path: root/src/kernel/block.c

/*
 * block.c
 *
 *  Created on: Jun 26, 2019
 *      Author: fred
 */

#include <net/sock.h>

#include "dnbd3.h"
#include "block.h"
#include "clientconfig.h"

#define dnbd3_sock_create(af,type,proto,sock) sock_create_kern(&init_net, (af) == HOST_IP4 ? AF_INET : AF_INET6, type, proto, sock)




void dnbd3_blk_fail_all_requests(dnbd3_device_t *dev)
{
	printk(KERN_DEBUG "dnbd3: fail all requests device %i\n", dev->minor);
}


int dnbd3_socket_connect(dnbd3_device_t *dev, dnbd3_sock *sock)
{
	int result = -EIO;
	struct request *req1 = NULL;
	struct timeval timeout;
	struct dnbd3_server_t *server = sock->server;
	dnbd3_request_t dnbd3_request;
	dnbd3_reply_t dnbd3_reply;
	struct msghdr msg;
	struct kvec iov[2];
	uint16_t rid;
	uint64_t reported_size;
	char *name;
	int mlen;
	serialized_buffer_t payload_buffer;

	printk(KERN_DEBUG "dnbd3: socket connect device %i\n", dev->minor);

	mutex_init(&sock->lock);
	mutex_lock(&sock->lock);
	if (sock->pending) {
		printk(KERN_DEBUG "dnbd3: socket still in request\n");
		while (sock->pending)
			schedule();
	}
	if (server->host.port == 0 || server->host.type == 0) {
		printk(KERN_ERR "dnbd3: host or port not set\n");
		goto error;
	}
	if (sock->sock) {
		printk(KERN_WARNING "dnbd3: socket already connected\n");
		goto error;
	}

	timeout.tv_sec = SOCKET_TIMEOUT_CLIENT_DATA;
	timeout.tv_usec = 0;

	req1 = kmalloc(sizeof(*req1), GFP_ATOMIC );
	if (!req1) {
		printk(KERN_ERR "dnbd3: kmalloc failed\n");
		goto error;
	}

	init_msghdr(msg);

	if (dnbd3_sock_create(server->host.type, SOCK_STREAM, IPPROTO_TCP, &sock->sock) < 0) {
		printk(KERN_ERR "dnbd3: could not create socket\n");
		goto error;
	}

	kernel_setsockopt(sock->sock, SOL_SOCKET, SO_SNDTIMEO, (char *)&timeout, sizeof(timeout));
	kernel_setsockopt(sock->sock, SOL_SOCKET, SO_RCVTIMEO, (char *)&timeout, sizeof(timeout));
	sock->sock->sk->sk_allocation = GFP_NOIO;
	if (server->host.type == HOST_IP4) {
		struct sockaddr_in sin;
		memset(&sin, 0, sizeof(sin));
		sin.sin_family = AF_INET;
		memcpy(&(sin.sin_addr), server->host.addr, 4);
		sin.sin_port = server->host.port;
		if (kernel_connect(sock->sock, (struct sockaddr *)&sin, sizeof(sin), 0) != 0) {
			printk(KERN_ERR "dnbd3: connection to host failed (ipv4)\n");
			goto error;
		}
	} else {
		struct sockaddr_in6 sin;
		memset(&sin, 0, sizeof(sin));
		sin.sin6_family = AF_INET6;
		memcpy(&(sin.sin6_addr), server->host.addr, 16);
		sin.sin6_port = server->host.port;
		if (kernel_connect(sock->sock, (struct sockaddr *)&sin, sizeof(sin), 0) != 0){
			printk(KERN_ERR "dnbd3: connection to host failed (ipv6)\n");
			goto error;
		}
	}
	// Request filesize
	dnbd3_request.magic = dnbd3_packet_magic;
	dnbd3_request.cmd = CMD_SELECT_IMAGE;
	iov[0].iov_base = &dnbd3_request;
	iov[0].iov_len = sizeof(dnbd3_request);
	serializer_reset_write(&payload_buffer);
	serializer_put_uint16(&payload_buffer, PROTOCOL_VERSION);
	serializer_put_string(&payload_buffer, dev->imgname);
	serializer_put_uint16(&payload_buffer, dev->rid);
	serializer_put_uint8(&payload_buffer, 0); // is_server = false
	iov[1].iov_base = &payload_buffer;
	dnbd3_request.size = iov[1].iov_len = serializer_get_written_length(&payload_buffer);
	fixup_request(dnbd3_request);
	mlen = sizeof(dnbd3_request) + iov[1].iov_len;
	if (kernel_sendmsg(sock->sock, &msg, iov, 2, mlen) != mlen) {
		printk(KERN_ERR "dnbd3: could not send CMD_SIZE_REQUEST\n");
		goto error;
	}
	// receive reply header
	iov[0].iov_base = &dnbd3_reply;
	iov[0].iov_len = sizeof(dnbd3_reply);
	if (kernel_recvmsg(sock->sock, &msg, iov, 1, sizeof(dnbd3_reply), msg.msg_flags) != sizeof(dnbd3_reply)) {
		printk(KERN_ERR "dnbd3: received corrupted reply header after CMD_SIZE_REQUEST\n");
		goto error;
	}

	// check reply header
	fixup_reply(dnbd3_reply);
	if (dnbd3_reply.cmd != CMD_SELECT_IMAGE ||
			dnbd3_reply.size < 3 ||
			dnbd3_reply.size > MAX_PAYLOAD ||
			dnbd3_reply.magic != dnbd3_packet_magic) {
		printk(KERN_ERR "dnbd3: received invalid reply to CMD_SIZE_REQUEST image does not exist on server\n");
		goto error;
	}

	// receive reply payload
	iov[0].iov_base = &payload_buffer;
	iov[0].iov_len = dnbd3_reply.size;
	if (kernel_recvmsg(sock->sock, &msg, iov, 1, dnbd3_reply.size, msg.msg_flags) != dnbd3_reply.size) {
		printk(KERN_ERR "dnbd3: could not read CMD_SELECT_IMAGE payload on handshake\n");
		goto error;
	}

	// handle/check reply payload
	serializer_reset_read(&payload_buffer, dnbd3_reply.size);
	server->protocol_version = serializer_get_uint16(&payload_buffer);
	if (server->protocol_version < MIN_SUPPORTED_SERVER) {
		printk(KERN_ERR "dnbd3: server version is lower than min supported version\n");
		goto error;
	}

	name = serializer_get_string(&payload_buffer);
	rid = serializer_get_uint16(&payload_buffer);
	if (dev->rid != rid && strcmp(name, dev->imgname) != 0) {
		printk(KERN_ERR "dnbd3: server offers image '%s', requested '%s'\n", name, dev->imgname);
		goto error;
	}

	reported_size = serializer_get_uint64(&payload_buffer);
	if (dev->reported_size == NULL) {
		if (reported_size < 4096) {
			printk(KERN_ERR "dnbd3: reported size by server is < 4096\n");
			goto error;
		}
		dev->reported_size = reported_size;
		set_capacity(dev->disk, dev->reported_size >> 9); /* 512 Byte blocks */
	} else if (dev->reported_size != reported_size) {
		printk(KERN_ERR "dnbd3: reported size by server is %llu but should be %llu\n", reported_size, dev->reported_size);
	}

	printk(KERN_DEBUG "dnbd3: connected to image %s, filesize %llu\n", dev->imgname, dev->reported_size);

	mutex_unlock(&sock->lock);

//	TODO add heartbeat
	//	timer_setup(&dev->hb_timer, dnbd3_net_heartbeat, 0);
	//	dev->hb_timer.expires = jiffies + HZ;
	//	add_timer(&dev->hb_timer);

	return 0;
error:
	if (sock->sock) {
		sock_release(sock->sock);
		sock->sock = NULL;
	}
	if (req1) {
		kfree(req1);
	}
	mutex_unlock(&sock->lock);
	return result;
}

int dnbd3_socket_disconnect(dnbd3_device_t *dev, dnbd3_sock *sock)
{
	printk(KERN_DEBUG "dnbd3: socket disconnect device %i\n", dev->minor);
	mutex_lock(&sock->lock);

	// clear heartbeat timer
//	del_timer(&dev->hb_timer);


	if (sock->sock) {
		kernel_sock_shutdown(sock->sock, SHUT_RDWR);
	}

	// clear socket
	if (sock->sock) {
		sock_release(sock->sock);
		sock->sock = NULL;
	}

	mutex_unlock(&sock->lock);
	mutex_destroy(&sock->lock);
	return 0;
}

int dnbd3_net_disconnect(struct dnbd3_device_t *dev) {
	int i;
	int result;
	for (i = 0; i < NUMBER_CONNECTIONS; i++) {
		if (dev->socks[i].sock) {
			if (dnbd3_socket_disconnect(dev, &dev->socks[i])) {
				result = -EIO;
			}
		}
	}
	return result;
}


static void printHost(struct dnbd3_host_t *host, char *msg)
{
	if (host->type == HOST_IP4) {
		printk(KERN_INFO "dnbd3: %s %pI4:%d", msg, host->addr, host->port);
	} else {
		printk(KERN_INFO "dnbd3: %s [%pI6]:%d", msg, host->addr, host->port);
	}
}

static void printServerList(struct dnbd3_device_t *dev)
{
	int i;
	printHost(&dev->initial_server.host, "initial server is");
	for (i = 0; i < NUMBER_SERVERS; i++) {
		if (dev->alt_servers[i].host.addr[0] != 0) {
			printHost(&dev->alt_servers[i].host, "alternative server is");
		}
	}
}

static int dnbd3_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
{
	int result = -100;
	dnbd3_device_t *dev = bdev->bd_disk->private_data;
	char *imgname = NULL;
	dnbd3_ioctl_t *msg = NULL;
	int i;

	printk(KERN_DEBUG "dnbd3: ioctl device %i, cmd %i, arg %lu\n", dev->minor, cmd, arg);
	//unsigned long irqflags;

//	while (dev->disconnecting) {
//		// do nothing
//	}

	if (arg != 0) {
		msg = kmalloc(sizeof(*msg), GFP_KERNEL);
		if (msg == NULL) return -ENOMEM;
		if (copy_from_user((char *)msg, (char *)arg, 2) != 0 || msg->len != sizeof(*msg)) {
			result = -ENOEXEC;
			goto cleanup_return;
		}
		if (copy_from_user((char *)msg, (char *)arg, sizeof(*msg)) != 0) {
			result = -ENOENT;
			goto cleanup_return;
		}
		if (msg->imgname != NULL && msg->imgnamelen > 0) {
			imgname = kmalloc(msg->imgnamelen + 1, GFP_KERNEL);
			if (imgname == NULL) {
				result = -ENOMEM;
				goto cleanup_return;
			}
			if (copy_from_user(imgname, msg->imgname, msg->imgnamelen) != 0) {
				result = -ENOENT;
				goto cleanup_return;
			}
			imgname[msg->imgnamelen] = '\0';

			printk(KERN_DEBUG "dnbd3: ioctl image name of len %i is %s\n", (int)msg->imgnamelen, imgname);
		}
	}

	switch (cmd) {
	case IOCTL_OPEN:
		printk(KERN_DEBUG "dnbd3: ioctl open\n");
		if (dev->imgname != NULL) {
			result = -EBUSY;
		} else if (imgname == NULL) {
			result = -EINVAL;
		} else if (msg == NULL) {
			result = -EINVAL;
		} else {
			if (sizeof(msg->host) != sizeof(dev->initial_server.host)) {
				printk(KERN_INFO "dnbd3: odd size bug#1 triggered in ioctl\n");
			}
			memcpy(&dev->initial_server.host, &msg->host, sizeof(msg->host));
			dev->initial_server.failures = 0;
//			memcpy(&dev->initial_server, &dev->cur_server, sizeof(dev->initial_server));
			dev->imgname = imgname;
			dev->rid = msg->rid;
			dev->use_server_provided_alts = msg->use_server_provided_alts;
			// Forget all alt servers on explicit connect, set first alt server to initial server
			memset(dev->alt_servers, 0, sizeof(dev->alt_servers[0])*NUMBER_SERVERS);
			memcpy(dev->alt_servers, &dev->initial_server, sizeof(dev->alt_servers[0]));
//#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
//			if (blk_queue->backing_dev_info != NULL) {
//				blk_queue->backing_dev_info->ra_pages = (msg->read_ahead_kb * 1024) / PAGE_SIZE;
//			}
//#else
//			blk_queue->backing_dev_info.ra_pages = (msg->read_ahead_kb * 1024) / PAGE_SIZE;
//#endif


			dev->socks[0].server = &dev->initial_server;
			if (dnbd3_socket_connect(dev, &dev->socks[0]) == 0) {
				printServerList(dev);
				result = 0;
				imgname = NULL; // Prevent kfree at the end
			} else {
				printk(KERN_ERR "dnbd3: failed to connect to initial server\n");
				result = -ENOENT;
				dev->imgname = NULL;
				dev->socks[0].server = NULL;
			}

		}
		break;

	case IOCTL_CLOSE:
		printk(KERN_DEBUG "dnbd3: ioctl close\n");
		dnbd3_blk_fail_all_requests(dev);
		dnbd3_net_disconnect(dev);
		dnbd3_blk_fail_all_requests(dev);
		set_capacity(dev->disk, 0);
		if (dev->imgname) {
			kfree(dev->imgname);
			dev->imgname = NULL;
		}
		break;

	case IOCTL_SWITCH:
		printk(KERN_DEBUG "dnbd3: ioctl switch\n");
		result = -EINVAL;
		break;

	case IOCTL_ADD_SRV:
	case IOCTL_REM_SRV:
		printk(KERN_DEBUG "dnbd3: ioctl add/rem srv\n");
		if (dev->imgname == NULL) {
			result = -ENOENT;
		} else if (dev->new_servers_num >= NUMBER_SERVERS) {
			result = -EAGAIN;
		} else if (msg == NULL) {
			result = -EINVAL;
		} else {
			memcpy(&dev->new_servers[dev->new_servers_num].host, &msg->host, sizeof(msg->host));
			dev->new_servers[dev->new_servers_num].failures = (cmd == IOCTL_ADD_SRV ? 0 : 1); // 0 = ADD, 1 = REM
			++dev->new_servers_num;
			result = 0;
		}
		break;

	case BLKFLSBUF:
		printk(KERN_DEBUG "dnbd3: ioctl blkflsbuf\n");
		result = 0;
		break;

	default:
		printk(KERN_DEBUG "dnbd3: ioctl unhandled cmd %d\n", cmd);
		result = -EIO;
		break;
	}

cleanup_return:
	if (msg) kfree(msg);
	if (imgname) kfree(imgname);
	return result;

}



struct block_device_operations dnbd3_fops =
{
	.owner = THIS_MODULE,
	.ioctl = dnbd3_ioctl,
	.compat_ioctl =	dnbd3_ioctl,
};