if connection is lost while connected modul now waits for new connection

3 files changed, 47 insertions, 22 deletions

diff --git a/src/kernel/mq.c b/src/kernel/mq.c
index 5b480de..f859c0f 100644
--- a/src/kernel/mq.c
+++ b/src/kernel/mq.c
@@ -23,6 +23,7 @@
 
 #include <linux/wait.h>
 #include <linux/blk-mq.h>
+#include <linux/delay.h>
 
 /**
  * dnbd3_busy_iter - iterator to set a bit for busy reqeust
@@ -81,10 +82,18 @@ static void dnbd3_busy_iter_requeue(struct request *req, void *priv, bool arg)
 {
 	struct dnbd3_sock *sock = priv;
 	struct dnbd3_cmd *cmd = blk_mq_rq_to_pdu(req);
+	if (&cmd->dnbd3->socks[cmd->index] != sock) {
+		return;
+	}
 	if (!mutex_trylock(&cmd->lock)) {
 		/* request is in sending, so we will not requeue */
 		return;
 	}
+	if (!cmd->dnbd3->connected) {
+		dnbd3_end_cmd(cmd, BLK_STS_IOERR);
+		mutex_unlock(&cmd->lock);
+		return;
+	}
 	debug_sock(sock, "requeue busy request %p", req);
 	dnbd3_requeue_cmd(cmd);
 	mutex_unlock(&cmd->lock);
@@ -100,7 +109,21 @@ void dndb3_reque_busy_requests(struct dnbd3_sock *sock)
 	blk_mq_tagset_busy_iter(set,(busy_tag_iter_fn *) dnbd3_busy_iter_requeue, sock);
 }
 
-
+static void dnbd3_wait_for_sockets(struct dnbd3_cmd * cmd) {
+	int i;
+	bool wait = true;
+	while (wait && cmd->dnbd3->connected) {
+		for (i = 0; i < cmd->dnbd3->number_connections; i++) {
+			if (dnbd3_is_sock_alive(cmd->dnbd3->socks[cmd->index])) {
+				wait = false;
+			}
+			if (wait && cmd->dnbd3->connected) {
+				debug("no socket alive sleeping");
+				msleep(100);
+			}
+		}
+	}
+}
 
 /**
  * dnbd3_requeue_cmd - requeue a command once
@@ -109,7 +132,9 @@ void dndb3_reque_busy_requests(struct dnbd3_sock *sock)
 void dnbd3_requeue_cmd(struct dnbd3_cmd *cmd)
 {
 	struct request *req = blk_mq_rq_from_pdu(cmd);
-	if (!cmd->requed && blk_queued_rq(req)) {
+	dnbd3_wait_for_sockets(cmd);
+
+	if (!cmd->requed/* && blk_queued_rq(req)*/) {
 		cmd->requed = true;
 		blk_mq_requeue_request(req, true);
 	}
@@ -127,6 +152,11 @@ void dnbd3_end_cmd(struct dnbd3_cmd *cmd, blk_status_t error)
 }
 
 
+void dnbd3_end_all_rq(struct dnbd3_device *dev)
+{
+	struct request_queue *q = dev->disk->queue;
+}
+
 /**
  * dnbd3_handle_cmd - handles a mq command
  * @cmd: the cmd to send
@@ -143,20 +173,16 @@ static int dnbd3_handle_cmd(struct dnbd3_cmd *cmd, int index)
 //			blk_rq_pos(req), blk_rq_bytes(req), index);
 
 
-	if (!dnbd3_is_sock_alive(*sock)) {
-//		warn_dev(dev, "attempted send on invalid socket %d", index);
-//		msleep(SOCKET_TIMEOUT_CLIENT_DATA * 1000);
+	if (!dev->socks || !dnbd3_is_sock_alive(*sock)) {
 
 		if (dev->connected) {
-//			info_dev(dev, "reset request to new socket");
+			info_dev(dev, "reset request to new socket");
 			dnbd3_requeue_cmd(cmd);
-			ret = 0;
-			goto out;
+			return 0;
 		} else {
 			error_dev(dev, "ending request device not connected");
 			dnbd3_end_cmd(cmd, BLK_STS_IOERR);
-			ret = -EIO;
-			goto out;
+			return -EIO;
 		}
 	}
 
diff --git a/src/kernel/net-txrx.c b/src/kernel/net-txrx.c
index 4ca3fa5..87a9169 100644
--- a/src/kernel/net-txrx.c
+++ b/src/kernel/net-txrx.c
@@ -163,6 +163,10 @@ int dnbd3_send_request(struct dnbd3_sock *sock, struct request *req,
 	iov[0].iov_base = &request;
 	iov[0].iov_len = sizeof(request);
 	lng = iov_num == 1 ? iov[0].iov_len : iov[0].iov_len + iov[1].iov_len;
+	if (unlikely(!sock->sock)) {
+		warn_sock(sock, "socket was shutdown while sending");
+		goto error;
+	}
 	result = kernel_sendmsg(sock->sock, &msg, iov, iov_num, lng);
 	if (result != lng) {
 		error_sock(sock, "connection to server lost");
@@ -336,8 +340,11 @@ int dnbd3_receive_cmd_get_block_mq(struct dnbd3_sock *sock,
 	if (!req || !blk_mq_request_started(req)) {
 		error_sock(sock, "unexpected reply (%d) %p", tag, req);
 		if (req) {
+			cmd = blk_mq_rq_to_pdu(req);
+			mutex_lock(&cmd->lock);
 			debug_sock(sock, "requeue request");
-			dnbd3_requeue_cmd(blk_mq_rq_to_pdu(req));
+			dnbd3_requeue_cmd(cmd);
+			mutex_unlock(&cmd->lock);
 		}
 		dnbd3_clear_socket(sock, reply, remaining);
 		return -EIO;
diff --git a/src/kernel/net.c b/src/kernel/net.c
index 801c15c..5f9846a 100644
--- a/src/kernel/net.c
+++ b/src/kernel/net.c
@@ -384,7 +384,7 @@ static void dnbd3_panic_worker(struct work_struct *work)
 		mutex_lock(&dev->device_lock);
 		if (dnbd3_adjust_connections(dev)) {
 			error_dev(dev, "failed to connect to any server");
-			dev->connected = false;
+//			dev->connected = false;
 
 		}
 		mutex_unlock(&dev->device_lock);
@@ -584,7 +584,7 @@ static void dnbd3_discovery_worker(struct work_struct *work)
 	mutex_lock(&dev->device_lock);
 	if (dnbd3_adjust_connections(dev)) {
 		error_dev(dev, "failed to connect to any server");
-		dev->connected = false;
+//		dev->connected = false;
 
 	}
 	mutex_unlock(&dev->device_lock);
@@ -836,14 +836,6 @@ int dnbd3_net_connect(struct dnbd3_device *dev)
 	dev->timer.expires = jiffies + HZ;
 	add_timer(&dev->timer);
 
-	// alt_server[0] is the initial server
-//	result = dnbd3_server_connect(dev, &dev->alt_servers[0]);
-//	if (result) {
-//		error_dev(dev, "failed to connect to initial server");
-//		result = -ENOENT;
-//		dev->imgname = NULL;
-//		dev->socks[0].server = NULL;
-//	}
 	return result;
 }
 
@@ -860,6 +852,7 @@ int dnbd3_net_disconnect(struct dnbd3_device *dev)
 		info_dev(dev, "not connected");
 	}
 
+	dev->connected = false;
 	del_timer_sync(&dev->timer);
 	/* be sure it does not recover while disconnecting */
 	cancel_work_sync(&dev->discovery_worker);
@@ -876,7 +869,6 @@ int dnbd3_net_disconnect(struct dnbd3_device *dev)
 	}
 	kfree(dev->socks);
 	dev->socks = NULL;
-	dev->connected = false;
 	return result;
 }