diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
index 9dcd5ddce94fb66e426697fa77fe273a3be329ea..7044ef76a19d52a1ccab6c23255092144d4f63f9 100644
--- a/drivers/block/nbd.c
+++ b/drivers/block/nbd.c
@@ -49,6 +49,8 @@ struct nbd_sock {
 	struct mutex tx_lock;
 	struct request *pending;
 	int sent;
+	bool dead;
+	int fallback_index;
 };
 
 #define NBD_TIMEDOUT			0
@@ -82,6 +84,7 @@ struct nbd_device {
 
 struct nbd_cmd {
 	struct nbd_device *nbd;
+	int index;
 	struct completion send_complete;
 };
 
@@ -124,6 +127,15 @@ static const char *nbdcmd_to_ascii(int cmd)
 	return "invalid";
 }
 
+static void nbd_mark_nsock_dead(struct nbd_sock *nsock)
+{
+	if (!nsock->dead)
+		kernel_sock_shutdown(nsock->sock, SHUT_RDWR);
+	nsock->dead = true;
+	nsock->pending = NULL;
+	nsock->sent = 0;
+}
+
 static int nbd_size_clear(struct nbd_device *nbd, struct block_device *bdev)
 {
 	if (bdev->bd_openers <= 1)
@@ -191,7 +203,31 @@ static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
 	struct nbd_cmd *cmd = blk_mq_rq_to_pdu(req);
 	struct nbd_device *nbd = cmd->nbd;
 
-	dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
+	if (nbd->num_connections > 1) {
+		dev_err_ratelimited(nbd_to_dev(nbd),
+				    "Connection timed out, retrying\n");
+		mutex_lock(&nbd->config_lock);
+		/*
+		 * Hooray we have more connections, requeue this IO, the submit
+		 * path will put it on a real connection.
+		 */
+		if (nbd->socks && nbd->num_connections > 1) {
+			if (cmd->index < nbd->num_connections) {
+				struct nbd_sock *nsock =
+					nbd->socks[cmd->index];
+				mutex_lock(&nsock->tx_lock);
+				nbd_mark_nsock_dead(nsock);
+				mutex_unlock(&nsock->tx_lock);
+			}
+			mutex_unlock(&nbd->config_lock);
+			blk_mq_requeue_request(req, true);
+			return BLK_EH_NOT_HANDLED;
+		}
+		mutex_unlock(&nbd->config_lock);
+	} else {
+		dev_err_ratelimited(nbd_to_dev(nbd),
+				    "Connection timed out\n");
+	}
 	set_bit(NBD_TIMEDOUT, &nbd->runtime_flags);
 	req->errors = -EIO;
 
@@ -301,6 +337,7 @@ static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 		}
 		iov_iter_advance(&from, sent);
 	}
+	cmd->index = index;
 	request.type = htonl(type);
 	if (type != NBD_CMD_FLUSH) {
 		request.from = cpu_to_be64((u64)blk_rq_pos(req) << 9);
@@ -328,7 +365,7 @@ static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 		}
 		dev_err_ratelimited(disk_to_dev(nbd->disk),
 			"Send control failed (result %d)\n", result);
-		return -EIO;
+		return -EAGAIN;
 	}
 send_pages:
 	if (type != NBD_CMD_WRITE)
@@ -370,7 +407,7 @@ static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 				dev_err(disk_to_dev(nbd->disk),
 					"Send data failed (result %d)\n",
 					result);
-				return -EIO;
+				return -EAGAIN;
 			}
 			/*
 			 * The completion might already have come in,
@@ -389,6 +426,12 @@ static int nbd_send_cmd(struct nbd_device *nbd, struct nbd_cmd *cmd, int index)
 	return 0;
 }
 
+static int nbd_disconnected(struct nbd_device *nbd)
+{
+	return test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) ||
+		test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags);
+}
+
 /* NULL returned = something went wrong, inform userspace */
 static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 {
@@ -405,8 +448,7 @@ static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 	iov_iter_kvec(&to, READ | ITER_KVEC, &iov, 1, sizeof(reply));
 	result = sock_xmit(nbd, index, 0, &to, MSG_WAITALL, NULL);
 	if (result <= 0) {
-		if (!test_bit(NBD_DISCONNECTED, &nbd->runtime_flags) &&
-		    !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
+		if (!nbd_disconnected(nbd))
 			dev_err(disk_to_dev(nbd->disk),
 				"Receive control failed (result %d)\n", result);
 		return ERR_PTR(result);
@@ -449,8 +491,19 @@ static struct nbd_cmd *nbd_read_stat(struct nbd_device *nbd, int index)
 			if (result <= 0) {
 				dev_err(disk_to_dev(nbd->disk), "Receive data failed (result %d)\n",
 					result);
-				req->errors = -EIO;
-				return cmd;
+				/*
+				 * If we've disconnected or we only have 1
+				 * connection then we need to make sure we
+				 * complete this request, otherwise error out
+				 * and let the timeout stuff handle resubmitting
+				 * this request onto another connection.
+				 */
+				if (nbd_disconnected(nbd) ||
+				    nbd->num_connections <= 1) {
+					req->errors = -EIO;
+					return cmd;
+				}
+				return ERR_PTR(-EIO);
 			}
 			dev_dbg(nbd_to_dev(nbd), "request %p: got %d bytes data\n",
 				cmd, bvec.bv_len);
@@ -495,19 +548,17 @@ static void recv_work(struct work_struct *work)
 	while (1) {
 		cmd = nbd_read_stat(nbd, args->index);
 		if (IS_ERR(cmd)) {
+			struct nbd_sock *nsock = nbd->socks[args->index];
+
+			mutex_lock(&nsock->tx_lock);
+			nbd_mark_nsock_dead(nsock);
+			mutex_unlock(&nsock->tx_lock);
 			ret = PTR_ERR(cmd);
 			break;
 		}
 
 		nbd_end_request(cmd);
 	}
-
-	/*
-	 * We got an error, shut everybody down if this wasn't the result of a
-	 * disconnect request.
-	 */
-	if (ret && !test_bit(NBD_DISCONNECT_REQUESTED, &nbd->runtime_flags))
-		sock_shutdown(nbd);
 	atomic_dec(&nbd->recv_threads);
 	wake_up(&nbd->recv_wq);
 }
@@ -531,6 +582,47 @@ static void nbd_clear_que(struct nbd_device *nbd)
 	dev_dbg(disk_to_dev(nbd->disk), "queue cleared\n");
 }
 
+static int find_fallback(struct nbd_device *nbd, int index)
+{
+	int new_index = -1;
+	struct nbd_sock *nsock = nbd->socks[index];
+	int fallback = nsock->fallback_index;
+
+	if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags))
+		return new_index;
+
+	if (nbd->num_connections <= 1) {
+		dev_err_ratelimited(disk_to_dev(nbd->disk),
+				    "Attempted send on invalid socket\n");
+		return new_index;
+	}
+
+	if (fallback >= 0 && fallback < nbd->num_connections &&
+	    !nbd->socks[fallback]->dead)
+		return fallback;
+
+	if (nsock->fallback_index < 0 ||
+	    nsock->fallback_index >= nbd->num_connections ||
+	    nbd->socks[nsock->fallback_index]->dead) {
+		int i;
+		for (i = 0; i < nbd->num_connections; i++) {
+			if (i == index)
+				continue;
+			if (!nbd->socks[i]->dead) {
+				new_index = i;
+				break;
+			}
+		}
+		nsock->fallback_index = new_index;
+		if (new_index < 0) {
+			dev_err_ratelimited(disk_to_dev(nbd->disk),
+					    "Dead connection, failed to find a fallback\n");
+			return new_index;
+		}
+	}
+	new_index = nsock->fallback_index;
+	return new_index;
+}
 
 static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 {
@@ -544,22 +636,16 @@ static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 				    "Attempted send on invalid socket\n");
 		return -EINVAL;
 	}
-
-	if (test_bit(NBD_DISCONNECTED, &nbd->runtime_flags)) {
-		dev_err_ratelimited(disk_to_dev(nbd->disk),
-				    "Attempted send on closed socket\n");
-		return -EINVAL;
-	}
-
 	req->errors = 0;
-
+again:
 	nsock = nbd->socks[index];
 	mutex_lock(&nsock->tx_lock);
-	if (unlikely(!nsock->sock)) {
+	if (nsock->dead) {
+		index = find_fallback(nbd, index);
 		mutex_unlock(&nsock->tx_lock);
-		dev_err_ratelimited(disk_to_dev(nbd->disk),
-				    "Attempted send on closed socket\n");
-		return -EINVAL;
+		if (index < 0)
+			return -EIO;
+		goto again;
 	}
 
 	/* Handle the case that we have a pending request that was partially
@@ -572,7 +658,18 @@ static int nbd_handle_cmd(struct nbd_cmd *cmd, int index)
 		ret = 0;
 		goto out;
 	}
+	/*
+	 * Some failures are related to the link going down, so anything that
+	 * returns EAGAIN can be retried on a different socket.
+	 */
 	ret = nbd_send_cmd(nbd, cmd, index);
+	if (ret == -EAGAIN) {
+		dev_err_ratelimited(disk_to_dev(nbd->disk),
+				    "Request send failed trying another connection\n");
+		nbd_mark_nsock_dead(nsock);
+		mutex_unlock(&nsock->tx_lock);
+		goto again;
+	}
 out:
 	mutex_unlock(&nsock->tx_lock);
 	return ret;
@@ -646,6 +743,8 @@ static int nbd_add_socket(struct nbd_device *nbd, struct block_device *bdev,
 
 	nbd->socks = socks;
 
+	nsock->fallback_index = -1;
+	nsock->dead = false;
 	mutex_init(&nsock->tx_lock);
 	nsock->sock = sock;
 	nsock->pending = NULL;