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/slab.h>
  29 #include <scsi/scsi.h>
  30 #include <scsi/scsi_cmnd.h>
  31 #include <scsi/scsi_eh.h>
  32 #include <scsi/scsi_dbg.h>
  33 #include "osd.h"
  34 #include "logging.h"
  35 #include "version_info.h"
  36 #include "vmbus.h"
  37 #include "storvsc_api.h"
  38
  39
  40 #define BLKVSC_MINORS   64
  41
  42 enum blkvsc_device_type {
  43         UNKNOWN_DEV_TYPE,
  44         HARDDISK_TYPE,
  45         DVD_TYPE,
  46 };
  47
  48 /*
  49  * This request ties the struct request and struct
  50  * blkvsc_request/hv_storvsc_request together A struct request may be
  51  * represented by 1 or more struct blkvsc_request
  52  */
  53 struct blkvsc_request_group {
  54         int outstanding;
  55         int status;
  56         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  57 };
  58
  59 struct blkvsc_request {
  60         /* blkvsc_request_group.blkvsc_req_list */
  61         struct list_head req_entry;
  62
  63         /* block_device_context.pending_list */
  64         struct list_head pend_entry;
  65
  66         /* This may be null if we generate a request internally */
  67         struct request *req;
  68
  69         struct block_device_context *dev;
  70
  71         /* The group this request is part of. Maybe null */
  72         struct blkvsc_request_group *group;
  73
  74         wait_queue_head_t wevent;
  75         int cond;
  76
  77         int write;
  78         sector_t sector_start;
  79         unsigned long sector_count;
  80
  81         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  82         unsigned char cmd_len;
  83         unsigned char cmnd[MAX_COMMAND_SIZE];
  84
  85         struct hv_storvsc_request request;
  86         /*
  87          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  88          * because - The extension buffer falls right here and is pointed to by
  89          * request.Extension;
  90          * Which sounds like a horrible idea, who designed this?
  91          */
  92 };
  93
  94 /* Per device structure */
  95 struct block_device_context {
  96         /* point back to our device context */
  97         struct vm_device *device_ctx;
  98         struct kmem_cache *request_pool;
  99         spinlock_t lock;
 100         struct gendisk *gd;
 101         enum blkvsc_device_type device_type;
 102         struct list_head pending_list;
 103
 104         unsigned char device_id[64];
 105         unsigned int device_id_len;
 106         int num_outstanding_reqs;
 107         int shutting_down;
 108         int media_not_present;
 109         unsigned int sector_size;
 110         sector_t capacity;
 111         unsigned int port;
 112         unsigned char path;
 113         unsigned char target;
 114         int users;
 115 };
 116
 117 /* Per driver */
 118 struct blkvsc_driver_context {
 119         /* !! These must be the first 2 fields !! */
 120         /* FIXME this is a bug! */
 121         struct driver_context drv_ctx;
 122         struct storvsc_driver_object drv_obj;
 123 };
 124
 125 /* Static decl */
 126 static int blkvsc_probe(struct device *dev);
 127 static int blkvsc_remove(struct device *device);
 128 static void blkvsc_shutdown(struct device *device);
 129
 130 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 131 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 132 static int blkvsc_media_changed(struct gendisk *gd);
 133 static int blkvsc_revalidate_disk(struct gendisk *gd);
 134 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 135 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 136                         unsigned cmd, unsigned long argument);
 137 static void blkvsc_request(struct request_queue *queue);
 138 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 139 static int blkvsc_do_request(struct block_device_context *blkdev,
 140                              struct request *req);
 141 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 142                 void (*request_completion)(struct hv_storvsc_request *));
 143 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 144 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 145 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 146 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 147 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 148 static int blkvsc_do_flush(struct block_device_context *blkdev);
 149 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 150 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 151
 152 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 153 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 154 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 155
 156 /* The one and only one */
 157 static struct blkvsc_driver_context g_blkvsc_drv;
 158
 159 static const struct block_device_operations block_ops = {
 160         .owner = THIS_MODULE,
 161         .open = blkvsc_open,
 162         .release = blkvsc_release,
 163         .media_changed = blkvsc_media_changed,
 164         .revalidate_disk = blkvsc_revalidate_disk,
 165         .getgeo = blkvsc_getgeo,
 166         .ioctl  = blkvsc_ioctl,
 167 };
 168
 169 /*
 170  * blkvsc_drv_init -  BlkVsc driver initialization.
 171  */
 172 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 173 {
 174         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 175         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 176         int ret;
 177
 178         vmbus_get_interface(&storvsc_drv_obj->Base.VmbusChannelInterface);
 179
 180         storvsc_drv_obj->RingBufferSize = blkvsc_ringbuffer_size;
 181
 182         /* Callback to client driver to complete the initialization */
 183         drv_init(&storvsc_drv_obj->Base);
 184
 185         drv_ctx->driver.name = storvsc_drv_obj->Base.name;
 186         memcpy(&drv_ctx->class_id, &storvsc_drv_obj->Base.deviceType,
 187                sizeof(struct hv_guid));
 188
 189         drv_ctx->probe = blkvsc_probe;
 190         drv_ctx->remove = blkvsc_remove;
 191         drv_ctx->shutdown = blkvsc_shutdown;
 192
 193         /* The driver belongs to vmbus */
 194         ret = vmbus_child_driver_register(drv_ctx);
 195
 196         DPRINT_EXIT(BLKVSC_DRV);
 197
 198         return ret;
 199 }
 200
 201 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 202 {
 203         struct device **curr = (struct device **)data;
 204         *curr = dev;
 205         return 1; /* stop iterating */
 206 }
 207
 208 static void blkvsc_drv_exit(void)
 209 {
 210         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 211         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 212         struct device *current_dev;
 213         int ret;
 214
 215         while (1) {
 216                 current_dev = NULL;
 217
 218                 /* Get the device */
 219                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
 220                                              (void *) &current_dev,
 221                                              blkvsc_drv_exit_cb);
 222
 223                 if (ret)
 224                         DPRINT_WARN(BLKVSC_DRV,
 225                                     "driver_for_each_device returned %d", ret);
 226
 227
 228                 if (current_dev == NULL)
 229                         break;
 230
 231                 /* Initiate removal from the top-down */
 232                 device_unregister(current_dev);
 233         }
 234
 235         if (storvsc_drv_obj->Base.OnCleanup)
 236                 storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
 237
 238         vmbus_child_driver_unregister(drv_ctx);
 239
 240         DPRINT_EXIT(BLKVSC_DRV);
 241
 242         return;
 243 }
 244
 245 /*
 246  * blkvsc_probe - Add a new device for this driver
 247  */
 248 static int blkvsc_probe(struct device *device)
 249 {
 250         struct driver_context *driver_ctx =
 251                                 driver_to_driver_context(device->driver);
 252         struct blkvsc_driver_context *blkvsc_drv_ctx =
 253                                 (struct blkvsc_driver_context *)driver_ctx;
 254         struct storvsc_driver_object *storvsc_drv_obj =
 255                                 &blkvsc_drv_ctx->drv_obj;
 256         struct vm_device *device_ctx = device_to_vm_device(device);
 257         struct hv_device *device_obj = &device_ctx->device_obj;
 258
 259         struct block_device_context *blkdev = NULL;
 260         struct storvsc_device_info device_info;
 261         int major = 0;
 262         int devnum = 0;
 263         int ret = 0;
 264         static int ide0_registered;
 265         static int ide1_registered;
 266
 267         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 268
 269         if (!storvsc_drv_obj->Base.OnDeviceAdd) {
 270                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 271                 ret = -1;
 272                 goto Cleanup;
 273         }
 274
 275         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 276         if (!blkdev) {
 277                 ret = -ENOMEM;
 278                 goto Cleanup;
 279         }
 280
 281         INIT_LIST_HEAD(&blkdev->pending_list);
 282
 283         /* Initialize what we can here */
 284         spin_lock_init(&blkdev->lock);
 285
 286         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 287         /*      sizeof(struct blkvsc_request)); */
 288
 289         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
 290                                         sizeof(struct blkvsc_request) +
 291                                         storvsc_drv_obj->RequestExtSize, 0,
 292                                         SLAB_HWCACHE_ALIGN, NULL);
 293         if (!blkdev->request_pool) {
 294                 ret = -ENOMEM;
 295                 goto Cleanup;
 296         }
 297
 298
 299         /* Call to the vsc driver to add the device */
 300         ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
 301         if (ret != 0) {
 302                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 303                 goto Cleanup;
 304         }
 305
 306         blkdev->device_ctx = device_ctx;
 307         /* this identified the device 0 or 1 */
 308         blkdev->target = device_info.TargetId;
 309         /* this identified the ide ctrl 0 or 1 */
 310         blkdev->path = device_info.PathId;
 311
 312         dev_set_drvdata(device, blkdev);
 313
 314         /* Calculate the major and device num */
 315         if (blkdev->path == 0) {
 316                 major = IDE0_MAJOR;
 317                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 318
 319                 if (!ide0_registered) {
 320                         ret = register_blkdev(major, "ide");
 321                         if (ret != 0) {
 322                                 DPRINT_ERR(BLKVSC_DRV,
 323                                            "register_blkdev() failed! ret %d",
 324                                            ret);
 325                                 goto Remove;
 326                         }
 327
 328                         ide0_registered = 1;
 329                 }
 330         } else if (blkdev->path == 1) {
 331                 major = IDE1_MAJOR;
 332                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 333
 334                 if (!ide1_registered) {
 335                         ret = register_blkdev(major, "ide");
 336                         if (ret != 0) {
 337                                 DPRINT_ERR(BLKVSC_DRV,
 338                                            "register_blkdev() failed! ret %d",
 339                                            ret);
 340                                 goto Remove;
 341                         }
 342
 343                         ide1_registered = 1;
 344                 }
 345         } else {
 346                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 347                 ret = -1;
 348                 goto Cleanup;
 349         }
 350
 351         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 352
 353         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 354         if (!blkdev->gd) {
 355                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 356                 ret = -1;
 357                 goto Cleanup;
 358         }
 359
 360         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 361
 362         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 363         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 364         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 365         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 366         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 367
 368         blkdev->gd->major = major;
 369         if (devnum == 1 || devnum == 3)
 370                 blkdev->gd->first_minor = BLKVSC_MINORS;
 371         else
 372                 blkdev->gd->first_minor = 0;
 373         blkdev->gd->fops = &block_ops;
 374         blkdev->gd->private_data = blkdev;
 375         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 376
 377         blkvsc_do_inquiry(blkdev);
 378         if (blkdev->device_type == DVD_TYPE) {
 379                 set_disk_ro(blkdev->gd, 1);
 380                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 381                 blkvsc_do_read_capacity(blkdev);
 382         } else {
 383                 blkvsc_do_read_capacity16(blkdev);
 384         }
 385
 386         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 387         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 388         /* go! */
 389         add_disk(blkdev->gd);
 390
 391         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 392                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 393                     blkdev->sector_size);
 394
 395         return ret;
 396
 397 Remove:
 398         storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 399
 400 Cleanup:
 401         if (blkdev) {
 402                 if (blkdev->request_pool) {
 403                         kmem_cache_destroy(blkdev->request_pool);
 404                         blkdev->request_pool = NULL;
 405                 }
 406                 kfree(blkdev);
 407                 blkdev = NULL;
 408         }
 409
 410         DPRINT_EXIT(BLKVSC_DRV);
 411
 412         return ret;
 413 }
 414
 415 static void blkvsc_shutdown(struct device *device)
 416 {
 417         struct block_device_context *blkdev = dev_get_drvdata(device);
 418         unsigned long flags;
 419
 420         if (!blkdev)
 421                 return;
 422
 423         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 424                    blkdev->users, blkdev->gd->disk_name);
 425
 426         spin_lock_irqsave(&blkdev->lock, flags);
 427
 428         blkdev->shutting_down = 1;
 429
 430         blk_stop_queue(blkdev->gd->queue);
 431
 432         spin_unlock_irqrestore(&blkdev->lock, flags);
 433
 434         while (blkdev->num_outstanding_reqs) {
 435                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 436                             blkdev->num_outstanding_reqs);
 437                 udelay(100);
 438         }
 439
 440         blkvsc_do_flush(blkdev);
 441
 442         spin_lock_irqsave(&blkdev->lock, flags);
 443
 444         blkvsc_cancel_pending_reqs(blkdev);
 445
 446         spin_unlock_irqrestore(&blkdev->lock, flags);
 447 }
 448
 449 static int blkvsc_do_flush(struct block_device_context *blkdev)
 450 {
 451         struct blkvsc_request *blkvsc_req;
 452
 453         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 454
 455         if (blkdev->device_type != HARDDISK_TYPE)
 456                 return 0;
 457
 458         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 459         if (!blkvsc_req)
 460                 return -ENOMEM;
 461
 462         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 463         init_waitqueue_head(&blkvsc_req->wevent);
 464         blkvsc_req->dev = blkdev;
 465         blkvsc_req->req = NULL;
 466         blkvsc_req->write = 0;
 467
 468         blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
 469         blkvsc_req->request.DataBuffer.Offset = 0;
 470         blkvsc_req->request.DataBuffer.Length = 0;
 471
 472         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 473         blkvsc_req->cmd_len = 10;
 474
 475         /*
 476          * Set this here since the completion routine may be invoked and
 477          * completed before we return
 478          */
 479         blkvsc_req->cond = 0;
 480         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 481
 482         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 483
 484         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 485
 486         return 0;
 487 }
 488
 489 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 490 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 491 {
 492         struct blkvsc_request *blkvsc_req;
 493         struct page *page_buf;
 494         unsigned char *buf;
 495         unsigned char device_type;
 496
 497         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 498
 499         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 500         if (!blkvsc_req)
 501                 return -ENOMEM;
 502
 503         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 504         page_buf = alloc_page(GFP_KERNEL);
 505         if (!page_buf) {
 506                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 507                 return -ENOMEM;
 508         }
 509
 510         init_waitqueue_head(&blkvsc_req->wevent);
 511         blkvsc_req->dev = blkdev;
 512         blkvsc_req->req = NULL;
 513         blkvsc_req->write = 0;
 514
 515         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 516         blkvsc_req->request.DataBuffer.Offset = 0;
 517         blkvsc_req->request.DataBuffer.Length = 64;
 518
 519         blkvsc_req->cmnd[0] = INQUIRY;
 520         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 521         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 522         blkvsc_req->cmnd[4] = 64;
 523         blkvsc_req->cmd_len = 6;
 524
 525         /*
 526          * Set this here since the completion routine may be invoked and
 527          * completed before we return
 528          */
 529         blkvsc_req->cond = 0;
 530
 531         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 532
 533         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 534                    blkvsc_req, blkvsc_req->cond);
 535
 536         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 537
 538         buf = kmap(page_buf);
 539
 540         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 541         /* be to le */
 542         device_type = buf[0] & 0x1F;
 543
 544         if (device_type == 0x0) {
 545                 blkdev->device_type = HARDDISK_TYPE;
 546         } else if (device_type == 0x5) {
 547                 blkdev->device_type = DVD_TYPE;
 548         } else {
 549                 /* TODO: this is currently unsupported device type */
 550                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 551         }
 552
 553         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 554
 555         blkdev->device_id_len = buf[7];
 556         if (blkdev->device_id_len > 64)
 557                 blkdev->device_id_len = 64;
 558
 559         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 560         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 561          * blkdev->device_id_len); */
 562
 563         kunmap(page_buf);
 564
 565         __free_page(page_buf);
 566
 567         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 568
 569         return 0;
 570 }
 571
 572 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 573 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 574 {
 575         struct blkvsc_request *blkvsc_req;
 576         struct page *page_buf;
 577         unsigned char *buf;
 578         struct scsi_sense_hdr sense_hdr;
 579
 580         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 581
 582         blkdev->sector_size = 0;
 583         blkdev->capacity = 0;
 584         blkdev->media_not_present = 0; /* assume a disk is present */
 585
 586         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 587         if (!blkvsc_req)
 588                 return -ENOMEM;
 589
 590         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 591         page_buf = alloc_page(GFP_KERNEL);
 592         if (!page_buf) {
 593                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 594                 return -ENOMEM;
 595         }
 596
 597         init_waitqueue_head(&blkvsc_req->wevent);
 598         blkvsc_req->dev = blkdev;
 599         blkvsc_req->req = NULL;
 600         blkvsc_req->write = 0;
 601
 602         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 603         blkvsc_req->request.DataBuffer.Offset = 0;
 604         blkvsc_req->request.DataBuffer.Length = 8;
 605
 606         blkvsc_req->cmnd[0] = READ_CAPACITY;
 607         blkvsc_req->cmd_len = 16;
 608
 609         /*
 610          * Set this here since the completion routine may be invoked
 611          * and completed before we return
 612          */
 613         blkvsc_req->cond = 0;
 614
 615         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 616
 617         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 618                    blkvsc_req, blkvsc_req->cond);
 619
 620         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 621
 622         /* check error */
 623         if (blkvsc_req->request.Status) {
 624                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 625                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 626
 627                 if (sense_hdr.asc == 0x3A) {
 628                         /* Medium not present */
 629                         blkdev->media_not_present = 1;
 630                 }
 631                 return 0;
 632         }
 633         buf = kmap(page_buf);
 634
 635         /* be to le */
 636         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 637                             (buf[2] << 8) | buf[3]) + 1;
 638         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 639                               (buf[6] << 8) | buf[7];
 640
 641         kunmap(page_buf);
 642
 643         __free_page(page_buf);
 644
 645         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 646
 647         return 0;
 648 }
 649
 650 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 651 {
 652         struct blkvsc_request *blkvsc_req;
 653         struct page *page_buf;
 654         unsigned char *buf;
 655         struct scsi_sense_hdr sense_hdr;
 656
 657         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 658
 659         blkdev->sector_size = 0;
 660         blkdev->capacity = 0;
 661         blkdev->media_not_present = 0; /* assume a disk is present */
 662
 663         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 664         if (!blkvsc_req)
 665                 return -ENOMEM;
 666
 667         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 668         page_buf = alloc_page(GFP_KERNEL);
 669         if (!page_buf) {
 670                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 671                 return -ENOMEM;
 672         }
 673
 674         init_waitqueue_head(&blkvsc_req->wevent);
 675         blkvsc_req->dev = blkdev;
 676         blkvsc_req->req = NULL;
 677         blkvsc_req->write = 0;
 678
 679         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 680         blkvsc_req->request.DataBuffer.Offset = 0;
 681         blkvsc_req->request.DataBuffer.Length = 12;
 682
 683         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 684         blkvsc_req->cmd_len = 16;
 685
 686         /*
 687          * Set this here since the completion routine may be invoked
 688          * and completed before we return
 689          */
 690         blkvsc_req->cond = 0;
 691
 692         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 693
 694         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 695                    blkvsc_req, blkvsc_req->cond);
 696
 697         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 698
 699         /* check error */
 700         if (blkvsc_req->request.Status) {
 701                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 702                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 703                 if (sense_hdr.asc == 0x3A) {
 704                         /* Medium not present */
 705                         blkdev->media_not_present = 1;
 706                 }
 707                 return 0;
 708         }
 709         buf = kmap(page_buf);
 710
 711         /* be to le */
 712         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 713         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 714
 715 #if 0
 716         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 717                             (buf[2] << 8) | buf[3]) + 1;
 718         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 719                               (buf[6] << 8) | buf[7];
 720 #endif
 721
 722         kunmap(page_buf);
 723
 724         __free_page(page_buf);
 725
 726         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 727
 728         return 0;
 729 }
 730
 731 /*
 732  * blkvsc_remove() - Callback when our device is removed
 733  */
 734 static int blkvsc_remove(struct device *device)
 735 {
 736         struct driver_context *driver_ctx =
 737                                 driver_to_driver_context(device->driver);
 738         struct blkvsc_driver_context *blkvsc_drv_ctx =
 739                                 (struct blkvsc_driver_context *)driver_ctx;
 740         struct storvsc_driver_object *storvsc_drv_obj =
 741                                 &blkvsc_drv_ctx->drv_obj;
 742         struct vm_device *device_ctx = device_to_vm_device(device);
 743         struct hv_device *device_obj = &device_ctx->device_obj;
 744         struct block_device_context *blkdev = dev_get_drvdata(device);
 745         unsigned long flags;
 746         int ret;
 747
 748         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 749
 750         if (!storvsc_drv_obj->Base.OnDeviceRemove) {
 751                 DPRINT_EXIT(BLKVSC_DRV);
 752                 return -1;
 753         }
 754
 755         /*
 756          * Call to the vsc driver to let it know that the device is being
 757          * removed
 758          */
 759         ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 760         if (ret != 0) {
 761                 /* TODO: */
 762                 DPRINT_ERR(BLKVSC_DRV,
 763                            "unable to remove blkvsc device (ret %d)", ret);
 764         }
 765
 766         /* Get to a known state */
 767         spin_lock_irqsave(&blkdev->lock, flags);
 768
 769         blkdev->shutting_down = 1;
 770
 771         blk_stop_queue(blkdev->gd->queue);
 772
 773         spin_unlock_irqrestore(&blkdev->lock, flags);
 774
 775         while (blkdev->num_outstanding_reqs) {
 776                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 777                             blkdev->num_outstanding_reqs);
 778                 udelay(100);
 779         }
 780
 781         blkvsc_do_flush(blkdev);
 782
 783         spin_lock_irqsave(&blkdev->lock, flags);
 784
 785         blkvsc_cancel_pending_reqs(blkdev);
 786
 787         spin_unlock_irqrestore(&blkdev->lock, flags);
 788
 789         blk_cleanup_queue(blkdev->gd->queue);
 790
 791         del_gendisk(blkdev->gd);
 792
 793         kmem_cache_destroy(blkdev->request_pool);
 794
 795         kfree(blkdev);
 796
 797         DPRINT_EXIT(BLKVSC_DRV);
 798
 799         return ret;
 800 }
 801
 802 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 803 {
 804         /* ASSERT(blkvsc_req->req); */
 805         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 806
 807         blkvsc_req->cmd_len = 16;
 808
 809         if (blkvsc_req->sector_start > 0xffffffff) {
 810                 if (rq_data_dir(blkvsc_req->req)) {
 811                         blkvsc_req->write = 1;
 812                         blkvsc_req->cmnd[0] = WRITE_16;
 813                 } else {
 814                         blkvsc_req->write = 0;
 815                         blkvsc_req->cmnd[0] = READ_16;
 816                 }
 817
 818                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 819
 820                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 821                                 cpu_to_be64(blkvsc_req->sector_start);
 822                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 823                                 cpu_to_be32(blkvsc_req->sector_count);
 824         } else if ((blkvsc_req->sector_count > 0xff) ||
 825                    (blkvsc_req->sector_start > 0x1fffff)) {
 826                 if (rq_data_dir(blkvsc_req->req)) {
 827                         blkvsc_req->write = 1;
 828                         blkvsc_req->cmnd[0] = WRITE_10;
 829                 } else {
 830                         blkvsc_req->write = 0;
 831                         blkvsc_req->cmnd[0] = READ_10;
 832                 }
 833
 834                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 835
 836                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 837                                 cpu_to_be32(blkvsc_req->sector_start);
 838                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 839                                 cpu_to_be16(blkvsc_req->sector_count);
 840         } else {
 841                 if (rq_data_dir(blkvsc_req->req)) {
 842                         blkvsc_req->write = 1;
 843                         blkvsc_req->cmnd[0] = WRITE_6;
 844                 } else {
 845                         blkvsc_req->write = 0;
 846                         blkvsc_req->cmnd[0] = READ_6;
 847                 }
 848
 849                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 850                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 851                 blkvsc_req->cmnd[1] &= 0x1f;
 852                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 853         }
 854 }
 855
 856 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 857                         void (*request_completion)(struct hv_storvsc_request *))
 858 {
 859         struct block_device_context *blkdev = blkvsc_req->dev;
 860         struct vm_device *device_ctx = blkdev->device_ctx;
 861         struct driver_context *driver_ctx =
 862                         driver_to_driver_context(device_ctx->device.driver);
 863         struct blkvsc_driver_context *blkvsc_drv_ctx =
 864                         (struct blkvsc_driver_context *)driver_ctx;
 865         struct storvsc_driver_object *storvsc_drv_obj =
 866                         &blkvsc_drv_ctx->drv_obj;
 867         struct hv_storvsc_request *storvsc_req;
 868         int ret;
 869
 870         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 871                    "req %p type %s start_sector %lu count %ld offset %d "
 872                    "len %d\n", blkvsc_req,
 873                    (blkvsc_req->write) ? "WRITE" : "READ",
 874                    (unsigned long) blkvsc_req->sector_start,
 875                    blkvsc_req->sector_count,
 876                    blkvsc_req->request.DataBuffer.Offset,
 877                    blkvsc_req->request.DataBuffer.Length);
 878 #if 0
 879         for (i = 0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++) {
 880                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 881                            "req %p pfn[%d] %llx\n",
 882                            blkvsc_req, i,
 883                            blkvsc_req->request.DataBuffer.PfnArray[i]);
 884         }
 885 #endif
 886
 887         storvsc_req = &blkvsc_req->request;
 888         storvsc_req->Extension = (void *)((unsigned long)blkvsc_req +
 889                                           sizeof(struct blkvsc_request));
 890
 891         storvsc_req->Type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 892
 893         storvsc_req->OnIOCompletion = request_completion;
 894         storvsc_req->Context = blkvsc_req;
 895
 896         storvsc_req->Host = blkdev->port;
 897         storvsc_req->Bus = blkdev->path;
 898         storvsc_req->TargetId = blkdev->target;
 899         storvsc_req->LunId = 0;  /* this is not really used at all */
 900
 901         storvsc_req->CdbLen = blkvsc_req->cmd_len;
 902         storvsc_req->Cdb = blkvsc_req->cmnd;
 903
 904         storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
 905         storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
 906
 907         ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj,
 908                                            &blkvsc_req->request);
 909         if (ret == 0)
 910                 blkdev->num_outstanding_reqs++;
 911
 912         return ret;
 913 }
 914
 915 /*
 916  * We break the request into 1 or more blkvsc_requests and submit
 917  * them.  If we cant submit them all, we put them on the
 918  * pending_list. The blkvsc_request() will work on the pending_list.
 919  */
 920 static int blkvsc_do_request(struct block_device_context *blkdev,
 921                              struct request *req)
 922 {
 923         struct bio *bio = NULL;
 924         struct bio_vec *bvec = NULL;
 925         struct bio_vec *prev_bvec = NULL;
 926         struct blkvsc_request *blkvsc_req = NULL;
 927         struct blkvsc_request *tmp;
 928         int databuf_idx = 0;
 929         int seg_idx = 0;
 930         sector_t start_sector;
 931         unsigned long num_sectors = 0;
 932         int ret = 0;
 933         int pending = 0;
 934         struct blkvsc_request_group *group = NULL;
 935
 936         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 937                   (unsigned long)blk_rq_pos(req));
 938
 939         /* Create a group to tie req to list of blkvsc_reqs */
 940         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 941         if (!group)
 942                 return -ENOMEM;
 943
 944         INIT_LIST_HEAD(&group->blkvsc_req_list);
 945         group->outstanding = group->status = 0;
 946
 947         start_sector = blk_rq_pos(req);
 948
 949         /* foreach bio in the request */
 950         if (req->bio) {
 951                 for (bio = req->bio; bio; bio = bio->bi_next) {
 952                         /*
 953                          * Map this bio into an existing or new storvsc request
 954                          */
 955                         bio_for_each_segment(bvec, bio, seg_idx) {
 956                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 957                                            "- req %p bio %p bvec %p seg_idx %d "
 958                                            "databuf_idx %d\n", req, bio, bvec,
 959                                            seg_idx, databuf_idx);
 960
 961                                 /* Get a new storvsc request */
 962                                 /* 1st-time */
 963                                 if ((!blkvsc_req) ||
 964                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 965                                     /* hole at the begin of page */
 966                                     || (bvec->bv_offset != 0) ||
 967                                     /* hold at the end of page */
 968                                     (prev_bvec &&
 969                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 970                                         /* submit the prev one */
 971                                         if (blkvsc_req) {
 972                                                 blkvsc_req->sector_start = start_sector;
 973                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 974
 975                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 976                                                 blkvsc_init_rw(blkvsc_req);
 977                                         }
 978
 979                                         /*
 980                                          * Create new blkvsc_req to represent
 981                                          * the current bvec
 982                                          */
 983                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 984                                         if (!blkvsc_req) {
 985                                                 /* free up everything */
 986                                                 list_for_each_entry_safe(
 987                                                         blkvsc_req, tmp,
 988                                                         &group->blkvsc_req_list,
 989                                                         req_entry) {
 990                                                         list_del(&blkvsc_req->req_entry);
 991                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 992                                                 }
 993
 994                                                 kmem_cache_free(blkdev->request_pool, group);
 995                                                 return -ENOMEM;
 996                                         }
 997
 998                                         memset(blkvsc_req, 0,
 999                                                sizeof(struct blkvsc_request));
1000
1001                                         blkvsc_req->dev = blkdev;
1002                                         blkvsc_req->req = req;
1003                                         blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
1004                                         blkvsc_req->request.DataBuffer.Length = 0;
1005
1006                                         /* Add to the group */
1007                                         blkvsc_req->group = group;
1008                                         blkvsc_req->group->outstanding++;
1009                                         list_add_tail(&blkvsc_req->req_entry,
1010                                                 &blkvsc_req->group->blkvsc_req_list);
1011
1012                                         start_sector += num_sectors;
1013                                         num_sectors = 0;
1014                                         databuf_idx = 0;
1015                                 }
1016
1017                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1018                                 blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
1019                                 blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
1020
1021                                 prev_bvec = bvec;
1022
1023                                 databuf_idx++;
1024                                 num_sectors += bvec->bv_len >> 9;
1025
1026                         } /* bio_for_each_segment */
1027
1028                 } /* rq_for_each_bio */
1029         }
1030
1031         /* Handle the last one */
1032         if (blkvsc_req) {
1033                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1034                            blkdev, req, blkvsc_req->group,
1035                            blkvsc_req->group->outstanding);
1036
1037                 blkvsc_req->sector_start = start_sector;
1038                 sector_div(blkvsc_req->sector_start,
1039                            (blkdev->sector_size >> 9));
1040
1041                 blkvsc_req->sector_count = num_sectors /
1042                                            (blkdev->sector_size >> 9);
1043
1044                 blkvsc_init_rw(blkvsc_req);
1045         }
1046
1047         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1048                 if (pending) {
1049                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1050                                    "pending_list - blkvsc_req %p start_sect %lu"
1051                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1052                                    (unsigned long)blkvsc_req->sector_start,
1053                                    blkvsc_req->sector_count,
1054                                    (unsigned long)start_sector,
1055                                    (unsigned long)num_sectors);
1056
1057                         list_add_tail(&blkvsc_req->pend_entry,
1058                                       &blkdev->pending_list);
1059                 } else {
1060                         ret = blkvsc_submit_request(blkvsc_req,
1061                                                     blkvsc_request_completion);
1062                         if (ret == -1) {
1063                                 pending = 1;
1064                                 list_add_tail(&blkvsc_req->pend_entry,
1065                                               &blkdev->pending_list);
1066                         }
1067
1068                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1069                                    "start_sect %lu sect_count %ld (%lu %ld) "
1070                                    "ret %d\n", blkvsc_req,
1071                                    (unsigned long)blkvsc_req->sector_start,
1072                                    blkvsc_req->sector_count,
1073                                    (unsigned long)start_sector,
1074                                    num_sectors, ret);
1075                 }
1076         }
1077
1078         return pending;
1079 }
1080
1081 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1082 {
1083         struct blkvsc_request *blkvsc_req =
1084                         (struct blkvsc_request *)request->Context;
1085         struct block_device_context *blkdev =
1086                         (struct block_device_context *)blkvsc_req->dev;
1087         struct scsi_sense_hdr sense_hdr;
1088
1089         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1090                    blkvsc_req);
1091
1092         blkdev->num_outstanding_reqs--;
1093
1094         if (blkvsc_req->request.Status)
1095                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1096                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1097                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1098
1099         blkvsc_req->cond = 1;
1100         wake_up_interruptible(&blkvsc_req->wevent);
1101 }
1102
1103 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1104 {
1105         struct blkvsc_request *blkvsc_req =
1106                         (struct blkvsc_request *)request->Context;
1107         struct block_device_context *blkdev =
1108                         (struct block_device_context *)blkvsc_req->dev;
1109         unsigned long flags;
1110         struct blkvsc_request *comp_req, *tmp;
1111
1112         /* ASSERT(blkvsc_req->group); */
1113
1114         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1115                    "sect_start %lu sect_count %ld len %d group outstd %d "
1116                    "total outstd %d\n",
1117                    blkdev, blkvsc_req, blkvsc_req->group,
1118                    (blkvsc_req->write) ? "WRITE" : "READ",
1119                    (unsigned long)blkvsc_req->sector_start,
1120                    blkvsc_req->sector_count,
1121                    blkvsc_req->request.DataBuffer.Length,
1122                    blkvsc_req->group->outstanding,
1123                    blkdev->num_outstanding_reqs);
1124
1125         spin_lock_irqsave(&blkdev->lock, flags);
1126
1127         blkdev->num_outstanding_reqs--;
1128         blkvsc_req->group->outstanding--;
1129
1130         /*
1131          * Only start processing when all the blkvsc_reqs are
1132          * completed. This guarantees no out-of-order blkvsc_req
1133          * completion when calling end_that_request_first()
1134          */
1135         if (blkvsc_req->group->outstanding == 0) {
1136                 list_for_each_entry_safe(comp_req, tmp,
1137                                          &blkvsc_req->group->blkvsc_req_list,
1138                                          req_entry) {
1139                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1140                                    "sect_start %lu sect_count %ld\n",
1141                                    comp_req,
1142                                    (unsigned long)comp_req->sector_start,
1143                                    comp_req->sector_count);
1144
1145                         list_del(&comp_req->req_entry);
1146
1147                         if (!__blk_end_request(comp_req->req,
1148                                 (!comp_req->request.Status ? 0 : -EIO),
1149                                 comp_req->sector_count * blkdev->sector_size)) {
1150                                 /*
1151                                  * All the sectors have been xferred ie the
1152                                  * request is done
1153                                  */
1154                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1155                                            comp_req->req);
1156                                 kmem_cache_free(blkdev->request_pool,
1157                                                 comp_req->group);
1158                         }
1159
1160                         kmem_cache_free(blkdev->request_pool, comp_req);
1161                 }
1162
1163                 if (!blkdev->shutting_down) {
1164                         blkvsc_do_pending_reqs(blkdev);
1165                         blk_start_queue(blkdev->gd->queue);
1166                         blkvsc_request(blkdev->gd->queue);
1167                 }
1168         }
1169
1170         spin_unlock_irqrestore(&blkdev->lock, flags);
1171 }
1172
1173 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1174 {
1175         struct blkvsc_request *pend_req, *tmp;
1176         struct blkvsc_request *comp_req, *tmp2;
1177
1178         int ret = 0;
1179
1180         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1181
1182         /* Flush the pending list first */
1183         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1184                                  pend_entry) {
1185                 /*
1186                  * The pend_req could be part of a partially completed
1187                  * request. If so, complete those req first until we
1188                  * hit the pend_req
1189                  */
1190                 list_for_each_entry_safe(comp_req, tmp2,
1191                                          &pend_req->group->blkvsc_req_list,
1192                                          req_entry) {
1193                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1194                                    "sect_start %lu sect_count %ld\n",
1195                                    comp_req,
1196                                    (unsigned long) comp_req->sector_start,
1197                                    comp_req->sector_count);
1198
1199                         if (comp_req == pend_req)
1200                                 break;
1201
1202                         list_del(&comp_req->req_entry);
1203
1204                         if (comp_req->req) {
1205                                 ret = __blk_end_request(comp_req->req,
1206                                         (!comp_req->request.Status ? 0 : -EIO),
1207                                         comp_req->sector_count *
1208                                         blkdev->sector_size);
1209
1210                                 /* FIXME: shouldn't this do more than return? */
1211                                 if (ret)
1212                                         goto out;
1213                         }
1214
1215                         kmem_cache_free(blkdev->request_pool, comp_req);
1216                 }
1217
1218                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1219                            pend_req);
1220
1221                 list_del(&pend_req->pend_entry);
1222
1223                 list_del(&pend_req->req_entry);
1224
1225                 if (comp_req->req) {
1226                         if (!__blk_end_request(pend_req->req, -EIO,
1227                                                pend_req->sector_count *
1228                                                blkdev->sector_size)) {
1229                                 /*
1230                                  * All the sectors have been xferred ie the
1231                                  * request is done
1232                                  */
1233                                 DPRINT_DBG(BLKVSC_DRV,
1234                                            "blkvsc_cancel_pending_reqs() - "
1235                                            "req %p COMPLETED\n", pend_req->req);
1236                                 kmem_cache_free(blkdev->request_pool,
1237                                                 pend_req->group);
1238                         }
1239                 }
1240
1241                 kmem_cache_free(blkdev->request_pool, pend_req);
1242         }
1243
1244 out:
1245         return ret;
1246 }
1247
1248 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1249 {
1250         struct blkvsc_request *pend_req, *tmp;
1251         int ret = 0;
1252
1253         /* Flush the pending list first */
1254         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1255                                  pend_entry) {
1256                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1257                            pend_req);
1258
1259                 ret = blkvsc_submit_request(pend_req,
1260                                             blkvsc_request_completion);
1261                 if (ret != 0)
1262                         break;
1263                 else
1264                         list_del(&pend_req->pend_entry);
1265         }
1266
1267         return ret;
1268 }
1269
1270 static void blkvsc_request(struct request_queue *queue)
1271 {
1272         struct block_device_context *blkdev = NULL;
1273         struct request *req;
1274         int ret = 0;
1275
1276         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1277         while ((req = blk_peek_request(queue)) != NULL) {
1278                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1279
1280                 blkdev = req->rq_disk->private_data;
1281                 if (blkdev->shutting_down || !blk_fs_request(req) ||
1282                     blkdev->media_not_present) {
1283                         __blk_end_request_cur(req, 0);
1284                         continue;
1285                 }
1286
1287                 ret = blkvsc_do_pending_reqs(blkdev);
1288
1289                 if (ret != 0) {
1290                         DPRINT_DBG(BLKVSC_DRV,
1291                                    "- stop queue - pending_list not empty\n");
1292                         blk_stop_queue(queue);
1293                         break;
1294                 }
1295
1296                 blk_start_request(req);
1297
1298                 ret = blkvsc_do_request(blkdev, req);
1299                 if (ret > 0) {
1300                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1301                         blk_stop_queue(queue);
1302                         break;
1303                 } else if (ret < 0) {
1304                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1305                         blk_requeue_request(queue, req);
1306                         blk_stop_queue(queue);
1307                         break;
1308                 }
1309         }
1310 }
1311
1312 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1313 {
1314         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1315
1316         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1317                    blkdev->gd->disk_name);
1318
1319         spin_lock(&blkdev->lock);
1320
1321         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1322                 spin_unlock(&blkdev->lock);
1323                 check_disk_change(bdev);
1324                 spin_lock(&blkdev->lock);
1325         }
1326
1327         blkdev->users++;
1328
1329         spin_unlock(&blkdev->lock);
1330         return 0;
1331 }
1332
1333 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1334 {
1335         struct block_device_context *blkdev = disk->private_data;
1336
1337         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1338                    blkdev->gd->disk_name);
1339
1340         spin_lock(&blkdev->lock);
1341         if (blkdev->users == 1) {
1342                 spin_unlock(&blkdev->lock);
1343                 blkvsc_do_flush(blkdev);
1344                 spin_lock(&blkdev->lock);
1345         }
1346
1347         blkdev->users--;
1348
1349         spin_unlock(&blkdev->lock);
1350         return 0;
1351 }
1352
1353 static int blkvsc_media_changed(struct gendisk *gd)
1354 {
1355         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1356         return 1;
1357 }
1358
1359 static int blkvsc_revalidate_disk(struct gendisk *gd)
1360 {
1361         struct block_device_context *blkdev = gd->private_data;
1362
1363         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1364
1365         if (blkdev->device_type == DVD_TYPE) {
1366                 blkvsc_do_read_capacity(blkdev);
1367                 set_capacity(blkdev->gd, blkdev->capacity *
1368                             (blkdev->sector_size/512));
1369                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1370         }
1371         return 0;
1372 }
1373
1374 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1375 {
1376         sector_t total_sectors = get_capacity(bd->bd_disk);
1377         sector_t cylinder_times_heads = 0;
1378         sector_t temp = 0;
1379
1380         int sectors_per_track = 0;
1381         int heads = 0;
1382         int cylinders = 0;
1383         int rem = 0;
1384
1385         if (total_sectors > (65535 * 16 * 255))
1386                 total_sectors = (65535 * 16 * 255);
1387
1388         if (total_sectors >= (65535 * 16 * 63)) {
1389                 sectors_per_track = 255;
1390                 heads = 16;
1391
1392                 cylinder_times_heads = total_sectors;
1393                 /* sector_div stores the quotient in cylinder_times_heads */
1394                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1395         } else {
1396                 sectors_per_track = 17;
1397
1398                 cylinder_times_heads = total_sectors;
1399                 /* sector_div stores the quotient in cylinder_times_heads */
1400                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1401
1402                 temp = cylinder_times_heads + 1023;
1403                 /* sector_div stores the quotient in temp */
1404                 rem = sector_div(temp, 1024);
1405
1406                 heads = temp;
1407
1408                 if (heads < 4)
1409                         heads = 4;
1410
1411
1412                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1413                         sectors_per_track = 31;
1414                         heads = 16;
1415
1416                         cylinder_times_heads = total_sectors;
1417                         /*
1418                          * sector_div stores the quotient in
1419                          * cylinder_times_heads
1420                          */
1421                         rem = sector_div(cylinder_times_heads,
1422                                          sectors_per_track);
1423                 }
1424
1425                 if (cylinder_times_heads >= (heads * 1024)) {
1426                         sectors_per_track = 63;
1427                         heads = 16;
1428
1429                         cylinder_times_heads = total_sectors;
1430                         /*
1431                          * sector_div stores the quotient in
1432                          * cylinder_times_heads
1433                          */
1434                         rem = sector_div(cylinder_times_heads,
1435                                          sectors_per_track);
1436                 }
1437         }
1438
1439         temp = cylinder_times_heads;
1440         /* sector_div stores the quotient in temp */
1441         rem = sector_div(temp, heads);
1442         cylinders = temp;
1443
1444         hg->heads = heads;
1445         hg->sectors = sectors_per_track;
1446         hg->cylinders = cylinders;
1447
1448         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1449                     sectors_per_track);
1450
1451     return 0;
1452 }
1453
1454 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1455                         unsigned cmd, unsigned long argument)
1456 {
1457 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1458         int ret;
1459
1460         switch (cmd) {
1461         /*
1462          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1463          * than just a GUID. Commented it out for now.
1464          */
1465 #if 0
1466         case HDIO_GET_IDENTITY:
1467                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1468                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1469                                  blkdev->device_id_len))
1470                         ret = -EFAULT;
1471                 break;
1472 #endif
1473         default:
1474                 ret = -EINVAL;
1475                 break;
1476         }
1477
1478         return ret;
1479 }
1480
1481 static int __init blkvsc_init(void)
1482 {
1483         int ret;
1484
1485         BUILD_BUG_ON(sizeof(sector_t) != 8);
1486
1487         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1488
1489         ret = blkvsc_drv_init(BlkVscInitialize);
1490
1491         DPRINT_EXIT(BLKVSC_DRV);
1492
1493         return ret;
1494 }
1495
1496 static void __exit blkvsc_exit(void)
1497 {
1498         blkvsc_drv_exit();
1499 }
1500
1501 MODULE_LICENSE("GPL");
1502 MODULE_VERSION(HV_DRV_VERSION);
1503 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1504 module_init(blkvsc_init);
1505 module_exit(blkvsc_exit);