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         wait_queue_head_t wevent;
  76         int cond;
  77
  78         int write;
  79         sector_t sector_start;
  80         unsigned long sector_count;
  81
  82         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  83         unsigned char cmd_len;
  84         unsigned char cmnd[MAX_COMMAND_SIZE];
  85
  86         struct hv_storvsc_request request;
  87 };
  88
  89 /* Per device structure */
  90 struct block_device_context {
  91         /* point back to our device context */
  92         struct hv_device *device_ctx;
  93         struct kmem_cache *request_pool;
  94         spinlock_t lock;
  95         struct gendisk *gd;
  96         enum blkvsc_device_type device_type;
  97         struct list_head pending_list;
  98
  99         unsigned char device_id[64];
 100         unsigned int device_id_len;
 101         int num_outstanding_reqs;
 102         int shutting_down;
 103         int media_not_present;
 104         unsigned int sector_size;
 105         sector_t capacity;
 106         unsigned int port;
 107         unsigned char path;
 108         unsigned char target;
 109         int users;
 110 };
 111
 112
 113 static const char *g_blk_driver_name = "blkvsc";
 114
 115 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 116 static const struct hv_guid g_blk_device_type = {
 117         .data = {
 118                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 119                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 120         }
 121 };
 122
 123 static int blk_vsc_on_device_add(struct hv_device *device,
 124                                 void *additional_info)
 125 {
 126         struct storvsc_device_info *device_info;
 127         int ret = 0;
 128
 129         device_info = (struct storvsc_device_info *)additional_info;
 130
 131         ret = stor_vsc_on_device_add(device, additional_info);
 132         if (ret != 0)
 133                 return ret;
 134
 135         /*
 136          * We need to use the device instance guid to set the path and target
 137          * id. For IDE devices, the device instance id is formatted as
 138          * <bus id> * - <device id> - 8899 - 000000000000.
 139          */
 140         device_info->path_id = device->dev_instance.data[3] << 24 |
 141                              device->dev_instance.data[2] << 16 |
 142                              device->dev_instance.data[1] << 8  |
 143                              device->dev_instance.data[0];
 144
 145         device_info->target_id = device->dev_instance.data[5] << 8 |
 146                                device->dev_instance.data[4];
 147
 148         return ret;
 149 }
 150
 151
 152 static int blk_vsc_initialize(struct hv_driver *driver)
 153 {
 154         struct storvsc_driver_object *stor_driver;
 155         int ret = 0;
 156
 157         stor_driver = hvdr_to_stordr(driver);
 158
 159         /* Make sure we are at least 2 pages since 1 page is used for control */
 160         /* ASSERT(stor_driver->RingBufferSize >= (PAGE_SIZE << 1)); */
 161
 162         driver->name = g_blk_driver_name;
 163         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 164
 165
 166         /*
 167          * Divide the ring buffer data size (which is 1 page less than the ring
 168          * buffer size since that page is reserved for the ring buffer indices)
 169          * by the max request size (which is
 170          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 171          */
 172         stor_driver->max_outstanding_req_per_channel =
 173                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 174                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 175                            sizeof(struct vstor_packet) + sizeof(u64),
 176                            sizeof(u64)));
 177
 178         DPRINT_INFO(BLKVSC, "max io outstd %u",
 179                     stor_driver->max_outstanding_req_per_channel);
 180
 181         /* Setup the dispatch table */
 182         stor_driver->base.dev_add = blk_vsc_on_device_add;
 183         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 184         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 185         stor_driver->on_io_request = stor_vsc_on_io_request;
 186
 187         return ret;
 188 }
 189
 190 /* Static decl */
 191 static DEFINE_MUTEX(blkvsc_mutex);
 192 static int blkvsc_probe(struct device *dev);
 193 static int blkvsc_remove(struct device *device);
 194 static void blkvsc_shutdown(struct device *device);
 195
 196 static int blkvsc_open(struct block_device *bdev,  fmode_t mode);
 197 static int blkvsc_release(struct gendisk *disk, fmode_t mode);
 198 static unsigned int blkvsc_check_events(struct gendisk *gd,
 199                                         unsigned int clearing);
 200 static int blkvsc_revalidate_disk(struct gendisk *gd);
 201 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg);
 202 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 203                         unsigned cmd, unsigned long argument);
 204 static void blkvsc_request(struct request_queue *queue);
 205 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 206 static int blkvsc_do_request(struct block_device_context *blkdev,
 207                              struct request *req);
 208 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 209                 void (*request_completion)(struct hv_storvsc_request *));
 210 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req);
 211 static void blkvsc_cmd_completion(struct hv_storvsc_request *request);
 212 static int blkvsc_do_inquiry(struct block_device_context *blkdev);
 213 static int blkvsc_do_read_capacity(struct block_device_context *blkdev);
 214 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev);
 215 static int blkvsc_do_flush(struct block_device_context *blkdev);
 216 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev);
 217 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev);
 218
 219 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 220 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 221 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 222
 223 /* The one and only one */
 224 static  struct storvsc_driver_object g_blkvsc_drv;
 225
 226 static const struct block_device_operations block_ops = {
 227         .owner = THIS_MODULE,
 228         .open = blkvsc_open,
 229         .release = blkvsc_release,
 230         .check_events = blkvsc_check_events,
 231         .revalidate_disk = blkvsc_revalidate_disk,
 232         .getgeo = blkvsc_getgeo,
 233         .ioctl  = blkvsc_ioctl,
 234 };
 235
 236 /*
 237  * blkvsc_drv_init -  BlkVsc driver initialization.
 238  */
 239 static int blkvsc_drv_init(void)
 240 {
 241         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 242         struct hv_driver *drv = &g_blkvsc_drv.base;
 243         int ret;
 244
 245         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 246
 247         drv->priv = storvsc_drv_obj;
 248
 249         /* Callback to client driver to complete the initialization */
 250         blk_vsc_initialize(&storvsc_drv_obj->base);
 251
 252         drv->driver.name = storvsc_drv_obj->base.name;
 253
 254         drv->driver.probe = blkvsc_probe;
 255         drv->driver.remove = blkvsc_remove;
 256         drv->driver.shutdown = blkvsc_shutdown;
 257
 258         /* The driver belongs to vmbus */
 259         ret = vmbus_child_driver_register(&drv->driver);
 260
 261         return ret;
 262 }
 263
 264 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 265 {
 266         struct device **curr = (struct device **)data;
 267         *curr = dev;
 268         return 1; /* stop iterating */
 269 }
 270
 271 static void blkvsc_drv_exit(void)
 272 {
 273         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 274         struct hv_driver *drv = &g_blkvsc_drv.base;
 275         struct device *current_dev;
 276         int ret;
 277
 278         while (1) {
 279                 current_dev = NULL;
 280
 281                 /* Get the device */
 282                 ret = driver_for_each_device(&drv->driver, NULL,
 283                                              (void *) &current_dev,
 284                                              blkvsc_drv_exit_cb);
 285
 286                 if (ret)
 287                         DPRINT_WARN(BLKVSC_DRV,
 288                                     "driver_for_each_device returned %d", ret);
 289
 290
 291                 if (current_dev == NULL)
 292                         break;
 293
 294                 /* Initiate removal from the top-down */
 295                 device_unregister(current_dev);
 296         }
 297
 298         if (storvsc_drv_obj->base.cleanup)
 299                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 300
 301         vmbus_child_driver_unregister(&drv->driver);
 302
 303         return;
 304 }
 305
 306 /*
 307  * blkvsc_probe - Add a new device for this driver
 308  */
 309 static int blkvsc_probe(struct device *device)
 310 {
 311         struct hv_driver *drv =
 312                                 drv_to_hv_drv(device->driver);
 313         struct storvsc_driver_object *storvsc_drv_obj =
 314                                 drv->priv;
 315         struct hv_device *device_obj = device_to_hv_device(device);
 316
 317         struct block_device_context *blkdev = NULL;
 318         struct storvsc_device_info device_info;
 319         int major = 0;
 320         int devnum = 0;
 321         int ret = 0;
 322         static int ide0_registered;
 323         static int ide1_registered;
 324
 325         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
 326
 327         if (!storvsc_drv_obj->base.dev_add) {
 328                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 329                 ret = -1;
 330                 goto Cleanup;
 331         }
 332
 333         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 334         if (!blkdev) {
 335                 ret = -ENOMEM;
 336                 goto Cleanup;
 337         }
 338
 339         INIT_LIST_HEAD(&blkdev->pending_list);
 340
 341         /* Initialize what we can here */
 342         spin_lock_init(&blkdev->lock);
 343
 344         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
 345         /*      sizeof(struct blkvsc_request)); */
 346
 347         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 348                                         sizeof(struct blkvsc_request), 0,
 349                                         SLAB_HWCACHE_ALIGN, NULL);
 350         if (!blkdev->request_pool) {
 351                 ret = -ENOMEM;
 352                 goto Cleanup;
 353         }
 354
 355
 356         /* Call to the vsc driver to add the device */
 357         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 358         if (ret != 0) {
 359                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
 360                 goto Cleanup;
 361         }
 362
 363         blkdev->device_ctx = device_obj;
 364         /* this identified the device 0 or 1 */
 365         blkdev->target = device_info.target_id;
 366         /* this identified the ide ctrl 0 or 1 */
 367         blkdev->path = device_info.path_id;
 368
 369         dev_set_drvdata(device, blkdev);
 370
 371         /* Calculate the major and device num */
 372         if (blkdev->path == 0) {
 373                 major = IDE0_MAJOR;
 374                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
 375
 376                 if (!ide0_registered) {
 377                         ret = register_blkdev(major, "ide");
 378                         if (ret != 0) {
 379                                 DPRINT_ERR(BLKVSC_DRV,
 380                                            "register_blkdev() failed! ret %d",
 381                                            ret);
 382                                 goto Remove;
 383                         }
 384
 385                         ide0_registered = 1;
 386                 }
 387         } else if (blkdev->path == 1) {
 388                 major = IDE1_MAJOR;
 389                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
 390
 391                 if (!ide1_registered) {
 392                         ret = register_blkdev(major, "ide");
 393                         if (ret != 0) {
 394                                 DPRINT_ERR(BLKVSC_DRV,
 395                                            "register_blkdev() failed! ret %d",
 396                                            ret);
 397                                 goto Remove;
 398                         }
 399
 400                         ide1_registered = 1;
 401                 }
 402         } else {
 403                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
 404                 ret = -1;
 405                 goto Cleanup;
 406         }
 407
 408         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
 409
 410         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 411         if (!blkdev->gd) {
 412                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
 413                 ret = -1;
 414                 goto Cleanup;
 415         }
 416
 417         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 418
 419         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 420         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 421         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 422         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 423         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 424
 425         blkdev->gd->major = major;
 426         if (devnum == 1 || devnum == 3)
 427                 blkdev->gd->first_minor = BLKVSC_MINORS;
 428         else
 429                 blkdev->gd->first_minor = 0;
 430         blkdev->gd->fops = &block_ops;
 431         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 432         blkdev->gd->private_data = blkdev;
 433         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 434         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
 435
 436         blkvsc_do_inquiry(blkdev);
 437         if (blkdev->device_type == DVD_TYPE) {
 438                 set_disk_ro(blkdev->gd, 1);
 439                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
 440                 blkvsc_do_read_capacity(blkdev);
 441         } else {
 442                 blkvsc_do_read_capacity16(blkdev);
 443         }
 444
 445         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 446         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 447         /* go! */
 448         add_disk(blkdev->gd);
 449
 450         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 451                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 452                     blkdev->sector_size);
 453
 454         return ret;
 455
 456 Remove:
 457         storvsc_drv_obj->base.dev_rm(device_obj);
 458
 459 Cleanup:
 460         if (blkdev) {
 461                 if (blkdev->request_pool) {
 462                         kmem_cache_destroy(blkdev->request_pool);
 463                         blkdev->request_pool = NULL;
 464                 }
 465                 kfree(blkdev);
 466                 blkdev = NULL;
 467         }
 468
 469         return ret;
 470 }
 471
 472 static void blkvsc_shutdown(struct device *device)
 473 {
 474         struct block_device_context *blkdev = dev_get_drvdata(device);
 475         unsigned long flags;
 476
 477         if (!blkdev)
 478                 return;
 479
 480         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 481                    blkdev->users, blkdev->gd->disk_name);
 482
 483         spin_lock_irqsave(&blkdev->lock, flags);
 484
 485         blkdev->shutting_down = 1;
 486
 487         blk_stop_queue(blkdev->gd->queue);
 488
 489         spin_unlock_irqrestore(&blkdev->lock, flags);
 490
 491         while (blkdev->num_outstanding_reqs) {
 492                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 493                             blkdev->num_outstanding_reqs);
 494                 udelay(100);
 495         }
 496
 497         blkvsc_do_flush(blkdev);
 498
 499         spin_lock_irqsave(&blkdev->lock, flags);
 500
 501         blkvsc_cancel_pending_reqs(blkdev);
 502
 503         spin_unlock_irqrestore(&blkdev->lock, flags);
 504 }
 505
 506 static int blkvsc_do_flush(struct block_device_context *blkdev)
 507 {
 508         struct blkvsc_request *blkvsc_req;
 509
 510         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 511
 512         if (blkdev->device_type != HARDDISK_TYPE)
 513                 return 0;
 514
 515         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 516         if (!blkvsc_req)
 517                 return -ENOMEM;
 518
 519         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 520         init_waitqueue_head(&blkvsc_req->wevent);
 521         blkvsc_req->dev = blkdev;
 522         blkvsc_req->req = NULL;
 523         blkvsc_req->write = 0;
 524
 525         blkvsc_req->request.extension.data_buffer.pfn_array[0] = 0;
 526         blkvsc_req->request.extension.data_buffer.offset = 0;
 527         blkvsc_req->request.extension.data_buffer.len = 0;
 528
 529         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 530         blkvsc_req->cmd_len = 10;
 531
 532         /*
 533          * Set this here since the completion routine may be invoked and
 534          * completed before we return
 535          */
 536         blkvsc_req->cond = 0;
 537         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 538
 539         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 540
 541         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 542
 543         return 0;
 544 }
 545
 546 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 547 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 548 {
 549         struct blkvsc_request *blkvsc_req;
 550         struct page *page_buf;
 551         unsigned char *buf;
 552         unsigned char device_type;
 553
 554         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 555
 556         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 557         if (!blkvsc_req)
 558                 return -ENOMEM;
 559
 560         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 561         page_buf = alloc_page(GFP_KERNEL);
 562         if (!page_buf) {
 563                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 564                 return -ENOMEM;
 565         }
 566
 567         init_waitqueue_head(&blkvsc_req->wevent);
 568         blkvsc_req->dev = blkdev;
 569         blkvsc_req->req = NULL;
 570         blkvsc_req->write = 0;
 571
 572         blkvsc_req->request.extension.data_buffer.pfn_array[0] =
 573         page_to_pfn(page_buf);
 574         blkvsc_req->request.extension.data_buffer.offset = 0;
 575         blkvsc_req->request.extension.data_buffer.len = 64;
 576
 577         blkvsc_req->cmnd[0] = INQUIRY;
 578         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 579         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 580         blkvsc_req->cmnd[4] = 64;
 581         blkvsc_req->cmd_len = 6;
 582
 583         /*
 584          * Set this here since the completion routine may be invoked and
 585          * completed before we return
 586          */
 587         blkvsc_req->cond = 0;
 588
 589         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 590
 591         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 592                    blkvsc_req, blkvsc_req->cond);
 593
 594         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 595
 596         buf = kmap(page_buf);
 597
 598         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 599         /* be to le */
 600         device_type = buf[0] & 0x1F;
 601
 602         if (device_type == 0x0) {
 603                 blkdev->device_type = HARDDISK_TYPE;
 604         } else if (device_type == 0x5) {
 605                 blkdev->device_type = DVD_TYPE;
 606         } else {
 607                 /* TODO: this is currently unsupported device type */
 608                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 609         }
 610
 611         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 612
 613         blkdev->device_id_len = buf[7];
 614         if (blkdev->device_id_len > 64)
 615                 blkdev->device_id_len = 64;
 616
 617         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 618         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 619          * blkdev->device_id_len); */
 620
 621         kunmap(page_buf);
 622
 623         __free_page(page_buf);
 624
 625         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 626
 627         return 0;
 628 }
 629
 630 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 631 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 632 {
 633         struct blkvsc_request *blkvsc_req;
 634         struct page *page_buf;
 635         unsigned char *buf;
 636         struct scsi_sense_hdr sense_hdr;
 637         struct vmscsi_request *vm_srb;
 638
 639         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 640
 641         blkdev->sector_size = 0;
 642         blkdev->capacity = 0;
 643         blkdev->media_not_present = 0; /* assume a disk is present */
 644
 645         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 646         if (!blkvsc_req)
 647                 return -ENOMEM;
 648
 649         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 650         page_buf = alloc_page(GFP_KERNEL);
 651         if (!page_buf) {
 652                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 653                 return -ENOMEM;
 654         }
 655
 656         vm_srb = &blkvsc_req->request.extension.vstor_packet.vm_srb;
 657         init_waitqueue_head(&blkvsc_req->wevent);
 658         blkvsc_req->dev = blkdev;
 659         blkvsc_req->req = NULL;
 660         blkvsc_req->write = 0;
 661
 662         blkvsc_req->request.extension.data_buffer.pfn_array[0] =
 663         page_to_pfn(page_buf);
 664         blkvsc_req->request.extension.data_buffer.offset = 0;
 665         blkvsc_req->request.extension.data_buffer.len = 8;
 666
 667         blkvsc_req->cmnd[0] = READ_CAPACITY;
 668         blkvsc_req->cmd_len = 16;
 669
 670         /*
 671          * Set this here since the completion routine may be invoked
 672          * and completed before we return
 673          */
 674         blkvsc_req->cond = 0;
 675
 676         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 677
 678         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 679                    blkvsc_req, blkvsc_req->cond);
 680
 681         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 682
 683         /* check error */
 684         if (vm_srb->scsi_status) {
 685                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 686                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 687
 688                 if (sense_hdr.asc == 0x3A) {
 689                         /* Medium not present */
 690                         blkdev->media_not_present = 1;
 691                 }
 692                 return 0;
 693         }
 694         buf = kmap(page_buf);
 695
 696         /* be to le */
 697         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 698                             (buf[2] << 8) | buf[3]) + 1;
 699         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 700                               (buf[6] << 8) | buf[7];
 701
 702         kunmap(page_buf);
 703
 704         __free_page(page_buf);
 705
 706         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 707
 708         return 0;
 709 }
 710
 711 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 712 {
 713         struct blkvsc_request *blkvsc_req;
 714         struct page *page_buf;
 715         unsigned char *buf;
 716         struct scsi_sense_hdr sense_hdr;
 717         struct vmscsi_request *vm_srb;
 718
 719         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 720
 721         blkdev->sector_size = 0;
 722         blkdev->capacity = 0;
 723         blkdev->media_not_present = 0; /* assume a disk is present */
 724
 725         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 726         if (!blkvsc_req)
 727                 return -ENOMEM;
 728
 729         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 730         vm_srb = &blkvsc_req->request.extension.vstor_packet.vm_srb;
 731         page_buf = alloc_page(GFP_KERNEL);
 732         if (!page_buf) {
 733                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 734                 return -ENOMEM;
 735         }
 736
 737         init_waitqueue_head(&blkvsc_req->wevent);
 738         blkvsc_req->dev = blkdev;
 739         blkvsc_req->req = NULL;
 740         blkvsc_req->write = 0;
 741
 742         blkvsc_req->request.extension.data_buffer.pfn_array[0] =
 743         page_to_pfn(page_buf);
 744         blkvsc_req->request.extension.data_buffer.offset = 0;
 745         blkvsc_req->request.extension.data_buffer.len = 12;
 746
 747         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 748         blkvsc_req->cmd_len = 16;
 749
 750         /*
 751          * Set this here since the completion routine may be invoked
 752          * and completed before we return
 753          */
 754         blkvsc_req->cond = 0;
 755
 756         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 757
 758         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete - cond %d\n",
 759                    blkvsc_req, blkvsc_req->cond);
 760
 761         wait_event_interruptible(blkvsc_req->wevent, blkvsc_req->cond);
 762
 763         /* check error */
 764         if (vm_srb->scsi_status) {
 765                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 766                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 767                 if (sense_hdr.asc == 0x3A) {
 768                         /* Medium not present */
 769                         blkdev->media_not_present = 1;
 770                 }
 771                 return 0;
 772         }
 773         buf = kmap(page_buf);
 774
 775         /* be to le */
 776         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 777         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 778
 779 #if 0
 780         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 781                             (buf[2] << 8) | buf[3]) + 1;
 782         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 783                               (buf[6] << 8) | buf[7];
 784 #endif
 785
 786         kunmap(page_buf);
 787
 788         __free_page(page_buf);
 789
 790         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 791
 792         return 0;
 793 }
 794
 795 /*
 796  * blkvsc_remove() - Callback when our device is removed
 797  */
 798 static int blkvsc_remove(struct device *device)
 799 {
 800         struct hv_driver *drv =
 801                                 drv_to_hv_drv(device->driver);
 802         struct storvsc_driver_object *storvsc_drv_obj =
 803                                 drv->priv;
 804         struct hv_device *device_obj = device_to_hv_device(device);
 805         struct block_device_context *blkdev = dev_get_drvdata(device);
 806         unsigned long flags;
 807         int ret;
 808
 809         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 810
 811         if (!storvsc_drv_obj->base.dev_rm)
 812                 return -1;
 813
 814         /*
 815          * Call to the vsc driver to let it know that the device is being
 816          * removed
 817          */
 818         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 819         if (ret != 0) {
 820                 /* TODO: */
 821                 DPRINT_ERR(BLKVSC_DRV,
 822                            "unable to remove blkvsc device (ret %d)", ret);
 823         }
 824
 825         /* Get to a known state */
 826         spin_lock_irqsave(&blkdev->lock, flags);
 827
 828         blkdev->shutting_down = 1;
 829
 830         blk_stop_queue(blkdev->gd->queue);
 831
 832         spin_unlock_irqrestore(&blkdev->lock, flags);
 833
 834         while (blkdev->num_outstanding_reqs) {
 835                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 836                             blkdev->num_outstanding_reqs);
 837                 udelay(100);
 838         }
 839
 840         blkvsc_do_flush(blkdev);
 841
 842         spin_lock_irqsave(&blkdev->lock, flags);
 843
 844         blkvsc_cancel_pending_reqs(blkdev);
 845
 846         spin_unlock_irqrestore(&blkdev->lock, flags);
 847
 848         blk_cleanup_queue(blkdev->gd->queue);
 849
 850         del_gendisk(blkdev->gd);
 851
 852         kmem_cache_destroy(blkdev->request_pool);
 853
 854         kfree(blkdev);
 855
 856         return ret;
 857 }
 858
 859 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 860 {
 861         /* ASSERT(blkvsc_req->req); */
 862         /* ASSERT(blkvsc_req->sector_count <= (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 863
 864         blkvsc_req->cmd_len = 16;
 865
 866         if (blkvsc_req->sector_start > 0xffffffff) {
 867                 if (rq_data_dir(blkvsc_req->req)) {
 868                         blkvsc_req->write = 1;
 869                         blkvsc_req->cmnd[0] = WRITE_16;
 870                 } else {
 871                         blkvsc_req->write = 0;
 872                         blkvsc_req->cmnd[0] = READ_16;
 873                 }
 874
 875                 blkvsc_req->cmnd[1] |=
 876                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 877
 878                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 879                                 cpu_to_be64(blkvsc_req->sector_start);
 880                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 881                                 cpu_to_be32(blkvsc_req->sector_count);
 882         } else if ((blkvsc_req->sector_count > 0xff) ||
 883                    (blkvsc_req->sector_start > 0x1fffff)) {
 884                 if (rq_data_dir(blkvsc_req->req)) {
 885                         blkvsc_req->write = 1;
 886                         blkvsc_req->cmnd[0] = WRITE_10;
 887                 } else {
 888                         blkvsc_req->write = 0;
 889                         blkvsc_req->cmnd[0] = READ_10;
 890                 }
 891
 892                 blkvsc_req->cmnd[1] |=
 893                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 894
 895                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 896                                 cpu_to_be32(blkvsc_req->sector_start);
 897                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 898                                 cpu_to_be16(blkvsc_req->sector_count);
 899         } else {
 900                 if (rq_data_dir(blkvsc_req->req)) {
 901                         blkvsc_req->write = 1;
 902                         blkvsc_req->cmnd[0] = WRITE_6;
 903                 } else {
 904                         blkvsc_req->write = 0;
 905                         blkvsc_req->cmnd[0] = READ_6;
 906                 }
 907
 908                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 909                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 910                 blkvsc_req->cmnd[1] &= 0x1f;
 911                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 912         }
 913 }
 914
 915 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 916                         void (*request_completion)(struct hv_storvsc_request *))
 917 {
 918         struct block_device_context *blkdev = blkvsc_req->dev;
 919         struct hv_device *device_ctx = blkdev->device_ctx;
 920         struct hv_driver *drv =
 921                         drv_to_hv_drv(device_ctx->device.driver);
 922         struct storvsc_driver_object *storvsc_drv_obj =
 923                         drv->priv;
 924         struct hv_storvsc_request *storvsc_req;
 925         struct vmscsi_request *vm_srb;
 926         int ret;
 927
 928         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 929                    "req %p type %s start_sector %lu count %ld offset %d "
 930                    "len %d\n", blkvsc_req,
 931                    (blkvsc_req->write) ? "WRITE" : "READ",
 932                    (unsigned long) blkvsc_req->sector_start,
 933                    blkvsc_req->sector_count,
 934                    blkvsc_req->request.extension.data_buffer.offset,
 935                    blkvsc_req->request.extension.data_buffer.len);
 936 #if 0
 937         for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
 938                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 939                            "req %p pfn[%d] %llx\n",
 940                            blkvsc_req, i,
 941                            blkvsc_req->request.data_buffer.pfn_array[i]);
 942         }
 943 #endif
 944
 945         storvsc_req = &blkvsc_req->request;
 946         vm_srb = &storvsc_req->extension.vstor_packet.vm_srb;
 947
 948         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 949
 950         storvsc_req->extension.on_io_completion = request_completion;
 951         storvsc_req->extension.context = blkvsc_req;
 952
 953         vm_srb->port_number = blkdev->port;
 954         vm_srb->path_id = blkdev->path;
 955         vm_srb->target_id = blkdev->target;
 956         vm_srb->lun = 0;         /* this is not really used at all */
 957
 958         vm_srb->cdb_length = blkvsc_req->cmd_len;
 959
 960         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 961
 962         storvsc_req->extension.sense_buffer = blkvsc_req->sense_buffer;
 963
 964         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 965                                            &blkvsc_req->request);
 966         if (ret == 0)
 967                 blkdev->num_outstanding_reqs++;
 968
 969         return ret;
 970 }
 971
 972 /*
 973  * We break the request into 1 or more blkvsc_requests and submit
 974  * them.  If we cant submit them all, we put them on the
 975  * pending_list. The blkvsc_request() will work on the pending_list.
 976  */
 977 static int blkvsc_do_request(struct block_device_context *blkdev,
 978                              struct request *req)
 979 {
 980         struct bio *bio = NULL;
 981         struct bio_vec *bvec = NULL;
 982         struct bio_vec *prev_bvec = NULL;
 983         struct blkvsc_request *blkvsc_req = NULL;
 984         struct blkvsc_request *tmp;
 985         int databuf_idx = 0;
 986         int seg_idx = 0;
 987         sector_t start_sector;
 988         unsigned long num_sectors = 0;
 989         int ret = 0;
 990         int pending = 0;
 991         struct blkvsc_request_group *group = NULL;
 992
 993         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 994                   (unsigned long)blk_rq_pos(req));
 995
 996         /* Create a group to tie req to list of blkvsc_reqs */
 997         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 998         if (!group)
 999                 return -ENOMEM;
1000
1001         INIT_LIST_HEAD(&group->blkvsc_req_list);
1002         group->outstanding = group->status = 0;
1003
1004         start_sector = blk_rq_pos(req);
1005
1006         /* foreach bio in the request */
1007         if (req->bio) {
1008                 for (bio = req->bio; bio; bio = bio->bi_next) {
1009                         /*
1010                          * Map this bio into an existing or new storvsc request
1011                          */
1012                         bio_for_each_segment(bvec, bio, seg_idx) {
1013                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
1014                                            "- req %p bio %p bvec %p seg_idx %d "
1015                                            "databuf_idx %d\n", req, bio, bvec,
1016                                            seg_idx, databuf_idx);
1017
1018                                 /* Get a new storvsc request */
1019                                 /* 1st-time */
1020                                 if ((!blkvsc_req) ||
1021                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
1022                                     /* hole at the begin of page */
1023                                     || (bvec->bv_offset != 0) ||
1024                                     /* hold at the end of page */
1025                                     (prev_bvec &&
1026                                      (prev_bvec->bv_len != PAGE_SIZE))) {
1027                                         /* submit the prev one */
1028                                         if (blkvsc_req) {
1029                                                 blkvsc_req->sector_start = start_sector;
1030                                                 sector_div(blkvsc_req->sector_start, (blkdev->sector_size >> 9));
1031
1032                                                 blkvsc_req->sector_count = num_sectors / (blkdev->sector_size >> 9);
1033                                                 blkvsc_init_rw(blkvsc_req);
1034                                         }
1035
1036                                         /*
1037                                          * Create new blkvsc_req to represent
1038                                          * the current bvec
1039                                          */
1040                                         blkvsc_req =
1041                                         kmem_cache_zalloc(
1042                                         blkdev->request_pool, GFP_ATOMIC);
1043                                         if (!blkvsc_req) {
1044                                                 /* free up everything */
1045                                                 list_for_each_entry_safe(
1046                                                         blkvsc_req, tmp,
1047                                                         &group->blkvsc_req_list,
1048                                                         req_entry) {
1049                                                         list_del(&blkvsc_req->req_entry);
1050                                                         kmem_cache_free(blkdev->request_pool, blkvsc_req);
1051                                                 }
1052
1053                                                 kmem_cache_free(blkdev->request_pool, group);
1054                                                 return -ENOMEM;
1055                                         }
1056
1057                                         memset(blkvsc_req, 0,
1058                                                sizeof(struct blkvsc_request));
1059
1060                                         blkvsc_req->dev = blkdev;
1061                                         blkvsc_req->req = req;
1062                                         blkvsc_req->request.extension.
1063                                         data_buffer.offset
1064                                         = bvec->bv_offset;
1065                                         blkvsc_req->request.extension.
1066                                         data_buffer.len = 0;
1067
1068                                         /* Add to the group */
1069                                         blkvsc_req->group = group;
1070                                         blkvsc_req->group->outstanding++;
1071                                         list_add_tail(&blkvsc_req->req_entry,
1072                                                 &blkvsc_req->group->blkvsc_req_list);
1073
1074                                         start_sector += num_sectors;
1075                                         num_sectors = 0;
1076                                         databuf_idx = 0;
1077                                 }
1078
1079                                 /* Add the curr bvec/segment to the curr blkvsc_req */
1080                                 blkvsc_req->request.extension.data_buffer.
1081                                         pfn_array[databuf_idx]
1082                                                 = page_to_pfn(bvec->bv_page);
1083                                 blkvsc_req->request.extension.data_buffer.len
1084                                         += bvec->bv_len;
1085
1086                                 prev_bvec = bvec;
1087
1088                                 databuf_idx++;
1089                                 num_sectors += bvec->bv_len >> 9;
1090
1091                         } /* bio_for_each_segment */
1092
1093                 } /* rq_for_each_bio */
1094         }
1095
1096         /* Handle the last one */
1097         if (blkvsc_req) {
1098                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1099                            blkdev, req, blkvsc_req->group,
1100                            blkvsc_req->group->outstanding);
1101
1102                 blkvsc_req->sector_start = start_sector;
1103                 sector_div(blkvsc_req->sector_start,
1104                            (blkdev->sector_size >> 9));
1105
1106                 blkvsc_req->sector_count = num_sectors /
1107                                            (blkdev->sector_size >> 9);
1108
1109                 blkvsc_init_rw(blkvsc_req);
1110         }
1111
1112         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1113                 if (pending) {
1114                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1115                                    "pending_list - blkvsc_req %p start_sect %lu"
1116                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1117                                    (unsigned long)blkvsc_req->sector_start,
1118                                    blkvsc_req->sector_count,
1119                                    (unsigned long)start_sector,
1120                                    (unsigned long)num_sectors);
1121
1122                         list_add_tail(&blkvsc_req->pend_entry,
1123                                       &blkdev->pending_list);
1124                 } else {
1125                         ret = blkvsc_submit_request(blkvsc_req,
1126                                                     blkvsc_request_completion);
1127                         if (ret == -1) {
1128                                 pending = 1;
1129                                 list_add_tail(&blkvsc_req->pend_entry,
1130                                               &blkdev->pending_list);
1131                         }
1132
1133                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1134                                    "start_sect %lu sect_count %ld (%lu %ld) "
1135                                    "ret %d\n", blkvsc_req,
1136                                    (unsigned long)blkvsc_req->sector_start,
1137                                    blkvsc_req->sector_count,
1138                                    (unsigned long)start_sector,
1139                                    num_sectors, ret);
1140                 }
1141         }
1142
1143         return pending;
1144 }
1145
1146 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
1147 {
1148         struct blkvsc_request *blkvsc_req =
1149                         (struct blkvsc_request *)request->extension.context;
1150         struct block_device_context *blkdev =
1151                         (struct block_device_context *)blkvsc_req->dev;
1152         struct scsi_sense_hdr sense_hdr;
1153         struct vmscsi_request *vm_srb;
1154
1155         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
1156                    blkvsc_req);
1157
1158         vm_srb = &blkvsc_req->request.extension.vstor_packet.vm_srb;
1159         blkdev->num_outstanding_reqs--;
1160
1161         if (vm_srb->scsi_status)
1162                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
1163                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
1164                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
1165
1166         blkvsc_req->cond = 1;
1167         wake_up_interruptible(&blkvsc_req->wevent);
1168 }
1169
1170 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1171 {
1172         struct blkvsc_request *blkvsc_req =
1173                         (struct blkvsc_request *)request->extension.context;
1174         struct block_device_context *blkdev =
1175                         (struct block_device_context *)blkvsc_req->dev;
1176         unsigned long flags;
1177         struct blkvsc_request *comp_req, *tmp;
1178         struct vmscsi_request *vm_srb;
1179
1180         /* ASSERT(blkvsc_req->group); */
1181
1182         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1183                    "sect_start %lu sect_count %ld len %d group outstd %d "
1184                    "total outstd %d\n",
1185                    blkdev, blkvsc_req, blkvsc_req->group,
1186                    (blkvsc_req->write) ? "WRITE" : "READ",
1187                    (unsigned long)blkvsc_req->sector_start,
1188                    blkvsc_req->sector_count,
1189                    blkvsc_req->request.extension.data_buffer.len,
1190                    blkvsc_req->group->outstanding,
1191                    blkdev->num_outstanding_reqs);
1192
1193         spin_lock_irqsave(&blkdev->lock, flags);
1194
1195         blkdev->num_outstanding_reqs--;
1196         blkvsc_req->group->outstanding--;
1197
1198         /*
1199          * Only start processing when all the blkvsc_reqs are
1200          * completed. This guarantees no out-of-order blkvsc_req
1201          * completion when calling end_that_request_first()
1202          */
1203         if (blkvsc_req->group->outstanding == 0) {
1204                 list_for_each_entry_safe(comp_req, tmp,
1205                                          &blkvsc_req->group->blkvsc_req_list,
1206                                          req_entry) {
1207                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1208                                    "sect_start %lu sect_count %ld\n",
1209                                    comp_req,
1210                                    (unsigned long)comp_req->sector_start,
1211                                    comp_req->sector_count);
1212
1213                         list_del(&comp_req->req_entry);
1214
1215                         vm_srb =
1216                         &comp_req->request.extension.vstor_packet.vm_srb;
1217                         if (!__blk_end_request(comp_req->req,
1218                                 (!vm_srb->scsi_status ? 0 : -EIO),
1219                                 comp_req->sector_count * blkdev->sector_size)) {
1220                                 /*
1221                                  * All the sectors have been xferred ie the
1222                                  * request is done
1223                                  */
1224                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1225                                            comp_req->req);
1226                                 kmem_cache_free(blkdev->request_pool,
1227                                                 comp_req->group);
1228                         }
1229
1230                         kmem_cache_free(blkdev->request_pool, comp_req);
1231                 }
1232
1233                 if (!blkdev->shutting_down) {
1234                         blkvsc_do_pending_reqs(blkdev);
1235                         blk_start_queue(blkdev->gd->queue);
1236                         blkvsc_request(blkdev->gd->queue);
1237                 }
1238         }
1239
1240         spin_unlock_irqrestore(&blkdev->lock, flags);
1241 }
1242
1243 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
1244 {
1245         struct blkvsc_request *pend_req, *tmp;
1246         struct blkvsc_request *comp_req, *tmp2;
1247         struct vmscsi_request *vm_srb;
1248
1249         int ret = 0;
1250
1251         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
1252
1253         /* Flush the pending list first */
1254         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1255                                  pend_entry) {
1256                 /*
1257                  * The pend_req could be part of a partially completed
1258                  * request. If so, complete those req first until we
1259                  * hit the pend_req
1260                  */
1261                 list_for_each_entry_safe(comp_req, tmp2,
1262                                          &pend_req->group->blkvsc_req_list,
1263                                          req_entry) {
1264                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1265                                    "sect_start %lu sect_count %ld\n",
1266                                    comp_req,
1267                                    (unsigned long) comp_req->sector_start,
1268                                    comp_req->sector_count);
1269
1270                         if (comp_req == pend_req)
1271                                 break;
1272
1273                         list_del(&comp_req->req_entry);
1274
1275                         if (comp_req->req) {
1276                                 vm_srb =
1277                                 &comp_req->request.extension.vstor_packet.
1278                                 vm_srb;
1279                                 ret = __blk_end_request(comp_req->req,
1280                                         (!vm_srb->scsi_status ? 0 : -EIO),
1281                                         comp_req->sector_count *
1282                                         blkdev->sector_size);
1283
1284                                 /* FIXME: shouldn't this do more than return? */
1285                                 if (ret)
1286                                         goto out;
1287                         }
1288
1289                         kmem_cache_free(blkdev->request_pool, comp_req);
1290                 }
1291
1292                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
1293                            pend_req);
1294
1295                 list_del(&pend_req->pend_entry);
1296
1297                 list_del(&pend_req->req_entry);
1298
1299                 if (comp_req->req) {
1300                         if (!__blk_end_request(pend_req->req, -EIO,
1301                                                pend_req->sector_count *
1302                                                blkdev->sector_size)) {
1303                                 /*
1304                                  * All the sectors have been xferred ie the
1305                                  * request is done
1306                                  */
1307                                 DPRINT_DBG(BLKVSC_DRV,
1308                                            "blkvsc_cancel_pending_reqs() - "
1309                                            "req %p COMPLETED\n", pend_req->req);
1310                                 kmem_cache_free(blkdev->request_pool,
1311                                                 pend_req->group);
1312                         }
1313                 }
1314
1315                 kmem_cache_free(blkdev->request_pool, pend_req);
1316         }
1317
1318 out:
1319         return ret;
1320 }
1321
1322 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1323 {
1324         struct blkvsc_request *pend_req, *tmp;
1325         int ret = 0;
1326
1327         /* Flush the pending list first */
1328         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1329                                  pend_entry) {
1330                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1331                            pend_req);
1332
1333                 ret = blkvsc_submit_request(pend_req,
1334                                             blkvsc_request_completion);
1335                 if (ret != 0)
1336                         break;
1337                 else
1338                         list_del(&pend_req->pend_entry);
1339         }
1340
1341         return ret;
1342 }
1343
1344 static void blkvsc_request(struct request_queue *queue)
1345 {
1346         struct block_device_context *blkdev = NULL;
1347         struct request *req;
1348         int ret = 0;
1349
1350         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1351         while ((req = blk_peek_request(queue)) != NULL) {
1352                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1353
1354                 blkdev = req->rq_disk->private_data;
1355                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1356                     blkdev->media_not_present) {
1357                         __blk_end_request_cur(req, 0);
1358                         continue;
1359                 }
1360
1361                 ret = blkvsc_do_pending_reqs(blkdev);
1362
1363                 if (ret != 0) {
1364                         DPRINT_DBG(BLKVSC_DRV,
1365                                    "- stop queue - pending_list not empty\n");
1366                         blk_stop_queue(queue);
1367                         break;
1368                 }
1369
1370                 blk_start_request(req);
1371
1372                 ret = blkvsc_do_request(blkdev, req);
1373                 if (ret > 0) {
1374                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1375                         blk_stop_queue(queue);
1376                         break;
1377                 } else if (ret < 0) {
1378                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1379                         blk_requeue_request(queue, req);
1380                         blk_stop_queue(queue);
1381                         break;
1382                 }
1383         }
1384 }
1385
1386 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
1387 {
1388         struct block_device_context *blkdev = bdev->bd_disk->private_data;
1389
1390         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1391                    blkdev->gd->disk_name);
1392
1393         mutex_lock(&blkvsc_mutex);
1394         spin_lock(&blkdev->lock);
1395
1396         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
1397                 spin_unlock(&blkdev->lock);
1398                 check_disk_change(bdev);
1399                 spin_lock(&blkdev->lock);
1400         }
1401
1402         blkdev->users++;
1403
1404         spin_unlock(&blkdev->lock);
1405         mutex_unlock(&blkvsc_mutex);
1406         return 0;
1407 }
1408
1409 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
1410 {
1411         struct block_device_context *blkdev = disk->private_data;
1412
1413         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
1414                    blkdev->gd->disk_name);
1415
1416         mutex_lock(&blkvsc_mutex);
1417         spin_lock(&blkdev->lock);
1418         if (blkdev->users == 1) {
1419                 spin_unlock(&blkdev->lock);
1420                 blkvsc_do_flush(blkdev);
1421                 spin_lock(&blkdev->lock);
1422         }
1423
1424         blkdev->users--;
1425
1426         spin_unlock(&blkdev->lock);
1427         mutex_unlock(&blkvsc_mutex);
1428         return 0;
1429 }
1430
1431 static unsigned int blkvsc_check_events(struct gendisk *gd,
1432                                         unsigned int clearing)
1433 {
1434         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1435         return DISK_EVENT_MEDIA_CHANGE;
1436 }
1437
1438 static int blkvsc_revalidate_disk(struct gendisk *gd)
1439 {
1440         struct block_device_context *blkdev = gd->private_data;
1441
1442         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1443
1444         if (blkdev->device_type == DVD_TYPE) {
1445                 blkvsc_do_read_capacity(blkdev);
1446                 set_capacity(blkdev->gd, blkdev->capacity *
1447                             (blkdev->sector_size/512));
1448                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
1449         }
1450         return 0;
1451 }
1452
1453 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
1454 {
1455         sector_t total_sectors = get_capacity(bd->bd_disk);
1456         sector_t cylinder_times_heads = 0;
1457         sector_t temp = 0;
1458
1459         int sectors_per_track = 0;
1460         int heads = 0;
1461         int cylinders = 0;
1462         int rem = 0;
1463
1464         if (total_sectors > (65535 * 16 * 255))
1465                 total_sectors = (65535 * 16 * 255);
1466
1467         if (total_sectors >= (65535 * 16 * 63)) {
1468                 sectors_per_track = 255;
1469                 heads = 16;
1470
1471                 cylinder_times_heads = total_sectors;
1472                 /* sector_div stores the quotient in cylinder_times_heads */
1473                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1474         } else {
1475                 sectors_per_track = 17;
1476
1477                 cylinder_times_heads = total_sectors;
1478                 /* sector_div stores the quotient in cylinder_times_heads */
1479                 rem = sector_div(cylinder_times_heads, sectors_per_track);
1480
1481                 temp = cylinder_times_heads + 1023;
1482                 /* sector_div stores the quotient in temp */
1483                 rem = sector_div(temp, 1024);
1484
1485                 heads = temp;
1486
1487                 if (heads < 4)
1488                         heads = 4;
1489
1490
1491                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
1492                         sectors_per_track = 31;
1493                         heads = 16;
1494
1495                         cylinder_times_heads = total_sectors;
1496                         /*
1497                          * sector_div stores the quotient in
1498                          * cylinder_times_heads
1499                          */
1500                         rem = sector_div(cylinder_times_heads,
1501                                          sectors_per_track);
1502                 }
1503
1504                 if (cylinder_times_heads >= (heads * 1024)) {
1505                         sectors_per_track = 63;
1506                         heads = 16;
1507
1508                         cylinder_times_heads = total_sectors;
1509                         /*
1510                          * sector_div stores the quotient in
1511                          * cylinder_times_heads
1512                          */
1513                         rem = sector_div(cylinder_times_heads,
1514                                          sectors_per_track);
1515                 }
1516         }
1517
1518         temp = cylinder_times_heads;
1519         /* sector_div stores the quotient in temp */
1520         rem = sector_div(temp, heads);
1521         cylinders = temp;
1522
1523         hg->heads = heads;
1524         hg->sectors = sectors_per_track;
1525         hg->cylinders = cylinders;
1526
1527         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
1528                     sectors_per_track);
1529
1530     return 0;
1531 }
1532
1533 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
1534                         unsigned cmd, unsigned long argument)
1535 {
1536 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
1537         int ret;
1538
1539         switch (cmd) {
1540         /*
1541          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
1542          * than just a GUID. Commented it out for now.
1543          */
1544 #if 0
1545         case HDIO_GET_IDENTITY:
1546                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
1547                 if (copy_to_user((void __user *)arg, blkdev->device_id,
1548                                  blkdev->device_id_len))
1549                         ret = -EFAULT;
1550                 break;
1551 #endif
1552         default:
1553                 ret = -EINVAL;
1554                 break;
1555         }
1556
1557         return ret;
1558 }
1559
1560 static int __init blkvsc_init(void)
1561 {
1562         int ret;
1563
1564         BUILD_BUG_ON(sizeof(sector_t) != 8);
1565
1566         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1567
1568         ret = blkvsc_drv_init();
1569
1570         return ret;
1571 }
1572
1573 static void __exit blkvsc_exit(void)
1574 {
1575         blkvsc_drv_exit();
1576 }
1577
1578 MODULE_LICENSE("GPL");
1579 MODULE_VERSION(HV_DRV_VERSION);
1580 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1581 module_init(blkvsc_init);
1582 module_exit(blkvsc_exit);