[PATCH v1 9/9] staging: android: binder: Add binder compat layer

[Colin Cross]

On Wed, Dec 4, 2013 at 3:21 PM, One Thousand Gnomes
<gnomes at lxorguk.ukuu.org.uk> wrote:
> On Wed, 4 Dec 2013 10:35:54 -0800
> Greg KH <gregkh at linuxfoundation.org> wrote:
>
>> On Wed, Dec 04, 2013 at 06:09:41PM +0000, Serban Constantinescu wrote:
>> > +#define size_helper(x) ({                                              \
>> > +   size_t __size;                                                      \
>> > +   if (!is_compat_task())                                              \
>> > +           __size = sizeof(x);                                         \
>> > +   else if (sizeof(x) == sizeof(struct flat_binder_object))            \
>> > +           __size = sizeof(struct compat_flat_binder_object);          \
>> > +   else if (sizeof(x) == sizeof(struct binder_transaction_data))       \
>> > +           __size = sizeof(struct compat_binder_transaction_data);     \
>> > +   else if (sizeof(x) == sizeof(size_t))                               \
>> > +           __size = sizeof(compat_size_t);                             \
>> > +   else                                                                \
>> > +            BUG();                                                     \
>> > +   __size;                                                             \
>> > +   })
>>
>> Ick.
>>
>> First off, no driver should ever be able to crash the kernel, which you
>> just did.
>
> And which would appear to mean that this code hasn't actually been
> tested - at least not properly with error cases ?
>
> You talk about type safety too but your code is already full of
> "put_user(node->ptr, (void * __user *)ptr))"

None of this (the patch series or the original code) is mine.  My
question was more of a general one on designing ioctls, as well as
concerns with changing the existing 32-bit api.