Re: [PATCH 1/3] drm/i915: allocate mock file pointer dynamically

[Arnd Bergmann]

On Mon, Mar 20, 2017 at 1:02 PM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Mon, Mar 20, 2017 at 1:00 PM, Joonas Lahtinen
> <joonas.lahtinen@xxxxxxxxxxxxxxx> wrote:
>> On ma, 2017-03-20 at 10:40 +0100, Arnd Bergmann wrote:
>
>>> diff --git a/drivers/gpu/drm/i915/selftests/mock_drm.c b/drivers/gpu/drm/i915/selftests/mock_drm.c
>>> index 113dec05c7dc..18514065c93d 100644
>>> --- a/drivers/gpu/drm/i915/selftests/mock_drm.c
>>> +++ b/drivers/gpu/drm/i915/selftests/mock_drm.c
>>> @@ -32,15 +32,15 @@ static inline struct inode fake_inode(struct drm_i915_private *i915)
>>>  struct drm_file *mock_file(struct drm_i915_private *i915)
>>>  {
>>> >     struct inode inode = fake_inode(i915);
>>> > -   struct file filp = {};
>>> > +   struct file *filp = kzalloc(sizeof(struct file), GFP_KERNEL);
>>> >     struct drm_file *file;
>>> >     int err;
>>>
>>
>>         filp = kzalloc(sizeof(*filp), GFP_KERNEL);
>>         if (unlikely(!filp)) {
>>                 err = -ENOMEM;
>>                 goto err;
>>         }
>>
>> And appropriate onion teardown in case drm_open fails, so that we don't
>> leak memory.
>
> Oops, of course you are right, sorry about that.

Actually it's even worse, as I had originally planned to also allocate the inode
dynamically, but for some reasons didn't do that in the patch I sent.

This version of the patch as I sent it is rather bogus, so please ignore it
(but not the other two patches I sent). If David's solution to this problem
can make it into 4.12, there is no problem at all. Another alternative
would be to use anon_inode_getfile(), again with proper error handling
and cleanup. This is a bit slower but gives us valid file and inode structures.

     Arnd