diff mbox series

misc: ocxl: fix possible name leak in ocxl_file_register_afu()

Message ID 20221111145929.2429271-1-yangyingliang@huawei.com (mailing list archive)
State Accepted
Commit 295faa17722a11cac8dbf51e4c9f9405a5e07ef1
Headers show
Series misc: ocxl: fix possible name leak in ocxl_file_register_afu() | expand

Commit Message

Yang Yingliang Nov. 11, 2022, 2:59 p.m. UTC
If device_register() returns error in ocxl_file_register_afu(),
the name allocated by dev_set_name() need be freed. As comment
of device_register() says, it should use put_device() to give
up the reference in the error path. So fix this by calling
put_device(), then the name can be freed in kobject_cleanup(),
and info is freed in info_release().

Fixes: 75ca758adbaf ("ocxl: Create a clear delineation between ocxl backend & frontend")
Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
---
 drivers/misc/ocxl/file.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

Comments

Frederic Barrat Nov. 14, 2022, 11:23 a.m. UTC | #1
On 11/11/2022 15:59, Yang Yingliang wrote:
> If device_register() returns error in ocxl_file_register_afu(),
> the name allocated by dev_set_name() need be freed. As comment
> of device_register() says, it should use put_device() to give
> up the reference in the error path. So fix this by calling
> put_device(), then the name can be freed in kobject_cleanup(),
> and info is freed in info_release().
> 
> Fixes: 75ca758adbaf ("ocxl: Create a clear delineation between ocxl backend & frontend")
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
> ---
>   drivers/misc/ocxl/file.c | 7 +++++--
>   1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/misc/ocxl/file.c b/drivers/misc/ocxl/file.c
> index d46dba2df5a1..452d5777a0e4 100644
> --- a/drivers/misc/ocxl/file.c
> +++ b/drivers/misc/ocxl/file.c
> @@ -541,8 +541,11 @@ int ocxl_file_register_afu(struct ocxl_afu *afu)
>   		goto err_put;
>   
>   	rc = device_register(&info->dev);
> -	if (rc)
> -		goto err_put;
> +	if (rc) {
> +		free_minor(info);
> +		put_device(&info->dev);
> +		return rc;
> +	}


While I agree that a put_device() is needed on that error path, the fix 
above is not correct as it forgets to release the afu reference and the 
memory allocated in info. That was taken care of by the jump to the 
err_put label, so it should be kept. Something like:

-	if (rc)
+	if (rc) {
+		put_device((&info->dev);
  		goto err_put;
+	}


   Fred
Yang Yingliang Nov. 14, 2022, 11:46 a.m. UTC | #2
Hi,

On 2022/11/14 19:23, Frederic Barrat wrote:
>
>
> On 11/11/2022 15:59, Yang Yingliang wrote:
>> If device_register() returns error in ocxl_file_register_afu(),
>> the name allocated by dev_set_name() need be freed. As comment
>> of device_register() says, it should use put_device() to give
>> up the reference in the error path. So fix this by calling
>> put_device(), then the name can be freed in kobject_cleanup(),
>> and info is freed in info_release().
>>
>> Fixes: 75ca758adbaf ("ocxl: Create a clear delineation between ocxl 
>> backend & frontend")
>> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
>> ---
>>   drivers/misc/ocxl/file.c | 7 +++++--
>>   1 file changed, 5 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/misc/ocxl/file.c b/drivers/misc/ocxl/file.c
>> index d46dba2df5a1..452d5777a0e4 100644
>> --- a/drivers/misc/ocxl/file.c
>> +++ b/drivers/misc/ocxl/file.c
>> @@ -541,8 +541,11 @@ int ocxl_file_register_afu(struct ocxl_afu *afu)
>>           goto err_put;
>>         rc = device_register(&info->dev);
>> -    if (rc)
>> -        goto err_put;
>> +    if (rc) {
>> +        free_minor(info);
>> +        put_device(&info->dev);
>> +        return rc;
>> +    }
>
>
> While I agree that a put_device() is needed on that error path, the 
> fix above is not correct as it forgets to release the afu reference 
> and the memory allocated in info. That was taken care of by the jump 
> to the err_put label, so it should be kept. Something like:
>
> -    if (rc)
> +    if (rc) {
> +        put_device((&info->dev);
>          goto err_put;
> +    }
The 'info' and the reference is released in info_release().

Here is call chain:
put_device()
   kobject_release()
     kobject_cleanup()
       device_release()
         info_release()

static void info_release(struct device *dev)
{
         struct ocxl_file_info *info = container_of(dev, struct 
ocxl_file_info, dev);

         ocxl_afu_put(info->afu);
         kfree(info);
}
So it don't need jump to the error label in this case.

Thanks,
Yang
>
>
>   Fred
>
> .
Frederic Barrat Nov. 14, 2022, 12:04 p.m. UTC | #3
On 14/11/2022 12:46, Yang Yingliang wrote:
> Hi,
> 
> On 2022/11/14 19:23, Frederic Barrat wrote:
>>
>>
>> On 11/11/2022 15:59, Yang Yingliang wrote:
>>> If device_register() returns error in ocxl_file_register_afu(),
>>> the name allocated by dev_set_name() need be freed. As comment
>>> of device_register() says, it should use put_device() to give
>>> up the reference in the error path. So fix this by calling
>>> put_device(), then the name can be freed in kobject_cleanup(),
>>> and info is freed in info_release().
>>>
>>> Fixes: 75ca758adbaf ("ocxl: Create a clear delineation between ocxl 
>>> backend & frontend")
>>> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>
>>> ---
>>>   drivers/misc/ocxl/file.c | 7 +++++--
>>>   1 file changed, 5 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/misc/ocxl/file.c b/drivers/misc/ocxl/file.c
>>> index d46dba2df5a1..452d5777a0e4 100644
>>> --- a/drivers/misc/ocxl/file.c
>>> +++ b/drivers/misc/ocxl/file.c
>>> @@ -541,8 +541,11 @@ int ocxl_file_register_afu(struct ocxl_afu *afu)
>>>           goto err_put;
>>>         rc = device_register(&info->dev);
>>> -    if (rc)
>>> -        goto err_put;
>>> +    if (rc) {
>>> +        free_minor(info);
>>> +        put_device(&info->dev);
>>> +        return rc;
>>> +    }
>>
>>
>> While I agree that a put_device() is needed on that error path, the 
>> fix above is not correct as it forgets to release the afu reference 
>> and the memory allocated in info. That was taken care of by the jump 
>> to the err_put label, so it should be kept. Something like:
>>
>> -    if (rc)
>> +    if (rc) {
>> +        put_device((&info->dev);
>>          goto err_put;
>> +    }
> The 'info' and the reference is released in info_release().
> 
> Here is call chain:
> put_device()
>    kobject_release()
>      kobject_cleanup()
>        device_release()
>          info_release()
> 
> static void info_release(struct device *dev)
> {
>          struct ocxl_file_info *info = container_of(dev, struct 
> ocxl_file_info, dev);
> 
>          ocxl_afu_put(info->afu);
>          kfree(info);
> }
> So it don't need jump to the error label in this case.


You're right, I went too fast and the patch is correct.
So:
Acked-by: Frederic Barrat <fbarrat@linux.ibm.com>

   Fred
Andrew Donnellan Nov. 21, 2022, 5:52 a.m. UTC | #4
On Fri, 2022-11-11 at 22:59 +0800, Yang Yingliang wrote:
> If device_register() returns error in ocxl_file_register_afu(),
> the name allocated by dev_set_name() need be freed. As comment
> of device_register() says, it should use put_device() to give
> up the reference in the error path. So fix this by calling
> put_device(), then the name can be freed in kobject_cleanup(),
> and info is freed in info_release().
> 
> Fixes: 75ca758adbaf ("ocxl: Create a clear delineation between ocxl
> backend & frontend")
> Signed-off-by: Yang Yingliang <yangyingliang@huawei.com>

Thanks for the fix - as you point out, put_device() should clean
everything up that needs cleaning up.

Acked-by: Andrew Donnellan <ajd@linux.ibm.com>

> ---
>  drivers/misc/ocxl/file.c | 7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/misc/ocxl/file.c b/drivers/misc/ocxl/file.c
> index d46dba2df5a1..452d5777a0e4 100644
> --- a/drivers/misc/ocxl/file.c
> +++ b/drivers/misc/ocxl/file.c
> @@ -541,8 +541,11 @@ int ocxl_file_register_afu(struct ocxl_afu *afu)
>                 goto err_put;
>  
>         rc = device_register(&info->dev);
> -       if (rc)
> -               goto err_put;
> +       if (rc) {
> +               free_minor(info);
> +               put_device(&info->dev);
> +               return rc;
> +       }
>  
>         rc = ocxl_sysfs_register_afu(info);
>         if (rc)
Michael Ellerman Nov. 30, 2022, 9:24 a.m. UTC | #5
On Fri, 11 Nov 2022 22:59:29 +0800, Yang Yingliang wrote:
> If device_register() returns error in ocxl_file_register_afu(),
> the name allocated by dev_set_name() need be freed. As comment
> of device_register() says, it should use put_device() to give
> up the reference in the error path. So fix this by calling
> put_device(), then the name can be freed in kobject_cleanup(),
> and info is freed in info_release().
> 
> [...]

Applied to powerpc/next.

[1/1] misc: ocxl: fix possible name leak in ocxl_file_register_afu()
      https://git.kernel.org/powerpc/c/295faa17722a11cac8dbf51e4c9f9405a5e07ef1

cheers
diff mbox series

Patch

diff --git a/drivers/misc/ocxl/file.c b/drivers/misc/ocxl/file.c
index d46dba2df5a1..452d5777a0e4 100644
--- a/drivers/misc/ocxl/file.c
+++ b/drivers/misc/ocxl/file.c
@@ -541,8 +541,11 @@  int ocxl_file_register_afu(struct ocxl_afu *afu)
 		goto err_put;
 
 	rc = device_register(&info->dev);
-	if (rc)
-		goto err_put;
+	if (rc) {
+		free_minor(info);
+		put_device(&info->dev);
+		return rc;
+	}
 
 	rc = ocxl_sysfs_register_afu(info);
 	if (rc)