[PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[Arvind Yadav]

if device_register() returned an error! Always use put_device()
to give up the reference initialized.

Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
---
change in v2:
        Fix use-after-free bug. move put_device() after cdev_del().

 drivers/mtd/ubi/vmt.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
index 3fd8d7f..93c6163 100644
--- a/drivers/mtd/ubi/vmt.c
+++ b/drivers/mtd/ubi/vmt.c
@@ -610,6 +610,7 @@ int ubi_add_volume(struct ubi_device *ubi, struct ubi_volume *vol)
 
 out_cdev:
 	cdev_del(&vol->cdev);
+	put_device(&vol->dev);
 	return err;
 }
 
-- 
1.9.1

Re: [PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[Richard Weinberger]

Am Donnerstag, 15. März 2018, 08:20:31 CET schrieb Arvind Yadav:
> if device_register() returned an error! Always use put_device()
> to give up the reference initialized.

Like DaveM said, there is no need to shout and use "!".
 
> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
> ---
> change in v2:
>         Fix use-after-free bug. move put_device() after cdev_del().
> 
>  drivers/mtd/ubi/vmt.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
> index 3fd8d7f..93c6163 100644
> --- a/drivers/mtd/ubi/vmt.c
> +++ b/drivers/mtd/ubi/vmt.c
> @@ -610,6 +610,7 @@ int ubi_add_volume(struct ubi_device *ubi, struct
> ubi_volume *vol)
> 
>  out_cdev:
>  	cdev_del(&vol->cdev);
> +	put_device(&vol->dev);
>  	return err;

The more I dig into device code, the more questions I have.
Why is cdev_del() not part of the release function?

Thanks,
//richard

Re: [PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[Arvind Yadav]

On Thursday 15 March 2018 01:25 PM, Richard Weinberger wrote:
> Am Donnerstag, 15. März 2018, 08:20:31 CET schrieb Arvind Yadav:
>> if device_register() returned an error! Always use put_device()
>> to give up the reference initialized.
> Like DaveM said, there is no need to shout and use "!".

I will fix this and send you update patch.
>   
>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
>> ---
>> change in v2:
>>          Fix use-after-free bug. move put_device() after cdev_del().
>>
>>   drivers/mtd/ubi/vmt.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
>> index 3fd8d7f..93c6163 100644
>> --- a/drivers/mtd/ubi/vmt.c
>> +++ b/drivers/mtd/ubi/vmt.c
>> @@ -610,6 +610,7 @@ int ubi_add_volume(struct ubi_device *ubi, struct
>> ubi_volume *vol)
>>
>>   out_cdev:
>>   	cdev_del(&vol->cdev);
>> +	put_device(&vol->dev);
>>   	return err;
> The more I dig into device code, the more questions I have.
> Why is cdev_del() not part of the release function?
>
> Thanks,
> //richard

Yes, It's should be a part release function.

~arvind

Re: [PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[arvindY]

On Thursday 15 March 2018 02:17 PM, Arvind Yadav wrote:
>
>
> On Thursday 15 March 2018 01:25 PM, Richard Weinberger wrote:
>> Am Donnerstag, 15. März 2018, 08:20:31 CET schrieb Arvind Yadav:
>>> if device_register() returned an error! Always use put_device()
>>> to give up the reference initialized.
>> Like DaveM said, there is no need to shout and use "!".
>
> I will fix this and send you update patch.
>>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
>>> ---
>>> change in v2:
>>>          Fix use-after-free bug. move put_device() after cdev_del().
>>>
>>>   drivers/mtd/ubi/vmt.c | 1 +
>>>   1 file changed, 1 insertion(+)
>>>
>>> diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
>>> index 3fd8d7f..93c6163 100644
>>> --- a/drivers/mtd/ubi/vmt.c
>>> +++ b/drivers/mtd/ubi/vmt.c
>>> @@ -610,6 +610,7 @@ int ubi_add_volume(struct ubi_device *ubi, struct
>>> ubi_volume *vol)
>>>
>>>   out_cdev:
>>>       cdev_del(&vol->cdev);
>>> +    put_device(&vol->dev);
>>>       return err;
>> The more I dig into device code, the more questions I have.
>> Why is cdev_del() not part of the release function?
>>
>> Thanks,
>> //richard
>
> Yes, It's should be a part release function.
>
> ~arvind

I was wrong, We can not add cdev_del() in release(vol_release)
function.
Function's ubi_create_volume and ubi_add_volume both are using
same release function to release a volume devices.
ubi_add_volume is registering character device for the volume.
So we will have to release character device here.

~arvind

Re: [PATCH 2/2 v2] mtd: ubi: use put_device() if device_register fail

[Richard Weinberger]

Arvind,

Am Donnerstag, 15. März 2018, 18:41:58 CET schrieb arvindY:
> On Thursday 15 March 2018 02:17 PM, Arvind Yadav wrote:
> > On Thursday 15 March 2018 01:25 PM, Richard Weinberger wrote:
> >> Am Donnerstag, 15. März 2018, 08:20:31 CET schrieb Arvind Yadav:
> >>> if device_register() returned an error! Always use put_device()
> >>> to give up the reference initialized.
> >> 
> >> Like DaveM said, there is no need to shout and use "!".
> > 
> > I will fix this and send you update patch.
> > 
> >>> Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
> >>> ---
> >>> 
> >>> change in v2:
> >>>          Fix use-after-free bug. move put_device() after cdev_del().
> >>>   
> >>>   drivers/mtd/ubi/vmt.c | 1 +
> >>>   1 file changed, 1 insertion(+)
> >>> 
> >>> diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
> >>> index 3fd8d7f..93c6163 100644
> >>> --- a/drivers/mtd/ubi/vmt.c
> >>> +++ b/drivers/mtd/ubi/vmt.c
> >>> @@ -610,6 +610,7 @@ int ubi_add_volume(struct ubi_device *ubi, struct
> >>> ubi_volume *vol)
> >>> 
> >>>   out_cdev:
> >>>       cdev_del(&vol->cdev);
> >>> 
> >>> +    put_device(&vol->dev);
> >>> 
> >>>       return err;
> >> 
> >> The more I dig into device code, the more questions I have.
> >> Why is cdev_del() not part of the release function?
> >> 
> >> Thanks,
> >> //richard
> > 
> > Yes, It's should be a part release function.
> > 
> > ~arvind
> 
> I was wrong, We can not add cdev_del() in release(vol_release)
> function.
> Function's ubi_create_volume and ubi_add_volume both are using
> same release function to release a volume devices.
> ubi_add_volume is registering character device for the volume.
> So we will have to release character device here.

This is not what I meant.
The question was whether we should free all this data structures from the 
device model's point of view.
That we have to massage UBI code for that is clear.

Thanks,
//richard