io_uring: do ctx initiated file note removal

Another preparation patch. When full quiesce is done on ctx exit, use
task_work infra to remove corresponding to the ctx io_uring->xa entries.
For that we use the back tctx map. Also use ->in_idle to prevent
removing it while we traversing ->xa on cancellation, just ignore it.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/io_uring.c

@@ -987,6 +987,7 @@ static const struct io_op_def io_op_defs[] = {
 	[IORING_OP_UNLINKAT] = {},
 };
 
+static void io_uring_del_task_file(unsigned long index);
 static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
 					 struct task_struct *task,
 					 struct files_struct *files);
@@ -8536,10 +8537,33 @@ static bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
 	return executed;
 }
 
+struct io_tctx_exit {
+	struct callback_head		task_work;
+	struct completion		completion;
+	unsigned long			index;
+};
+
+static void io_tctx_exit_cb(struct callback_head *cb)
+{
+	struct io_uring_task *tctx = current->io_uring;
+	struct io_tctx_exit *work;
+
+	work = container_of(cb, struct io_tctx_exit, task_work);
+	/*
+	 * When @in_idle, we're in cancellation and it's racy to remove the
+	 * node. It'll be removed by the end of cancellation, just ignore it.
+	 */
+	if (!atomic_read(&tctx->in_idle))
+		io_uring_del_task_file(work->index);
+	complete(&work->completion);
+}
+
 static void io_ring_exit_work(struct work_struct *work)
 {
-	struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
+	struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
-					       exit_work);
+	struct io_tctx_exit exit;
+	struct io_tctx_node *node;
+	int ret;
 
 	/*
 	 * If we're doing polled IO and end up having requests being
@@ -8550,6 +8574,26 @@ static void io_ring_exit_work(struct work_struct *work)
 	do {
 		io_uring_try_cancel_requests(ctx, NULL, NULL);
 	} while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
+
+	mutex_lock(&ctx->uring_lock);
+	while (!list_empty(&ctx->tctx_list)) {
+		node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
+					ctx_node);
+		exit.index = (unsigned long)node->file;
+		init_completion(&exit.completion);
+		init_task_work(&exit.task_work, io_tctx_exit_cb);
+		ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
+		if (WARN_ON_ONCE(ret))
+			continue;
+		wake_up_process(node->task);
+
+		mutex_unlock(&ctx->uring_lock);
+		wait_for_completion(&exit.completion);
+		cond_resched();
+		mutex_lock(&ctx->uring_lock);
+	}
+	mutex_unlock(&ctx->uring_lock);
+
 	io_ring_ctx_free(ctx);
 }