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