linux-kernel - Re: [RFC v1 0/3] Address potential user-after-free on module unload

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 6 Feb 2019 09:30:29 -0800
From:   Dmitry Torokhov <dmitry.torokhov@...il.com>
To:     Greg KH <gregkh@...uxfoundation.org>
Cc:     Sven Van Asbroeck <thesven73@...il.com>,
        Kees Cook <keescook@...omium.org>, Tejun Heo <tj@...nel.org>,
        Lai Jiangshan <jiangshanlai@...il.com>,
        LKML <linux-kernel@...r.kernel.org>,
        Sebastian Reichel <sre@...nel.org>
Subject: Re: [RFC v1 0/3] Address potential user-after-free on module unload

On Wed, Feb 6, 2019 at 8:47 AM Greg KH <gregkh@...uxfoundation.org> wrote:
>
> On Tue, Feb 05, 2019 at 02:12:31PM -0500, Sven Van Asbroeck wrote:
> > On Tue, Feb 5, 2019 at 1:43 PM Greg KH <gregkh@...uxfoundation.org> wrote:
> > >
> > >
> > > It really should happen when the device is removed (if it is a driver
> > > that binds to a device.)
> >
> > Absolutely. That's why I'm advocating adding a devm_init_work(),
> > which will take care of this automatically.
> >
> > But it's of course not universally applicable. Not all drivers use devm.
>
> Ick, no, watch out for devm() calls.  Odds are this is _NOT_ what you
> want to do for a device.  Remember when devm calls get freed (hint, not
> at driver unbind/unload, but at device structure removal.


??? We unwind devm on probe() failure and after remove() is called.
The device can live on.

>
>
> By creating a work queue, you are suddenly tying module code to a device
> memory structure lifespan, both of which are totally independant.
>
> It's the same issue with the devm irq call, that has been nothing but a
> nightmare as everyone gets it wrong.  Try to learn from our past
> mistakes please :)


Yeah. But devm irq gave most trouble because we did not have enough
devm APIs so we often ended up with mixed devm/non-devm usage and that
is what was causing most of the issues. If we can switch everything to
devm then devm irq is not that troublesome.

I have 2+ drivers that currently use devm_add_action_or_reset() to
install action canceling work, they could be switched to
devm_init_work().

Thanks,

-- 
Dmitry