linux-kernel - Re: [PATCH] binfmt_elf: use check_mul

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <202506092053.827AD89DC5@keescook>
Date: Mon, 9 Jun 2025 21:04:36 -0700
From: Kees Cook <kees@...nel.org>
To: Pranav Tyagi <pranav.tyagi03@...il.com>
Cc: viro@...iv.linux.org.uk, brauner@...nel.org, jack@...e.cz,
	linux-fsdevel@...r.kernel.org, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org, skhan@...uxfoundation.org,
	linux-kernel-mentees@...ts.linux.dev
Subject: Re: [PATCH] binfmt_elf: use check_mul_overflow() for size calc

On Sat, Jun 07, 2025 at 01:58:44PM +0530, Pranav Tyagi wrote:
> Use check_mul_overflow() to safely compute the total size of ELF program
> headers instead of relying on direct multiplication.
> 
> Directly multiplying sizeof(struct elf_phdr) with e_phnum risks integer
> overflow, especially on 32-bit systems or with malformed ELF binaries
> crafted to trigger wrap-around. If an overflow occurs, kmalloc() could
> allocate insufficient memory, potentially leading to out-of-bound
> accesses, memory corruption or security vulnerabilities.
> 
> Using check_mul_overflow() ensures the multiplication is performed
> safely and detects overflows before memory allocation. This change makes
> the function more robust when handling untrusted or corrupted binaries.
> 
> Signed-off-by: Pranav Tyagi <pranav.tyagi03@...il.com>
> Link: https://github.com/KSPP/linux/issues/92
> ---
>  fs/binfmt_elf.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index a43363d593e5..774e705798b8 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -518,7 +518,10 @@ static struct elf_phdr *load_elf_phdrs(const struct elfhdr *elf_ex,
>  
>  	/* Sanity check the number of program headers... */
>  	/* ...and their total size. */
> -	size = sizeof(struct elf_phdr) * elf_ex->e_phnum;

size is unsigned int, which has a maximum value of 4,294,967,295.

elf_ex->e_phnum is a u16 (2 bytes) and will not be changing:

$ pahole -C elf64_hdr */fs/binfmt_elf.o
struct elf64_hdr {
	...
        Elf64_Half                 e_phnum;              /*    56     2 */
	...
$ pahole -C Elf64_Half */fs/binfmt_elf.o
typedef __u16 Elf64_Half;

So it has a maximum value of 65,535.

sizeof(struct elf_phdr) is a fixed value, 56:

$ pahole -C elf64_phdr */fs/binfmt_elf.o
struct elf64_phdr {
	...
        /* size: 56, cachelines: 1, members: 8 */
        /* last cacheline: 56 bytes */
};

So the maximum product of the two is 3,669,960.

It is not possible for this calculation to overflow.

> +	
> +	if (check_mul_overflow(sizeof(struct elf_phdr), elf_ex->e_phnum, &size))
> +		goto out;
> +

You can even see that the entire check would be elided by the compiler:

#include <elf.h>

unsigned int unchecked(Elf64_Ehdr *elf_ex)
{
    unsigned int size;

    size = sizeof(Elf64_Phdr) * elf_ex->e_phnum;

    return size;
}

unsigned int checked(Elf64_Ehdr *elf_ex)
{
    unsigned int size;

    if (__builtin_mul_overflow(sizeof(Elf64_Phdr), elf_ex->e_phnum, &size))
        return 0;

    return size;
}

...produces this assembler, identical for both functions:

unchecked:
        movzx   eax, WORD PTR [rdi+56]
        imul    eax, eax, 56
        ret
checked:
        movzx   eax, WORD PTR [rdi+56]
        imul    eax, eax, 56
        ret


https://godbolt.org/z/hTEef8cT9

-Kees

-- 
Kees Cook