linux kernel implementation?

[Nicholas Murphy]

Actually, no...for performance reasons.  I don't propose actually  
merging GMP into the kernel source tree.  I'm just after a version  
that will compile into a kernel module for experimental purposes.   
Specifically, I'm experimenting with cryptographically signing and  
verifying network packets, and it's too expensive to do a userland  
transition every time.

Thanks,
Nick

On Dec 13, 2006, at 9:02 AM, Linas Vepstas wrote:

> On Tue, Dec 12, 2006 at 11:11:40PM -0800, Nicholas Murphy wrote:
>> Has anyone ever tried to port GMP for use within the Linux kernel?
>> If not, any idea what would be involved?  Would it just be a matter
>> of stripping out printf's and replacing malloc with kmalloc, etc.?
>
> Well, although there's some basic crypto in the kernel, I doubt
> the kernel hackers would allow gmp code into the kernel. Most
> want less, not more, code in the kernel.  Thus, the canonical
> rejoinder is to do it in user space, and communicate with kernel
> with a netlink.
>
> I presme the point of putting it in the kernel is to gain additional
> security. For that, I suggest that the gmp-using server run in its
> own security domain (if using selinux), or, what surely is easier, run
> it in its own virtual domain (using vserver).  If its acceptable to
> run only on the latest hardware, then look at using the hypervisor
> features to isolate your server from the bad guys.
>
> --linas