If device_register() fails, both put_device() and kfree()
are called, ending with a double free of the scmi_dev.
Calling kfree() is needed only when a failure happens between the
allocation of the scmi_dev and its registration, so move it to
there and remove it from the error flow.
Fixes: 46edb8d1322c ("firmware: arm_scmi: provide the mandatory device release callback")
Signed-off-by: Wen Yang <[email protected]>
Cc: Sudeep Holla <[email protected]>
Cc: [email protected]
Cc: [email protected]
---
drivers/firmware/arm_scmi/bus.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/firmware/arm_scmi/bus.c b/drivers/firmware/arm_scmi/bus.c
index 92f843ea..7a30952 100644
--- a/drivers/firmware/arm_scmi/bus.c
+++ b/drivers/firmware/arm_scmi/bus.c
@@ -135,8 +135,10 @@ struct scmi_device *
return NULL;
id = ida_simple_get(&scmi_bus_id, 1, 0, GFP_KERNEL);
- if (id < 0)
- goto free_mem;
+ if (id < 0) {
+ kfree(scmi_dev);
+ return NULL;
+ }
scmi_dev->id = id;
scmi_dev->protocol_id = protocol;
@@ -154,8 +156,6 @@ struct scmi_device *
put_dev:
put_device(&scmi_dev->dev);
ida_simple_remove(&scmi_bus_id, id);
-free_mem:
- kfree(scmi_dev);
return NULL;
}
--
1.8.3.1
On Mon, Nov 25, 2019 at 11:54:09PM +0800, Wen Yang wrote:
> If device_register() fails, both put_device() and kfree()
> are called, ending with a double free of the scmi_dev.
>
Correct.
> Calling kfree() is needed only when a failure happens between the
> allocation of the scmi_dev and its registration, so move it to
> there and remove it from the error flow.
>
kstrdup_const can fail and in that case device is not yet registered,
so we need to free. Since device_register() calls put_device() on failure
too, I would just drop it as it's unnecessary, not sure why I have added
it in the first place. Can you re-spin the patch dropping put_device
and renaming put_dev label to something like free_const.
--
Regards,
Sudeep
On 2019/11/26 12:13 上午, Sudeep Holla wrote:
> On Mon, Nov 25, 2019 at 11:54:09PM +0800, Wen Yang wrote:
>> If device_register() fails, both put_device() and kfree()
>> are called, ending with a double free of the scmi_dev.
>>
>
> Correct.
>
>> Calling kfree() is needed only when a failure happens between the
>> allocation of the scmi_dev and its registration, so move it to
>> there and remove it from the error flow.
>>
>
> kstrdup_const can fail and in that case device is not yet registered,
> so we need to free. Since device_register() calls put_device() on failure
> too, I would just drop it as it's unnecessary, not sure why I have added
> it in the first place. Can you re-spin the patch dropping put_device
> and renaming put_dev label to something like free_const.
>
> --
> Regards,
> Sudeep
>
Hi Sudeep,
Thanks for your comments.
Let's check the code like this:
int device_register(struct device *dev)
{
device_initialize(dev); --> Initialize kobj-> kref to 1
return device_add(dev);
}
int device_add(struct device *dev)
{
...
dev = get_device(dev); --> kobj-> kref increases by 1
...
done:
put_device(dev); --> kobj-> kref decreases by 1 and is now 1
return error;
...
}
So we also need to call put_device (),
and the original patch should be fine.
Please kindly help to check again, thank you.
--
Regards,
Wen
On Tue, Nov 26, 2019 at 06:24:47PM +0800, Wen Yang wrote:
>
>
> On 2019/11/26 12:13 上午, Sudeep Holla wrote:
> > On Mon, Nov 25, 2019 at 11:54:09PM +0800, Wen Yang wrote:
> > > If device_register() fails, both put_device() and kfree()
> > > are called, ending with a double free of the scmi_dev.
> > >
> >
> > Correct.
> >
> > > Calling kfree() is needed only when a failure happens between the
> > > allocation of the scmi_dev and its registration, so move it to
> > > there and remove it from the error flow.
> > >
> >
> > kstrdup_const can fail and in that case device is not yet registered,
> > so we need to free. Since device_register() calls put_device() on failure
> > too, I would just drop it as it's unnecessary, not sure why I have added
> > it in the first place. Can you re-spin the patch dropping put_device
> > and renaming put_dev label to something like free_const.
> >
Please ignore the above completely. I have made some changes locally and
got completely confused when I looked at your patch and compared with
the modified context locally.
>
> Hi Sudeep,
> Thanks for your comments.
> Let's check the code like this:
>
> int device_register(struct device *dev)
> {
> device_initialize(dev); --> Initialize kobj-> kref to 1
> return device_add(dev);
> }
>
> int device_add(struct device *dev)
> {
> ...
> dev = get_device(dev); --> kobj-> kref increases by 1
> ...
> done:
> put_device(dev); --> kobj-> kref decreases by 1 and is now 1
> return error;
> ...
> }
>
> So we also need to call put_device (),
> and the original patch should be fine.
> Please kindly help to check again, thank you.
>
You are right, sorry for the confusion. I will apply your original patch.
Thanks again for the patch.
--
Regards,
Sudeep