[sheepdog] [PATCH v2 2/4] sheepdev: linux kernel module of block device driver for sheepdog

[levin li]

From: levin li <xingke.lwp at taobao.com>

This module provides function for users to take sheepdog VDIs as block
devices in linux, users can register a VDI to its kernel space, it just
like that a new hard disk is added to the computer, users can create
partitions for the disk, format the disk or mount the disk, it provides
users a efficient way to use sheepdog as distributed storage system.

The usage is easy, after install the module sheepdev.ko, it creates a proc
entry '/proc/entry', you can write into the proc entry file to control
the driver.

Add a new block device from an existing sheepdog VDI:

# echo "add 127.0.0.1:7070 linux" > /proc/sheep

It would create a block device /dev/sheepa, you can format/mount this device:

# mkfs.ext4 /dev/sheepa
# mount -t ext4 /sheep/sheepa test

Remove a block device from the kernel:

# echo "del sheepa" > /proc/sheep

Signed-off-by: levin li <xingke.lwp at taobao.com>
---
 sheepdev/connect.c |  178 ++++++++++++
 sheepdev/device.c  |  774 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 sheepdev/proc.c    |  158 +++++++++++
 sheepdev/sheep.c   |  176 ++++++++++++
 sheepdev/sheep.h   |  119 ++++++++
 5 files changed, 1405 insertions(+), 0 deletions(-)
 create mode 100644 sheepdev/connect.c
 create mode 100644 sheepdev/device.c
 create mode 100644 sheepdev/proc.c
 create mode 100644 sheepdev/sheep.c
 create mode 100644 sheepdev/sheep.h

diff --git a/sheepdev/connect.c b/sheepdev/connect.c
new file mode 100644
index 0000000..009a7b9
--- /dev/null
+++ b/sheepdev/connect.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright (C) 2012 Taobao Inc.
+ *
+ * Levin Li <xingke.lwp at taobao.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sheep.h"
+#include "sheepdog_proto.h"
+
+int connect_to(struct socket **sock, const char *ip_addr, int port)
+{
+	int ret;
+	struct sockaddr_in addr;
+
+	ret = sock_create(AF_INET, SOCK_STREAM, IPPROTO_TCP, sock);
+	if (ret) {
+		DBPRT("fail to create socket\n");
+		return ret;
+	}
+
+	memset(&addr, 0, sizeof(addr));
+	addr.sin_family = AF_INET;
+	addr.sin_port = htons(port);
+	addr.sin_addr.s_addr = in_aton(ip_addr);
+
+	ret = (*sock)->ops->connect(*sock, (struct sockaddr *)&addr,
+				 sizeof(addr), 0);
+
+	if (!ret)
+		DBPRT("connected to %s:%d\n", ip_addr, port);
+
+	return ret;
+}
+
+int do_read(struct socket *sock, char *buf, const size_t length)
+{
+	struct msghdr msg;
+	struct iovec iov;
+	int ret = 0, received = 0, left = length;
+	mm_segment_t oldmm;
+
+	memset(&msg, 0, sizeof(msg));
+	msg.msg_iov = &iov;
+	msg.msg_iovlen = 1;
+
+	while (left > 0) {
+		oldmm = get_fs();
+		set_fs(KERNEL_DS);
+		msg.msg_iov->iov_base = buf + received;
+		msg.msg_iov->iov_len = left;
+		ret = sock_recvmsg(sock, &msg, left, MSG_WAITALL);
+		set_fs(oldmm);
+		if (ret <= 0)
+			break;
+		left -= ret;
+		received += ret;
+	}
+
+	return ret;
+}
+
+static void forward_iov(struct msghdr *msg, int len)
+{
+	while (msg->msg_iov->iov_len <= len) {
+		len -= msg->msg_iov->iov_len;
+		msg->msg_iov++;
+		msg->msg_iovlen--;
+	}
+
+	msg->msg_iov->iov_base = (char *) msg->msg_iov->iov_base + len;
+	msg->msg_iov->iov_len -= len;
+}
+
+
+static int do_write(struct socket *sock, struct msghdr *msg, int len)
+{
+	int ret;
+	mm_segment_t oldmm;
+
+rewrite:
+	oldmm = get_fs();
+	set_fs(KERNEL_DS);
+	ret = sock_sendmsg(sock, msg, len);
+	set_fs(oldmm);
+
+	if (ret < 0) {
+		if (ret == -EINTR)
+			goto rewrite;
+		if (ret == -EBUSY) {
+			DBPRT("busy\n");
+			goto rewrite;
+		}
+		DBPRT("failed to write to socket: %d\n", ret);
+		return -EFAULT;
+	}
+
+	len -= ret;
+	if (len) {
+		forward_iov(msg, ret);
+		goto rewrite;
+	}
+
+	return 0;
+}
+
+int send_req(struct socket *sock, struct sd_req *hdr, void *data,
+	     unsigned int wlen)
+{
+	int ret;
+	struct msghdr msg;
+	struct iovec iov[2];
+
+	memset(&msg, 0, sizeof(msg));
+
+	msg.msg_iov = iov;
+
+	msg.msg_iovlen = 1;
+	iov[0].iov_base = hdr;
+	iov[0].iov_len = sizeof(*hdr);
+
+	if (wlen) {
+		msg.msg_iovlen++;
+		iov[1].iov_base = data;
+		iov[1].iov_len = wlen;
+	}
+
+	ret = do_write(sock, &msg, sizeof(*hdr) + wlen);
+	if (ret) {
+		DBPRT("failed to send request %x, %d\n", hdr->opcode, wlen);
+		ret = -EFAULT;
+	}
+
+	return ret;
+}
+
+int exec_req(struct socket *sock, struct sd_req *hdr, void *data)
+{
+	int ret;
+	struct sd_rsp *rsp = (struct sd_rsp *)hdr;
+	unsigned int wlen, rlen;
+
+	if (hdr->flags & SD_FLAG_CMD_WRITE) {
+		wlen = hdr->data_length;
+		rlen = 0;
+	} else {
+		wlen = 0;
+		rlen = hdr->data_length;
+	}
+
+	if (send_req(sock, hdr, data, wlen))
+		return -EFAULT;
+
+	ret = do_read(sock, (char *)rsp, sizeof(*rsp));
+	if (ret < 0) {
+		DBPRT("failed to read a response\n");
+		return -EFAULT;
+	}
+
+	if (rlen > rsp->data_length)
+		rlen = rsp->data_length;
+
+	if (rlen) {
+		ret = do_read(sock, data, rlen);
+		if (ret < 0) {
+			DBPRT("failed to read the response data\n");
+			return -EFAULT;
+		}
+	}
+
+	return 0;
+}
diff --git a/sheepdev/device.c b/sheepdev/device.c
new file mode 100644
index 0000000..8d2dd4d
--- /dev/null
+++ b/sheepdev/device.c
@@ -0,0 +1,774 @@
+/*
+ * Copyright (C) 2012 Taobao Inc.
+ *
+ * Levin Li <xingke.lwp at taobao.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/wait.h>
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/fs.h>
+#include <linux/genhd.h>
+#include <linux/blkdev.h>
+#include <linux/hdreg.h>
+#include <linux/proc_fs.h>
+#include <linux/kthread.h>
+#include "sheep.h"
+
+static int sheepdev_major;
+spinlock_t devices_lock;
+struct list_head dev_list;
+static unsigned long *device_bitmap;
+
+static void sheepdev_get(struct sheepdev *dev)
+{
+	atomic_inc(&dev->struct_refcnt);
+}
+
+static void sheepdev_put(struct sheepdev *dev)
+{
+	if (atomic_dec_and_test(&dev->struct_refcnt))
+		kfree(dev);
+}
+
+static int add_request(struct sheepdev *dev, struct request *req, uint64_t oid,
+		       int idx)
+{
+	struct sheep_request *s_req = kmalloc(sizeof(*s_req), GFP_KERNEL);
+	if (!s_req)
+		return -EIO;
+
+	s_req->req_id = dev->req_id;
+	s_req->req = req;
+	s_req->oid = oid;
+	s_req->idx = idx;
+	INIT_LIST_HEAD(&s_req->list);
+
+	spin_lock_irq(&dev->fin_lock);
+	list_add_tail(&s_req->list, &dev->finish_list);
+	spin_unlock_irq(&dev->fin_lock);
+
+	if (dev->req_id > UINT_MAX)
+		dev->req_id = 1;
+	else
+		dev->req_id++;
+
+	return 0;
+}
+
+static void sheep_end_request(struct request *req, int ret)
+{
+	struct request_queue *q = req->q;
+	unsigned long flags;
+
+	spin_lock_irqsave(q->queue_lock, flags);
+	__blk_end_request_all(req, ret);
+	spin_unlock_irqrestore(q->queue_lock, flags);
+}
+
+static struct obj_state_entry *obj_state_rb_insert(struct rb_root *root,
+						   struct obj_state_entry *new)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct obj_state_entry *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct obj_state_entry, node);
+
+		if (new->idx < entry->idx)
+			p = &(*p)->rb_left;
+		else if (new->idx > entry->idx)
+			p = &(*p)->rb_right;
+		else
+			return entry;
+	}
+	rb_link_node(&new->node, parent, p);
+	rb_insert_color(&new->node, root);
+
+	return NULL;
+}
+
+static struct obj_state_entry *obj_state_rb_search(struct rb_root *root,
+						   uint32_t idx)
+{
+	struct rb_node *n = root->rb_node;
+	struct obj_state_entry *t;
+
+	while (n) {
+		t = rb_entry(n, struct obj_state_entry, node);
+
+		if (idx < t->idx)
+			n = n->rb_left;
+		else if (idx > t->idx)
+			n = n->rb_right;
+		else
+			return t;
+	}
+
+	return NULL;
+}
+
+static void set_obj_state(struct sheepdev *dev, int idx, int state)
+{
+	struct obj_state_entry *old, *new;
+
+	new = kmalloc(sizeof(*new), GFP_KERNEL);
+	if (!new)
+		DBPRT("[%s] No-Mem\n", __func__);
+
+	new->idx = idx;
+	new->state = state;
+
+	old = obj_state_rb_insert(&dev->obj_state_tree, new);
+	if (old) {
+		kfree(new);
+		old->state = state;
+	}
+}
+
+static int get_obj_state(struct sheepdev *dev, int idx)
+{
+	struct obj_state_entry *entry;
+
+	entry = obj_state_rb_search(&dev->obj_state_tree, idx);
+	if (entry)
+		return entry->state;
+
+	DBPRT("No obj-state entry found for [%x:%d]\n",
+	      dev->vid, idx);
+
+	return -ENOENT;
+}
+
+static int remove_obj_state(struct sheepdev *dev, int idx)
+{
+	struct rb_root *root = &dev->obj_state_tree;
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	struct obj_state_entry *entry;
+
+	while (*p) {
+		parent = *p;
+		entry = rb_entry(parent, struct obj_state_entry, node);
+
+		if (idx < entry->idx)
+			p = &(*p)->rb_left;
+		else if (idx > entry->idx)
+			p = &(*p)->rb_right;
+		else {
+			rb_erase(parent, root);
+			kfree(entry);
+			return 0;
+		}
+	}
+
+	return -ENOENT;
+}
+
+static int sheep_handle_request(struct request *req)
+{
+	struct req_iterator iter;
+	struct bio_vec *bvec;
+	struct gendisk *disk = req->rq_disk;
+	struct sheepdev *dev = disk->private_data;
+	unsigned long sector = blk_rq_pos(req);
+	unsigned long offset = sector * KERNEL_SECTOR_SIZE;
+	unsigned long nbytes = blk_rq_bytes(req);
+	int idx = offset / SHEEP_OBJECT_SIZE + 1;
+	uint64_t oid = vid_to_data_oid(dev->vid, idx);
+	uint64_t off = offset % SHEEP_OBJECT_SIZE;
+	uint64_t cow_oid = 0;
+	int ret = 0, len = 0, create = 0;
+	int write = rq_data_dir(req);
+	void *sheep_buf = NULL;
+
+	/* For read requests */
+	if (!write && !dev->inode->data_vdi_id[idx]) {
+		rq_for_each_segment(bvec, req, iter) {
+			void *addr = kmap(bvec->bv_page);
+			memset(addr + bvec->bv_offset, 0, bvec->bv_len);
+			kunmap(bvec->bv_page);
+		}
+		sheep_end_request(req, 0);
+		return 0;
+	} else if (!write) {
+		oid = vid_to_data_oid(dev->inode->data_vdi_id[idx], idx);
+		ret = send_read_req(dev, oid, nbytes, off);
+		if (ret)
+			return -EIO;
+
+		ret = add_request(dev, req, oid, idx);
+		if (ret)
+			return -EIO;
+
+		return 0;
+	}
+
+	/* For write requests */
+	sheep_buf = kmalloc(nbytes, GFP_KERNEL);
+	if (!sheep_buf)
+		return -EIO;
+
+	spin_lock(&dev->creating_lock);
+	if (!dev->inode->data_vdi_id[idx]) {
+		if (get_obj_state(dev, idx) == OBJ_STATE_CREATING) {
+			/* Wait for pending inode-update to complete */
+			spin_unlock(&dev->creating_lock);
+			wait_event_interruptible(dev->creating_wait,
+						 object_ready(dev, idx));
+		} else {
+			set_obj_state(dev, idx, OBJ_STATE_CREATING);
+			create = 1;
+			spin_unlock(&dev->creating_lock);
+		}
+	} else if (!object_ready(dev, idx)) {
+		/*
+		 * Now we check the rbtree to determine whether to wait for
+		 * copy-on-write done or to invoke copy-on-write for this object
+		 */
+		if (get_obj_state(dev, idx) == OBJ_STATE_COWING) {
+			/* Wait inode to be updated */
+			spin_unlock(&dev->creating_lock);
+			wait_event_interruptible(dev->creating_wait,
+						 object_ready(dev, idx));
+		} else {
+			set_obj_state(dev, idx, OBJ_STATE_COWING);
+			cow_oid = vid_to_data_oid(dev->inode->data_vdi_id[idx], idx);
+			create = 1;
+			spin_unlock(&dev->creating_lock);
+		}
+	} else
+		spin_unlock(&dev->creating_lock);
+
+
+	rq_for_each_segment(bvec, req, iter) {
+		void *addr = kmap(bvec->bv_page);
+
+		memcpy(sheep_buf + len, addr + bvec->bv_offset, bvec->bv_len);
+		len += bvec->bv_len;
+
+		if (rq_iter_last(req, iter)) {
+			ret = send_write_req(dev, oid, cow_oid, sheep_buf, len, off,
+					     create);
+			if (ret != SD_RES_SUCCESS) {
+				kunmap(bvec->bv_page);
+				ret = -EIO;
+				goto out;
+			}
+
+			ret = add_request(dev, req, oid, idx);
+			if (ret) {
+				kunmap(bvec->bv_page);
+				ret = -EIO;
+				goto out;
+			}
+
+			if (!create)
+				goto done;
+
+			/* For create/cow operations we need to update inode */
+			oid = vid_to_vdi_oid(dev->vid);
+			off = offsetof(struct sheepdog_inode, data_vdi_id);
+			off += sizeof(uint32_t) * idx;
+			ret = send_write_req(dev, oid, 0, (char *)&dev->vid,
+					     sizeof(dev->vid), off, 0);
+			if (ret != SD_RES_SUCCESS) {
+				kunmap(bvec->bv_page);
+				ret = -EIO;
+				goto out;
+			}
+
+			ret = add_request(dev, req, oid, idx);
+			if (ret) {
+				kunmap(bvec->bv_page);
+				ret = -EIO;
+				goto out;
+			}
+done:;
+		}
+
+		kunmap(bvec->bv_page);
+	}
+
+out:
+	kfree(sheep_buf);
+	return ret;
+}
+
+static void sheep_io_request(struct request_queue *rq)
+{
+	struct request *req;
+	struct gendisk *disk;
+	struct sheepdev *dev;
+
+	while ((req = blk_fetch_request(rq)) != NULL) {
+
+		disk = req->rq_disk;
+		dev = disk->private_data;
+
+		if (req->cmd_type != REQ_TYPE_FS) {
+			DBPRT("Skip non-fs request\n");
+			__blk_end_request_all(req, -EIO);
+		}
+
+		spin_lock(&dev->req_lock);
+		list_add_tail(&req->queuelist, &dev->pending_list);
+		spin_unlock(&dev->req_lock);
+
+		wake_up_interruptible(&dev->req_wait);
+	}
+}
+
+static int process_request(void *data)
+{
+	struct sheepdev *dev = (struct sheepdev *)data;
+	struct request *req;
+	int ret;
+
+	sheepdev_get(dev);
+
+	while (!kthread_should_stop() || !list_empty(&dev->pending_list)) {
+		wait_event_interruptible(dev->req_wait,
+					 !list_empty(&dev->pending_list) ||
+					 kthread_should_stop());
+
+		spin_lock(&dev->req_lock);
+		if (list_empty(&dev->pending_list)) {
+			spin_unlock(&dev->req_lock);
+			continue;
+		}
+
+		req = list_entry(dev->pending_list.next, struct request,
+				 queuelist);
+		list_del_init(&req->queuelist);
+		spin_unlock(&dev->req_lock);
+
+		ret = sheep_handle_request(req);
+		if (ret)
+			sheep_end_request(req, ret);
+		else
+			wake_up_interruptible(&dev->fin_wait);
+	}
+
+	sheepdev_put(dev);
+
+	return 0;
+}
+
+static int sheepdev_open(struct block_device *blkdev, fmode_t mode)
+{
+	struct gendisk *disk = blkdev->bd_disk;
+	struct sheepdev *dev = disk->private_data;
+
+	spin_lock(&dev->dev_lock);
+	dev->device_refcnt++;
+	spin_unlock(&dev->dev_lock);
+
+	return 0;
+}
+
+static int sheepdev_release(struct gendisk *disk, fmode_t mode)
+{
+	struct sheepdev *dev = disk->private_data;
+
+	spin_lock(&dev->dev_lock);
+	dev->device_refcnt--;
+	spin_unlock(&dev->dev_lock);
+
+	return 0;
+}
+
+static struct block_device_operations sheepdev_ops = {
+	.owner = THIS_MODULE,
+	.open = sheepdev_open,
+	.release = sheepdev_release,
+};
+
+static int sheep_add_disk(struct sheepdev *dev)
+{
+	int ret;
+	struct request_queue *queue;
+
+	dev->disk = alloc_disk(SHEEP_BLKDEV_MINORS);
+	if (!dev->disk) {
+		DBPRT("allocate gendisk failure\n");
+		ret = -EBUSY;
+		return ret;
+	}
+	queue = blk_init_queue(sheep_io_request, &dev->que_lock);
+	/* 4M boundary */
+	blk_queue_segment_boundary(queue, 0x3fffff);
+	dev->disk->major = sheepdev_major;
+	dev->disk->first_minor = dev->minor * SHEEP_BLKDEV_MINORS;
+	dev->disk->queue = queue;
+	dev->disk->fops = &sheepdev_ops;
+	dev->disk->private_data = dev;
+	snprintf(dev->disk->disk_name, sizeof(dev->disk->disk_name),
+		 SHEEP_BLKDEV_NAME"%c", dev->minor + 'a');
+
+	set_capacity(dev->disk, dev->sectors);
+	add_disk(dev->disk);
+
+	return 0;
+}
+
+static struct sheep_request *find_request(struct sheepdev *dev, int id)
+{
+	struct sheep_request *req, *t;
+
+	spin_lock_irq(&dev->fin_lock);
+	list_for_each_entry_safe(req, t, &dev->finish_list, list) {
+		if (req->req_id != id)
+			continue;
+		list_del_init(&req->list);
+		spin_unlock_irq(&dev->fin_lock);
+		return req;
+	}
+	spin_unlock_irq(&dev->fin_lock);
+
+	return NULL;
+}
+
+static int read_reply(struct sheepdev *dev, int *req_id, int *result,
+		      void **data)
+{
+	int ret;
+	struct sd_rsp rsp;
+	void *buf = NULL;
+
+	*result = 0;
+	*req_id = 0;
+	*data = NULL;
+
+	ret = do_read(dev->sock, (char *)&rsp, sizeof(rsp));
+	if (ret < 0) {
+		DBPRT("failed to read response\n");
+		return -EIO;
+	}
+
+	if (rsp.data_length > 0) {
+		buf = kmalloc(rsp.data_length, GFP_KERNEL);
+		if (!buf) {
+			DBPRT("No-mem\n");
+			return -ENOMEM;
+		}
+
+		ret = do_read(dev->sock, buf, rsp.data_length);
+		if (ret != rsp.data_length) {
+			kfree(buf);
+			return -EIO;
+		}
+	}
+
+	*req_id = rsp.id;
+	*result = rsp.result;
+	*data = buf;
+
+	return 0;
+}
+
+static void cleanup_finish_list(struct sheepdev *dev)
+{
+	struct sheep_request *req, *t;
+
+	spin_lock(&dev->fin_lock);
+	list_for_each_entry_safe(req, t, &dev->finish_list, list) {
+		list_del_init(&req->list);
+		sheep_end_request(req->req, -EIO);
+		kfree(req);
+	}
+
+	spin_unlock(&dev->fin_lock);
+}
+
+static int process_response(void *data)
+{
+	struct sheepdev *dev = data;
+	struct sheep_request *sheep_req;
+	struct request *req;
+	int ret, req_id, res;
+	int obj_state;
+
+	sheepdev_get(dev);
+
+	while (!kthread_should_stop() || !list_empty(&dev->finish_list)) {
+		void *buf = NULL;
+
+		wait_event_interruptible(dev->fin_wait,
+					 !list_empty(&dev->finish_list) ||
+					 kthread_should_stop());
+
+		spin_lock_irq(&dev->fin_lock);
+		if (list_empty(&dev->finish_list)) {
+			spin_unlock_irq(&dev->fin_lock);
+			continue;
+		}
+		spin_unlock_irq(&dev->fin_lock);
+
+		ret = read_reply(dev, &req_id, &res, &buf);
+		if (ret) {
+			cleanup_finish_list(dev);
+			continue;
+		}
+
+		sheep_req = find_request(dev, req_id);
+		if (!sheep_req)
+			goto next;
+		req = sheep_req->req;
+
+		if (rq_data_dir(req)) {
+			int idx;
+
+			res = (res != SD_RES_SUCCESS) ? -EIO : 0;
+			if (sheep_req->oid == vid_to_vdi_oid(dev->vid)) {
+				/* inode-update response */
+				idx = sheep_req->idx;
+			} else {
+				/* oridinary write response */
+				idx = data_oid_to_idx(sheep_req->oid);
+
+				/* obj already exist */
+				if (dev->inode->data_vdi_id[idx] == dev->vid) {
+					sheep_end_request(req, res);
+					goto next;
+				}
+			}
+
+			/* inode-update response */
+			spin_lock(&dev->creating_lock);
+			obj_state = get_obj_state(dev, idx);
+			if (obj_state == OBJ_STATE_OK) {
+				/*
+				 * Both obj-write and inode-update are complete
+				 * we can end the write request and wake other
+				 * requests waiting for this object.
+				 */
+				remove_obj_state(dev, idx);
+				dev->inode->data_vdi_id[idx] = dev->vid;
+				spin_unlock(&dev->creating_lock);
+
+				sheep_end_request(req, res);
+				wake_up_interruptible(&dev->creating_wait);
+
+				goto next;
+			} else {
+				/*
+				 * wait for obj-write or inode-update to complete
+				 */
+				set_obj_state(dev, idx, OBJ_STATE_OK);
+			}
+			spin_unlock(&dev->creating_lock);
+
+		} else {
+			int len = 0;
+			struct req_iterator iter;
+			struct bio_vec *bvec;
+
+			if (res != SD_RES_SUCCESS) {
+				sheep_end_request(req, -EIO);
+				goto next;
+			}
+
+			rq_for_each_segment(bvec, req, iter) {
+				void *addr = kmap(bvec->bv_page);
+				memcpy(addr + bvec->bv_offset, buf + len,
+				       bvec->bv_len);
+				len += bvec->bv_len;
+				kunmap(bvec->bv_page);
+			}
+			sheep_end_request(req, 0);
+		}
+next:
+		kfree(buf);
+		kfree(sheep_req);
+	}
+
+	sheepdev_put(dev);
+	return 0;
+}
+
+static int sheep_dev_setup(struct sheepdev *dev)
+{
+	int ret;
+
+	ret = sheep_vdi_setup(dev);
+	if (ret)
+		return ret;
+
+	spin_lock_init(&dev->que_lock);
+	spin_lock_init(&dev->req_lock);
+	spin_lock_init(&dev->fin_lock);
+	spin_lock_init(&dev->dev_lock);
+	spin_lock_init(&dev->creating_lock);
+	init_waitqueue_head(&dev->req_wait);
+	init_waitqueue_head(&dev->fin_wait);
+	init_waitqueue_head(&dev->creating_wait);
+	INIT_LIST_HEAD(&dev->pending_list);
+	INIT_LIST_HEAD(&dev->finish_list);
+	INIT_LIST_HEAD(&dev->dev_list);
+
+	dev->req_id = 1;
+	dev->req_thread = kthread_run(process_request, dev,
+				      "sheep_req");
+	dev->fin_thread = kthread_run(process_response, dev,
+				      "sheep_fin");
+
+	ret = sheep_add_disk(dev);
+	if (ret) {
+		return ret;
+	}
+
+	return 0;
+}
+
+int sheep_add_device(const char *addr, int port, const char *vdiname,
+		     int snapshot_id, const char *snapshot_tag)
+{
+	struct sheepdev *dev;
+	int ret = 0;
+
+	DBPRT("[%s:%d] vdiname: %s, snapshot id: %d, snapshot tag: %s\n",
+	      addr, port, vdiname, snapshot_id, snapshot_tag);
+
+	dev = kmalloc(sizeof(*dev), GFP_KERNEL);
+	if (!dev)
+		return -ENOMEM;
+
+	memset(dev, 0, sizeof(*dev));
+	dev->port = port;
+	dev->snapshot_id = snapshot_id;
+	memcpy(dev->ip_addr, addr, sizeof(dev->ip_addr));
+	strcpy(dev->vdiname, vdiname);
+	strcpy(dev->snapshot_tag, snapshot_tag);
+
+	spin_lock(&devices_lock);
+	dev->minor = find_next_zero_bit(device_bitmap, SHEEP_BLKDEV_MINORS, 0);
+	set_bit(dev->minor, device_bitmap);
+	spin_unlock(&devices_lock);
+
+	ret = sheep_dev_setup(dev);
+	if (ret) {
+		clear_bit(dev->minor, device_bitmap);
+		goto out;
+	} else {
+		sheepdev_get(dev);
+		spin_lock(&devices_lock);
+		list_add_tail(&dev->dev_list, &dev_list);
+		spin_unlock(&devices_lock);
+	}
+
+	return ret;
+out:
+	kfree(dev);
+	return ret;
+}
+
+static void remove_device(struct sheepdev *dev)
+{
+	DBPRT("remove device /dev/%s\n", dev->disk->disk_name);
+
+	kthread_stop(dev->req_thread);
+	kthread_stop(dev->fin_thread);
+	wake_up_interruptible(&dev->req_wait);
+	wake_up_interruptible(&dev->fin_wait);
+
+	blk_cleanup_queue(dev->disk->queue);
+	del_gendisk(dev->disk);
+	put_disk(dev->disk);
+
+	clear_bit(dev->minor, device_bitmap);
+	inet_release(dev->sock);
+
+	sheepdev_put(dev);
+}
+
+int sheep_remove_device(const char *dev_name)
+{
+	struct sheepdev *dev, *t;
+	int ret = 0;
+
+	spin_lock(&devices_lock);
+	list_for_each_entry_safe(dev, t, &dev_list, dev_list) {
+		if (strcmp(dev_name, dev->disk->disk_name) != 0)
+			continue;
+
+		spin_lock(&dev->dev_lock);
+		if (dev->device_refcnt) {
+			spin_unlock(&dev->dev_lock);
+			ret = -EBUSY;
+		} else {
+			spin_unlock(&dev->dev_lock);
+			list_del_init(&dev->dev_list);
+			remove_device(dev);
+		}
+
+		break;
+	}
+	spin_unlock(&devices_lock);
+
+	return ret;
+}
+
+static int __init sheep_module_init(void)
+{
+	int ret;
+
+	DBPRT("Block device driver for Sheepdog\n");
+
+	spin_lock_init(&devices_lock);
+	INIT_LIST_HEAD(&dev_list);
+	device_bitmap = kmalloc(SHEEP_BLKDEV_MINORS / 8, GFP_KERNEL);
+	if (!device_bitmap)
+		return -ENOMEM;
+	memset(device_bitmap, 0, SHEEP_BLKDEV_MINORS / 8);
+
+	ret = sheep_proc_init();
+	if (ret)
+		return ret;
+
+	sheepdev_major = register_blkdev(0, SHEEP_BLKDEV_NAME);
+	if (sheepdev_major < 0) {
+		ret = sheepdev_major;
+		goto error;
+	}
+
+	return 0;
+
+error:
+	sheep_proc_destroy();
+	return ret;
+}
+
+static void __exit sheep_module_exit(void)
+{
+	struct sheepdev *dev, *t;
+
+	list_for_each_entry_safe(dev, t, &dev_list, dev_list) {
+		list_del_init(&dev->dev_list);
+		remove_device(dev);
+	}
+
+	sheep_proc_destroy();
+	unregister_blkdev(sheepdev_major, SHEEP_BLKDEV_NAME);
+	kfree(device_bitmap);
+
+	DBPRT("Sheepdog Block Device Removed.\n");
+}
+
+module_init(sheep_module_init);
+module_exit(sheep_module_exit);
+
+MODULE_LICENSE("GPL");
diff --git a/sheepdev/proc.c b/sheepdev/proc.c
new file mode 100644
index 0000000..a2a46de
--- /dev/null
+++ b/sheepdev/proc.c
@@ -0,0 +1,158 @@
+/*
+ * Copyright (C) 2012 Taobao Inc.
+ *
+ * Levin Li <xingke.lwp at taobao.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+#include "sheep.h"
+
+static struct proc_dir_entry *sheep_proc_entry;
+
+#define MAX_CMD_LEN 64
+
+static int process_add_command(char *buf, int len)
+{
+	int i, ret;
+	char addr[16];
+	char snapshot_tag[SD_MAX_VDI_TAG_LEN];
+	char vdiname[SD_MAX_VDI_LEN];
+	int snapshot_id = 0, port;
+
+	/* parse ip address */
+	for (i = 0; buf[i] != '\0' && buf[i] != '\n' &&
+	     buf[i] != ' ' && buf[i] != ':' && i < len; i++);
+
+	if (buf[i] != ' ' && buf[i] != ':')
+		return -EINVAL;
+
+	if (i + 1 > sizeof(addr))
+		return -EINVAL;
+
+	memcpy(addr, buf, i);
+	addr[i] = '\0';
+
+	/* parse port */
+	if (buf[i] == ' ') {
+		port = SD_LISTEN_PORT;
+		buf = &buf[i + 1];
+	} else {
+		/* start from ':' to ' ' */
+		char *tmp = &buf[i + 1];
+		len -= (i + 1);
+		for (i = 0; tmp[i] != ' ' && tmp[i] != '\0' &&
+		     tmp[i] != '\n' && i < len; i++);
+		if (tmp[i] != ' ') {
+			return -EINVAL;
+		}
+		tmp[i] = '\0';
+		buf = &tmp[i + 1];
+		port = simple_strtol(tmp, NULL, 10);
+	}
+
+	/* parse vdiname and snapshot id */
+	for (i = 0; buf[i] != '\0' && buf[i] != ':' && buf[i] != '\n'; i++);
+	memcpy(vdiname, buf, i);
+	vdiname[i] = '\0';
+
+	*snapshot_tag = '\0';
+	if (buf[i] == ':') {
+		char *p;
+		buf = &buf[i + 1];
+		snapshot_id = simple_strtol(buf, &p, 10);
+		if (buf == p) {
+			snapshot_id = 0;
+			if (buf[strlen(buf) - 1] == '\n')
+				buf[strlen(buf) - 1] = '\0';
+			strncpy(snapshot_tag, buf, sizeof(snapshot_tag));
+		}
+	}
+
+	ret = sheep_add_device(addr, port, vdiname, snapshot_id, snapshot_tag);
+
+	return ret;
+}
+
+static int process_del_command(char *buf, int len)
+{
+	int ret;
+
+	if (buf[len - 1] != '\n')
+		return -EINVAL;
+	buf[len - 1] = '\0';
+
+	ret = sheep_remove_device(buf);
+
+	return ret;
+}
+
+static ssize_t sheep_proc_write(struct file *filp, const char __user *buf,
+				size_t len, loff_t *offset)
+{
+	char *kern_buf, cmd_buf[MAX_CMD_LEN];
+	int i, ret;
+
+	kern_buf = kmalloc(len, GFP_KERNEL);
+	if (!kern_buf)
+		return -ENOMEM;
+
+	if (copy_from_user(kern_buf, buf, len)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	for (i = 0; kern_buf[i] != '\0' && kern_buf[i] != '\n' &&
+	     kern_buf[i] != ' ' && i < len; i++);
+
+	if (i > MAX_CMD_LEN || kern_buf[i] != ' ') {
+		ret = -EINVAL;
+		goto out;
+	}
+	memcpy(cmd_buf, kern_buf, i);
+	cmd_buf[i] = '\0';
+	if (strcmp(cmd_buf, "add") == 0) {
+		ret = process_add_command(&kern_buf[i + 1], len - i - 1);
+		if (ret)
+			goto out;
+	} else if (strcmp(cmd_buf, "del") == 0) {
+		ret = process_del_command(&kern_buf[i + 1], len - i - 1);
+		if (ret)
+			goto out;
+
+	} else {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = len;
+out:
+	kfree(kern_buf);
+	return ret;
+}
+
+static struct file_operations sheep_proc_fops = {
+	.write = sheep_proc_write,
+};
+
+int sheep_proc_init(void)
+{
+	/* create proc entry for sheep control */
+	sheep_proc_entry = create_proc_entry(PROC_ENTRY_NAME,
+					     S_IFREG | S_IRUGO | S_IWUGO, NULL);
+	if (!sheep_proc_entry)
+		return -ENOMEM;
+
+	sheep_proc_entry->proc_fops = &sheep_proc_fops;
+
+	return 0;
+}
+
+void sheep_proc_destroy(void)
+{
+	remove_proc_entry(PROC_ENTRY_NAME, NULL);
+}
diff --git a/sheepdev/sheep.c b/sheepdev/sheep.c
new file mode 100644
index 0000000..61e761e
--- /dev/null
+++ b/sheepdev/sheep.c
@@ -0,0 +1,176 @@
+/*
+ * Copyright (C) 2012 Taobao Inc.
+ *
+ * Levin Li <xingke.lwp at taobao.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "sheep.h"
+
+static void sd_init_req(struct sd_req *req, uint8_t opcode)
+{
+	memset(req, 0, sizeof(*req));
+	req->opcode = opcode;
+}
+
+static int read_object(struct sheepdev *dev, uint64_t oid, void *data,
+		       unsigned int datalen, uint64_t offset)
+{
+	struct sd_req hdr;
+	struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
+	int ret;
+
+	sd_init_req(&hdr, SD_OP_READ_OBJ);
+	hdr.id = 0;
+	hdr.data_length = datalen;
+
+	hdr.obj.oid = oid;
+	hdr.obj.offset = offset;
+
+	ret = exec_req(dev->sock, &hdr, data);
+
+	if (ret < 0) {
+		DBPRT("Failed to read object %llx\n", oid);
+		return SD_RES_EIO;
+	}
+
+	if (rsp->result != SD_RES_SUCCESS) {
+		DBPRT("Failed to read object %llx,%d\n", oid,
+		      rsp->result);
+		return SD_RES_EIO;
+	}
+
+	return SD_RES_SUCCESS;
+}
+
+int send_read_req(struct sheepdev *dev, uint64_t oid,
+		  unsigned int datalen, uint64_t offset)
+{
+	struct sd_req hdr;
+	int ret;
+
+	sd_init_req(&hdr, SD_OP_READ_OBJ);
+	hdr.id = dev->req_id;
+	hdr.data_length = datalen;
+
+	hdr.obj.oid = oid;
+	hdr.obj.offset = offset;
+
+	ret = send_req(dev->sock, &hdr, NULL, 0);
+
+	if (ret < 0) {
+		DBPRT("Failed to read object %llx\n", oid);
+		return SD_RES_EIO;
+	}
+
+	return SD_RES_SUCCESS;
+}
+
+int send_write_req(struct sheepdev *dev, uint64_t oid, uint64_t cow_oid,
+		   void *data, unsigned int datalen, uint64_t offset,
+		   int create)
+{
+	struct sd_req hdr;
+	int ret;
+
+	if (create)
+		sd_init_req(&hdr, SD_OP_CREATE_AND_WRITE_OBJ);
+	else
+		sd_init_req(&hdr, SD_OP_WRITE_OBJ);
+
+	hdr.id = dev->req_id;
+	hdr.data_length = datalen;
+	hdr.flags = SD_FLAG_CMD_WRITE | SD_FLAG_CMD_DIRECT;
+	if (cow_oid)
+		hdr.flags |= SD_FLAG_CMD_COW;
+
+	hdr.obj.oid = oid;
+	hdr.obj.cow_oid = cow_oid;
+	hdr.obj.offset = offset;
+	hdr.obj.copies = dev->inode->nr_copies;
+
+	ret = send_req(dev->sock, &hdr, data, datalen);
+
+	if (ret < 0) {
+		DBPRT("Failed to write object %llx\n", oid);
+		return SD_RES_EIO;
+	}
+
+	return SD_RES_SUCCESS;
+}
+
+static int find_vdi_name(struct sheepdev *dev, const char *vdiname,
+			 uint32_t snapid, const char *tag)
+{
+	int ret;
+	struct sd_req hdr;
+	struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
+	char buf[SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN];
+
+	memset(buf, 0, sizeof(buf));
+	strncpy(buf, vdiname, SD_MAX_VDI_LEN);
+	strncpy(buf + SD_MAX_VDI_LEN, tag, SD_MAX_VDI_TAG_LEN);
+
+	sd_init_req(&hdr, SD_OP_LOCK_VDI);
+	hdr.data_length = SD_MAX_VDI_LEN + SD_MAX_VDI_TAG_LEN;
+	hdr.flags = SD_FLAG_CMD_WRITE;
+	hdr.vdi.snapid = snapid;
+
+	ret = exec_req(dev->sock, &hdr, buf);
+	if (ret)
+		return -EIO;
+
+	if (rsp->result != SD_RES_SUCCESS) {
+		DBPRT("Cannot get VDI info for %s %d %s\n",
+		      vdiname, snapid, tag);
+		return -EIO;
+	}
+
+	dev->vid = rsp->vdi.vdi_id;
+
+	return 0;
+}
+
+int sheep_vdi_setup(struct sheepdev *dev)
+{
+	int ret;
+	struct sheepdog_inode *inode;
+
+	inode = vmalloc(sizeof(*inode));
+	if (!inode)
+		return -ENOMEM;
+	memset(inode, 0 , sizeof(*inode));
+
+	ret = connect_to(&dev->sock, dev->ip_addr, dev->port);
+	if (ret) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	ret = find_vdi_name(dev, dev->vdiname, dev->snapshot_id,
+			    dev->snapshot_tag);
+	if (ret)
+		goto out;
+
+	ret = read_object(dev, vid_to_vdi_oid(dev->vid), inode,
+			  SD_INODE_SIZE, 0);
+	if (ret != SD_RES_SUCCESS) {
+		ret = -EIO;
+		goto out;
+	}
+
+	dev->size = inode->vdi_size - SHEEP_OBJECT_SIZE;
+	dev->sectors = dev->size / KERNEL_SECTOR_SIZE;
+	dev->inode = inode;
+
+	return 0;
+out:
+	vfree(inode);
+	return ret;
+}
diff --git a/sheepdev/sheep.h b/sheepdev/sheep.h
new file mode 100644
index 0000000..edbb724
--- /dev/null
+++ b/sheepdev/sheep.h
@@ -0,0 +1,119 @@
+/*
+ * Copyright (C) 2012 Taobao Inc.
+ *
+ * Levin Li <xingke.lwp at taobao.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version
+ * 2 as published by the Free Software Foundation.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef __SHEEP_H_
+#define __SHEEP_H_
+
+#include <linux/socket.h>
+#include <linux/net.h>
+#include <net/sock.h>
+#include <linux/tcp.h>
+#include <linux/socket.h>
+#include <linux/slab.h>
+#include <linux/in.h>
+#include <linux/inet.h>
+#include <linux/list.h>
+#include <linux/rbtree.h>
+#include <asm/atomic.h>
+#include <net/inet_common.h>
+#include "sheepdog_proto.h"
+
+#define SHEEP_OBJECT_SIZE (4 * 1024 * 1024)
+
+#define SHEEP_BLKDEV_NAME "sheep"
+#define PROC_ENTRY_NAME "sheep"
+#define KERNEL_SECTOR_SIZE 512
+#define SHEEP_BLKDEV_MINORS 1024
+
+#define DBPRT(fmt, args...) printk(KERN_DEBUG "sheep: " fmt, ##args)
+
+struct sheepdev {
+	struct gendisk *disk;
+	struct socket *sock;
+	char ip_addr[16];
+	char vdiname[SD_MAX_VDI_LEN];
+	char snapshot_tag[SD_MAX_VDI_TAG_LEN];
+	unsigned int snapshot_id;
+	unsigned int port;
+	unsigned int minor;
+	unsigned int req_id;
+	unsigned int vid;
+	unsigned long size;
+	unsigned long sectors;
+	atomic_t struct_refcnt;
+	unsigned int device_refcnt;
+	spinlock_t dev_lock;
+	spinlock_t req_lock;
+	spinlock_t fin_lock;
+	spinlock_t que_lock;
+	spinlock_t creating_lock;
+	struct rb_root obj_state_tree;
+	struct task_struct *req_thread;
+	struct task_struct *fin_thread;
+	wait_queue_head_t req_wait;
+	wait_queue_head_t fin_wait;
+	wait_queue_head_t creating_wait;
+	struct list_head pending_list;
+	struct list_head finish_list;
+	struct list_head dev_list;
+	struct sheepdog_inode *inode;
+};
+
+struct sheep_request {
+	int req_id;
+	int idx; /* idx is only used when update inode */
+	uint64_t oid;
+	struct request *req;
+	struct list_head list;
+};
+
+#define OBJ_STATE_CREATING 1
+#define OBJ_STATE_COWING   2
+#define OBJ_STATE_OK       3
+
+struct obj_state_entry {
+	int idx;
+	int state;
+	struct rb_node node;
+};
+
+/* connect.c */
+int connect_to(struct socket **sock, const char *addr, int port);
+int send_req(struct socket *sock, struct sd_req *hdr, void *data,
+	     unsigned int wlen);
+int do_read(struct socket *sock, char *buf, const size_t length);
+int exec_req(struct socket *sock, struct sd_req *hdr, void *data);
+
+/* proc.c */
+int sheep_proc_init(void);
+void sheep_proc_destroy(void);
+
+/* sheep.c */
+int send_read_req(struct sheepdev *sheepdev, uint64_t oid,
+		  unsigned int datalen, uint64_t offset);
+int send_write_req(struct sheepdev *sheepdev, uint64_t oid, uint64_t cow_oid,
+		   void *data, unsigned int datalen, uint64_t offset,
+		   int create);
+int sheep_vdi_setup(struct sheepdev *sheep_dev);
+
+/* device.c */
+int sheep_add_device(const char *addr, int port, const char *vdiname,
+		     int snapshot_id, const char *snapshot_tag);
+int sheep_remove_device(const char *dev_name);
+
+static inline int object_ready(struct sheepdev *dev, int idx)
+{
+	return dev->inode->data_vdi_id[idx] == dev->vid;
+}
+
+#endif
-- 
1.7.1