Re: [patch V4 10/15] x86/entry: Use generic syscall entry function

[Kees Cook <keescook@chromium.org>]

On Tue, Jul 21, 2020 at 12:57:16PM +0200, Thomas Gleixner wrote:
> Replace the syscall entry work handling with the generic version. Provide
> the necessary helper inlines to handle the real architecture specific
> parts, e.g. ptrace.
> 
> Use a temporary define for idtentry_enter_user which will be cleaned up
> seperately.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

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

Though, notes and a comment below...

> +/* Check that the stack and regs on entry from user mode are sane. */
> +static __always_inline void arch_check_user_regs(struct pt_regs *regs)
> +{
> +	if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) {
> +		/*
> +		 * Make sure that the entry code gave us a sensible EFLAGS
> +		 * register.  Native because we want to check the actual CPU
> +		 * state, not the interrupt state as imagined by Xen.
> +		 */
> +		unsigned long flags = native_save_fl();
> +		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
> +				      X86_EFLAGS_NT));

push, pop, bit test

> +
> +		/* We think we came from user mode. Make sure pt_regs agrees. */
> +		WARN_ON_ONCE(!user_mode(regs));

memory deref, bit test

> +
> +		/*
> +		 * All entries from user mode (except #DF) should be on the
> +		 * normal thread stack and should have user pt_regs in the
> +		 * correct location.
> +		 */
> +		WARN_ON_ONCE(!on_thread_stack());

per-cpu deref, subtract, test

> +		WARN_ON_ONCE(regs != task_pt_regs(current));

memory deref, test

> +	}
> +}

This doesn't look very expensive, and they certain indicate really bad
conditions. Does this need to be behind a CONFIG? (Whatever the answer,
we can probably make those changes in a later series -- some of these
also look not arch-specific...)

-- 
Kees Cook