nvmet: use a private workqueue instead of the system workqueue

Any attempt to flush kernel-global WQs has possibility of deadlock
so we should simply stop using them, instead introduce nvmet_wq
which is the generic nvmet workqueue for work elements that
don't explicitly require a dedicated workqueue (by the mere fact
that they are using the system_wq).

Changes were done using the following replaces:

- s/schedule_work(/queue_work(nvmet_wq, /g
- s/schedule_delayed_work(/queue_delayed_work(nvmet_wq, /g
- s/flush_scheduled_work()/flush_workqueue(nvmet_wq)/g

Reported-by: Tetsuo Handa <[e-mail]>
Signed-off-by: Sagi Grimberg <[e-mail]>
Reviewed-by: Chaitanya Kulkarni <[e-mail]>
Signed-off-by: Christoph Hellwig <[e-mail]>

drivers/nvme/target/admin-cmd.c

···
 	ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
 	mutex_unlock(&ctrl->lock);
 
-	schedule_work(&ctrl->async_event_work);
+	queue_work(nvmet_wq, &ctrl->async_event_work);
 }
 
 void nvmet_execute_keep_alive(struct nvmet_req *req)

drivers/nvme/target/configfs.c

···
 	struct nvmet_port *port = to_nvmet_port(item);
 
 	/* Let inflight controllers teardown complete */
-	flush_scheduled_work();
+	flush_workqueue(nvmet_wq);
 	list_del(&port->global_entry);
 
 	kfree(port->ana_state);

drivers/nvme/target/core.c

···
 static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX];
 static DEFINE_IDA(cntlid_ida);
 
+struct workqueue_struct *nvmet_wq;
+EXPORT_SYMBOL_GPL(nvmet_wq);
+
 /*
  * This read/write semaphore is used to synchronize access to configuration
  * information on a target system that will result in discovery log page
···
 	list_add_tail(&aen->entry, &ctrl->async_events);
 	mutex_unlock(&ctrl->lock);
 
-	schedule_work(&ctrl->async_event_work);
+	queue_work(nvmet_wq, &ctrl->async_event_work);
 }
 
 static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
···
 	if (reset_tbkas) {
 		pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
 			ctrl->cntlid);
-		schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+		queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
 		return;
 	}
 
···
 	pr_debug("ctrl %d start keep-alive timer for %d secs\n",
 		ctrl->cntlid, ctrl->kato);
 
-	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
+	queue_delayed_work(nvmet_wq, &ctrl->ka_work, ctrl->kato * HZ);
 }
 
 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
···
 	mutex_lock(&ctrl->lock);
 	if (!(ctrl->csts & NVME_CSTS_CFS)) {
 		ctrl->csts |= NVME_CSTS_CFS;
-		schedule_work(&ctrl->fatal_err_work);
+		queue_work(nvmet_wq, &ctrl->fatal_err_work);
 	}
 	mutex_unlock(&ctrl->lock);
 }
···
 		goto out_free_zbd_work_queue;
 	}
 
+	nvmet_wq = alloc_workqueue("nvmet-wq", WQ_MEM_RECLAIM, 0);
+	if (!nvmet_wq) {
+		error = -ENOMEM;
+		goto out_free_buffered_work_queue;
+	}
+
 	error = nvmet_init_discovery();
 	if (error)
-		goto out_free_work_queue;
+		goto out_free_nvmet_work_queue;
 
 	error = nvmet_init_configfs();
 	if (error)
···
 
 out_exit_discovery:
 	nvmet_exit_discovery();
-out_free_work_queue:
+out_free_nvmet_work_queue:
+	destroy_workqueue(nvmet_wq);
+out_free_buffered_work_queue:
 	destroy_workqueue(buffered_io_wq);
 out_free_zbd_work_queue:
 	destroy_workqueue(zbd_wq);
···
 	nvmet_exit_configfs();
 	nvmet_exit_discovery();
 	ida_destroy(&cntlid_ida);
+	destroy_workqueue(nvmet_wq);
 	destroy_workqueue(buffered_io_wq);
 	destroy_workqueue(zbd_wq);
 

drivers/nvme/target/fc.c

···
 	list_for_each_entry_rcu(assoc, &tgtport->assoc_list, a_list) {
 		if (!nvmet_fc_tgt_a_get(assoc))
 			continue;
-		if (!schedule_work(&assoc->del_work))
+		if (!queue_work(nvmet_wq, &assoc->del_work))
 			/* already deleting - release local reference */
 			nvmet_fc_tgt_a_put(assoc);
 	}
···
 			continue;
 		assoc->hostport->invalid = 1;
 		noassoc = false;
-		if (!schedule_work(&assoc->del_work))
+		if (!queue_work(nvmet_wq, &assoc->del_work))
 			/* already deleting - release local reference */
 			nvmet_fc_tgt_a_put(assoc);
 	}
···
 		nvmet_fc_tgtport_put(tgtport);
 
 		if (found_ctrl) {
-			if (!schedule_work(&assoc->del_work))
+			if (!queue_work(nvmet_wq, &assoc->del_work))
 				/* already deleting - release local reference */
 				nvmet_fc_tgt_a_put(assoc);
 			return;
···
 	iod->rqstdatalen = lsreqbuf_len;
 	iod->hosthandle = hosthandle;
 
-	schedule_work(&iod->work);
+	queue_work(nvmet_wq, &iod->work);
 
 	return 0;
 }

drivers/nvme/target/fcloop.c

···
 		spin_lock(&rport->lock);
 		list_add_tail(&rport->ls_list, &tls_req->ls_list);
 		spin_unlock(&rport->lock);
-		schedule_work(&rport->ls_work);
+		queue_work(nvmet_wq, &rport->ls_work);
 		return ret;
 	}
 
···
 		spin_lock(&rport->lock);
 		list_add_tail(&rport->ls_list, &tls_req->ls_list);
 		spin_unlock(&rport->lock);
-		schedule_work(&rport->ls_work);
+		queue_work(nvmet_wq, &rport->ls_work);
 	}
 
 	return 0;
···
 		spin_lock(&tport->lock);
 		list_add_tail(&tport->ls_list, &tls_req->ls_list);
 		spin_unlock(&tport->lock);
-		schedule_work(&tport->ls_work);
+		queue_work(nvmet_wq, &tport->ls_work);
 		return ret;
 	}
 
···
 		spin_lock(&tport->lock);
 		list_add_tail(&tport->ls_list, &tls_req->ls_list);
 		spin_unlock(&tport->lock);
-		schedule_work(&tport->ls_work);
+		queue_work(nvmet_wq, &tport->ls_work);
 	}
 
 	return 0;
···
 	tgt_rscn->tport = tgtport->private;
 	INIT_WORK(&tgt_rscn->work, fcloop_tgt_rscn_work);
 
-	schedule_work(&tgt_rscn->work);
+	queue_work(nvmet_wq, &tgt_rscn->work);
 }
 
 static void
···
 	INIT_WORK(&tfcp_req->tio_done_work, fcloop_tgt_fcprqst_done_work);
 	kref_init(&tfcp_req->ref);
 
-	schedule_work(&tfcp_req->fcp_rcv_work);
+	queue_work(nvmet_wq, &tfcp_req->fcp_rcv_work);
 
 	return 0;
 }
···
 {
 	struct fcloop_fcpreq *tfcp_req = tgt_fcp_req_to_fcpreq(tgt_fcpreq);
 
-	schedule_work(&tfcp_req->tio_done_work);
+	queue_work(nvmet_wq, &tfcp_req->tio_done_work);
 }
 
 static void
···
 
 	if (abortio)
 		/* leave the reference while the work item is scheduled */
-		WARN_ON(!schedule_work(&tfcp_req->abort_rcv_work));
+		WARN_ON(!queue_work(nvmet_wq, &tfcp_req->abort_rcv_work));
 	else  {
 		/*
 		 * as the io has already had the done callback made,

drivers/nvme/target/io-cmd-file.c

···
 	if (!nvmet_check_transfer_len(req, 0))
 		return;
 	INIT_WORK(&req->f.work, nvmet_file_flush_work);
-	schedule_work(&req->f.work);
+	queue_work(nvmet_wq, &req->f.work);
 }
 
 static void nvmet_file_execute_discard(struct nvmet_req *req)
···
 	if (!nvmet_check_data_len_lte(req, nvmet_dsm_len(req)))
 		return;
 	INIT_WORK(&req->f.work, nvmet_file_dsm_work);
-	schedule_work(&req->f.work);
+	queue_work(nvmet_wq, &req->f.work);
 }
 
 static void nvmet_file_write_zeroes_work(struct work_struct *w)
···
 	if (!nvmet_check_transfer_len(req, 0))
 		return;
 	INIT_WORK(&req->f.work, nvmet_file_write_zeroes_work);
-	schedule_work(&req->f.work);
+	queue_work(nvmet_wq, &req->f.work);
 }
 
 u16 nvmet_file_parse_io_cmd(struct nvmet_req *req)

drivers/nvme/target/loop.c

···
 		iod->req.transfer_len = blk_rq_payload_bytes(req);
 	}
 
-	schedule_work(&iod->work);
+	queue_work(nvmet_wq, &iod->work);
 	return BLK_STS_OK;
 }
 
···
 		return;
 	}
 
-	schedule_work(&iod->work);
+	queue_work(nvmet_wq, &iod->work);
 }
 
 static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl,

drivers/nvme/target/nvmet.h

···
 
 extern struct workqueue_struct *buffered_io_wq;
 extern struct workqueue_struct *zbd_wq;
+extern struct workqueue_struct *nvmet_wq;
 
 static inline void nvmet_set_result(struct nvmet_req *req, u32 result)
 {

drivers/nvme/target/passthru.c

···
 	if (req->p.use_workqueue || effects) {
 		INIT_WORK(&req->p.work, nvmet_passthru_execute_cmd_work);
 		req->p.rq = rq;
-		schedule_work(&req->p.work);
+		queue_work(nvmet_wq, &req->p.work);
 	} else {
 		rq->end_io_data = req;
 		blk_execute_rq_nowait(rq, false, nvmet_passthru_req_done);

drivers/nvme/target/rdma.c

···
 
 	if (queue->host_qid == 0) {
 		/* Let inflight controller teardown complete */
-		flush_scheduled_work();
+		flush_workqueue(nvmet_wq);
 	}
 
 	ret = nvmet_rdma_cm_accept(cm_id, queue, &event->param.conn);
···
 
 	if (disconnect) {
 		rdma_disconnect(queue->cm_id);
-		schedule_work(&queue->release_work);
+		queue_work(nvmet_wq, &queue->release_work);
 	}
 }
 
···
 	mutex_unlock(&nvmet_rdma_queue_mutex);
 
 	pr_err("failed to connect queue %d\n", queue->idx);
-	schedule_work(&queue->release_work);
+	queue_work(nvmet_wq, &queue->release_work);
 }
 
 /**
···
 		if (!queue) {
 			struct nvmet_rdma_port *port = cm_id->context;
 
-			schedule_delayed_work(&port->repair_work, 0);
+			queue_delayed_work(nvmet_wq, &port->repair_work, 0);
 			break;
 		}
 		fallthrough;
···
 	nvmet_rdma_disable_port(port);
 	ret = nvmet_rdma_enable_port(port);
 	if (ret)
-		schedule_delayed_work(&port->repair_work, 5 * HZ);
+		queue_delayed_work(nvmet_wq, &port->repair_work, 5 * HZ);
 }
 
 static int nvmet_rdma_add_port(struct nvmet_port *nport)
···
 	}
 	mutex_unlock(&nvmet_rdma_queue_mutex);
 
-	flush_scheduled_work();
+	flush_workqueue(nvmet_wq);
 }
 
 static struct ib_client nvmet_rdma_ib_client = {

drivers/nvme/target/tcp.c

···
 	spin_lock(&queue->state_lock);
 	if (queue->state != NVMET_TCP_Q_DISCONNECTING) {
 		queue->state = NVMET_TCP_Q_DISCONNECTING;
-		schedule_work(&queue->release_work);
+		queue_work(nvmet_wq, &queue->release_work);
 	}
 	spin_unlock(&queue->state_lock);
 }
···
 		goto out;
 
 	if (sk->sk_state == TCP_LISTEN)
-		schedule_work(&port->accept_work);
+		queue_work(nvmet_wq, &port->accept_work);
 out:
 	read_unlock_bh(&sk->sk_callback_lock);
 }
···
 
 	if (sq->qid == 0) {
 		/* Let inflight controller teardown complete */
-		flush_scheduled_work();
+		flush_workqueue(nvmet_wq);
 	}
 
 	queue->nr_cmds = sq->size * 2;
···
 
 	nvmet_unregister_transport(&nvmet_tcp_ops);
 
-	flush_scheduled_work();
+	flush_workqueue(nvmet_wq);
 	mutex_lock(&nvmet_tcp_queue_mutex);
 	list_for_each_entry(queue, &nvmet_tcp_queue_list, queue_list)
 		kernel_sock_shutdown(queue->sock, SHUT_RDWR);
 	mutex_unlock(&nvmet_tcp_queue_mutex);
-	flush_scheduled_work();
+	flush_workqueue(nvmet_wq);
 
 	destroy_workqueue(nvmet_tcp_wq);
 }