exec: make de_thread() killable

Change de_thread() to use KILLABLE rather than UNINTERRUPTIBLE while
waiting for other threads.  The only complication is that we should
clear ->group_exit_task and ->notify_count before we return, and we
should do this under tasklist_lock.  -EAGAIN is used to match the
initial signal_group_exit() check/return, it doesn't really matter.

This fixes the (unlikely) race with coredump.  de_thread() checks
signal_group_exit() before it starts to kill the subthreads, but this
can't help if another CLONE_VM (but non CLONE_THREAD) task starts the
coredumping after de_thread() unlocks ->siglock.  In this case the
killed sub-thread can block in exit_mm() waiting for coredump_finish(),
execing thread waits for that sub-thead, and the coredumping thread
waits for execing thread.  Deadlock.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 14 insertions, 2 deletions

diff --git a/fs/exec.c b/fs/exec.c
index 9824473..19f4fb8 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -878,9 +878,11 @@ static int de_thread(struct task_struct *tsk)
 		sig->notify_count--;
 
 	while (sig->notify_count) {
-		__set_current_state(TASK_UNINTERRUPTIBLE);
+		__set_current_state(TASK_KILLABLE);
 		spin_unlock_irq(lock);
 		schedule();
+		if (unlikely(__fatal_signal_pending(tsk)))
+			goto killed;
 		spin_lock_irq(lock);
 	}
 	spin_unlock_irq(lock);
@@ -898,9 +900,11 @@ static int de_thread(struct task_struct *tsk)
 			write_lock_irq(&tasklist_lock);
 			if (likely(leader->exit_state))
 				break;
-			__set_current_state(TASK_UNINTERRUPTIBLE);
+			__set_current_state(TASK_KILLABLE);
 			write_unlock_irq(&tasklist_lock);
 			schedule();
+			if (unlikely(__fatal_signal_pending(tsk)))
+				goto killed;
 		}
 
 		/*
@@ -994,6 +998,14 @@ no_thread_group:
 
 	BUG_ON(!thread_group_leader(tsk));
 	return 0;
+
+killed:
+	/* protects against exit_notify() and __exit_signal() */
+	read_lock(&tasklist_lock);
+	sig->group_exit_task = NULL;
+	sig->notify_count = 0;
+	read_unlock(&tasklist_lock);
+	return -EAGAIN;
 }
 
 char *get_task_comm(char *buf, struct task_struct *tsk)