put_device calls release function which do kfree() inside.
So following use of sch&cdev would cause use-after-free bugs.
Fix these by simply adjusting the position of put_device.
Fixes: 37db8985b211 ("s390/cio: add basic protected virtualization support")
Fixes: 74bd0d859dc3 ("s390/cio: fix unlocked access of online member")
Reported-by: Hulk Robot <[email protected]>
Signed-off-by: Qinglang Miao <[email protected]>
---
drivers/s390/cio/device.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/s390/cio/device.c b/drivers/s390/cio/device.c
index b29fe8d50..69492417b 100644
--- a/drivers/s390/cio/device.c
+++ b/drivers/s390/cio/device.c
@@ -1664,10 +1664,10 @@ void __init ccw_device_destroy_console(struct ccw_device *cdev)
struct io_subchannel_private *io_priv = to_io_private(sch);
set_io_private(sch, NULL);
- put_device(&sch->dev);
- put_device(&cdev->dev);
dma_free_coherent(&sch->dev, sizeof(*io_priv->dma_area),
io_priv->dma_area, io_priv->dma_area_dma);
+ put_device(&sch->dev);
+ put_device(&cdev->dev);
kfree(io_priv);
}
@@ -1774,8 +1774,8 @@ static int ccw_device_remove(struct device *dev)
ret, cdev->private->dev_id.ssid,
cdev->private->dev_id.devno);
/* Give up reference obtained in ccw_device_set_online(). */
- put_device(&cdev->dev);
spin_lock_irq(cdev->ccwlock);
+ put_device(&cdev->dev);
}
ccw_device_set_timeout(cdev, 0);
cdev->drv = NULL;
--
2.23.0
On Fri, 20 Nov 2020 15:48:49 +0800
Qinglang Miao <[email protected]> wrote:
> put_device calls release function which do kfree() inside.
> So following use of sch&cdev would cause use-after-free bugs.
>
> Fix these by simply adjusting the position of put_device.
>
> Fixes: 37db8985b211 ("s390/cio: add basic protected virtualization support")
> Fixes: 74bd0d859dc3 ("s390/cio: fix unlocked access of online member")
> Reported-by: Hulk Robot <[email protected]>
> Signed-off-by: Qinglang Miao <[email protected]>
> ---
> drivers/s390/cio/device.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/s390/cio/device.c b/drivers/s390/cio/device.c
> index b29fe8d50..69492417b 100644
> --- a/drivers/s390/cio/device.c
> +++ b/drivers/s390/cio/device.c
> @@ -1664,10 +1664,10 @@ void __init ccw_device_destroy_console(struct ccw_device *cdev)
> struct io_subchannel_private *io_priv = to_io_private(sch);
>
> set_io_private(sch, NULL);
> - put_device(&sch->dev);
> - put_device(&cdev->dev);
> dma_free_coherent(&sch->dev, sizeof(*io_priv->dma_area),
> io_priv->dma_area, io_priv->dma_area_dma);
> + put_device(&sch->dev);
> + put_device(&cdev->dev);
That change looks reasonable.
> kfree(io_priv);
> }
>
> @@ -1774,8 +1774,8 @@ static int ccw_device_remove(struct device *dev)
> ret, cdev->private->dev_id.ssid,
> cdev->private->dev_id.devno);
> /* Give up reference obtained in ccw_device_set_online(). */
> - put_device(&cdev->dev);
> spin_lock_irq(cdev->ccwlock);
> + put_device(&cdev->dev);
As the comment above states, the put_device() gives up the reference
obtained in ccw_device_set_online(). There's at least one more
reference remaining (held by the caller of the remove function). Moving
the put_device() does not fix anything here.
> }
> ccw_device_set_timeout(cdev, 0);
> cdev->drv = NULL;
?? 2020/11/20 15:55, Cornelia Huck ะด??:
> On Fri, 20 Nov 2020 15:48:49 +0800
> Qinglang Miao <[email protected]> wrote:
>
>> put_device calls release function which do kfree() inside.
>> So following use of sch&cdev would cause use-after-free bugs.
>>
>> Fix these by simply adjusting the position of put_device.
>>
>> Fixes: 37db8985b211 ("s390/cio: add basic protected virtualization support")
>> Fixes: 74bd0d859dc3 ("s390/cio: fix unlocked access of online member")
>> Reported-by: Hulk Robot <[email protected]>
>> Signed-off-by: Qinglang Miao <[email protected]>
>> ---
>> drivers/s390/cio/device.c | 6 +++---
>> 1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/s390/cio/device.c b/drivers/s390/cio/device.c
>> index b29fe8d50..69492417b 100644
>> --- a/drivers/s390/cio/device.c
>> +++ b/drivers/s390/cio/device.c
>> @@ -1664,10 +1664,10 @@ void __init ccw_device_destroy_console(struct ccw_device *cdev)
>> struct io_subchannel_private *io_priv = to_io_private(sch);
>>
>> set_io_private(sch, NULL);
>> - put_device(&sch->dev);
>> - put_device(&cdev->dev);
>> dma_free_coherent(&sch->dev, sizeof(*io_priv->dma_area),
>> io_priv->dma_area, io_priv->dma_area_dma);
>> + put_device(&sch->dev);
>> + put_device(&cdev->dev);
>
> That change looks reasonable.
>
>> kfree(io_priv);
>> }
>>
>> @@ -1774,8 +1774,8 @@ static int ccw_device_remove(struct device *dev)
>> ret, cdev->private->dev_id.ssid,
>> cdev->private->dev_id.devno);
>> /* Give up reference obtained in ccw_device_set_online(). */
>> - put_device(&cdev->dev);
>> spin_lock_irq(cdev->ccwlock);
>> + put_device(&cdev->dev);
>
> As the comment above states, the put_device() gives up the reference
> obtained in ccw_device_set_online(). There's at least one more
> reference remaining (held by the caller of the remove function). Moving
> the put_device() does not fix anything here.
Hi, Cornelia
Sorry for the delayed reply.
Your suggestion is reasonable, there is a mistake in this patch for I
didn't notice that there would be at least one more reference remaining.
So I sent a new patch only to move put_device after dma_free_coherent. I
put the link as below:
https://lore.kernel.org/lkml/[email protected]/
Thanks!
>
>> }
>> ccw_device_set_timeout(cdev, 0);
>> cdev->drv = NULL;
>
> .
>