drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "osd.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 /*
  50  * This request ties the struct request and struct
  51  * blkvsc_request/hv_storvsc_request together A struct request may be
  52  * represented by 1 or more struct blkvsc_request
  53  */
  54 struct blkvsc_request_group {
  55         int outstanding;
  56         int status;
  57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  58 };
  59
  60 struct blkvsc_request {
  61         /* blkvsc_request_group.blkvsc_req_list */
  62         struct list_head req_entry;
  63
  64         /* block_device_context.pending_list */
  65         struct list_head pend_entry;
  66
  67         /* This may be null if we generate a request internally */
  68         struct request *req;
  69
  70         struct block_device_context *dev;
  71
  72         /* The group this request is part of. Maybe null */
  73         struct blkvsc_request_group *group;
  74
  75         wait_queue_head_t wevent;
  76         int cond;
  77
  78         int write;
  79         sector_t sector_start;
  80         unsigned long sector_count;
  81
  82         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  83         unsigned char cmd_len;
  84         unsigned char cmnd[MAX_COMMAND_SIZE];
  85
  86         struct hv_storvsc_request request;
  87         /*
  88          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  89          * because - The extension buffer falls right here and is pointed to by
  90          * request.Extension;
  91          * Which sounds like a horrible idea, who designed this?
  92          */
  93 };
  94
  95 /* Per device structure */
  96 struct block_device_context {
  97         /* point back to our device context */
  98         struct vm_device *device_ctx;
  99         struct kmem_cache *request_pool;
 100         spinlock_t lock;
 101         struct gendisk *gd;
 102         enum blkvsc_device_type device_type;
 103         struct list_head pending_list;
 104
 105         unsigned char device_id[64];
 106         unsigned int device_id_len;
 107         int num_outstanding_reqs;
 108         int shutting_down;
 109         int media_not_present;
 110         unsigned int sector_size;
 111         sector_t capacity;
 112         unsigned int port;
 113         unsigned char path;
 114         unsigned char target;
 115         int users;
 116 };
 117
 118 /* Per driver */
 119 struct blkvsc_driver_context {
 120         /* !! These must be the first 2 fields !! */
 121         /* FIXME this is a bug! */
 122         struct driver_context drv_ctx;
 123         struct storvsc_driver_object drv_obj;
 124 };
 125
 126 /* Static decl */
 127 static DEFINE_MUTEX(blkvsc_mutex);
 128 static int blkvsc_probe(struct device *dev);
 129 static int blkvsc_remove(struct device *device);
 130 static void blkvsc_shutdown(struct device *device);
 131
 132 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 133 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 134 static int blkvsc_media_changed(struct gendisk *gd);
 135 static int blkvsc_revalidate_disk(struct gendisk *gd);
 136 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 137 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 138                         unsigned cmd, unsigned long argument);
 139 static void blkvsc_request(struct request_queue *queue);
 140 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 141 static int blkvsc_do_request(struct block_device_context *blkdev,
 142                              struct request *req);
 143 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 144                 void (*request_completion)(struct hv_storvsc_request *));
 145 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 146 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 147 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 148 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 149 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 150 static int blkvsc_do_flush(struct block_device_context *blkdev);
 151 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 152 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 153
 154 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 155 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 156 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 157
 158 /* The one and only one */
 159 static struct blkvsc_driver_context g_blkvsc_drv;
 160
 161 static const struct block_device_operations block_ops = {
 162         .owner = THIS_MODULE,
 163         .open = blkvsc_open,
 164         .release = blkvsc_release,
 165         .media_changed = blkvsc_media_changed,
 166         .revalidate_disk = blkvsc_revalidate_disk,
 167         .getgeo = blkvsc_getgeo,
 168         .ioctl  = blkvsc_ioctl,
 169 };
 170
 171 /*
 172  * blkvsc_drv_init -  BlkVsc driver initialization.
 173  */
 174 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 175 {
 176         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 177         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 178         int ret;
 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         return ret;
 197 }
 198
 199 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 200 {
 201         struct device **curr = (struct device **)data;
 202         *curr = dev;
 203         return 1; /* stop iterating */
 204 }
 205
 206 static void blkvsc_drv_exit(void)
 207 {
 208         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 209         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 210         struct device *current_dev;
 211         int ret;
 212
 213         while (1) {
 214                 current_dev = NULL;
 215
 216                 /* Get the device */
 217                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
 218                                              (void *) &current_dev,
 219                                              blkvsc_drv_exit_cb);
 220
 221                 if (ret)
 222                         DPRINT_WARN(BLKVSC_DRV,
 223                                     "driver_for_each_device returned %d", ret);
 224
 225
 226                 if (current_dev == NULL)
 227                         break;
 228
 229                 /* Initiate removal from the top-down */
 230                 device_unregister(current_dev);
 231         }
 232
 233         if (storvsc_drv_obj->Base.OnCleanup)
 234                 storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
 235
 236         vmbus_child_driver_unregister(drv_ctx);
 237
 238         return;
 239 }
 240
 241 /*
 242  * blkvsc_probe - Add a new device for this driver
 243  */
 244 static int blkvsc_probe(struct device *device)
 245 {
 246         struct driver_context *driver_ctx =
 247                                 driver_to_driver_context(device->driver);
 248         struct blkvsc_driver_context *blkvsc_drv_ctx =
 249                                 (struct blkvsc_driver_context *)driver_ctx;
 250         struct storvsc_driver_object *storvsc_drv_obj =
 251                                 &blkvsc_drv_ctx->drv_obj;
 252         struct vm_device *device_ctx = device_to_vm_device(device);
 253         struct hv_device *device_obj = &device_ctx->device_obj;
 254
 255         struct block_device_context *blkdev = NULL;
 256         struct storvsc_device_info device_info;
 257         int major = 0;
 258         int devnum = 0;
 259         int ret = 0;
 260         static int ide0_registered;
 261         static int ide1_registered;
 262
 263         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 264
 265         if (!storvsc_drv_obj->Base.OnDeviceAdd) {
 266                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 267                 ret = -1;
 268                 goto Cleanup;
 269         }
 270
 271         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 272         if (!blkdev) {
 273                 ret = -ENOMEM;
 274                 goto Cleanup;
 275         }
 276
 277         INIT_LIST_HEAD(&blkdev->pending_list);
 278
 279         /* Initialize what we can here */
 280         spin_lock_init(&blkdev->lock);
 281
 282         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 283         /*      sizeof(struct blkvsc_request)); */
 284
 285         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
 286                                         sizeof(struct blkvsc_request) +
 287                                         storvsc_drv_obj->RequestExtSize, 0,
 288                                         SLAB_HWCACHE_ALIGN, NULL);
 289         if (!blkdev->request_pool) {
 290                 ret = -ENOMEM;
 291                 goto Cleanup;
 292         }
 293
 294
 295         /* Call to the vsc driver to add the device */
 296         ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
 297         if (ret != 0) {
 298                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 299                 goto Cleanup;
 300         }
 301
 302         blkdev->device_ctx = device_ctx;
 303         /* this identified the device 0 or 1 */
 304         blkdev->target = device_info.TargetId;
 305         /* this identified the ide ctrl 0 or 1 */
 306         blkdev->path = device_info.PathId;
 307
 308         dev_set_drvdata(device, blkdev);
 309
 310         /* Calculate the major and device num */
 311         if (blkdev->path == 0) {
 312                 major = IDE0_MAJOR;
 313                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 314
 315                 if (!ide0_registered) {
 316                         ret = register_blkdev(major, "ide");
 317                         if (ret != 0) {
 318                                 DPRINT_ERR(BLKVSC_DRV,
 319                                            "register_blkdev() failed! ret %d",
 320                                            ret);
 321                                 goto Remove;
 322                         }
 323
 324                         ide0_registered = 1;
 325                 }
 326         } else if (blkdev->path == 1) {
 327                 major = IDE1_MAJOR;
 328                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 329
 330                 if (!ide1_registered) {
 331                         ret = register_blkdev(major, "ide");
 332                         if (ret != 0) {
 333                                 DPRINT_ERR(BLKVSC_DRV,
 334                                            "register_blkdev() failed! ret %d",
 335                                            ret);
 336                                 goto Remove;
 337                         }
 338
 339                         ide1_registered = 1;
 340                 }
 341         } else {
 342                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 343                 ret = -1;
 344                 goto Cleanup;
 345         }
 346
 347         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 348
 349         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 350         if (!blkdev->gd) {
 351                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 352                 ret = -1;
 353                 goto Cleanup;
 354         }
 355
 356         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 357
 358         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 359         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 360         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 361         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 362         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 363
 364         blkdev->gd->major = major;
 365         if (devnum == 1 || devnum == 3)
 366                 blkdev->gd->first_minor = BLKVSC_MINORS;
 367         else
 368                 blkdev->gd->first_minor = 0;
 369         blkdev->gd->fops = &block_ops;
 370         blkdev->gd->private_data = blkdev;
 371         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 372
 373         blkvsc_do_inquiry(blkdev);
 374         if (blkdev->device_type == DVD_TYPE) {
 375                 set_disk_ro(blkdev->gd, 1);
 376                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 377                 blkvsc_do_read_capacity(blkdev);
 378         } else {
 379                 blkvsc_do_read_capacity16(blkdev);
 380         }
 381
 382         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 383         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 384         /* go! */
 385         add_disk(blkdev->gd);
 386
 387         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 388                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 389                     blkdev->sector_size);
 390
 391         return ret;
 392
 393 Remove:
 394         storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 395
 396 Cleanup:
 397         if (blkdev) {
 398                 if (blkdev->request_pool) {
 399                         kmem_cache_destroy(blkdev->request_pool);
 400                         blkdev->request_pool = NULL;
 401                 }
 402                 kfree(blkdev);
 403                 blkdev = NULL;
 404         }
 405
 406         return ret;
 407 }
 408
 409 static void blkvsc_shutdown(struct device *device)
 410 {
 411         struct block_device_context *blkdev = dev_get_drvdata(device);
 412         unsigned long flags;
 413
 414         if (!blkdev)
 415                 return;
 416
 417         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 418                    blkdev->users, blkdev->gd->disk_name);
 419
 420         spin_lock_irqsave(&blkdev->lock, flags);
 421
 422         blkdev->shutting_down = 1;
 423
 424         blk_stop_queue(blkdev->gd->queue);
 425
 426         spin_unlock_irqrestore(&blkdev->lock, flags);
 427
 428         while (blkdev->num_outstanding_reqs) {
 429                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 430                             blkdev->num_outstanding_reqs);
 431                 udelay(100);
 432         }
 433
 434         blkvsc_do_flush(blkdev);
 435
 436         spin_lock_irqsave(&blkdev->lock, flags);
 437
 438         blkvsc_cancel_pending_reqs(blkdev);
 439
 440         spin_unlock_irqrestore(&blkdev->lock, flags);
 441 }
 442
 443 static int blkvsc_do_flush(struct block_device_context *blkdev)
 444 {
 445         struct blkvsc_request *blkvsc_req;
 446
 447         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 448
 449         if (blkdev->device_type != HARDDISK_TYPE)
 450                 return 0;
 451
 452         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 453         if (!blkvsc_req)
 454                 return -ENOMEM;
 455
 456         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 457         init_waitqueue_head(&blkvsc_req->wevent);
 458         blkvsc_req->dev = blkdev;
 459         blkvsc_req->req = NULL;
 460         blkvsc_req->write = 0;
 461
 462         blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
 463         blkvsc_req->request.DataBuffer.Offset = 0;
 464         blkvsc_req->request.DataBuffer.Length = 0;
 465
 466         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 467         blkvsc_req->cmd_len = 10;
 468
 469         /*
 470          * Set this here since the completion routine may be invoked and
 471          * completed before we return
 472          */
 473         blkvsc_req->cond = 0;
 474         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 475
 476         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 477
 478         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 479
 480         return 0;
 481 }
 482
 483 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 484 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 485 {
 486         struct blkvsc_request *blkvsc_req;
 487         struct page *page_buf;
 488         unsigned char *buf;
 489         unsigned char device_type;
 490
 491         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 492
 493         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 494         if (!blkvsc_req)
 495                 return -ENOMEM;
 496
 497         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 498         page_buf = alloc_page(GFP_KERNEL);
 499         if (!page_buf) {
 500                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 501                 return -ENOMEM;
 502         }
 503
 504         init_waitqueue_head(&blkvsc_req->wevent);
 505         blkvsc_req->dev = blkdev;
 506         blkvsc_req->req = NULL;
 507         blkvsc_req->write = 0;
 508
 509         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 510         blkvsc_req->request.DataBuffer.Offset = 0;
 511         blkvsc_req->request.DataBuffer.Length = 64;
 512
 513         blkvsc_req->cmnd[0] = INQUIRY;
 514         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 515         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 516         blkvsc_req->cmnd[4] = 64;
 517         blkvsc_req->cmd_len = 6;
 518
 519         /*
 520          * Set this here since the completion routine may be invoked and
 521          * completed before we return
 522          */
 523         blkvsc_req->cond = 0;
 524
 525         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 526
 527         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 528                    blkvsc_req, blkvsc_req->cond);
 529
 530         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 531
 532         buf = kmap(page_buf);
 533
 534         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 535         /* be to le */
 536         device_type = buf[0] & 0x1F;
 537
 538         if (device_type == 0x0) {
 539                 blkdev->device_type = HARDDISK_TYPE;
 540         } else if (device_type == 0x5) {
 541                 blkdev->device_type = DVD_TYPE;
 542         } else {
 543                 /* TODO: this is currently unsupported device type */
 544                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 545         }
 546
 547         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 548
 549         blkdev->device_id_len = buf[7];
 550         if (blkdev->device_id_len > 64)
 551                 blkdev->device_id_len = 64;
 552
 553         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 554         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 555          * blkdev->device_id_len); */
 556
 557         kunmap(page_buf);
 558
 559         __free_page(page_buf);
 560
 561         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 562
 563         return 0;
 564 }
 565
 566 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 567 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 568 {
 569         struct blkvsc_request *blkvsc_req;
 570         struct page *page_buf;
 571         unsigned char *buf;
 572         struct scsi_sense_hdr sense_hdr;
 573
 574         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 575
 576         blkdev->sector_size = 0;
 577         blkdev->capacity = 0;
 578         blkdev->media_not_present = 0; /* assume a disk is present */
 579
 580         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 581         if (!blkvsc_req)
 582                 return -ENOMEM;
 583
 584         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 585         page_buf = alloc_page(GFP_KERNEL);
 586         if (!page_buf) {
 587                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 588                 return -ENOMEM;
 589         }
 590
 591         init_waitqueue_head(&blkvsc_req->wevent);
 592         blkvsc_req->dev = blkdev;
 593         blkvsc_req->req = NULL;
 594         blkvsc_req->write = 0;
 595
 596         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 597         blkvsc_req->request.DataBuffer.Offset = 0;
 598         blkvsc_req->request.DataBuffer.Length = 8;
 599
 600         blkvsc_req->cmnd[0] = READ_CAPACITY;
 601         blkvsc_req->cmd_len = 16;
 602
 603         /*
 604          * Set this here since the completion routine may be invoked
 605          * and completed before we return
 606          */
 607         blkvsc_req->cond = 0;
 608
 609         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 610
 611         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 612                    blkvsc_req, blkvsc_req->cond);
 613
 614         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 615
 616         /* check error */
 617         if (blkvsc_req->request.Status) {
 618                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 619                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 620
 621                 if (sense_hdr.asc == 0x3A) {
 622                         /* Medium not present */
 623                         blkdev->media_not_present = 1;
 624                 }
 625                 return 0;
 626         }
 627         buf = kmap(page_buf);
 628
 629         /* be to le */
 630         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 631                             (buf[2] << 8) | buf[3]) + 1;
 632         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 633                               (buf[6] << 8) | buf[7];
 634
 635         kunmap(page_buf);
 636
 637         __free_page(page_buf);
 638
 639         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 640
 641         return 0;
 642 }
 643
 644 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 645 {
 646         struct blkvsc_request *blkvsc_req;
 647         struct page *page_buf;
 648         unsigned char *buf;
 649         struct scsi_sense_hdr sense_hdr;
 650
 651         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 652
 653         blkdev->sector_size = 0;
 654         blkdev->capacity = 0;
 655         blkdev->media_not_present = 0; /* assume a disk is present */
 656
 657         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 658         if (!blkvsc_req)
 659                 return -ENOMEM;
 660
 661         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 662         page_buf = alloc_page(GFP_KERNEL);
 663         if (!page_buf) {
 664                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 665                 return -ENOMEM;
 666         }
 667
 668         init_waitqueue_head(&blkvsc_req->wevent);
 669         blkvsc_req->dev = blkdev;
 670         blkvsc_req->req = NULL;
 671         blkvsc_req->write = 0;
 672
 673         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 674         blkvsc_req->request.DataBuffer.Offset = 0;
 675         blkvsc_req->request.DataBuffer.Length = 12;
 676
 677         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 678         blkvsc_req->cmd_len = 16;
 679
 680         /*
 681          * Set this here since the completion routine may be invoked
 682          * and completed before we return
 683          */
 684         blkvsc_req->cond = 0;
 685
 686         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 687
 688         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 689                    blkvsc_req, blkvsc_req->cond);
 690
 691         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 692
 693         /* check error */
 694         if (blkvsc_req->request.Status) {
 695                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 696                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 697                 if (sense_hdr.asc == 0x3A) {
 698                         /* Medium not present */
 699                         blkdev->media_not_present = 1;
 700                 }
 701                 return 0;
 702         }
 703         buf = kmap(page_buf);
 704
 705         /* be to le */
 706         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 707         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 708
 709 #if 0
 710         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 711                             (buf[2] << 8) | buf[3]) + 1;
 712         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 713                               (buf[6] << 8) | buf[7];
 714 #endif
 715
 716         kunmap(page_buf);
 717
 718         __free_page(page_buf);
 719
 720         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 721
 722         return 0;
 723 }
 724
 725 /*
 726  * blkvsc_remove() - Callback when our device is removed
 727  */
 728 static int blkvsc_remove(struct device *device)
 729 {
 730         struct driver_context *driver_ctx =
 731                                 driver_to_driver_context(device->driver);
 732         struct blkvsc_driver_context *blkvsc_drv_ctx =
 733                                 (struct blkvsc_driver_context *)driver_ctx;
 734         struct storvsc_driver_object *storvsc_drv_obj =
 735                                 &blkvsc_drv_ctx->drv_obj;
 736         struct vm_device *device_ctx = device_to_vm_device(device);
 737         struct hv_device *device_obj = &device_ctx->device_obj;
 738         struct block_device_context *blkdev = dev_get_drvdata(device);
 739         unsigned long flags;
 740         int ret;
 741
 742         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 743
 744         if (!storvsc_drv_obj->Base.OnDeviceRemove)
 745                 return -1;
 746
 747         /*
 748          * Call to the vsc driver to let it know that the device is being
 749          * removed
 750          */
 751         ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 752         if (ret != 0) {
 753                 /* TODO: */
 754                 DPRINT_ERR(BLKVSC_DRV,
 755                            "unable to remove blkvsc device (ret %d)", ret);
 756         }
 757
 758         /* Get to a known state */
 759         spin_lock_irqsave(&blkdev->lock, flags);
 760
 761         blkdev->shutting_down = 1;
 762
 763         blk_stop_queue(blkdev->gd->queue);
 764
 765         spin_unlock_irqrestore(&blkdev->lock, flags);
 766
 767         while (blkdev->num_outstanding_reqs) {
 768                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 769                             blkdev->num_outstanding_reqs);
 770                 udelay(100);
 771         }
 772
 773         blkvsc_do_flush(blkdev);
 774
 775         spin_lock_irqsave(&blkdev->lock, flags);
 776
 777         blkvsc_cancel_pending_reqs(blkdev);
 778
 779         spin_unlock_irqrestore(&blkdev->lock, flags);
 780
 781         blk_cleanup_queue(blkdev->gd->queue);
 782
 783         del_gendisk(blkdev->gd);
 784
 785         kmem_cache_destroy(blkdev->request_pool);
 786
 787         kfree(blkdev);
 788
 789         return ret;
 790 }
 791
 792 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 793 {
 794         /* ASSERT(blkvsc_req->req); */
 795         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 796
 797         blkvsc_req->cmd_len = 16;
 798
 799         if (blkvsc_req->sector_start > 0xffffffff) {
 800                 if (rq_data_dir(blkvsc_req->req)) {
 801                         blkvsc_req->write = 1;
 802                         blkvsc_req->cmnd[0] = WRITE_16;
 803                 } else {
 804                         blkvsc_req->write = 0;
 805                         blkvsc_req->cmnd[0] = READ_16;
 806                 }
 807
 808                 blkvsc_req->cmnd[1] |=
 809                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 810
 811                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 812                                 cpu_to_be64(blkvsc_req->sector_start);
 813                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 814                                 cpu_to_be32(blkvsc_req->sector_count);
 815         } else if ((blkvsc_req->sector_count > 0xff) ||
 816                    (blkvsc_req->sector_start > 0x1fffff)) {
 817                 if (rq_data_dir(blkvsc_req->req)) {
 818                         blkvsc_req->write = 1;
 819                         blkvsc_req->cmnd[0] = WRITE_10;
 820                 } else {
 821                         blkvsc_req->write = 0;
 822                         blkvsc_req->cmnd[0] = READ_10;
 823                 }
 824
 825                 blkvsc_req->cmnd[1] |=
 826                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 827
 828                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 829                                 cpu_to_be32(blkvsc_req->sector_start);
 830                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 831                                 cpu_to_be16(blkvsc_req->sector_count);
 832         } else {
 833                 if (rq_data_dir(blkvsc_req->req)) {
 834                         blkvsc_req->write = 1;
 835                         blkvsc_req->cmnd[0] = WRITE_6;
 836                 } else {
 837                         blkvsc_req->write = 0;
 838                         blkvsc_req->cmnd[0] = READ_6;
 839                 }
 840
 841                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 842                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 843                 blkvsc_req->cmnd[1] &= 0x1f;
 844                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 845         }
 846 }
 847
 848 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 849                         void (*request_completion)(struct hv_storvsc_request *))
 850 {
 851         struct block_device_context *blkdev = blkvsc_req->dev;
 852         struct vm_device *device_ctx = blkdev->device_ctx;
 853         struct driver_context *driver_ctx =
 854                         driver_to_driver_context(device_ctx->device.driver);
 855         struct blkvsc_driver_context *blkvsc_drv_ctx =
 856                         (struct blkvsc_driver_context *)driver_ctx;
 857         struct storvsc_driver_object *storvsc_drv_obj =
 858                         &blkvsc_drv_ctx->drv_obj;
 859         struct hv_storvsc_request *storvsc_req;
 860         int ret;
 861
 862         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 863                    "req %p type %s start_sector %lu count %ld offset %d "
 864                    "len %d\n", blkvsc_req,
 865                    (blkvsc_req->write) ? "WRITE" : "READ",
 866                    (unsigned long) blkvsc_req->sector_start,
 867                    blkvsc_req->sector_count,
 868                    blkvsc_req->request.DataBuffer.Offset,
 869                    blkvsc_req->request.DataBuffer.Length);
 870 #if 0
 871         for (i = 0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++) {
 872                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 873                            "req %p pfn[%d] %llx\n",
 874                            blkvsc_req, i,
 875                            blkvsc_req->request.DataBuffer.PfnArray[i]);
 876         }
 877 #endif
 878
 879         storvsc_req = &blkvsc_req->request;
 880         storvsc_req->Extension = (void *)((unsigned long)blkvsc_req +
 881                                           sizeof(struct blkvsc_request));
 882
 883         storvsc_req->Type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 884
 885         storvsc_req->OnIOCompletion = request_completion;
 886         storvsc_req->Context = blkvsc_req;
 887
 888         storvsc_req->Host = blkdev->port;
 889         storvsc_req->Bus = blkdev->path;
 890         storvsc_req->TargetId = blkdev->target;
 891         storvsc_req->LunId = 0;  /* this is not really used at all */
 892
 893         storvsc_req->CdbLen = blkvsc_req->cmd_len;
 894         storvsc_req->Cdb = blkvsc_req->cmnd;
 895
 896         storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
 897         storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
 898
 899         ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj,
 900                                            &blkvsc_req->request);
 901         if (ret == 0)
 902                 blkdev->num_outstanding_reqs++;
 903
 904         return ret;
 905 }
 906
 907 /*
 908  * We break the request into 1 or more blkvsc_requests and submit
 909  * them.  If we cant submit them all, we put them on the
 910  * pending_list. The blkvsc_request() will work on the pending_list.
 911  */
 912 static int blkvsc_do_request(struct block_device_context *blkdev,
 913                              struct request *req)
 914 {
 915         struct bio *bio = NULL;
 916         struct bio_vec *bvec = NULL;
 917         struct bio_vec *prev_bvec = NULL;
 918         struct blkvsc_request *blkvsc_req = NULL;
 919         struct blkvsc_request *tmp;
 920         int databuf_idx = 0;
 921         int seg_idx = 0;
 922         sector_t start_sector;
 923         unsigned long num_sectors = 0;
 924         int ret = 0;
 925         int pending = 0;
 926         struct blkvsc_request_group *group = NULL;
 927
 928         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 929                   (unsigned long)blk_rq_pos(req));
 930
 931         /* Create a group to tie req to list of blkvsc_reqs */
 932         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 933         if (!group)
 934                 return -ENOMEM;
 935
 936         INIT_LIST_HEAD(&group->blkvsc_req_list);
 937         group->outstanding = group->status = 0;
 938
 939         start_sector = blk_rq_pos(req);
 940
 941         /* foreach bio in the request */
 942         if (req->bio) {
 943                 for (bio = req->bio; bio; bio = bio->bi_next) {
 944                         /*
 945                          * Map this bio into an existing or new storvsc request
 946                          */
 947                         bio_for_each_segment(bvec, bio, seg_idx) {
 948                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 949                                            "- req %p bio %p bvec %p seg_idx %d "
 950                                            "databuf_idx %d\n", req, bio, bvec,
 951                                            seg_idx, databuf_idx);
 952
 953                                 /* Get a new storvsc request */
 954                                 /* 1st-time */
 955                                 if ((!blkvsc_req) ||
 956                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 957                                     /* hole at the begin of page */
 958                                     || (bvec->bv_offset != 0) ||
 959                                     /* hold at the end of page */
 960                                     (prev_bvec &&
 961                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 962                                         /* submit the prev one */
 963                                         if (blkvsc_req) {
 964                                                 blkvsc_req->sector_start = start_sector;
 965                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 966
 967                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 968                                                 blkvsc_init_rw(blkvsc_req);
 969                                         }
 970
 971                                         /*
 972                                          * Create new blkvsc_req to represent
 973                                          * the current bvec
 974                                          */
 975                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 976                                         if (!blkvsc_req) {
 977                                                 /* free up everything */
 978                                                 list_for_each_entry_safe(
 979                                                         blkvsc_req, tmp,
 980                                                         &group->blkvsc_req_list,
 981                                                         req_entry) {
 982                                                         list_del(&blkvsc_req->req_entry);
 983                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 984                                                 }
 985
 986                                                 kmem_cache_free(blkdev->request_pool, group);
 987                                                 return -ENOMEM;
 988                                         }
 989
 990                                         memset(blkvsc_req, 0,
 991                                                sizeof(struct blkvsc_request));
 992
 993                                         blkvsc_req->dev = blkdev;
 994                                         blkvsc_req->req = req;
 995                                         blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
 996                                         blkvsc_req->request.DataBuffer.Length = 0;
 997
 998                                         /* Add to the group */
 999                                         blkvsc_req->group = group;
1000                                         blkvsc_req->group->outstanding++;
1001                                         list_add_tail(&blkvsc_req->req_entry,
1002                                                 &blkvsc_req->group->blkvsc_req_list);
1003
1004                                         start_sector += num_sectors;
1005                                         num_sectors = 0;
1006                                         databuf_idx = 0;
1007                                 }
1008
1009                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1010                                 blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
1011                                 blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
1012
1013                                 prev_bvec = bvec;
1014
1015                                 databuf_idx++;
1016                                 num_sectors += bvec->bv_len >> 9;
1017
1018                         } /* bio_for_each_segment */
1019
1020                 } /* rq_for_each_bio */
1021         }
1022
1023         /* Handle the last one */
1024         if (blkvsc_req) {
1025                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1026                            blkdev, req, blkvsc_req->group,
1027                            blkvsc_req->group->outstanding);
1028
1029                 blkvsc_req->sector_start = start_sector;
1030                 sector_div(blkvsc_req->sector_start,
1031                            (blkdev->sector_size >> 9));
1032
1033                 blkvsc_req->sector_count = num_sectors /
1034                                            (blkdev->sector_size >> 9);
1035
1036                 blkvsc_init_rw(blkvsc_req);
1037         }
1038
1039         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1040                 if (pending) {
1041                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1042                                    "pending_list - blkvsc_req %p start_sect %lu"
1043                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1044                                    (unsigned long)blkvsc_req->sector_start,
1045                                    blkvsc_req->sector_count,
1046                                    (unsigned long)start_sector,
1047                                    (unsigned long)num_sectors);
1048
1049                         list_add_tail(&blkvsc_req->pend_entry,
1050                                       &blkdev->pending_list);
1051                 } else {
1052                         ret = blkvsc_submit_request(blkvsc_req,
1053                                                     blkvsc_request_completion);
1054                         if (ret == -1) {
1055                                 pending = 1;
1056                                 list_add_tail(&blkvsc_req->pend_entry,
1057                                               &blkdev->pending_list);
1058                         }
1059
1060                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1061                                    "start_sect %lu sect_count %ld (%lu %ld) "
1062                                    "ret %d\n", blkvsc_req,
1063                                    (unsigned long)blkvsc_req->sector_start,
1064                                    blkvsc_req->sector_count,
1065                                    (unsigned long)start_sector,
1066                                    num_sectors, ret);
1067                 }
1068         }
1069
1070         return pending;
1071 }
1072
1073 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1074 {
1075         struct blkvsc_request *blkvsc_req =
1076                         (struct blkvsc_request *)request->Context;
1077         struct block_device_context *blkdev =
1078                         (struct block_device_context *)blkvsc_req->dev;
1079         struct scsi_sense_hdr sense_hdr;
1080
1081         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1082                    blkvsc_req);
1083
1084         blkdev->num_outstanding_reqs--;
1085
1086         if (blkvsc_req->request.Status)
1087                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1088                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1089                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1090
1091         blkvsc_req->cond = 1;
1092         wake_up_interruptible(&blkvsc_req->wevent);
1093 }
1094
1095 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1096 {
1097         struct blkvsc_request *blkvsc_req =
1098                         (struct blkvsc_request *)request->Context;
1099         struct block_device_context *blkdev =
1100                         (struct block_device_context *)blkvsc_req->dev;
1101         unsigned long flags;
1102         struct blkvsc_request *comp_req, *tmp;
1103
1104         /* ASSERT(blkvsc_req->group); */
1105
1106         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1107                    "sect_start %lu sect_count %ld len %d group outstd %d "
1108                    "total outstd %d\n",
1109                    blkdev, blkvsc_req, blkvsc_req->group,
1110                    (blkvsc_req->write) ? "WRITE" : "READ",
1111                    (unsigned long)blkvsc_req->sector_start,
1112                    blkvsc_req->sector_count,
1113                    blkvsc_req->request.DataBuffer.Length,
1114                    blkvsc_req->group->outstanding,
1115                    blkdev->num_outstanding_reqs);
1116
1117         spin_lock_irqsave(&blkdev->lock, flags);
1118
1119         blkdev->num_outstanding_reqs--;
1120         blkvsc_req->group->outstanding--;
1121
1122         /*
1123          * Only start processing when all the blkvsc_reqs are
1124          * completed. This guarantees no out-of-order blkvsc_req
1125          * completion when calling end_that_request_first()
1126          */
1127         if (blkvsc_req->group->outstanding == 0) {
1128                 list_for_each_entry_safe(comp_req, tmp,
1129                                          &blkvsc_req->group->blkvsc_req_list,
1130                                          req_entry) {
1131                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1132                                    "sect_start %lu sect_count %ld\n",
1133                                    comp_req,
1134                                    (unsigned long)comp_req->sector_start,
1135                                    comp_req->sector_count);
1136
1137                         list_del(&comp_req->req_entry);
1138
1139                         if (!__blk_end_request(comp_req->req,
1140                                 (!comp_req->request.Status ? 0 : -EIO),
1141                                 comp_req->sector_count * blkdev->sector_size)) {
1142                                 /*
1143                                  * All the sectors have been xferred ie the
1144                                  * request is done
1145                                  */
1146                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1147                                            comp_req->req);
1148                                 kmem_cache_free(blkdev->request_pool,
1149                                                 comp_req->group);
1150                         }
1151
1152                         kmem_cache_free(blkdev->request_pool, comp_req);
1153                 }
1154
1155                 if (!blkdev->shutting_down) {
1156                         blkvsc_do_pending_reqs(blkdev);
1157                         blk_start_queue(blkdev->gd->queue);
1158                         blkvsc_request(blkdev->gd->queue);
1159                 }
1160         }
1161
1162         spin_unlock_irqrestore(&blkdev->lock, flags);
1163 }
1164
1165 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1166 {
1167         struct blkvsc_request *pend_req, *tmp;
1168         struct blkvsc_request *comp_req, *tmp2;
1169
1170         int ret = 0;
1171
1172         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1173
1174         /* Flush the pending list first */
1175         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1176                                  pend_entry) {
1177                 /*
1178                  * The pend_req could be part of a partially completed
1179                  * request. If so, complete those req first until we
1180                  * hit the pend_req
1181                  */
1182                 list_for_each_entry_safe(comp_req, tmp2,
1183                                          &pend_req->group->blkvsc_req_list,
1184                                          req_entry) {
1185                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1186                                    "sect_start %lu sect_count %ld\n",
1187                                    comp_req,
1188                                    (unsigned long) comp_req->sector_start,
1189                                    comp_req->sector_count);
1190
1191                         if (comp_req == pend_req)
1192                                 break;
1193
1194                         list_del(&comp_req->req_entry);
1195
1196                         if (comp_req->req) {
1197                                 ret = __blk_end_request(comp_req->req,
1198                                         (!comp_req->request.Status ? 0 : -EIO),
1199                                         comp_req->sector_count *
1200                                         blkdev->sector_size);
1201
1202                                 /* FIXME: shouldn't this do more than return? */
1203                                 if (ret)
1204                                         goto out;
1205                         }
1206
1207                         kmem_cache_free(blkdev->request_pool, comp_req);
1208                 }
1209
1210                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1211                            pend_req);
1212
1213                 list_del(&pend_req->pend_entry);
1214
1215                 list_del(&pend_req->req_entry);
1216
1217                 if (comp_req->req) {
1218                         if (!__blk_end_request(pend_req->req, -EIO,
1219                                                pend_req->sector_count *
1220                                                blkdev->sector_size)) {
1221                                 /*
1222                                  * All the sectors have been xferred ie the
1223                                  * request is done
1224                                  */
1225                                 DPRINT_DBG(BLKVSC_DRV,
1226                                            "blkvsc_cancel_pending_reqs() - "
1227                                            "req %p COMPLETED\n", pend_req->req);
1228                                 kmem_cache_free(blkdev->request_pool,
1229                                                 pend_req->group);
1230                         }
1231                 }
1232
1233                 kmem_cache_free(blkdev->request_pool, pend_req);
1234         }
1235
1236 out:
1237         return ret;
1238 }
1239
1240 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1241 {
1242         struct blkvsc_request *pend_req, *tmp;
1243         int ret = 0;
1244
1245         /* Flush the pending list first */
1246         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1247                                  pend_entry) {
1248                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1249                            pend_req);
1250
1251                 ret = blkvsc_submit_request(pend_req,
1252                                             blkvsc_request_completion);
1253                 if (ret != 0)
1254                         break;
1255                 else
1256                         list_del(&pend_req->pend_entry);
1257         }
1258
1259         return ret;
1260 }
1261
1262 static void blkvsc_request(struct request_queue *queue)
1263 {
1264         struct block_device_context *blkdev = NULL;
1265         struct request *req;
1266         int ret = 0;
1267
1268         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1269         while ((req = blk_peek_request(queue)) != NULL) {
1270                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1271
1272                 blkdev = req->rq_disk->private_data;
1273                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1274                     blkdev->media_not_present) {
1275                         __blk_end_request_cur(req, 0);
1276                         continue;
1277                 }
1278
1279                 ret = blkvsc_do_pending_reqs(blkdev);
1280
1281                 if (ret != 0) {
1282                         DPRINT_DBG(BLKVSC_DRV,
1283                                    "- stop queue - pending_list not empty\n");
1284                         blk_stop_queue(queue);
1285                         break;
1286                 }
1287
1288                 blk_start_request(req);
1289
1290                 ret = blkvsc_do_request(blkdev, req);
1291                 if (ret > 0) {
1292                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1293                         blk_stop_queue(queue);
1294                         break;
1295                 } else if (ret < 0) {
1296                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1297                         blk_requeue_request(queue, req);
1298                         blk_stop_queue(queue);
1299                         break;
1300                 }
1301         }
1302 }
1303
1304 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1305 {
1306         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1307
1308         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1309                    blkdev->gd->disk_name);
1310
1311         mutex_lock(&blkvsc_mutex);
1312         spin_lock(&blkdev->lock);
1313
1314         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1315                 spin_unlock(&blkdev->lock);
1316                 check_disk_change(bdev);
1317                 spin_lock(&blkdev->lock);
1318         }
1319
1320         blkdev->users++;
1321
1322         spin_unlock(&blkdev->lock);
1323         mutex_unlock(&blkvsc_mutex);
1324         return 0;
1325 }
1326
1327 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1328 {
1329         struct block_device_context *blkdev = disk->private_data;
1330
1331         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1332                    blkdev->gd->disk_name);
1333
1334         mutex_lock(&blkvsc_mutex);
1335         spin_lock(&blkdev->lock);
1336         if (blkdev->users == 1) {
1337                 spin_unlock(&blkdev->lock);
1338                 blkvsc_do_flush(blkdev);
1339                 spin_lock(&blkdev->lock);
1340         }
1341
1342         blkdev->users--;
1343
1344         spin_unlock(&blkdev->lock);
1345         mutex_unlock(&blkvsc_mutex);
1346         return 0;
1347 }
1348
1349 static int blkvsc_media_changed(struct gendisk *gd)
1350 {
1351         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1352         return 1;
1353 }
1354
1355 static int blkvsc_revalidate_disk(struct gendisk *gd)
1356 {
1357         struct block_device_context *blkdev = gd->private_data;
1358
1359         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1360
1361         if (blkdev->device_type == DVD_TYPE) {
1362                 blkvsc_do_read_capacity(blkdev);
1363                 set_capacity(blkdev->gd, blkdev->capacity *
1364                             (blkdev->sector_size/512));
1365                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1366         }
1367         return 0;
1368 }
1369
1370 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1371 {
1372         sector_t total_sectors = get_capacity(bd->bd_disk);
1373         sector_t cylinder_times_heads = 0;
1374         sector_t temp = 0;
1375
1376         int sectors_per_track = 0;
1377         int heads = 0;
1378         int cylinders = 0;
1379         int rem = 0;
1380
1381         if (total_sectors > (65535 * 16 * 255))
1382                 total_sectors = (65535 * 16 * 255);
1383
1384         if (total_sectors >= (65535 * 16 * 63)) {
1385                 sectors_per_track = 255;
1386                 heads = 16;
1387
1388                 cylinder_times_heads = total_sectors;
1389                 /* sector_div stores the quotient in cylinder_times_heads */
1390                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1391         } else {
1392                 sectors_per_track = 17;
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
1398                 temp = cylinder_times_heads + 1023;
1399                 /* sector_div stores the quotient in temp */
1400                 rem = sector_div(temp, 1024);
1401
1402                 heads = temp;
1403
1404                 if (heads < 4)
1405                         heads = 4;
1406
1407
1408                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1409                         sectors_per_track = 31;
1410                         heads = 16;
1411
1412                         cylinder_times_heads = total_sectors;
1413                         /*
1414                          * sector_div stores the quotient in
1415                          * cylinder_times_heads
1416                          */
1417                         rem = sector_div(cylinder_times_heads,
1418                                          sectors_per_track);
1419                 }
1420
1421                 if (cylinder_times_heads >= (heads * 1024)) {
1422                         sectors_per_track = 63;
1423                         heads = 16;
1424
1425                         cylinder_times_heads = total_sectors;
1426                         /*
1427                          * sector_div stores the quotient in
1428                          * cylinder_times_heads
1429                          */
1430                         rem = sector_div(cylinder_times_heads,
1431                                          sectors_per_track);
1432                 }
1433         }
1434
1435         temp = cylinder_times_heads;
1436         /* sector_div stores the quotient in temp */
1437         rem = sector_div(temp, heads);
1438         cylinders = temp;
1439
1440         hg->heads = heads;
1441         hg->sectors = sectors_per_track;
1442         hg->cylinders = cylinders;
1443
1444         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1445                     sectors_per_track);
1446
1447     return 0;
1448 }
1449
1450 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1451                         unsigned cmd, unsigned long argument)
1452 {
1453 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1454         int ret;
1455
1456         switch (cmd) {
1457         /*
1458          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1459          * than just a GUID. Commented it out for now.
1460          */
1461 #if 0
1462         case HDIO_GET_IDENTITY:
1463                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1464                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1465                                  blkdev->device_id_len))
1466                         ret = -EFAULT;
1467                 break;
1468 #endif
1469         default:
1470                 ret = -EINVAL;
1471                 break;
1472         }
1473
1474         return ret;
1475 }
1476
1477 static int __init blkvsc_init(void)
1478 {
1479         int ret;
1480
1481         BUILD_BUG_ON(sizeof(sector_t) != 8);
1482
1483         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1484
1485         ret = blkvsc_drv_init(BlkVscInitialize);
1486
1487         return ret;
1488 }
1489
1490 static void __exit blkvsc_exit(void)
1491 {
1492         blkvsc_drv_exit();
1493 }
1494
1495 MODULE_LICENSE("GPL");
1496 MODULE_VERSION(HV_DRV_VERSION);
1497 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1498 module_init(blkvsc_init);
1499 module_exit(blkvsc_exit);