Re: [PATCH] x86/ptrace: Remove questionable TS_COMPAT usage in ptrace

[Oleg Nesterov]

On 06/19, Andy Lutomirski wrote:
>
> On Sat, Jun 18, 2016 at 10:02 AM, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
> Step 1: for 4.7 and for -stable, introduce TS_I386_REGS_POKED.  Set it
> in putreg32.  Use it in syscall_get_error, get_nr_restart_syscall,
> etc.  Clear it in do_signal.

do_signal() won't be necessarily called...

> I wonder if we could actually get away with doing syscall restart
> processing before ptrace invocation.

How? this doesn't look possible or I misunderstood.

How about the simple change below for now? IIRC 32-bit task can't use
"syscall" so if syscall_get_nr() >= 0 then even the wrong TS_COMPAT is
not that bad, even if it "leaks" to user-mode.

nobody should use, say, in_ia32_syscall() unless we know that "in syscall"
is actually true. Hmm, arch/x86/kernel/uprobes.c does and this is wrong
regardless, I'll send the fix.

Oleg.

--- x/arch/x86/kernel/ptrace.c
+++ x/arch/x86/kernel/ptrace.c
@@ -930,7 +930,7 @@ static int putreg32(struct task_struct *
 		 * exit from a 32-bit syscall with TS_COMPAT still set.
 		 */
 		regs->orig_ax = value;
-		if (syscall_get_nr(child, regs) >= 0)
+		if (syscall_get_nr(child, regs) >= 0 && !user_64bit_mode(regs))
 			task_thread_info(child)->status |= TS_COMPAT;
 		break;