signals: don't abuse __flush_signals() in selinux_bprm_committed_creds()

selinux_bprm_committed_creds()->__flush_signals() is not right, we
shouldn't clear TIF_SIGPENDING unconditionally. There can be other
reasons for signal_pending(): freezing(), JOBCTL_PENDING_MASK, and
potentially more.

Also change this code to check fatal_signal_pending() rather than
SIGNAL_GROUP_EXIT, it looks a bit better.

Now we can kill __flush_signals() before it finds another buggy user.

Note: this code looks racy, we can flush a signal which was sent after
the task SID has been updated.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Paul Moore <pmoore@redhat.com>

1 files changed, 4 insertions, 2 deletions

diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index 99c4a00cce4e..8abbd548ece9 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c
@@ -2416,10 +2416,12 @@ static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
 		for (i = 0; i < 3; i++)
 			do_setitimer(i, &itimer, NULL);
 		spin_lock_irq(&current->sighand->siglock);
-		if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
-			__flush_signals(current);
+		if (!fatal_signal_pending(current)) {
+			flush_sigqueue(&current->pending);
+			flush_sigqueue(&current->signal->shared_pending);
 			flush_signal_handlers(current, 1);
 			sigemptyset(&current->blocked);
+			recalc_sigpending();
 		}
 		spin_unlock_irq(&current->sighand->siglock);
 	}