Re: [PATCH] Don't mlock guardpage if the stack is growing up

[Zdenek Kabelac <zkabelac@redhat.com>]

Dne 12.5.2011 04:12, Linus Torvalds napsal(a):
> On Wed, May 11, 2011 at 1:42 AM, Milan Broz <mbroz@redhat.com> wrote:
>>
>> Another one is cryptsetup [..]
> 
> Quite frankly, all security-related uses should always be happy about
> a "MCL_SPARSE" model, since there is no point in ever bringing in
> pages that haven't been used. The whole (and only) point of
> mlock[all]() for them is the "avoid to push to disk" issue.
> 
> I do wonder if we really should ever do the page-in at all. We might
> simply be better off always just saying "we'll lock pages you've
> touched, that's it".
> 


For LVM we need to ensure the code which might ever be executed during  disk
suspend state must be paged and locked in - thus we would need MCL_SPARSE only
on several selected 'unneeded' libraries - as we are obviously not really able
to select which part of glibc might be needed during all code path (though I
guess we may find some limits).  But if we are sure that some libraries and
locale files will never be used during suspend state - we do not care about
those pages at all.

So it's not like we would always need only MCL_SPARSE all the time - we would
probably need to have some control to switch i.e.  glibc into MCL_ALL.

Zdenek