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
 110 static const char *g_blk_driver_name = "blkvsc";
 111
 112 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 113 static const struct hv_guid g_blk_device_type = {
 114         .data = {
 115                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 116                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 117         }
 118 };
 119
 120 static int blk_vsc_on_device_add(struct hv_device *device,
 121                                 void *additional_info)
 122 {
 123         struct storvsc_device_info *device_info;
 124         int ret = 0;
 125
 126         device_info = (struct storvsc_device_info *)additional_info;
 127
 128         ret = stor_vsc_on_device_add(device, additional_info);
 129         if (ret != 0)
 130                 return ret;
 131
 132         /*
 133          * We need to use the device instance guid to set the path and target
 134          * id. For IDE devices, the device instance id is formatted as
 135          * <bus id> * - <device id> - 8899 - 000000000000.
 136          */
 137         device_info->path_id = device->dev_instance.data[3] << 24 |
 138                              device->dev_instance.data[2] << 16 |
 139                              device->dev_instance.data[1] << 8  |
 140                              device->dev_instance.data[0];
 141
 142         device_info->target_id = device->dev_instance.data[5] << 8 |
 143                                device->dev_instance.data[4];
 144
 145         return ret;
 146 }
 147
 148
 149 static int blk_vsc_initialize(struct hv_driver *driver)
 150 {
 151         struct storvsc_driver_object *stor_driver;
 152         int ret = 0;
 153
 154         stor_driver = hvdr_to_stordr(driver);
 155
 156         /* Make sure we are at least 2 pages since 1 page is used for control */
 157         /* ASSERT(stor_driver->RingBufferSize >= (PAGE_SIZE << 1)); */
 158
 159         driver->name = g_blk_driver_name;
 160         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 161
 162
 163         /*
 164          * Divide the ring buffer data size (which is 1 page less than the ring
 165          * buffer size since that page is reserved for the ring buffer indices)
 166          * by the max request size (which is
 167          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 168          */
 169         stor_driver->max_outstanding_req_per_channel =
 170                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 171                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 172                            sizeof(struct vstor_packet) + sizeof(u64),
 173                            sizeof(u64)));
 174
 175         DPRINT_INFO(BLKVSC, "max io outstd %u",
 176                     stor_driver->max_outstanding_req_per_channel);
 177
 178         /* Setup the dispatch table */
 179         stor_driver->base.dev_add = blk_vsc_on_device_add;
 180         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 181         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 182         stor_driver->on_io_request = stor_vsc_on_io_request;
 183
 184         return ret;
 185 }
 186
 187
 188 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 189                         void (*request_completion)(struct hv_storvsc_request *))
 190 {
 191         struct block_device_context *blkdev = blkvsc_req->dev;
 192         struct hv_device *device_ctx = blkdev->device_ctx;
 193         struct hv_driver *drv =
 194                         drv_to_hv_drv(device_ctx->device.driver);
 195         struct storvsc_driver_object *storvsc_drv_obj =
 196                         drv->priv;
 197         struct hv_storvsc_request *storvsc_req;
 198         struct vmscsi_request *vm_srb;
 199         int ret;
 200
 201         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 202                    "req %p type %s start_sector %lu count %ld offset %d "
 203                    "len %d\n", blkvsc_req,
 204                    (blkvsc_req->write) ? "WRITE" : "READ",
 205                    (unsigned long) blkvsc_req->sector_start,
 206                    blkvsc_req->sector_count,
 207                    blkvsc_req->request.data_buffer.offset,
 208                    blkvsc_req->request.data_buffer.len);
 209 #if 0
 210         for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
 211                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 212                            "req %p pfn[%d] %llx\n",
 213                            blkvsc_req, i,
 214                            blkvsc_req->request.data_buffer.pfn_array[i]);
 215         }
 216 #endif
 217
 218         storvsc_req = &blkvsc_req->request;
 219         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 220
 221         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 222
 223         storvsc_req->on_io_completion = request_completion;
 224         storvsc_req->context = blkvsc_req;
 225
 226         vm_srb->port_number = blkdev->port;
 227         vm_srb->path_id = blkdev->path;
 228         vm_srb->target_id = blkdev->target;
 229         vm_srb->lun = 0;         /* this is not really used at all */
 230
 231         vm_srb->cdb_length = blkvsc_req->cmd_len;
 232
 233         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 234
 235         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 236
 237         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 238                                            &blkvsc_req->request);
 239         if (ret == 0)
 240                 blkdev->num_outstanding_reqs++;
 241
 242         return ret;
 243 }
 244
 245
 246 static unsigned int blkvsc_check_events(struct gendisk *gd,
 247                                         unsigned int clearing)
 248 {
 249         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
 250         return DISK_EVENT_MEDIA_CHANGE;
 251 }
 252
 253
 254 /* Static decl */
 255 static DEFINE_MUTEX(blkvsc_mutex);
 256 static int blkvsc_probe(struct device *dev);
 257 static int blkvsc_remove(struct device *device);
 258 static void blkvsc_shutdown(struct device *device);
 259
 260 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 261 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 262 static int blkvsc_revalidate_disk(struct gendisk *gd);
 263 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 264 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 265                         unsigned cmd, unsigned long argument);
 266 static void blkvsc_request(struct request_queue *queue);
 267 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 268 static int blkvsc_do_request(struct block_device_context *blkdev,
 269                              struct request *req);
 270 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 271 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 272 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 273 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 274 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 275 static int blkvsc_do_flush(struct block_device_context *blkdev);
 276 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 277 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 278
 279 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 280 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 281 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 282
 283 /* The one and only one */
 284 static  struct storvsc_driver_object g_blkvsc_drv;
 285
 286 static const struct block_device_operations block_ops = {
 287         .owner = THIS_MODULE,
 288         .open = blkvsc_open,
 289         .release = blkvsc_release,
 290         .check_events = blkvsc_check_events,
 291         .revalidate_disk = blkvsc_revalidate_disk,
 292         .getgeo = blkvsc_getgeo,
 293         .ioctl  = blkvsc_ioctl,
 294 };
 295
 296 /*
 297  * blkvsc_drv_init -  BlkVsc driver initialization.
 298  */
 299 static int blkvsc_drv_init(void)
 300 {
 301         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 302         struct hv_driver *drv = &g_blkvsc_drv.base;
 303         int ret;
 304
 305         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 306
 307         drv->priv = storvsc_drv_obj;
 308
 309         /* Callback to client driver to complete the initialization */
 310         blk_vsc_initialize(&storvsc_drv_obj->base);
 311
 312         drv->driver.name = storvsc_drv_obj->base.name;
 313
 314         drv->driver.probe = blkvsc_probe;
 315         drv->driver.remove = blkvsc_remove;
 316         drv->driver.shutdown = blkvsc_shutdown;
 317
 318         /* The driver belongs to vmbus */
 319         ret = vmbus_child_driver_register(&drv->driver);
 320
 321         return ret;
 322 }
 323
 324 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 325 {
 326         struct device **curr = (struct device **)data;
 327         *curr = dev;
 328         return 1; /* stop iterating */
 329 }
 330
 331 static void blkvsc_drv_exit(void)
 332 {
 333         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 334         struct hv_driver *drv = &g_blkvsc_drv.base;
 335         struct device *current_dev;
 336         int ret;
 337
 338         while (1) {
 339                 current_dev = NULL;
 340
 341                 /* Get the device */
 342                 ret = driver_for_each_device(&drv->driver, NULL,
 343                                              (void *) &current_dev,
 344                                              blkvsc_drv_exit_cb);
 345
 346                 if (ret)
 347                         DPRINT_WARN(BLKVSC_DRV,
 348                                     "driver_for_each_device returned %d", ret);
 349
 350
 351                 if (current_dev == NULL)
 352                         break;
 353
 354                 /* Initiate removal from the top-down */
 355                 device_unregister(current_dev);
 356         }
 357
 358         if (storvsc_drv_obj->base.cleanup)
 359                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 360
 361         vmbus_child_driver_unregister(&drv->driver);
 362
 363         return;
 364 }
 365
 366 /*
 367  * blkvsc_probe - Add a new device for this driver
 368  */
 369 static int blkvsc_probe(struct device *device)
 370 {
 371         struct hv_driver *drv =
 372                                 drv_to_hv_drv(device->driver);
 373         struct storvsc_driver_object *storvsc_drv_obj =
 374                                 drv->priv;
 375         struct hv_device *device_obj = device_to_hv_device(device);
 376
 377         struct block_device_context *blkdev = NULL;
 378         struct storvsc_device_info device_info;
 379         int major = 0;
 380         int devnum = 0;
 381         int ret = 0;
 382         static int ide0_registered;
 383         static int ide1_registered;
 384
 385         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 386
 387         if (!storvsc_drv_obj->base.dev_add) {
 388                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 389                 ret = -1;
 390                 goto Cleanup;
 391         }
 392
 393         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 394         if (!blkdev) {
 395                 ret = -ENOMEM;
 396                 goto Cleanup;
 397         }
 398
 399         INIT_LIST_HEAD(&blkdev->pending_list);
 400
 401         /* Initialize what we can here */
 402         spin_lock_init(&blkdev->lock);
 403
 404         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 405         /*      sizeof(struct blkvsc_request)); */
 406
 407         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 408                                         sizeof(struct blkvsc_request), 0,
 409                                         SLAB_HWCACHE_ALIGN, NULL);
 410         if (!blkdev->request_pool) {
 411                 ret = -ENOMEM;
 412                 goto Cleanup;
 413         }
 414
 415
 416         /* Call to the vsc driver to add the device */
 417         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 418         if (ret != 0) {
 419                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 420                 goto Cleanup;
 421         }
 422
 423         blkdev->device_ctx = device_obj;
 424         /* this identified the device 0 or 1 */
 425         blkdev->target = device_info.target_id;
 426         /* this identified the ide ctrl 0 or 1 */
 427         blkdev->path = device_info.path_id;
 428
 429         dev_set_drvdata(device, blkdev);
 430
 431         /* Calculate the major and device num */
 432         if (blkdev->path == 0) {
 433                 major = IDE0_MAJOR;
 434                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 435
 436                 if (!ide0_registered) {
 437                         ret = register_blkdev(major, "ide");
 438                         if (ret != 0) {
 439                                 DPRINT_ERR(BLKVSC_DRV,
 440                                            "register_blkdev() failed! ret %d",
 441                                            ret);
 442                                 goto Remove;
 443                         }
 444
 445                         ide0_registered = 1;
 446                 }
 447         } else if (blkdev->path == 1) {
 448                 major = IDE1_MAJOR;
 449                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 450
 451                 if (!ide1_registered) {
 452                         ret = register_blkdev(major, "ide");
 453                         if (ret != 0) {
 454                                 DPRINT_ERR(BLKVSC_DRV,
 455                                            "register_blkdev() failed! ret %d",
 456                                            ret);
 457                                 goto Remove;
 458                         }
 459
 460                         ide1_registered = 1;
 461                 }
 462         } else {
 463                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 464                 ret = -1;
 465                 goto Cleanup;
 466         }
 467
 468         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 469
 470         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 471         if (!blkdev->gd) {
 472                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 473                 ret = -1;
 474                 goto Cleanup;
 475         }
 476
 477         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 478
 479         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 480         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 481         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 482         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 483         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 484
 485         blkdev->gd->major = major;
 486         if (devnum == 1 || devnum == 3)
 487                 blkdev->gd->first_minor = BLKVSC_MINORS;
 488         else
 489                 blkdev->gd->first_minor = 0;
 490         blkdev->gd->fops = &block_ops;
 491         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 492         blkdev->gd->private_data = blkdev;
 493         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 494         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 495
 496         blkvsc_do_inquiry(blkdev);
 497         if (blkdev->device_type == DVD_TYPE) {
 498                 set_disk_ro(blkdev->gd, 1);
 499                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 500                 blkvsc_do_read_capacity(blkdev);
 501         } else {
 502                 blkvsc_do_read_capacity16(blkdev);
 503         }
 504
 505         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 506         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 507         /* go! */
 508         add_disk(blkdev->gd);
 509
 510         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 511                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 512                     blkdev->sector_size);
 513
 514         return ret;
 515
 516 Remove:
 517         storvsc_drv_obj->base.dev_rm(device_obj);
 518
 519 Cleanup:
 520         if (blkdev) {
 521                 if (blkdev->request_pool) {
 522                         kmem_cache_destroy(blkdev->request_pool);
 523                         blkdev->request_pool = NULL;
 524                 }
 525                 kfree(blkdev);
 526                 blkdev = NULL;
 527         }
 528
 529         return ret;
 530 }
 531
 532 static void blkvsc_shutdown(struct device *device)
 533 {
 534         struct block_device_context *blkdev = dev_get_drvdata(device);
 535         unsigned long flags;
 536
 537         if (!blkdev)
 538                 return;
 539
 540         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 541                    blkdev->users, blkdev->gd->disk_name);
 542
 543         spin_lock_irqsave(&blkdev->lock, flags);
 544
 545         blkdev->shutting_down = 1;
 546
 547         blk_stop_queue(blkdev->gd->queue);
 548
 549         spin_unlock_irqrestore(&blkdev->lock, flags);
 550
 551         while (blkdev->num_outstanding_reqs) {
 552                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 553                             blkdev->num_outstanding_reqs);
 554                 udelay(100);
 555         }
 556
 557         blkvsc_do_flush(blkdev);
 558
 559         spin_lock_irqsave(&blkdev->lock, flags);
 560
 561         blkvsc_cancel_pending_reqs(blkdev);
 562
 563         spin_unlock_irqrestore(&blkdev->lock, flags);
 564 }
 565
 566 static int blkvsc_do_flush(struct block_device_context *blkdev)
 567 {
 568         struct blkvsc_request *blkvsc_req;
 569
 570         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 571
 572         if (blkdev->device_type != HARDDISK_TYPE)
 573                 return 0;
 574
 575         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 576         if (!blkvsc_req)
 577                 return -ENOMEM;
 578
 579         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 580         init_completion(&blkvsc_req->request.wait_event);
 581         blkvsc_req->dev = blkdev;
 582         blkvsc_req->req = NULL;
 583         blkvsc_req->write = 0;
 584
 585         blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 586         blkvsc_req->request.data_buffer.offset = 0;
 587         blkvsc_req->request.data_buffer.len = 0;
 588
 589         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 590         blkvsc_req->cmd_len = 10;
 591
 592         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 593
 594         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 595
 596         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 597
 598         return 0;
 599 }
 600
 601 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 602 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 603 {
 604         struct blkvsc_request *blkvsc_req;
 605         struct page *page_buf;
 606         unsigned char *buf;
 607         unsigned char device_type;
 608
 609         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 610
 611         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 612         if (!blkvsc_req)
 613                 return -ENOMEM;
 614
 615         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 616         page_buf = alloc_page(GFP_KERNEL);
 617         if (!page_buf) {
 618                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 619                 return -ENOMEM;
 620         }
 621
 622         init_completion(&blkvsc_req->request.wait_event);
 623         blkvsc_req->dev = blkdev;
 624         blkvsc_req->req = NULL;
 625         blkvsc_req->write = 0;
 626
 627         blkvsc_req->request.data_buffer.pfn_array[0] =
 628         page_to_pfn(page_buf);
 629         blkvsc_req->request.data_buffer.offset = 0;
 630         blkvsc_req->request.data_buffer.len = 64;
 631
 632         blkvsc_req->cmnd[0] = INQUIRY;
 633         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 634         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 635         blkvsc_req->cmnd[4] = 64;
 636         blkvsc_req->cmd_len = 6;
 637
 638         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 639
 640         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 641                    blkvsc_req);
 642
 643         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 644
 645         buf = kmap(page_buf);
 646
 647         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 648         /* be to le */
 649         device_type = buf[0] & 0x1F;
 650
 651         if (device_type == 0x0) {
 652                 blkdev->device_type = HARDDISK_TYPE;
 653         } else if (device_type == 0x5) {
 654                 blkdev->device_type = DVD_TYPE;
 655         } else {
 656                 /* TODO: this is currently unsupported device type */
 657                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 658         }
 659
 660         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 661
 662         blkdev->device_id_len = buf[7];
 663         if (blkdev->device_id_len > 64)
 664                 blkdev->device_id_len = 64;
 665
 666         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 667         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 668          * blkdev->device_id_len); */
 669
 670         kunmap(page_buf);
 671
 672         __free_page(page_buf);
 673
 674         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 675
 676         return 0;
 677 }
 678
 679 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 680 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 681 {
 682         struct blkvsc_request *blkvsc_req;
 683         struct page *page_buf;
 684         unsigned char *buf;
 685         struct scsi_sense_hdr sense_hdr;
 686         struct vmscsi_request *vm_srb;
 687
 688         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 689
 690         blkdev->sector_size = 0;
 691         blkdev->capacity = 0;
 692         blkdev->media_not_present = 0; /* assume a disk is present */
 693
 694         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 695         if (!blkvsc_req)
 696                 return -ENOMEM;
 697
 698         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 699         page_buf = alloc_page(GFP_KERNEL);
 700         if (!page_buf) {
 701                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 702                 return -ENOMEM;
 703         }
 704
 705         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 706         init_completion(&blkvsc_req->request.wait_event);
 707         blkvsc_req->dev = blkdev;
 708         blkvsc_req->req = NULL;
 709         blkvsc_req->write = 0;
 710
 711         blkvsc_req->request.data_buffer.pfn_array[0] =
 712         page_to_pfn(page_buf);
 713         blkvsc_req->request.data_buffer.offset = 0;
 714         blkvsc_req->request.data_buffer.len = 8;
 715
 716         blkvsc_req->cmnd[0] = READ_CAPACITY;
 717         blkvsc_req->cmd_len = 16;
 718
 719         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 720
 721         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 722                    blkvsc_req);
 723
 724         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 725
 726         /* check error */
 727         if (vm_srb->scsi_status) {
 728                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 729                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 730
 731                 if (sense_hdr.asc == 0x3A) {
 732                         /* Medium not present */
 733                         blkdev->media_not_present = 1;
 734                 }
 735                 return 0;
 736         }
 737         buf = kmap(page_buf);
 738
 739         /* be to le */
 740         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 741                             (buf[2] << 8) | buf[3]) + 1;
 742         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 743                               (buf[6] << 8) | buf[7];
 744
 745         kunmap(page_buf);
 746
 747         __free_page(page_buf);
 748
 749         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 750
 751         return 0;
 752 }
 753
 754 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 755 {
 756         struct blkvsc_request *blkvsc_req;
 757         struct page *page_buf;
 758         unsigned char *buf;
 759         struct scsi_sense_hdr sense_hdr;
 760         struct vmscsi_request *vm_srb;
 761
 762         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 763
 764         blkdev->sector_size = 0;
 765         blkdev->capacity = 0;
 766         blkdev->media_not_present = 0; /* assume a disk is present */
 767
 768         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 769         if (!blkvsc_req)
 770                 return -ENOMEM;
 771
 772         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 773         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 774         page_buf = alloc_page(GFP_KERNEL);
 775         if (!page_buf) {
 776                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 777                 return -ENOMEM;
 778         }
 779
 780         init_completion(&blkvsc_req->request.wait_event);
 781         blkvsc_req->dev = blkdev;
 782         blkvsc_req->req = NULL;
 783         blkvsc_req->write = 0;
 784
 785         blkvsc_req->request.data_buffer.pfn_array[0] =
 786         page_to_pfn(page_buf);
 787         blkvsc_req->request.data_buffer.offset = 0;
 788         blkvsc_req->request.data_buffer.len = 12;
 789
 790         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 791         blkvsc_req->cmd_len = 16;
 792
 793         /*
 794          * Set this here since the completion routine may be invoked
 795          * and completed before we return
 796          */
 797
 798         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 799
 800         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 801                    blkvsc_req);
 802
 803         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 804
 805         /* check error */
 806         if (vm_srb->scsi_status) {
 807                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 808                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 809                 if (sense_hdr.asc == 0x3A) {
 810                         /* Medium not present */
 811                         blkdev->media_not_present = 1;
 812                 }
 813                 return 0;
 814         }
 815         buf = kmap(page_buf);
 816
 817         /* be to le */
 818         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 819         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 820
 821 #if 0
 822         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 823                             (buf[2] << 8) | buf[3]) + 1;
 824         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 825                               (buf[6] << 8) | buf[7];
 826 #endif
 827
 828         kunmap(page_buf);
 829
 830         __free_page(page_buf);
 831
 832         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 833
 834         return 0;
 835 }
 836
 837 /*
 838  * blkvsc_remove() - Callback when our device is removed
 839  */
 840 static int blkvsc_remove(struct device *device)
 841 {
 842         struct hv_driver *drv =
 843                                 drv_to_hv_drv(device->driver);
 844         struct storvsc_driver_object *storvsc_drv_obj =
 845                                 drv->priv;
 846         struct hv_device *device_obj = device_to_hv_device(device);
 847         struct block_device_context *blkdev = dev_get_drvdata(device);
 848         unsigned long flags;
 849         int ret;
 850
 851         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 852
 853         if (!storvsc_drv_obj->base.dev_rm)
 854                 return -1;
 855
 856         /*
 857          * Call to the vsc driver to let it know that the device is being
 858          * removed
 859          */
 860         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 861         if (ret != 0) {
 862                 /* TODO: */
 863                 DPRINT_ERR(BLKVSC_DRV,
 864                            "unable to remove blkvsc device (ret %d)", ret);
 865         }
 866
 867         /* Get to a known state */
 868         spin_lock_irqsave(&blkdev->lock, flags);
 869
 870         blkdev->shutting_down = 1;
 871
 872         blk_stop_queue(blkdev->gd->queue);
 873
 874         spin_unlock_irqrestore(&blkdev->lock, flags);
 875
 876         while (blkdev->num_outstanding_reqs) {
 877                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 878                             blkdev->num_outstanding_reqs);
 879                 udelay(100);
 880         }
 881
 882         blkvsc_do_flush(blkdev);
 883
 884         spin_lock_irqsave(&blkdev->lock, flags);
 885
 886         blkvsc_cancel_pending_reqs(blkdev);
 887
 888         spin_unlock_irqrestore(&blkdev->lock, flags);
 889
 890         blk_cleanup_queue(blkdev->gd->queue);
 891
 892         del_gendisk(blkdev->gd);
 893
 894         kmem_cache_destroy(blkdev->request_pool);
 895
 896         kfree(blkdev);
 897
 898         return ret;
 899 }
 900
 901 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 902 {
 903         /* ASSERT(blkvsc_req->req); */
 904         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 905
 906         blkvsc_req->cmd_len = 16;
 907
 908         if (blkvsc_req->sector_start > 0xffffffff) {
 909                 if (rq_data_dir(blkvsc_req->req)) {
 910                         blkvsc_req->write = 1;
 911                         blkvsc_req->cmnd[0] = WRITE_16;
 912                 } else {
 913                         blkvsc_req->write = 0;
 914                         blkvsc_req->cmnd[0] = READ_16;
 915                 }
 916
 917                 blkvsc_req->cmnd[1] |=
 918                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 919
 920                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 921                                 cpu_to_be64(blkvsc_req->sector_start);
 922                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 923                                 cpu_to_be32(blkvsc_req->sector_count);
 924         } else if ((blkvsc_req->sector_count > 0xff) ||
 925                    (blkvsc_req->sector_start > 0x1fffff)) {
 926                 if (rq_data_dir(blkvsc_req->req)) {
 927                         blkvsc_req->write = 1;
 928                         blkvsc_req->cmnd[0] = WRITE_10;
 929                 } else {
 930                         blkvsc_req->write = 0;
 931                         blkvsc_req->cmnd[0] = READ_10;
 932                 }
 933
 934                 blkvsc_req->cmnd[1] |=
 935                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 936
 937                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 938                                 cpu_to_be32(blkvsc_req->sector_start);
 939                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 940                                 cpu_to_be16(blkvsc_req->sector_count);
 941         } else {
 942                 if (rq_data_dir(blkvsc_req->req)) {
 943                         blkvsc_req->write = 1;
 944                         blkvsc_req->cmnd[0] = WRITE_6;
 945                 } else {
 946                         blkvsc_req->write = 0;
 947                         blkvsc_req->cmnd[0] = READ_6;
 948                 }
 949
 950                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 951                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 952                 blkvsc_req->cmnd[1] &= 0x1f;
 953                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 954         }
 955 }
 956
 957
 958 /*
 959  * We break the request into 1 or more blkvsc_requests and submit
 960  * them.  If we can't submit them all, we put them on the
 961  * pending_list. The blkvsc_request() will work on the pending_list.
 962  */
 963 static int blkvsc_do_request(struct block_device_context *blkdev,
 964                              struct request *req)
 965 {
 966         struct bio *bio = NULL;
 967         struct bio_vec *bvec = NULL;
 968         struct bio_vec *prev_bvec = NULL;
 969         struct blkvsc_request *blkvsc_req = NULL;
 970         struct blkvsc_request *tmp;
 971         int databuf_idx = 0;
 972         int seg_idx = 0;
 973         sector_t start_sector;
 974         unsigned long num_sectors = 0;
 975         int ret = 0;
 976         int pending = 0;
 977         struct blkvsc_request_group *group = NULL;
 978
 979         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 980                   (unsigned long)blk_rq_pos(req));
 981
 982         /* Create a group to tie req to list of blkvsc_reqs */
 983         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 984         if (!group)
 985                 return -ENOMEM;
 986
 987         INIT_LIST_HEAD(&group->blkvsc_req_list);
 988         group->outstanding = group->status = 0;
 989
 990         start_sector = blk_rq_pos(req);
 991
 992         /* foreach bio in the request */
 993         if (req->bio) {
 994                 for (bio = req->bio; bio; bio = bio->bi_next) {
 995                         /*
 996                          * Map this bio into an existing or new storvsc request
 997                          */
 998                         bio_for_each_segment(bvec, bio, seg_idx) {
 999                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
1000                                            "- req %p bio %p bvec %p seg_idx %d "
1001                                            "databuf_idx %d\n", req, bio, bvec,
1002                                            seg_idx, databuf_idx);
1003
1004                                 /* Get a new storvsc request */
1005                                 /* 1st-time */
1006                                 if ((!blkvsc_req) ||
1007                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
1008                                     /* hole at the begin of page */
1009                                     || (bvec->bv_offset != 0) ||
1010                                     /* hold at the end of page */
1011                                     (prev_bvec &&
1012                                      (prev_bvec->bv_len != PAGE_SIZE))) {
1013                                         /* submit the prev one */
1014                                         if (blkvsc_req) {
1015                                                 blkvsc_req->sector_start = start_sector;
1016                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
1017
1018                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
1019                                                 blkvsc_init_rw(blkvsc_req);
1020                                         }
1021
1022                                         /*
1023                                          * Create new blkvsc_req to represent
1024                                          * the current bvec
1025                                          */
1026                                         blkvsc_req =
1027                                         kmem_cache_zalloc(
1028                                         blkdev->request_pool, GFP_ATOMIC);
1029                                         if (!blkvsc_req) {
1030                                                 /* free up everything */
1031                                                 list_for_each_entry_safe(
1032                                                         blkvsc_req, tmp,
1033                                                         &group->blkvsc_req_list,
1034                                                         req_entry) {
1035                                                         list_del(&blkvsc_req->req_entry);
1036                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
1037                                                 }
1038
1039                                                 kmem_cache_free(blkdev->request_pool, group);
1040                                                 return -ENOMEM;
1041                                         }
1042
1043                                         memset(blkvsc_req, 0,
1044                                                sizeof(struct blkvsc_request));
1045
1046                                         blkvsc_req->dev = blkdev;
1047                                         blkvsc_req->req = req;
1048                                         blkvsc_req->request.
1049                                         data_buffer.offset
1050                                         = bvec->bv_offset;
1051                                         blkvsc_req->request.
1052                                         data_buffer.len = 0;
1053
1054                                         /* Add to the group */
1055                                         blkvsc_req->group = group;
1056                                         blkvsc_req->group->outstanding++;
1057                                         list_add_tail(&blkvsc_req->req_entry,
1058                                                 &blkvsc_req->group->blkvsc_req_list);
1059
1060                                         start_sector += num_sectors;
1061                                         num_sectors = 0;
1062                                         databuf_idx = 0;
1063                                 }
1064
1065                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1066                                 blkvsc_req->request.data_buffer.
1067                                         pfn_array[databuf_idx]
1068                                                 = page_to_pfn(bvec->bv_page);
1069                                 blkvsc_req->request.data_buffer.len
1070                                         += bvec->bv_len;
1071
1072                                 prev_bvec = bvec;
1073
1074                                 databuf_idx++;
1075                                 num_sectors += bvec->bv_len >> 9;
1076
1077                         } /* bio_for_each_segment */
1078
1079                 } /* rq_for_each_bio */
1080         }
1081
1082         /* Handle the last one */
1083         if (blkvsc_req) {
1084                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1085                            blkdev, req, blkvsc_req->group,
1086                            blkvsc_req->group->outstanding);
1087
1088                 blkvsc_req->sector_start = start_sector;
1089                 sector_div(blkvsc_req->sector_start,
1090                            (blkdev->sector_size >> 9));
1091
1092                 blkvsc_req->sector_count = num_sectors /
1093                                            (blkdev->sector_size >> 9);
1094
1095                 blkvsc_init_rw(blkvsc_req);
1096         }
1097
1098         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1099                 if (pending) {
1100                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1101                                    "pending_list - blkvsc_req %p start_sect %lu"
1102                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1103                                    (unsigned long)blkvsc_req->sector_start,
1104                                    blkvsc_req->sector_count,
1105                                    (unsigned long)start_sector,
1106                                    (unsigned long)num_sectors);
1107
1108                         list_add_tail(&blkvsc_req->pend_entry,
1109                                       &blkdev->pending_list);
1110                 } else {
1111                         ret = blkvsc_submit_request(blkvsc_req,
1112                                                     blkvsc_request_completion);
1113                         if (ret == -1) {
1114                                 pending = 1;
1115                                 list_add_tail(&blkvsc_req->pend_entry,
1116                                               &blkdev->pending_list);
1117                         }
1118
1119                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1120                                    "start_sect %lu sect_count %ld (%lu %ld) "
1121                                    "ret %d\n", blkvsc_req,
1122                                    (unsigned long)blkvsc_req->sector_start,
1123                                    blkvsc_req->sector_count,
1124                                    (unsigned long)start_sector,
1125                                    num_sectors, ret);
1126                 }
1127         }
1128
1129         return pending;
1130 }
1131
1132 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1133 {
1134         struct blkvsc_request *blkvsc_req =
1135                         (struct blkvsc_request *)request->context;
1136         struct block_device_context *blkdev =
1137                         (struct block_device_context *)blkvsc_req->dev;
1138         struct scsi_sense_hdr sense_hdr;
1139         struct vmscsi_request *vm_srb;
1140
1141         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1142                    blkvsc_req);
1143
1144         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
1145         blkdev->num_outstanding_reqs--;
1146
1147         if (vm_srb->scsi_status)
1148                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1149                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1150                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1151
1152         complete(&blkvsc_req->request.wait_event);
1153 }
1154
1155 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1156 {
1157         struct blkvsc_request *blkvsc_req =
1158                         (struct blkvsc_request *)request->context;
1159         struct block_device_context *blkdev =
1160                         (struct block_device_context *)blkvsc_req->dev;
1161         unsigned long flags;
1162         struct blkvsc_request *comp_req, *tmp;
1163         struct vmscsi_request *vm_srb;
1164
1165         /* ASSERT(blkvsc_req->group); */
1166
1167         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1168                    "sect_start %lu sect_count %ld len %d group outstd %d "
1169                    "total outstd %d\n",
1170                    blkdev, blkvsc_req, blkvsc_req->group,
1171                    (blkvsc_req->write) ? "WRITE" : "READ",
1172                    (unsigned long)blkvsc_req->sector_start,
1173                    blkvsc_req->sector_count,
1174                    blkvsc_req->request.data_buffer.len,
1175                    blkvsc_req->group->outstanding,
1176                    blkdev->num_outstanding_reqs);
1177
1178         spin_lock_irqsave(&blkdev->lock, flags);
1179
1180         blkdev->num_outstanding_reqs--;
1181         blkvsc_req->group->outstanding--;
1182
1183         /*
1184          * Only start processing when all the blkvsc_reqs are
1185          * completed. This guarantees no out-of-order blkvsc_req
1186          * completion when calling end_that_request_first()
1187          */
1188         if (blkvsc_req->group->outstanding == 0) {
1189                 list_for_each_entry_safe(comp_req, tmp,
1190                                          &blkvsc_req->group->blkvsc_req_list,
1191                                          req_entry) {
1192                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1193                                    "sect_start %lu sect_count %ld\n",
1194                                    comp_req,
1195                                    (unsigned long)comp_req->sector_start,
1196                                    comp_req->sector_count);
1197
1198                         list_del(&comp_req->req_entry);
1199
1200                         vm_srb =
1201                         &comp_req->request.vstor_packet.vm_srb;
1202                         if (!__blk_end_request(comp_req->req,
1203                                 (!vm_srb->scsi_status ? 0 : -EIO),
1204                                 comp_req->sector_count * blkdev->sector_size)) {
1205                                 /*
1206                                  * All the sectors have been xferred ie the
1207                                  * request is done
1208                                  */
1209                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1210                                            comp_req->req);
1211                                 kmem_cache_free(blkdev->request_pool,
1212                                                 comp_req->group);
1213                         }
1214
1215                         kmem_cache_free(blkdev->request_pool, comp_req);
1216                 }
1217
1218                 if (!blkdev->shutting_down) {
1219                         blkvsc_do_pending_reqs(blkdev);
1220                         blk_start_queue(blkdev->gd->queue);
1221                         blkvsc_request(blkdev->gd->queue);
1222                 }
1223         }
1224
1225         spin_unlock_irqrestore(&blkdev->lock, flags);
1226 }
1227
1228 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1229 {
1230         struct blkvsc_request *pend_req, *tmp;
1231         struct blkvsc_request *comp_req, *tmp2;
1232         struct vmscsi_request *vm_srb;
1233
1234         int ret = 0;
1235
1236         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1237
1238         /* Flush the pending list first */
1239         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1240                                  pend_entry) {
1241                 /*
1242                  * The pend_req could be part of a partially completed
1243                  * request. If so, complete those req first until we
1244                  * hit the pend_req
1245                  */
1246                 list_for_each_entry_safe(comp_req, tmp2,
1247                                          &pend_req->group->blkvsc_req_list,
1248                                          req_entry) {
1249                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1250                                    "sect_start %lu sect_count %ld\n",
1251                                    comp_req,
1252                                    (unsigned long) comp_req->sector_start,
1253                                    comp_req->sector_count);
1254
1255                         if (comp_req == pend_req)
1256                                 break;
1257
1258                         list_del(&comp_req->req_entry);
1259
1260                         if (comp_req->req) {
1261                                 vm_srb =
1262                                 &comp_req->request.vstor_packet.
1263                                 vm_srb;
1264                                 ret = __blk_end_request(comp_req->req,
1265                                         (!vm_srb->scsi_status ? 0 : -EIO),
1266                                         comp_req->sector_count *
1267                                         blkdev->sector_size);
1268
1269                                 /* FIXME: shouldn't this do more than return? */
1270                                 if (ret)
1271                                         goto out;
1272                         }
1273
1274                         kmem_cache_free(blkdev->request_pool, comp_req);
1275                 }
1276
1277                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1278                            pend_req);
1279
1280                 list_del(&pend_req->pend_entry);
1281
1282                 list_del(&pend_req->req_entry);
1283
1284                 if (comp_req->req) {
1285                         if (!__blk_end_request(pend_req->req, -EIO,
1286                                                pend_req->sector_count *
1287                                                blkdev->sector_size)) {
1288                                 /*
1289                                  * All the sectors have been xferred ie the
1290                                  * request is done
1291                                  */
1292                                 DPRINT_DBG(BLKVSC_DRV,
1293                                            "blkvsc_cancel_pending_reqs() - "
1294                                            "req %p COMPLETED\n", pend_req->req);
1295                                 kmem_cache_free(blkdev->request_pool,
1296                                                 pend_req->group);
1297                         }
1298                 }
1299
1300                 kmem_cache_free(blkdev->request_pool, pend_req);
1301         }
1302
1303 out:
1304         return ret;
1305 }
1306
1307 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1308 {
1309         struct blkvsc_request *pend_req, *tmp;
1310         int ret = 0;
1311
1312         /* Flush the pending list first */
1313         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1314                                  pend_entry) {
1315                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1316                            pend_req);
1317
1318                 ret = blkvsc_submit_request(pend_req,
1319                                             blkvsc_request_completion);
1320                 if (ret != 0)
1321                         break;
1322                 else
1323                         list_del(&pend_req->pend_entry);
1324         }
1325
1326         return ret;
1327 }
1328
1329 static void blkvsc_request(struct request_queue *queue)
1330 {
1331         struct block_device_context *blkdev = NULL;
1332         struct request *req;
1333         int ret = 0;
1334
1335         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1336         while ((req = blk_peek_request(queue)) != NULL) {
1337                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1338
1339                 blkdev = req->rq_disk->private_data;
1340                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1341                     blkdev->media_not_present) {
1342                         __blk_end_request_cur(req, 0);
1343                         continue;
1344                 }
1345
1346                 ret = blkvsc_do_pending_reqs(blkdev);
1347
1348                 if (ret != 0) {
1349                         DPRINT_DBG(BLKVSC_DRV,
1350                                    "- stop queue - pending_list not empty\n");
1351                         blk_stop_queue(queue);
1352                         break;
1353                 }
1354
1355                 blk_start_request(req);
1356
1357                 ret = blkvsc_do_request(blkdev, req);
1358                 if (ret > 0) {
1359                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1360                         blk_stop_queue(queue);
1361                         break;
1362                 } else if (ret < 0) {
1363                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1364                         blk_requeue_request(queue, req);
1365                         blk_stop_queue(queue);
1366                         break;
1367                 }
1368         }
1369 }
1370
1371 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1372 {
1373         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1374
1375         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1376                    blkdev->gd->disk_name);
1377
1378         mutex_lock(&blkvsc_mutex);
1379         spin_lock(&blkdev->lock);
1380
1381         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1382                 spin_unlock(&blkdev->lock);
1383                 check_disk_change(bdev);
1384                 spin_lock(&blkdev->lock);
1385         }
1386
1387         blkdev->users++;
1388
1389         spin_unlock(&blkdev->lock);
1390         mutex_unlock(&blkvsc_mutex);
1391         return 0;
1392 }
1393
1394 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1395 {
1396         struct block_device_context *blkdev = disk->private_data;
1397
1398         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1399                    blkdev->gd->disk_name);
1400
1401         mutex_lock(&blkvsc_mutex);
1402         spin_lock(&blkdev->lock);
1403         if (blkdev->users == 1) {
1404                 spin_unlock(&blkdev->lock);
1405                 blkvsc_do_flush(blkdev);
1406                 spin_lock(&blkdev->lock);
1407         }
1408
1409         blkdev->users--;
1410
1411         spin_unlock(&blkdev->lock);
1412         mutex_unlock(&blkvsc_mutex);
1413         return 0;
1414 }
1415
1416 static int blkvsc_revalidate_disk(struct gendisk *gd)
1417 {
1418         struct block_device_context *blkdev = gd->private_data;
1419
1420         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1421
1422         if (blkdev->device_type == DVD_TYPE) {
1423                 blkvsc_do_read_capacity(blkdev);
1424                 set_capacity(blkdev->gd, blkdev->capacity *
1425                             (blkdev->sector_size/512));
1426                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1427         }
1428         return 0;
1429 }
1430
1431 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1432 {
1433         sector_t total_sectors = get_capacity(bd->bd_disk);
1434         sector_t cylinder_times_heads = 0;
1435         sector_t temp = 0;
1436
1437         int sectors_per_track = 0;
1438         int heads = 0;
1439         int cylinders = 0;
1440         int rem = 0;
1441
1442         if (total_sectors > (65535 * 16 * 255))
1443                 total_sectors = (65535 * 16 * 255);
1444
1445         if (total_sectors >= (65535 * 16 * 63)) {
1446                 sectors_per_track = 255;
1447                 heads = 16;
1448
1449                 cylinder_times_heads = total_sectors;
1450                 /* sector_div stores the quotient in cylinder_times_heads */
1451                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1452         } else {
1453                 sectors_per_track = 17;
1454
1455                 cylinder_times_heads = total_sectors;
1456                 /* sector_div stores the quotient in cylinder_times_heads */
1457                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1458
1459                 temp = cylinder_times_heads + 1023;
1460                 /* sector_div stores the quotient in temp */
1461                 rem = sector_div(temp, 1024);
1462
1463                 heads = temp;
1464
1465                 if (heads < 4)
1466                         heads = 4;
1467
1468
1469                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1470                         sectors_per_track = 31;
1471                         heads = 16;
1472
1473                         cylinder_times_heads = total_sectors;
1474                         /*
1475                          * sector_div stores the quotient in
1476                          * cylinder_times_heads
1477                          */
1478                         rem = sector_div(cylinder_times_heads,
1479                                          sectors_per_track);
1480                 }
1481
1482                 if (cylinder_times_heads >= (heads * 1024)) {
1483                         sectors_per_track = 63;
1484                         heads = 16;
1485
1486                         cylinder_times_heads = total_sectors;
1487                         /*
1488                          * sector_div stores the quotient in
1489                          * cylinder_times_heads
1490                          */
1491                         rem = sector_div(cylinder_times_heads,
1492                                          sectors_per_track);
1493                 }
1494         }
1495
1496         temp = cylinder_times_heads;
1497         /* sector_div stores the quotient in temp */
1498         rem = sector_div(temp, heads);
1499         cylinders = temp;
1500
1501         hg->heads = heads;
1502         hg->sectors = sectors_per_track;
1503         hg->cylinders = cylinders;
1504
1505         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1506                     sectors_per_track);
1507
1508     return 0;
1509 }
1510
1511 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1512                         unsigned cmd, unsigned long argument)
1513 {
1514 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1515         int ret;
1516
1517         switch (cmd) {
1518         /*
1519          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1520          * than just a GUID. Commented it out for now.
1521          */
1522 #if 0
1523         case HDIO_GET_IDENTITY:
1524                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1525                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1526                                  blkdev->device_id_len))
1527                         ret = -EFAULT;
1528                 break;
1529 #endif
1530         default:
1531                 ret = -EINVAL;
1532                 break;
1533         }
1534
1535         return ret;
1536 }
1537
1538 static int __init blkvsc_init(void)
1539 {
1540         int ret;
1541
1542         BUILD_BUG_ON(sizeof(sector_t) != 8);
1543
1544         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1545
1546         ret = blkvsc_drv_init();
1547
1548         return ret;
1549 }
1550
1551 static void __exit blkvsc_exit(void)
1552 {
1553         blkvsc_drv_exit();
1554 }
1555
1556 MODULE_LICENSE("GPL");
1557 MODULE_VERSION(HV_DRV_VERSION);
1558 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1559 module_init(blkvsc_init);
1560 module_exit(blkvsc_exit);