Staging: hv: add the Hyper-V virtual block driver
authorHank Janssen <hjanssen@microsoft.com>
Mon, 13 Jul 2009 22:33:02 +0000 (15:33 -0700)
committerGreg Kroah-Hartman <gregkh@suse.de>
Tue, 15 Sep 2009 19:01:43 +0000 (12:01 -0700)
This is the virtual block driver when running Linux on top of Hyper-V.

Signed-off-by: Hank Janssen <hjanssen@microsoft.com>
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
drivers/staging/hv/BlkVsc.c [new file with mode: 0644]
drivers/staging/hv/blkvsc_drv.c [new file with mode: 0644]

diff --git a/drivers/staging/hv/BlkVsc.c b/drivers/staging/hv/BlkVsc.c
new file mode 100644 (file)
index 0000000..58b9698
--- /dev/null
@@ -0,0 +1,107 @@
+/*
+ *
+ * Copyright (c) 2009, Microsoft Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ *
+ * Authors:
+ *   Hank Janssen  <hjanssen@microsoft.com>
+ *
+ */
+
+
+#include "../storvsc/StorVsc.c"
+
+static const char* gBlkDriverName="blkvsc";
+
+//{32412632-86cb-44a2-9b5c-50d1417354f5}
+static const GUID gBlkVscDeviceType={
+       .Data = {0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44, 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5}
+};
+
+// Static routines
+static int
+BlkVscOnDeviceAdd(
+       DEVICE_OBJECT   *Device,
+       void                    *AdditionalInfo
+       );
+
+
+int
+BlkVscInitialize(
+       DRIVER_OBJECT *Driver
+       )
+{
+       STORVSC_DRIVER_OBJECT* storDriver = (STORVSC_DRIVER_OBJECT*)Driver;
+       int ret=0;
+
+       DPRINT_ENTER(BLKVSC);
+
+       // Make sure we are at least 2 pages since 1 page is used for control
+       ASSERT(storDriver->RingBufferSize >= (PAGE_SIZE << 1));
+
+       Driver->name = gBlkDriverName;
+       memcpy(&Driver->deviceType, &gBlkVscDeviceType, sizeof(GUID));
+
+       storDriver->RequestExtSize                      = sizeof(STORVSC_REQUEST_EXTENSION);
+       // Divide the ring buffer data size (which is 1 page less than the ring buffer size since that page is reserved for the ring buffer indices)
+       // by the max request size (which is VMBUS_CHANNEL_PACKET_MULITPAGE_BUFFER + VSTOR_PACKET + UINT64)
+       storDriver->MaxOutstandingRequestsPerChannel =
+               ((storDriver->RingBufferSize - PAGE_SIZE) / ALIGN_UP(MAX_MULTIPAGE_BUFFER_PACKET + sizeof(VSTOR_PACKET) + sizeof(UINT64),sizeof(UINT64)));
+
+       DPRINT_INFO(BLKVSC, "max io outstd %u", storDriver->MaxOutstandingRequestsPerChannel);
+
+       // Setup the dispatch table
+       storDriver->Base.OnDeviceAdd            = BlkVscOnDeviceAdd;
+       storDriver->Base.OnDeviceRemove         = StorVscOnDeviceRemove;
+       storDriver->Base.OnCleanup                      = StorVscOnCleanup;
+
+       storDriver->OnIORequest                         = StorVscOnIORequest;
+
+       DPRINT_EXIT(BLKVSC);
+
+       return ret;
+}
+
+int
+BlkVscOnDeviceAdd(
+       DEVICE_OBJECT   *Device,
+       void                    *AdditionalInfo
+       )
+{
+       int ret=0;
+       STORVSC_DEVICE_INFO *deviceInfo = (STORVSC_DEVICE_INFO*)AdditionalInfo;
+
+       DPRINT_ENTER(BLKVSC);
+
+       ret = StorVscOnDeviceAdd(Device, AdditionalInfo);
+
+       if (ret != 0)
+       {
+               DPRINT_EXIT(BLKVSC);
+
+               return ret;
+       }
+
+       // We need to use the device instance guid to set the path and target id. For IDE devices, the
+       // device instance id is formatted as <bus id> - <device id> - 8899 - 000000000000.
+       deviceInfo->PathId = Device->deviceInstance.Data[3] << 24 | Device->deviceInstance.Data[2] << 16 |
+               Device->deviceInstance.Data[1] << 8 |Device->deviceInstance.Data[0];
+
+       deviceInfo->TargetId = Device->deviceInstance.Data[5] << 8 | Device->deviceInstance.Data[4];
+
+       DPRINT_EXIT(BLKVSC);
+
+       return ret;
+}
diff --git a/drivers/staging/hv/blkvsc_drv.c b/drivers/staging/hv/blkvsc_drv.c
new file mode 100644 (file)
index 0000000..2f67981
--- /dev/null
@@ -0,0 +1,1547 @@
+/*
+ *
+ * Copyright (c) 2009, Microsoft Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
+ * Place - Suite 330, Boston, MA 02111-1307 USA.
+ *
+ * Authors:
+ *   Hank Janssen  <hjanssen@microsoft.com>
+ *
+ */
+
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/blkdev.h>
+#include <linux/major.h>
+#include <linux/delay.h>
+#include <linux/hdreg.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_eh.h>
+#include <scsi/scsi_dbg.h>
+
+#include "logging.h"
+#include "vmbus.h"
+
+#include "StorVscApi.h"
+
+//
+// #defines
+//
+#define BLKVSC_MINORS  64
+
+//
+// Data types
+//
+enum blkvsc_device_type {
+       UNKNOWN_DEV_TYPE,
+       HARDDISK_TYPE,
+       DVD_TYPE,
+};
+
+// This request ties the struct request and struct blkvsc_request/STORVSC_REQUEST together
+// A struct request may be represented by 1 or more struct blkvsc_request
+struct blkvsc_request_group {
+       int                                     outstanding;
+       int                                     status;
+
+       struct list_head        blkvsc_req_list;        // list of blkvsc_requests
+};
+
+
+struct blkvsc_request {
+       struct list_head        req_entry;                      // blkvsc_request_group.blkvsc_req_list
+
+       struct list_head        pend_entry;                     // block_device_context.pending_list
+
+       struct request          *req;                           // This may be null if we generate a request internally
+       struct block_device_context     *dev;
+       struct blkvsc_request_group     *group;         // The group this request is part of. Maybe null
+
+       wait_queue_head_t       wevent;
+       int cond;
+
+       int                                     write;
+       sector_t                        sector_start;
+       unsigned long           sector_count;
+
+       unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
+       unsigned char cmd_len;
+       unsigned char cmnd[MAX_COMMAND_SIZE];
+
+       STORVSC_REQUEST         request;
+       // !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap, because -
+       // The extension buffer falls right here and is pointed to by request.Extension;
+};
+
+// Per device structure
+struct block_device_context {
+       struct device_context   *device_ctx; // point back to our device context
+       struct kmem_cache       *request_pool;
+       spinlock_t                              lock;
+       struct gendisk                  *gd;
+       enum blkvsc_device_type device_type;
+       struct list_head                pending_list;
+
+       unsigned char                   device_id[64];
+       unsigned int                    device_id_len;
+       int                                             num_outstanding_reqs;
+       int                                             shutting_down;
+       int                                             media_not_present;
+       unsigned int                    sector_size;
+       sector_t                                capacity;
+       unsigned int                    port;
+       unsigned char                   path;
+       unsigned char                   target;
+       int                                             users;
+};
+
+// Per driver
+struct blkvsc_driver_context {
+       // !! These must be the first 2 fields !!
+       struct driver_context   drv_ctx;
+       STORVSC_DRIVER_OBJECT   drv_obj;
+};
+
+// Static decl
+static int blkvsc_probe(struct device *dev);
+static int blkvsc_remove(struct device *device);
+static void blkvsc_shutdown(struct device *device);
+
+static int blkvsc_open(struct inode *inode, struct file *filep);
+static int blkvsc_release(struct inode *inode, struct file *filep);
+static int blkvsc_media_changed(struct gendisk *gd);
+static int blkvsc_revalidate_disk(struct gendisk *gd);
+static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
+static int blkvsc_ioctl(struct inode *inode, struct file *filep, unsigned cmd, unsigned long arg);
+
+static void blkvsc_request(struct request_queue *queue);
+static void blkvsc_request_completion(STORVSC_REQUEST* request);
+static int blkvsc_do_request(struct block_device_context *blkdev, struct request *req);
+static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req, void (*request_completion)(STORVSC_REQUEST*) );
+static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
+static void blkvsc_cmd_completion(STORVSC_REQUEST* request);
+static int blkvsc_do_inquiry(struct block_device_context *blkdev);
+static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
+static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
+static int blkvsc_do_flush(struct block_device_context *blkdev);
+static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
+static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
+
+
+static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
+
+// The one and only one
+static struct blkvsc_driver_context g_blkvsc_drv;
+
+
+static struct block_device_operations block_ops =
+{
+       .owner = THIS_MODULE,
+       .open = blkvsc_open,
+       .release = blkvsc_release,
+       .media_changed = blkvsc_media_changed,
+       .revalidate_disk = blkvsc_revalidate_disk,
+       .getgeo = blkvsc_getgeo,
+       .ioctl  = blkvsc_ioctl,
+};
+
+/*++
+
+Name:  blkvsc_drv_init()
+
+Desc:  BlkVsc driver initialization.
+
+--*/
+int blkvsc_drv_init(PFN_DRIVERINITIALIZE pfn_drv_init)
+{
+       int ret=0;
+       STORVSC_DRIVER_OBJECT *storvsc_drv_obj=&g_blkvsc_drv.drv_obj;
+       struct driver_context *drv_ctx=&g_blkvsc_drv.drv_ctx;
+
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       vmbus_get_interface(&storvsc_drv_obj->Base.VmbusChannelInterface);
+
+       storvsc_drv_obj->RingBufferSize = blkvsc_ringbuffer_size;
+
+       // Callback to client driver to complete the initialization
+       pfn_drv_init(&storvsc_drv_obj->Base);
+
+       drv_ctx->driver.name = storvsc_drv_obj->Base.name;
+       memcpy(&drv_ctx->class_id, &storvsc_drv_obj->Base.deviceType, sizeof(GUID));
+
+#if defined(KERNEL_2_6_5) || defined(KERNEL_2_6_9)
+       drv_ctx->driver.probe = blkvsc_probe;
+       drv_ctx->driver.remove = blkvsc_remove;
+#else
+       drv_ctx->probe = blkvsc_probe;
+       drv_ctx->remove = blkvsc_remove;
+       drv_ctx->shutdown = blkvsc_shutdown;
+#endif
+
+       // The driver belongs to vmbus
+       vmbus_child_driver_register(drv_ctx);
+
+       DPRINT_EXIT(BLKVSC_DRV);
+
+       return ret;
+}
+
+
+static int blkvsc_drv_exit_cb(struct device *dev, void *data)
+{
+       struct device **curr = (struct device **)data;
+       *curr = dev;
+       return 1; // stop iterating
+}
+
+/*++
+
+Name:  blkvsc_drv_exit()
+
+Desc:
+
+--*/
+void blkvsc_drv_exit(void)
+{
+       STORVSC_DRIVER_OBJECT *storvsc_drv_obj=&g_blkvsc_drv.drv_obj;
+       struct driver_context *drv_ctx=&g_blkvsc_drv.drv_ctx;
+
+       struct device *current_dev=NULL;
+
+#if defined(KERNEL_2_6_5) || defined(KERNEL_2_6_9)
+#define driver_for_each_device(drv, start, data, fn) \
+       struct list_head *ptr, *n; \
+       list_for_each_safe(ptr, n, &((drv)->devices)) {\
+               struct device *curr_dev;\
+               curr_dev = list_entry(ptr, struct device, driver_list);\
+               fn(curr_dev, data);\
+       }
+#endif // KERNEL_2_6_9
+
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       while (1)
+       {
+               current_dev = NULL;
+
+               // Get the device
+               driver_for_each_device(&drv_ctx->driver, NULL, (void*)&current_dev, blkvsc_drv_exit_cb);
+
+               if (current_dev == NULL)
+                       break;
+
+               // Initiate removal from the top-down
+               device_unregister(current_dev);
+       }
+
+       if (storvsc_drv_obj->Base.OnCleanup)
+               storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
+
+       vmbus_child_driver_unregister(drv_ctx);
+
+       DPRINT_EXIT(BLKVSC_DRV);
+
+       return;
+}
+
+/*++
+
+Name:  blkvsc_probe()
+
+Desc:  Add a new device for this driver
+
+--*/
+static int blkvsc_probe(struct device *device)
+{
+       int ret=0;
+
+       struct driver_context *driver_ctx = driver_to_driver_context(device->driver);
+       struct blkvsc_driver_context *blkvsc_drv_ctx = (struct blkvsc_driver_context*)driver_ctx;
+       STORVSC_DRIVER_OBJECT* storvsc_drv_obj = &blkvsc_drv_ctx->drv_obj;
+
+       struct device_context *device_ctx = device_to_device_context(device);
+       DEVICE_OBJECT* device_obj = &device_ctx->device_obj;
+
+       struct block_device_context *blkdev=NULL;
+       STORVSC_DEVICE_INFO device_info;
+       int major=0;
+       int devnum=0;
+
+       static int ide0_registered=0;
+       static int ide1_registered=0;
+
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
+
+       if (!storvsc_drv_obj->Base.OnDeviceAdd)
+       {
+               DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
+
+               ret = -1;
+               goto Cleanup;
+       }
+
+       blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
+       if (!blkdev)
+       {
+               ret = -ENOMEM;
+               goto Cleanup;
+       }
+
+       INIT_LIST_HEAD(&blkdev->pending_list);
+
+       // Initialize what we can here
+       spin_lock_init(&blkdev->lock);
+
+       ASSERT(sizeof(struct blkvsc_request_group) <= sizeof(struct blkvsc_request));
+
+#ifdef KERNEL_2_6_27
+        blkdev->request_pool = kmem_cache_create(device_ctx->device.bus_id,
+                sizeof(struct blkvsc_request) + storvsc_drv_obj->RequestExtSize, 0,
+                SLAB_HWCACHE_ALIGN, NULL);
+#else
+       blkdev->request_pool = kmem_cache_create(device_ctx->device.bus_id,
+               sizeof(struct blkvsc_request) + storvsc_drv_obj->RequestExtSize, 0,
+               SLAB_HWCACHE_ALIGN, NULL, NULL);
+#endif
+       if (!blkdev->request_pool)
+       {
+               ret = -ENOMEM;
+               goto Cleanup;
+       }
+
+
+       // Call to the vsc driver to add the device
+       ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
+       if (ret != 0)
+       {
+               DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
+               goto Cleanup;
+       }
+
+       blkdev->device_ctx = device_ctx;
+       blkdev->target = device_info.TargetId; // this identified the device 0 or 1
+       blkdev->path = device_info.PathId; // this identified the ide ctrl 0 or 1
+
+       device->driver_data = blkdev;
+
+       // Calculate the major and device num
+       if (blkdev->path == 0)
+       {
+               major = IDE0_MAJOR;
+               devnum = blkdev->path + blkdev->target;         // 0 or 1
+
+               if (!ide0_registered)
+               {
+                       ret = register_blkdev(major, "ide");
+                       if (ret != 0)
+                       {
+                               DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
+                               goto Remove;
+                       }
+
+                       ide0_registered = 1;
+               }
+       }
+       else if (blkdev->path == 1)
+       {
+               major = IDE1_MAJOR;
+               devnum = blkdev->path + blkdev->target + 1; // 2 or 3
+
+               if (!ide1_registered)
+               {
+                       ret = register_blkdev(major, "ide");
+                       if (ret != 0)
+                       {
+                               DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
+                               goto Remove;
+                       }
+
+                       ide1_registered = 1;
+               }
+
+       }
+       else
+       {
+               DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
+               ret = -1;
+               goto Cleanup;
+       }
+
+       DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
+
+       blkdev->gd = alloc_disk(BLKVSC_MINORS);
+       if (!blkdev->gd)
+       {
+               DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
+               ret = -1;
+               goto Cleanup;
+       }
+
+       blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
+
+       blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
+       blk_queue_max_phys_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
+       blk_queue_max_hw_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
+       blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
+       blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
+       blk_queue_dma_alignment(blkdev->gd->queue, 511);
+
+       blkdev->gd->major = major;
+       if (devnum == 1 || devnum == 3)
+               blkdev->gd->first_minor = BLKVSC_MINORS;
+       else
+               blkdev->gd->first_minor = 0;
+       blkdev->gd->fops = &block_ops;
+       blkdev->gd->private_data = blkdev;
+       sprintf(blkdev->gd->disk_name, "hd%c", 'a'+ devnum);
+
+       blkvsc_do_inquiry(blkdev);
+       if (blkdev->device_type == DVD_TYPE)
+       {
+               set_disk_ro(blkdev->gd, 1);
+               blkdev->gd->flags |= GENHD_FL_REMOVABLE;
+               blkvsc_do_read_capacity(blkdev);
+       }
+       else
+       {
+               blkvsc_do_read_capacity16(blkdev);
+       }
+
+       set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
+       blk_queue_hardsect_size(blkdev->gd->queue, blkdev->sector_size);
+       // go!
+       add_disk(blkdev->gd);
+
+       DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %llu sector_size %d", blkdev->gd->disk_name, blkdev->capacity, blkdev->sector_size);
+
+       return ret;
+
+Remove:
+       storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
+
+Cleanup:
+       if (blkdev)
+       {
+               if (blkdev->request_pool)
+               {
+                       kmem_cache_destroy(blkdev->request_pool);
+                       blkdev->request_pool = NULL;
+               }
+               kfree(blkdev);
+               blkdev = NULL;
+       }
+
+       DPRINT_EXIT(BLKVSC_DRV);
+
+       return ret;
+}
+
+static void blkvsc_shutdown(struct device *device)
+{
+       struct block_device_context *blkdev = (struct block_device_context*)device->driver_data;
+       unsigned long flags;
+
+       if (!blkdev)
+               return;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n", blkdev->users, blkdev->gd->disk_name);
+
+       spin_lock_irqsave(&blkdev->lock, flags);
+
+       blkdev->shutting_down = 1;
+
+       blk_stop_queue(blkdev->gd->queue);
+
+       spin_unlock_irqrestore(&blkdev->lock, flags);
+
+       while (blkdev->num_outstanding_reqs)
+       {
+               DPRINT_INFO(STORVSC, "waiting for %d requests to complete...", blkdev->num_outstanding_reqs);
+
+               udelay(100);
+       }
+
+       blkvsc_do_flush(blkdev);
+
+       spin_lock_irqsave(&blkdev->lock, flags);
+
+       blkvsc_cancel_pending_reqs(blkdev);
+
+       spin_unlock_irqrestore(&blkdev->lock, flags);
+}
+
+static int blkvsc_do_flush(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *blkvsc_req=NULL;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
+
+       if (blkdev->device_type != HARDDISK_TYPE)
+               return 0;
+
+       blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
+       if (!blkvsc_req)
+       {
+               return -ENOMEM;
+       }
+
+       memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
+       init_waitqueue_head(&blkvsc_req->wevent);
+       blkvsc_req->dev = blkdev;
+       blkvsc_req->req = NULL;
+       blkvsc_req->write = 0;
+
+       blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
+       blkvsc_req->request.DataBuffer.Offset = 0;
+       blkvsc_req->request.DataBuffer.Length = 0;
+
+       blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
+       blkvsc_req->cmd_len = 10;
+
+       // Set this here since the completion routine may be invoked and completed before we return
+       blkvsc_req->cond =0;
+       blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
+
+       wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
+
+       kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+
+       return 0;
+}
+
+// Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd)
+static int blkvsc_do_inquiry(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *blkvsc_req=NULL;
+       struct page *page_buf;
+       unsigned char *buf;
+       unsigned char device_type;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
+
+       blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
+       if (!blkvsc_req)
+       {
+               return -ENOMEM;
+       }
+
+       memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
+       page_buf = alloc_page(GFP_KERNEL);
+       if (!page_buf)
+       {
+               kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+               return -ENOMEM;
+       }
+
+       init_waitqueue_head(&blkvsc_req->wevent);
+       blkvsc_req->dev = blkdev;
+       blkvsc_req->req = NULL;
+       blkvsc_req->write = 0;
+
+       blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
+       blkvsc_req->request.DataBuffer.Offset = 0;
+       blkvsc_req->request.DataBuffer.Length = 64;
+
+       blkvsc_req->cmnd[0] = INQUIRY;
+       blkvsc_req->cmnd[1] = 0x1;              // Get product data
+       blkvsc_req->cmnd[2] = 0x83;             // mode page 83
+       blkvsc_req->cmnd[4] = 64;
+       blkvsc_req->cmd_len = 6;
+
+       // Set this here since the completion routine may be invoked and completed before we return
+       blkvsc_req->cond =0;
+
+       blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
+
+       DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n", blkvsc_req, blkvsc_req->cond);
+
+       wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
+
+       buf = kmap(page_buf);
+
+       //PrintBytes(buf, 64);
+       // be to le
+       device_type = buf[0] & 0x1F;
+
+       if (device_type == 0x0)
+       {
+               blkdev->device_type = HARDDISK_TYPE;
+       }
+       else if (device_type == 0x5)
+       {
+               blkdev->device_type = DVD_TYPE;
+       }
+       else
+       {
+               // TODO: this is currently unsupported device type
+               blkdev->device_type = UNKNOWN_DEV_TYPE;
+       }
+
+       DPRINT_DBG(BLKVSC_DRV, "device type %d \n", device_type);
+
+       blkdev->device_id_len = buf[7];
+       if (blkdev->device_id_len > 64)
+               blkdev->device_id_len = 64;
+
+       memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
+       //PrintBytes(blkdev->device_id, blkdev->device_id_len);
+
+       kunmap(page_buf);
+
+       __free_page(page_buf);
+
+       kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+
+       return 0;
+}
+
+// Do a scsi READ_CAPACITY cmd here to get the size of the disk
+static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *blkvsc_req=NULL;
+       struct page *page_buf;
+       unsigned char *buf;
+       struct scsi_sense_hdr sense_hdr;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
+
+       blkdev->sector_size = 0;
+       blkdev->capacity = 0;
+       blkdev->media_not_present = 0; // assume a disk is present
+
+       blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
+       if (!blkvsc_req)
+       {
+               return -ENOMEM;
+       }
+
+       memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
+       page_buf = alloc_page(GFP_KERNEL);
+       if (!page_buf)
+       {
+               kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+               return -ENOMEM;
+       }
+
+       init_waitqueue_head(&blkvsc_req->wevent);
+       blkvsc_req->dev = blkdev;
+       blkvsc_req->req = NULL;
+       blkvsc_req->write = 0;
+
+       blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
+       blkvsc_req->request.DataBuffer.Offset = 0;
+       blkvsc_req->request.DataBuffer.Length = 8;
+
+       blkvsc_req->cmnd[0] = READ_CAPACITY;
+       blkvsc_req->cmd_len = 16;
+
+       // Set this here since the completion routine may be invoked and completed before we return
+       blkvsc_req->cond =0;
+
+       blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
+
+       DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n", blkvsc_req, blkvsc_req->cond);
+
+       wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
+
+       // check error
+       if (blkvsc_req->request.Status)
+       {
+               scsi_normalize_sense(blkvsc_req->sense_buffer, SCSI_SENSE_BUFFERSIZE, &sense_hdr);
+
+               if (sense_hdr.asc == 0x3A) // Medium not present
+               {
+                       blkdev->media_not_present = 1;
+               }
+
+               return 0;
+       }
+       buf = kmap(page_buf);
+
+       // be to le
+       blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3]) + 1;
+       blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) | (buf[6] << 8) | buf[7];
+
+       kunmap(page_buf);
+
+       __free_page(page_buf);
+
+       kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+
+       return 0;
+}
+
+
+static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *blkvsc_req=NULL;
+       struct page *page_buf;
+       unsigned char *buf;
+       struct scsi_sense_hdr sense_hdr;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
+
+       blkdev->sector_size = 0;
+       blkdev->capacity = 0;
+       blkdev->media_not_present = 0; // assume a disk is present
+
+       blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
+       if (!blkvsc_req)
+       {
+               return -ENOMEM;
+       }
+
+       memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
+       page_buf = alloc_page(GFP_KERNEL);
+       if (!page_buf)
+       {
+               kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+               return -ENOMEM;
+       }
+
+       init_waitqueue_head(&blkvsc_req->wevent);
+       blkvsc_req->dev = blkdev;
+       blkvsc_req->req = NULL;
+       blkvsc_req->write = 0;
+
+       blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
+       blkvsc_req->request.DataBuffer.Offset = 0;
+       blkvsc_req->request.DataBuffer.Length = 12;
+
+       blkvsc_req->cmnd[0] = 0x9E; //READ_CAPACITY16;
+       blkvsc_req->cmd_len = 16;
+
+       // Set this here since the completion routine may be invoked and completed before we return
+       blkvsc_req->cond =0;
+
+       blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
+
+       DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n", blkvsc_req, blkvsc_req->cond);
+
+       wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
+
+       // check error
+       if (blkvsc_req->request.Status)
+       {
+               scsi_normalize_sense(blkvsc_req->sense_buffer, SCSI_SENSE_BUFFERSIZE, &sense_hdr);
+
+               if (sense_hdr.asc == 0x3A) // Medium not present
+               {
+                       blkdev->media_not_present = 1;
+               }
+
+               return 0;
+       }
+       buf = kmap(page_buf);
+
+       // be to le
+       blkdev->capacity = be64_to_cpu(*(unsigned long long*) &buf[0]) + 1;
+       blkdev->sector_size = be32_to_cpu(*(unsigned int*)&buf[8]);
+
+       //blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) | (buf[2] << 8) | buf[3]) + 1;
+       //blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) | (buf[6] << 8) | buf[7];
+
+       kunmap(page_buf);
+
+       __free_page(page_buf);
+
+       kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
+
+       return 0;
+}
+
+/*++
+
+Name:  blkvsc_remove()
+
+Desc:  Callback when our device is removed
+
+--*/
+static int blkvsc_remove(struct device *device)
+{
+       int ret=0;
+
+       struct driver_context *driver_ctx = driver_to_driver_context(device->driver);
+       struct blkvsc_driver_context *blkvsc_drv_ctx = (struct blkvsc_driver_context*)driver_ctx;
+       STORVSC_DRIVER_OBJECT* storvsc_drv_obj = &blkvsc_drv_ctx->drv_obj;
+
+       struct device_context *device_ctx = device_to_device_context(device);
+       DEVICE_OBJECT* device_obj = &device_ctx->device_obj;
+       struct block_device_context *blkdev = (struct block_device_context*)device->driver_data;
+       unsigned long flags;
+
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
+
+       if (!storvsc_drv_obj->Base.OnDeviceRemove)
+       {
+               DPRINT_EXIT(BLKVSC_DRV);
+               return -1;
+       }
+
+       // Call to the vsc driver to let it know that the device is being removed
+       ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
+       if (ret != 0)
+       {
+               // TODO:
+               DPRINT_ERR(BLKVSC_DRV, "unable to remove blkvsc device (ret %d)", ret);
+       }
+
+       // Get to a known state
+       spin_lock_irqsave(&blkdev->lock, flags);
+
+       blkdev->shutting_down = 1;
+
+       blk_stop_queue(blkdev->gd->queue);
+
+       spin_unlock_irqrestore(&blkdev->lock, flags);
+
+       while (blkdev->num_outstanding_reqs)
+       {
+               DPRINT_INFO(STORVSC, "waiting for %d requests to complete...", blkdev->num_outstanding_reqs);
+
+               udelay(100);
+       }
+
+       blkvsc_do_flush(blkdev);
+
+       spin_lock_irqsave(&blkdev->lock, flags);
+
+       blkvsc_cancel_pending_reqs(blkdev);
+
+       spin_unlock_irqrestore(&blkdev->lock, flags);
+
+       blk_cleanup_queue(blkdev->gd->queue);
+
+       del_gendisk(blkdev->gd);
+
+       kmem_cache_destroy(blkdev->request_pool);
+
+       kfree(blkdev);
+
+       DPRINT_EXIT(BLKVSC_DRV);
+
+       return ret;
+}
+
+static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
+{
+       ASSERT(blkvsc_req->req);
+       ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8));
+
+       blkvsc_req->cmd_len = 16;
+
+       if (blkvsc_req->sector_start > 0xffffffff)
+       {
+               if (rq_data_dir(blkvsc_req->req))
+               {
+                       blkvsc_req->write = 1;
+                       blkvsc_req->cmnd[0] = WRITE_16;
+               }
+               else
+               {
+                       blkvsc_req->write = 0;
+                       blkvsc_req->cmnd[0] = READ_16;
+               }
+
+               blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
+
+               *(unsigned long long*)&blkvsc_req->cmnd[2] = cpu_to_be64(blkvsc_req->sector_start);
+               *(unsigned int*)&blkvsc_req->cmnd[10] = cpu_to_be32(blkvsc_req->sector_count);
+       }
+       else if ((blkvsc_req->sector_count > 0xff) || (blkvsc_req->sector_start > 0x1fffff))
+       {
+               if (rq_data_dir(blkvsc_req->req))
+               {
+                       blkvsc_req->write = 1;
+                       blkvsc_req->cmnd[0] = WRITE_10;
+               }
+               else
+               {
+                       blkvsc_req->write = 0;
+                       blkvsc_req->cmnd[0] = READ_10;
+               }
+
+               blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
+
+               *(unsigned int *)&blkvsc_req->cmnd[2] = cpu_to_be32(blkvsc_req->sector_start);
+               *(unsigned short*)&blkvsc_req->cmnd[7] = cpu_to_be16(blkvsc_req->sector_count);
+    }
+       else
+       {
+               if (rq_data_dir(blkvsc_req->req))
+               {
+                       blkvsc_req->write = 1;
+                       blkvsc_req->cmnd[0] = WRITE_6;
+               }
+               else
+               {
+                       blkvsc_req->write = 0;
+                       blkvsc_req->cmnd[0] = READ_6;
+               }
+
+               *(unsigned int *)&blkvsc_req->cmnd[1] = cpu_to_be32(blkvsc_req->sector_start) >> 8;
+               blkvsc_req->cmnd[1] &= 0x1f;
+               blkvsc_req->cmnd[4] = (unsigned char) blkvsc_req->sector_count;
+       }
+}
+
+static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req, void (*request_completion)(STORVSC_REQUEST*) )
+{
+       struct block_device_context *blkdev = blkvsc_req->dev;
+       struct device_context *device_ctx=blkdev->device_ctx;
+       struct driver_context *driver_ctx = driver_to_driver_context(device_ctx->device.driver);
+       struct blkvsc_driver_context *blkvsc_drv_ctx = (struct blkvsc_driver_context*)driver_ctx;
+       STORVSC_DRIVER_OBJECT* storvsc_drv_obj = &blkvsc_drv_ctx->drv_obj;
+       int ret =0;
+
+       STORVSC_REQUEST *storvsc_req;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - req %p type %s start_sector %llu count %d offset %d len %d\n",
+               blkvsc_req,
+               (blkvsc_req->write)?"WRITE":"READ",
+               blkvsc_req->sector_start,
+               blkvsc_req->sector_count,
+               blkvsc_req->request.DataBuffer.Offset,
+               blkvsc_req->request.DataBuffer.Length);
+
+       /*for (i=0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++)
+       {
+               DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - req %p pfn[%d] %llx\n",
+               blkvsc_req,
+               i,
+               blkvsc_req->request.DataBuffer.PfnArray[i]);
+       }*/
+
+       storvsc_req = &blkvsc_req->request;
+       storvsc_req->Extension = (void*)((unsigned long)blkvsc_req + sizeof(struct blkvsc_request));
+
+       storvsc_req->Type = blkvsc_req->write? WRITE_TYPE : READ_TYPE;
+
+       storvsc_req->OnIOCompletion = request_completion;
+       storvsc_req->Context = blkvsc_req;
+
+       storvsc_req->Host = blkdev->port;
+       storvsc_req->Bus = blkdev->path;
+       storvsc_req->TargetId = blkdev->target;
+       storvsc_req->LunId = 0;  // this is not really used at all
+
+       storvsc_req->CdbLen = blkvsc_req->cmd_len;
+       storvsc_req->Cdb = blkvsc_req->cmnd;
+
+       storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
+       storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
+
+       ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj, &blkvsc_req->request);
+       if (ret == 0)
+       {
+               blkdev->num_outstanding_reqs++;
+       }
+
+       return ret;
+}
+
+//
+// We break the request into 1 or more blkvsc_requests and submit them.
+// If we cant submit them all, we put them on the pending_list. The
+// blkvsc_request() will work on the pending_list.
+//
+static int blkvsc_do_request(struct block_device_context *blkdev, struct request *req)
+{
+       struct bio *bio=NULL;
+       struct bio_vec *bvec=NULL;
+       struct bio_vec *prev_bvec=NULL;
+
+       struct blkvsc_request *blkvsc_req=NULL;
+       struct blkvsc_request *tmp;
+       int databuf_idx=0;
+       int seg_idx=0;
+
+       sector_t start_sector;
+       unsigned long num_sectors = 0;
+       int ret=0;
+       int pending=0;
+       struct blkvsc_request_group *group=NULL;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %llu \n", blkdev, req, req->sector);
+
+       // Create a group to tie req to list of blkvsc_reqs
+       group = (struct blkvsc_request_group*)kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
+       if (!group)
+       {
+               return -ENOMEM;
+       }
+
+       INIT_LIST_HEAD(&group->blkvsc_req_list);
+       group->outstanding = group->status = 0;
+
+       start_sector = req->sector;
+
+       // foreach bio in the request
+       if (req->bio)
+        for (bio = req->bio; bio; bio = bio->bi_next)
+       {
+               // Map this bio into an existing or new storvsc request
+               bio_for_each_segment (bvec, bio, seg_idx)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() - req %p bio %p bvec %p seg_idx %d databuf_idx %d\n",
+                                                       req, bio, bvec, seg_idx, databuf_idx);
+
+                       // Get a new storvsc request
+                       if ( (!blkvsc_req) ||                                                                   // 1st-time
+                                (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT) ||
+                                (bvec->bv_offset != 0) ||                                                      // hole at the begin of page
+                                (prev_bvec && (prev_bvec->bv_len != PAGE_SIZE)) )      // hold at the end of page
+                       {
+                               // submit the prev one
+                               if (blkvsc_req)
+                               {
+                                       blkvsc_req->sector_start = start_sector;
+                                       sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
+
+                                       blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
+
+                                       blkvsc_init_rw(blkvsc_req);
+                               }
+
+                               // Create new blkvsc_req to represent the current bvec
+                               blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
+                               if (!blkvsc_req)
+                               {
+                                       // free up everything
+                                       list_for_each_entry_safe(blkvsc_req, tmp, &group->blkvsc_req_list, req_entry)
+                                       {
+                                               list_del(&blkvsc_req->req_entry);
+                                               kmem_cache_free(blkdev->request_pool, blkvsc_req);
+                                       }
+
+                                       kmem_cache_free(blkdev->request_pool, group);
+                                       return -ENOMEM;
+                               }
+
+                               memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
+
+                               blkvsc_req->dev = blkdev;
+                               blkvsc_req->req = req;
+                               blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
+                               blkvsc_req->request.DataBuffer.Length = 0;
+
+                               // Add to the group
+                               blkvsc_req->group = group;
+                               blkvsc_req->group->outstanding++;
+                               list_add_tail(&blkvsc_req->req_entry, &blkvsc_req->group->blkvsc_req_list);
+
+                               start_sector += num_sectors;
+                               num_sectors = 0;
+                               databuf_idx = 0;
+                       }
+
+                       // Add the curr bvec/segment to the curr blkvsc_req
+                       blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
+                       blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
+
+                       prev_bvec = bvec;
+
+                       databuf_idx++;
+                       num_sectors += bvec->bv_len >> 9;
+
+               } // bio_for_each_segment
+
+       } // rq_for_each_bio
+
+       // Handle the last one
+       if (blkvsc_req)
+       {
+               DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n", blkdev, req, blkvsc_req->group, blkvsc_req->group->outstanding);
+
+               blkvsc_req->sector_start = start_sector;
+               sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
+
+               blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
+
+               blkvsc_init_rw(blkvsc_req);
+       }
+
+       list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry)
+       {
+               if (pending)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to pending_list - blkvsc_req %p start_sect %llu sect_count %d (%llu %d)\n",
+                               blkvsc_req, blkvsc_req->sector_start, blkvsc_req->sector_count, start_sector, num_sectors);
+
+                       list_add_tail(&blkvsc_req->pend_entry, &blkdev->pending_list);
+               }
+               else
+               {
+                       ret = blkvsc_submit_request(blkvsc_req, blkvsc_request_completion);
+                       if (ret == -1)
+                       {
+                               pending = 1;
+                               list_add_tail(&blkvsc_req->pend_entry, &blkdev->pending_list);
+                       }
+
+                       DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p start_sect %llu sect_count %d (%llu %d) ret %d\n",
+                               blkvsc_req, blkvsc_req->sector_start, blkvsc_req->sector_count, start_sector, num_sectors, ret);
+               }
+       }
+
+       return pending;
+}
+
+static void blkvsc_cmd_completion(STORVSC_REQUEST* request)
+{
+       struct blkvsc_request *blkvsc_req=(struct blkvsc_request*)request->Context;
+       struct block_device_context *blkdev = (struct block_device_context*)blkvsc_req->dev;
+
+       struct scsi_sense_hdr sense_hdr;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n", blkvsc_req);
+
+       blkdev->num_outstanding_reqs--;
+
+       if (blkvsc_req->request.Status)
+       {
+               if (scsi_normalize_sense(blkvsc_req->sense_buffer, SCSI_SENSE_BUFFERSIZE, &sense_hdr))
+               {
+                       scsi_print_sense_hdr("blkvsc", &sense_hdr);
+               }
+       }
+
+       blkvsc_req->cond =1;
+       wake_up_interruptible(&blkvsc_req->wevent);
+}
+
+static void blkvsc_request_completion(STORVSC_REQUEST* request)
+{
+       struct blkvsc_request *blkvsc_req=(struct blkvsc_request*)request->Context;
+       struct block_device_context *blkdev = (struct block_device_context*)blkvsc_req->dev;
+       unsigned long flags;
+       struct blkvsc_request *comp_req, *tmp;
+
+       ASSERT(blkvsc_req->group);
+
+       DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s sect_start %llu sect_count %d len %d group outstd %d total outstd %d\n",
+               blkdev,
+               blkvsc_req,
+               blkvsc_req->group,
+               (blkvsc_req->write)?"WRITE":"READ",
+               blkvsc_req->sector_start,
+               blkvsc_req->sector_count,
+               blkvsc_req->request.DataBuffer.Length,
+               blkvsc_req->group->outstanding,
+               blkdev->num_outstanding_reqs);
+
+       spin_lock_irqsave(&blkdev->lock, flags);
+
+       blkdev->num_outstanding_reqs--;
+       blkvsc_req->group->outstanding--;
+
+       // Only start processing when all the blkvsc_reqs are completed. This guarantees no out-of-order
+       // blkvsc_req completion when calling end_that_request_first()
+       if (blkvsc_req->group->outstanding == 0)
+       {
+               list_for_each_entry_safe(comp_req, tmp, &blkvsc_req->group->blkvsc_req_list, req_entry)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p sect_start %llu sect_count %d \n",
+                               comp_req,
+                               comp_req->sector_start,
+                               comp_req->sector_count);
+
+                       list_del(&comp_req->req_entry);
+
+#ifdef KERNEL_2_6_27
+                       if (!__blk_end_request(
+                               comp_req->req,
+                               (!comp_req->request.Status ? 0: -EIO),
+                               comp_req->sector_count * blkdev->sector_size))
+                       {
+                               //All the sectors have been xferred ie the request is done
+                               DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n", comp_req->req);
+                               kmem_cache_free(blkdev->request_pool, comp_req->group);
+                       }
+#else
+                       if (!end_that_request_first(comp_req->req, !comp_req->request.Status,  (comp_req->sector_count * (blkdev->sector_size >> 9))))
+                       {
+                               //All the sectors have been xferred ie the request is done
+                               DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n", comp_req->req);
+
+                               end_that_request_last(comp_req->req, !comp_req->request.Status);
+
+                               kmem_cache_free(blkdev->request_pool, comp_req->group);
+                       }
+#endif
+
+                       kmem_cache_free(blkdev->request_pool, comp_req);
+               }
+
+               if (!blkdev->shutting_down)
+               {
+                       blkvsc_do_pending_reqs(blkdev);
+                       blk_start_queue(blkdev->gd->queue);
+                       blkvsc_request(blkdev->gd->queue);
+               }
+       }
+
+       spin_unlock_irqrestore(&blkdev->lock, flags);
+}
+
+static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *pend_req, *tmp;
+       struct blkvsc_request *comp_req, *tmp2;
+
+       int ret=0;
+
+       DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
+
+       // Flush the pending list first
+       list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list, pend_entry)
+       {
+               // The pend_req could be part of a partially completed request. If so, complete those req first
+               // until we hit the pend_req
+               list_for_each_entry_safe(comp_req, tmp2, &pend_req->group->blkvsc_req_list, req_entry)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p sect_start %llu sect_count %d \n",
+                               comp_req,
+                               comp_req->sector_start,
+                               comp_req->sector_count);
+
+                       if (comp_req == pend_req)
+                               break;
+
+                       list_del(&comp_req->req_entry);
+
+                       if (comp_req->req)
+                       {
+#ifdef KERNEL_2_6_27
+                       ret = __blk_end_request(
+                           comp_req->req,
+                           (!comp_req->request.Status ? 0 : -EIO),
+                           comp_req->sector_count * blkdev->sector_size);
+#else
+                       ret = end_that_request_first(comp_req->req, !comp_req->request.Status,  (comp_req->sector_count * (blkdev->sector_size >> 9)));
+#endif
+                       ASSERT(ret != 0);
+                       }
+
+                       kmem_cache_free(blkdev->request_pool, comp_req);
+               }
+
+               DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n", pend_req);
+
+               list_del(&pend_req->pend_entry);
+
+               list_del(&pend_req->req_entry);
+
+               if (comp_req->req)
+               {
+#ifdef KERNEL_2_6_27
+               if (!__blk_end_request(
+                       pend_req->req,
+                       -EIO,
+                       pend_req->sector_count * blkdev->sector_size))
+               {
+                       //All the sectors have been xferred ie the request is done
+                       DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs() - req %p COMPLETED\n", pend_req->req);
+                       kmem_cache_free(blkdev->request_pool, pend_req->group);
+               }
+#else
+               if (!end_that_request_first(pend_req->req, 0,  (pend_req->sector_count * (blkdev->sector_size >> 9))))
+               {
+                       //All the sectors have been xferred ie the request is done
+                       DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs() - req %p COMPLETED\n", pend_req->req);
+
+                       end_that_request_last(pend_req->req, 0);
+
+                       kmem_cache_free(blkdev->request_pool, pend_req->group);
+               }
+#endif
+               }
+
+               kmem_cache_free(blkdev->request_pool, pend_req);
+       }
+
+       return ret;
+}
+
+static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
+{
+       struct blkvsc_request *pend_req, *tmp;
+       int ret=0;
+
+       // Flush the pending list first
+       list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list, pend_entry)
+       {
+               DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n", pend_req);
+
+               ret = blkvsc_submit_request(pend_req, blkvsc_request_completion);
+               if (ret != 0)
+               {
+                       break;
+               }
+               else
+               {
+                       list_del(&pend_req->pend_entry);
+               }
+       }
+
+       return ret;
+}
+
+static void blkvsc_request(struct request_queue *queue)
+{
+       struct block_device_context *blkdev = NULL;
+       struct request *req;
+       int ret=0;
+
+       DPRINT_DBG(BLKVSC_DRV, "- enter \n");
+       while ((req = elv_next_request(queue)) != NULL)
+       {
+               DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
+
+               blkdev = req->rq_disk->private_data;
+               if (blkdev->shutting_down || !blk_fs_request(req) || blkdev->media_not_present) {
+                       end_request(req, 0);
+                       continue;
+               }
+
+               ret = blkvsc_do_pending_reqs(blkdev);
+
+               if (ret != 0)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "- stop queue - pending_list not empty\n");
+                       blk_stop_queue(queue);
+                       break;
+               }
+
+               blkdev_dequeue_request(req);
+
+               ret = blkvsc_do_request(blkdev, req);
+               if (ret > 0)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
+                       blk_stop_queue(queue);
+                       break;
+               }
+               else if (ret < 0)
+               {
+                       DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
+                       blk_requeue_request(queue, req);
+                       blk_stop_queue(queue);
+                       break;
+               }
+       }
+}
+
+static int blkvsc_open(struct inode *inode, struct file *filep)
+{
+       struct block_device_context *blkdev = inode->i_bdev->bd_disk->private_data;
+
+       DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users, blkdev->gd->disk_name);
+
+       spin_lock(&blkdev->lock);
+
+       if (!blkdev->users && blkdev->device_type == DVD_TYPE)
+       {
+               spin_unlock(&blkdev->lock);
+               check_disk_change(inode->i_bdev);
+               spin_lock(&blkdev->lock);
+       }
+
+       blkdev->users++;
+
+       spin_unlock(&blkdev->lock);
+       return 0;
+}
+
+static int blkvsc_release(struct inode *inode, struct file *filep)
+{
+       struct block_device_context *blkdev = inode->i_bdev->bd_disk->private_data;
+
+       DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users, blkdev->gd->disk_name);
+
+       spin_lock(&blkdev->lock);
+       if (blkdev->users == 1)
+       {
+               spin_unlock(&blkdev->lock);
+               blkvsc_do_flush(blkdev);
+               spin_lock(&blkdev->lock);
+       }
+
+       blkdev->users--;
+
+       spin_unlock(&blkdev->lock);
+       return 0;
+}
+
+static int blkvsc_media_changed(struct gendisk *gd)
+{
+       DPRINT_DBG(BLKVSC_DRV, "- enter\n");
+
+       return 1;
+}
+
+static int blkvsc_revalidate_disk(struct gendisk *gd)
+{
+       struct block_device_context *blkdev = gd->private_data;
+
+       DPRINT_DBG(BLKVSC_DRV, "- enter\n");
+
+       if (blkdev->device_type == DVD_TYPE)
+       {
+               blkvsc_do_read_capacity(blkdev);
+               set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
+               blk_queue_hardsect_size(gd->queue, blkdev->sector_size);
+       }
+       return 0;
+}
+
+int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
+{
+       sector_t total_sectors = get_capacity(bd->bd_disk);
+       sector_t cylinder_times_heads=0;
+       sector_t temp=0;
+
+       int sectors_per_track=0;
+       int heads=0;
+       int cylinders=0;
+       int rem=0;
+
+    if (total_sectors > (65535 * 16 * 255)) {
+        total_sectors = (65535 * 16 * 255);
+    }
+
+    if (total_sectors >= (65535 * 16 * 63)) {
+        sectors_per_track = 255;
+        heads = 16;
+
+               cylinder_times_heads = total_sectors;
+               rem = sector_div(cylinder_times_heads, sectors_per_track); // sector_div stores the quotient in cylinder_times_heads
+    }
+       else
+       {
+        sectors_per_track = 17;
+
+               cylinder_times_heads = total_sectors;
+        rem = sector_div(cylinder_times_heads, sectors_per_track);     // sector_div stores the quotient in cylinder_times_heads
+
+               temp = cylinder_times_heads + 1023;
+               rem = sector_div(temp, 1024);   // sector_div stores the quotient in temp
+
+               heads = temp;
+
+        if (heads < 4) {
+            heads = 4;
+        }
+
+        if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
+            sectors_per_track = 31;
+            heads = 16;
+
+                       cylinder_times_heads = total_sectors;
+            rem = sector_div(cylinder_times_heads, sectors_per_track); // sector_div stores the quotient in cylinder_times_heads
+        }
+
+        if (cylinder_times_heads >= (heads * 1024)) {
+            sectors_per_track = 63;
+            heads = 16;
+
+                       cylinder_times_heads = total_sectors;
+            rem = sector_div(cylinder_times_heads, sectors_per_track); // sector_div stores the quotient in cylinder_times_heads
+        }
+    }
+
+       temp = cylinder_times_heads;
+    rem = sector_div(temp, heads); // sector_div stores the quotient in temp
+       cylinders = temp;
+
+       hg->heads = heads;
+    hg->sectors = sectors_per_track;
+    hg->cylinders = cylinders;
+
+       DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads, sectors_per_track);
+
+    return 0;
+}
+
+static int blkvsc_ioctl(struct inode *inode, struct file *filep, unsigned cmd, unsigned long arg)
+{
+       struct block_device *bd = inode->i_bdev;
+       struct block_device_context *blkdev = bd->bd_disk->private_data;
+       int ret=0;
+
+       switch (cmd)
+       {
+       // TODO: I think there is certain format for HDIO_GET_IDENTITY rather than just
+       // a GUID. Commented it out for now.
+       /*case HDIO_GET_IDENTITY:
+               DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
+
+               if (copy_to_user((void __user *)arg, blkdev->device_id, blkdev->device_id_len))
+               {
+                       ret = -EFAULT;
+               }
+
+               break;*/
+       default:
+               ret = -EINVAL;
+               break;
+       }
+
+       return ret;
+}
+
+
+MODULE_LICENSE("GPL");
+
+static int __init blkvsc_init(void)
+{
+       int ret;
+
+       ASSERT(sizeof(sector_t) == 8); // Make sure CONFIG_LBD is set
+
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
+
+       ret = blkvsc_drv_init(BlkVscInitialize);
+
+       DPRINT_EXIT(BLKVSC_DRV);
+
+       return ret;
+}
+
+static void __exit blkvsc_exit(void)
+{
+       DPRINT_ENTER(BLKVSC_DRV);
+
+       blkvsc_drv_exit();
+
+       DPRINT_ENTER(BLKVSC_DRV);
+}
+
+module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
+
+module_init(blkvsc_init);
+module_exit(blkvsc_exit);
+
+// eof