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

[Kees Cook <keescook@chromium.org>]

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>

-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