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

[Kees Cook <keescook@chromium.org>]

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?

-Kees

> 
> -Kees
> 
> > ---
> >  fs/binfmt_elf.c | 18 ++++++++++++++----
> >  1 file changed, 14 insertions(+), 4 deletions(-)
> > 
> > diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> > index f8c7f26f1fbb..0caaad9eddd1 100644
> > --- a/fs/binfmt_elf.c
> > +++ b/fs/binfmt_elf.c
> > @@ -402,19 +402,29 @@ static unsigned long elf_map(struct file *filep, unsigned long addr,
> >  static unsigned long total_mapping_size(const struct elf_phdr *cmds, int nr)
> >  {
> >  	int i, first_idx = -1, last_idx = -1;
> > +	unsigned long min_vaddr = ULONG_MAX, max_vaddr = 0;
> >  
> >  	for (i = 0; i < nr; i++) {
> >  		if (cmds[i].p_type == PT_LOAD) {
> > -			last_idx = i;
> > -			if (first_idx == -1)
> > +			/*
> > +			 * The PT_LOAD segments are not necessarily ordered
> > +			 * by vaddr. Make sure that we get the segment with
> > +			 * minimum vaddr (maximum vaddr respectively)
> > +			 */
> > +			if (cmds[i].p_vaddr <= min_vaddr) {
> >  				first_idx = i;
> > +				min_vaddr = cmds[i].p_vaddr;
> > +			}
> > +			if (cmds[i].p_vaddr >= max_vaddr) {
> > +				last_idx = i;
> > +				max_vaddr = cmds[i].p_vaddr;
> > +			}
> >  		}
> >  	}
> >  	if (first_idx == -1)
> >  		return 0;
> >  
> > -	return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
> > -				ELF_PAGESTART(cmds[first_idx].p_vaddr);
> > +	return max_vaddr + cmds[last_idx].p_memsz - ELF_PAGESTART(min_vaddr);
> >  }
> >  
> >  static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
> > -- 
> > 2.34.1
> 
> -- 
> Kees Cook

-- 
Kees Cook