Re: [PATCH] An attempt to have an unlimitedly extendable sys_clone

[Pavel Emelyanov <xemul-GEFAQzZX7r8dnm+yROfE0A@public.gmane.org>]

Daniel Hokka Zakrisson wrote:
> Pavel Emelyanov wrote:
>> [snip]
>>
>>>> +static struct long_clone_arg *get_long_clone_arg(int __user
>>>> *child_tidptr)
>>>> +{
>>>> +	int size;
>>>> +	struct long_clone_arg *carg;
>>>> +
>>>> +	if (get_user(size, child_tidptr))
>>>> +		return ERR_PTR(-EFAULT);
>>>> +
>>>> +	if (size > sizeof(struct long_clone_arg))
>>>> +		return ERR_PTR(-EINVAL);
>>> This means that software built against a newer kernel won't work on an
>>> older one. Surely that's not intended?
>> It is intended. If I ask an old kernel to clone the mq namespace, but
>> it doesn't support such, that I'd better like to get an -EINVAL error
>> rather than be silently held in an old global namespace.
> 
> That rules out using the struct for things like child_tidptr, the desired
> pid for C/R, etc. I think the version Dave Hansen proposed would be

A new system call? Ok - what arguments should it take?

> better, or if it's really just for bits, use an array rather than a struct
> to make that obvious.

It's for anything you'd like to tell to clone(). In the nearest future - for
more bits to clone new namespaces. In the far future - for pid to create the
task with (for c/r jobs). In the the far-far future - for anything you will
need then.

>> [snip]
>>
>