Re: [PATCH v3] net: bluetooth: hci_sock: Use 'const u32 *' instead of 'void *' for 2nd parameter of hci_test_bit()

On 2/10/15 00:57, David Laight wrote:

From: Marcel Holtmann

quoted

Hi David,

quoted

quoted

So at present, in kernel part, we can only say the original authors
intended to do like this. And only within kernel part, it can not cause
issue. I guess, original authors originally knew what we talk about.

I've just searched for hci_u*filter it is all horrid.
Look at the code for get/set_sockopt of HCI_FILTER.
Someone seems to have done a complete 'bodge job' of fixing the user interface
for apps (32bit and 64bit) on 64bit kernels.

we are actually fully aware of that. Just keep in mind that this code is from 2.4.6 kernel and with
that most likely 14 years old by now. Unfortunately it ended up as userspace API.

AFAICT the userspace API is always __u32[2] the long[2] is never exposed
to userspace (certainly not through the socket option code).

Quite why all the casts were added - instead of changing the type
is probably hidden in the annals of the source repo.
Some serious archaeology might be in order.

For me, I guess what you said above sounds reasonable. But excuse me,
I am not familiar with the related modules outside Linux kernel (at
present, it is out of my border).

I guess, I am not the suitable member to make the related fix patch for
this issue. Welcome any other members to send the fix patch for it.

And for me, if "hci_u*filter" is related with the modules outside kernel
(I guess, it should be), I recommend to put it to the related "uapi"
header.


Thanks.
-- 
Chen Gang

Open, share, and attitude like air, water, and life which God blessed