changed strategy to connect to diffrent sockets

4 files changed, 175 insertions, 381 deletions

diff --git a/src/kernel/core.c b/src/kernel/core.c
index 8bb23d9..6f0dab4 100644
--- a/src/kernel/core.c
+++ b/src/kernel/core.c
@@ -72,6 +72,8 @@ static void dnbd3_requeue_cmd(struct dnbd3_cmd *cmd)
 	if (!cmd->requed) {
 		cmd->requed = true;
 		blk_mq_requeue_request(req, true);
+	} else {
+		blk_mq_end_request(req, BLK_STS_IOERR);
 	}
 }
 
@@ -84,47 +86,22 @@ static int dnbd3_handle_cmd(struct dnbd3_cmd *cmd, int index)
 {
 	struct request *req = blk_mq_rq_from_pdu(cmd);
 	struct dnbd3_device *dev = cmd->dnbd3;
-	struct dnbd3_sock *sock = NULL;
-	bool first_try = true;
+	struct dnbd3_sock *sock = &dev->socks[index];
 	int ret = -1;
-	int i;
-	int sock_alive = 0;
 
-	debug_dev(dev, "handle request at position %lu, size %d, index %d",
-			blk_rq_pos(req), blk_rq_bytes(req), index);
+//	debug_dev(dev, "handle request at position %lu, size %d, index %d",
+//			blk_rq_pos(req), blk_rq_bytes(req), index);
 
-again:
 
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (dnbd3_is_sock_alive(dev->socks[i])) {
-			if (index == sock_alive) {
-				sock = &dev->socks[i];
-			}
-			sock_alive++;
-		}
-	}
+	if (!(sock->server && sock->sock && !sock->panic)) {
+//		warn_dev(dev, "attempted send on invalid socket %d", index);
+		msleep(SOCKET_TIMEOUT_CLIENT_DATA * 1000);
 
-	if (!sock) { // TODO let the mq queues be the same number as NUMBER_CONNECTIONS than handle them
-		warn_dev(dev, "index is %d but no socket was found", index);
-		dev_err_ratelimited(disk_to_dev(dev->disk), "attempted send on invalid socket\n");
-		if (sock_alive > 0) {
-			blk_mq_update_nr_hw_queues(&dev->tag_set, sock_alive);
-			dnbd3_requeue_cmd(cmd);
-			ret = 0;
-			goto out;
-		}
-		if (first_try) {
-			debug_dev(dev, "no socket found, going to sleep");
-			msleep(SOCKET_TIMEOUT_CLIENT_DATA * 1000);
-			first_try = false;
-			goto again;
-		}
-		error_dev(dev, "failed to find a socket, end request");
-		blk_mq_end_request(req, BLK_STS_IOERR);
-		return -EINVAL;
+		dnbd3_requeue_cmd(cmd);
+		ret = 0;
+		goto out;
 	}
 
-
 	cmd->status = BLK_STS_OK;
 
 	mutex_lock(&sock->tx_lock);
@@ -164,7 +141,7 @@ static blk_status_t dnbd3_queue_rq(struct blk_mq_hw_ctx *hctx,
 	int ret;
 
 	mutex_lock(&cmd->lock);
-	cmd->requed = false;
+//	cmd->requed = false;
 
 	ret = dnbd3_handle_cmd(cmd, hctx->queue_num);
 	if (ret < 0) {
@@ -290,7 +267,6 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode,
 		}
 	}
 
-	mutex_lock(&dev->device_lock);
 	switch (cmd) {
 	case IOCTL_OPEN:
 		debug_dev(dev, "ioctl open");
@@ -301,13 +277,14 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode,
 		} else if (msg == NULL) {
 			result = -EINVAL;
 		} else {
+			mutex_lock(&dev->device_lock);
 			if (sizeof(msg->host) != sizeof(dev->initial_server.host)) {
 				warn_dev(dev, "odd size bug#1 triggered in ioctl");
 			}
 			memcpy(&dev->initial_server.host, &msg->host,
 					sizeof(msg->host));
 			dev->initial_server.failures = 0;
-			dnbd3_set_rtt_unreachable(&dev->initial_server)
+			dnbd3_set_rtt_unknown(&dev->initial_server)
 			dev->imgname = imgname;
 			dev->rid = msg->rid;
 			dev->use_server_provided_alts =
@@ -320,6 +297,9 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode,
 					* NUMBER_SERVERS);
 			memcpy(dev->alt_servers, &dev->initial_server,
 					sizeof(dev->alt_servers[0]));
+
+			mutex_unlock(&dev->device_lock);
+
 			result = dnbd3_net_connect(dev);
 			imgname = NULL;
 		}
@@ -352,12 +332,14 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode,
 		} else if (msg == NULL) {
 			result = -EINVAL;
 		} else {
+			mutex_lock(&dev->device_lock);
 			memcpy(&dev->new_servers[dev->new_servers_num].host,
 					&msg->host, sizeof(msg->host));
 			/* 0 = ADD, 1 = REM */
 			dev->new_servers[dev->new_servers_num].failures =
 					(cmd == IOCTL_ADD_SRV ? 0 : 1);
-			++dev->new_servers_num;
+			dev->new_servers_num++;
+			mutex_unlock(&dev->device_lock);
 			result = 0;
 		}
 		break;
@@ -372,7 +354,6 @@ static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode,
 		result = -EIO;
 		break;
 	}
-	mutex_unlock(&dev->device_lock);
 error:
 	if (msg) kfree(msg);
 	if (imgname) kfree(imgname);
@@ -435,7 +416,7 @@ int dnbd3_add_device(struct dnbd3_device *dev, int minor)
 	dev->disk = disk;
 	dev->tag_set.ops = &dnbd3_mq_ops;
 	/* this can be changed later with blk_mq_update_nr_hw_queues() */
-	dev->tag_set.nr_hw_queues = 1;
+	dev->tag_set.nr_hw_queues = NUMBER_CONNECTIONS;
 	dev->tag_set.queue_depth = 128;
 	dev->tag_set.numa_node = NUMA_NO_NODE;
 	dev->tag_set.cmd_size = sizeof(struct dnbd3_cmd);
diff --git a/src/kernel/dnbd3.h b/src/kernel/dnbd3.h
index 2c3ec39..8fabc35 100644
--- a/src/kernel/dnbd3.h
+++ b/src/kernel/dnbd3.h
@@ -39,6 +39,9 @@
 
 
 
+
+#define RTT_FACTOR(rtt) (((rtt) * 3) / 2)
+
 /**
  * limit to which the other connected servers are only allowed to be that worser
  * then the best rtt
@@ -72,6 +75,7 @@ extern struct workqueue_struct *dnbd3_wq;
 struct dnbd3_server {
 	dnbd3_host_t host;
 	uint64_t rtts[4];
+	uint64_t avg_rtt;
 	uint16_t protocol_version;
 	uint16_t failures; // TODO failures runterzaehlen wenn wieder verbindung moeglich, seltener messen bei hohem failure count
 
@@ -208,7 +212,14 @@ struct dnbd3_cmd {
 
 #define dnbd3_set_rtt_unreachable(server) \
 	(server)->rtts[0] = (server)->rtts[1] = (server)->rtts[2] \
-			= (server)->rtts[3] = RTT_UNREACHABLE;
+			= (server)->rtts[3] = (server)->avg_rtt = \
+			RTT_UNREACHABLE;
+
+
+#define dnbd3_set_rtt_unknown(server) \
+	(server)->rtts[0] = (server)->rtts[1] = (server)->rtts[2] \
+			= (server)->rtts[3] = (server)->avg_rtt = \
+			RTT_UNREACHABLE / 2;
 
 
 /**
diff --git a/src/kernel/net.c b/src/kernel/net.c
index 5478e95..35c49b5 100644
--- a/src/kernel/net.c
+++ b/src/kernel/net.c
@@ -22,6 +22,7 @@
 
 #include <net/sock.h>
 #include <linux/wait.h>
+#include <linux/sort.h>
 
 #include "net.h"
 #include "utils.h"
@@ -65,46 +66,11 @@
 		(h).msg_flags = MSG_WAITALL | MSG_NOSIGNAL; \
 	} while (0)
 
-
-#if NUMBER_CONNECTIONS == 1
-#define dnbd3_pl_socket_connect(dev, server) \
-	dnbd3_socket_connect(dev, server)
-
-#define dnbd3_pl_socket_disconnect(sock) \
-	dnbd3_socket_disconnect(sock)
-
-
-#else
-#define dnbd3_pl_socket_connect(sock, server) \
-	do {\
-		int i; \
-		for (i = 0 < NUMBER_PARALLEL_CONNECTIONS; i++) { \
-			dnbd3_socket_connect(dev, server); \
-			dnbd3_socket_connect((sock) + \
-					(i * sizeof(struct dnbd3_sock)), \
-					(server)); \
-		} \
-	} while (0)
-
-
-#define dnbd3_pl_socket_disconnect(sock) \
-	do { \
-		int i; \
-		for (i = 0 < NUMBER_PARALLEL_CONNECTIONS; i++) { \
-			dnbd3_socket_disconnect((sock) + \
-					(i * sizeof(struct dnbd3_sock))); \
-		} \
-	} while (0)
-
-#endif
-
-static int dnbd3_server_connect(struct dnbd3_device *dev,
-		struct dnbd3_server *server);
+static int __dnbd3_socket_connect(struct dnbd3_sock *sock,
+		struct dnbd3_server * server);
 static int dnbd3_socket_connect(struct dnbd3_sock *sock,
 		struct dnbd3_server * server);
 static int dnbd3_socket_disconnect(struct dnbd3_sock *sock);
-static int dnbd3_server_disconnect(struct dnbd3_device *dev,
-		struct dnbd3_server *server);
 
 
 /*
@@ -216,8 +182,8 @@ int dnbd3_send_request(struct dnbd3_sock *sock, struct request *req,
 		goto error;
 	}
 
-	sock->pending = NULL;
 error:
+	sock->pending = NULL;
 	return result;
 }
 
@@ -738,176 +704,107 @@ static void dnbd3_keepalive_worker(struct work_struct *work)
 }
 
 /**
- * dnbd3_find_best_alt_server - find best alternative server
- * @dev: the device where to search for alternative servers
- *
- * searches for an alternative server which has an rtt better than RTT_THRESOLD
- * of the best connected server
+ * dnbd3_compare_servers - comparator for the server
+ * @lhs: left hand sign
+ * @rhs: right hand sign
  */
-static struct dnbd3_server *dnbd3_find_best_alt_server(struct dnbd3_device *dev)
-{
-	int i, j;
-	uint64_t rtt = 0;
-	uint64_t best_rtt = RTT_UNREACHABLE;
-	uint64_t current_best_rtt = RTT_UNREACHABLE;
-	struct dnbd3_server *best_server = NULL;
-	struct dnbd3_server *server = NULL;
-
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (dnbd3_is_sock_alive(dev->socks[i])) {
-			rtt = dnbd3_avg_rtt(dev->socks[i].server);
-			if (rtt <= current_best_rtt) {
-				current_best_rtt = rtt;
-			}
-		}
+static int dnbd3_compare_servers(const void *lhs, const void *rhs) {
+	uint64_t l, r;
+	struct dnbd3_server *lhs_server = *((struct dnbd3_server **) lhs);
+	struct dnbd3_server *rhs_server = *((struct dnbd3_server **) rhs);
+
+	l = lhs_server->host.type != 0 ? lhs_server->avg_rtt
+			: RTT_UNREACHABLE + 1;
+	r = rhs_server->host.type != 0 ? rhs_server->avg_rtt
+			: RTT_UNREACHABLE + 1;
+	if (l < r) {
+		return -1;
+	} else if (l > r) {
+		return 1;
 	}
-
-	best_rtt = RTT_THRESOULD_LIMIT(current_best_rtt);
-	debug_dev(dev, "best connected rtt is %llu, searching for rtt better than %llu",
-			current_best_rtt, best_rtt);
-
-	for (i = 0; i < NUMBER_SERVERS; i++) {
-		if (dev->alt_servers[i].host.type != 0) {
-			rtt = dnbd3_avg_rtt(&dev->alt_servers[i]);
-			if (rtt <= best_rtt) {
-				server = &dev->alt_servers[i];
-				/*  check if already connected */
-				for (j = 0; j < NUMBER_CONNECTIONS; j++) {
-					if (server == dev->socks[j].server) {
-						server = NULL;
-						break;
-					}
-				}
-				if (server) {
-					best_server = server;
-					best_rtt = rtt;
-				}
-			}
-		}
-	}
-	if (best_server) {
-		debug_server(dev, best_server, "found best alt server with rtt %llu",
-				best_rtt);
-	} else {
-		debug_dev(dev, "did not find any alternative server");
-	}
-	return best_server;
+	return 0;
 }
 
 /**
- * dnbd3_better_rtt - checks if the rtt is better
- * @new_server: the server to check
- * @existing_server: current server
+ * dnbd3_sort_server - sort the alt server according to their avg rtt
+ * @dev: the dndb3 device
  *
- * checks if the rtt is better than RTT_THRESHOLD_FACTOR
+ * the returned array has to be freed with kfree
  */
-static bool dnbd3_better_rtt(struct dnbd3_server *new_server,
-		struct dnbd3_server *existing_server) {
-	uint64_t new_rtt = dnbd3_avg_rtt(new_server);
-	uint64_t existing_rtt = dnbd3_avg_rtt(existing_server);
+static struct dnbd3_server **dnbd3_sort_server(struct dnbd3_device *dev) {
+	int i;
+	struct dnbd3_server **sorted_servers = kmalloc(NUMBER_SERVERS * sizeof(struct dnbd3_device *), GFP_KERNEL);
+	if (!sorted_servers) {
+		return NULL;
+	}
 
-	if (new_rtt < RTT_THRESHOLD_FACTOR(existing_rtt)) {
-		return true;
+	for (i = 0; i < NUMBER_SERVERS; i++) {
+		sorted_servers[i] = &dev->alt_servers[i];
 	}
-	return false;
+	sort(sorted_servers, NUMBER_SERVERS, sizeof(struct dnbd3_device *),
+			&dnbd3_compare_servers, NULL);
+
+	return sorted_servers;
 }
 
 /**
- * dnbd3_adjust_connections - adjust the connections of the device
- * @dev: the device
+ * dnbd3_adjust_connections - create a connection plan and connect
+ * @dev: the dnbd3 device
  *
- * 1. connect empty sockets if best alternative server is found
- * 2. replace slow socket with better server if available
- * 3. remove socket if one is slow
+ * 1. sort the alt server after the avg rtt
+ * 2. create a connection plan
+ * 3. connect the plan
  */
-static void dnbd3_adjust_connections(struct dnbd3_device *dev) {
-	int i;
-	int sock_alive = 0;
-	uint64_t rtt;
-	uint64_t best_rtt = RTT_UNREACHABLE;
-	struct dnbd3_server *server, *existing_server;
+static int dnbd3_adjust_connections(struct dnbd3_device *dev) {
+	int i, j, fallback;
+	struct dnbd3_server *plan[NUMBER_CONNECTIONS];
+	struct dnbd3_server **servers = dnbd3_sort_server(dev);
 
-	/* connect empty sockets */ //TODO check if panic worker connects at the same time, than we would have a problem
-	sock_alive = 0;
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (!dnbd3_is_sock_alive(dev->socks[i])) {
-			server = dnbd3_find_best_alt_server(dev);
-			if (server) {
-				if (dnbd3_server_connect(dev, server) == 0) {
-					sock_alive++;
+	if (servers && servers[0]->host.type != 0) {
+		mutex_lock(&dev->device_lock);
+		plan[0] = servers[0];
+		fallback = 0;
+		j = 1;
+		debug_dev(dev, "connection plan:");
+		debug_server(dev, plan[0], "server 0 with rtt %llu:",
+				plan[0]->avg_rtt);
+		for (i = 1; i < NUMBER_CONNECTIONS; i++) {
+			if (servers[j]->host.type != 0) {
+				if (RTT_FACTOR(plan[i - 1]->avg_rtt) >
+						servers[j]->avg_rtt) {
+					plan[i] = servers[j];
+					j++;
 				} else {
-					warn_server(dev, server, "failed to connect");
+					plan[i] = plan[fallback];
+					fallback++;
 				}
+			} else {
+				plan[i] = plan[fallback];
+				fallback++;
 			}
-		} else {
-			sock_alive++;
+			debug_server(dev, plan[i], "server %d with rtt %llu:",
+					i, plan[i]->avg_rtt);
 		}
-	}
+		kfree(servers);
 
-	/* replace socket with better server */
-	if (sock_alive == NUMBER_CONNECTIONS) {
 		for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-			if (dnbd3_is_sock_alive(dev->socks[i])) {
-				server = dnbd3_find_best_alt_server(dev);
-				existing_server = dev->socks[i].server;
-				if (server && dnbd3_better_rtt(server, dev->socks[i].server)) {
+			if (plan[i] != dev->socks[i].server) {
+				if (dnbd3_is_sock_alive(dev->socks[i])) {
 					dnbd3_socket_disconnect(&dev->socks[i]);
-
-					/* if it fails reconnect to existing */
-					if (dnbd3_server_connect(dev, server) != 0) {
-						warn_server(dev, server, "failed to connect");
-						dnbd3_server_connect(dev, existing_server);
-					}
-				}
-			}
-		}
-	}
-
-	/* remove a socket if it is much slower than the others */
-	if (sock_alive > 1) {
-		for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-			if (dnbd3_is_sock_alive(dev->socks[i])) {
-				rtt = dnbd3_avg_rtt(dev->socks[i].server);
-				if (rtt <= best_rtt) {
-					best_rtt = rtt;
 				}
+				dnbd3_socket_connect(&dev->socks[i], plan[i]);
 			}
 		}
-		for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-			if (dnbd3_is_sock_alive(dev->socks[i])) {
-				rtt = dnbd3_avg_rtt(dev->socks[i].server);
-				if (rtt > RTT_THRESOULD_LIMIT(best_rtt)) {
-					info_sock(&dev->socks[i], "removing connection with rtt %llu", rtt);
-					dnbd3_socket_disconnect(&dev->socks[i]);
-					sock_alive--;
-				}
-			}
+		mutex_unlock(&dev->device_lock);
+		return 0;
+	} else { /* there is nothing to connect */
+		if (servers) {
+			kfree(servers);
 		}
+		return -ENONET;
 	}
 
-	info_dev(dev, "connected to %d/%d sockets", sock_alive, NUMBER_CONNECTIONS);
-}
 
-/**
- * dnbd3_panic_connect - connect to the first available server
- * @dev: the device
- */
-static int dnbd3_panic_connect(struct dnbd3_device *dev) {
-	int result, i;
-	result = dnbd3_server_connect(dev, &dev->initial_server);
-	if (result) {
-		for (i = 0; i < NUMBER_SERVERS; i++) {
-			if (dev->alt_servers[i].host.type != 0) {
-				result = dnbd3_server_connect(dev,
-						&dev->alt_servers[i]);
-				if (!result) {
-					info_server(dev, &dev->alt_servers[i], "found server to connect to");
-					break;
-				}
-			}
-		}
-	}
-	return result;
 }
 
 /**
@@ -921,47 +818,27 @@ static int dnbd3_panic_connect(struct dnbd3_device *dev) {
 static void dnbd3_panic_worker(struct work_struct *work)
 {
 	struct dnbd3_device *dev;
-	struct dnbd3_sock *panicked_sock = NULL;
-	struct dnbd3_server *new_server, *panicked_server;
+	bool panic = false;
 	int i;
 	int sock_alive = 0;
 	dev = container_of(work, struct dnbd3_device, panic_worker);
 	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
 		if (dev->socks[i].panic) {
-			panicked_sock = &dev->socks[i];
+			panic = true;
+			dnbd3_set_rtt_unreachable(dev->socks[i].server);
+			dnbd3_socket_disconnect(&dev->socks[i]);
+
 		} else if (dnbd3_is_sock_alive(dev->socks[i])) {
 			sock_alive++;
 		}
 	}
 
-	if (panicked_sock) {
-		warn_sock(panicked_sock, "panicked, connections still alive %d",
-				sock_alive);
-		panicked_server = panicked_sock->server;
-		new_server = dnbd3_find_best_alt_server(dev);
 
-		dnbd3_socket_disconnect(panicked_sock);
-		dnbd3_set_rtt_unreachable(panicked_server);
+	if (panic) {
+		warn_dev(dev, "panicked, connections still alive %d",
+				sock_alive);
 
-		if (new_server != NULL && new_server != panicked_server) {
-			info_server(dev, new_server, "found replacement");
-			if (!dnbd3_server_connect(dev, new_server)) {
-				sock_alive++;
-			}
-		} else if (sock_alive > 0) {
-			info_dev(dev, "found no replacement server but still connected to %d servers",
-					sock_alive);
-		}
-	}
-
-	if (sock_alive == 0) {
-		warn_dev(dev, "did not find a good server, trying to connect to any available server");
-
-		if (dnbd3_panic_connect(dev)) {
-			error_dev(dev, "could not connect to any server");
-		} else {
-			info_dev(dev, "found server to connect to");
-		}
+		dnbd3_adjust_connections(dev);
 	}
 }
 
@@ -985,7 +862,7 @@ static int dnbd3_meassure_rtt(struct dnbd3_device *dev,
 			.server = server
 	};
 
-	result = dnbd3_socket_connect(&sock, server);
+	result = __dnbd3_socket_connect(&sock, server);
 	if (result) {
 		error_sock(&sock, "socket connect failed in rtt measurement");
 		goto error;
@@ -1044,6 +921,7 @@ static int dnbd3_meassure_rtt(struct dnbd3_device *dev,
 
 error:
 	sock.server->rtts[dev->discovery_count % 4] = rtt;
+	sock.server->avg_rtt = dnbd3_avg_rtt(sock.server);
 	if (result <= 0) {
 		server->failures++;
 	}
@@ -1092,8 +970,9 @@ static void dnbd3_merge_new_server(struct dnbd3_device *dev,
 		if (new_server->failures == 1) { /* remove is requested */
 			info_server(dev, new_server,
 					"remove server is requested");
-			dnbd3_server_disconnect(dev, existing_server);
+			// adjust connection will remove it later
 			existing_server->host.type = 0;
+			dnbd3_set_rtt_unreachable(existing_server);
 		}
 //		existing_server->failures = 0; // reset failure count
 		return;
@@ -1110,9 +989,10 @@ static void dnbd3_merge_new_server(struct dnbd3_device *dev,
 	info_server(dev, free_server, "got new alternative server");
 	free_server->failures = 0;
 	free_server->protocol_version = 0;
-	dnbd3_set_rtt_unreachable(free_server);
+	dnbd3_set_rtt_unknown(free_server);
 }
 
+
 /**
  * dnbd3_discovery_worker - handle discovery
  * @work: the work used to get the dndb3_device
@@ -1168,11 +1048,11 @@ static void dnbd3_discovery_worker(struct work_struct *work)
 
 
 /**
- * dnbd3_socket_connect - connect a socket to a server
+ * __dnbd3_socket_connect - internal connect a socket to a server
  * @sock: the socket to connect
  * @server: the server
  */
-static int dnbd3_socket_connect(struct dnbd3_sock *sock,
+static int __dnbd3_socket_connect(struct dnbd3_sock *sock,
 		struct dnbd3_server *server)
 {
 	int result = 0;
@@ -1237,44 +1117,27 @@ error:
 }
 
 /**
- * dnbd3_server_connect - connect a server to a device
- * @dev: the device
+ * dnbd3_socket_connect - connect a socket to a server
+ * @sock: the socket
  * @server: the server to connect
  *
- * 1. connects the server to a free socket if available
+ * 1. connects the server to the socket
  * 2. select the image
  * 3. start receiver_worker and keepalive_worker
- * 4. if it is the first connection start timer, panic_worker and
- *    keepalive_worker
- * 5. update the mq queues to the number of sockets alive
  */
-static int dnbd3_server_connect(struct dnbd3_device *dev,
+static int dnbd3_socket_connect(struct dnbd3_sock *sock,
 		struct dnbd3_server *server)
 {
-	int i;
-	int sock_alive = 0;
 	int result = -EIO;
 	dnbd3_reply_t reply;
-	struct dnbd3_sock *sock = NULL;
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (!dnbd3_is_sock_alive(dev->socks[i])) {
-			sock = &dev->socks[i];
-			break;
-		}
-	}
-	if (sock == NULL) {
-		warn_server(dev, server, "could not connect to socket, to many connections");
-		return -EIO;
-	}
+	struct dnbd3_device *dev = sock->device;
 	sock->server = server;
 
-
 	debug_sock(sock, "socket connect");
 
-
 	mutex_init(&sock->tx_lock);
 	mutex_lock(&sock->tx_lock);
-	result = dnbd3_socket_connect(sock, server);
+	result = __dnbd3_socket_connect(sock, server);
 	if (result) {
 		error_sock(sock, "connection to socket failed");
 		result = -EIO;
@@ -1294,6 +1157,7 @@ static int dnbd3_server_connect(struct dnbd3_device *dev,
 
 	result = dnbd3_send_request_cmd(sock, CMD_SELECT_IMAGE);
 	if (result <= 0) {
+
 		error_sock(sock, "connection to image %s failed", dev->imgname);
 		result = -EIO;
 		goto error;
@@ -1328,23 +1192,7 @@ static int dnbd3_server_connect(struct dnbd3_device *dev,
 
 	INIT_WORK(&sock->keepalive_worker, dnbd3_keepalive_worker);
 
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (dev->socks[i].sock  && dev->socks[i].server) {
-			sock_alive++;
-		}
-	}
-	/* if first socket to connect, start timer and workers */
-	if (sock_alive == 1) {
-		debug_sock(sock, "first connection to server, starting workers");
-		INIT_WORK(&dev->discovery_worker, dnbd3_discovery_worker);
-		INIT_WORK(&dev->panic_worker, dnbd3_panic_worker);
-		timer_setup(&dev->timer, dnbd3_timer, 0);
-		dev->timer.expires = jiffies + HZ;
-		add_timer(&dev->timer);
-	}
-	blk_mq_update_nr_hw_queues(&dev->tag_set, sock_alive);
-
-	/* request alternative servers receiver will handle this */
+	/* TODO not on every connect? request alternative servers receiver will handle this */
 	if (dnbd3_send_request_cmd(sock, CMD_GET_SERVERS) <= 0) {
 		error_sock(sock, "failed to get servers in discovery");
 	}
@@ -1352,6 +1200,7 @@ static int dnbd3_server_connect(struct dnbd3_device *dev,
 	return 0;
 error:
 	server->failures++;
+	sock->panic = true;
 	if (sock->sock) {
 		kernel_sock_shutdown(sock->sock, SHUT_RDWR);
 		cancel_work_sync(&sock->receive_worker);
@@ -1362,41 +1211,13 @@ error:
 	return result;
 }
 
-static void dnbd3_clear_req(struct request *req, void *data, bool reserved)
-{
-	struct dnbd3_cmd *cmd = blk_mq_rq_to_pdu(req);
-
-	cmd->status = BLK_STS_IOERR;
-	blk_mq_complete_request(req);
-}
 
 /**
- * dnbd3_socket_disconnect - disconnect a socket or server
+ * dnbd3_socket_disconnect - disconnect a socket
  * @sock: the socket to disconnect
- *
- * 1. update nr of mq queues
- * 2. if last socket remove timer
- * 3. disconnect socket
  */
 static int dnbd3_socket_disconnect(struct dnbd3_sock *sock)
 {
-	int i;
-	struct dnbd3_device *dev = sock->device;
-	int sock_alive = 0;
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (dnbd3_is_sock_alive(dev->socks[i])) {
-			sock_alive++;
-		}
-	}
-	if (sock_alive <= 1) {
-		info_sock(sock, "shutting down last socket and stopping timer");
-		del_timer_sync(&dev->timer);
-		/*
-		 * do not wait for discovery and panic worker as they may have
-		 * called this method
-		 */
-
-	}
 	cancel_work_sync(&sock->keepalive_worker);
 
 	debug_sock(sock, "socket disconnect");
@@ -1425,50 +1246,18 @@ static int dnbd3_socket_disconnect(struct dnbd3_sock *sock)
 		sock_release(sock->sock);
 		sock->sock = NULL;
 	}
-	debug_sock(sock, "update nr hw queues to %d", sock_alive -1);
-//	blk_mq_quiesce_queue(dev->disk->queue);
-//	debug_sock(sock, "busy iter");
-//	blk_mq_tagset_busy_iter(&dev->tag_set, dnbd3_clear_req, NULL);
-//	debug_sock(sock, "update");
-	//TODO can we just stop the failed queue?
-	blk_mq_update_nr_hw_queues(&dev->tag_set, sock_alive - 1);
-//	debug_sock(sock, "unquisco");
-//	blk_mq_unquiesce_queue(dev->disk->queue);
-//	debug_sock(sock, "donee");
 
 	sock->server = NULL;
 	sock->panic = false;
 	return 0;
 }
 
-/**
- * dnbd3_server_disconnect - disconnect a server from a socket
- * @dev: the device
- * @server: the server to disconnect
- */
-static int dnbd3_server_disconnect(struct dnbd3_device *dev,
-		struct dnbd3_server *server)
-{
-	int i;
-	struct dnbd3_sock *sock = NULL;
-	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
-		if (dev->socks[i].server == server) {
-			sock = &dev->socks[i];
-		}
-	}
-	if (!sock || !sock->sock) {
-		warn_dev(dev, "could not find socket to disconnect");
-		return -EIO;
-	}
-	return dnbd3_socket_disconnect(sock);
-}
-
 
 /**
  * dnbd3_net_connect - connect device
  * @dev: the device to connect
  *
- * dnbd3_device.alt_servers[0] must set
+ * dnbd3_device.alt_servers[0] must be set
  */
 int dnbd3_net_connect(struct dnbd3_device *dev)
 {
@@ -1479,14 +1268,27 @@ int dnbd3_net_connect(struct dnbd3_device *dev)
 		return -ENONET;
 	}
 
-	// alt_server[0] is the initial server
-	result = dnbd3_server_connect(dev, &dev->alt_servers[0]);
+	result = dnbd3_adjust_connections(dev);
 	if (result) {
 		error_dev(dev, "failed to connect to initial server");
-		result = -ENOENT;
-		dev->imgname = NULL;
-		dev->socks[0].server = NULL;
+		return -ENOENT;
 	}
+
+	debug_dev(dev, "connected, starting workers");
+	INIT_WORK(&dev->discovery_worker, dnbd3_discovery_worker);
+	INIT_WORK(&dev->panic_worker, dnbd3_panic_worker);
+	timer_setup(&dev->timer, dnbd3_timer, 0);
+	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;
 }
 
diff --git a/src/kernel/sysfs.c b/src/kernel/sysfs.c
index 1588dfb..490ffbd 100644
--- a/src/kernel/sysfs.c
+++ b/src/kernel/sysfs.c
@@ -46,18 +46,18 @@ ssize_t show_connected_servers(char *buf, struct dnbd3_device *dev)
 		if (dev->socks[i].sock) {
 			if (dev->socks[i].server->host.type == HOST_IP4) {
 				ret = MIN(snprintf(buf, size, "%pI4,%d,%llu,%d\n",
-									dev->socks[i].server->host.addr,
-				                   (int)ntohs(dev->socks[i].server->host.port),
-				                   (unsigned long long)((dev->socks[i].server->rtts[0] + dev->socks[i].server->rtts[1] + dev->socks[i].server->rtts[2] + dev->socks[i].server->rtts[3]) / 4),
-				                   (int)dev->socks[i].server->failures)
+						dev->socks[i].server->host.addr,
+				                (int)ntohs(dev->socks[i].server->host.port),
+						dev->socks[i].server->avg_rtt,
+						(int)dev->socks[i].server->failures)
 				          , size);
 
 			} else {
 				ret = MIN(snprintf(buf, size, "%pI6,%d,%llu,%d\n",
-									dev->socks[i].server->host.addr,
-				                   (int)ntohs(dev->socks[i].server->host.port),
-				                   (unsigned long long)((dev->socks[i].server->rtts[0] + dev->socks[i].server->rtts[1] + dev->socks[i].server->rtts[2] + dev->socks[i].server->rtts[3]) / 4),
-				                   (int)dev->socks[i].server->failures)
+						dev->socks[i].server->host.addr,
+						(int)ntohs(dev->socks[i].server->host.port),
+						dev->socks[i].server->avg_rtt,
+						(int)dev->socks[i].server->failures)
 				          , size);
 
 			}
@@ -81,9 +81,10 @@ ssize_t show_connected_servers(char *buf, struct dnbd3_device *dev)
 ssize_t show_alt_server_num(char *buf, struct dnbd3_device *dev)
 {
 	int i, num = 0;
-	for (i = 0; i < NUMBER_SERVERS; ++i)
-	{
-		if (dev->alt_servers[i].host.type) ++num;
+	for (i = 0; i < NUMBER_SERVERS; i++) {
+		if (dev->alt_servers[i].host.type) {
+			num++;
+		}
 	}
 	return MIN(snprintf(buf, PAGE_SIZE, "%d\n", num), PAGE_SIZE);
 }
@@ -91,28 +92,27 @@ ssize_t show_alt_server_num(char *buf, struct dnbd3_device *dev)
 ssize_t show_alt_servers(char *buf, struct dnbd3_device *dev)
 {
 	int i, size = PAGE_SIZE, ret;
-	for (i = 0; i < NUMBER_SERVERS; ++i)
-	{
-		if (dev->alt_servers[i].host.type == HOST_IP4)
+	for (i = 0; i < NUMBER_SERVERS; ++i) {
+		if (dev->alt_servers[i].host.type == HOST_IP4) {
 			ret = MIN(snprintf(buf, size, "%pI4,%d,%llu,%d\n",
 			                   dev->alt_servers[i].host.addr,
 			                   (int)ntohs(dev->alt_servers[i].host.port),
-			                   (unsigned long long)((dev->alt_servers[i].rtts[0] + dev->alt_servers[i].rtts[1] + dev->alt_servers[i].rtts[2] + dev->alt_servers[i].rtts[3]) / 4),
+					   dev->alt_servers[i].avg_rtt,
 			                   (int)dev->alt_servers[i].failures)
 			          , size);
-		else if (dev->alt_servers[i].host.type == HOST_IP6)
+		} else if (dev->alt_servers[i].host.type == HOST_IP6) {
 			ret = MIN(snprintf(buf, size, "%pI6,%d,%llu,%d\n",
 			                   dev->alt_servers[i].host.addr,
 			                   (int)ntohs(dev->alt_servers[i].host.port),
-			                   (unsigned long long)((dev->alt_servers[i].rtts[0] + dev->alt_servers[i].rtts[1] + dev->alt_servers[i].rtts[2] + dev->alt_servers[i].rtts[3]) / 4),
+					   dev->alt_servers[i].avg_rtt,
 			                   (int)dev->alt_servers[i].failures)
 			          , size);
-		else
+		} else {
 			continue;
+		}
 		size -= ret;
 		buf += ret;
-		if (size <= 0)
-		{
+		if (size <= 0) {
 			size = 0;
 			break;
 		}