core/entry: Report syscall correctly for trace and audit

On v5.8 when doing seccomp syscall rewrites (e.g. getpid into getppid
as seen in the seccomp selftests), trace (and audit) correctly see the
rewritten syscall on entry and exit:

	seccomp_bpf-1307  [000] .... 22974.874393: sys_enter: NR 110 (...
	seccomp_bpf-1307  [000] .N.. 22974.874401: sys_exit: NR 110 = 1304

With mainline we see a mismatched enter and exit (the original syscall
is incorrectly visible on entry):

	seccomp_bpf-1030  [000] ....    21.806766: sys_enter: NR 39 (...
	seccomp_bpf-1030  [000] ....    21.806767: sys_exit: NR 110 = 1027

When ptrace or seccomp change the syscall, this needs to be visible to
trace and audit at that time as well. Update the syscall earlier so they
see the correct value.

Fixes: d88d59b ("core/entry: Respect syscall number rewrites")
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20200912005826.586171-1-keescook@chromium.org

kernel/entry/common.c

@@ -60,13 +60,15 @@ static long syscall_trace_enter(struct pt_regs *regs, long syscall,
 			return ret;
 	}
 
+	/* Either of the above might have changed the syscall number */
+	syscall = syscall_get_nr(current, regs);
+
 	if (unlikely(ti_work & _TIF_SYSCALL_TRACEPOINT))
 		trace_sys_enter(regs, syscall);
 
 	syscall_enter_audit(regs, syscall);
 
-	/* The above might have changed the syscall number */
-	return ret ? : syscall_get_nr(current, regs);
+	return ret ? : syscall;
 }
 
 static __always_inline long