Input: synaptics-rmi4 - Really fix attn_data use-after-free
diff mbox series

Message ID 20200427145537.1.Ic8f898e0147beeee2c005ee7b20f1aebdef1e7eb@changeid
State Accepted
Commit d5a5e5b5fa7b86c05bf073acc0ba98fa280174ec
Headers show
Series
  • Input: synaptics-rmi4 - Really fix attn_data use-after-free
Related show

Commit Message

Evan Green April 27, 2020, 9:55 p.m. UTC
Fix a use-after-free noticed by running with KASAN enabled. If
rmi_irq_fn() is run twice in a row, then rmi_f11_attention() (among
others) will end up reading from drvdata->attn_data.data, which was
freed and left dangling in rmi_irq_fn().

Commit 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by
KASAN") correctly identified and analyzed this bug. However the attempted
fix only NULLed out a local variable, missing the fact that
drvdata->attn_data is a struct, not a pointer.

NULL out the correct pointer in the driver data to prevent the attention
functions from copying from it.

Fixes: 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by KASAN")
Fixes: b908d3cd812a ("Input: synaptics-rmi4 - allow to add attention data")

Signed-off-by: Evan Green <evgreen@chromium.org>

Cc: stable@vger.kernel.org
Cc: Nick Desaulniers <nick.desaulniers@gmail.com>
---

 drivers/input/rmi4/rmi_driver.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Dmitry Torokhov April 28, 2020, 1:11 a.m. UTC | #1
On Mon, Apr 27, 2020 at 02:55:48PM -0700, Evan Green wrote:
> Fix a use-after-free noticed by running with KASAN enabled. If
> rmi_irq_fn() is run twice in a row, then rmi_f11_attention() (among
> others) will end up reading from drvdata->attn_data.data, which was
> freed and left dangling in rmi_irq_fn().
> 
> Commit 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by
> KASAN") correctly identified and analyzed this bug. However the attempted
> fix only NULLed out a local variable, missing the fact that
> drvdata->attn_data is a struct, not a pointer.
> 
> NULL out the correct pointer in the driver data to prevent the attention
> functions from copying from it.
> 
> Fixes: 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by KASAN")
> Fixes: b908d3cd812a ("Input: synaptics-rmi4 - allow to add attention data")
> 
> Signed-off-by: Evan Green <evgreen@chromium.org>
> 
> Cc: stable@vger.kernel.org
> Cc: Nick Desaulniers <nick.desaulniers@gmail.com>

Ugh, this is all kind of ugly, but I guess that's what we have now.
Applied, thank you.

> ---
> 
>  drivers/input/rmi4/rmi_driver.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/input/rmi4/rmi_driver.c b/drivers/input/rmi4/rmi_driver.c
> index 190b9974526bb..c18e1a25bca6e 100644
> --- a/drivers/input/rmi4/rmi_driver.c
> +++ b/drivers/input/rmi4/rmi_driver.c
> @@ -205,7 +205,7 @@ static irqreturn_t rmi_irq_fn(int irq, void *dev_id)
>  
>  	if (count) {
>  		kfree(attn_data.data);
> -		attn_data.data = NULL;
> +		drvdata->attn_data.data = NULL;
>  	}
>  
>  	if (!kfifo_is_empty(&drvdata->attn_fifo))
> -- 
> 2.24.1
>
Evan Green April 28, 2020, 3:53 p.m. UTC | #2
On Mon, Apr 27, 2020 at 6:11 PM Dmitry Torokhov
<dmitry.torokhov@gmail.com> wrote:
>
> On Mon, Apr 27, 2020 at 02:55:48PM -0700, Evan Green wrote:
> > Fix a use-after-free noticed by running with KASAN enabled. If
> > rmi_irq_fn() is run twice in a row, then rmi_f11_attention() (among
> > others) will end up reading from drvdata->attn_data.data, which was
> > freed and left dangling in rmi_irq_fn().
> >
> > Commit 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by
> > KASAN") correctly identified and analyzed this bug. However the attempted
> > fix only NULLed out a local variable, missing the fact that
> > drvdata->attn_data is a struct, not a pointer.
> >
> > NULL out the correct pointer in the driver data to prevent the attention
> > functions from copying from it.
> >
> > Fixes: 55edde9fff1a ("Input: synaptics-rmi4 - prevent UAF reported by KASAN")
> > Fixes: b908d3cd812a ("Input: synaptics-rmi4 - allow to add attention data")
> >
> > Signed-off-by: Evan Green <evgreen@chromium.org>
> >
> > Cc: stable@vger.kernel.org
> > Cc: Nick Desaulniers <nick.desaulniers@gmail.com>
>
> Ugh, this is all kind of ugly, but I guess that's what we have now.
> Applied, thank you.

Thanks Dmitry. There are other bits that sketch me out in here as
well, but I didn't get a chance to really figure out if they're a
problem. We call rmi_process_interrupt_requests(), which results in
reads from that same attn_data.data pointer, in a few different
places. Some of those calls are outside the irq handling path, like
the in rmi_enable_irq() and rmi_enable_sensor(). Can they race with
the irq handling path? (Meaning they'd be doing those attn_data.data
reads as rmi_irq_fn() is kfreeing the data?) There are a smattering of
mutexes around, but I'm not sure if they're trying to protect this.

If I can find some time I'll try to submit a patch. Anyone is welcome
to beat me to it though.
-Evan

Patch
diff mbox series

diff --git a/drivers/input/rmi4/rmi_driver.c b/drivers/input/rmi4/rmi_driver.c
index 190b9974526bb..c18e1a25bca6e 100644
--- a/drivers/input/rmi4/rmi_driver.c
+++ b/drivers/input/rmi4/rmi_driver.c
@@ -205,7 +205,7 @@  static irqreturn_t rmi_irq_fn(int irq, void *dev_id)
 
 	if (count) {
 		kfree(attn_data.data);
-		attn_data.data = NULL;
+		drvdata->attn_data.data = NULL;
 	}
 
 	if (!kfifo_is_empty(&drvdata->attn_fifo))