Re: [PATCH v2] chardev: fix error handling in cdev_device_add()

[Yang Yingliang]

Hi Greg,

On 2022/10/25 21:37, Greg KH wrote:
> On Tue, Oct 25, 2022 at 09:20:12PM +0800, Yang Yingliang wrote:
> > Hi, Greg
> >
> > On 2022/10/25 19:50, Greg KH wrote:
> > > On Tue, Oct 25, 2022 at 07:39:57PM +0800, Yang Yingliang wrote:
> > > > While doing fault injection test, I got the following report:
> > > >
> > > > ------------[ cut here ]------------
> > > > kobject: '(null)' (0000000039956980): is not initialized, yet kobject_put() is being called.
> > > > WARNING: CPU: 3 PID: 6306 at kobject_put+0x23d/0x4e0
> > > > CPU: 3 PID: 6306 Comm: 283 Tainted: G        W          6.1.0-rc2-00005-g307c1086d7c9 #1253
> > > > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1ubuntu1.1 04/01/2014
> > > > RIP: 0010:kobject_put+0x23d/0x4e0
> > > > Call Trace:
> > > >    <TASK>
> > > >    cdev_device_add+0x15e/0x1b0
> > > >    __iio_device_register+0x13b4/0x1af0 [industrialio]
> > > >    __devm_iio_device_register+0x22/0x90 [industrialio]
> > > >    max517_probe+0x3d8/0x6b4 [max517]
> > > >    i2c_device_probe+0xa81/0xc00
> > > >
> > > > When device_add() is injected fault and returns error, if dev->devt is not set,
> > > > cdev_add() is not called, cdev_del() is not needed. Fix this by checking dev->devt
> > > > in error path.
> > > Nit, please wrap your changelog text at 72 columns.
> > >
> > > > Fixes: 233ed09d7fda ("chardev: add helper function to register char devs with a struct device")
> > > > Signed-off-by: Yang Yingliang <yangyingliang@xxxxxxxxxx>
> > > > ---
> > > > v1 -> v2:
> > > >     Add information to update commit message.
> > > >     v1 link: https://lore.kernel.org/lkml/1959fa74-b06c-b8bc-d14f-b71e5c4290ee@xxxxxxxxxx/T/
> > > > ---
> > > >    fs/char_dev.c | 2 +-
> > > >    1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/fs/char_dev.c b/fs/char_dev.c
> > > > index ba0ded7842a7..3f667292608c 100644
> > > > --- a/fs/char_dev.c
> > > > +++ b/fs/char_dev.c
> > > > @@ -547,7 +547,7 @@ int cdev_device_add(struct cdev *cdev, struct device *dev)
> > > >    	}
> > > >    	rc = device_add(dev);
> > > > -	if (rc)
> > > > +	if (rc && dev->devt)
> > > No, this is a layering violation and one that you do not know is really
> > > going to be true or not.  the devt being present, or not, should not be
> > > an issue of if the device_add failed or not.  This isn't correct, sorry.
> > Do you mean it's not a bug or the warn can be ignored or it's bug in driver
> > ?
> > I see devt is checked before calling cdev_del() in cdev_device_del().
> Ah!  The core doesn't set devt, the caller has that set.  That makes
> more sense now, sorry for the confusion on my side.
>
> Yes, this looks correct, the diff didn't have the full context and I was
> confused.
>
> I'll go queue this up, very nice work.
>
> greg k-h
I didn't find this patch in your trees, does it been merged?

Thanks,
Yang
> .