Skip to contentSkip to conversation

exec: simplify the new ->sighand allocation

PATCH In review Oleg Nesterov 2 participants 2 messages Aug 18, 2007

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 3/5 exec: simplify the new ->sighand allocation

de_thread() pre-allocates newsighand to make sure that exec() can't fail after killing all sub-threads. Imho, this buys nothing, but complicates the code:

- this is (mostly) needed to handle CLONE_SIGHAND without CLONE_THREAD tasks, this is very unlikely (if ever used) case

- unless we already have some serious problems, GFP_KERNEL allocation should not fail

- ENOMEM still can happen after de_thread(), ->sighand is not the last object we have to allocate

Change the code to allocate the new ->sighand on demand.

fs/exec.c renamed from fs/exec.c~3_ALLOC +9 -15
@@ -774,7 +774,7 @@ static int exec_mmap(struct mm_struct *m
 static int de_thread(struct task_struct *tsk)
 {
 	struct signal_struct *sig = tsk->signal;
-	struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
+	struct sighand_struct *oldsighand = tsk->sighand;
 	spinlock_t *lock = &oldsighand->siglock;
 	struct task_struct *leader = NULL;
 	int count;
@@ -789,10 +789,6 @@ static int de_thread(struct task_struct 
 		return 0;
 	}
 
-	newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
-	if (!newsighand)
-		return -ENOMEM;
-
 	if (thread_group_empty(tsk))
 		goto no_thread_group;
 
@@ -809,7 +805,6 @@ static int de_thread(struct task_struct 
 		 */
 		spin_unlock_irq(lock);
 		read_unlock(&tasklist_lock);
-		kmem_cache_free(sighand_cachep, newsighand);
 		return -EAGAIN;
 	}
 
@@ -928,17 +923,16 @@ no_thread_group:
 	if (leader)
 		release_task(leader);
 
-	if (atomic_read(&oldsighand->count) == 1) {
-		/*
-		 * Now that we nuked the rest of the thread group,
-		 * it turns out we are not sharing sighand any more either.
-		 * So we can just keep it.
-		 */
-		kmem_cache_free(sighand_cachep, newsighand);
-	} else {
+	if (atomic_read(&oldsighand->count) != 1) {
+		struct sighand_struct *newsighand;
 		/*
-		 * Move our state over to newsighand and switch it in.
+		 * This ->sighand is shared with the CLONE_SIGHAND
+		 * but not CLONE_THREAD task, switch to the new one.
 		 */
+		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
+		if (!newsighand)
+			return -ENOMEM;
+
 		atomic_set(&newsighand->count, 1);
 		memcpy(newsighand->action, oldsighand->action,
 		       sizeof(newsighand->action));
#1 lore raw
Quoted 2 lines

- ENOMEM still can happen after de_thread(), ->sighand is not the last object we have to allocate

As long as this is true, I think it's the incontrovertible argument there is no reason not to simplify it. (Not that I don't also agree with your other reasons).

Thanks,
Roland