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 decl */
 247 static DEFINE_MUTEX(blkvsc_mutex);
 248 static int blkvsc_probe(struct device *dev);
 249 static int blkvsc_remove(struct device *device);
 250 static void blkvsc_shutdown(struct device *device);
 251
 252 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 253 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 254 static unsigned int blkvsc_check_events(struct gendisk *gd,
 255                                         unsigned int clearing);
 256 static int blkvsc_revalidate_disk(struct gendisk *gd);
 257 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 258 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 259                         unsigned cmd, unsigned long argument);
 260 static void blkvsc_request(struct request_queue *queue);
 261 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 262 static int blkvsc_do_request(struct block_device_context *blkdev,
 263                              struct request *req);
 264 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 265 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 266 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 267 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 268 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 269 static int blkvsc_do_flush(struct block_device_context *blkdev);
 270 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 271 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 272
 273 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 274 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 275 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 276
 277 /* The one and only one */
 278 static  struct storvsc_driver_object g_blkvsc_drv;
 279
 280 static const struct block_device_operations block_ops = {
 281         .owner = THIS_MODULE,
 282         .open = blkvsc_open,
 283         .release = blkvsc_release,
 284         .check_events = blkvsc_check_events,
 285         .revalidate_disk = blkvsc_revalidate_disk,
 286         .getgeo = blkvsc_getgeo,
 287         .ioctl  = blkvsc_ioctl,
 288 };
 289
 290 /*
 291  * blkvsc_drv_init -  BlkVsc driver initialization.
 292  */
 293 static int blkvsc_drv_init(void)
 294 {
 295         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 296         struct hv_driver *drv = &g_blkvsc_drv.base;
 297         int ret;
 298
 299         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 300
 301         drv->priv = storvsc_drv_obj;
 302
 303         /* Callback to client driver to complete the initialization */
 304         blk_vsc_initialize(&storvsc_drv_obj->base);
 305
 306         drv->driver.name = storvsc_drv_obj->base.name;
 307
 308         drv->driver.probe = blkvsc_probe;
 309         drv->driver.remove = blkvsc_remove;
 310         drv->driver.shutdown = blkvsc_shutdown;
 311
 312         /* The driver belongs to vmbus */
 313         ret = vmbus_child_driver_register(&drv->driver);
 314
 315         return ret;
 316 }
 317
 318 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 319 {
 320         struct device **curr = (struct device **)data;
 321         *curr = dev;
 322         return 1; /* stop iterating */
 323 }
 324
 325 static void blkvsc_drv_exit(void)
 326 {
 327         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 328         struct hv_driver *drv = &g_blkvsc_drv.base;
 329         struct device *current_dev;
 330         int ret;
 331
 332         while (1) {
 333                 current_dev = NULL;
 334
 335                 /* Get the device */
 336                 ret = driver_for_each_device(&drv->driver, NULL,
 337                                              (void *) &current_dev,
 338                                              blkvsc_drv_exit_cb);
 339
 340                 if (ret)
 341                         DPRINT_WARN(BLKVSC_DRV,
 342                                     "driver_for_each_device returned %d", ret);
 343
 344
 345                 if (current_dev == NULL)
 346                         break;
 347
 348                 /* Initiate removal from the top-down */
 349                 device_unregister(current_dev);
 350         }
 351
 352         if (storvsc_drv_obj->base.cleanup)
 353                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 354
 355         vmbus_child_driver_unregister(&drv->driver);
 356
 357         return;
 358 }
 359
 360 /*
 361  * blkvsc_probe - Add a new device for this driver
 362  */
 363 static int blkvsc_probe(struct device *device)
 364 {
 365         struct hv_driver *drv =
 366                                 drv_to_hv_drv(device->driver);
 367         struct storvsc_driver_object *storvsc_drv_obj =
 368                                 drv->priv;
 369         struct hv_device *device_obj = device_to_hv_device(device);
 370
 371         struct block_device_context *blkdev = NULL;
 372         struct storvsc_device_info device_info;
 373         int major = 0;
 374         int devnum = 0;
 375         int ret = 0;
 376         static int ide0_registered;
 377         static int ide1_registered;
 378
 379         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 380
 381         if (!storvsc_drv_obj->base.dev_add) {
 382                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 383                 ret = -1;
 384                 goto Cleanup;
 385         }
 386
 387         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 388         if (!blkdev) {
 389                 ret = -ENOMEM;
 390                 goto Cleanup;
 391         }
 392
 393         INIT_LIST_HEAD(&blkdev->pending_list);
 394
 395         /* Initialize what we can here */
 396         spin_lock_init(&blkdev->lock);
 397
 398         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 399         /*      sizeof(struct blkvsc_request)); */
 400
 401         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 402                                         sizeof(struct blkvsc_request), 0,
 403                                         SLAB_HWCACHE_ALIGN, NULL);
 404         if (!blkdev->request_pool) {
 405                 ret = -ENOMEM;
 406                 goto Cleanup;
 407         }
 408
 409
 410         /* Call to the vsc driver to add the device */
 411         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 412         if (ret != 0) {
 413                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 414                 goto Cleanup;
 415         }
 416
 417         blkdev->device_ctx = device_obj;
 418         /* this identified the device 0 or 1 */
 419         blkdev->target = device_info.target_id;
 420         /* this identified the ide ctrl 0 or 1 */
 421         blkdev->path = device_info.path_id;
 422
 423         dev_set_drvdata(device, blkdev);
 424
 425         /* Calculate the major and device num */
 426         if (blkdev->path == 0) {
 427                 major = IDE0_MAJOR;
 428                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 429
 430                 if (!ide0_registered) {
 431                         ret = register_blkdev(major, "ide");
 432                         if (ret != 0) {
 433                                 DPRINT_ERR(BLKVSC_DRV,
 434                                            "register_blkdev() failed! ret %d",
 435                                            ret);
 436                                 goto Remove;
 437                         }
 438
 439                         ide0_registered = 1;
 440                 }
 441         } else if (blkdev->path == 1) {
 442                 major = IDE1_MAJOR;
 443                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 444
 445                 if (!ide1_registered) {
 446                         ret = register_blkdev(major, "ide");
 447                         if (ret != 0) {
 448                                 DPRINT_ERR(BLKVSC_DRV,
 449                                            "register_blkdev() failed! ret %d",
 450                                            ret);
 451                                 goto Remove;
 452                         }
 453
 454                         ide1_registered = 1;
 455                 }
 456         } else {
 457                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 458                 ret = -1;
 459                 goto Cleanup;
 460         }
 461
 462         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 463
 464         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 465         if (!blkdev->gd) {
 466                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 467                 ret = -1;
 468                 goto Cleanup;
 469         }
 470
 471         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 472
 473         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 474         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 475         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 476         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 477         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 478
 479         blkdev->gd->major = major;
 480         if (devnum == 1 || devnum == 3)
 481                 blkdev->gd->first_minor = BLKVSC_MINORS;
 482         else
 483                 blkdev->gd->first_minor = 0;
 484         blkdev->gd->fops = &block_ops;
 485         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 486         blkdev->gd->private_data = blkdev;
 487         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 488         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 489
 490         blkvsc_do_inquiry(blkdev);
 491         if (blkdev->device_type == DVD_TYPE) {
 492                 set_disk_ro(blkdev->gd, 1);
 493                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 494                 blkvsc_do_read_capacity(blkdev);
 495         } else {
 496                 blkvsc_do_read_capacity16(blkdev);
 497         }
 498
 499         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 500         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 501         /* go! */
 502         add_disk(blkdev->gd);
 503
 504         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 505                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 506                     blkdev->sector_size);
 507
 508         return ret;
 509
 510 Remove:
 511         storvsc_drv_obj->base.dev_rm(device_obj);
 512
 513 Cleanup:
 514         if (blkdev) {
 515                 if (blkdev->request_pool) {
 516                         kmem_cache_destroy(blkdev->request_pool);
 517                         blkdev->request_pool = NULL;
 518                 }
 519                 kfree(blkdev);
 520                 blkdev = NULL;
 521         }
 522
 523         return ret;
 524 }
 525
 526 static void blkvsc_shutdown(struct device *device)
 527 {
 528         struct block_device_context *blkdev = dev_get_drvdata(device);
 529         unsigned long flags;
 530
 531         if (!blkdev)
 532                 return;
 533
 534         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 535                    blkdev->users, blkdev->gd->disk_name);
 536
 537         spin_lock_irqsave(&blkdev->lock, flags);
 538
 539         blkdev->shutting_down = 1;
 540
 541         blk_stop_queue(blkdev->gd->queue);
 542
 543         spin_unlock_irqrestore(&blkdev->lock, flags);
 544
 545         while (blkdev->num_outstanding_reqs) {
 546                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 547                             blkdev->num_outstanding_reqs);
 548                 udelay(100);
 549         }
 550
 551         blkvsc_do_flush(blkdev);
 552
 553         spin_lock_irqsave(&blkdev->lock, flags);
 554
 555         blkvsc_cancel_pending_reqs(blkdev);
 556
 557         spin_unlock_irqrestore(&blkdev->lock, flags);
 558 }
 559
 560 static int blkvsc_do_flush(struct block_device_context *blkdev)
 561 {
 562         struct blkvsc_request *blkvsc_req;
 563
 564         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 565
 566         if (blkdev->device_type != HARDDISK_TYPE)
 567                 return 0;
 568
 569         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 570         if (!blkvsc_req)
 571                 return -ENOMEM;
 572
 573         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 574         init_completion(&blkvsc_req->request.wait_event);
 575         blkvsc_req->dev = blkdev;
 576         blkvsc_req->req = NULL;
 577         blkvsc_req->write = 0;
 578
 579         blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 580         blkvsc_req->request.data_buffer.offset = 0;
 581         blkvsc_req->request.data_buffer.len = 0;
 582
 583         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 584         blkvsc_req->cmd_len = 10;
 585
 586         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 587
 588         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 589
 590         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 591
 592         return 0;
 593 }
 594
 595 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 596 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 597 {
 598         struct blkvsc_request *blkvsc_req;
 599         struct page *page_buf;
 600         unsigned char *buf;
 601         unsigned char device_type;
 602
 603         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 604
 605         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 606         if (!blkvsc_req)
 607                 return -ENOMEM;
 608
 609         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 610         page_buf = alloc_page(GFP_KERNEL);
 611         if (!page_buf) {
 612                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 613                 return -ENOMEM;
 614         }
 615
 616         init_completion(&blkvsc_req->request.wait_event);
 617         blkvsc_req->dev = blkdev;
 618         blkvsc_req->req = NULL;
 619         blkvsc_req->write = 0;
 620
 621         blkvsc_req->request.data_buffer.pfn_array[0] =
 622         page_to_pfn(page_buf);
 623         blkvsc_req->request.data_buffer.offset = 0;
 624         blkvsc_req->request.data_buffer.len = 64;
 625
 626         blkvsc_req->cmnd[0] = INQUIRY;
 627         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 628         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 629         blkvsc_req->cmnd[4] = 64;
 630         blkvsc_req->cmd_len = 6;
 631
 632         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 633
 634         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 635                    blkvsc_req);
 636
 637         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 638
 639         buf = kmap(page_buf);
 640
 641         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 642         /* be to le */
 643         device_type = buf[0] & 0x1F;
 644
 645         if (device_type == 0x0) {
 646                 blkdev->device_type = HARDDISK_TYPE;
 647         } else if (device_type == 0x5) {
 648                 blkdev->device_type = DVD_TYPE;
 649         } else {
 650                 /* TODO: this is currently unsupported device type */
 651                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 652         }
 653
 654         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 655
 656         blkdev->device_id_len = buf[7];
 657         if (blkdev->device_id_len > 64)
 658                 blkdev->device_id_len = 64;
 659
 660         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 661         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 662          * blkdev->device_id_len); */
 663
 664         kunmap(page_buf);
 665
 666         __free_page(page_buf);
 667
 668         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 669
 670         return 0;
 671 }
 672
 673 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 674 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 675 {
 676         struct blkvsc_request *blkvsc_req;
 677         struct page *page_buf;
 678         unsigned char *buf;
 679         struct scsi_sense_hdr sense_hdr;
 680         struct vmscsi_request *vm_srb;
 681
 682         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 683
 684         blkdev->sector_size = 0;
 685         blkdev->capacity = 0;
 686         blkdev->media_not_present = 0; /* assume a disk is present */
 687
 688         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 689         if (!blkvsc_req)
 690                 return -ENOMEM;
 691
 692         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 693         page_buf = alloc_page(GFP_KERNEL);
 694         if (!page_buf) {
 695                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 696                 return -ENOMEM;
 697         }
 698
 699         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 700         init_completion(&blkvsc_req->request.wait_event);
 701         blkvsc_req->dev = blkdev;
 702         blkvsc_req->req = NULL;
 703         blkvsc_req->write = 0;
 704
 705         blkvsc_req->request.data_buffer.pfn_array[0] =
 706         page_to_pfn(page_buf);
 707         blkvsc_req->request.data_buffer.offset = 0;
 708         blkvsc_req->request.data_buffer.len = 8;
 709
 710         blkvsc_req->cmnd[0] = READ_CAPACITY;
 711         blkvsc_req->cmd_len = 16;
 712
 713         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 714
 715         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 716                    blkvsc_req);
 717
 718         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 719
 720         /* check error */
 721         if (vm_srb->scsi_status) {
 722                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 723                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 724
 725                 if (sense_hdr.asc == 0x3A) {
 726                         /* Medium not present */
 727                         blkdev->media_not_present = 1;
 728                 }
 729                 return 0;
 730         }
 731         buf = kmap(page_buf);
 732
 733         /* be to le */
 734         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 735                             (buf[2] << 8) | buf[3]) + 1;
 736         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 737                               (buf[6] << 8) | buf[7];
 738
 739         kunmap(page_buf);
 740
 741         __free_page(page_buf);
 742
 743         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 744
 745         return 0;
 746 }
 747
 748 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 749 {
 750         struct blkvsc_request *blkvsc_req;
 751         struct page *page_buf;
 752         unsigned char *buf;
 753         struct scsi_sense_hdr sense_hdr;
 754         struct vmscsi_request *vm_srb;
 755
 756         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 757
 758         blkdev->sector_size = 0;
 759         blkdev->capacity = 0;
 760         blkdev->media_not_present = 0; /* assume a disk is present */
 761
 762         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 763         if (!blkvsc_req)
 764                 return -ENOMEM;
 765
 766         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 767         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 768         page_buf = alloc_page(GFP_KERNEL);
 769         if (!page_buf) {
 770                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 771                 return -ENOMEM;
 772         }
 773
 774         init_completion(&blkvsc_req->request.wait_event);
 775         blkvsc_req->dev = blkdev;
 776         blkvsc_req->req = NULL;
 777         blkvsc_req->write = 0;
 778
 779         blkvsc_req->request.data_buffer.pfn_array[0] =
 780         page_to_pfn(page_buf);
 781         blkvsc_req->request.data_buffer.offset = 0;
 782         blkvsc_req->request.data_buffer.len = 12;
 783
 784         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 785         blkvsc_req->cmd_len = 16;
 786
 787         /*
 788          * Set this here since the completion routine may be invoked
 789          * and completed before we return
 790          */
 791
 792         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 793
 794         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 795                    blkvsc_req);
 796
 797         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 798
 799         /* check error */
 800         if (vm_srb->scsi_status) {
 801                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 802                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 803                 if (sense_hdr.asc == 0x3A) {
 804                         /* Medium not present */
 805                         blkdev->media_not_present = 1;
 806                 }
 807                 return 0;
 808         }
 809         buf = kmap(page_buf);
 810
 811         /* be to le */
 812         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 813         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 814
 815 #if 0
 816         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 817                             (buf[2] << 8) | buf[3]) + 1;
 818         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 819                               (buf[6] << 8) | buf[7];
 820 #endif
 821
 822         kunmap(page_buf);
 823
 824         __free_page(page_buf);
 825
 826         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 827
 828         return 0;
 829 }
 830
 831 /*
 832  * blkvsc_remove() - Callback when our device is removed
 833  */
 834 static int blkvsc_remove(struct device *device)
 835 {
 836         struct hv_driver *drv =
 837                                 drv_to_hv_drv(device->driver);
 838         struct storvsc_driver_object *storvsc_drv_obj =
 839                                 drv->priv;
 840         struct hv_device *device_obj = device_to_hv_device(device);
 841         struct block_device_context *blkdev = dev_get_drvdata(device);
 842         unsigned long flags;
 843         int ret;
 844
 845         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 846
 847         if (!storvsc_drv_obj->base.dev_rm)
 848                 return -1;
 849
 850         /*
 851          * Call to the vsc driver to let it know that the device is being
 852          * removed
 853          */
 854         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 855         if (ret != 0) {
 856                 /* TODO: */
 857                 DPRINT_ERR(BLKVSC_DRV,
 858                            "unable to remove blkvsc device (ret %d)", ret);
 859         }
 860
 861         /* Get to a known state */
 862         spin_lock_irqsave(&blkdev->lock, flags);
 863
 864         blkdev->shutting_down = 1;
 865
 866         blk_stop_queue(blkdev->gd->queue);
 867
 868         spin_unlock_irqrestore(&blkdev->lock, flags);
 869
 870         while (blkdev->num_outstanding_reqs) {
 871                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 872                             blkdev->num_outstanding_reqs);
 873                 udelay(100);
 874         }
 875
 876         blkvsc_do_flush(blkdev);
 877
 878         spin_lock_irqsave(&blkdev->lock, flags);
 879
 880         blkvsc_cancel_pending_reqs(blkdev);
 881
 882         spin_unlock_irqrestore(&blkdev->lock, flags);
 883
 884         blk_cleanup_queue(blkdev->gd->queue);
 885
 886         del_gendisk(blkdev->gd);
 887
 888         kmem_cache_destroy(blkdev->request_pool);
 889
 890         kfree(blkdev);
 891
 892         return ret;
 893 }
 894
 895 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 896 {
 897         /* ASSERT(blkvsc_req->req); */
 898         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 899
 900         blkvsc_req->cmd_len = 16;
 901
 902         if (blkvsc_req->sector_start > 0xffffffff) {
 903                 if (rq_data_dir(blkvsc_req->req)) {
 904                         blkvsc_req->write = 1;
 905                         blkvsc_req->cmnd[0] = WRITE_16;
 906                 } else {
 907                         blkvsc_req->write = 0;
 908                         blkvsc_req->cmnd[0] = READ_16;
 909                 }
 910
 911                 blkvsc_req->cmnd[1] |=
 912                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 913
 914                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 915                                 cpu_to_be64(blkvsc_req->sector_start);
 916                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 917                                 cpu_to_be32(blkvsc_req->sector_count);
 918         } else if ((blkvsc_req->sector_count > 0xff) ||
 919                    (blkvsc_req->sector_start > 0x1fffff)) {
 920                 if (rq_data_dir(blkvsc_req->req)) {
 921                         blkvsc_req->write = 1;
 922                         blkvsc_req->cmnd[0] = WRITE_10;
 923                 } else {
 924                         blkvsc_req->write = 0;
 925                         blkvsc_req->cmnd[0] = READ_10;
 926                 }
 927
 928                 blkvsc_req->cmnd[1] |=
 929                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 930
 931                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 932                                 cpu_to_be32(blkvsc_req->sector_start);
 933                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 934                                 cpu_to_be16(blkvsc_req->sector_count);
 935         } else {
 936                 if (rq_data_dir(blkvsc_req->req)) {
 937                         blkvsc_req->write = 1;
 938                         blkvsc_req->cmnd[0] = WRITE_6;
 939                 } else {
 940                         blkvsc_req->write = 0;
 941                         blkvsc_req->cmnd[0] = READ_6;
 942                 }
 943
 944                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 945                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 946                 blkvsc_req->cmnd[1] &= 0x1f;
 947                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 948         }
 949 }
 950
 951
 952 /*
 953  * We break the request into 1 or more blkvsc_requests and submit
 954  * them.  If we can't submit them all, we put them on the
 955  * pending_list. The blkvsc_request() will work on the pending_list.
 956  */
 957 static int blkvsc_do_request(struct block_device_context *blkdev,
 958                              struct request *req)
 959 {
 960         struct bio *bio = NULL;
 961         struct bio_vec *bvec = NULL;
 962         struct bio_vec *prev_bvec = NULL;
 963         struct blkvsc_request *blkvsc_req = NULL;
 964         struct blkvsc_request *tmp;
 965         int databuf_idx = 0;
 966         int seg_idx = 0;
 967         sector_t start_sector;
 968         unsigned long num_sectors = 0;
 969         int ret = 0;
 970         int pending = 0;
 971         struct blkvsc_request_group *group = NULL;
 972
 973         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 974                   (unsigned long)blk_rq_pos(req));
 975
 976         /* Create a group to tie req to list of blkvsc_reqs */
 977         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 978         if (!group)
 979                 return -ENOMEM;
 980
 981         INIT_LIST_HEAD(&group->blkvsc_req_list);
 982         group->outstanding = group->status = 0;
 983
 984         start_sector = blk_rq_pos(req);
 985
 986         /* foreach bio in the request */
 987         if (req->bio) {
 988                 for (bio = req->bio; bio; bio = bio->bi_next) {
 989                         /*
 990                          * Map this bio into an existing or new storvsc request
 991                          */
 992                         bio_for_each_segment(bvec, bio, seg_idx) {
 993                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 994                                            "- req %p bio %p bvec %p seg_idx %d "
 995                                            "databuf_idx %d\n", req, bio, bvec,
 996                                            seg_idx, databuf_idx);
 997
 998                                 /* Get a new storvsc request */
 999                                 /* 1st-time */
1000                                 if ((!blkvsc_req) ||
1001                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
1002                                     /* hole at the begin of page */
1003                                     || (bvec->bv_offset != 0) ||
1004                                     /* hold at the end of page */
1005                                     (prev_bvec &&
1006                                      (prev_bvec->bv_len != PAGE_SIZE))) {
1007                                         /* submit the prev one */
1008                                         if (blkvsc_req) {
1009                                                 blkvsc_req->sector_start = start_sector;
1010                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
1011
1012                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
1013                                                 blkvsc_init_rw(blkvsc_req);
1014                                         }
1015
1016                                         /*
1017                                          * Create new blkvsc_req to represent
1018                                          * the current bvec
1019                                          */
1020                                         blkvsc_req =
1021                                         kmem_cache_zalloc(
1022                                         blkdev->request_pool, GFP_ATOMIC);
1023                                         if (!blkvsc_req) {
1024                                                 /* free up everything */
1025                                                 list_for_each_entry_safe(
1026                                                         blkvsc_req, tmp,
1027                                                         &group->blkvsc_req_list,
1028                                                         req_entry) {
1029                                                         list_del(&blkvsc_req->req_entry);
1030                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
1031                                                 }
1032
1033                                                 kmem_cache_free(blkdev->request_pool, group);
1034                                                 return -ENOMEM;
1035                                         }
1036
1037                                         memset(blkvsc_req, 0,
1038                                                sizeof(struct blkvsc_request));
1039
1040                                         blkvsc_req->dev = blkdev;
1041                                         blkvsc_req->req = req;
1042                                         blkvsc_req->request.
1043                                         data_buffer.offset
1044                                         = bvec->bv_offset;
1045                                         blkvsc_req->request.
1046                                         data_buffer.len = 0;
1047
1048                                         /* Add to the group */
1049                                         blkvsc_req->group = group;
1050                                         blkvsc_req->group->outstanding++;
1051                                         list_add_tail(&blkvsc_req->req_entry,
1052                                                 &blkvsc_req->group->blkvsc_req_list);
1053
1054                                         start_sector += num_sectors;
1055                                         num_sectors = 0;
1056                                         databuf_idx = 0;
1057                                 }
1058
1059                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1060                                 blkvsc_req->request.data_buffer.
1061                                         pfn_array[databuf_idx]
1062                                                 = page_to_pfn(bvec->bv_page);
1063                                 blkvsc_req->request.data_buffer.len
1064                                         += bvec->bv_len;
1065
1066                                 prev_bvec = bvec;
1067
1068                                 databuf_idx++;
1069                                 num_sectors += bvec->bv_len >> 9;
1070
1071                         } /* bio_for_each_segment */
1072
1073                 } /* rq_for_each_bio */
1074         }
1075
1076         /* Handle the last one */
1077         if (blkvsc_req) {
1078                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1079                            blkdev, req, blkvsc_req->group,
1080                            blkvsc_req->group->outstanding);
1081
1082                 blkvsc_req->sector_start = start_sector;
1083                 sector_div(blkvsc_req->sector_start,
1084                            (blkdev->sector_size >> 9));
1085
1086                 blkvsc_req->sector_count = num_sectors /
1087                                            (blkdev->sector_size >> 9);
1088
1089                 blkvsc_init_rw(blkvsc_req);
1090         }
1091
1092         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1093                 if (pending) {
1094                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1095                                    "pending_list - blkvsc_req %p start_sect %lu"
1096                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1097                                    (unsigned long)blkvsc_req->sector_start,
1098                                    blkvsc_req->sector_count,
1099                                    (unsigned long)start_sector,
1100                                    (unsigned long)num_sectors);
1101
1102                         list_add_tail(&blkvsc_req->pend_entry,
1103                                       &blkdev->pending_list);
1104                 } else {
1105                         ret = blkvsc_submit_request(blkvsc_req,
1106                                                     blkvsc_request_completion);
1107                         if (ret == -1) {
1108                                 pending = 1;
1109                                 list_add_tail(&blkvsc_req->pend_entry,
1110                                               &blkdev->pending_list);
1111                         }
1112
1113                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1114                                    "start_sect %lu sect_count %ld (%lu %ld) "
1115                                    "ret %d\n", blkvsc_req,
1116                                    (unsigned long)blkvsc_req->sector_start,
1117                                    blkvsc_req->sector_count,
1118                                    (unsigned long)start_sector,
1119                                    num_sectors, ret);
1120                 }
1121         }
1122
1123         return pending;
1124 }
1125
1126 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1127 {
1128         struct blkvsc_request *blkvsc_req =
1129                         (struct blkvsc_request *)request->context;
1130         struct block_device_context *blkdev =
1131                         (struct block_device_context *)blkvsc_req->dev;
1132         struct scsi_sense_hdr sense_hdr;
1133         struct vmscsi_request *vm_srb;
1134
1135         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1136                    blkvsc_req);
1137
1138         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
1139         blkdev->num_outstanding_reqs--;
1140
1141         if (vm_srb->scsi_status)
1142                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1143                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1144                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1145
1146         complete(&blkvsc_req->request.wait_event);
1147 }
1148
1149 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1150 {
1151         struct blkvsc_request *blkvsc_req =
1152                         (struct blkvsc_request *)request->context;
1153         struct block_device_context *blkdev =
1154                         (struct block_device_context *)blkvsc_req->dev;
1155         unsigned long flags;
1156         struct blkvsc_request *comp_req, *tmp;
1157         struct vmscsi_request *vm_srb;
1158
1159         /* ASSERT(blkvsc_req->group); */
1160
1161         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1162                    "sect_start %lu sect_count %ld len %d group outstd %d "
1163                    "total outstd %d\n",
1164                    blkdev, blkvsc_req, blkvsc_req->group,
1165                    (blkvsc_req->write) ? "WRITE" : "READ",
1166                    (unsigned long)blkvsc_req->sector_start,
1167                    blkvsc_req->sector_count,
1168                    blkvsc_req->request.data_buffer.len,
1169                    blkvsc_req->group->outstanding,
1170                    blkdev->num_outstanding_reqs);
1171
1172         spin_lock_irqsave(&blkdev->lock, flags);
1173
1174         blkdev->num_outstanding_reqs--;
1175         blkvsc_req->group->outstanding--;
1176
1177         /*
1178          * Only start processing when all the blkvsc_reqs are
1179          * completed. This guarantees no out-of-order blkvsc_req
1180          * completion when calling end_that_request_first()
1181          */
1182         if (blkvsc_req->group->outstanding == 0) {
1183                 list_for_each_entry_safe(comp_req, tmp,
1184                                          &blkvsc_req->group->blkvsc_req_list,
1185                                          req_entry) {
1186                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1187                                    "sect_start %lu sect_count %ld\n",
1188                                    comp_req,
1189                                    (unsigned long)comp_req->sector_start,
1190                                    comp_req->sector_count);
1191
1192                         list_del(&comp_req->req_entry);
1193
1194                         vm_srb =
1195                         &comp_req->request.vstor_packet.vm_srb;
1196                         if (!__blk_end_request(comp_req->req,
1197                                 (!vm_srb->scsi_status ? 0 : -EIO),
1198                                 comp_req->sector_count * blkdev->sector_size)) {
1199                                 /*
1200                                  * All the sectors have been xferred ie the
1201                                  * request is done
1202                                  */
1203                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1204                                            comp_req->req);
1205                                 kmem_cache_free(blkdev->request_pool,
1206                                                 comp_req->group);
1207                         }
1208
1209                         kmem_cache_free(blkdev->request_pool, comp_req);
1210                 }
1211
1212                 if (!blkdev->shutting_down) {
1213                         blkvsc_do_pending_reqs(blkdev);
1214                         blk_start_queue(blkdev->gd->queue);
1215                         blkvsc_request(blkdev->gd->queue);
1216                 }
1217         }
1218
1219         spin_unlock_irqrestore(&blkdev->lock, flags);
1220 }
1221
1222 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1223 {
1224         struct blkvsc_request *pend_req, *tmp;
1225         struct blkvsc_request *comp_req, *tmp2;
1226         struct vmscsi_request *vm_srb;
1227
1228         int ret = 0;
1229
1230         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1231
1232         /* Flush the pending list first */
1233         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1234                                  pend_entry) {
1235                 /*
1236                  * The pend_req could be part of a partially completed
1237                  * request. If so, complete those req first until we
1238                  * hit the pend_req
1239                  */
1240                 list_for_each_entry_safe(comp_req, tmp2,
1241                                          &pend_req->group->blkvsc_req_list,
1242                                          req_entry) {
1243                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1244                                    "sect_start %lu sect_count %ld\n",
1245                                    comp_req,
1246                                    (unsigned long) comp_req->sector_start,
1247                                    comp_req->sector_count);
1248
1249                         if (comp_req == pend_req)
1250                                 break;
1251
1252                         list_del(&comp_req->req_entry);
1253
1254                         if (comp_req->req) {
1255                                 vm_srb =
1256                                 &comp_req->request.vstor_packet.
1257                                 vm_srb;
1258                                 ret = __blk_end_request(comp_req->req,
1259                                         (!vm_srb->scsi_status ? 0 : -EIO),
1260                                         comp_req->sector_count *
1261                                         blkdev->sector_size);
1262
1263                                 /* FIXME: shouldn't this do more than return? */
1264                                 if (ret)
1265                                         goto out;
1266                         }
1267
1268                         kmem_cache_free(blkdev->request_pool, comp_req);
1269                 }
1270
1271                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1272                            pend_req);
1273
1274                 list_del(&pend_req->pend_entry);
1275
1276                 list_del(&pend_req->req_entry);
1277
1278                 if (comp_req->req) {
1279                         if (!__blk_end_request(pend_req->req, -EIO,
1280                                                pend_req->sector_count *
1281                                                blkdev->sector_size)) {
1282                                 /*
1283                                  * All the sectors have been xferred ie the
1284                                  * request is done
1285                                  */
1286                                 DPRINT_DBG(BLKVSC_DRV,
1287                                            "blkvsc_cancel_pending_reqs() - "
1288                                            "req %p COMPLETED\n", pend_req->req);
1289                                 kmem_cache_free(blkdev->request_pool,
1290                                                 pend_req->group);
1291                         }
1292                 }
1293
1294                 kmem_cache_free(blkdev->request_pool, pend_req);
1295         }
1296
1297 out:
1298         return ret;
1299 }
1300
1301 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1302 {
1303         struct blkvsc_request *pend_req, *tmp;
1304         int ret = 0;
1305
1306         /* Flush the pending list first */
1307         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1308                                  pend_entry) {
1309                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1310                            pend_req);
1311
1312                 ret = blkvsc_submit_request(pend_req,
1313                                             blkvsc_request_completion);
1314                 if (ret != 0)
1315                         break;
1316                 else
1317                         list_del(&pend_req->pend_entry);
1318         }
1319
1320         return ret;
1321 }
1322
1323 static void blkvsc_request(struct request_queue *queue)
1324 {
1325         struct block_device_context *blkdev = NULL;
1326         struct request *req;
1327         int ret = 0;
1328
1329         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1330         while ((req = blk_peek_request(queue)) != NULL) {
1331                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1332
1333                 blkdev = req->rq_disk->private_data;
1334                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1335                     blkdev->media_not_present) {
1336                         __blk_end_request_cur(req, 0);
1337                         continue;
1338                 }
1339
1340                 ret = blkvsc_do_pending_reqs(blkdev);
1341
1342                 if (ret != 0) {
1343                         DPRINT_DBG(BLKVSC_DRV,
1344                                    "- stop queue - pending_list not empty\n");
1345                         blk_stop_queue(queue);
1346                         break;
1347                 }
1348
1349                 blk_start_request(req);
1350
1351                 ret = blkvsc_do_request(blkdev, req);
1352                 if (ret > 0) {
1353                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1354                         blk_stop_queue(queue);
1355                         break;
1356                 } else if (ret < 0) {
1357                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1358                         blk_requeue_request(queue, req);
1359                         blk_stop_queue(queue);
1360                         break;
1361                 }
1362         }
1363 }
1364
1365 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1366 {
1367         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1368
1369         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1370                    blkdev->gd->disk_name);
1371
1372         mutex_lock(&blkvsc_mutex);
1373         spin_lock(&blkdev->lock);
1374
1375         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1376                 spin_unlock(&blkdev->lock);
1377                 check_disk_change(bdev);
1378                 spin_lock(&blkdev->lock);
1379         }
1380
1381         blkdev->users++;
1382
1383         spin_unlock(&blkdev->lock);
1384         mutex_unlock(&blkvsc_mutex);
1385         return 0;
1386 }
1387
1388 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1389 {
1390         struct block_device_context *blkdev = disk->private_data;
1391
1392         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1393                    blkdev->gd->disk_name);
1394
1395         mutex_lock(&blkvsc_mutex);
1396         spin_lock(&blkdev->lock);
1397         if (blkdev->users == 1) {
1398                 spin_unlock(&blkdev->lock);
1399                 blkvsc_do_flush(blkdev);
1400                 spin_lock(&blkdev->lock);
1401         }
1402
1403         blkdev->users--;
1404
1405         spin_unlock(&blkdev->lock);
1406         mutex_unlock(&blkvsc_mutex);
1407         return 0;
1408 }
1409
1410 static unsigned int blkvsc_check_events(struct gendisk *gd,
1411                                         unsigned int clearing)
1412 {
1413         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1414         return DISK_EVENT_MEDIA_CHANGE;
1415 }
1416
1417 static int blkvsc_revalidate_disk(struct gendisk *gd)
1418 {
1419         struct block_device_context *blkdev = gd->private_data;
1420
1421         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1422
1423         if (blkdev->device_type == DVD_TYPE) {
1424                 blkvsc_do_read_capacity(blkdev);
1425                 set_capacity(blkdev->gd, blkdev->capacity *
1426                             (blkdev->sector_size/512));
1427                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1428         }
1429         return 0;
1430 }
1431
1432 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1433 {
1434         sector_t total_sectors = get_capacity(bd->bd_disk);
1435         sector_t cylinder_times_heads = 0;
1436         sector_t temp = 0;
1437
1438         int sectors_per_track = 0;
1439         int heads = 0;
1440         int cylinders = 0;
1441         int rem = 0;
1442
1443         if (total_sectors > (65535 * 16 * 255))
1444                 total_sectors = (65535 * 16 * 255);
1445
1446         if (total_sectors >= (65535 * 16 * 63)) {
1447                 sectors_per_track = 255;
1448                 heads = 16;
1449
1450                 cylinder_times_heads = total_sectors;
1451                 /* sector_div stores the quotient in cylinder_times_heads */
1452                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1453         } else {
1454                 sectors_per_track = 17;
1455
1456                 cylinder_times_heads = total_sectors;
1457                 /* sector_div stores the quotient in cylinder_times_heads */
1458                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1459
1460                 temp = cylinder_times_heads + 1023;
1461                 /* sector_div stores the quotient in temp */
1462                 rem = sector_div(temp, 1024);
1463
1464                 heads = temp;
1465
1466                 if (heads < 4)
1467                         heads = 4;
1468
1469
1470                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1471                         sectors_per_track = 31;
1472                         heads = 16;
1473
1474                         cylinder_times_heads = total_sectors;
1475                         /*
1476                          * sector_div stores the quotient in
1477                          * cylinder_times_heads
1478                          */
1479                         rem = sector_div(cylinder_times_heads,
1480                                          sectors_per_track);
1481                 }
1482
1483                 if (cylinder_times_heads >= (heads * 1024)) {
1484                         sectors_per_track = 63;
1485                         heads = 16;
1486
1487                         cylinder_times_heads = total_sectors;
1488                         /*
1489                          * sector_div stores the quotient in
1490                          * cylinder_times_heads
1491                          */
1492                         rem = sector_div(cylinder_times_heads,
1493                                          sectors_per_track);
1494                 }
1495         }
1496
1497         temp = cylinder_times_heads;
1498         /* sector_div stores the quotient in temp */
1499         rem = sector_div(temp, heads);
1500         cylinders = temp;
1501
1502         hg->heads = heads;
1503         hg->sectors = sectors_per_track;
1504         hg->cylinders = cylinders;
1505
1506         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1507                     sectors_per_track);
1508
1509     return 0;
1510 }
1511
1512 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1513                         unsigned cmd, unsigned long argument)
1514 {
1515 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1516         int ret;
1517
1518         switch (cmd) {
1519         /*
1520          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1521          * than just a GUID. Commented it out for now.
1522          */
1523 #if 0
1524         case HDIO_GET_IDENTITY:
1525                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1526                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1527                                  blkdev->device_id_len))
1528                         ret = -EFAULT;
1529                 break;
1530 #endif
1531         default:
1532                 ret = -EINVAL;
1533                 break;
1534         }
1535
1536         return ret;
1537 }
1538
1539 static int __init blkvsc_init(void)
1540 {
1541         int ret;
1542
1543         BUILD_BUG_ON(sizeof(sector_t) != 8);
1544
1545         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1546
1547         ret = blkvsc_drv_init();
1548
1549         return ret;
1550 }
1551
1552 static void __exit blkvsc_exit(void)
1553 {
1554         blkvsc_drv_exit();
1555 }
1556
1557 MODULE_LICENSE("GPL");
1558 MODULE_VERSION(HV_DRV_VERSION);
1559 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1560 module_init(blkvsc_init);
1561 module_exit(blkvsc_exit);