drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "hv_api.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 /*
  50  * This request ties the struct request and struct
  51  * blkvsc_request/hv_storvsc_request together A struct request may be
  52  * represented by 1 or more struct blkvsc_request
  53  */
  54 struct blkvsc_request_group {
  55         int outstanding;
  56         int status;
  57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  58 };
  59
  60 struct blkvsc_request {
  61         /* blkvsc_request_group.blkvsc_req_list */
  62         struct list_head req_entry;
  63
  64         /* block_device_context.pending_list */
  65         struct list_head pend_entry;
  66
  67         /* This may be null if we generate a request internally */
  68         struct request *req;
  69
  70         struct block_device_context *dev;
  71
  72         /* The group this request is part of. Maybe null */
  73         struct blkvsc_request_group *group;
  74
  75         wait_queue_head_t wevent;
  76         int cond;
  77
  78         int write;
  79         sector_t sector_start;
  80         unsigned long sector_count;
  81
  82         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  83         unsigned char cmd_len;
  84         unsigned char cmnd[MAX_COMMAND_SIZE];
  85
  86         struct hv_storvsc_request request;
  87         /*
  88          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  89          * because - The extension buffer falls right here and is pointed to by
  90          * request.Extension;
  91          * Which sounds like a horrible idea, who designed this?
  92          */
  93 };
  94
  95 /* Per device structure */
  96 struct block_device_context {
  97         /* point back to our device context */
  98         struct hv_device *device_ctx;
  99         struct kmem_cache *request_pool;
 100         spinlock_t lock;
 101         struct gendisk *gd;
 102         enum blkvsc_device_type device_type;
 103         struct list_head pending_list;
 104
 105         unsigned char device_id[64];
 106         unsigned int device_id_len;
 107         int num_outstanding_reqs;
 108         int shutting_down;
 109         int media_not_present;
 110         unsigned int sector_size;
 111         sector_t capacity;
 112         unsigned int port;
 113         unsigned char path;
 114         unsigned char target;
 115         int users;
 116 };
 117
 118
 119 /* Static decl */
 120 static DEFINE_MUTEX(blkvsc_mutex);
 121 static int blkvsc_probe(struct device *dev);
 122 static int blkvsc_remove(struct device *device);
 123 static void blkvsc_shutdown(struct device *device);
 124
 125 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 126 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 127 static unsigned int blkvsc_check_events(struct gendisk *gd,
 128                                         unsigned int clearing);
 129 static int blkvsc_revalidate_disk(struct gendisk *gd);
 130 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 131 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 132                         unsigned cmd, unsigned long argument);
 133 static void blkvsc_request(struct request_queue *queue);
 134 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 135 static int blkvsc_do_request(struct block_device_context *blkdev,
 136                              struct request *req);
 137 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 138                 void (*request_completion)(struct hv_storvsc_request *));
 139 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 140 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 141 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 142 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 143 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 144 static int blkvsc_do_flush(struct block_device_context *blkdev);
 145 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 146 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 147
 148 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 149 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 150 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 151
 152 /* The one and only one */
 153 static  struct storvsc_driver_object g_blkvsc_drv;
 154
 155 static const struct block_device_operations block_ops = {
 156         .owner = THIS_MODULE,
 157         .open = blkvsc_open,
 158         .release = blkvsc_release,
 159         .check_events = blkvsc_check_events,
 160         .revalidate_disk = blkvsc_revalidate_disk,
 161         .getgeo = blkvsc_getgeo,
 162         .ioctl  = blkvsc_ioctl,
 163 };
 164
 165 /*
 166  * blkvsc_drv_init -  BlkVsc driver initialization.
 167  */
 168 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 169 {
 170         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 171         struct hv_driver *drv = &g_blkvsc_drv.base;
 172         int ret;
 173
 174         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 175
 176         drv->priv = storvsc_drv_obj;
 177
 178         /* Callback to client driver to complete the initialization */
 179         drv_init(&storvsc_drv_obj->base);
 180
 181         drv->driver.name = storvsc_drv_obj->base.name;
 182
 183         drv->driver.probe = blkvsc_probe;
 184         drv->driver.remove = blkvsc_remove;
 185         drv->driver.shutdown = blkvsc_shutdown;
 186
 187         /* The driver belongs to vmbus */
 188         ret = vmbus_child_driver_register(&drv->driver);
 189
 190         return ret;
 191 }
 192
 193 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 194 {
 195         struct device **curr = (struct device **)data;
 196         *curr = dev;
 197         return 1; /* stop iterating */
 198 }
 199
 200 static void blkvsc_drv_exit(void)
 201 {
 202         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 203         struct hv_driver *drv = &g_blkvsc_drv.base;
 204         struct device *current_dev;
 205         int ret;
 206
 207         while (1) {
 208                 current_dev = NULL;
 209
 210                 /* Get the device */
 211                 ret = driver_for_each_device(&drv->driver, NULL,
 212                                              (void *) &current_dev,
 213                                              blkvsc_drv_exit_cb);
 214
 215                 if (ret)
 216                         DPRINT_WARN(BLKVSC_DRV,
 217                                     "driver_for_each_device returned %d", ret);
 218
 219
 220                 if (current_dev == NULL)
 221                         break;
 222
 223                 /* Initiate removal from the top-down */
 224                 device_unregister(current_dev);
 225         }
 226
 227         if (storvsc_drv_obj->base.cleanup)
 228                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 229
 230         vmbus_child_driver_unregister(&drv->driver);
 231
 232         return;
 233 }
 234
 235 /*
 236  * blkvsc_probe - Add a new device for this driver
 237  */
 238 static int blkvsc_probe(struct device *device)
 239 {
 240         struct hv_driver *drv =
 241                                 drv_to_hv_drv(device->driver);
 242         struct storvsc_driver_object *storvsc_drv_obj =
 243                                 drv->priv;
 244         struct hv_device *device_obj = device_to_hv_device(device);
 245
 246         struct block_device_context *blkdev = NULL;
 247         struct storvsc_device_info device_info;
 248         int major = 0;
 249         int devnum = 0;
 250         int ret = 0;
 251         static int ide0_registered;
 252         static int ide1_registered;
 253
 254         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 255
 256         if (!storvsc_drv_obj->base.dev_add) {
 257                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 258                 ret = -1;
 259                 goto Cleanup;
 260         }
 261
 262         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 263         if (!blkdev) {
 264                 ret = -ENOMEM;
 265                 goto Cleanup;
 266         }
 267
 268         INIT_LIST_HEAD(&blkdev->pending_list);
 269
 270         /* Initialize what we can here */
 271         spin_lock_init(&blkdev->lock);
 272
 273         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 274         /*      sizeof(struct blkvsc_request)); */
 275
 276         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 277                                         sizeof(struct blkvsc_request) +
 278                                         storvsc_drv_obj->request_ext_size, 0,
 279                                         SLAB_HWCACHE_ALIGN, NULL);
 280         if (!blkdev->request_pool) {
 281                 ret = -ENOMEM;
 282                 goto Cleanup;
 283         }
 284
 285
 286         /* Call to the vsc driver to add the device */
 287         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 288         if (ret != 0) {
 289                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 290                 goto Cleanup;
 291         }
 292
 293         blkdev->device_ctx = device_obj;
 294         /* this identified the device 0 or 1 */
 295         blkdev->target = device_info.target_id;
 296         /* this identified the ide ctrl 0 or 1 */
 297         blkdev->path = device_info.path_id;
 298
 299         dev_set_drvdata(device, blkdev);
 300
 301         /* Calculate the major and device num */
 302         if (blkdev->path == 0) {
 303                 major = IDE0_MAJOR;
 304                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 305
 306                 if (!ide0_registered) {
 307                         ret = register_blkdev(major, "ide");
 308                         if (ret != 0) {
 309                                 DPRINT_ERR(BLKVSC_DRV,
 310                                            "register_blkdev() failed! ret %d",
 311                                            ret);
 312                                 goto Remove;
 313                         }
 314
 315                         ide0_registered = 1;
 316                 }
 317         } else if (blkdev->path == 1) {
 318                 major = IDE1_MAJOR;
 319                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 320
 321                 if (!ide1_registered) {
 322                         ret = register_blkdev(major, "ide");
 323                         if (ret != 0) {
 324                                 DPRINT_ERR(BLKVSC_DRV,
 325                                            "register_blkdev() failed! ret %d",
 326                                            ret);
 327                                 goto Remove;
 328                         }
 329
 330                         ide1_registered = 1;
 331                 }
 332         } else {
 333                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 334                 ret = -1;
 335                 goto Cleanup;
 336         }
 337
 338         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 339
 340         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 341         if (!blkdev->gd) {
 342                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 343                 ret = -1;
 344                 goto Cleanup;
 345         }
 346
 347         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 348
 349         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 350         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 351         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 352         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 353         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 354
 355         blkdev->gd->major = major;
 356         if (devnum == 1 || devnum == 3)
 357                 blkdev->gd->first_minor = BLKVSC_MINORS;
 358         else
 359                 blkdev->gd->first_minor = 0;
 360         blkdev->gd->fops = &block_ops;
 361         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 362         blkdev->gd->private_data = blkdev;
 363         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 364         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 365
 366         blkvsc_do_inquiry(blkdev);
 367         if (blkdev->device_type == DVD_TYPE) {
 368                 set_disk_ro(blkdev->gd, 1);
 369                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 370                 blkvsc_do_read_capacity(blkdev);
 371         } else {
 372                 blkvsc_do_read_capacity16(blkdev);
 373         }
 374
 375         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 376         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 377         /* go! */
 378         add_disk(blkdev->gd);
 379
 380         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 381                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 382                     blkdev->sector_size);
 383
 384         return ret;
 385
 386 Remove:
 387         storvsc_drv_obj->base.dev_rm(device_obj);
 388
 389 Cleanup:
 390         if (blkdev) {
 391                 if (blkdev->request_pool) {
 392                         kmem_cache_destroy(blkdev->request_pool);
 393                         blkdev->request_pool = NULL;
 394                 }
 395                 kfree(blkdev);
 396                 blkdev = NULL;
 397         }
 398
 399         return ret;
 400 }
 401
 402 static void blkvsc_shutdown(struct device *device)
 403 {
 404         struct block_device_context *blkdev = dev_get_drvdata(device);
 405         unsigned long flags;
 406
 407         if (!blkdev)
 408                 return;
 409
 410         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 411                    blkdev->users, blkdev->gd->disk_name);
 412
 413         spin_lock_irqsave(&blkdev->lock, flags);
 414
 415         blkdev->shutting_down = 1;
 416
 417         blk_stop_queue(blkdev->gd->queue);
 418
 419         spin_unlock_irqrestore(&blkdev->lock, flags);
 420
 421         while (blkdev->num_outstanding_reqs) {
 422                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 423                             blkdev->num_outstanding_reqs);
 424                 udelay(100);
 425         }
 426
 427         blkvsc_do_flush(blkdev);
 428
 429         spin_lock_irqsave(&blkdev->lock, flags);
 430
 431         blkvsc_cancel_pending_reqs(blkdev);
 432
 433         spin_unlock_irqrestore(&blkdev->lock, flags);
 434 }
 435
 436 static int blkvsc_do_flush(struct block_device_context *blkdev)
 437 {
 438         struct blkvsc_request *blkvsc_req;
 439
 440         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 441
 442         if (blkdev->device_type != HARDDISK_TYPE)
 443                 return 0;
 444
 445         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 446         if (!blkvsc_req)
 447                 return -ENOMEM;
 448
 449         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 450         init_waitqueue_head(&blkvsc_req->wevent);
 451         blkvsc_req->dev = blkdev;
 452         blkvsc_req->req = NULL;
 453         blkvsc_req->write = 0;
 454
 455         blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 456         blkvsc_req->request.data_buffer.offset = 0;
 457         blkvsc_req->request.data_buffer.len = 0;
 458
 459         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 460         blkvsc_req->cmd_len = 10;
 461
 462         /*
 463          * Set this here since the completion routine may be invoked and
 464          * completed before we return
 465          */
 466         blkvsc_req->cond = 0;
 467         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 468
 469         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 470
 471         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 472
 473         return 0;
 474 }
 475
 476 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 477 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 478 {
 479         struct blkvsc_request *blkvsc_req;
 480         struct page *page_buf;
 481         unsigned char *buf;
 482         unsigned char device_type;
 483
 484         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 485
 486         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 487         if (!blkvsc_req)
 488                 return -ENOMEM;
 489
 490         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 491         page_buf = alloc_page(GFP_KERNEL);
 492         if (!page_buf) {
 493                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 494                 return -ENOMEM;
 495         }
 496
 497         init_waitqueue_head(&blkvsc_req->wevent);
 498         blkvsc_req->dev = blkdev;
 499         blkvsc_req->req = NULL;
 500         blkvsc_req->write = 0;
 501
 502         blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
 503         blkvsc_req->request.data_buffer.offset = 0;
 504         blkvsc_req->request.data_buffer.len = 64;
 505
 506         blkvsc_req->cmnd[0] = INQUIRY;
 507         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 508         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 509         blkvsc_req->cmnd[4] = 64;
 510         blkvsc_req->cmd_len = 6;
 511
 512         /*
 513          * Set this here since the completion routine may be invoked and
 514          * completed before we return
 515          */
 516         blkvsc_req->cond = 0;
 517
 518         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 519
 520         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 521                    blkvsc_req, blkvsc_req->cond);
 522
 523         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 524
 525         buf = kmap(page_buf);
 526
 527         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 528         /* be to le */
 529         device_type = buf[0] & 0x1F;
 530
 531         if (device_type == 0x0) {
 532                 blkdev->device_type = HARDDISK_TYPE;
 533         } else if (device_type == 0x5) {
 534                 blkdev->device_type = DVD_TYPE;
 535         } else {
 536                 /* TODO: this is currently unsupported device type */
 537                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 538         }
 539
 540         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 541
 542         blkdev->device_id_len = buf[7];
 543         if (blkdev->device_id_len > 64)
 544                 blkdev->device_id_len = 64;
 545
 546         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 547         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 548          * blkdev->device_id_len); */
 549
 550         kunmap(page_buf);
 551
 552         __free_page(page_buf);
 553
 554         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 555
 556         return 0;
 557 }
 558
 559 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 560 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 561 {
 562         struct blkvsc_request *blkvsc_req;
 563         struct page *page_buf;
 564         unsigned char *buf;
 565         struct scsi_sense_hdr sense_hdr;
 566
 567         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 568
 569         blkdev->sector_size = 0;
 570         blkdev->capacity = 0;
 571         blkdev->media_not_present = 0; /* assume a disk is present */
 572
 573         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 574         if (!blkvsc_req)
 575                 return -ENOMEM;
 576
 577         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 578         page_buf = alloc_page(GFP_KERNEL);
 579         if (!page_buf) {
 580                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 581                 return -ENOMEM;
 582         }
 583
 584         init_waitqueue_head(&blkvsc_req->wevent);
 585         blkvsc_req->dev = blkdev;
 586         blkvsc_req->req = NULL;
 587         blkvsc_req->write = 0;
 588
 589         blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
 590         blkvsc_req->request.data_buffer.offset = 0;
 591         blkvsc_req->request.data_buffer.len = 8;
 592
 593         blkvsc_req->cmnd[0] = READ_CAPACITY;
 594         blkvsc_req->cmd_len = 16;
 595
 596         /*
 597          * Set this here since the completion routine may be invoked
 598          * and completed before we return
 599          */
 600         blkvsc_req->cond = 0;
 601
 602         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 603
 604         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 605                    blkvsc_req, blkvsc_req->cond);
 606
 607         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 608
 609         /* check error */
 610         if (blkvsc_req->request.status) {
 611                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 612                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 613
 614                 if (sense_hdr.asc == 0x3A) {
 615                         /* Medium not present */
 616                         blkdev->media_not_present = 1;
 617                 }
 618                 return 0;
 619         }
 620         buf = kmap(page_buf);
 621
 622         /* be to le */
 623         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 624                             (buf[2] << 8) | buf[3]) + 1;
 625         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 626                               (buf[6] << 8) | buf[7];
 627
 628         kunmap(page_buf);
 629
 630         __free_page(page_buf);
 631
 632         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 633
 634         return 0;
 635 }
 636
 637 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 638 {
 639         struct blkvsc_request *blkvsc_req;
 640         struct page *page_buf;
 641         unsigned char *buf;
 642         struct scsi_sense_hdr sense_hdr;
 643
 644         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 645
 646         blkdev->sector_size = 0;
 647         blkdev->capacity = 0;
 648         blkdev->media_not_present = 0; /* assume a disk is present */
 649
 650         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 651         if (!blkvsc_req)
 652                 return -ENOMEM;
 653
 654         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 655         page_buf = alloc_page(GFP_KERNEL);
 656         if (!page_buf) {
 657                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 658                 return -ENOMEM;
 659         }
 660
 661         init_waitqueue_head(&blkvsc_req->wevent);
 662         blkvsc_req->dev = blkdev;
 663         blkvsc_req->req = NULL;
 664         blkvsc_req->write = 0;
 665
 666         blkvsc_req->request.data_buffer.pfn_array[0] = page_to_pfn(page_buf);
 667         blkvsc_req->request.data_buffer.offset = 0;
 668         blkvsc_req->request.data_buffer.len = 12;
 669
 670         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 671         blkvsc_req->cmd_len = 16;
 672
 673         /*
 674          * Set this here since the completion routine may be invoked
 675          * and completed before we return
 676          */
 677         blkvsc_req->cond = 0;
 678
 679         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 680
 681         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 682                    blkvsc_req, blkvsc_req->cond);
 683
 684         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 685
 686         /* check error */
 687         if (blkvsc_req->request.status) {
 688                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 689                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 690                 if (sense_hdr.asc == 0x3A) {
 691                         /* Medium not present */
 692                         blkdev->media_not_present = 1;
 693                 }
 694                 return 0;
 695         }
 696         buf = kmap(page_buf);
 697
 698         /* be to le */
 699         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 700         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 701
 702 #if 0
 703         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 704                             (buf[2] << 8) | buf[3]) + 1;
 705         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 706                               (buf[6] << 8) | buf[7];
 707 #endif
 708
 709         kunmap(page_buf);
 710
 711         __free_page(page_buf);
 712
 713         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 714
 715         return 0;
 716 }
 717
 718 /*
 719  * blkvsc_remove() - Callback when our device is removed
 720  */
 721 static int blkvsc_remove(struct device *device)
 722 {
 723         struct hv_driver *drv =
 724                                 drv_to_hv_drv(device->driver);
 725         struct storvsc_driver_object *storvsc_drv_obj =
 726                                 drv->priv;
 727         struct hv_device *device_obj = device_to_hv_device(device);
 728         struct block_device_context *blkdev = dev_get_drvdata(device);
 729         unsigned long flags;
 730         int ret;
 731
 732         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 733
 734         if (!storvsc_drv_obj->base.dev_rm)
 735                 return -1;
 736
 737         /*
 738          * Call to the vsc driver to let it know that the device is being
 739          * removed
 740          */
 741         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 742         if (ret != 0) {
 743                 /* TODO: */
 744                 DPRINT_ERR(BLKVSC_DRV,
 745                            "unable to remove blkvsc device (ret %d)", ret);
 746         }
 747
 748         /* Get to a known state */
 749         spin_lock_irqsave(&blkdev->lock, flags);
 750
 751         blkdev->shutting_down = 1;
 752
 753         blk_stop_queue(blkdev->gd->queue);
 754
 755         spin_unlock_irqrestore(&blkdev->lock, flags);
 756
 757         while (blkdev->num_outstanding_reqs) {
 758                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 759                             blkdev->num_outstanding_reqs);
 760                 udelay(100);
 761         }
 762
 763         blkvsc_do_flush(blkdev);
 764
 765         spin_lock_irqsave(&blkdev->lock, flags);
 766
 767         blkvsc_cancel_pending_reqs(blkdev);
 768
 769         spin_unlock_irqrestore(&blkdev->lock, flags);
 770
 771         blk_cleanup_queue(blkdev->gd->queue);
 772
 773         del_gendisk(blkdev->gd);
 774
 775         kmem_cache_destroy(blkdev->request_pool);
 776
 777         kfree(blkdev);
 778
 779         return ret;
 780 }
 781
 782 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 783 {
 784         /* ASSERT(blkvsc_req->req); */
 785         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 786
 787         blkvsc_req->cmd_len = 16;
 788
 789         if (blkvsc_req->sector_start > 0xffffffff) {
 790                 if (rq_data_dir(blkvsc_req->req)) {
 791                         blkvsc_req->write = 1;
 792                         blkvsc_req->cmnd[0] = WRITE_16;
 793                 } else {
 794                         blkvsc_req->write = 0;
 795                         blkvsc_req->cmnd[0] = READ_16;
 796                 }
 797
 798                 blkvsc_req->cmnd[1] |=
 799                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 800
 801                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 802                                 cpu_to_be64(blkvsc_req->sector_start);
 803                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 804                                 cpu_to_be32(blkvsc_req->sector_count);
 805         } else if ((blkvsc_req->sector_count > 0xff) ||
 806                    (blkvsc_req->sector_start > 0x1fffff)) {
 807                 if (rq_data_dir(blkvsc_req->req)) {
 808                         blkvsc_req->write = 1;
 809                         blkvsc_req->cmnd[0] = WRITE_10;
 810                 } else {
 811                         blkvsc_req->write = 0;
 812                         blkvsc_req->cmnd[0] = READ_10;
 813                 }
 814
 815                 blkvsc_req->cmnd[1] |=
 816                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 817
 818                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 819                                 cpu_to_be32(blkvsc_req->sector_start);
 820                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 821                                 cpu_to_be16(blkvsc_req->sector_count);
 822         } else {
 823                 if (rq_data_dir(blkvsc_req->req)) {
 824                         blkvsc_req->write = 1;
 825                         blkvsc_req->cmnd[0] = WRITE_6;
 826                 } else {
 827                         blkvsc_req->write = 0;
 828                         blkvsc_req->cmnd[0] = READ_6;
 829                 }
 830
 831                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 832                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 833                 blkvsc_req->cmnd[1] &= 0x1f;
 834                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 835         }
 836 }
 837
 838 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 839                         void (*request_completion)(struct hv_storvsc_request *))
 840 {
 841         struct block_device_context *blkdev = blkvsc_req->dev;
 842         struct hv_device *device_ctx = blkdev->device_ctx;
 843         struct hv_driver *drv =
 844                         drv_to_hv_drv(device_ctx->device.driver);
 845         struct storvsc_driver_object *storvsc_drv_obj =
 846                         drv->priv;
 847         struct hv_storvsc_request *storvsc_req;
 848         int ret;
 849
 850         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 851                    "req %p type %s start_sector %lu count %ld offset %d "
 852                    "len %d\n", blkvsc_req,
 853                    (blkvsc_req->write) ? "WRITE" : "READ",
 854                    (unsigned long) blkvsc_req->sector_start,
 855                    blkvsc_req->sector_count,
 856                    blkvsc_req->request.data_buffer.offset,
 857                    blkvsc_req->request.data_buffer.len);
 858 #if 0
 859         for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
 860                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 861                            "req %p pfn[%d] %llx\n",
 862                            blkvsc_req, i,
 863                            blkvsc_req->request.data_buffer.pfn_array[i]);
 864         }
 865 #endif
 866
 867         storvsc_req = &blkvsc_req->request;
 868         storvsc_req->extension = (void *)((unsigned long)blkvsc_req +
 869                                           sizeof(struct blkvsc_request));
 870
 871         storvsc_req->type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 872
 873         storvsc_req->on_io_completion = request_completion;
 874         storvsc_req->context = blkvsc_req;
 875
 876         storvsc_req->host = blkdev->port;
 877         storvsc_req->bus = blkdev->path;
 878         storvsc_req->target_id = blkdev->target;
 879         storvsc_req->lun_id = 0;         /* this is not really used at all */
 880
 881         storvsc_req->cdb_len = blkvsc_req->cmd_len;
 882         storvsc_req->cdb = blkvsc_req->cmnd;
 883
 884         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 885         storvsc_req->sense_buffer_size = SCSI_SENSE_BUFFERSIZE;
 886
 887         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 888                                            &blkvsc_req->request);
 889         if (ret == 0)
 890                 blkdev->num_outstanding_reqs++;
 891
 892         return ret;
 893 }
 894
 895 /*
 896  * We break the request into 1 or more blkvsc_requests and submit
 897  * them.  If we cant submit them all, we put them on the
 898  * pending_list. The blkvsc_request() will work on the pending_list.
 899  */
 900 static int blkvsc_do_request(struct block_device_context *blkdev,
 901                              struct request *req)
 902 {
 903         struct bio *bio = NULL;
 904         struct bio_vec *bvec = NULL;
 905         struct bio_vec *prev_bvec = NULL;
 906         struct blkvsc_request *blkvsc_req = NULL;
 907         struct blkvsc_request *tmp;
 908         int databuf_idx = 0;
 909         int seg_idx = 0;
 910         sector_t start_sector;
 911         unsigned long num_sectors = 0;
 912         int ret = 0;
 913         int pending = 0;
 914         struct blkvsc_request_group *group = NULL;
 915
 916         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 917                   (unsigned long)blk_rq_pos(req));
 918
 919         /* Create a group to tie req to list of blkvsc_reqs */
 920         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 921         if (!group)
 922                 return -ENOMEM;
 923
 924         INIT_LIST_HEAD(&group->blkvsc_req_list);
 925         group->outstanding = group->status = 0;
 926
 927         start_sector = blk_rq_pos(req);
 928
 929         /* foreach bio in the request */
 930         if (req->bio) {
 931                 for (bio = req->bio; bio; bio = bio->bi_next) {
 932                         /*
 933                          * Map this bio into an existing or new storvsc request
 934                          */
 935                         bio_for_each_segment(bvec, bio, seg_idx) {
 936                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 937                                            "- req %p bio %p bvec %p seg_idx %d "
 938                                            "databuf_idx %d\n", req, bio, bvec,
 939                                            seg_idx, databuf_idx);
 940
 941                                 /* Get a new storvsc request */
 942                                 /* 1st-time */
 943                                 if ((!blkvsc_req) ||
 944                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 945                                     /* hole at the begin of page */
 946                                     || (bvec->bv_offset != 0) ||
 947                                     /* hold at the end of page */
 948                                     (prev_bvec &&
 949                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 950                                         /* submit the prev one */
 951                                         if (blkvsc_req) {
 952                                                 blkvsc_req->sector_start = start_sector;
 953                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 954
 955                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 956                                                 blkvsc_init_rw(blkvsc_req);
 957                                         }
 958
 959                                         /*
 960                                          * Create new blkvsc_req to represent
 961                                          * the current bvec
 962                                          */
 963                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 964                                         if (!blkvsc_req) {
 965                                                 /* free up everything */
 966                                                 list_for_each_entry_safe(
 967                                                         blkvsc_req, tmp,
 968                                                         &group->blkvsc_req_list,
 969                                                         req_entry) {
 970                                                         list_del(&blkvsc_req->req_entry);
 971                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 972                                                 }
 973
 974                                                 kmem_cache_free(blkdev->request_pool, group);
 975                                                 return -ENOMEM;
 976                                         }
 977
 978                                         memset(blkvsc_req, 0,
 979                                                sizeof(struct blkvsc_request));
 980
 981                                         blkvsc_req->dev = blkdev;
 982                                         blkvsc_req->req = req;
 983                                         blkvsc_req->request.data_buffer.offset
 984                                                 = bvec->bv_offset;
 985                                         blkvsc_req->request.data_buffer.len
 986                                                 = 0;
 987
 988                                         /* Add to the group */
 989                                         blkvsc_req->group = group;
 990                                         blkvsc_req->group->outstanding++;
 991                                         list_add_tail(&blkvsc_req->req_entry,
 992                                                 &blkvsc_req->group->blkvsc_req_list);
 993
 994                                         start_sector += num_sectors;
 995                                         num_sectors = 0;
 996                                         databuf_idx = 0;
 997                                 }
 998
 999                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1000                                 blkvsc_req->request.data_buffer.
1001                                         pfn_array[databuf_idx]
1002                                                 = page_to_pfn(bvec->bv_page);
1003                                 blkvsc_req->request.data_buffer.len
1004                                         += bvec->bv_len;
1005
1006                                 prev_bvec = bvec;
1007
1008                                 databuf_idx++;
1009                                 num_sectors += bvec->bv_len >> 9;
1010
1011                         } /* bio_for_each_segment */
1012
1013                 } /* rq_for_each_bio */
1014         }
1015
1016         /* Handle the last one */
1017         if (blkvsc_req) {
1018                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1019                            blkdev, req, blkvsc_req->group,
1020                            blkvsc_req->group->outstanding);
1021
1022                 blkvsc_req->sector_start = start_sector;
1023                 sector_div(blkvsc_req->sector_start,
1024                            (blkdev->sector_size >> 9));
1025
1026                 blkvsc_req->sector_count = num_sectors /
1027                                            (blkdev->sector_size >> 9);
1028
1029                 blkvsc_init_rw(blkvsc_req);
1030         }
1031
1032         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1033                 if (pending) {
1034                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1035                                    "pending_list - blkvsc_req %p start_sect %lu"
1036                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1037                                    (unsigned long)blkvsc_req->sector_start,
1038                                    blkvsc_req->sector_count,
1039                                    (unsigned long)start_sector,
1040                                    (unsigned long)num_sectors);
1041
1042                         list_add_tail(&blkvsc_req->pend_entry,
1043                                       &blkdev->pending_list);
1044                 } else {
1045                         ret = blkvsc_submit_request(blkvsc_req,
1046                                                     blkvsc_request_completion);
1047                         if (ret == -1) {
1048                                 pending = 1;
1049                                 list_add_tail(&blkvsc_req->pend_entry,
1050                                               &blkdev->pending_list);
1051                         }
1052
1053                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1054                                    "start_sect %lu sect_count %ld (%lu %ld) "
1055                                    "ret %d\n", blkvsc_req,
1056                                    (unsigned long)blkvsc_req->sector_start,
1057                                    blkvsc_req->sector_count,
1058                                    (unsigned long)start_sector,
1059                                    num_sectors, ret);
1060                 }
1061         }
1062
1063         return pending;
1064 }
1065
1066 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1067 {
1068         struct blkvsc_request *blkvsc_req =
1069                         (struct blkvsc_request *)request->context;
1070         struct block_device_context *blkdev =
1071                         (struct block_device_context *)blkvsc_req->dev;
1072         struct scsi_sense_hdr sense_hdr;
1073
1074         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1075                    blkvsc_req);
1076
1077         blkdev->num_outstanding_reqs--;
1078
1079         if (blkvsc_req->request.status)
1080                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1081                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1082                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1083
1084         blkvsc_req->cond = 1;
1085         wake_up_interruptible(&blkvsc_req->wevent);
1086 }
1087
1088 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1089 {
1090         struct blkvsc_request *blkvsc_req =
1091                         (struct blkvsc_request *)request->context;
1092         struct block_device_context *blkdev =
1093                         (struct block_device_context *)blkvsc_req->dev;
1094         unsigned long flags;
1095         struct blkvsc_request *comp_req, *tmp;
1096
1097         /* ASSERT(blkvsc_req->group); */
1098
1099         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1100                    "sect_start %lu sect_count %ld len %d group outstd %d "
1101                    "total outstd %d\n",
1102                    blkdev, blkvsc_req, blkvsc_req->group,
1103                    (blkvsc_req->write) ? "WRITE" : "READ",
1104                    (unsigned long)blkvsc_req->sector_start,
1105                    blkvsc_req->sector_count,
1106                    blkvsc_req->request.data_buffer.len,
1107                    blkvsc_req->group->outstanding,
1108                    blkdev->num_outstanding_reqs);
1109
1110         spin_lock_irqsave(&blkdev->lock, flags);
1111
1112         blkdev->num_outstanding_reqs--;
1113         blkvsc_req->group->outstanding--;
1114
1115         /*
1116          * Only start processing when all the blkvsc_reqs are
1117          * completed. This guarantees no out-of-order blkvsc_req
1118          * completion when calling end_that_request_first()
1119          */
1120         if (blkvsc_req->group->outstanding == 0) {
1121                 list_for_each_entry_safe(comp_req, tmp,
1122                                          &blkvsc_req->group->blkvsc_req_list,
1123                                          req_entry) {
1124                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1125                                    "sect_start %lu sect_count %ld\n",
1126                                    comp_req,
1127                                    (unsigned long)comp_req->sector_start,
1128                                    comp_req->sector_count);
1129
1130                         list_del(&comp_req->req_entry);
1131
1132                         if (!__blk_end_request(comp_req->req,
1133                                 (!comp_req->request.status ? 0 : -EIO),
1134                                 comp_req->sector_count * blkdev->sector_size)) {
1135                                 /*
1136                                  * All the sectors have been xferred ie the
1137                                  * request is done
1138                                  */
1139                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1140                                            comp_req->req);
1141                                 kmem_cache_free(blkdev->request_pool,
1142                                                 comp_req->group);
1143                         }
1144
1145                         kmem_cache_free(blkdev->request_pool, comp_req);
1146                 }
1147
1148                 if (!blkdev->shutting_down) {
1149                         blkvsc_do_pending_reqs(blkdev);
1150                         blk_start_queue(blkdev->gd->queue);
1151                         blkvsc_request(blkdev->gd->queue);
1152                 }
1153         }
1154
1155         spin_unlock_irqrestore(&blkdev->lock, flags);
1156 }
1157
1158 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1159 {
1160         struct blkvsc_request *pend_req, *tmp;
1161         struct blkvsc_request *comp_req, *tmp2;
1162
1163         int ret = 0;
1164
1165         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1166
1167         /* Flush the pending list first */
1168         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1169                                  pend_entry) {
1170                 /*
1171                  * The pend_req could be part of a partially completed
1172                  * request. If so, complete those req first until we
1173                  * hit the pend_req
1174                  */
1175                 list_for_each_entry_safe(comp_req, tmp2,
1176                                          &pend_req->group->blkvsc_req_list,
1177                                          req_entry) {
1178                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1179                                    "sect_start %lu sect_count %ld\n",
1180                                    comp_req,
1181                                    (unsigned long) comp_req->sector_start,
1182                                    comp_req->sector_count);
1183
1184                         if (comp_req == pend_req)
1185                                 break;
1186
1187                         list_del(&comp_req->req_entry);
1188
1189                         if (comp_req->req) {
1190                                 ret = __blk_end_request(comp_req->req,
1191                                         (!comp_req->request.status ? 0 : -EIO),
1192                                         comp_req->sector_count *
1193                                         blkdev->sector_size);
1194
1195                                 /* FIXME: shouldn't this do more than return? */
1196                                 if (ret)
1197                                         goto out;
1198                         }
1199
1200                         kmem_cache_free(blkdev->request_pool, comp_req);
1201                 }
1202
1203                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1204                            pend_req);
1205
1206                 list_del(&pend_req->pend_entry);
1207
1208                 list_del(&pend_req->req_entry);
1209
1210                 if (comp_req->req) {
1211                         if (!__blk_end_request(pend_req->req, -EIO,
1212                                                pend_req->sector_count *
1213                                                blkdev->sector_size)) {
1214                                 /*
1215                                  * All the sectors have been xferred ie the
1216                                  * request is done
1217                                  */
1218                                 DPRINT_DBG(BLKVSC_DRV,
1219                                            "blkvsc_cancel_pending_reqs() - "
1220                                            "req %p COMPLETED\n", pend_req->req);
1221                                 kmem_cache_free(blkdev->request_pool,
1222                                                 pend_req->group);
1223                         }
1224                 }
1225
1226                 kmem_cache_free(blkdev->request_pool, pend_req);
1227         }
1228
1229 out:
1230         return ret;
1231 }
1232
1233 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1234 {
1235         struct blkvsc_request *pend_req, *tmp;
1236         int ret = 0;
1237
1238         /* Flush the pending list first */
1239         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1240                                  pend_entry) {
1241                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1242                            pend_req);
1243
1244                 ret = blkvsc_submit_request(pend_req,
1245                                             blkvsc_request_completion);
1246                 if (ret != 0)
1247                         break;
1248                 else
1249                         list_del(&pend_req->pend_entry);
1250         }
1251
1252         return ret;
1253 }
1254
1255 static void blkvsc_request(struct request_queue *queue)
1256 {
1257         struct block_device_context *blkdev = NULL;
1258         struct request *req;
1259         int ret = 0;
1260
1261         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1262         while ((req = blk_peek_request(queue)) != NULL) {
1263                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1264
1265                 blkdev = req->rq_disk->private_data;
1266                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1267                     blkdev->media_not_present) {
1268                         __blk_end_request_cur(req, 0);
1269                         continue;
1270                 }
1271
1272                 ret = blkvsc_do_pending_reqs(blkdev);
1273
1274                 if (ret != 0) {
1275                         DPRINT_DBG(BLKVSC_DRV,
1276                                    "- stop queue - pending_list not empty\n");
1277                         blk_stop_queue(queue);
1278                         break;
1279                 }
1280
1281                 blk_start_request(req);
1282
1283                 ret = blkvsc_do_request(blkdev, req);
1284                 if (ret > 0) {
1285                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1286                         blk_stop_queue(queue);
1287                         break;
1288                 } else if (ret < 0) {
1289                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1290                         blk_requeue_request(queue, req);
1291                         blk_stop_queue(queue);
1292                         break;
1293                 }
1294         }
1295 }
1296
1297 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1298 {
1299         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1300
1301         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1302                    blkdev->gd->disk_name);
1303
1304         mutex_lock(&blkvsc_mutex);
1305         spin_lock(&blkdev->lock);
1306
1307         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1308                 spin_unlock(&blkdev->lock);
1309                 check_disk_change(bdev);
1310                 spin_lock(&blkdev->lock);
1311         }
1312
1313         blkdev->users++;
1314
1315         spin_unlock(&blkdev->lock);
1316         mutex_unlock(&blkvsc_mutex);
1317         return 0;
1318 }
1319
1320 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1321 {
1322         struct block_device_context *blkdev = disk->private_data;
1323
1324         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1325                    blkdev->gd->disk_name);
1326
1327         mutex_lock(&blkvsc_mutex);
1328         spin_lock(&blkdev->lock);
1329         if (blkdev->users == 1) {
1330                 spin_unlock(&blkdev->lock);
1331                 blkvsc_do_flush(blkdev);
1332                 spin_lock(&blkdev->lock);
1333         }
1334
1335         blkdev->users--;
1336
1337         spin_unlock(&blkdev->lock);
1338         mutex_unlock(&blkvsc_mutex);
1339         return 0;
1340 }
1341
1342 static unsigned int blkvsc_check_events(struct gendisk *gd,
1343                                         unsigned int clearing)
1344 {
1345         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1346         return DISK_EVENT_MEDIA_CHANGE;
1347 }
1348
1349 static int blkvsc_revalidate_disk(struct gendisk *gd)
1350 {
1351         struct block_device_context *blkdev = gd->private_data;
1352
1353         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1354
1355         if (blkdev->device_type == DVD_TYPE) {
1356                 blkvsc_do_read_capacity(blkdev);
1357                 set_capacity(blkdev->gd, blkdev->capacity *
1358                             (blkdev->sector_size/512));
1359                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1360         }
1361         return 0;
1362 }
1363
1364 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1365 {
1366         sector_t total_sectors = get_capacity(bd->bd_disk);
1367         sector_t cylinder_times_heads = 0;
1368         sector_t temp = 0;
1369
1370         int sectors_per_track = 0;
1371         int heads = 0;
1372         int cylinders = 0;
1373         int rem = 0;
1374
1375         if (total_sectors > (65535 * 16 * 255))
1376                 total_sectors = (65535 * 16 * 255);
1377
1378         if (total_sectors >= (65535 * 16 * 63)) {
1379                 sectors_per_track = 255;
1380                 heads = 16;
1381
1382                 cylinder_times_heads = total_sectors;
1383                 /* sector_div stores the quotient in cylinder_times_heads */
1384                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1385         } else {
1386                 sectors_per_track = 17;
1387
1388                 cylinder_times_heads = total_sectors;
1389                 /* sector_div stores the quotient in cylinder_times_heads */
1390                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1391
1392                 temp = cylinder_times_heads + 1023;
1393                 /* sector_div stores the quotient in temp */
1394                 rem = sector_div(temp, 1024);
1395
1396                 heads = temp;
1397
1398                 if (heads < 4)
1399                         heads = 4;
1400
1401
1402                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1403                         sectors_per_track = 31;
1404                         heads = 16;
1405
1406                         cylinder_times_heads = total_sectors;
1407                         /*
1408                          * sector_div stores the quotient in
1409                          * cylinder_times_heads
1410                          */
1411                         rem = sector_div(cylinder_times_heads,
1412                                          sectors_per_track);
1413                 }
1414
1415                 if (cylinder_times_heads >= (heads * 1024)) {
1416                         sectors_per_track = 63;
1417                         heads = 16;
1418
1419                         cylinder_times_heads = total_sectors;
1420                         /*
1421                          * sector_div stores the quotient in
1422                          * cylinder_times_heads
1423                          */
1424                         rem = sector_div(cylinder_times_heads,
1425                                          sectors_per_track);
1426                 }
1427         }
1428
1429         temp = cylinder_times_heads;
1430         /* sector_div stores the quotient in temp */
1431         rem = sector_div(temp, heads);
1432         cylinders = temp;
1433
1434         hg->heads = heads;
1435         hg->sectors = sectors_per_track;
1436         hg->cylinders = cylinders;
1437
1438         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1439                     sectors_per_track);
1440
1441     return 0;
1442 }
1443
1444 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1445                         unsigned cmd, unsigned long argument)
1446 {
1447 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1448         int ret;
1449
1450         switch (cmd) {
1451         /*
1452          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1453          * than just a GUID. Commented it out for now.
1454          */
1455 #if 0
1456         case HDIO_GET_IDENTITY:
1457                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1458                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1459                                  blkdev->device_id_len))
1460                         ret = -EFAULT;
1461                 break;
1462 #endif
1463         default:
1464                 ret = -EINVAL;
1465                 break;
1466         }
1467
1468         return ret;
1469 }
1470
1471 static int __init blkvsc_init(void)
1472 {
1473         int ret;
1474
1475         BUILD_BUG_ON(sizeof(sector_t) != 8);
1476
1477         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1478
1479         ret = blkvsc_drv_init(blk_vsc_initialize);
1480
1481         return ret;
1482 }
1483
1484 static void __exit blkvsc_exit(void)
1485 {
1486         blkvsc_drv_exit();
1487 }
1488
1489 MODULE_LICENSE("GPL");
1490 MODULE_VERSION(HV_DRV_VERSION);
1491 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1492 module_init(blkvsc_init);
1493 module_exit(blkvsc_exit);