From mboxrd@z Thu Jan  1 00:00:00 1970
From: Mark Rutland <mark.rutland@arm.com>
Subject: Re: [PATCH v2 1/4] syscalls: Restore address limit after a syscall
Date: Thu, 9 Mar 2017 12:09:55 +0000
Message-ID: <20170309120955.GA6320@leverpostej>
References: <20170309012456.5631-1-thgarnie@google.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Return-path: <kernel-hardening-return-6961-glkh-kernel-hardening=m.gmane.org@lists.openwall.com>
List-Post: <mailto:kernel-hardening@lists.openwall.com>
List-Help: <mailto:kernel-hardening-help@lists.openwall.com>
List-Unsubscribe: <mailto:kernel-hardening-unsubscribe@lists.openwall.com>
List-Subscribe: <mailto:kernel-hardening-subscribe@lists.openwall.com>
Content-Disposition: inline
In-Reply-To: <20170309012456.5631-1-thgarnie@google.com>
To: Thomas Garnier <thgarnie@google.com>
Cc: David Howells <dhowells@redhat.com>, Dave Hansen <dave.hansen@intel.com>, Arnd Bergmann <arnd@arndb.de>, Al Viro <viro@zeniv.linux.org.uk>, =?utf-8?B?UmVuw6k=?= Nyffenegger <mail@renenyffenegger.ch>, Andrew Morton <akpm@linux-foundation.org>, Kees Cook <keescook@chromium.org>, "Paul E . McKenney" <paulmck@linux.vnet.ibm.com>, "David S . Miller" <davem@davemloft.net>, Andy Lutomirski <luto@kernel.org>, Ard Biesheuvel <ard.biesheuvel@linaro.org>, Nicolas Pitre <nicolas.pitre@linaro.org>, Petr Mladek <pmladek@suse.com>, Sebastian Andrzej Siewior <bigeasy@linutronix.de>, Sergey Senozhatsky <sergey.senozhatsky@gmail.com>, Helge Deller <deller@gmx.de>, Rik van Riel <riel@redhat.com>, Ingo Molnar <mingo@kernel.org>, Oleg Nesterov <oleg@redhat.com>, John Stultz <john.stultz@linaro.org>, Thomas Gleixner <tglx@linutronix.de>, Pavel Tikhomirov <ptikhomirov@virtuozzo.com>, Fre
List-Id: linux-api@vger.kernel.org

Hi,

On Wed, Mar 08, 2017 at 05:24:53PM -0800, Thomas Garnier wrote:
> This patch ensures a syscall does not return to user-mode with a kernel
> address limit. If that happened, a process can corrupt kernel-mode
> memory and elevate privileges.
> 
> For example, it would mitigation this bug:
> 
> - https://bugs.chromium.org/p/project-zero/issues/detail?id=990
> 
> If the CONFIG_BUG_ON_DATA_CORRUPTION option is enabled, an incorrect
> state will result in a BUG_ON.
> 
> The CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE option is also
> added so each architecture can optimize this change.

> +#ifndef CONFIG_ARCH_NO_SYSCALL_VERIFY_PRE_USERMODE_STATE
> +static inline bool has_user_ds(void) {
> +	bool ret = segment_eq(get_fs(), USER_DS);
> +	// Prevent re-ordering the call
> +	barrier();

What ordering are we trying to ensure, that isn't otherwise given?

We expect get_fs() and set_fs() to be ordered w.r.t. each other and
w.r.t. uaccess uses, or we'd need barriers all over the place.

Given that, I can't see why we need a barrier here. So this needs a
better comment, at least.

> +	return ret;
> +}
> +#else
> +static inline bool has_user_ds(void) {
> +	return false;
> +}
> +#endif

It would be simpler to wrap the call entirely, e.g. have:

#ifdef CONFIG_WHATEVER
static inline void verify_pre_usermode_state(void)
{
	if (segment_eq(get_fs(), USER_DS))
		__verify_pre_usermode_state();
}
#else
static inline void verify_pre_usermode_state(void) { }
#endif

> @@ -199,7 +215,10 @@ extern struct trace_event_functions exit_syscall_print_funcs;
>  	asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__));	\
>  	asmlinkage long SyS##name(__MAP(x,__SC_LONG,__VA_ARGS__))	\
>  	{								\
> +		bool user_caller = has_user_ds();			\
>  		long ret = SYSC##name(__MAP(x,__SC_CAST,__VA_ARGS__));	\
> +		if (user_caller)					\
> +			verify_pre_usermode_state();			\

... then we can unconditionally use verify_pre_usermode_state() here ... 

>  		__MAP(x,__SC_TEST,__VA_ARGS__);				\
>  		__PROTECT(x, ret,__MAP(x,__SC_ARGS,__VA_ARGS__));	\
>  		return ret;						\

[...]

> +/* Called before coming back to user-mode */
> +asmlinkage void verify_pre_usermode_state(void)

... and we just prepend a couple of underscores here.

> +{
> +	if (CHECK_DATA_CORRUPTION(!segment_eq(get_fs(), USER_DS),
> +				  "incorrect get_fs() on user-mode return"))
> +		set_fs(USER_DS);
> +}

Thanks,
Mark.