[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20221024135640.73e5eddf@kernel.org>
Date: Mon, 24 Oct 2022 13:56:40 -0700
From: Jakub Kicinski <kuba@...nel.org>
To: Thomas Osterried <thomas@...erried.de>
Cc: Eric Dumazet <edumazet@...gle.com>,
"David S . Miller" <davem@...emloft.net>,
Paolo Abeni <pabeni@...hat.com>,
Bernard Pidoux <f6bvp@...e.fr>,
Duoming Zhou <duoming@....edu.cn>, netdev@...r.kernel.org,
linux-hams@...r.kernel.org
Subject: Re: [AX25] patch did not fix -- was: ax25: fix incorrect
dev_tracker usage
On Mon, 24 Oct 2022 20:00:00 +0200 Thomas Osterried wrote:
> II) What consequences has the tracker counter?
>
> As far as I can see by kernel messages, the netdev tracker forces the
> kernel to wait
> (on ifdown (i.e. ifconfig ax0 down)
> or rmmod (i.e. rmmod bpqether or rmmod ax25) )
> until all references to the network device are freed.
> If there's a bug (refcount > 0 or < 0), kernel obviously waits for ever.
Small correction here - the wait is when netdev is unregistered,
which often happens on rmmod, but also when user asks for a sw
netdev to be deleted, or HW device is removed, etc.
> III) Is it only to track sessions initiated from userspace?
>
> I think no.
Correct, the trackers track the references taken on the device.
Doesn't really matter if the reference is somehow tracable to
a user request or not.
Sorry for lack of input on the actual ax25 problem, I had looked
briefly in September and it wasn't obvious how things work :S
Powered by blists - more mailing lists