Re: [PATCH] Fix argument checking in sched_setaffinity

[Andi Kleen <ak@muc.de>]

On Tue, Aug 31, 2004 at 06:36:55PM -0700, Paul Jackson wrote:
> Looks good - thanks, Andi.
> 
> I notice that you didn't bother with the fractional byte that is handled
> by 'endmask' in mm/mempolicy.c:get_nodes().  But I really don't give a
> hoot - either way is fine by me.

It is not needed because this function gets bytes instead of bits.


> I've written a couple of code snippets that manage to intuit the size of
> the kernel's cpumask dynamically from user space, by probing with
> various sched_getaffinity() calls.  But since your patch only changes
> the errors generated by sched_setaffinity() [that's "set", not "get"], I
> will not experience any grief from this subtle change in the kernel's
> API.
> 
> Should you lock hotplug before calling get_user_cpu_mask(), since
> get_user_cpu_mask() depends on cpu_online_mask()?

Good point yes, that is missing. 

However Linus has already thrown the code out and replaced
it with something more broken :-/

-Andi