drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "hv_api.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 /*
  50  * This request ties the struct request and struct
  51  * blkvsc_request/hv_storvsc_request together A struct request may be
  52  * represented by 1 or more struct blkvsc_request
  53  */
  54 struct blkvsc_request_group {
  55         int outstanding;
  56         int status;
  57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  58 };
  59
  60 struct blkvsc_request {
  61         /* blkvsc_request_group.blkvsc_req_list */
  62         struct list_head req_entry;
  63
  64         /* block_device_context.pending_list */
  65         struct list_head pend_entry;
  66
  67         /* This may be null if we generate a request internally */
  68         struct request *req;
  69
  70         struct block_device_context *dev;
  71
  72         /* The group this request is part of. Maybe null */
  73         struct blkvsc_request_group *group;
  74
  75         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
  86 /* Per device structure */
  87 struct block_device_context {
  88         /* point back to our device context */
  89         struct hv_device *device_ctx;
  90         struct kmem_cache *request_pool;
  91         spinlock_t lock;
  92         struct gendisk *gd;
  93         enum blkvsc_device_type device_type;
  94         struct list_head pending_list;
  95
  96         unsigned char device_id[64];
  97         unsigned int device_id_len;
  98         int num_outstanding_reqs;
  99         int shutting_down;
 100         int media_not_present;
 101         unsigned int sector_size;
 102         sector_t capacity;
 103         unsigned int port;
 104         unsigned char path;
 105         unsigned char target;
 106         int users;
 107 };
 108
 109 static DEFINE_MUTEX(blkvsc_mutex);
 110
 111 static const char *g_blk_driver_name = "blkvsc";
 112
 113 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 114 static const struct hv_guid g_blk_device_type = {
 115         .data = {
 116                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 117                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 118         }
 119 };
 120
 121 static int blk_vsc_on_device_add(struct hv_device *device,
 122                                 void *additional_info)
 123 {
 124         struct storvsc_device_info *device_info;
 125         int ret = 0;
 126
 127         device_info = (struct storvsc_device_info *)additional_info;
 128
 129         ret = stor_vsc_on_device_add(device, additional_info);
 130         if (ret != 0)
 131                 return ret;
 132
 133         /*
 134          * We need to use the device instance guid to set the path and target
 135          * id. For IDE devices, the device instance id is formatted as
 136          * <bus id> * - <device id> - 8899 - 000000000000.
 137          */
 138         device_info->path_id = device->dev_instance.data[3] << 24 |
 139                              device->dev_instance.data[2] << 16 |
 140                              device->dev_instance.data[1] << 8  |
 141                              device->dev_instance.data[0];
 142
 143         device_info->target_id = device->dev_instance.data[5] << 8 |
 144                                device->dev_instance.data[4];
 145
 146         return ret;
 147 }
 148
 149
 150 static int blk_vsc_initialize(struct hv_driver *driver)
 151 {
 152         struct storvsc_driver_object *stor_driver;
 153         int ret = 0;
 154
 155         stor_driver = hvdr_to_stordr(driver);
 156
 157         /* Make sure we are at least 2 pages since 1 page is used for control */
 158         /* ASSERT(stor_driver->RingBufferSize >= (PAGE_SIZE << 1)); */
 159
 160         driver->name = g_blk_driver_name;
 161         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 162
 163
 164         /*
 165          * Divide the ring buffer data size (which is 1 page less than the ring
 166          * buffer size since that page is reserved for the ring buffer indices)
 167          * by the max request size (which is
 168          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 169          */
 170         stor_driver->max_outstanding_req_per_channel =
 171                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 172                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 173                            sizeof(struct vstor_packet) + sizeof(u64),
 174                            sizeof(u64)));
 175
 176         DPRINT_INFO(BLKVSC, "max io outstd %u",
 177                     stor_driver->max_outstanding_req_per_channel);
 178
 179         /* Setup the dispatch table */
 180         stor_driver->base.dev_add = blk_vsc_on_device_add;
 181         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 182         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 183         stor_driver->on_io_request = stor_vsc_on_io_request;
 184
 185         return ret;
 186 }
 187
 188
 189 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 190                         void (*request_completion)(struct hv_storvsc_request *))
 191 {
 192         struct block_device_context *blkdev = blkvsc_req->dev;
 193         struct hv_device *device_ctx = blkdev->device_ctx;
 194         struct hv_driver *drv =
 195                         drv_to_hv_drv(device_ctx->device.driver);
 196         struct storvsc_driver_object *storvsc_drv_obj =
 197                         drv->priv;
 198         struct hv_storvsc_request *storvsc_req;
 199         struct vmscsi_request *vm_srb;
 200         int ret;
 201
 202         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 203                    "req %p type %s start_sector %lu count %ld offset %d "
 204                    "len %d\n", blkvsc_req,
 205                    (blkvsc_req->write) ? "WRITE" : "READ",
 206                    (unsigned long) blkvsc_req->sector_start,
 207                    blkvsc_req->sector_count,
 208                    blkvsc_req->request.data_buffer.offset,
 209                    blkvsc_req->request.data_buffer.len);
 210 #if 0
 211         for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
 212                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 213                            "req %p pfn[%d] %llx\n",
 214                            blkvsc_req, i,
 215                            blkvsc_req->request.data_buffer.pfn_array[i]);
 216         }
 217 #endif
 218
 219         storvsc_req = &blkvsc_req->request;
 220         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 221
 222         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 223
 224         storvsc_req->on_io_completion = request_completion;
 225         storvsc_req->context = blkvsc_req;
 226
 227         vm_srb->port_number = blkdev->port;
 228         vm_srb->path_id = blkdev->path;
 229         vm_srb->target_id = blkdev->target;
 230         vm_srb->lun = 0;         /* this is not really used at all */
 231
 232         vm_srb->cdb_length = blkvsc_req->cmd_len;
 233
 234         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 235
 236         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 237
 238         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 239                                            &blkvsc_req->request);
 240         if (ret == 0)
 241                 blkdev->num_outstanding_reqs++;
 242
 243         return ret;
 244 }
 245
 246
 247 static unsigned int blkvsc_check_events(struct gendisk *gd,
 248                                         unsigned int clearing)
 249 {
 250         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
 251         return DISK_EVENT_MEDIA_CHANGE;
 252 }
 253
 254
 255 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
 256 {
 257         struct block_device_context *blkdev = bdev->bd_disk->private_data;
 258
 259         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
 260                    blkdev->gd->disk_name);
 261
 262         mutex_lock(&blkvsc_mutex);
 263         spin_lock(&blkdev->lock);
 264
 265         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
 266                 spin_unlock(&blkdev->lock);
 267                 check_disk_change(bdev);
 268                 spin_lock(&blkdev->lock);
 269         }
 270
 271         blkdev->users++;
 272
 273         spin_unlock(&blkdev->lock);
 274         mutex_unlock(&blkvsc_mutex);
 275         return 0;
 276 }
 277
 278 /* Static decl */
 279 static int blkvsc_probe(struct device *dev);
 280 static int blkvsc_remove(struct device *device);
 281 static void blkvsc_shutdown(struct device *device);
 282
 283 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 284 static int blkvsc_revalidate_disk(struct gendisk *gd);
 285 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 286 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 287                         unsigned cmd, unsigned long argument);
 288 static void blkvsc_request(struct request_queue *queue);
 289 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 290 static int blkvsc_do_request(struct block_device_context *blkdev,
 291                              struct request *req);
 292 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 293 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 294 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 295 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 296 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 297 static int blkvsc_do_flush(struct block_device_context *blkdev);
 298 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 299 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 300
 301 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 302 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 303 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 304
 305 /* The one and only one */
 306 static  struct storvsc_driver_object g_blkvsc_drv;
 307
 308 static const struct block_device_operations block_ops = {
 309         .owner = THIS_MODULE,
 310         .open = blkvsc_open,
 311         .release = blkvsc_release,
 312         .check_events = blkvsc_check_events,
 313         .revalidate_disk = blkvsc_revalidate_disk,
 314         .getgeo = blkvsc_getgeo,
 315         .ioctl  = blkvsc_ioctl,
 316 };
 317
 318 /*
 319  * blkvsc_drv_init -  BlkVsc driver initialization.
 320  */
 321 static int blkvsc_drv_init(void)
 322 {
 323         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 324         struct hv_driver *drv = &g_blkvsc_drv.base;
 325         int ret;
 326
 327         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 328
 329         drv->priv = storvsc_drv_obj;
 330
 331         /* Callback to client driver to complete the initialization */
 332         blk_vsc_initialize(&storvsc_drv_obj->base);
 333
 334         drv->driver.name = storvsc_drv_obj->base.name;
 335
 336         drv->driver.probe = blkvsc_probe;
 337         drv->driver.remove = blkvsc_remove;
 338         drv->driver.shutdown = blkvsc_shutdown;
 339
 340         /* The driver belongs to vmbus */
 341         ret = vmbus_child_driver_register(&drv->driver);
 342
 343         return ret;
 344 }
 345
 346 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 347 {
 348         struct device **curr = (struct device **)data;
 349         *curr = dev;
 350         return 1; /* stop iterating */
 351 }
 352
 353 static void blkvsc_drv_exit(void)
 354 {
 355         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 356         struct hv_driver *drv = &g_blkvsc_drv.base;
 357         struct device *current_dev;
 358         int ret;
 359
 360         while (1) {
 361                 current_dev = NULL;
 362
 363                 /* Get the device */
 364                 ret = driver_for_each_device(&drv->driver, NULL,
 365                                              (void *) &current_dev,
 366                                              blkvsc_drv_exit_cb);
 367
 368                 if (ret)
 369                         DPRINT_WARN(BLKVSC_DRV,
 370                                     "driver_for_each_device returned %d", ret);
 371
 372
 373                 if (current_dev == NULL)
 374                         break;
 375
 376                 /* Initiate removal from the top-down */
 377                 device_unregister(current_dev);
 378         }
 379
 380         if (storvsc_drv_obj->base.cleanup)
 381                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 382
 383         vmbus_child_driver_unregister(&drv->driver);
 384
 385         return;
 386 }
 387
 388 /*
 389  * blkvsc_probe - Add a new device for this driver
 390  */
 391 static int blkvsc_probe(struct device *device)
 392 {
 393         struct hv_driver *drv =
 394                                 drv_to_hv_drv(device->driver);
 395         struct storvsc_driver_object *storvsc_drv_obj =
 396                                 drv->priv;
 397         struct hv_device *device_obj = device_to_hv_device(device);
 398
 399         struct block_device_context *blkdev = NULL;
 400         struct storvsc_device_info device_info;
 401         int major = 0;
 402         int devnum = 0;
 403         int ret = 0;
 404         static int ide0_registered;
 405         static int ide1_registered;
 406
 407         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 408
 409         if (!storvsc_drv_obj->base.dev_add) {
 410                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 411                 ret = -1;
 412                 goto Cleanup;
 413         }
 414
 415         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 416         if (!blkdev) {
 417                 ret = -ENOMEM;
 418                 goto Cleanup;
 419         }
 420
 421         INIT_LIST_HEAD(&blkdev->pending_list);
 422
 423         /* Initialize what we can here */
 424         spin_lock_init(&blkdev->lock);
 425
 426         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 427         /*      sizeof(struct blkvsc_request)); */
 428
 429         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 430                                         sizeof(struct blkvsc_request), 0,
 431                                         SLAB_HWCACHE_ALIGN, NULL);
 432         if (!blkdev->request_pool) {
 433                 ret = -ENOMEM;
 434                 goto Cleanup;
 435         }
 436
 437
 438         /* Call to the vsc driver to add the device */
 439         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 440         if (ret != 0) {
 441                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 442                 goto Cleanup;
 443         }
 444
 445         blkdev->device_ctx = device_obj;
 446         /* this identified the device 0 or 1 */
 447         blkdev->target = device_info.target_id;
 448         /* this identified the ide ctrl 0 or 1 */
 449         blkdev->path = device_info.path_id;
 450
 451         dev_set_drvdata(device, blkdev);
 452
 453         /* Calculate the major and device num */
 454         if (blkdev->path == 0) {
 455                 major = IDE0_MAJOR;
 456                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 457
 458                 if (!ide0_registered) {
 459                         ret = register_blkdev(major, "ide");
 460                         if (ret != 0) {
 461                                 DPRINT_ERR(BLKVSC_DRV,
 462                                            "register_blkdev() failed! ret %d",
 463                                            ret);
 464                                 goto Remove;
 465                         }
 466
 467                         ide0_registered = 1;
 468                 }
 469         } else if (blkdev->path == 1) {
 470                 major = IDE1_MAJOR;
 471                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 472
 473                 if (!ide1_registered) {
 474                         ret = register_blkdev(major, "ide");
 475                         if (ret != 0) {
 476                                 DPRINT_ERR(BLKVSC_DRV,
 477                                            "register_blkdev() failed! ret %d",
 478                                            ret);
 479                                 goto Remove;
 480                         }
 481
 482                         ide1_registered = 1;
 483                 }
 484         } else {
 485                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 486                 ret = -1;
 487                 goto Cleanup;
 488         }
 489
 490         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 491
 492         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 493         if (!blkdev->gd) {
 494                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 495                 ret = -1;
 496                 goto Cleanup;
 497         }
 498
 499         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 500
 501         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 502         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 503         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 504         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 505         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 506
 507         blkdev->gd->major = major;
 508         if (devnum == 1 || devnum == 3)
 509                 blkdev->gd->first_minor = BLKVSC_MINORS;
 510         else
 511                 blkdev->gd->first_minor = 0;
 512         blkdev->gd->fops = &block_ops;
 513         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 514         blkdev->gd->private_data = blkdev;
 515         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 516         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 517
 518         blkvsc_do_inquiry(blkdev);
 519         if (blkdev->device_type == DVD_TYPE) {
 520                 set_disk_ro(blkdev->gd, 1);
 521                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 522                 blkvsc_do_read_capacity(blkdev);
 523         } else {
 524                 blkvsc_do_read_capacity16(blkdev);
 525         }
 526
 527         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 528         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 529         /* go! */
 530         add_disk(blkdev->gd);
 531
 532         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 533                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 534                     blkdev->sector_size);
 535
 536         return ret;
 537
 538 Remove:
 539         storvsc_drv_obj->base.dev_rm(device_obj);
 540
 541 Cleanup:
 542         if (blkdev) {
 543                 if (blkdev->request_pool) {
 544                         kmem_cache_destroy(blkdev->request_pool);
 545                         blkdev->request_pool = NULL;
 546                 }
 547                 kfree(blkdev);
 548                 blkdev = NULL;
 549         }
 550
 551         return ret;
 552 }
 553
 554 static void blkvsc_shutdown(struct device *device)
 555 {
 556         struct block_device_context *blkdev = dev_get_drvdata(device);
 557         unsigned long flags;
 558
 559         if (!blkdev)
 560                 return;
 561
 562         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 563                    blkdev->users, blkdev->gd->disk_name);
 564
 565         spin_lock_irqsave(&blkdev->lock, flags);
 566
 567         blkdev->shutting_down = 1;
 568
 569         blk_stop_queue(blkdev->gd->queue);
 570
 571         spin_unlock_irqrestore(&blkdev->lock, flags);
 572
 573         while (blkdev->num_outstanding_reqs) {
 574                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 575                             blkdev->num_outstanding_reqs);
 576                 udelay(100);
 577         }
 578
 579         blkvsc_do_flush(blkdev);
 580
 581         spin_lock_irqsave(&blkdev->lock, flags);
 582
 583         blkvsc_cancel_pending_reqs(blkdev);
 584
 585         spin_unlock_irqrestore(&blkdev->lock, flags);
 586 }
 587
 588 static int blkvsc_do_flush(struct block_device_context *blkdev)
 589 {
 590         struct blkvsc_request *blkvsc_req;
 591
 592         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 593
 594         if (blkdev->device_type != HARDDISK_TYPE)
 595                 return 0;
 596
 597         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 598         if (!blkvsc_req)
 599                 return -ENOMEM;
 600
 601         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 602         init_completion(&blkvsc_req->request.wait_event);
 603         blkvsc_req->dev = blkdev;
 604         blkvsc_req->req = NULL;
 605         blkvsc_req->write = 0;
 606
 607         blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 608         blkvsc_req->request.data_buffer.offset = 0;
 609         blkvsc_req->request.data_buffer.len = 0;
 610
 611         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 612         blkvsc_req->cmd_len = 10;
 613
 614         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 615
 616         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 617
 618         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 619
 620         return 0;
 621 }
 622
 623 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 624 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 625 {
 626         struct blkvsc_request *blkvsc_req;
 627         struct page *page_buf;
 628         unsigned char *buf;
 629         unsigned char device_type;
 630
 631         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 632
 633         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 634         if (!blkvsc_req)
 635                 return -ENOMEM;
 636
 637         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 638         page_buf = alloc_page(GFP_KERNEL);
 639         if (!page_buf) {
 640                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 641                 return -ENOMEM;
 642         }
 643
 644         init_completion(&blkvsc_req->request.wait_event);
 645         blkvsc_req->dev = blkdev;
 646         blkvsc_req->req = NULL;
 647         blkvsc_req->write = 0;
 648
 649         blkvsc_req->request.data_buffer.pfn_array[0] =
 650         page_to_pfn(page_buf);
 651         blkvsc_req->request.data_buffer.offset = 0;
 652         blkvsc_req->request.data_buffer.len = 64;
 653
 654         blkvsc_req->cmnd[0] = INQUIRY;
 655         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 656         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 657         blkvsc_req->cmnd[4] = 64;
 658         blkvsc_req->cmd_len = 6;
 659
 660         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 661
 662         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 663                    blkvsc_req);
 664
 665         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 666
 667         buf = kmap(page_buf);
 668
 669         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 670         /* be to le */
 671         device_type = buf[0] & 0x1F;
 672
 673         if (device_type == 0x0) {
 674                 blkdev->device_type = HARDDISK_TYPE;
 675         } else if (device_type == 0x5) {
 676                 blkdev->device_type = DVD_TYPE;
 677         } else {
 678                 /* TODO: this is currently unsupported device type */
 679                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 680         }
 681
 682         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 683
 684         blkdev->device_id_len = buf[7];
 685         if (blkdev->device_id_len > 64)
 686                 blkdev->device_id_len = 64;
 687
 688         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 689         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 690          * blkdev->device_id_len); */
 691
 692         kunmap(page_buf);
 693
 694         __free_page(page_buf);
 695
 696         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 697
 698         return 0;
 699 }
 700
 701 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 702 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 703 {
 704         struct blkvsc_request *blkvsc_req;
 705         struct page *page_buf;
 706         unsigned char *buf;
 707         struct scsi_sense_hdr sense_hdr;
 708         struct vmscsi_request *vm_srb;
 709
 710         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 711
 712         blkdev->sector_size = 0;
 713         blkdev->capacity = 0;
 714         blkdev->media_not_present = 0; /* assume a disk is present */
 715
 716         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 717         if (!blkvsc_req)
 718                 return -ENOMEM;
 719
 720         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 721         page_buf = alloc_page(GFP_KERNEL);
 722         if (!page_buf) {
 723                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 724                 return -ENOMEM;
 725         }
 726
 727         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 728         init_completion(&blkvsc_req->request.wait_event);
 729         blkvsc_req->dev = blkdev;
 730         blkvsc_req->req = NULL;
 731         blkvsc_req->write = 0;
 732
 733         blkvsc_req->request.data_buffer.pfn_array[0] =
 734         page_to_pfn(page_buf);
 735         blkvsc_req->request.data_buffer.offset = 0;
 736         blkvsc_req->request.data_buffer.len = 8;
 737
 738         blkvsc_req->cmnd[0] = READ_CAPACITY;
 739         blkvsc_req->cmd_len = 16;
 740
 741         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 742
 743         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 744                    blkvsc_req);
 745
 746         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 747
 748         /* check error */
 749         if (vm_srb->scsi_status) {
 750                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 751                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 752
 753                 if (sense_hdr.asc == 0x3A) {
 754                         /* Medium not present */
 755                         blkdev->media_not_present = 1;
 756                 }
 757                 return 0;
 758         }
 759         buf = kmap(page_buf);
 760
 761         /* be to le */
 762         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 763                             (buf[2] << 8) | buf[3]) + 1;
 764         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 765                               (buf[6] << 8) | buf[7];
 766
 767         kunmap(page_buf);
 768
 769         __free_page(page_buf);
 770
 771         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 772
 773         return 0;
 774 }
 775
 776 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 777 {
 778         struct blkvsc_request *blkvsc_req;
 779         struct page *page_buf;
 780         unsigned char *buf;
 781         struct scsi_sense_hdr sense_hdr;
 782         struct vmscsi_request *vm_srb;
 783
 784         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 785
 786         blkdev->sector_size = 0;
 787         blkdev->capacity = 0;
 788         blkdev->media_not_present = 0; /* assume a disk is present */
 789
 790         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 791         if (!blkvsc_req)
 792                 return -ENOMEM;
 793
 794         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 795         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 796         page_buf = alloc_page(GFP_KERNEL);
 797         if (!page_buf) {
 798                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 799                 return -ENOMEM;
 800         }
 801
 802         init_completion(&blkvsc_req->request.wait_event);
 803         blkvsc_req->dev = blkdev;
 804         blkvsc_req->req = NULL;
 805         blkvsc_req->write = 0;
 806
 807         blkvsc_req->request.data_buffer.pfn_array[0] =
 808         page_to_pfn(page_buf);
 809         blkvsc_req->request.data_buffer.offset = 0;
 810         blkvsc_req->request.data_buffer.len = 12;
 811
 812         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 813         blkvsc_req->cmd_len = 16;
 814
 815         /*
 816          * Set this here since the completion routine may be invoked
 817          * and completed before we return
 818          */
 819
 820         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 821
 822         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 823                    blkvsc_req);
 824
 825         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 826
 827         /* check error */
 828         if (vm_srb->scsi_status) {
 829                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 830                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 831                 if (sense_hdr.asc == 0x3A) {
 832                         /* Medium not present */
 833                         blkdev->media_not_present = 1;
 834                 }
 835                 return 0;
 836         }
 837         buf = kmap(page_buf);
 838
 839         /* be to le */
 840         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 841         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 842
 843 #if 0
 844         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 845                             (buf[2] << 8) | buf[3]) + 1;
 846         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 847                               (buf[6] << 8) | buf[7];
 848 #endif
 849
 850         kunmap(page_buf);
 851
 852         __free_page(page_buf);
 853
 854         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 855
 856         return 0;
 857 }
 858
 859 /*
 860  * blkvsc_remove() - Callback when our device is removed
 861  */
 862 static int blkvsc_remove(struct device *device)
 863 {
 864         struct hv_driver *drv =
 865                                 drv_to_hv_drv(device->driver);
 866         struct storvsc_driver_object *storvsc_drv_obj =
 867                                 drv->priv;
 868         struct hv_device *device_obj = device_to_hv_device(device);
 869         struct block_device_context *blkdev = dev_get_drvdata(device);
 870         unsigned long flags;
 871         int ret;
 872
 873         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 874
 875         if (!storvsc_drv_obj->base.dev_rm)
 876                 return -1;
 877
 878         /*
 879          * Call to the vsc driver to let it know that the device is being
 880          * removed
 881          */
 882         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 883         if (ret != 0) {
 884                 /* TODO: */
 885                 DPRINT_ERR(BLKVSC_DRV,
 886                            "unable to remove blkvsc device (ret %d)", ret);
 887         }
 888
 889         /* Get to a known state */
 890         spin_lock_irqsave(&blkdev->lock, flags);
 891
 892         blkdev->shutting_down = 1;
 893
 894         blk_stop_queue(blkdev->gd->queue);
 895
 896         spin_unlock_irqrestore(&blkdev->lock, flags);
 897
 898         while (blkdev->num_outstanding_reqs) {
 899                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 900                             blkdev->num_outstanding_reqs);
 901                 udelay(100);
 902         }
 903
 904         blkvsc_do_flush(blkdev);
 905
 906         spin_lock_irqsave(&blkdev->lock, flags);
 907
 908         blkvsc_cancel_pending_reqs(blkdev);
 909
 910         spin_unlock_irqrestore(&blkdev->lock, flags);
 911
 912         blk_cleanup_queue(blkdev->gd->queue);
 913
 914         del_gendisk(blkdev->gd);
 915
 916         kmem_cache_destroy(blkdev->request_pool);
 917
 918         kfree(blkdev);
 919
 920         return ret;
 921 }
 922
 923 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 924 {
 925         /* ASSERT(blkvsc_req->req); */
 926         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 927
 928         blkvsc_req->cmd_len = 16;
 929
 930         if (blkvsc_req->sector_start > 0xffffffff) {
 931                 if (rq_data_dir(blkvsc_req->req)) {
 932                         blkvsc_req->write = 1;
 933                         blkvsc_req->cmnd[0] = WRITE_16;
 934                 } else {
 935                         blkvsc_req->write = 0;
 936                         blkvsc_req->cmnd[0] = READ_16;
 937                 }
 938
 939                 blkvsc_req->cmnd[1] |=
 940                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 941
 942                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 943                                 cpu_to_be64(blkvsc_req->sector_start);
 944                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 945                                 cpu_to_be32(blkvsc_req->sector_count);
 946         } else if ((blkvsc_req->sector_count > 0xff) ||
 947                    (blkvsc_req->sector_start > 0x1fffff)) {
 948                 if (rq_data_dir(blkvsc_req->req)) {
 949                         blkvsc_req->write = 1;
 950                         blkvsc_req->cmnd[0] = WRITE_10;
 951                 } else {
 952                         blkvsc_req->write = 0;
 953                         blkvsc_req->cmnd[0] = READ_10;
 954                 }
 955
 956                 blkvsc_req->cmnd[1] |=
 957                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 958
 959                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 960                                 cpu_to_be32(blkvsc_req->sector_start);
 961                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 962                                 cpu_to_be16(blkvsc_req->sector_count);
 963         } else {
 964                 if (rq_data_dir(blkvsc_req->req)) {
 965                         blkvsc_req->write = 1;
 966                         blkvsc_req->cmnd[0] = WRITE_6;
 967                 } else {
 968                         blkvsc_req->write = 0;
 969                         blkvsc_req->cmnd[0] = READ_6;
 970                 }
 971
 972                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 973                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 974                 blkvsc_req->cmnd[1] &= 0x1f;
 975                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 976         }
 977 }
 978
 979
 980 /*
 981  * We break the request into 1 or more blkvsc_requests and submit
 982  * them.  If we can't submit them all, we put them on the
 983  * pending_list. The blkvsc_request() will work on the pending_list.
 984  */
 985 static int blkvsc_do_request(struct block_device_context *blkdev,
 986                              struct request *req)
 987 {
 988         struct bio *bio = NULL;
 989         struct bio_vec *bvec = NULL;
 990         struct bio_vec *prev_bvec = NULL;
 991         struct blkvsc_request *blkvsc_req = NULL;
 992         struct blkvsc_request *tmp;
 993         int databuf_idx = 0;
 994         int seg_idx = 0;
 995         sector_t start_sector;
 996         unsigned long num_sectors = 0;
 997         int ret = 0;
 998         int pending = 0;
 999         struct blkvsc_request_group *group = NULL;
1000
1001         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
1002                   (unsigned long)blk_rq_pos(req));
1003
1004         /* Create a group to tie req to list of blkvsc_reqs */
1005         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
1006         if (!group)
1007                 return -ENOMEM;
1008
1009         INIT_LIST_HEAD(&group->blkvsc_req_list);
1010         group->outstanding = group->status = 0;
1011
1012         start_sector = blk_rq_pos(req);
1013
1014         /* foreach bio in the request */
1015         if (req->bio) {
1016                 for (bio = req->bio; bio; bio = bio->bi_next) {
1017                         /*
1018                          * Map this bio into an existing or new storvsc request
1019                          */
1020                         bio_for_each_segment(bvec, bio, seg_idx) {
1021                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
1022                                            "- req %p bio %p bvec %p seg_idx %d "
1023                                            "databuf_idx %d\n", req, bio, bvec,
1024                                            seg_idx, databuf_idx);
1025
1026                                 /* Get a new storvsc request */
1027                                 /* 1st-time */
1028                                 if ((!blkvsc_req) ||
1029                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
1030                                     /* hole at the begin of page */
1031                                     || (bvec->bv_offset != 0) ||
1032                                     /* hold at the end of page */
1033                                     (prev_bvec &&
1034                                      (prev_bvec->bv_len != PAGE_SIZE))) {
1035                                         /* submit the prev one */
1036                                         if (blkvsc_req) {
1037                                                 blkvsc_req->sector_start = start_sector;
1038                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
1039
1040                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
1041                                                 blkvsc_init_rw(blkvsc_req);
1042                                         }
1043
1044                                         /*
1045                                          * Create new blkvsc_req to represent
1046                                          * the current bvec
1047                                          */
1048                                         blkvsc_req =
1049                                         kmem_cache_zalloc(
1050                                         blkdev->request_pool, GFP_ATOMIC);
1051                                         if (!blkvsc_req) {
1052                                                 /* free up everything */
1053                                                 list_for_each_entry_safe(
1054                                                         blkvsc_req, tmp,
1055                                                         &group->blkvsc_req_list,
1056                                                         req_entry) {
1057                                                         list_del(&blkvsc_req->req_entry);
1058                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
1059                                                 }
1060
1061                                                 kmem_cache_free(blkdev->request_pool, group);
1062                                                 return -ENOMEM;
1063                                         }
1064
1065                                         memset(blkvsc_req, 0,
1066                                                sizeof(struct blkvsc_request));
1067
1068                                         blkvsc_req->dev = blkdev;
1069                                         blkvsc_req->req = req;
1070                                         blkvsc_req->request.
1071                                         data_buffer.offset
1072                                         = bvec->bv_offset;
1073                                         blkvsc_req->request.
1074                                         data_buffer.len = 0;
1075
1076                                         /* Add to the group */
1077                                         blkvsc_req->group = group;
1078                                         blkvsc_req->group->outstanding++;
1079                                         list_add_tail(&blkvsc_req->req_entry,
1080                                                 &blkvsc_req->group->blkvsc_req_list);
1081
1082                                         start_sector += num_sectors;
1083                                         num_sectors = 0;
1084                                         databuf_idx = 0;
1085                                 }
1086
1087                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1088                                 blkvsc_req->request.data_buffer.
1089                                         pfn_array[databuf_idx]
1090                                                 = page_to_pfn(bvec->bv_page);
1091                                 blkvsc_req->request.data_buffer.len
1092                                         += bvec->bv_len;
1093
1094                                 prev_bvec = bvec;
1095
1096                                 databuf_idx++;
1097                                 num_sectors += bvec->bv_len >> 9;
1098
1099                         } /* bio_for_each_segment */
1100
1101                 } /* rq_for_each_bio */
1102         }
1103
1104         /* Handle the last one */
1105         if (blkvsc_req) {
1106                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1107                            blkdev, req, blkvsc_req->group,
1108                            blkvsc_req->group->outstanding);
1109
1110                 blkvsc_req->sector_start = start_sector;
1111                 sector_div(blkvsc_req->sector_start,
1112                            (blkdev->sector_size >> 9));
1113
1114                 blkvsc_req->sector_count = num_sectors /
1115                                            (blkdev->sector_size >> 9);
1116
1117                 blkvsc_init_rw(blkvsc_req);
1118         }
1119
1120         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1121                 if (pending) {
1122                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1123                                    "pending_list - blkvsc_req %p start_sect %lu"
1124                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1125                                    (unsigned long)blkvsc_req->sector_start,
1126                                    blkvsc_req->sector_count,
1127                                    (unsigned long)start_sector,
1128                                    (unsigned long)num_sectors);
1129
1130                         list_add_tail(&blkvsc_req->pend_entry,
1131                                       &blkdev->pending_list);
1132                 } else {
1133                         ret = blkvsc_submit_request(blkvsc_req,
1134                                                     blkvsc_request_completion);
1135                         if (ret == -1) {
1136                                 pending = 1;
1137                                 list_add_tail(&blkvsc_req->pend_entry,
1138                                               &blkdev->pending_list);
1139                         }
1140
1141                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1142                                    "start_sect %lu sect_count %ld (%lu %ld) "
1143                                    "ret %d\n", blkvsc_req,
1144                                    (unsigned long)blkvsc_req->sector_start,
1145                                    blkvsc_req->sector_count,
1146                                    (unsigned long)start_sector,
1147                                    num_sectors, ret);
1148                 }
1149         }
1150
1151         return pending;
1152 }
1153
1154 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1155 {
1156         struct blkvsc_request *blkvsc_req =
1157                         (struct blkvsc_request *)request->context;
1158         struct block_device_context *blkdev =
1159                         (struct block_device_context *)blkvsc_req->dev;
1160         struct scsi_sense_hdr sense_hdr;
1161         struct vmscsi_request *vm_srb;
1162
1163         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1164                    blkvsc_req);
1165
1166         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
1167         blkdev->num_outstanding_reqs--;
1168
1169         if (vm_srb->scsi_status)
1170                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1171                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1172                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1173
1174         complete(&blkvsc_req->request.wait_event);
1175 }
1176
1177 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1178 {
1179         struct blkvsc_request *blkvsc_req =
1180                         (struct blkvsc_request *)request->context;
1181         struct block_device_context *blkdev =
1182                         (struct block_device_context *)blkvsc_req->dev;
1183         unsigned long flags;
1184         struct blkvsc_request *comp_req, *tmp;
1185         struct vmscsi_request *vm_srb;
1186
1187         /* ASSERT(blkvsc_req->group); */
1188
1189         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1190                    "sect_start %lu sect_count %ld len %d group outstd %d "
1191                    "total outstd %d\n",
1192                    blkdev, blkvsc_req, blkvsc_req->group,
1193                    (blkvsc_req->write) ? "WRITE" : "READ",
1194                    (unsigned long)blkvsc_req->sector_start,
1195                    blkvsc_req->sector_count,
1196                    blkvsc_req->request.data_buffer.len,
1197                    blkvsc_req->group->outstanding,
1198                    blkdev->num_outstanding_reqs);
1199
1200         spin_lock_irqsave(&blkdev->lock, flags);
1201
1202         blkdev->num_outstanding_reqs--;
1203         blkvsc_req->group->outstanding--;
1204
1205         /*
1206          * Only start processing when all the blkvsc_reqs are
1207          * completed. This guarantees no out-of-order blkvsc_req
1208          * completion when calling end_that_request_first()
1209          */
1210         if (blkvsc_req->group->outstanding == 0) {
1211                 list_for_each_entry_safe(comp_req, tmp,
1212                                          &blkvsc_req->group->blkvsc_req_list,
1213                                          req_entry) {
1214                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1215                                    "sect_start %lu sect_count %ld\n",
1216                                    comp_req,
1217                                    (unsigned long)comp_req->sector_start,
1218                                    comp_req->sector_count);
1219
1220                         list_del(&comp_req->req_entry);
1221
1222                         vm_srb =
1223                         &comp_req->request.vstor_packet.vm_srb;
1224                         if (!__blk_end_request(comp_req->req,
1225                                 (!vm_srb->scsi_status ? 0 : -EIO),
1226                                 comp_req->sector_count * blkdev->sector_size)) {
1227                                 /*
1228                                  * All the sectors have been xferred ie the
1229                                  * request is done
1230                                  */
1231                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1232                                            comp_req->req);
1233                                 kmem_cache_free(blkdev->request_pool,
1234                                                 comp_req->group);
1235                         }
1236
1237                         kmem_cache_free(blkdev->request_pool, comp_req);
1238                 }
1239
1240                 if (!blkdev->shutting_down) {
1241                         blkvsc_do_pending_reqs(blkdev);
1242                         blk_start_queue(blkdev->gd->queue);
1243                         blkvsc_request(blkdev->gd->queue);
1244                 }
1245         }
1246
1247         spin_unlock_irqrestore(&blkdev->lock, flags);
1248 }
1249
1250 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1251 {
1252         struct blkvsc_request *pend_req, *tmp;
1253         struct blkvsc_request *comp_req, *tmp2;
1254         struct vmscsi_request *vm_srb;
1255
1256         int ret = 0;
1257
1258         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1259
1260         /* Flush the pending list first */
1261         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1262                                  pend_entry) {
1263                 /*
1264                  * The pend_req could be part of a partially completed
1265                  * request. If so, complete those req first until we
1266                  * hit the pend_req
1267                  */
1268                 list_for_each_entry_safe(comp_req, tmp2,
1269                                          &pend_req->group->blkvsc_req_list,
1270                                          req_entry) {
1271                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1272                                    "sect_start %lu sect_count %ld\n",
1273                                    comp_req,
1274                                    (unsigned long) comp_req->sector_start,
1275                                    comp_req->sector_count);
1276
1277                         if (comp_req == pend_req)
1278                                 break;
1279
1280                         list_del(&comp_req->req_entry);
1281
1282                         if (comp_req->req) {
1283                                 vm_srb =
1284                                 &comp_req->request.vstor_packet.
1285                                 vm_srb;
1286                                 ret = __blk_end_request(comp_req->req,
1287                                         (!vm_srb->scsi_status ? 0 : -EIO),
1288                                         comp_req->sector_count *
1289                                         blkdev->sector_size);
1290
1291                                 /* FIXME: shouldn't this do more than return? */
1292                                 if (ret)
1293                                         goto out;
1294                         }
1295
1296                         kmem_cache_free(blkdev->request_pool, comp_req);
1297                 }
1298
1299                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1300                            pend_req);
1301
1302                 list_del(&pend_req->pend_entry);
1303
1304                 list_del(&pend_req->req_entry);
1305
1306                 if (comp_req->req) {
1307                         if (!__blk_end_request(pend_req->req, -EIO,
1308                                                pend_req->sector_count *
1309                                                blkdev->sector_size)) {
1310                                 /*
1311                                  * All the sectors have been xferred ie the
1312                                  * request is done
1313                                  */
1314                                 DPRINT_DBG(BLKVSC_DRV,
1315                                            "blkvsc_cancel_pending_reqs() - "
1316                                            "req %p COMPLETED\n", pend_req->req);
1317                                 kmem_cache_free(blkdev->request_pool,
1318                                                 pend_req->group);
1319                         }
1320                 }
1321
1322                 kmem_cache_free(blkdev->request_pool, pend_req);
1323         }
1324
1325 out:
1326         return ret;
1327 }
1328
1329 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1330 {
1331         struct blkvsc_request *pend_req, *tmp;
1332         int ret = 0;
1333
1334         /* Flush the pending list first */
1335         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1336                                  pend_entry) {
1337                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1338                            pend_req);
1339
1340                 ret = blkvsc_submit_request(pend_req,
1341                                             blkvsc_request_completion);
1342                 if (ret != 0)
1343                         break;
1344                 else
1345                         list_del(&pend_req->pend_entry);
1346         }
1347
1348         return ret;
1349 }
1350
1351 static void blkvsc_request(struct request_queue *queue)
1352 {
1353         struct block_device_context *blkdev = NULL;
1354         struct request *req;
1355         int ret = 0;
1356
1357         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1358         while ((req = blk_peek_request(queue)) != NULL) {
1359                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1360
1361                 blkdev = req->rq_disk->private_data;
1362                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1363                     blkdev->media_not_present) {
1364                         __blk_end_request_cur(req, 0);
1365                         continue;
1366                 }
1367
1368                 ret = blkvsc_do_pending_reqs(blkdev);
1369
1370                 if (ret != 0) {
1371                         DPRINT_DBG(BLKVSC_DRV,
1372                                    "- stop queue - pending_list not empty\n");
1373                         blk_stop_queue(queue);
1374                         break;
1375                 }
1376
1377                 blk_start_request(req);
1378
1379                 ret = blkvsc_do_request(blkdev, req);
1380                 if (ret > 0) {
1381                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1382                         blk_stop_queue(queue);
1383                         break;
1384                 } else if (ret < 0) {
1385                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1386                         blk_requeue_request(queue, req);
1387                         blk_stop_queue(queue);
1388                         break;
1389                 }
1390         }
1391 }
1392
1393 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1394 {
1395         struct block_device_context *blkdev = disk->private_data;
1396
1397         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1398                    blkdev->gd->disk_name);
1399
1400         mutex_lock(&blkvsc_mutex);
1401         spin_lock(&blkdev->lock);
1402         if (blkdev->users == 1) {
1403                 spin_unlock(&blkdev->lock);
1404                 blkvsc_do_flush(blkdev);
1405                 spin_lock(&blkdev->lock);
1406         }
1407
1408         blkdev->users--;
1409
1410         spin_unlock(&blkdev->lock);
1411         mutex_unlock(&blkvsc_mutex);
1412         return 0;
1413 }
1414
1415 static int blkvsc_revalidate_disk(struct gendisk *gd)
1416 {
1417         struct block_device_context *blkdev = gd->private_data;
1418
1419         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1420
1421         if (blkdev->device_type == DVD_TYPE) {
1422                 blkvsc_do_read_capacity(blkdev);
1423                 set_capacity(blkdev->gd, blkdev->capacity *
1424                             (blkdev->sector_size/512));
1425                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1426         }
1427         return 0;
1428 }
1429
1430 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1431 {
1432         sector_t total_sectors = get_capacity(bd->bd_disk);
1433         sector_t cylinder_times_heads = 0;
1434         sector_t temp = 0;
1435
1436         int sectors_per_track = 0;
1437         int heads = 0;
1438         int cylinders = 0;
1439         int rem = 0;
1440
1441         if (total_sectors > (65535 * 16 * 255))
1442                 total_sectors = (65535 * 16 * 255);
1443
1444         if (total_sectors >= (65535 * 16 * 63)) {
1445                 sectors_per_track = 255;
1446                 heads = 16;
1447
1448                 cylinder_times_heads = total_sectors;
1449                 /* sector_div stores the quotient in cylinder_times_heads */
1450                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1451         } else {
1452                 sectors_per_track = 17;
1453
1454                 cylinder_times_heads = total_sectors;
1455                 /* sector_div stores the quotient in cylinder_times_heads */
1456                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1457
1458                 temp = cylinder_times_heads + 1023;
1459                 /* sector_div stores the quotient in temp */
1460                 rem = sector_div(temp, 1024);
1461
1462                 heads = temp;
1463
1464                 if (heads < 4)
1465                         heads = 4;
1466
1467
1468                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1469                         sectors_per_track = 31;
1470                         heads = 16;
1471
1472                         cylinder_times_heads = total_sectors;
1473                         /*
1474                          * sector_div stores the quotient in
1475                          * cylinder_times_heads
1476                          */
1477                         rem = sector_div(cylinder_times_heads,
1478                                          sectors_per_track);
1479                 }
1480
1481                 if (cylinder_times_heads >= (heads * 1024)) {
1482                         sectors_per_track = 63;
1483                         heads = 16;
1484
1485                         cylinder_times_heads = total_sectors;
1486                         /*
1487                          * sector_div stores the quotient in
1488                          * cylinder_times_heads
1489                          */
1490                         rem = sector_div(cylinder_times_heads,
1491                                          sectors_per_track);
1492                 }
1493         }
1494
1495         temp = cylinder_times_heads;
1496         /* sector_div stores the quotient in temp */
1497         rem = sector_div(temp, heads);
1498         cylinders = temp;
1499
1500         hg->heads = heads;
1501         hg->sectors = sectors_per_track;
1502         hg->cylinders = cylinders;
1503
1504         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1505                     sectors_per_track);
1506
1507     return 0;
1508 }
1509
1510 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1511                         unsigned cmd, unsigned long argument)
1512 {
1513 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1514         int ret;
1515
1516         switch (cmd) {
1517         /*
1518          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1519          * than just a GUID. Commented it out for now.
1520          */
1521 #if 0
1522         case HDIO_GET_IDENTITY:
1523                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1524                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1525                                  blkdev->device_id_len))
1526                         ret = -EFAULT;
1527                 break;
1528 #endif
1529         default:
1530                 ret = -EINVAL;
1531                 break;
1532         }
1533
1534         return ret;
1535 }
1536
1537 static int __init blkvsc_init(void)
1538 {
1539         int ret;
1540
1541         BUILD_BUG_ON(sizeof(sector_t) != 8);
1542
1543         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1544
1545         ret = blkvsc_drv_init();
1546
1547         return ret;
1548 }
1549
1550 static void __exit blkvsc_exit(void)
1551 {
1552         blkvsc_drv_exit();
1553 }
1554
1555 MODULE_LICENSE("GPL");
1556 MODULE_VERSION(HV_DRV_VERSION);
1557 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1558 module_init(blkvsc_init);
1559 module_exit(blkvsc_exit);