Re: [PATCHv5 03/19] mm/ksm: Do not merge pages with different KeyIDs

[Kirill A. Shutemov]

On Thu, Jul 19, 2018 at 07:02:34AM -0700, Dave Hansen wrote:
> On 07/19/2018 12:32 AM, Kirill A. Shutemov wrote:
> > On Wed, Jul 18, 2018 at 10:38:27AM -0700, Dave Hansen wrote:
> >> On 07/17/2018 04:20 AM, Kirill A. Shutemov wrote:
> >>> Pages encrypted with different encryption keys are not allowed to be
> >>> merged by KSM. Otherwise it would cross security boundary.
> >> Let's say I'm using plain AES (not AES-XTS).  I use the same key in two
> >> keyid slots.  I map a page with the first keyid and another with the
> >> other keyid.
> >>
> >> Won't they have the same cipertext?  Why shouldn't we KSM them?
> > We compare plain text, not ciphertext. And for good reason.
> 
> What's the reason?  Probably good to talk about it for those playing
> along at home.

I'll update commit message.

> > Comparing ciphertext would only make KSM successful for AES-ECB that
> > doesn't dependent on physical address of the page.
> > 
> > MKTME only supports AES-XTS (no plans to support AES-ECB). It effectively
> > disables KSM if we go with comparing ciphertext.
> 
> But what's the security boundary that is violated?  You are talking
> about some practical concerns (KSM scanning inefficiency) which is a far
> cry from being any kind of security issue.

As with zero page, my initial reasoning was that mixing pages from
different secrutiy domains is bad idea.

-- 
 Kirill A. Shutemov