Re: [PATCH -next] ovl: respect underlying filesystem's get_unmapped_area()

[Kefeng Wang <wangkefeng.wang@huawei.com>]

On 2024/12/6 20:58, Amir Goldstein wrote:
> On Fri, Dec 6, 2024 at 11:45 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>
...
>>
>> So maybe use mm_get_unmapped_area() instead of __get_unmapped_area(),
>> something like below,
>>
>> +static unsigned long ovl_get_unmapped_area(struct file *file,
>> +               unsigned long addr, unsigned long len, unsigned long pgoff,
>> +               unsigned long flags)
>> +{
>> +       struct file *realfile;
>> +       const struct cred *old_cred;
>> +
>> +       realfile = ovl_real_file(file);
>> +       if (IS_ERR(realfile))
>> +               return PTR_ERR(realfile);
>> +
>> +       if (realfile->f_op->get_unmapped_area) {
>> +               unsigned long ret;
>> +
>> +               old_cred = ovl_override_creds(file_inode(file)->i_sb);
>> +               ret = realfile->f_op->get_unmapped_area(realfile, addr, len,
>> +                                                       pgoff, flags);
>> +               ovl_revert_creds(old_cred);
>> +
>> +               if (ret)
>> +                       return ret;
>> +       }
>> +
>> +       return mm_get_unmapped_area(current->mm, file, addr, len, pgoff,
>> flags);
>> +}
>>
>> Correct me If I'm wrong.
>>
> 
> You just need to be aware of the fact that between ovl_get_unmapped_area()
> and ovl_mmap(), ovl_real_file(file) could change from the lower file, to the
> upper file due to another operation that initiated copy-up.

Not sure about this part(I have very little knowledge of ovl), do you
mean that we could not use ovl_real_file()?  The ovl_mmap() using
realfile = file->private_data, we may use similar way in
ovl_get_unmapped_area(). but I may have misunderstood.

Thanks.