>> I'm not really convinced it's an acceptable fix in itself? Why should we
>> ever have to call netdev notifiers, that seems odd to me. Also why would
>> that prevent a crash? Wouldn't we just drop packets for the mesh if we
>> aren't joined in a mesh any more? Or something like that  Accessing the
>> priv flags and then calling the netdev notifiers seems really strange to
>> me. If this was necessary, then wouldn't netif_carrier_off() do it
>> internally?
>
> OK I'll take a look at this and figure out a cleaner solution then. Thanks.

After reverting this patch I can no longer trigger the crash it was
supposed to fix. Either something changed in the rebase, or it was
never needed. Will drop from the resubmission for now.

Thanks,
--
Thomas