From: Roland McGrath <roland@redhat.com>
Date: Fri, 22 Aug 2008 12:21:43 -0700
Subject: [utrace] signal interception breaks systemtap uprobes
Message-id: 20080822192143.BAF0415426C@magilla.localdomain
O-Subject: [RHEL5.3 PATCH] fix utrace signal interception breaking systemtap uprobes (RHBZ#459786)
Bugzilla: 459786

This fixes a bug in the utrace code, the cause of stap SWBZ#6828 / RHBZ#459786.
The bug affects a code path only taken when using the utrace kernel API,
never used by ptrace.  RHEL5.3's systemtap will exploit this code path.

The bug hits when intercepting a signal that has a handler installed with
the SA_ONESHOT flag (aka SA_RESETHAND).  It causes the traced process to
jump to 0 instead of the proper signal handler entry point.

The fix has been tested by systemtap developers.

Signed-off-by: Roland McGrath <roland@redhat.com>

diff --git a/kernel/utrace.c b/kernel/utrace.c
index 5083288..7eaff6a 100644
--- a/kernel/utrace.c
+++ b/kernel/utrace.c
@@ -2108,8 +2108,15 @@ utrace_get_signal(struct task_struct *tsk, struct pt_regs *regs,
 		 * The handler will run.  We do the SA_ONESHOT work here
 		 * since the normal path will only touch *return_ka now.
 		 */
-		if (return_ka->sa.sa_flags & SA_ONESHOT)
-			ka->sa.sa_handler = SIG_DFL;
+		signal.signr = info->si_signo;
+		if (likely(signal.signr) &&
+		    unlikely(return_ka->sa.sa_flags & SA_ONESHOT)) {
+			return_ka->sa.sa_flags &= ~SA_ONESHOT;
+			if (likely(valid_signal(signal.signr))) {
+				ka = &tsk->sighand->action[signal.signr - 1];
+				ka->sa.sa_handler = SIG_DFL;
+			}
+		}
 		break;
 
 	case UTRACE_SIGNAL_TSTP: