Re: [PATCH 0/3] kasan: Fix ordering between MTE tag colouring and page->flags

[Catalin Marinas <catalin.marinas@arm.com>]

Hi Andrey,

Sorry, I got distracted by the merging window.

On Tue, May 31, 2022 at 07:16:03PM +0200, Andrey Konovalov wrote:
> On Thu, May 26, 2022 at 2:24 PM Catalin Marinas <catalin.marinas@arm.com> wrote:
> > If we skip unpoisoning (not just poisoning as we already do) for user
> > pages, we should reset the tags in page->flags. Whether __GFP_ZEROTAGS
> > is passed is complementary, depending on the reason for allocation.
> 
> [...]
> 
> > Currently if __GFP_ZEROTAGS is passed, the unpoisoning is skipped but I
> > think we should have just added __GFP_SKIP_KASAN_UNPOISON instead and
> > not add a new argument to should_skip_kasan_unpoison(). If we decide to
> > always skip unpoisoning, something like below on top of the vanilla
> > kernel:
> 
> [...]
> 
> > With the above, we can wire up page_kasan_tag_reset() to the
> > __GFP_SKIP_KASAN_UNPOISON check without any additional flags.
> 
> This would make __GFP_SKIP_KASAN_UNPOISON do two logically unrelated
> things: skip setting memory tags and reset page tags. This seems
> weird.

Not entirely weird, it depends on how you look at it. After allocation,
you expect the accesses to page_address() to work, irrespective of the
GFP flags. __kasan_unpoison_pages() ensures that the page->flags match
the written tag without a new GFP flag to set the page->flags. If you
skip the unpoisoning something should reset the page->flags tag to
ensure an accessible page_address(). I find it weirder that you need
another GFP flag to pretty much say 'give me an accessible page'.

> I think it makes more sense to split __GFP_ZEROTAGS into
> __GFP_ZERO_MEMORY_TAGS and __GFP_ZERO_PAGE_TAGS: the first one does
> tag_clear_highpage() without page_kasan_tag_reset() and the second one
> does page_kasan_tag_reset() in post_alloc_hook(). Then, add
> __GFP_ZERO_PAGE_TAGS to GFP_HIGHUSER_MOVABLE along with
> __GFP_SKIP_KASAN_UNPOISON and __GFP_SKIP_KASAN_POISON. And replace
> __GFP_ZEROTAGS with __GFP_ZERO_MEMORY_TAGS in
> alloc_zeroed_user_highpage_movable().

As above, my preference would be to avoid a new flag, just wire this up
to __GFP_SKIP_KASAN_UNPOISON. But if you do want fine-grained control, I
can add the above.

Thanks.

-- 
Catalin