Re: [PATCH] Fix open/close race in saa7134

[Marcin Slusarz]

On Mon, Jun 23, 2008 at 11:54:32AM -0700, Arjan van de Ven wrote:
> On Mon, 23 Jun 2008 20:49:42 +0200
> Marcin Slusarz <marcin.slusarz@xxxxxxxxx> wrote:
> 
> > On Sun, Jun 22, 2008 at 10:58:32AM -0700, Arjan van de Ven wrote:
> > > On Sun, 22 Jun 2008 19:33:37 +0200
> > > Marcin Slusarz <marcin.slusarz@xxxxxxxxx> wrote:
> > > 
> > > > On Sun, Jun 22, 2008 at 10:05:07AM -0700, Arjan van de Ven wrote:
> > > > > 
> > > > > From: Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>
> > > > > Date: Sun, 22 Jun 2008 10:03:02 -0700
> > > > > Subject: [PATCH] Fix open/close race in saa7134
> > > > > 
> > > > > The saa7134 driver uses a (non-atomic) variable in an attempt to
> > > > > only allow one opener of the device (how it deals with sending
> > > > > the fd over unix sockets I don't know).
> > > > > 
> > > > > Unfortunately, the release function first decrements this
> > > > > variable, and THEN goes on to disable more of the device. This
> > > > > allows for a race where another opener of the device comes in
> > > > > after the decrement of the variable, configures the hardware
> > > > > just to then see the hardware be disabled by the rest of the
> > > > > release function.
> > > > 
> > > > Simplier fix:
> > > > http://lkml.org/lkml/2008/6/9/308
> > > > But I don't remember whether it was applied or not...
> > > 
> > > 
> > > the patch might be simpler, but it's not fully correct...
> > > the decrement is non-atomic and not protected by any lock
> > > whatsoever.
> > 
> > It's not atomic, but it don't have to, because there's only one
> > thread which "owns" the device.
> 
> this is not correct!
> 
> this is a close->open race, not an open->close!
> which means the guy who's closing it and the guy who's then opening it
> again do not have to be the same guy

If dev->empress_users could be > 1 then ok - it could break, but it can only be 1 or 0.
If it's 1 you won't open the device. If it's 0 you won't reach ts_close.

If you still see the race, please show me the sequence, because I don't
(of course when decrementing is the last operation of ts_close).

Marcin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/