Re: [PATCH] paravirt.h

[Zachary Amsden <zach@vmware.com>]

Andi Kleen wrote:
> On Tuesday 22 August 2006 16:25, Adrian Bunk wrote:
>   
>> On Tue, Aug 22, 2006 at 03:50:57PM +0200, Andi Kleen wrote:
>>     
>>>> this would need a "const after boot" section; which is really not hard
>>>> to make and probably useful for a lot more things.... todo++
>>>>         
>>> except for anything that needs tlb entries in user space. And it only gives you
>>> false sense of security. --todo
>>>       
>> What's the alternative?
>>     
>
> The alternative is to not protect it, since protecting it doesn't
> offer any significant additional security over not protecting it.
>   

Didn't someone point out yet that if you are vulnerable to someone 
loading a kernel module of their choosing, you lose, plain and simple?  
You don't need paravirt-ops to implement a rootkit, and it doesn't make 
it any easier, and write protecting it is totally useless.  How do you 
think VMware runs on Linux?  It takes over the hardware entirely, loads 
a hypervisor, and starts running in a completely different world.  And 
it doesn't even need to use a single _GPL'd export to do that.

Write protection is great as a debug option to find accidental memory 
corruptions.  It is useless as a technique to prevent subversion.  Um 
hello, you're already at CPL-0.  Just rewrite the page tables already.

>> Change it from a struct to a compile time choice?
>>     
>
> One of the design goals of paravirt-ops was to allow single binaries
> that run on both native hardware and on hypervisors. So that would
> be a non starter.

Strongly agree.

Zach