Re: Linux 6.2-rc1

From: Guenter Roeck <linux@roeck-us.net>
To: Kees Cook <kees@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Jaegeuk Kim <jaegeuk@kernel.org>, Chao Yu <chao@kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Vlastimil Babka <vbabka@suse.cz>,
	Peter Zijlstra <peterz@infradead.org>,
	Nick Desaulniers <ndesaulniers@google.com>,
	Kees Cook <keescook@chromium.org>
Subject: Re: Linux 6.2-rc1
Date: Wed, 28 Dec 2022 06:44:08 -0800	[thread overview]
Message-ID: <20221228144408.GA873415@roeck-us.net> (raw)
In-Reply-To: <DE30CDA1-CFF4-49FE-B653-F4CA831EAA12@kernel.org>

On Tue, Dec 27, 2022 at 07:40:30PM -0800, Kees Cook wrote:
> On December 26, 2022 9:52:12 PM PST, Guenter Roeck <linux@roeck-us.net> wrote:
> >On Mon, Dec 26, 2022 at 05:32:28PM -0800, Kees Cook wrote:
> >> On December 26, 2022 4:29:41 PM PST, Guenter Roeck <linux@roeck-us.net> wrote:
> >> >On Mon, Dec 26, 2022 at 01:03:59PM -0800, Kees Cook wrote:
> >> >> On December 26, 2022 12:56:29 PM PST, Linus Torvalds <torvalds@linux-foundation.org> wrote:
> >> >> >On Mon, Dec 26, 2022 at 11:52 AM Guenter Roeck <linux@roeck-us.net> wrote:
> >> >> >>
> >> >> >> fs/f2fs/inline.c: In function 'f2fs_move_inline_dirents':
> >> >> >> include/linux/fortify-string.h:59:33: error: '__builtin_memset' pointer overflow between offset [28, 898293814] and size [-898293787, -1] [-Werror=array-bounds]
> >> >> >> fs/f2fs/inline.c:430:9: note: in expansion of macro 'memset'
> >> >> >>   430 |         memset(dst.bitmap + src.nr_bitmap, 0, dst.nr_bitmap - src.nr_bitmap);
> >> >> >>       |         ^~~~~~
> >> >> >
> >> >> >Well, that's unfortunate.
> >> >> 
> >> >> I'll look into this.
> >> >> 
> >> >
> >> >I did some more testing. The problem is seen with gcc 11.3.0, but not with
> >> >gcc 12.2.0 nor with gcc 10.3.0.
> >> 
> >> That's what I'd expect: 10 didn't have variable range tracking wired up to -Warray-bounds, 11 does, and we disable -Warray-bounds on 12 because of 3 separate 12-only GCC bugs.
> >> 
> >> > gcc bug ? Should I switch to gcc 12.2.0 for
> >> >powerpc when build testing the latest kernel ?
> >> 
> >> Sure? But that'll just hide it. I suspect GCC has found a way for dst.nr_bitmap to be compile-time 27, so the size is always negative.
> >> 
> >dst.nr_bitmap is initialized with SIZE_OF_DENTRY_BITMAP,
> >which is defined as:
> >
> >#define NR_DENTRY_IN_BLOCK      214     /* the number of dentry in a block */
> >#define SIZE_OF_DIR_ENTRY       11      /* by byte */
> >#define SIZE_OF_DENTRY_BITMAP   ((NR_DENTRY_IN_BLOCK + BITS_PER_BYTE - 1) / \
> >                                        BITS_PER_BYTE)
> >
> >((214 + 8 - 1) / 8 = 27, so dst.nr_bitmap is indeed compile-time 27.
> >
> >Not sure how would know that src.nr_bitmap can be > 27, though.
> >Am I missing something ?
> 
> I think it's saying it can't rule out it being larger? I.e. there is no obvious bounds checking for it. Perhaps:
> 
> if (src.nr_bitmap > dst.nr_bitmap) {
>     err = -EFSCORRUPTED;
> 		goto out;
> }
> 

After going through all calculations, using maximum values (or minimum
values where appropriate) everywhere, I calculated that src.nr_bitmap
is always <= 24. The actual inode is sanity checked in
fs/f2fs/inode.c:sanity_check_inode().

Also, why is this only seen when I try to build powerpc test images ?

Thanks,
Guenter