Re: [PATCH] kfence: unpoison pool region before use

From: Andrey Konovalov <andreyknvl@gmail.com>
To: Marco Elver <elver@google.com>, Peter Collingbourne <pcc@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>,
	Alexander Potapenko <glider@google.com>,
	Evgenii Stepanov <eugenis@google.com>,
	Linux Memory Management List <linux-mm@kvack.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>
Subject: Re: [PATCH] kfence: unpoison pool region before use
Date: Sun, 4 Apr 2021 14:09:27 +0200	[thread overview]
Message-ID: <CA+fCnZeYpBx5KwjBTr7H_kU=kSYhxiFdG9r=Wn0dBJ-fiV_9oA@mail.gmail.com> (raw)
In-Reply-To: <CA+fCnZd4BaejuyyWuT4xeiEyY1J0-6RWiyP3_u+w-xdOrALd9w@mail.gmail.com>

On Sun, Apr 4, 2021 at 1:52 AM Andrey Konovalov <andreyknvl@gmail.com> wrote:
>
> On Sun, Apr 4, 2021 at 12:31 AM Marco Elver <elver@google.com> wrote:
> >
> > However, given the above, I think we need to explain this in the
> > commit message (which also makes the dependency between these 2
> > patches clear) and add a comment above the new kasan_unpoison_range().
> > That is, if we still think this is the right fix -- I'm not entirely
> > sure it is.
> >
> > Because what I gather from "kasan: initialize shadow to TAG_INVALID
> > for SW_TAGS", is the requirement that "0xFF pointer tag is a match-all
> > tag, it doesn't matter what tag the accessed memory has".
> >
> > While KFENCE memory is accessible through the slab API, and in this
> > case ksize() calling kasan_check_byte() leading to a failure, the
> > kasan_check_byte() call is part of the public KASAN API. Which means
> > that if some subsystem decides to memblock_alloc() some memory, and
> > wishes to use kasan_check_byte() on that memory but with an untagged
> > pointer, will get the same problem as KFENCE: with generic and HW_TAGS
> > mode everything is fine, but with SW_TAGS mode things break.
>
> It makes sense to allow this function to operate on any kind of
> memory, including memory that hasn't been previously marked by KASAN.
>
> > To me this indicates the fix is not with KFENCE, but should be in
> > mm/kasan/sw_tags.c:kasan_byte_accessible(), which should not load the
> > shadow when the pointer is untagged.
>
> The problem isn't in accessing shadow per se. Looking at
> kasan_byte_accessible() (in both sw_tags.c and kasan.h), the return
> statement there seems just wrong and redundant.

(Technically, it's not wrong. But it was written under the assumption
that no accesses to KASAN_TAG_INVALID memory are valid. It's not the
case with KFENCE (without built-in KASAN annotations). And there might
be other places where this isn't the case as well, though we haven't
encountered any yet.)