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 <scsi/scsi.h>
  29 #include <scsi/scsi_cmnd.h>
  30 #include <scsi/scsi_eh.h>
  31 #include <scsi/scsi_dbg.h>
  32 #include "osd.h"
  33 #include "logging.h"
  34 #include "VersionInfo.h"
  35 #include "vmbus.h"
  36 #include "StorVscApi.h"
  37
  38
  39 #define BLKVSC_MINORS   64
  40
  41 enum blkvsc_device_type {
  42         UNKNOWN_DEV_TYPE,
  43         HARDDISK_TYPE,
  44         DVD_TYPE,
  45 };
  46
  47 /*
  48  * This request ties the struct request and struct
  49  * blkvsc_request/hv_storvsc_request together A struct request may be
  50  * represented by 1 or more struct blkvsc_request
  51  */
  52 struct blkvsc_request_group {
  53         int outstanding;
  54         int status;
  55         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  56 };
  57
  58 struct blkvsc_request {
  59         /* blkvsc_request_group.blkvsc_req_list */
  60         struct list_head req_entry;
  61
  62         /* block_device_context.pending_list */
  63         struct list_head pend_entry;
  64
  65         /* This may be null if we generate a request internally */
  66         struct request *req;
  67
  68         struct block_device_context *dev;
  69
  70         /* The group this request is part of. Maybe null */
  71         struct blkvsc_request_group *group;
  72
  73         wait_queue_head_t wevent;
  74         int cond;
  75
  76         int write;
  77         sector_t sector_start;
  78         unsigned long sector_count;
  79
  80         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  81         unsigned char cmd_len;
  82         unsigned char cmnd[MAX_COMMAND_SIZE];
  83
  84         struct hv_storvsc_request request;
  85         /*
  86          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  87          * because - The extension buffer falls right here and is pointed to by
  88          * request.Extension;
  89          * Which sounds like a horrible idea, who designed this?
  90          */
  91 };
  92
  93 /* Per device structure */
  94 struct block_device_context {
  95         /* point back to our device context */
  96         struct vm_device *device_ctx;
  97         struct kmem_cache *request_pool;
  98         spinlock_t lock;
  99         struct gendisk *gd;
 100         enum blkvsc_device_type device_type;
 101         struct list_head pending_list;
 102
 103         unsigned char device_id[64];
 104         unsigned int device_id_len;
 105         int num_outstanding_reqs;
 106         int shutting_down;
 107         int media_not_present;
 108         unsigned int sector_size;
 109         sector_t capacity;
 110         unsigned int port;
 111         unsigned char path;
 112         unsigned char target;
 113         int users;
 114 };
 115
 116 /* Per driver */
 117 struct blkvsc_driver_context {
 118         /* !! These must be the first 2 fields !! */
 119         /* FIXME this is a bug! */
 120         struct driver_context drv_ctx;
 121         struct storvsc_driver_object drv_obj;
 122 };
 123
 124 /* Static decl */
 125 static int blkvsc_probe(struct device *dev);
 126 static int blkvsc_remove(struct device *device);
 127 static void blkvsc_shutdown(struct device *device);
 128
 129 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 130 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 131 static int blkvsc_media_changed(struct gendisk *gd);
 132 static int blkvsc_revalidate_disk(struct gendisk *gd);
 133 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 134 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 135                         unsigned cmd, unsigned long argument);
 136 static void blkvsc_request(struct request_queue *queue);
 137 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 138 static int blkvsc_do_request(struct block_device_context *blkdev,
 139                              struct request *req);
 140 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 141                 void (*request_completion)(struct hv_storvsc_request *));
 142 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 143 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 144 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 145 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 146 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 147 static int blkvsc_do_flush(struct block_device_context *blkdev);
 148 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 149 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 150
 151
 152 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 153
 154 /* The one and only one */
 155 static struct blkvsc_driver_context g_blkvsc_drv;
 156
 157 static struct block_device_operations block_ops = {
 158         .owner = THIS_MODULE,
 159         .open = blkvsc_open,
 160         .release = blkvsc_release,
 161         .media_changed = blkvsc_media_changed,
 162         .revalidate_disk = blkvsc_revalidate_disk,
 163         .getgeo = blkvsc_getgeo,
 164         .ioctl  = blkvsc_ioctl,
 165 };
 166
 167 /**
 168  * blkvsc_drv_init -  BlkVsc driver initialization.
 169  */
 170 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 171 {
 172         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 173         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 174         int ret;
 175
 176         DPRINT_ENTER(BLKVSC_DRV);
 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         DPRINT_ENTER(BLKVSC_DRV);
 216
 217         while (1) {
 218                 current_dev = NULL;
 219
 220                 /* Get the device */
 221                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
 222                                              (void *) &current_dev,
 223                                              blkvsc_drv_exit_cb);
 224
 225                 if (ret)
 226                         DPRINT_WARN(BLKVSC_DRV,
 227                                     "driver_for_each_device returned %d", ret);
 228
 229
 230                 if (current_dev == NULL)
 231                         break;
 232
 233                 /* Initiate removal from the top-down */
 234                 device_unregister(current_dev);
 235         }
 236
 237         if (storvsc_drv_obj->Base.OnCleanup)
 238                 storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
 239
 240         vmbus_child_driver_unregister(drv_ctx);
 241
 242         DPRINT_EXIT(BLKVSC_DRV);
 243
 244         return;
 245 }
 246
 247 /**
 248  * blkvsc_probe - Add a new device for this driver
 249  */
 250 static int blkvsc_probe(struct device *device)
 251 {
 252         struct driver_context *driver_ctx =
 253                                 driver_to_driver_context(device->driver);
 254         struct blkvsc_driver_context *blkvsc_drv_ctx =
 255                                 (struct blkvsc_driver_context *)driver_ctx;
 256         struct storvsc_driver_object *storvsc_drv_obj =
 257                                 &blkvsc_drv_ctx->drv_obj;
 258         struct vm_device *device_ctx = device_to_vm_device(device);
 259         struct hv_device *device_obj = &device_ctx->device_obj;
 260
 261         struct block_device_context *blkdev = NULL;
 262         struct storvsc_device_info device_info;
 263         int major = 0;
 264         int devnum = 0;
 265         int ret = 0;
 266         static int ide0_registered;
 267         static int ide1_registered;
 268
 269         DPRINT_ENTER(BLKVSC_DRV);
 270
 271         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 272
 273         if (!storvsc_drv_obj->Base.OnDeviceAdd) {
 274                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 275                 ret = -1;
 276                 goto Cleanup;
 277         }
 278
 279         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 280         if (!blkdev) {
 281                 ret = -ENOMEM;
 282                 goto Cleanup;
 283         }
 284
 285         INIT_LIST_HEAD(&blkdev->pending_list);
 286
 287         /* Initialize what we can here */
 288         spin_lock_init(&blkdev->lock);
 289
 290         ASSERT(sizeof(struct blkvsc_request_group) <=
 291                 sizeof(struct blkvsc_request));
 292
 293         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
 294                                         sizeof(struct blkvsc_request) +
 295                                         storvsc_drv_obj->RequestExtSize, 0,
 296                                         SLAB_HWCACHE_ALIGN, NULL);
 297         if (!blkdev->request_pool) {
 298                 ret = -ENOMEM;
 299                 goto Cleanup;
 300         }
 301
 302
 303         /* Call to the vsc driver to add the device */
 304         ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
 305         if (ret != 0) {
 306                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 307                 goto Cleanup;
 308         }
 309
 310         blkdev->device_ctx = device_ctx;
 311         /* this identified the device 0 or 1 */
 312         blkdev->target = device_info.TargetId;
 313         /* this identified the ide ctrl 0 or 1 */
 314         blkdev->path = device_info.PathId;
 315
 316         dev_set_drvdata(device, blkdev);
 317
 318         /* Calculate the major and device num */
 319         if (blkdev->path == 0) {
 320                 major = IDE0_MAJOR;
 321                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 322
 323                 if (!ide0_registered) {
 324                         ret = register_blkdev(major, "ide");
 325                         if (ret != 0) {
 326                                 DPRINT_ERR(BLKVSC_DRV,
 327                                            "register_blkdev() failed! ret %d",
 328                                            ret);
 329                                 goto Remove;
 330                         }
 331
 332                         ide0_registered = 1;
 333                 }
 334         } else if (blkdev->path == 1) {
 335                 major = IDE1_MAJOR;
 336                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 337
 338                 if (!ide1_registered) {
 339                         ret = register_blkdev(major, "ide");
 340                         if (ret != 0) {
 341                                 DPRINT_ERR(BLKVSC_DRV,
 342                                            "register_blkdev() failed! ret %d",
 343                                            ret);
 344                                 goto Remove;
 345                         }
 346
 347                         ide1_registered = 1;
 348                 }
 349         } else {
 350                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 351                 ret = -1;
 352                 goto Cleanup;
 353         }
 354
 355         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 356
 357         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 358         if (!blkdev->gd) {
 359                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 360                 ret = -1;
 361                 goto Cleanup;
 362         }
 363
 364         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 365
 366         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 367         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 368         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 369         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 370         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 371
 372         blkdev->gd->major = major;
 373         if (devnum == 1 || devnum == 3)
 374                 blkdev->gd->first_minor = BLKVSC_MINORS;
 375         else
 376                 blkdev->gd->first_minor = 0;
 377         blkdev->gd->fops = &block_ops;
 378         blkdev->gd->private_data = blkdev;
 379         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 380
 381         blkvsc_do_inquiry(blkdev);
 382         if (blkdev->device_type == DVD_TYPE) {
 383                 set_disk_ro(blkdev->gd, 1);
 384                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 385                 blkvsc_do_read_capacity(blkdev);
 386         } else {
 387                 blkvsc_do_read_capacity16(blkdev);
 388         }
 389
 390         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 391         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 392         /* go! */
 393         add_disk(blkdev->gd);
 394
 395         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 396                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 397                     blkdev->sector_size);
 398
 399         return ret;
 400
 401 Remove:
 402         storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 403
 404 Cleanup:
 405         if (blkdev) {
 406                 if (blkdev->request_pool) {
 407                         kmem_cache_destroy(blkdev->request_pool);
 408                         blkdev->request_pool = NULL;
 409                 }
 410                 kfree(blkdev);
 411                 blkdev = NULL;
 412         }
 413
 414         DPRINT_EXIT(BLKVSC_DRV);
 415
 416         return ret;
 417 }
 418
 419 static void blkvsc_shutdown(struct device *device)
 420 {
 421         struct block_device_context *blkdev = dev_get_drvdata(device);
 422         unsigned long flags;
 423
 424         if (!blkdev)
 425                 return;
 426
 427         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 428                    blkdev->users, blkdev->gd->disk_name);
 429
 430         spin_lock_irqsave(&blkdev->lock, flags);
 431
 432         blkdev->shutting_down = 1;
 433
 434         blk_stop_queue(blkdev->gd->queue);
 435
 436         spin_unlock_irqrestore(&blkdev->lock, flags);
 437
 438         while (blkdev->num_outstanding_reqs) {
 439                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 440                             blkdev->num_outstanding_reqs);
 441                 udelay(100);
 442         }
 443
 444         blkvsc_do_flush(blkdev);
 445
 446         spin_lock_irqsave(&blkdev->lock, flags);
 447
 448         blkvsc_cancel_pending_reqs(blkdev);
 449
 450         spin_unlock_irqrestore(&blkdev->lock, flags);
 451 }
 452
 453 static int blkvsc_do_flush(struct block_device_context *blkdev)
 454 {
 455         struct blkvsc_request *blkvsc_req;
 456
 457         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 458
 459         if (blkdev->device_type != HARDDISK_TYPE)
 460                 return 0;
 461
 462         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 463         if (!blkvsc_req)
 464                 return -ENOMEM;
 465
 466         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 467         init_waitqueue_head(&blkvsc_req->wevent);
 468         blkvsc_req->dev = blkdev;
 469         blkvsc_req->req = NULL;
 470         blkvsc_req->write = 0;
 471
 472         blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
 473         blkvsc_req->request.DataBuffer.Offset = 0;
 474         blkvsc_req->request.DataBuffer.Length = 0;
 475
 476         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 477         blkvsc_req->cmd_len = 10;
 478
 479         /*
 480          * Set this here since the completion routine may be invoked and
 481          * completed before we return
 482          */
 483         blkvsc_req->cond = 0;
 484         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 485
 486         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 487
 488         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 489
 490         return 0;
 491 }
 492
 493 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 494 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 495 {
 496         struct blkvsc_request *blkvsc_req;
 497         struct page *page_buf;
 498         unsigned char *buf;
 499         unsigned char device_type;
 500
 501         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 502
 503         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 504         if (!blkvsc_req)
 505                 return -ENOMEM;
 506
 507         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 508         page_buf = alloc_page(GFP_KERNEL);
 509         if (!page_buf) {
 510                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 511                 return -ENOMEM;
 512         }
 513
 514         init_waitqueue_head(&blkvsc_req->wevent);
 515         blkvsc_req->dev = blkdev;
 516         blkvsc_req->req = NULL;
 517         blkvsc_req->write = 0;
 518
 519         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 520         blkvsc_req->request.DataBuffer.Offset = 0;
 521         blkvsc_req->request.DataBuffer.Length = 64;
 522
 523         blkvsc_req->cmnd[0] = INQUIRY;
 524         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 525         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 526         blkvsc_req->cmnd[4] = 64;
 527         blkvsc_req->cmd_len = 6;
 528
 529         /*
 530          * Set this here since the completion routine may be invoked and
 531          * completed before we return
 532          */
 533         blkvsc_req->cond = 0;
 534
 535         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 536
 537         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 538                    blkvsc_req, blkvsc_req->cond);
 539
 540         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 541
 542         buf = kmap(page_buf);
 543
 544         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 545         /* be to le */
 546         device_type = buf[0] & 0x1F;
 547
 548         if (device_type == 0x0) {
 549                 blkdev->device_type = HARDDISK_TYPE;
 550         } else if (device_type == 0x5) {
 551                 blkdev->device_type = DVD_TYPE;
 552         } else {
 553                 /* TODO: this is currently unsupported device type */
 554                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 555         }
 556
 557         DPRINT_DBG(BLKVSC_DRV, "device type %d \n", device_type);
 558
 559         blkdev->device_id_len = buf[7];
 560         if (blkdev->device_id_len > 64)
 561                 blkdev->device_id_len = 64;
 562
 563         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 564         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 565          * blkdev->device_id_len); */
 566
 567         kunmap(page_buf);
 568
 569         __free_page(page_buf);
 570
 571         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 572
 573         return 0;
 574 }
 575
 576 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 577 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 578 {
 579         struct blkvsc_request *blkvsc_req;
 580         struct page *page_buf;
 581         unsigned char *buf;
 582         struct scsi_sense_hdr sense_hdr;
 583
 584         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 585
 586         blkdev->sector_size = 0;
 587         blkdev->capacity = 0;
 588         blkdev->media_not_present = 0; /* assume a disk is present */
 589
 590         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 591         if (!blkvsc_req)
 592                 return -ENOMEM;
 593
 594         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 595         page_buf = alloc_page(GFP_KERNEL);
 596         if (!page_buf) {
 597                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 598                 return -ENOMEM;
 599         }
 600
 601         init_waitqueue_head(&blkvsc_req->wevent);
 602         blkvsc_req->dev = blkdev;
 603         blkvsc_req->req = NULL;
 604         blkvsc_req->write = 0;
 605
 606         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 607         blkvsc_req->request.DataBuffer.Offset = 0;
 608         blkvsc_req->request.DataBuffer.Length = 8;
 609
 610         blkvsc_req->cmnd[0] = READ_CAPACITY;
 611         blkvsc_req->cmd_len = 16;
 612
 613         /*
 614          * Set this here since the completion routine may be invoked
 615          * and completed before we return
 616          */
 617         blkvsc_req->cond = 0;
 618
 619         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 620
 621         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 622                    blkvsc_req, blkvsc_req->cond);
 623
 624         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 625
 626         /* check error */
 627         if (blkvsc_req->request.Status) {
 628                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 629                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 630
 631                 if (sense_hdr.asc == 0x3A) {
 632                         /* Medium not present */
 633                         blkdev->media_not_present = 1;
 634                 }
 635                 return 0;
 636         }
 637         buf = kmap(page_buf);
 638
 639         /* be to le */
 640         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 641                             (buf[2] << 8) | buf[3]) + 1;
 642         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 643                               (buf[6] << 8) | buf[7];
 644
 645         kunmap(page_buf);
 646
 647         __free_page(page_buf);
 648
 649         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 650
 651         return 0;
 652 }
 653
 654 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 655 {
 656         struct blkvsc_request *blkvsc_req;
 657         struct page *page_buf;
 658         unsigned char *buf;
 659         struct scsi_sense_hdr sense_hdr;
 660
 661         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 662
 663         blkdev->sector_size = 0;
 664         blkdev->capacity = 0;
 665         blkdev->media_not_present = 0; /* assume a disk is present */
 666
 667         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 668         if (!blkvsc_req)
 669                 return -ENOMEM;
 670
 671         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 672         page_buf = alloc_page(GFP_KERNEL);
 673         if (!page_buf) {
 674                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 675                 return -ENOMEM;
 676         }
 677
 678         init_waitqueue_head(&blkvsc_req->wevent);
 679         blkvsc_req->dev = blkdev;
 680         blkvsc_req->req = NULL;
 681         blkvsc_req->write = 0;
 682
 683         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 684         blkvsc_req->request.DataBuffer.Offset = 0;
 685         blkvsc_req->request.DataBuffer.Length = 12;
 686
 687         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 688         blkvsc_req->cmd_len = 16;
 689
 690         /*
 691          * Set this here since the completion routine may be invoked
 692          * and completed before we return
 693          */
 694         blkvsc_req->cond = 0;
 695
 696         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 697
 698         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 699                    blkvsc_req, blkvsc_req->cond);
 700
 701         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 702
 703         /* check error */
 704         if (blkvsc_req->request.Status) {
 705                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 706                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 707                 if (sense_hdr.asc == 0x3A) {
 708                         /* Medium not present */
 709                         blkdev->media_not_present = 1;
 710                 }
 711                 return 0;
 712         }
 713         buf = kmap(page_buf);
 714
 715         /* be to le */
 716         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 717         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 718
 719 #if 0
 720         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 721                             (buf[2] << 8) | buf[3]) + 1;
 722         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 723                               (buf[6] << 8) | buf[7];
 724 #endif
 725
 726         kunmap(page_buf);
 727
 728         __free_page(page_buf);
 729
 730         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 731
 732         return 0;
 733 }
 734
 735 /**
 736  * blkvsc_remove() - Callback when our device is removed
 737  */
 738 static int blkvsc_remove(struct device *device)
 739 {
 740         struct driver_context *driver_ctx =
 741                                 driver_to_driver_context(device->driver);
 742         struct blkvsc_driver_context *blkvsc_drv_ctx =
 743                                 (struct blkvsc_driver_context *)driver_ctx;
 744         struct storvsc_driver_object *storvsc_drv_obj =
 745                                 &blkvsc_drv_ctx->drv_obj;
 746         struct vm_device *device_ctx = device_to_vm_device(device);
 747         struct hv_device *device_obj = &device_ctx->device_obj;
 748         struct block_device_context *blkdev = dev_get_drvdata(device);
 749         unsigned long flags;
 750         int ret;
 751
 752         DPRINT_ENTER(BLKVSC_DRV);
 753
 754         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 755
 756         if (!storvsc_drv_obj->Base.OnDeviceRemove) {
 757                 DPRINT_EXIT(BLKVSC_DRV);
 758                 return -1;
 759         }
 760
 761         /*
 762          * Call to the vsc driver to let it know that the device is being
 763          * removed
 764          */
 765         ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 766         if (ret != 0) {
 767                 /* TODO: */
 768                 DPRINT_ERR(BLKVSC_DRV,
 769                            "unable to remove blkvsc device (ret %d)", ret);
 770         }
 771
 772         /* Get to a known state */
 773         spin_lock_irqsave(&blkdev->lock, flags);
 774
 775         blkdev->shutting_down = 1;
 776
 777         blk_stop_queue(blkdev->gd->queue);
 778
 779         spin_unlock_irqrestore(&blkdev->lock, flags);
 780
 781         while (blkdev->num_outstanding_reqs) {
 782                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 783                             blkdev->num_outstanding_reqs);
 784                 udelay(100);
 785         }
 786
 787         blkvsc_do_flush(blkdev);
 788
 789         spin_lock_irqsave(&blkdev->lock, flags);
 790
 791         blkvsc_cancel_pending_reqs(blkdev);
 792
 793         spin_unlock_irqrestore(&blkdev->lock, flags);
 794
 795         blk_cleanup_queue(blkdev->gd->queue);
 796
 797         del_gendisk(blkdev->gd);
 798
 799         kmem_cache_destroy(blkdev->request_pool);
 800
 801         kfree(blkdev);
 802
 803         DPRINT_EXIT(BLKVSC_DRV);
 804
 805         return ret;
 806 }
 807
 808 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 809 {
 810         ASSERT(blkvsc_req->req);
 811         ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8));
 812
 813         blkvsc_req->cmd_len = 16;
 814
 815         if (blkvsc_req->sector_start > 0xffffffff) {
 816                 if (rq_data_dir(blkvsc_req->req)) {
 817                         blkvsc_req->write = 1;
 818                         blkvsc_req->cmnd[0] = WRITE_16;
 819                 } else {
 820                         blkvsc_req->write = 0;
 821                         blkvsc_req->cmnd[0] = READ_16;
 822                 }
 823
 824                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 825
 826                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 827                                 cpu_to_be64(blkvsc_req->sector_start);
 828                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 829                                 cpu_to_be32(blkvsc_req->sector_count);
 830         } else if ((blkvsc_req->sector_count > 0xff) ||
 831                    (blkvsc_req->sector_start > 0x1fffff)) {
 832                 if (rq_data_dir(blkvsc_req->req)) {
 833                         blkvsc_req->write = 1;
 834                         blkvsc_req->cmnd[0] = WRITE_10;
 835                 } else {
 836                         blkvsc_req->write = 0;
 837                         blkvsc_req->cmnd[0] = READ_10;
 838                 }
 839
 840                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 841
 842                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 843                                 cpu_to_be32(blkvsc_req->sector_start);
 844                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 845                                 cpu_to_be16(blkvsc_req->sector_count);
 846         } else {
 847                 if (rq_data_dir(blkvsc_req->req)) {
 848                         blkvsc_req->write = 1;
 849                         blkvsc_req->cmnd[0] = WRITE_6;
 850                 } else {
 851                         blkvsc_req->write = 0;
 852                         blkvsc_req->cmnd[0] = READ_6;
 853                 }
 854
 855                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 856                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 857                 blkvsc_req->cmnd[1] &= 0x1f;
 858                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 859         }
 860 }
 861
 862 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 863                         void (*request_completion)(struct hv_storvsc_request *))
 864 {
 865         struct block_device_context *blkdev = blkvsc_req->dev;
 866         struct vm_device *device_ctx = blkdev->device_ctx;
 867         struct driver_context *driver_ctx =
 868                         driver_to_driver_context(device_ctx->device.driver);
 869         struct blkvsc_driver_context *blkvsc_drv_ctx =
 870                         (struct blkvsc_driver_context *)driver_ctx;
 871         struct storvsc_driver_object *storvsc_drv_obj =
 872                         &blkvsc_drv_ctx->drv_obj;
 873         struct hv_storvsc_request *storvsc_req;
 874         int ret;
 875
 876         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 877                    "req %p type %s start_sector %lu count %ld offset %d "
 878                    "len %d\n", blkvsc_req,
 879                    (blkvsc_req->write) ? "WRITE" : "READ",
 880                    (unsigned long) blkvsc_req->sector_start,
 881                    blkvsc_req->sector_count,
 882                    blkvsc_req->request.DataBuffer.Offset,
 883                    blkvsc_req->request.DataBuffer.Length);
 884 #if 0
 885         for (i = 0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++) {
 886                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 887                            "req %p pfn[%d] %llx\n",
 888                            blkvsc_req, i,
 889                            blkvsc_req->request.DataBuffer.PfnArray[i]);
 890         }
 891 #endif
 892
 893         storvsc_req = &blkvsc_req->request;
 894         storvsc_req->Extension = (void *)((unsigned long)blkvsc_req +
 895                                           sizeof(struct blkvsc_request));
 896
 897         storvsc_req->Type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 898
 899         storvsc_req->OnIOCompletion = request_completion;
 900         storvsc_req->Context = blkvsc_req;
 901
 902         storvsc_req->Host = blkdev->port;
 903         storvsc_req->Bus = blkdev->path;
 904         storvsc_req->TargetId = blkdev->target;
 905         storvsc_req->LunId = 0;  /* this is not really used at all */
 906
 907         storvsc_req->CdbLen = blkvsc_req->cmd_len;
 908         storvsc_req->Cdb = blkvsc_req->cmnd;
 909
 910         storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
 911         storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
 912
 913         ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj,
 914                                            &blkvsc_req->request);
 915         if (ret == 0)
 916                 blkdev->num_outstanding_reqs++;
 917
 918         return ret;
 919 }
 920
 921 /*
 922  * We break the request into 1 or more blkvsc_requests and submit
 923  * them.  If we cant submit them all, we put them on the
 924  * pending_list. The blkvsc_request() will work on the pending_list.
 925  */
 926 static int blkvsc_do_request(struct block_device_context *blkdev,
 927                              struct request *req)
 928 {
 929         struct bio *bio = NULL;
 930         struct bio_vec *bvec = NULL;
 931         struct bio_vec *prev_bvec = NULL;
 932         struct blkvsc_request *blkvsc_req = NULL;
 933         struct blkvsc_request *tmp;
 934         int databuf_idx = 0;
 935         int seg_idx = 0;
 936         sector_t start_sector;
 937         unsigned long num_sectors = 0;
 938         int ret = 0;
 939         int pending = 0;
 940         struct blkvsc_request_group *group = NULL;
 941
 942         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu \n", blkdev, req,
 943                   (unsigned long)blk_rq_pos(req));
 944
 945         /* Create a group to tie req to list of blkvsc_reqs */
 946         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 947         if (!group)
 948                 return -ENOMEM;
 949
 950         INIT_LIST_HEAD(&group->blkvsc_req_list);
 951         group->outstanding = group->status = 0;
 952
 953         start_sector = blk_rq_pos(req);
 954
 955         /* foreach bio in the request */
 956         if (req->bio) {
 957                 for (bio = req->bio; bio; bio = bio->bi_next) {
 958                         /*
 959                          * Map this bio into an existing or new storvsc request
 960                          */
 961                         bio_for_each_segment(bvec, bio, seg_idx) {
 962                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 963                                            "- req %p bio %p bvec %p seg_idx %d "
 964                                            "databuf_idx %d\n", req, bio, bvec,
 965                                            seg_idx, databuf_idx);
 966
 967                                 /* Get a new storvsc request */
 968                                 /* 1st-time */
 969                                 if ((!blkvsc_req) ||
 970                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 971                                     /* hole at the begin of page */
 972                                     || (bvec->bv_offset != 0) ||
 973                                     /* hold at the end of page */
 974                                     (prev_bvec &&
 975                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 976                                         /* submit the prev one */
 977                                         if (blkvsc_req) {
 978                                                 blkvsc_req->sector_start = start_sector;
 979                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 980
 981                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 982                                                 blkvsc_init_rw(blkvsc_req);
 983                                         }
 984
 985                                         /*
 986                                          * Create new blkvsc_req to represent
 987                                          * the current bvec
 988                                          */
 989                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 990                                         if (!blkvsc_req) {
 991                                                 /* free up everything */
 992                                                 list_for_each_entry_safe(
 993                                                         blkvsc_req, tmp,
 994                                                         &group->blkvsc_req_list,
 995                                                         req_entry) {
 996                                                         list_del(&blkvsc_req->req_entry);
 997                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 998                                                 }
 999
1000                                                 kmem_cache_free(blkdev->request_pool, group);
1001                                                 return -ENOMEM;
1002                                         }
1003
1004                                         memset(blkvsc_req, 0,
1005                                                sizeof(struct blkvsc_request));
1006
1007                                         blkvsc_req->dev = blkdev;
1008                                         blkvsc_req->req = req;
1009                                         blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
1010                                         blkvsc_req->request.DataBuffer.Length = 0;
1011
1012                                         /* Add to the group */
1013                                         blkvsc_req->group = group;
1014                                         blkvsc_req->group->outstanding++;
1015                                         list_add_tail(&blkvsc_req->req_entry,
1016                                                 &blkvsc_req->group->blkvsc_req_list);
1017
1018                                         start_sector += num_sectors;
1019                                         num_sectors = 0;
1020                                         databuf_idx = 0;
1021                                 }
1022
1023                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1024                                 blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
1025                                 blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
1026
1027                                 prev_bvec = bvec;
1028
1029                                 databuf_idx++;
1030                                 num_sectors += bvec->bv_len >> 9;
1031
1032                         } /* bio_for_each_segment */
1033
1034                 } /* rq_for_each_bio */
1035         }
1036
1037         /* Handle the last one */
1038         if (blkvsc_req) {
1039                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1040                            blkdev, req, blkvsc_req->group,
1041                            blkvsc_req->group->outstanding);
1042
1043                 blkvsc_req->sector_start = start_sector;
1044                 sector_div(blkvsc_req->sector_start,
1045                            (blkdev->sector_size >> 9));
1046
1047                 blkvsc_req->sector_count = num_sectors /
1048                                            (blkdev->sector_size >> 9);
1049
1050                 blkvsc_init_rw(blkvsc_req);
1051         }
1052
1053         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1054                 if (pending) {
1055                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1056                                    "pending_list - blkvsc_req %p start_sect %lu"
1057                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1058                                    (unsigned long)blkvsc_req->sector_start,
1059                                    blkvsc_req->sector_count,
1060                                    (unsigned long)start_sector,
1061                                    (unsigned long)num_sectors);
1062
1063                         list_add_tail(&blkvsc_req->pend_entry,
1064                                       &blkdev->pending_list);
1065                 } else {
1066                         ret = blkvsc_submit_request(blkvsc_req,
1067                                                     blkvsc_request_completion);
1068                         if (ret == -1) {
1069                                 pending = 1;
1070                                 list_add_tail(&blkvsc_req->pend_entry,
1071                                               &blkdev->pending_list);
1072                         }
1073
1074                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1075                                    "start_sect %lu sect_count %ld (%lu %ld) "
1076                                    "ret %d\n", blkvsc_req,
1077                                    (unsigned long)blkvsc_req->sector_start,
1078                                    blkvsc_req->sector_count,
1079                                    (unsigned long)start_sector,
1080                                    num_sectors, ret);
1081                 }
1082         }
1083
1084         return pending;
1085 }
1086
1087 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1088 {
1089         struct blkvsc_request *blkvsc_req =
1090                         (struct blkvsc_request *)request->Context;
1091         struct block_device_context *blkdev =
1092                         (struct block_device_context *)blkvsc_req->dev;
1093         struct scsi_sense_hdr sense_hdr;
1094
1095         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1096                    blkvsc_req);
1097
1098         blkdev->num_outstanding_reqs--;
1099
1100         if (blkvsc_req->request.Status)
1101                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1102                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1103                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1104
1105         blkvsc_req->cond = 1;
1106         wake_up_interruptible(&blkvsc_req->wevent);
1107 }
1108
1109 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1110 {
1111         struct blkvsc_request *blkvsc_req =
1112                         (struct blkvsc_request *)request->Context;
1113         struct block_device_context *blkdev =
1114                         (struct block_device_context *)blkvsc_req->dev;
1115         unsigned long flags;
1116         struct blkvsc_request *comp_req, *tmp;
1117
1118         ASSERT(blkvsc_req->group);
1119
1120         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1121                    "sect_start %lu sect_count %ld len %d group outstd %d "
1122                    "total outstd %d\n",
1123                    blkdev, blkvsc_req, blkvsc_req->group,
1124                    (blkvsc_req->write) ? "WRITE" : "READ",
1125                    (unsigned long)blkvsc_req->sector_start,
1126                    blkvsc_req->sector_count,
1127                    blkvsc_req->request.DataBuffer.Length,
1128                    blkvsc_req->group->outstanding,
1129                    blkdev->num_outstanding_reqs);
1130
1131         spin_lock_irqsave(&blkdev->lock, flags);
1132
1133         blkdev->num_outstanding_reqs--;
1134         blkvsc_req->group->outstanding--;
1135
1136         /*
1137          * Only start processing when all the blkvsc_reqs are
1138          * completed. This guarantees no out-of-order blkvsc_req
1139          * completion when calling end_that_request_first()
1140          */
1141         if (blkvsc_req->group->outstanding == 0) {
1142                 list_for_each_entry_safe(comp_req, tmp,
1143                                          &blkvsc_req->group->blkvsc_req_list,
1144                                          req_entry) {
1145                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1146                                    "sect_start %lu sect_count %ld \n",
1147                                    comp_req,
1148                                    (unsigned long)comp_req->sector_start,
1149                                    comp_req->sector_count);
1150
1151                         list_del(&comp_req->req_entry);
1152
1153                         if (!__blk_end_request(comp_req->req,
1154                                 (!comp_req->request.Status ? 0 : -EIO),
1155                                 comp_req->sector_count * blkdev->sector_size)) {
1156                                 /*
1157                                  * All the sectors have been xferred ie the
1158                                  * request is done
1159                                  */
1160                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1161                                            comp_req->req);
1162                                 kmem_cache_free(blkdev->request_pool,
1163                                                 comp_req->group);
1164                         }
1165
1166                         kmem_cache_free(blkdev->request_pool, comp_req);
1167                 }
1168
1169                 if (!blkdev->shutting_down) {
1170                         blkvsc_do_pending_reqs(blkdev);
1171                         blk_start_queue(blkdev->gd->queue);
1172                         blkvsc_request(blkdev->gd->queue);
1173                 }
1174         }
1175
1176         spin_unlock_irqrestore(&blkdev->lock, flags);
1177 }
1178
1179 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1180 {
1181         struct blkvsc_request *pend_req, *tmp;
1182         struct blkvsc_request *comp_req, *tmp2;
1183
1184         int ret = 0;
1185
1186         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1187
1188         /* Flush the pending list first */
1189         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1190                                  pend_entry) {
1191                 /*
1192                  * The pend_req could be part of a partially completed
1193                  * request. If so, complete those req first until we
1194                  * hit the pend_req
1195                  */
1196                 list_for_each_entry_safe(comp_req, tmp2,
1197                                          &pend_req->group->blkvsc_req_list,
1198                                          req_entry) {
1199                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1200                                    "sect_start %lu sect_count %ld \n",
1201                                    comp_req,
1202                                    (unsigned long) comp_req->sector_start,
1203                                    comp_req->sector_count);
1204
1205                         if (comp_req == pend_req)
1206                                 break;
1207
1208                         list_del(&comp_req->req_entry);
1209
1210                         if (comp_req->req) {
1211                                 ret = __blk_end_request(comp_req->req,
1212                                         (!comp_req->request.Status ? 0 : -EIO),
1213                                         comp_req->sector_count *
1214                                         blkdev->sector_size);
1215                                 ASSERT(ret != 0);
1216                         }
1217
1218                         kmem_cache_free(blkdev->request_pool, comp_req);
1219                 }
1220
1221                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1222                            pend_req);
1223
1224                 list_del(&pend_req->pend_entry);
1225
1226                 list_del(&pend_req->req_entry);
1227
1228                 if (comp_req->req) {
1229                         if (!__blk_end_request(pend_req->req, -EIO,
1230                                                pend_req->sector_count *
1231                                                blkdev->sector_size)) {
1232                                 /*
1233                                  * All the sectors have been xferred ie the
1234                                  * request is done
1235                                  */
1236                                 DPRINT_DBG(BLKVSC_DRV,
1237                                            "blkvsc_cancel_pending_reqs() - "
1238                                            "req %p COMPLETED\n", pend_req->req);
1239                                 kmem_cache_free(blkdev->request_pool,
1240                                                 pend_req->group);
1241                         }
1242                 }
1243
1244                 kmem_cache_free(blkdev->request_pool, pend_req);
1245         }
1246
1247         return ret;
1248 }
1249
1250 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1251 {
1252         struct blkvsc_request *pend_req, *tmp;
1253         int ret = 0;
1254
1255         /* Flush the pending list first */
1256         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1257                                  pend_entry) {
1258                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1259                            pend_req);
1260
1261                 ret = blkvsc_submit_request(pend_req,
1262                                             blkvsc_request_completion);
1263                 if (ret != 0)
1264                         break;
1265                 else
1266                         list_del(&pend_req->pend_entry);
1267         }
1268
1269         return ret;
1270 }
1271
1272 static void blkvsc_request(struct request_queue *queue)
1273 {
1274         struct block_device_context *blkdev = NULL;
1275         struct request *req;
1276         int ret = 0;
1277
1278         DPRINT_DBG(BLKVSC_DRV, "- enter \n");
1279         while ((req = blk_peek_request(queue)) != NULL) {
1280                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1281
1282                 blkdev = req->rq_disk->private_data;
1283                 if (blkdev->shutting_down || !blk_fs_request(req) ||
1284                     blkdev->media_not_present) {
1285                         __blk_end_request_cur(req, 0);
1286                         continue;
1287                 }
1288
1289                 ret = blkvsc_do_pending_reqs(blkdev);
1290
1291                 if (ret != 0) {
1292                         DPRINT_DBG(BLKVSC_DRV,
1293                                    "- stop queue - pending_list not empty\n");
1294                         blk_stop_queue(queue);
1295                         break;
1296                 }
1297
1298                 blk_start_request(req);
1299
1300                 ret = blkvsc_do_request(blkdev, req);
1301                 if (ret > 0) {
1302                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1303                         blk_stop_queue(queue);
1304                         break;
1305                 } else if (ret < 0) {
1306                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1307                         blk_requeue_request(queue, req);
1308                         blk_stop_queue(queue);
1309                         break;
1310                 }
1311         }
1312 }
1313
1314 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1315 {
1316         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1317
1318         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1319                    blkdev->gd->disk_name);
1320
1321         spin_lock(&blkdev->lock);
1322
1323         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1324                 spin_unlock(&blkdev->lock);
1325                 check_disk_change(bdev);
1326                 spin_lock(&blkdev->lock);
1327         }
1328
1329         blkdev->users++;
1330
1331         spin_unlock(&blkdev->lock);
1332         return 0;
1333 }
1334
1335 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1336 {
1337         struct block_device_context *blkdev = disk->private_data;
1338
1339         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1340                    blkdev->gd->disk_name);
1341
1342         spin_lock(&blkdev->lock);
1343         if (blkdev->users == 1) {
1344                 spin_unlock(&blkdev->lock);
1345                 blkvsc_do_flush(blkdev);
1346                 spin_lock(&blkdev->lock);
1347         }
1348
1349         blkdev->users--;
1350
1351         spin_unlock(&blkdev->lock);
1352         return 0;
1353 }
1354
1355 static int blkvsc_media_changed(struct gendisk *gd)
1356 {
1357         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1358         return 1;
1359 }
1360
1361 static int blkvsc_revalidate_disk(struct gendisk *gd)
1362 {
1363         struct block_device_context *blkdev = gd->private_data;
1364
1365         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1366
1367         if (blkdev->device_type == DVD_TYPE) {
1368                 blkvsc_do_read_capacity(blkdev);
1369                 set_capacity(blkdev->gd, blkdev->capacity *
1370                             (blkdev->sector_size/512));
1371                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1372         }
1373         return 0;
1374 }
1375
1376 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1377 {
1378         sector_t total_sectors = get_capacity(bd->bd_disk);
1379         sector_t cylinder_times_heads = 0;
1380         sector_t temp = 0;
1381
1382         int sectors_per_track = 0;
1383         int heads = 0;
1384         int cylinders = 0;
1385         int rem = 0;
1386
1387         if (total_sectors > (65535 * 16 * 255))
1388                 total_sectors = (65535 * 16 * 255);
1389
1390         if (total_sectors >= (65535 * 16 * 63)) {
1391                 sectors_per_track = 255;
1392                 heads = 16;
1393
1394                 cylinder_times_heads = total_sectors;
1395                 /* sector_div stores the quotient in cylinder_times_heads */
1396                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1397         } else {
1398                 sectors_per_track = 17;
1399
1400                 cylinder_times_heads = total_sectors;
1401                 /* sector_div stores the quotient in cylinder_times_heads */
1402                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1403
1404                 temp = cylinder_times_heads + 1023;
1405                 /* sector_div stores the quotient in temp */
1406                 rem = sector_div(temp, 1024);
1407
1408                 heads = temp;
1409
1410                 if (heads < 4)
1411                         heads = 4;
1412
1413
1414                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1415                         sectors_per_track = 31;
1416                         heads = 16;
1417
1418                         cylinder_times_heads = total_sectors;
1419                         /*
1420                          * sector_div stores the quotient in
1421                          * cylinder_times_heads
1422                          */
1423                         rem = sector_div(cylinder_times_heads,
1424                                          sectors_per_track);
1425                 }
1426
1427                 if (cylinder_times_heads >= (heads * 1024)) {
1428                         sectors_per_track = 63;
1429                         heads = 16;
1430
1431                         cylinder_times_heads = total_sectors;
1432                         /*
1433                          * sector_div stores the quotient in
1434                          * cylinder_times_heads
1435                          */
1436                         rem = sector_div(cylinder_times_heads,
1437                                          sectors_per_track);
1438                 }
1439         }
1440
1441         temp = cylinder_times_heads;
1442         /* sector_div stores the quotient in temp */
1443         rem = sector_div(temp, heads);
1444         cylinders = temp;
1445
1446         hg->heads = heads;
1447         hg->sectors = sectors_per_track;
1448         hg->cylinders = cylinders;
1449
1450         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1451                     sectors_per_track);
1452
1453     return 0;
1454 }
1455
1456 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1457                         unsigned cmd, unsigned long argument)
1458 {
1459 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1460         int ret;
1461
1462         switch (cmd) {
1463         /*
1464          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1465          * than just a GUID. Commented it out for now.
1466          */
1467 #if 0
1468         case HDIO_GET_IDENTITY:
1469                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1470                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1471                                  blkdev->device_id_len))
1472                         ret = -EFAULT;
1473                 break;
1474 #endif
1475         default:
1476                 ret = -EINVAL;
1477                 break;
1478         }
1479
1480         return ret;
1481 }
1482
1483 static int __init blkvsc_init(void)
1484 {
1485         int ret;
1486
1487         ASSERT(sizeof(sector_t) == 8); /* Make sure CONFIG_LBD is set */
1488
1489         DPRINT_ENTER(BLKVSC_DRV);
1490
1491         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1492
1493         ret = blkvsc_drv_init(BlkVscInitialize);
1494
1495         DPRINT_EXIT(BLKVSC_DRV);
1496
1497         return ret;
1498 }
1499
1500 static void __exit blkvsc_exit(void)
1501 {
1502         DPRINT_ENTER(BLKVSC_DRV);
1503         blkvsc_drv_exit();
1504         DPRINT_ENTER(BLKVSC_DRV);
1505 }
1506
1507 MODULE_LICENSE("GPL");
1508 MODULE_VERSION(HV_DRV_VERSION);
1509 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
1510 module_init(blkvsc_init);
1511 module_exit(blkvsc_exit);