exec: RT sub-thread can livelock and monopolize CPU on exec

de_thread() yields waiting for ->group_leader to be a zombie. This deadlocks
if an rt-prio execer shares the same cpu with ->group_leader. Change the code
to use ->group_exit_task/notify_count mechanics.

This patch certainly uglifies the code, perhaps someone can suggest something
better.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2 files changed, 22 insertions, 16 deletions

diff --git a/fs/exec.c b/fs/exec.c
index ab5a4a3ece46..aa470a93540a 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -801,16 +801,15 @@ static int de_thread(struct task_struct *tsk)
                         hrtimer_restart(&sig->real_timer);
                 spin_lock_irq(lock);
         }
+
+        sig->notify_count = count;
+        sig->group_exit_task = tsk;
         while (atomic_read(&sig->count) > count) {
-                sig->group_exit_task = tsk;
-                sig->notify_count = count;
                 __set_current_state(TASK_UNINTERRUPTIBLE);
                 spin_unlock_irq(lock);
                 schedule();
                 spin_lock_irq(lock);
         }
-        sig->group_exit_task = NULL;
-        sig->notify_count = 0;
         spin_unlock_irq(lock);
 
         /*
@@ -819,14 +818,17 @@ static int de_thread(struct task_struct *tsk)
          * and to assume its PID:
          */
         if (!thread_group_leader(tsk)) {
-                /*
-                 * Wait for the thread group leader to be a zombie.
-                 * It should already be zombie at this point, most
-                 * of the time.
-                 */
                 leader = tsk->group_leader;
-                while (leader->exit_state != EXIT_ZOMBIE)
-                        yield();
+
+                sig->notify_count = -1;
+                for (;;) {
+                        write_lock_irq(&tasklist_lock);
+                        if (likely(leader->exit_state))
+                                break;
+                        __set_current_state(TASK_UNINTERRUPTIBLE);
+                        write_unlock_irq(&tasklist_lock);
+                        schedule();
+                }
 
                 /*
                  * The only record we have of the real-time age of a
@@ -840,8 +842,6 @@ static int de_thread(struct task_struct *tsk)
                  */
                 tsk->start_time = leader->start_time;
 
-                write_lock_irq(&tasklist_lock);
-
                 BUG_ON(leader->tgid != tsk->tgid);
                 BUG_ON(tsk->pid == tsk->tgid);
                 /*
@@ -874,6 +874,8 @@ static int de_thread(struct task_struct *tsk)
                 write_unlock_irq(&tasklist_lock);
         }
 
+        sig->group_exit_task = NULL;
+        sig->notify_count = 0;
         /*
          * There may be one thread left which is just exiting,
          * but it's safe to stop telling the group to kill themselves.
diff --git a/kernel/exit.c b/kernel/exit.c
index 25f6805be5fe..4c108df88a37 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -92,10 +92,9 @@ static void __exit_signal(struct task_struct *tsk)
                  * If there is any task waiting for the group exit
                  * then notify it:
                  */
-                if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
+                if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
                         wake_up_process(sig->group_exit_task);
-                        sig->group_exit_task = NULL;
-                }
+
                 if (tsk == sig->curr_target)
                         sig->curr_target = next_thread(tsk);
                 /*
@@ -827,6 +826,11 @@ static void exit_notify(struct task_struct *tsk)
                 state = EXIT_DEAD;
         tsk->exit_state = state;
 
+        if (thread_group_leader(tsk) &&
+            tsk->signal->notify_count < 0 &&
+            tsk->signal->group_exit_task)
+                wake_up_process(tsk->signal->group_exit_task);
+
         write_unlock_irq(&tasklist_lock);
 
         list_for_each_safe(_p, _n, &ptrace_dead) {