diff mbox series

[v3,2/2] kunit: avoid memory leak on device register error

Message ID 20240418210236.194190-3-wander@redhat.com
State New
Headers show
Series kunit: fix minor error path mistakes | expand

Commit Message

Wander Lairson Costa April 18, 2024, 9:02 p.m. UTC
If the device register fails, free the allocated memory before
returning.

Signed-off-by: Wander Lairson Costa <wander@redhat.com>
Fixes: d03c720e03bd ("kunit: Add APIs for managing devices")
---
 lib/kunit/device.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

Comments

Wander Lairson Costa April 19, 2024, 12:30 p.m. UTC | #1
On Fri, Apr 19, 2024 at 1:59 AM David Gow <davidgow@google.com> wrote:
>
> On Fri, 19 Apr 2024 at 05:02, Wander Lairson Costa <wander@redhat.com> wrote:
> >
> > If the device register fails, free the allocated memory before
> > returning.
> >
> > Signed-off-by: Wander Lairson Costa <wander@redhat.com>
> > Fixes: d03c720e03bd ("kunit: Add APIs for managing devices")
> > ---
>
> Thanks.
>
> I'm not sure this is correct, though... Shouldn't put_device() free this for us?
>
> The documentation for device_register() says to never free a device
> after device_register() has been called, even if it fails:
> https://docs.kernel.org/driver-api/infrastructure.html#c.device_register
>
> Or am I missing something?
>

I am not freeing the device object passed to device_register, but its
parent structure.

As a side note, the behavior of device_register() seems
counterintuitive and error-prone, IMO. If the function returns an
error, it should ensure it leaks no resource and shouldn't require the
caller to do any cleanup.

> Cheers,
> -- David
>
>
> >  lib/kunit/device.c | 11 ++++++-----
> >  1 file changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/lib/kunit/device.c b/lib/kunit/device.c
> > index 25c81ed465fb..bc2e2032e505 100644
> > --- a/lib/kunit/device.c
> > +++ b/lib/kunit/device.c
> > @@ -119,10 +119,8 @@ static struct kunit_device *kunit_device_register_internal(struct kunit *test,
> >         kunit_dev->owner = test;
> >
> >         err = dev_set_name(&kunit_dev->dev, "%s.%s", test->name, name);
> > -       if (err) {
> > -               kfree(kunit_dev);
> > -               return ERR_PTR(err);
> > -       }
> > +       if (err)
> > +               goto error;
> >
> >         kunit_dev->dev.release = kunit_device_release;
> >         kunit_dev->dev.bus = &kunit_bus_type;
> > @@ -131,7 +129,7 @@ static struct kunit_device *kunit_device_register_internal(struct kunit *test,
> >         err = device_register(&kunit_dev->dev);
> >         if (err) {
> >                 put_device(&kunit_dev->dev);
> > -               return ERR_PTR(err);
> > +               goto error;
> >         }
> >
> >         kunit_dev->dev.dma_mask = &kunit_dev->dev.coherent_dma_mask;
> > @@ -140,6 +138,9 @@ static struct kunit_device *kunit_device_register_internal(struct kunit *test,
> >         kunit_add_action(test, device_unregister_wrapper, &kunit_dev->dev);
> >
> >         return kunit_dev;
> > +error:
> > +       kfree(kunit_dev);
> > +       return ERR_PTR(err);
> >  }
> >
> >  /*
> > --
> > 2.44.0
> >
Greg KH April 19, 2024, 1:59 p.m. UTC | #2
On Fri, Apr 19, 2024 at 09:30:06AM -0300, Wander Lairson Costa wrote:
> As a side note, the behavior of device_register() seems
> counterintuitive and error-prone, IMO. If the function returns an
> error, it should ensure it leaks no resource and shouldn't require the
> caller to do any cleanup.

I too want a pony, but that's not the way the code works here, sorry.
It's always been like this, and has always been a problem, but last time
I looked, there was no way to really fix this.  That's why we document
it a lot to make sure people don't get the error paths wrong here.  I
know it's a pain :(

sorry,

greg k-h
diff mbox series

Patch

diff --git a/lib/kunit/device.c b/lib/kunit/device.c
index 25c81ed465fb..bc2e2032e505 100644
--- a/lib/kunit/device.c
+++ b/lib/kunit/device.c
@@ -119,10 +119,8 @@  static struct kunit_device *kunit_device_register_internal(struct kunit *test,
 	kunit_dev->owner = test;
 
 	err = dev_set_name(&kunit_dev->dev, "%s.%s", test->name, name);
-	if (err) {
-		kfree(kunit_dev);
-		return ERR_PTR(err);
-	}
+	if (err)
+		goto error;
 
 	kunit_dev->dev.release = kunit_device_release;
 	kunit_dev->dev.bus = &kunit_bus_type;
@@ -131,7 +129,7 @@  static struct kunit_device *kunit_device_register_internal(struct kunit *test,
 	err = device_register(&kunit_dev->dev);
 	if (err) {
 		put_device(&kunit_dev->dev);
-		return ERR_PTR(err);
+		goto error;
 	}
 
 	kunit_dev->dev.dma_mask = &kunit_dev->dev.coherent_dma_mask;
@@ -140,6 +138,9 @@  static struct kunit_device *kunit_device_register_internal(struct kunit *test,
 	kunit_add_action(test, device_unregister_wrapper, &kunit_dev->dev);
 
 	return kunit_dev;
+error:
+	kfree(kunit_dev);
+	return ERR_PTR(err);
 }
 
 /*