Re: [PATCH] elf: Relax assumptions about vaddr ordering

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 26 Jan 2022 22:31:42 -0800 Kees Cook <keescook@chromium.org> wrote:

> On Wed, Jan 26, 2022 at 08:50:15AM -0800, Kees Cook wrote:
> > On Wed, Jan 26, 2022 at 05:25:20PM +0100, Magnus Groß wrote:
> > > From ff4dde97e82727727bda711f2367c05663498b24 Mon Sep 17 00:00:00 2001
> > > From: =?UTF-8?q?Magnus=20Gro=C3=9F?= <magnus.gross@rwth-aachen.de>
> > > Date: Wed, 26 Jan 2022 16:35:07 +0100
> > > Subject: [PATCH] elf: Relax assumptions about vaddr ordering
> > > MIME-Version: 1.0
> > > Content-Type: text/plain; charset=UTF-8
> > > Content-Transfer-Encoding: 8bit
> > > 
> > > Commit 5f501d555653 ("binfmt_elf: reintroduce using
> > > MAP_FIXED_NOREPLACE") introduced a regression, where the kernel now
> > > assumes that PT_LOAD segments are ordered by vaddr in load_elf_binary().
> > > 
> > > Specifically consider an ELF binary with the following PT_LOAD segments:
> > > 
> > > Type  Offset   VirtAddr   PhysAddr   FileSiz  MemSiz    Flg Align
> > > LOAD  0x000000 0x08000000 0x08000000 0x474585 0x474585  R E 0x1000
> > > LOAD  0x475000 0x08475000 0x08475000 0x090a4  0xc6c10   RW  0x1000
> > > LOAD  0x47f000 0x00010000 0x00010000 0x00000  0x7ff0000     0x1000
> > > 
> > > Note how the last segment is actually the first segment and vice versa.
> > > 
> > > Since total_mapping_size() only computes the difference between the
> > > first and the last segment in the order that they appear, it will return
> > > a size of 0 in this case, thus causing load_elf_binary() to fail, which
> > > did not happen before that change.
> > > 
> > > Strictly speaking total_mapping_size() made that assumption already
> > > before that patch, but the issue did not appear because the old
> > > load_addr_set guards never allowed this call to total_mapping_size().
> > > 
> > > Instead of fixing this by reverting to the old load_addr_set logic, we
> > > fix this by comparing the correct first and last segments in
> > > total_mapping_size().
> > 
> > Ah, nice. Yeah, this is good.
> > 
> > > Signed-off-by: Magnus Groß <magnus.gross@rwth-aachen.de>
> > 
> > Fixes: 5f501d555653 ("binfmt_elf: reintroduce using MAP_FIXED_NOREPLACE")
> > Cc: stable@vger.kernel.org
> > Acked-by: Kees Cook <keescook@chromium.org>
> 
> Andrew, can you pick this up too?
> 

It conflicts significantly with Alexey's "ELF: fix overflow in total
mapping size calculation".