io_uring: make SQPOLL thread parking saner

We have this weird true/false return from parking, and then some of the
callers decide to look at that. It can lead to unbalanced parks and
sqd locking. Have the callers check the thread status once it's parked.
We know we have the lock at that point, so it's either valid or it's NULL.

Fix race with parking on thread exit. We need to be careful here with
ordering of the sdq->lock and the IO_SQ_THREAD_SHOULD_PARK bit.

Rename sqd->completion to sqd->parked to reflect that this is the only
thing this completion event doesn.

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

Author: axboe

fs/io_uring.c

@@ -274,7 +274,7 @@ struct io_sq_data {
 
 	unsigned long		state;
 	struct completion	startup;
-	struct completion	completion;
+	struct completion	parked;
 	struct completion	exited;
 };
 
@@ -6656,7 +6656,7 @@ static void io_sq_thread_parkme(struct io_sq_data *sqd)
 		 * wait_task_inactive().
 		 */
 		preempt_disable();
-		complete(&sqd->completion);
+		complete(&sqd->parked);
 		schedule_preempt_disabled();
 		preempt_enable();
 	}
@@ -6751,14 +6751,18 @@ static int io_sq_thread(void *data)
 
 	io_run_task_work();
 
-	if (io_sq_thread_should_park(sqd))
-		io_sq_thread_parkme(sqd);
-
 	/*
-	 * Clear thread under lock so that concurrent parks work correctly
+	 * Ensure that we park properly if racing with someone trying to park
+	 * while we're exiting. If we fail to grab the lock, check park and
+	 * park if necessary. The ordering with the park bit and the lock
+	 * ensures that we catch this reliably.
 	 */
-	complete(&sqd->completion);
-	mutex_lock(&sqd->lock);
+	if (!mutex_trylock(&sqd->lock)) {
+		if (io_sq_thread_should_park(sqd))
+			io_sq_thread_parkme(sqd);
+		mutex_lock(&sqd->lock);
+	}
+
 	sqd->thread = NULL;
 	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
 		ctx->sqo_exec = 1;
@@ -7067,29 +7071,25 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
 static void io_sq_thread_unpark(struct io_sq_data *sqd)
 	__releases(&sqd->lock)
 {
-	if (!sqd->thread)
-		return;
 	if (sqd->thread == current)
 		return;
 	clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
-	wake_up_state(sqd->thread, TASK_PARKED);
+	if (sqd->thread)
+		wake_up_state(sqd->thread, TASK_PARKED);
 	mutex_unlock(&sqd->lock);
 }
 
-static bool io_sq_thread_park(struct io_sq_data *sqd)
+static void io_sq_thread_park(struct io_sq_data *sqd)
 	__acquires(&sqd->lock)
 {
 	if (sqd->thread == current)
-		return true;
+		return;
+	set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
 	mutex_lock(&sqd->lock);
-	if (!sqd->thread) {
-		mutex_unlock(&sqd->lock);
-		return false;
+	if (sqd->thread) {
+		wake_up_process(sqd->thread);
+		wait_for_completion(&sqd->parked);
 	}
-	set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
-	wake_up_process(sqd->thread);
-	wait_for_completion(&sqd->completion);
-	return true;
 }
 
 static void io_sq_thread_stop(struct io_sq_data *sqd)
@@ -7185,7 +7185,7 @@ static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
 	mutex_init(&sqd->lock);
 	init_waitqueue_head(&sqd->wait);
 	init_completion(&sqd->startup);
-	init_completion(&sqd->completion);
+	init_completion(&sqd->parked);
 	init_completion(&sqd->exited);
 	return sqd;
 }
@@ -7829,7 +7829,7 @@ static int io_sq_thread_fork(struct io_sq_data *sqd, struct io_ring_ctx *ctx)
 	int ret;
 
 	clear_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
-	reinit_completion(&sqd->completion);
+	reinit_completion(&sqd->parked);
 	ctx->sqo_exec = 0;
 	sqd->task_pid = current->pid;
 	tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
@@ -8712,19 +8712,17 @@ static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 					  struct files_struct *files)
 {
 	struct task_struct *task = current;
-	bool did_park = false;
 
 	if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
 		/* never started, nothing to cancel */
 		if (ctx->flags & IORING_SETUP_R_DISABLED) {
 			io_sq_offload_start(ctx);
 			return;
 		}
-		did_park = io_sq_thread_park(ctx->sq_data);
-		if (did_park) {
-			task = ctx->sq_data->thread;
+		io_sq_thread_park(ctx->sq_data);
+		task = ctx->sq_data->thread;
+		if (task)
 			atomic_inc(&task->io_uring->in_idle);
-		}
 	}
 
 	io_cancel_defer_files(ctx, task, files);
@@ -8733,10 +8731,10 @@ static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
 	if (!files)
 		io_uring_try_cancel_requests(ctx, task, NULL);
 
-	if (did_park) {
+	if (task)
 		atomic_dec(&task->io_uring->in_idle);
+	if (ctx->sq_data)
 		io_sq_thread_unpark(ctx->sq_data);
-	}
 }
 
 /*
@@ -8836,15 +8834,12 @@ static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
 
 	if (!sqd)
 		return;
-	if (!io_sq_thread_park(sqd))
-		return;
-	tctx = ctx->sq_data->thread->io_uring;
-	/* can happen on fork/alloc failure, just ignore that state */
-	if (!tctx) {
+	io_sq_thread_park(sqd);
+	if (!sqd->thread || !sqd->thread->io_uring) {
 		io_sq_thread_unpark(sqd);
 		return;
 	}
-
+	tctx = ctx->sq_data->thread->io_uring;
 	atomic_inc(&tctx->in_idle);
 	do {
 		/* read completions before cancelations */