io-wq: provide an io_wq_put_and_exit() helper

If we put the io-wq from io_uring, we really want it to exit. Provide
a helper that does that for us. Couple that with not having the manager
hold a reference to the 'wq' and the normal SQPOLL exit will tear down
the io-wq context appropriate.

On the io-wq side, our wq context is per task, so only the task itself
is manipulating ->manager and hence it's safe to check and clear without
any extra locking. We just need to ensure that the manager task stays
around, in case it exits.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

3 files changed, 21 insertions, 11 deletions

diff --git a/fs/io-wq.c b/fs/io-wq.c
index f0b7e9ff63faa3..1407ba74ffc3d8 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -749,7 +749,7 @@ static int io_wq_manager(void *data)
 	sprintf(buf, "iou-mgr-%d", wq->task_pid);
 	set_task_comm(current, buf);
 	current->flags |= PF_IO_WORKER;
-	wq->manager = current;
+	wq->manager = get_task_struct(current);
 
 	complete(&wq->started);
 
@@ -771,9 +771,7 @@ static int io_wq_manager(void *data)
 	/* we might not ever have created any workers */
 	if (atomic_read(&wq->worker_refs))
 		wait_for_completion(&wq->worker_done);
-	wq->manager = NULL;
 	complete(&wq->exited);
-	io_wq_put(wq);
 	do_exit(0);
 }
 
@@ -816,8 +814,6 @@ static int io_wq_fork_manager(struct io_wq *wq)
 		return 0;
 
 	reinit_completion(&wq->worker_done);
-	clear_bit(IO_WQ_BIT_EXIT, &wq->state);
-	refcount_inc(&wq->refs);
 	current->flags |= PF_IO_WORKER;
 	ret = io_wq_fork_thread(io_wq_manager, wq);
 	current->flags &= ~PF_IO_WORKER;
@@ -1089,6 +1085,16 @@ err_wq:
 	return ERR_PTR(ret);
 }
 
+static void io_wq_destroy_manager(struct io_wq *wq)
+{
+	if (wq->manager) {
+		wake_up_process(wq->manager);
+		wait_for_completion(&wq->exited);
+		put_task_struct(wq->manager);
+		wq->manager = NULL;
+	}
+}
+
 static void io_wq_destroy(struct io_wq *wq)
 {
 	int node;
@@ -1096,10 +1102,7 @@ static void io_wq_destroy(struct io_wq *wq)
 	cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
 
 	set_bit(IO_WQ_BIT_EXIT, &wq->state);
-	if (wq->manager) {
-		wake_up_process(wq->manager);
-		wait_for_completion(&wq->exited);
-	}
+	io_wq_destroy_manager(wq);
 
 	spin_lock_irq(&wq->hash->wait.lock);
 	for_each_node(node) {
@@ -1112,7 +1115,6 @@ static void io_wq_destroy(struct io_wq *wq)
 	io_wq_put_hash(wq->hash);
 	kfree(wq->wqes);
 	kfree(wq);
-
 }
 
 void io_wq_put(struct io_wq *wq)
@@ -1121,6 +1123,13 @@ void io_wq_put(struct io_wq *wq)
 		io_wq_destroy(wq);
 }
 
+void io_wq_put_and_exit(struct io_wq *wq)
+{
+	set_bit(IO_WQ_BIT_EXIT, &wq->state);
+	io_wq_destroy_manager(wq);
+	io_wq_put(wq);
+}
+
 static bool io_wq_worker_affinity(struct io_worker *worker, void *data)
 {
 	struct task_struct *task = worker->task;
diff --git a/fs/io-wq.h b/fs/io-wq.h
index b6ca12b60c35c2..f6ef433df8a8fa 100644
--- a/fs/io-wq.h
+++ b/fs/io-wq.h
@@ -114,6 +114,7 @@ struct io_wq_data {
 
 struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data);
 void io_wq_put(struct io_wq *wq);
+void io_wq_put_and_exit(struct io_wq *wq);
 
 void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work);
 void io_wq_hash_work(struct io_wq_work *work, void *val);
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 904bf0fecc36a8..cb65e54c1b09da 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -8857,7 +8857,7 @@ void __io_uring_files_cancel(struct files_struct *files)
 	if (files) {
 		io_uring_remove_task_files(tctx);
 		if (tctx->io_wq) {
-			io_wq_put(tctx->io_wq);
+			io_wq_put_and_exit(tctx->io_wq);
 			tctx->io_wq = NULL;
 		}
 	}