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 "vmbus.h"
  35 #include "StorVscApi.h"
  36
  37
  38 #define BLKVSC_MINORS   64
  39
  40 enum blkvsc_device_type {
  41         UNKNOWN_DEV_TYPE,
  42         HARDDISK_TYPE,
  43         DVD_TYPE,
  44 };
  45
  46 /*
  47  * This request ties the struct request and struct
  48  * blkvsc_request/hv_storvsc_request together A struct request may be
  49  * represented by 1 or more struct blkvsc_request
  50  */
  51 struct blkvsc_request_group {
  52         int outstanding;
  53         int status;
  54         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  55 };
  56
  57 struct blkvsc_request {
  58         /* blkvsc_request_group.blkvsc_req_list */
  59         struct list_head req_entry;
  60
  61         /* block_device_context.pending_list */
  62         struct list_head pend_entry;
  63
  64         /* This may be null if we generate a request internally */
  65         struct request *req;
  66
  67         struct block_device_context *dev;
  68
  69         /* The group this request is part of. Maybe null */
  70         struct blkvsc_request_group *group;
  71
  72         wait_queue_head_t wevent;
  73         int cond;
  74
  75         int write;
  76         sector_t sector_start;
  77         unsigned long sector_count;
  78
  79         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  80         unsigned char cmd_len;
  81         unsigned char cmnd[MAX_COMMAND_SIZE];
  82
  83         struct hv_storvsc_request request;
  84         /*
  85          * !!!DO NOT ADD ANYTHING BELOW HERE!!! Otherwise, memory can overlap,
  86          * because - The extension buffer falls right here and is pointed to by
  87          * request.Extension;
  88          * Which sounds like a horrible idea, who designed this?
  89          */
  90 };
  91
  92 /* Per device structure */
  93 struct block_device_context {
  94         /* point back to our device context */
  95         struct device_context *device_ctx;
  96         struct kmem_cache *request_pool;
  97         spinlock_t lock;
  98         struct gendisk *gd;
  99         enum blkvsc_device_type device_type;
 100         struct list_head pending_list;
 101
 102         unsigned char device_id[64];
 103         unsigned int device_id_len;
 104         int num_outstanding_reqs;
 105         int shutting_down;
 106         int media_not_present;
 107         unsigned int sector_size;
 108         sector_t capacity;
 109         unsigned int port;
 110         unsigned char path;
 111         unsigned char target;
 112         int users;
 113 };
 114
 115 /* Per driver */
 116 struct blkvsc_driver_context {
 117         /* !! These must be the first 2 fields !! */
 118         /* FIXME this is a bug! */
 119         struct driver_context drv_ctx;
 120         struct storvsc_driver_object drv_obj;
 121 };
 122
 123 /* Static decl */
 124 static int blkvsc_probe(struct device *dev);
 125 static int blkvsc_remove(struct device *device);
 126 static void blkvsc_shutdown(struct device *device);
 127
 128 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 129 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 130 static int blkvsc_media_changed(struct gendisk *gd);
 131 static int blkvsc_revalidate_disk(struct gendisk *gd);
 132 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 133 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 134                         unsigned cmd, unsigned long argument);
 135 static void blkvsc_request(struct request_queue *queue);
 136 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 137 static int blkvsc_do_request(struct block_device_context *blkdev,
 138                              struct request *req);
 139 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 140                 void (*request_completion)(struct hv_storvsc_request *));
 141 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 142 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 143 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 144 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 145 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 146 static int blkvsc_do_flush(struct block_device_context *blkdev);
 147 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 148 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 149
 150
 151 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 152
 153 /* The one and only one */
 154 static struct blkvsc_driver_context g_blkvsc_drv;
 155
 156 static struct block_device_operations block_ops = {
 157         .owner = THIS_MODULE,
 158         .open = blkvsc_open,
 159         .release = blkvsc_release,
 160         .media_changed = blkvsc_media_changed,
 161         .revalidate_disk = blkvsc_revalidate_disk,
 162         .getgeo = blkvsc_getgeo,
 163         .ioctl  = blkvsc_ioctl,
 164 };
 165
 166 /**
 167  * blkvsc_drv_init -  BlkVsc driver initialization.
 168  */
 169 static int blkvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 170 {
 171         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 172         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 173         int ret;
 174
 175         DPRINT_ENTER(BLKVSC_DRV);
 176
 177         vmbus_get_interface(&storvsc_drv_obj->Base.VmbusChannelInterface);
 178
 179         storvsc_drv_obj->RingBufferSize = blkvsc_ringbuffer_size;
 180
 181         /* Callback to client driver to complete the initialization */
 182         drv_init(&storvsc_drv_obj->Base);
 183
 184         drv_ctx->driver.name = storvsc_drv_obj->Base.name;
 185         memcpy(&drv_ctx->class_id, &storvsc_drv_obj->Base.deviceType,
 186                sizeof(struct hv_guid));
 187
 188         drv_ctx->probe = blkvsc_probe;
 189         drv_ctx->remove = blkvsc_remove;
 190         drv_ctx->shutdown = blkvsc_shutdown;
 191
 192         /* The driver belongs to vmbus */
 193         ret = vmbus_child_driver_register(drv_ctx);
 194
 195         DPRINT_EXIT(BLKVSC_DRV);
 196
 197         return ret;
 198 }
 199
 200 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 201 {
 202         struct device **curr = (struct device **)data;
 203         *curr = dev;
 204         return 1; /* stop iterating */
 205 }
 206
 207 static void blkvsc_drv_exit(void)
 208 {
 209         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv.drv_obj;
 210         struct driver_context *drv_ctx = &g_blkvsc_drv.drv_ctx;
 211         struct device *current_dev;
 212         int ret;
 213
 214         DPRINT_ENTER(BLKVSC_DRV);
 215
 216         while (1) {
 217                 current_dev = NULL;
 218
 219                 /* Get the device */
 220                 ret = driver_for_each_device(&drv_ctx->driver, NULL,
 221                                              (void *) &current_dev,
 222                                              blkvsc_drv_exit_cb);
 223
 224                 if (ret)
 225                         DPRINT_WARN(BLKVSC_DRV,
 226                                     "driver_for_each_device returned %d", ret);
 227
 228
 229                 if (current_dev == NULL)
 230                         break;
 231
 232                 /* Initiate removal from the top-down */
 233                 device_unregister(current_dev);
 234         }
 235
 236         if (storvsc_drv_obj->Base.OnCleanup)
 237                 storvsc_drv_obj->Base.OnCleanup(&storvsc_drv_obj->Base);
 238
 239         vmbus_child_driver_unregister(drv_ctx);
 240
 241         DPRINT_EXIT(BLKVSC_DRV);
 242
 243         return;
 244 }
 245
 246 /**
 247  * blkvsc_probe - Add a new device for this driver
 248  */
 249 static int blkvsc_probe(struct device *device)
 250 {
 251         struct driver_context *driver_ctx =
 252                                 driver_to_driver_context(device->driver);
 253         struct blkvsc_driver_context *blkvsc_drv_ctx =
 254                                 (struct blkvsc_driver_context *)driver_ctx;
 255         struct storvsc_driver_object *storvsc_drv_obj =
 256                                 &blkvsc_drv_ctx->drv_obj;
 257         struct device_context *device_ctx = device_to_device_context(device);
 258         struct hv_device *device_obj = &device_ctx->device_obj;
 259
 260         struct block_device_context *blkdev = NULL;
 261         struct storvsc_device_info device_info;
 262         int major = 0;
 263         int devnum = 0;
 264         int ret = 0;
 265         static int ide0_registered;
 266         static int ide1_registered;
 267
 268         DPRINT_ENTER(BLKVSC_DRV);
 269
 270         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 271
 272         if (!storvsc_drv_obj->Base.OnDeviceAdd) {
 273                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 274                 ret = -1;
 275                 goto Cleanup;
 276         }
 277
 278         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 279         if (!blkdev) {
 280                 ret = -ENOMEM;
 281                 goto Cleanup;
 282         }
 283
 284         INIT_LIST_HEAD(&blkdev->pending_list);
 285
 286         /* Initialize what we can here */
 287         spin_lock_init(&blkdev->lock);
 288
 289         ASSERT(sizeof(struct blkvsc_request_group) <=
 290                 sizeof(struct blkvsc_request));
 291
 292         blkdev->request_pool = kmem_cache_create(dev_name(&device_ctx->device),
 293                                         sizeof(struct blkvsc_request) +
 294                                         storvsc_drv_obj->RequestExtSize, 0,
 295                                         SLAB_HWCACHE_ALIGN, NULL);
 296         if (!blkdev->request_pool) {
 297                 ret = -ENOMEM;
 298                 goto Cleanup;
 299         }
 300
 301
 302         /* Call to the vsc driver to add the device */
 303         ret = storvsc_drv_obj->Base.OnDeviceAdd(device_obj, &device_info);
 304         if (ret != 0) {
 305                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 306                 goto Cleanup;
 307         }
 308
 309         blkdev->device_ctx = device_ctx;
 310         /* this identified the device 0 or 1 */
 311         blkdev->target = device_info.TargetId;
 312         /* this identified the ide ctrl 0 or 1 */
 313         blkdev->path = device_info.PathId;
 314
 315         dev_set_drvdata(device, blkdev);
 316
 317         /* Calculate the major and device num */
 318         if (blkdev->path == 0) {
 319                 major = IDE0_MAJOR;
 320                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 321
 322                 if (!ide0_registered) {
 323                         ret = register_blkdev(major, "ide");
 324                         if (ret != 0) {
 325                                 DPRINT_ERR(BLKVSC_DRV,
 326                                            "register_blkdev() failed! ret %d",
 327                                            ret);
 328                                 goto Remove;
 329                         }
 330
 331                         ide0_registered = 1;
 332                 }
 333         } else if (blkdev->path == 1) {
 334                 major = IDE1_MAJOR;
 335                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 336
 337                 if (!ide1_registered) {
 338                         ret = register_blkdev(major, "ide");
 339                         if (ret != 0) {
 340                                 DPRINT_ERR(BLKVSC_DRV,
 341                                            "register_blkdev() failed! ret %d",
 342                                            ret);
 343                                 goto Remove;
 344                         }
 345
 346                         ide1_registered = 1;
 347                 }
 348         } else {
 349                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 350                 ret = -1;
 351                 goto Cleanup;
 352         }
 353
 354         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 355
 356         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 357         if (!blkdev->gd) {
 358                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 359                 ret = -1;
 360                 goto Cleanup;
 361         }
 362
 363         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 364
 365         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 366         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 367         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 368         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 369         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 370
 371         blkdev->gd->major = major;
 372         if (devnum == 1 || devnum == 3)
 373                 blkdev->gd->first_minor = BLKVSC_MINORS;
 374         else
 375                 blkdev->gd->first_minor = 0;
 376         blkdev->gd->fops = &block_ops;
 377         blkdev->gd->private_data = blkdev;
 378         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 379
 380         blkvsc_do_inquiry(blkdev);
 381         if (blkdev->device_type == DVD_TYPE) {
 382                 set_disk_ro(blkdev->gd, 1);
 383                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 384                 blkvsc_do_read_capacity(blkdev);
 385         } else {
 386                 blkvsc_do_read_capacity16(blkdev);
 387         }
 388
 389         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 390         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 391         /* go! */
 392         add_disk(blkdev->gd);
 393
 394         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 395                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 396                     blkdev->sector_size);
 397
 398         return ret;
 399
 400 Remove:
 401         storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 402
 403 Cleanup:
 404         if (blkdev) {
 405                 if (blkdev->request_pool) {
 406                         kmem_cache_destroy(blkdev->request_pool);
 407                         blkdev->request_pool = NULL;
 408                 }
 409                 kfree(blkdev);
 410                 blkdev = NULL;
 411         }
 412
 413         DPRINT_EXIT(BLKVSC_DRV);
 414
 415         return ret;
 416 }
 417
 418 static void blkvsc_shutdown(struct device *device)
 419 {
 420         struct block_device_context *blkdev = dev_get_drvdata(device);
 421         unsigned long flags;
 422
 423         if (!blkdev)
 424                 return;
 425
 426         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 427                    blkdev->users, blkdev->gd->disk_name);
 428
 429         spin_lock_irqsave(&blkdev->lock, flags);
 430
 431         blkdev->shutting_down = 1;
 432
 433         blk_stop_queue(blkdev->gd->queue);
 434
 435         spin_unlock_irqrestore(&blkdev->lock, flags);
 436
 437         while (blkdev->num_outstanding_reqs) {
 438                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 439                             blkdev->num_outstanding_reqs);
 440                 udelay(100);
 441         }
 442
 443         blkvsc_do_flush(blkdev);
 444
 445         spin_lock_irqsave(&blkdev->lock, flags);
 446
 447         blkvsc_cancel_pending_reqs(blkdev);
 448
 449         spin_unlock_irqrestore(&blkdev->lock, flags);
 450 }
 451
 452 static int blkvsc_do_flush(struct block_device_context *blkdev)
 453 {
 454         struct blkvsc_request *blkvsc_req;
 455
 456         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 457
 458         if (blkdev->device_type != HARDDISK_TYPE)
 459                 return 0;
 460
 461         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 462         if (!blkvsc_req)
 463                 return -ENOMEM;
 464
 465         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 466         init_waitqueue_head(&blkvsc_req->wevent);
 467         blkvsc_req->dev = blkdev;
 468         blkvsc_req->req = NULL;
 469         blkvsc_req->write = 0;
 470
 471         blkvsc_req->request.DataBuffer.PfnArray[0] = 0;
 472         blkvsc_req->request.DataBuffer.Offset = 0;
 473         blkvsc_req->request.DataBuffer.Length = 0;
 474
 475         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 476         blkvsc_req->cmd_len = 10;
 477
 478         /*
 479          * Set this here since the completion routine may be invoked and
 480          * completed before we return
 481          */
 482         blkvsc_req->cond = 0;
 483         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 484
 485         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 486
 487         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 488
 489         return 0;
 490 }
 491
 492 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 493 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 494 {
 495         struct blkvsc_request *blkvsc_req;
 496         struct page *page_buf;
 497         unsigned char *buf;
 498         unsigned char device_type;
 499
 500         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 501
 502         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 503         if (!blkvsc_req)
 504                 return -ENOMEM;
 505
 506         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 507         page_buf = alloc_page(GFP_KERNEL);
 508         if (!page_buf) {
 509                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 510                 return -ENOMEM;
 511         }
 512
 513         init_waitqueue_head(&blkvsc_req->wevent);
 514         blkvsc_req->dev = blkdev;
 515         blkvsc_req->req = NULL;
 516         blkvsc_req->write = 0;
 517
 518         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 519         blkvsc_req->request.DataBuffer.Offset = 0;
 520         blkvsc_req->request.DataBuffer.Length = 64;
 521
 522         blkvsc_req->cmnd[0] = INQUIRY;
 523         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 524         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 525         blkvsc_req->cmnd[4] = 64;
 526         blkvsc_req->cmd_len = 6;
 527
 528         /*
 529          * Set this here since the completion routine may be invoked and
 530          * completed before we return
 531          */
 532         blkvsc_req->cond = 0;
 533
 534         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 535
 536         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 537                    blkvsc_req, blkvsc_req->cond);
 538
 539         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 540
 541         buf = kmap(page_buf);
 542
 543         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 544         /* be to le */
 545         device_type = buf[0] & 0x1F;
 546
 547         if (device_type == 0x0) {
 548                 blkdev->device_type = HARDDISK_TYPE;
 549         } else if (device_type == 0x5) {
 550                 blkdev->device_type = DVD_TYPE;
 551         } else {
 552                 /* TODO: this is currently unsupported device type */
 553                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 554         }
 555
 556         DPRINT_DBG(BLKVSC_DRV, "device type %d \n", device_type);
 557
 558         blkdev->device_id_len = buf[7];
 559         if (blkdev->device_id_len > 64)
 560                 blkdev->device_id_len = 64;
 561
 562         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 563         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 564          * blkdev->device_id_len); */
 565
 566         kunmap(page_buf);
 567
 568         __free_page(page_buf);
 569
 570         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 571
 572         return 0;
 573 }
 574
 575 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 576 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 577 {
 578         struct blkvsc_request *blkvsc_req;
 579         struct page *page_buf;
 580         unsigned char *buf;
 581         struct scsi_sense_hdr sense_hdr;
 582
 583         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 584
 585         blkdev->sector_size = 0;
 586         blkdev->capacity = 0;
 587         blkdev->media_not_present = 0; /* assume a disk is present */
 588
 589         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 590         if (!blkvsc_req)
 591                 return -ENOMEM;
 592
 593         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 594         page_buf = alloc_page(GFP_KERNEL);
 595         if (!page_buf) {
 596                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 597                 return -ENOMEM;
 598         }
 599
 600         init_waitqueue_head(&blkvsc_req->wevent);
 601         blkvsc_req->dev = blkdev;
 602         blkvsc_req->req = NULL;
 603         blkvsc_req->write = 0;
 604
 605         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 606         blkvsc_req->request.DataBuffer.Offset = 0;
 607         blkvsc_req->request.DataBuffer.Length = 8;
 608
 609         blkvsc_req->cmnd[0] = READ_CAPACITY;
 610         blkvsc_req->cmd_len = 16;
 611
 612         /*
 613          * Set this here since the completion routine may be invoked
 614          * and completed before we return
 615          */
 616         blkvsc_req->cond = 0;
 617
 618         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 619
 620         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 621                    blkvsc_req, blkvsc_req->cond);
 622
 623         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 624
 625         /* check error */
 626         if (blkvsc_req->request.Status) {
 627                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 628                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 629
 630                 if (sense_hdr.asc == 0x3A) {
 631                         /* Medium not present */
 632                         blkdev->media_not_present = 1;
 633                 }
 634                 return 0;
 635         }
 636         buf = kmap(page_buf);
 637
 638         /* be to le */
 639         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 640                             (buf[2] << 8) | buf[3]) + 1;
 641         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 642                               (buf[6] << 8) | buf[7];
 643
 644         kunmap(page_buf);
 645
 646         __free_page(page_buf);
 647
 648         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 649
 650         return 0;
 651 }
 652
 653 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 654 {
 655         struct blkvsc_request *blkvsc_req;
 656         struct page *page_buf;
 657         unsigned char *buf;
 658         struct scsi_sense_hdr sense_hdr;
 659
 660         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 661
 662         blkdev->sector_size = 0;
 663         blkdev->capacity = 0;
 664         blkdev->media_not_present = 0; /* assume a disk is present */
 665
 666         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_KERNEL);
 667         if (!blkvsc_req)
 668                 return -ENOMEM;
 669
 670         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 671         page_buf = alloc_page(GFP_KERNEL);
 672         if (!page_buf) {
 673                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 674                 return -ENOMEM;
 675         }
 676
 677         init_waitqueue_head(&blkvsc_req->wevent);
 678         blkvsc_req->dev = blkdev;
 679         blkvsc_req->req = NULL;
 680         blkvsc_req->write = 0;
 681
 682         blkvsc_req->request.DataBuffer.PfnArray[0] = page_to_pfn(page_buf);
 683         blkvsc_req->request.DataBuffer.Offset = 0;
 684         blkvsc_req->request.DataBuffer.Length = 12;
 685
 686         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 687         blkvsc_req->cmd_len = 16;
 688
 689         /*
 690          * Set this here since the completion routine may be invoked
 691          * and completed before we return
 692          */
 693         blkvsc_req->cond = 0;
 694
 695         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 696
 697         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 698                    blkvsc_req, blkvsc_req->cond);
 699
 700         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 701
 702         /* check error */
 703         if (blkvsc_req->request.Status) {
 704                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 705                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 706                 if (sense_hdr.asc == 0x3A) {
 707                         /* Medium not present */
 708                         blkdev->media_not_present = 1;
 709                 }
 710                 return 0;
 711         }
 712         buf = kmap(page_buf);
 713
 714         /* be to le */
 715         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 716         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 717
 718 #if 0
 719         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 720                             (buf[2] << 8) | buf[3]) + 1;
 721         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 722                               (buf[6] << 8) | buf[7];
 723 #endif
 724
 725         kunmap(page_buf);
 726
 727         __free_page(page_buf);
 728
 729         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 730
 731         return 0;
 732 }
 733
 734 /**
 735  * blkvsc_remove() - Callback when our device is removed
 736  */
 737 static int blkvsc_remove(struct device *device)
 738 {
 739         struct driver_context *driver_ctx =
 740                                 driver_to_driver_context(device->driver);
 741         struct blkvsc_driver_context *blkvsc_drv_ctx =
 742                                 (struct blkvsc_driver_context *)driver_ctx;
 743         struct storvsc_driver_object *storvsc_drv_obj =
 744                                 &blkvsc_drv_ctx->drv_obj;
 745         struct device_context *device_ctx = device_to_device_context(device);
 746         struct hv_device *device_obj = &device_ctx->device_obj;
 747         struct block_device_context *blkdev = dev_get_drvdata(device);
 748         unsigned long flags;
 749         int ret;
 750
 751         DPRINT_ENTER(BLKVSC_DRV);
 752
 753         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 754
 755         if (!storvsc_drv_obj->Base.OnDeviceRemove) {
 756                 DPRINT_EXIT(BLKVSC_DRV);
 757                 return -1;
 758         }
 759
 760         /*
 761          * Call to the vsc driver to let it know that the device is being
 762          * removed
 763          */
 764         ret = storvsc_drv_obj->Base.OnDeviceRemove(device_obj);
 765         if (ret != 0) {
 766                 /* TODO: */
 767                 DPRINT_ERR(BLKVSC_DRV,
 768                            "unable to remove blkvsc device (ret %d)", ret);
 769         }
 770
 771         /* Get to a known state */
 772         spin_lock_irqsave(&blkdev->lock, flags);
 773
 774         blkdev->shutting_down = 1;
 775
 776         blk_stop_queue(blkdev->gd->queue);
 777
 778         spin_unlock_irqrestore(&blkdev->lock, flags);
 779
 780         while (blkdev->num_outstanding_reqs) {
 781                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 782                             blkdev->num_outstanding_reqs);
 783                 udelay(100);
 784         }
 785
 786         blkvsc_do_flush(blkdev);
 787
 788         spin_lock_irqsave(&blkdev->lock, flags);
 789
 790         blkvsc_cancel_pending_reqs(blkdev);
 791
 792         spin_unlock_irqrestore(&blkdev->lock, flags);
 793
 794         blk_cleanup_queue(blkdev->gd->queue);
 795
 796         del_gendisk(blkdev->gd);
 797
 798         kmem_cache_destroy(blkdev->request_pool);
 799
 800         kfree(blkdev);
 801
 802         DPRINT_EXIT(BLKVSC_DRV);
 803
 804         return ret;
 805 }
 806
 807 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 808 {
 809         ASSERT(blkvsc_req->req);
 810         ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8));
 811
 812         blkvsc_req->cmd_len = 16;
 813
 814         if (blkvsc_req->sector_start > 0xffffffff) {
 815                 if (rq_data_dir(blkvsc_req->req)) {
 816                         blkvsc_req->write = 1;
 817                         blkvsc_req->cmnd[0] = WRITE_16;
 818                 } else {
 819                         blkvsc_req->write = 0;
 820                         blkvsc_req->cmnd[0] = READ_16;
 821                 }
 822
 823                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 824
 825                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 826                                 cpu_to_be64(blkvsc_req->sector_start);
 827                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 828                                 cpu_to_be32(blkvsc_req->sector_count);
 829         } else if ((blkvsc_req->sector_count > 0xff) ||
 830                    (blkvsc_req->sector_start > 0x1fffff)) {
 831                 if (rq_data_dir(blkvsc_req->req)) {
 832                         blkvsc_req->write = 1;
 833                         blkvsc_req->cmnd[0] = WRITE_10;
 834                 } else {
 835                         blkvsc_req->write = 0;
 836                         blkvsc_req->cmnd[0] = READ_10;
 837                 }
 838
 839                 blkvsc_req->cmnd[1] |= blk_fua_rq(blkvsc_req->req) ? 0x8 : 0;
 840
 841                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 842                                 cpu_to_be32(blkvsc_req->sector_start);
 843                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 844                                 cpu_to_be16(blkvsc_req->sector_count);
 845         } else {
 846                 if (rq_data_dir(blkvsc_req->req)) {
 847                         blkvsc_req->write = 1;
 848                         blkvsc_req->cmnd[0] = WRITE_6;
 849                 } else {
 850                         blkvsc_req->write = 0;
 851                         blkvsc_req->cmnd[0] = READ_6;
 852                 }
 853
 854                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 855                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 856                 blkvsc_req->cmnd[1] &= 0x1f;
 857                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 858         }
 859 }
 860
 861 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 862                         void (*request_completion)(struct hv_storvsc_request *))
 863 {
 864         struct block_device_context *blkdev = blkvsc_req->dev;
 865         struct device_context *device_ctx = blkdev->device_ctx;
 866         struct driver_context *driver_ctx =
 867                         driver_to_driver_context(device_ctx->device.driver);
 868         struct blkvsc_driver_context *blkvsc_drv_ctx =
 869                         (struct blkvsc_driver_context *)driver_ctx;
 870         struct storvsc_driver_object *storvsc_drv_obj =
 871                         &blkvsc_drv_ctx->drv_obj;
 872         struct hv_storvsc_request *storvsc_req;
 873         int ret;
 874
 875         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 876                    "req %p type %s start_sector %lu count %ld offset %d "
 877                    "len %d\n", blkvsc_req,
 878                    (blkvsc_req->write) ? "WRITE" : "READ",
 879                    (unsigned long) blkvsc_req->sector_start,
 880                    blkvsc_req->sector_count,
 881                    blkvsc_req->request.DataBuffer.Offset,
 882                    blkvsc_req->request.DataBuffer.Length);
 883 #if 0
 884         for (i = 0; i < (blkvsc_req->request.DataBuffer.Length >> 12); i++) {
 885                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 886                            "req %p pfn[%d] %llx\n",
 887                            blkvsc_req, i,
 888                            blkvsc_req->request.DataBuffer.PfnArray[i]);
 889         }
 890 #endif
 891
 892         storvsc_req = &blkvsc_req->request;
 893         storvsc_req->Extension = (void *)((unsigned long)blkvsc_req +
 894                                           sizeof(struct blkvsc_request));
 895
 896         storvsc_req->Type = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 897
 898         storvsc_req->OnIOCompletion = request_completion;
 899         storvsc_req->Context = blkvsc_req;
 900
 901         storvsc_req->Host = blkdev->port;
 902         storvsc_req->Bus = blkdev->path;
 903         storvsc_req->TargetId = blkdev->target;
 904         storvsc_req->LunId = 0;  /* this is not really used at all */
 905
 906         storvsc_req->CdbLen = blkvsc_req->cmd_len;
 907         storvsc_req->Cdb = blkvsc_req->cmnd;
 908
 909         storvsc_req->SenseBuffer = blkvsc_req->sense_buffer;
 910         storvsc_req->SenseBufferSize = SCSI_SENSE_BUFFERSIZE;
 911
 912         ret = storvsc_drv_obj->OnIORequest(&blkdev->device_ctx->device_obj,
 913                                            &blkvsc_req->request);
 914         if (ret == 0)
 915                 blkdev->num_outstanding_reqs++;
 916
 917         return ret;
 918 }
 919
 920 /*
 921  * We break the request into 1 or more blkvsc_requests and submit
 922  * them.  If we cant submit them all, we put them on the
 923  * pending_list. The blkvsc_request() will work on the pending_list.
 924  */
 925 static int blkvsc_do_request(struct block_device_context *blkdev,
 926                              struct request *req)
 927 {
 928         struct bio *bio = NULL;
 929         struct bio_vec *bvec = NULL;
 930         struct bio_vec *prev_bvec = NULL;
 931         struct blkvsc_request *blkvsc_req = NULL;
 932         struct blkvsc_request *tmp;
 933         int databuf_idx = 0;
 934         int seg_idx = 0;
 935         sector_t start_sector;
 936         unsigned long num_sectors = 0;
 937         int ret = 0;
 938         int pending = 0;
 939         struct blkvsc_request_group *group = NULL;
 940
 941         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu \n", blkdev, req,
 942                   (unsigned long)blk_rq_pos(req));
 943
 944         /* Create a group to tie req to list of blkvsc_reqs */
 945         group = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 946         if (!group)
 947                 return -ENOMEM;
 948
 949         INIT_LIST_HEAD(&group->blkvsc_req_list);
 950         group->outstanding = group->status = 0;
 951
 952         start_sector = blk_rq_pos(req);
 953
 954         /* foreach bio in the request */
 955         if (req->bio) {
 956                 for (bio = req->bio; bio; bio = bio->bi_next) {
 957                         /*
 958                          * Map this bio into an existing or new storvsc request
 959                          */
 960                         bio_for_each_segment(bvec, bio, seg_idx) {
 961                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 962                                            "- req %p bio %p bvec %p seg_idx %d "
 963                                            "databuf_idx %d\n", req, bio, bvec,
 964                                            seg_idx, databuf_idx);
 965
 966                                 /* Get a new storvsc request */
 967                                 /* 1st-time */
 968                                 if ((!blkvsc_req) ||
 969                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 970                                     /* hole at the begin of page */
 971                                     || (bvec->bv_offset != 0) ||
 972                                     /* hold at the end of page */
 973                                     (prev_bvec &&
 974                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 975                                         /* submit the prev one */
 976                                         if (blkvsc_req) {
 977                                                 blkvsc_req->sector_start = start_sector;
 978                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
 979
 980                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
 981                                                 blkvsc_init_rw(blkvsc_req);
 982                                         }
 983
 984                                         /*
 985                                          * Create new blkvsc_req to represent
 986                                          * the current bvec
 987                                          */
 988                                         blkvsc_req = kmem_cache_alloc(blkdev->request_pool, GFP_ATOMIC);
 989                                         if (!blkvsc_req) {
 990                                                 /* free up everything */
 991                                                 list_for_each_entry_safe(
 992                                                         blkvsc_req, tmp,
 993                                                         &group->blkvsc_req_list,
 994                                                         req_entry) {
 995                                                         list_del(&blkvsc_req->req_entry);
 996                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
 997                                                 }
 998
 999                                                 kmem_cache_free(blkdev->request_pool, group);
1000                                                 return -ENOMEM;
1001                                         }
1002
1003                                         memset(blkvsc_req, 0,
1004                                                sizeof(struct blkvsc_request));
1005
1006                                         blkvsc_req->dev = blkdev;
1007                                         blkvsc_req->req = req;
1008                                         blkvsc_req->request.DataBuffer.Offset = bvec->bv_offset;
1009                                         blkvsc_req->request.DataBuffer.Length = 0;
1010
1011                                         /* Add to the group */
1012                                         blkvsc_req->group = group;
1013                                         blkvsc_req->group->outstanding++;
1014                                         list_add_tail(&blkvsc_req->req_entry,
1015                                                 &blkvsc_req->group->blkvsc_req_list);
1016
1017                                         start_sector += num_sectors;
1018                                         num_sectors = 0;
1019                                         databuf_idx = 0;
1020                                 }
1021
1022                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1023                                 blkvsc_req->request.DataBuffer.PfnArray[databuf_idx] = page_to_pfn(bvec->bv_page);
1024                                 blkvsc_req->request.DataBuffer.Length += bvec->bv_len;
1025
1026                                 prev_bvec = bvec;
1027
1028                                 databuf_idx++;
1029                                 num_sectors += bvec->bv_len >> 9;
1030
1031                         } /* bio_for_each_segment */
1032
1033                 } /* rq_for_each_bio */
1034         }
1035
1036         /* Handle the last one */
1037         if (blkvsc_req) {
1038                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1039                            blkdev, req, blkvsc_req->group,
1040                            blkvsc_req->group->outstanding);
1041
1042                 blkvsc_req->sector_start = start_sector;
1043                 sector_div(blkvsc_req->sector_start,
1044                            (blkdev->sector_size >> 9));
1045
1046                 blkvsc_req->sector_count = num_sectors /
1047                                            (blkdev->sector_size >> 9);
1048
1049                 blkvsc_init_rw(blkvsc_req);
1050         }
1051
1052         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1053                 if (pending) {
1054                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1055                                    "pending_list - blkvsc_req %p start_sect %lu"
1056                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1057                                    (unsigned long)blkvsc_req->sector_start,
1058                                    blkvsc_req->sector_count,
1059                                    (unsigned long)start_sector,
1060                                    (unsigned long)num_sectors);
1061
1062                         list_add_tail(&blkvsc_req->pend_entry,
1063                                       &blkdev->pending_list);
1064                 } else {
1065                         ret = blkvsc_submit_request(blkvsc_req,
1066                                                     blkvsc_request_completion);
1067                         if (ret == -1) {
1068                                 pending = 1;
1069                                 list_add_tail(&blkvsc_req->pend_entry,
1070                                               &blkdev->pending_list);
1071                         }
1072
1073                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1074                                    "start_sect %lu sect_count %ld (%lu %ld) "
1075                                    "ret %d\n", blkvsc_req,
1076                                    (unsigned long)blkvsc_req->sector_start,
1077                                    blkvsc_req->sector_count,
1078                                    (unsigned long)start_sector,
1079                                    num_sectors, ret);
1080                 }
1081         }
1082
1083         return pending;
1084 }
1085
1086 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1087 {
1088         struct blkvsc_request *blkvsc_req =
1089                         (struct blkvsc_request *)request->Context;
1090         struct block_device_context *blkdev =
1091                         (struct block_device_context *)blkvsc_req->dev;
1092         struct scsi_sense_hdr sense_hdr;
1093
1094         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1095                    blkvsc_req);
1096
1097         blkdev->num_outstanding_reqs--;
1098
1099         if (blkvsc_req->request.Status)
1100                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1101                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1102                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1103
1104         blkvsc_req->cond = 1;
1105         wake_up_interruptible(&blkvsc_req->wevent);
1106 }
1107
1108 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1109 {
1110         struct blkvsc_request *blkvsc_req =
1111                         (struct blkvsc_request *)request->Context;
1112         struct block_device_context *blkdev =
1113                         (struct block_device_context *)blkvsc_req->dev;
1114         unsigned long flags;
1115         struct blkvsc_request *comp_req, *tmp;
1116
1117         ASSERT(blkvsc_req->group);
1118
1119         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1120                    "sect_start %lu sect_count %ld len %d group outstd %d "
1121                    "total outstd %d\n",
1122                    blkdev, blkvsc_req, blkvsc_req->group,
1123                    (blkvsc_req->write) ? "WRITE" : "READ",
1124                    (unsigned long)blkvsc_req->sector_start,
1125                    blkvsc_req->sector_count,
1126                    blkvsc_req->request.DataBuffer.Length,
1127                    blkvsc_req->group->outstanding,
1128                    blkdev->num_outstanding_reqs);
1129
1130         spin_lock_irqsave(&blkdev->lock, flags);
1131
1132         blkdev->num_outstanding_reqs--;
1133         blkvsc_req->group->outstanding--;
1134
1135         /*
1136          * Only start processing when all the blkvsc_reqs are
1137          * completed. This guarantees no out-of-order blkvsc_req
1138          * completion when calling end_that_request_first()
1139          */
1140         if (blkvsc_req->group->outstanding == 0) {
1141                 list_for_each_entry_safe(comp_req, tmp,
1142                                          &blkvsc_req->group->blkvsc_req_list,
1143                                          req_entry) {
1144                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1145                                    "sect_start %lu sect_count %ld \n",
1146                                    comp_req,
1147                                    (unsigned long)comp_req->sector_start,
1148                                    comp_req->sector_count);
1149
1150                         list_del(&comp_req->req_entry);
1151
1152                         if (!__blk_end_request(comp_req->req,
1153                                 (!comp_req->request.Status ? 0 : -EIO),
1154                                 comp_req->sector_count * blkdev->sector_size)) {
1155                                 /*
1156                                  * All the sectors have been xferred ie the
1157                                  * request is done
1158                                  */
1159                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1160                                            comp_req->req);
1161                                 kmem_cache_free(blkdev->request_pool,
1162                                                 comp_req->group);
1163                         }
1164
1165                         kmem_cache_free(blkdev->request_pool, comp_req);
1166                 }
1167
1168                 if (!blkdev->shutting_down) {
1169                         blkvsc_do_pending_reqs(blkdev);
1170                         blk_start_queue(blkdev->gd->queue);
1171                         blkvsc_request(blkdev->gd->queue);
1172                 }
1173         }
1174
1175         spin_unlock_irqrestore(&blkdev->lock, flags);
1176 }
1177
1178 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1179 {
1180         struct blkvsc_request *pend_req, *tmp;
1181         struct blkvsc_request *comp_req, *tmp2;
1182
1183         int ret = 0;
1184
1185         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1186
1187         /* Flush the pending list first */
1188         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1189                                  pend_entry) {
1190                 /*
1191                  * The pend_req could be part of a partially completed
1192                  * request. If so, complete those req first until we
1193                  * hit the pend_req
1194                  */
1195                 list_for_each_entry_safe(comp_req, tmp2,
1196                                          &pend_req->group->blkvsc_req_list,
1197                                          req_entry) {
1198                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1199                                    "sect_start %lu sect_count %ld \n",
1200                                    comp_req,
1201                                    (unsigned long) comp_req->sector_start,
1202                                    comp_req->sector_count);
1203
1204                         if (comp_req == pend_req)
1205                                 break;
1206
1207                         list_del(&comp_req->req_entry);
1208
1209                         if (comp_req->req) {
1210                                 ret = __blk_end_request(comp_req->req,
1211                                         (!comp_req->request.Status ? 0 : -EIO),
1212                                         comp_req->sector_count *
1213                                         blkdev->sector_size);
1214                                 ASSERT(ret != 0);
1215                         }
1216
1217                         kmem_cache_free(blkdev->request_pool, comp_req);
1218                 }
1219
1220                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1221                            pend_req);
1222
1223                 list_del(&pend_req->pend_entry);
1224
1225                 list_del(&pend_req->req_entry);
1226
1227                 if (comp_req->req) {
1228                         if (!__blk_end_request(pend_req->req, -EIO,
1229                                                pend_req->sector_count *
1230                                                blkdev->sector_size)) {
1231                                 /*
1232                                  * All the sectors have been xferred ie the
1233                                  * request is done
1234                                  */
1235                                 DPRINT_DBG(BLKVSC_DRV,
1236                                            "blkvsc_cancel_pending_reqs() - "
1237                                            "req %p COMPLETED\n", pend_req->req);
1238                                 kmem_cache_free(blkdev->request_pool,
1239                                                 pend_req->group);
1240                         }
1241                 }
1242
1243                 kmem_cache_free(blkdev->request_pool, pend_req);
1244         }
1245
1246         return ret;
1247 }
1248
1249 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1250 {
1251         struct blkvsc_request *pend_req, *tmp;
1252         int ret = 0;
1253
1254         /* Flush the pending list first */
1255         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1256                                  pend_entry) {
1257                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1258                            pend_req);
1259
1260                 ret = blkvsc_submit_request(pend_req,
1261                                             blkvsc_request_completion);
1262                 if (ret != 0)
1263                         break;
1264                 else
1265                         list_del(&pend_req->pend_entry);
1266         }
1267
1268         return ret;
1269 }
1270
1271 static void blkvsc_request(struct request_queue *queue)
1272 {
1273         struct block_device_context *blkdev = NULL;
1274         struct request *req;
1275         int ret = 0;
1276
1277         DPRINT_DBG(BLKVSC_DRV, "- enter \n");
1278         while ((req = blk_peek_request(queue)) != NULL) {
1279                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1280
1281                 blkdev = req->rq_disk->private_data;
1282                 if (blkdev->shutting_down || !blk_fs_request(req) ||
1283                     blkdev->media_not_present) {
1284                         __blk_end_request_cur(req, 0);
1285                         continue;
1286                 }
1287
1288                 ret = blkvsc_do_pending_reqs(blkdev);
1289
1290                 if (ret != 0) {
1291                         DPRINT_DBG(BLKVSC_DRV,
1292                                    "- stop queue - pending_list not empty\n");
1293                         blk_stop_queue(queue);
1294                         break;
1295                 }
1296
1297                 blk_start_request(req);
1298
1299                 ret = blkvsc_do_request(blkdev, req);
1300                 if (ret > 0) {
1301                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1302                         blk_stop_queue(queue);
1303                         break;
1304                 } else if (ret < 0) {
1305                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1306                         blk_requeue_request(queue, req);
1307                         blk_stop_queue(queue);
1308                         break;
1309                 }
1310         }
1311 }
1312
1313 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1314 {
1315         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1316
1317         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1318                    blkdev->gd->disk_name);
1319
1320         spin_lock(&blkdev->lock);
1321
1322         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1323                 spin_unlock(&blkdev->lock);
1324                 check_disk_change(bdev);
1325                 spin_lock(&blkdev->lock);
1326         }
1327
1328         blkdev->users++;
1329
1330         spin_unlock(&blkdev->lock);
1331         return 0;
1332 }
1333
1334 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1335 {
1336         struct block_device_context *blkdev = disk->private_data;
1337
1338         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1339                    blkdev->gd->disk_name);
1340
1341         spin_lock(&blkdev->lock);
1342         if (blkdev->users == 1) {
1343                 spin_unlock(&blkdev->lock);
1344                 blkvsc_do_flush(blkdev);
1345                 spin_lock(&blkdev->lock);
1346         }
1347
1348         blkdev->users--;
1349
1350         spin_unlock(&blkdev->lock);
1351         return 0;
1352 }
1353
1354 static int blkvsc_media_changed(struct gendisk *gd)
1355 {
1356         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1357         return 1;
1358 }
1359
1360 static int blkvsc_revalidate_disk(struct gendisk *gd)
1361 {
1362         struct block_device_context *blkdev = gd->private_data;
1363
1364         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1365
1366         if (blkdev->device_type == DVD_TYPE) {
1367                 blkvsc_do_read_capacity(blkdev);
1368                 set_capacity(blkdev->gd, blkdev->capacity *
1369                             (blkdev->sector_size/512));
1370                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1371         }
1372         return 0;
1373 }
1374
1375 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1376 {
1377         sector_t total_sectors = get_capacity(bd->bd_disk);
1378         sector_t cylinder_times_heads = 0;
1379         sector_t temp = 0;
1380
1381         int sectors_per_track = 0;
1382         int heads = 0;
1383         int cylinders = 0;
1384         int rem = 0;
1385
1386         if (total_sectors > (65535 * 16 * 255))
1387                 total_sectors = (65535 * 16 * 255);
1388
1389         if (total_sectors >= (65535 * 16 * 63)) {
1390                 sectors_per_track = 255;
1391                 heads = 16;
1392
1393                 cylinder_times_heads = total_sectors;
1394                 /* sector_div stores the quotient in cylinder_times_heads */
1395                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1396         } else {
1397                 sectors_per_track = 17;
1398
1399                 cylinder_times_heads = total_sectors;
1400                 /* sector_div stores the quotient in cylinder_times_heads */
1401                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1402
1403                 temp = cylinder_times_heads + 1023;
1404                 /* sector_div stores the quotient in temp */
1405                 rem = sector_div(temp, 1024);
1406
1407                 heads = temp;
1408
1409                 if (heads < 4)
1410                         heads = 4;
1411
1412
1413                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1414                         sectors_per_track = 31;
1415                         heads = 16;
1416
1417                         cylinder_times_heads = total_sectors;
1418                         /*
1419                          * sector_div stores the quotient in
1420                          * cylinder_times_heads
1421                          */
1422                         rem = sector_div(cylinder_times_heads,
1423                                          sectors_per_track);
1424                 }
1425
1426                 if (cylinder_times_heads >= (heads * 1024)) {
1427                         sectors_per_track = 63;
1428                         heads = 16;
1429
1430                         cylinder_times_heads = total_sectors;
1431                         /*
1432                          * sector_div stores the quotient in
1433                          * cylinder_times_heads
1434                          */
1435                         rem = sector_div(cylinder_times_heads,
1436                                          sectors_per_track);
1437                 }
1438         }
1439
1440         temp = cylinder_times_heads;
1441         /* sector_div stores the quotient in temp */
1442         rem = sector_div(temp, heads);
1443         cylinders = temp;
1444
1445         hg->heads = heads;
1446         hg->sectors = sectors_per_track;
1447         hg->cylinders = cylinders;
1448
1449         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1450                     sectors_per_track);
1451
1452     return 0;
1453 }
1454
1455 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1456                         unsigned cmd, unsigned long argument)
1457 {
1458 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1459         int ret;
1460
1461         switch (cmd) {
1462         /*
1463          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1464          * than just a GUID. Commented it out for now.
1465          */
1466 #if 0
1467         case HDIO_GET_IDENTITY:
1468                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1469                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1470                                  blkdev->device_id_len))
1471                         ret = -EFAULT;
1472                 break;
1473 #endif
1474         default:
1475                 ret = -EINVAL;
1476                 break;
1477         }
1478
1479         return ret;
1480 }
1481
1482 static int __init blkvsc_init(void)
1483 {
1484         int ret;
1485
1486         ASSERT(sizeof(sector_t) == 8); /* Make sure CONFIG_LBD is set */
1487
1488         DPRINT_ENTER(BLKVSC_DRV);
1489
1490         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1491
1492         ret = blkvsc_drv_init(BlkVscInitialize);
1493
1494         DPRINT_EXIT(BLKVSC_DRV);
1495
1496         return ret;
1497 }
1498
1499 static void __exit blkvsc_exit(void)
1500 {
1501         DPRINT_ENTER(BLKVSC_DRV);
1502         blkvsc_drv_exit();
1503         DPRINT_ENTER(BLKVSC_DRV);
1504 }
1505
1506 MODULE_LICENSE("GPL");
1507 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
1508 module_init(blkvsc_init);
1509 module_exit(blkvsc_exit);