Received-SPF: pass (google.com: domain of linux-kernel+bounces-66119-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Wed, 14 Feb 2024 15:24:42 -0800
From: Kees Cook <keescook@chromium.org>
To: Jann Horn <jannh@google.com>
Cc: Paul Moore <paul@paul-moore.com>, James Morris <jmorris@namei.org>,
	"Serge E. Hallyn" <serge@hallyn.com>,
	Casey Schaufler <casey@schaufler-ca.com>,
	John Johansen <john.johansen@canonical.com>,
	linux-security-module@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] security: fix integer overflow in lsm_set_self_attr()
 syscall
Message-ID: <202402141453.FEC8EF722@keescook>
References: <20240214160538.1086089-1-jannh@google.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240214160538.1086089-1-jannh@google.com>

On Wed, Feb 14, 2024 at 05:05:38PM +0100, Jann Horn wrote:
> security_setselfattr() has an integer overflow bug that leads to
> out-of-bounds access when userspace provides bogus input:
> `lctx->ctx_len + sizeof(*lctx)` is checked against `lctx->len` (and,
> redundantly, also against `size`), but there are no checks on
> `lctx->ctx_len`.
> Therefore, userspace can provide an `lsm_ctx` with `->ctx_len` set to a
> value between `-sizeof(struct lsm_ctx)` and -1, and this bogus `->ctx_len`
> will then be passed to an LSM module as a buffer length, causing LSM
> modules to perform out-of-bounds accesses.

Ugh. Thanks for catching this. I continue to want to get the unsigned
integer overflow sanitizer rolled out, which would have caught this.

> 
> The following reproducer will demonstrate this under ASAN (if AppArmor is
> loaded as an LSM):
> ```
> #define _GNU_SOURCE
> #include <unistd.h>
> #include <stdint.h>
> #include <stdlib.h>
> #include <sys/syscall.h>
> 
> struct lsm_ctx {
>   uint64_t id;
>   uint64_t flags;
>   uint64_t len;
>   uint64_t ctx_len;
>   char ctx[];
> };
> 
> int main(void) {
>   size_t size = sizeof(struct lsm_ctx);
>   struct lsm_ctx *ctx = malloc(size);
>   ctx->id = 104/*LSM_ID_APPARMOR*/;
>   ctx->flags = 0;
>   ctx->len = size;
>   ctx->ctx_len = -sizeof(struct lsm_ctx);
>   syscall(
>     460/*__NR_lsm_set_self_attr*/,
>     /*attr=*/  100/*LSM_ATTR_CURRENT*/,
>     /*ctx=*/   ctx,
>     /*size=*/  size,
>     /*flags=*/ 0
>   );
> }
> ```
> 
> (I'm including an ASAN splat in the patch notes sent to the list.)
> 
> Fixes: a04a1198088a ("LSM: syscalls for current process attributes")
> Signed-off-by: Jann Horn <jannh@google.com>

Reviewed-by: Kees Cook <keescook@chromium.org>

> ---
> ASAN splat from the reproducer:
> ==================================================================
> BUG: KASAN: slab-out-of-bounds in do_setattr (security/apparmor/lsm.c:860) 
> Read of size 1 at addr ffff888006163abf by task setselfattr/548

I'd rather prefer that this splat (or some portion of it) stay in the
actual commit log. It makes it easier to scan for sanitizer-related
fixes, etc.

-Kees

-- 
Kees Cook