Pileus Git - ~andy/linux/blob - drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "hv_api.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 /*
  50  * This request ties the struct request and struct
  51  * blkvsc_request/hv_storvsc_request together A struct request may be
  52  * represented by 1 or more struct blkvsc_request
  53  */
  54 struct blkvsc_request_group {
  55         int outstanding;
  56         int status;
  57         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  58 };
  59
  60 struct blkvsc_request {
  61         /* blkvsc_request_group.blkvsc_req_list */
  62         struct list_head req_entry;
  63
  64         /* block_device_context.pending_list */
  65         struct list_head pend_entry;
  66
  67         /* This may be null if we generate a request internally */
  68         struct request *req;
  69
  70         struct block_device_context *dev;
  71
  72         /* The group this request is part of. Maybe null */
  73         struct blkvsc_request_group *group;
  74
  75         int write;
  76         sector_t sector_start;
  77         unsigned long sector_count;
  78
  79         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  80         unsigned char cmd_len;
  81         unsigned char cmnd[MAX_COMMAND_SIZE];
  82
  83         struct hv_storvsc_request request;
  84 };
  85
  86 /* Per device structure */
  87 struct block_device_context {
  88         /* point back to our device context */
  89         struct hv_device *device_ctx;
  90         struct kmem_cache *request_pool;
  91         spinlock_t lock;
  92         struct gendisk *gd;
  93         enum blkvsc_device_type device_type;
  94         struct list_head pending_list;
  95
  96         unsigned char device_id[64];
  97         unsigned int device_id_len;
  98         int num_outstanding_reqs;
  99         int shutting_down;
 100         int media_not_present;
 101         unsigned int sector_size;
 102         sector_t capacity;
 103         unsigned int port;
 104         unsigned char path;
 105         unsigned char target;
 106         int users;
 107 };
 108
 109 static DEFINE_MUTEX(blkvsc_mutex);
 110
 111 static const char *g_blk_driver_name = "blkvsc";
 112
 113 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 114 static const struct hv_guid g_blk_device_type = {
 115         .data = {
 116                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 117                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 118         }
 119 };
 120
 121 /*
 122  * There is a circular dependency involving blkvsc_request_completion()
 123  * and blkvsc_do_request().
 124  */
 125 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 126
 127 static int blk_vsc_on_device_add(struct hv_device *device,
 128                                 void *additional_info)
 129 {
 130         struct storvsc_device_info *device_info;
 131         int ret = 0;
 132
 133         device_info = (struct storvsc_device_info *)additional_info;
 134
 135         ret = stor_vsc_on_device_add(device, additional_info);
 136         if (ret != 0)
 137                 return ret;
 138
 139         /*
 140          * We need to use the device instance guid to set the path and target
 141          * id. For IDE devices, the device instance id is formatted as
 142          * <bus id> * - <device id> - 8899 - 000000000000.
 143          */
 144         device_info->path_id = device->dev_instance.data[3] << 24 |
 145                              device->dev_instance.data[2] << 16 |
 146                              device->dev_instance.data[1] << 8  |
 147                              device->dev_instance.data[0];
 148
 149         device_info->target_id = device->dev_instance.data[5] << 8 |
 150                                device->dev_instance.data[4];
 151
 152         return ret;
 153 }
 154
 155
 156 static int blk_vsc_initialize(struct hv_driver *driver)
 157 {
 158         struct storvsc_driver_object *stor_driver;
 159         int ret = 0;
 160
 161         stor_driver = hvdr_to_stordr(driver);
 162
 163         /* Make sure we are at least 2 pages since 1 page is used for control */
 164         /* ASSERT(stor_driver->RingBufferSize >= (PAGE_SIZE << 1)); */
 165
 166         driver->name = g_blk_driver_name;
 167         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 168
 169
 170         /*
 171          * Divide the ring buffer data size (which is 1 page less than the ring
 172          * buffer size since that page is reserved for the ring buffer indices)
 173          * by the max request size (which is
 174          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 175          */
 176         stor_driver->max_outstanding_req_per_channel =
 177                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 178                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 179                            sizeof(struct vstor_packet) + sizeof(u64),
 180                            sizeof(u64)));
 181
 182         DPRINT_INFO(BLKVSC, "max io outstd %u",
 183                     stor_driver->max_outstanding_req_per_channel);
 184
 185         /* Setup the dispatch table */
 186         stor_driver->base.dev_add = blk_vsc_on_device_add;
 187         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 188         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 189         stor_driver->on_io_request = stor_vsc_on_io_request;
 190
 191         return ret;
 192 }
 193
 194
 195 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 196                         void (*request_completion)(struct hv_storvsc_request *))
 197 {
 198         struct block_device_context *blkdev = blkvsc_req->dev;
 199         struct hv_device *device_ctx = blkdev->device_ctx;
 200         struct hv_driver *drv =
 201                         drv_to_hv_drv(device_ctx->device.driver);
 202         struct storvsc_driver_object *storvsc_drv_obj =
 203                         drv->priv;
 204         struct hv_storvsc_request *storvsc_req;
 205         struct vmscsi_request *vm_srb;
 206         int ret;
 207
 208         DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 209                    "req %p type %s start_sector %lu count %ld offset %d "
 210                    "len %d\n", blkvsc_req,
 211                    (blkvsc_req->write) ? "WRITE" : "READ",
 212                    (unsigned long) blkvsc_req->sector_start,
 213                    blkvsc_req->sector_count,
 214                    blkvsc_req->request.data_buffer.offset,
 215                    blkvsc_req->request.data_buffer.len);
 216 #if 0
 217         for (i = 0; i < (blkvsc_req->request.data_buffer.len >> 12); i++) {
 218                 DPRINT_DBG(BLKVSC_DRV, "blkvsc_submit_request() - "
 219                            "req %p pfn[%d] %llx\n",
 220                            blkvsc_req, i,
 221                            blkvsc_req->request.data_buffer.pfn_array[i]);
 222         }
 223 #endif
 224
 225         storvsc_req = &blkvsc_req->request;
 226         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 227
 228         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 229
 230         storvsc_req->on_io_completion = request_completion;
 231         storvsc_req->context = blkvsc_req;
 232
 233         vm_srb->port_number = blkdev->port;
 234         vm_srb->path_id = blkdev->path;
 235         vm_srb->target_id = blkdev->target;
 236         vm_srb->lun = 0;         /* this is not really used at all */
 237
 238         vm_srb->cdb_length = blkvsc_req->cmd_len;
 239
 240         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 241
 242         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 243
 244         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 245                                            &blkvsc_req->request);
 246         if (ret == 0)
 247                 blkdev->num_outstanding_reqs++;
 248
 249         return ret;
 250 }
 251
 252
 253 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
 254 {
 255         struct block_device_context *blkdev = bdev->bd_disk->private_data;
 256
 257         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
 258                    blkdev->gd->disk_name);
 259
 260         mutex_lock(&blkvsc_mutex);
 261         spin_lock(&blkdev->lock);
 262
 263         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
 264                 spin_unlock(&blkdev->lock);
 265                 check_disk_change(bdev);
 266                 spin_lock(&blkdev->lock);
 267         }
 268
 269         blkdev->users++;
 270
 271         spin_unlock(&blkdev->lock);
 272         mutex_unlock(&blkvsc_mutex);
 273         return 0;
 274 }
 275
 276
 277 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
 278 {
 279         sector_t total_sectors = get_capacity(bd->bd_disk);
 280         sector_t cylinder_times_heads = 0;
 281         sector_t temp = 0;
 282
 283         int sectors_per_track = 0;
 284         int heads = 0;
 285         int cylinders = 0;
 286         int rem = 0;
 287
 288         if (total_sectors > (65535 * 16 * 255))
 289                 total_sectors = (65535 * 16 * 255);
 290
 291         if (total_sectors >= (65535 * 16 * 63)) {
 292                 sectors_per_track = 255;
 293                 heads = 16;
 294
 295                 cylinder_times_heads = total_sectors;
 296                 /* sector_div stores the quotient in cylinder_times_heads */
 297                 rem = sector_div(cylinder_times_heads, sectors_per_track);
 298         } else {
 299                 sectors_per_track = 17;
 300
 301                 cylinder_times_heads = total_sectors;
 302                 /* sector_div stores the quotient in cylinder_times_heads */
 303                 rem = sector_div(cylinder_times_heads, sectors_per_track);
 304
 305                 temp = cylinder_times_heads + 1023;
 306                 /* sector_div stores the quotient in temp */
 307                 rem = sector_div(temp, 1024);
 308
 309                 heads = temp;
 310
 311                 if (heads < 4)
 312                         heads = 4;
 313
 314
 315                 if (cylinder_times_heads >= (heads * 1024) || (heads > 16)) {
 316                         sectors_per_track = 31;
 317                         heads = 16;
 318
 319                         cylinder_times_heads = total_sectors;
 320                         /*
 321                          * sector_div stores the quotient in
 322                          * cylinder_times_heads
 323                          */
 324                         rem = sector_div(cylinder_times_heads,
 325                                          sectors_per_track);
 326                 }
 327
 328                 if (cylinder_times_heads >= (heads * 1024)) {
 329                         sectors_per_track = 63;
 330                         heads = 16;
 331
 332                         cylinder_times_heads = total_sectors;
 333                         /*
 334                          * sector_div stores the quotient in
 335                          * cylinder_times_heads
 336                          */
 337                         rem = sector_div(cylinder_times_heads,
 338                                          sectors_per_track);
 339                 }
 340         }
 341
 342         temp = cylinder_times_heads;
 343         /* sector_div stores the quotient in temp */
 344         rem = sector_div(temp, heads);
 345         cylinders = temp;
 346
 347         hg->heads = heads;
 348         hg->sectors = sectors_per_track;
 349         hg->cylinders = cylinders;
 350
 351         DPRINT_INFO(BLKVSC_DRV, "CHS (%d, %d, %d)", cylinders, heads,
 352                     sectors_per_track);
 353
 354         return 0;
 355 }
 356
 357
 358 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 359 {
 360         /* ASSERT(blkvsc_req->req); */
 361         /* ASSERT(blkvsc_req->sector_count <=
 362            (MAX_MULTIPAGE_BUFFER_COUNT*8)); */
 363
 364         blkvsc_req->cmd_len = 16;
 365
 366         if (blkvsc_req->sector_start > 0xffffffff) {
 367                 if (rq_data_dir(blkvsc_req->req)) {
 368                         blkvsc_req->write = 1;
 369                         blkvsc_req->cmnd[0] = WRITE_16;
 370                 } else {
 371                         blkvsc_req->write = 0;
 372                         blkvsc_req->cmnd[0] = READ_16;
 373                 }
 374
 375                 blkvsc_req->cmnd[1] |=
 376                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 377
 378                 *(unsigned long long *)&blkvsc_req->cmnd[2] =
 379                                 cpu_to_be64(blkvsc_req->sector_start);
 380                 *(unsigned int *)&blkvsc_req->cmnd[10] =
 381                                 cpu_to_be32(blkvsc_req->sector_count);
 382         } else if ((blkvsc_req->sector_count > 0xff) ||
 383                    (blkvsc_req->sector_start > 0x1fffff)) {
 384                 if (rq_data_dir(blkvsc_req->req)) {
 385                         blkvsc_req->write = 1;
 386                         blkvsc_req->cmnd[0] = WRITE_10;
 387                 } else {
 388                         blkvsc_req->write = 0;
 389                         blkvsc_req->cmnd[0] = READ_10;
 390                 }
 391
 392                 blkvsc_req->cmnd[1] |=
 393                         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 394
 395                 *(unsigned int *)&blkvsc_req->cmnd[2] =
 396                                 cpu_to_be32(blkvsc_req->sector_start);
 397                 *(unsigned short *)&blkvsc_req->cmnd[7] =
 398                                 cpu_to_be16(blkvsc_req->sector_count);
 399         } else {
 400                 if (rq_data_dir(blkvsc_req->req)) {
 401                         blkvsc_req->write = 1;
 402                         blkvsc_req->cmnd[0] = WRITE_6;
 403                 } else {
 404                         blkvsc_req->write = 0;
 405                         blkvsc_req->cmnd[0] = READ_6;
 406                 }
 407
 408                 *(unsigned int *)&blkvsc_req->cmnd[1] =
 409                                 cpu_to_be32(blkvsc_req->sector_start) >> 8;
 410                 blkvsc_req->cmnd[1] &= 0x1f;
 411                 blkvsc_req->cmnd[4] = (unsigned char)blkvsc_req->sector_count;
 412         }
 413 }
 414
 415
 416 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 417                         unsigned cmd, unsigned long argument)
 418 {
 419 /*      struct block_device_context *blkdev = bd->bd_disk->private_data; */
 420         int ret;
 421
 422         switch (cmd) {
 423         /*
 424          * TODO: I think there is certain format for HDIO_GET_IDENTITY rather
 425          * than just a GUID. Commented it out for now.
 426          */
 427 #if 0
 428         case HDIO_GET_IDENTITY:
 429                 DPRINT_INFO(BLKVSC_DRV, "HDIO_GET_IDENTITY\n");
 430                 if (copy_to_user((void __user *)arg, blkdev->device_id,
 431                                  blkdev->device_id_len))
 432                         ret = -EFAULT;
 433                 break;
 434 #endif
 435         default:
 436                 ret = -EINVAL;
 437                 break;
 438         }
 439
 440         return ret;
 441 }
 442
 443 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
 444 {
 445         struct blkvsc_request *blkvsc_req =
 446                         (struct blkvsc_request *)request->context;
 447         struct block_device_context *blkdev =
 448                         (struct block_device_context *)blkvsc_req->dev;
 449         struct scsi_sense_hdr sense_hdr;
 450         struct vmscsi_request *vm_srb;
 451
 452         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cmd_completion() - req %p\n",
 453                    blkvsc_req);
 454
 455         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 456         blkdev->num_outstanding_reqs--;
 457
 458         if (vm_srb->scsi_status)
 459                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
 460                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 461                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
 462
 463         complete(&blkvsc_req->request.wait_event);
 464 }
 465
 466 static int blkvsc_do_flush(struct block_device_context *blkdev)
 467 {
 468         struct blkvsc_request *blkvsc_req;
 469
 470         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_flush()\n");
 471
 472         if (blkdev->device_type != HARDDISK_TYPE)
 473                 return 0;
 474
 475         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 476         if (!blkvsc_req)
 477                 return -ENOMEM;
 478
 479         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 480         init_completion(&blkvsc_req->request.wait_event);
 481         blkvsc_req->dev = blkdev;
 482         blkvsc_req->req = NULL;
 483         blkvsc_req->write = 0;
 484
 485         blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 486         blkvsc_req->request.data_buffer.offset = 0;
 487         blkvsc_req->request.data_buffer.len = 0;
 488
 489         blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 490         blkvsc_req->cmd_len = 10;
 491
 492         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 493
 494         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 495
 496         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 497
 498         return 0;
 499 }
 500
 501
 502 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
 503 {
 504         struct blkvsc_request *pend_req, *tmp;
 505         struct blkvsc_request *comp_req, *tmp2;
 506         struct vmscsi_request *vm_srb;
 507
 508         int ret = 0;
 509
 510         DPRINT_DBG(BLKVSC_DRV, "blkvsc_cancel_pending_reqs()");
 511
 512         /* Flush the pending list first */
 513         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 514                                  pend_entry) {
 515                 /*
 516                  * The pend_req could be part of a partially completed
 517                  * request. If so, complete those req first until we
 518                  * hit the pend_req
 519                  */
 520                 list_for_each_entry_safe(comp_req, tmp2,
 521                                          &pend_req->group->blkvsc_req_list,
 522                                          req_entry) {
 523                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
 524                                    "sect_start %lu sect_count %ld\n",
 525                                    comp_req,
 526                                    (unsigned long) comp_req->sector_start,
 527                                    comp_req->sector_count);
 528
 529                         if (comp_req == pend_req)
 530                                 break;
 531
 532                         list_del(&comp_req->req_entry);
 533
 534                         if (comp_req->req) {
 535                                 vm_srb =
 536                                 &comp_req->request.vstor_packet.
 537                                 vm_srb;
 538                                 ret = __blk_end_request(comp_req->req,
 539                                         (!vm_srb->scsi_status ? 0 : -EIO),
 540                                         comp_req->sector_count *
 541                                         blkdev->sector_size);
 542
 543                                 /* FIXME: shouldn't this do more than return? */
 544                                 if (ret)
 545                                         goto out;
 546                         }
 547
 548                         kmem_cache_free(blkdev->request_pool, comp_req);
 549                 }
 550
 551                 DPRINT_DBG(BLKVSC_DRV, "cancelling pending request - %p\n",
 552                            pend_req);
 553
 554                 list_del(&pend_req->pend_entry);
 555
 556                 list_del(&pend_req->req_entry);
 557
 558                 if (comp_req->req) {
 559                         if (!__blk_end_request(pend_req->req, -EIO,
 560                                                pend_req->sector_count *
 561                                                blkdev->sector_size)) {
 562                                 /*
 563                                  * All the sectors have been xferred ie the
 564                                  * request is done
 565                                  */
 566                                 DPRINT_DBG(BLKVSC_DRV,
 567                                            "blkvsc_cancel_pending_reqs() - "
 568                                            "req %p COMPLETED\n", pend_req->req);
 569                                 kmem_cache_free(blkdev->request_pool,
 570                                                 pend_req->group);
 571                         }
 572                 }
 573
 574                 kmem_cache_free(blkdev->request_pool, pend_req);
 575         }
 576
 577 out:
 578         return ret;
 579 }
 580
 581
 582 /*
 583  * blkvsc_remove() - Callback when our device is removed
 584  */
 585 static int blkvsc_remove(struct device *device)
 586 {
 587         struct hv_driver *drv =
 588                                 drv_to_hv_drv(device->driver);
 589         struct storvsc_driver_object *storvsc_drv_obj =
 590                                 drv->priv;
 591         struct hv_device *device_obj = device_to_hv_device(device);
 592         struct block_device_context *blkdev = dev_get_drvdata(device);
 593         unsigned long flags;
 594         int ret;
 595
 596         DPRINT_DBG(BLKVSC_DRV, "blkvsc_remove()\n");
 597
 598         if (!storvsc_drv_obj->base.dev_rm)
 599                 return -1;
 600
 601         /*
 602          * Call to the vsc driver to let it know that the device is being
 603          * removed
 604          */
 605         ret = storvsc_drv_obj->base.dev_rm(device_obj);
 606         if (ret != 0) {
 607                 /* TODO: */
 608                 DPRINT_ERR(BLKVSC_DRV,
 609                            "unable to remove blkvsc device (ret %d)", ret);
 610         }
 611
 612         /* Get to a known state */
 613         spin_lock_irqsave(&blkdev->lock, flags);
 614
 615         blkdev->shutting_down = 1;
 616
 617         blk_stop_queue(blkdev->gd->queue);
 618
 619         spin_unlock_irqrestore(&blkdev->lock, flags);
 620
 621         while (blkdev->num_outstanding_reqs) {
 622                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 623                             blkdev->num_outstanding_reqs);
 624                 udelay(100);
 625         }
 626
 627         blkvsc_do_flush(blkdev);
 628
 629         spin_lock_irqsave(&blkdev->lock, flags);
 630
 631         blkvsc_cancel_pending_reqs(blkdev);
 632
 633         spin_unlock_irqrestore(&blkdev->lock, flags);
 634
 635         blk_cleanup_queue(blkdev->gd->queue);
 636
 637         del_gendisk(blkdev->gd);
 638
 639         kmem_cache_destroy(blkdev->request_pool);
 640
 641         kfree(blkdev);
 642
 643         return ret;
 644 }
 645
 646 static void blkvsc_shutdown(struct device *device)
 647 {
 648         struct block_device_context *blkdev = dev_get_drvdata(device);
 649         unsigned long flags;
 650
 651         if (!blkdev)
 652                 return;
 653
 654         DPRINT_DBG(BLKVSC_DRV, "blkvsc_shutdown - users %d disk %s\n",
 655                    blkdev->users, blkdev->gd->disk_name);
 656
 657         spin_lock_irqsave(&blkdev->lock, flags);
 658
 659         blkdev->shutting_down = 1;
 660
 661         blk_stop_queue(blkdev->gd->queue);
 662
 663         spin_unlock_irqrestore(&blkdev->lock, flags);
 664
 665         while (blkdev->num_outstanding_reqs) {
 666                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 667                             blkdev->num_outstanding_reqs);
 668                 udelay(100);
 669         }
 670
 671         blkvsc_do_flush(blkdev);
 672
 673         spin_lock_irqsave(&blkdev->lock, flags);
 674
 675         blkvsc_cancel_pending_reqs(blkdev);
 676
 677         spin_unlock_irqrestore(&blkdev->lock, flags);
 678 }
 679
 680 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
 681 {
 682         struct block_device_context *blkdev = disk->private_data;
 683
 684         DPRINT_DBG(BLKVSC_DRV, "- users %d disk %s\n", blkdev->users,
 685                    blkdev->gd->disk_name);
 686
 687         mutex_lock(&blkvsc_mutex);
 688         spin_lock(&blkdev->lock);
 689         if (blkdev->users == 1) {
 690                 spin_unlock(&blkdev->lock);
 691                 blkvsc_do_flush(blkdev);
 692                 spin_lock(&blkdev->lock);
 693         }
 694
 695         blkdev->users--;
 696
 697         spin_unlock(&blkdev->lock);
 698         mutex_unlock(&blkvsc_mutex);
 699         return 0;
 700 }
 701
 702
 703 /* Do a scsi READ_CAPACITY cmd here to get the size of the disk */
 704 static int blkvsc_do_read_capacity(struct block_device_context *blkdev)
 705 {
 706         struct blkvsc_request *blkvsc_req;
 707         struct page *page_buf;
 708         unsigned char *buf;
 709         struct scsi_sense_hdr sense_hdr;
 710         struct vmscsi_request *vm_srb;
 711
 712         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity()\n");
 713
 714         blkdev->sector_size = 0;
 715         blkdev->capacity = 0;
 716         blkdev->media_not_present = 0; /* assume a disk is present */
 717
 718         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 719         if (!blkvsc_req)
 720                 return -ENOMEM;
 721
 722         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 723         page_buf = alloc_page(GFP_KERNEL);
 724         if (!page_buf) {
 725                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 726                 return -ENOMEM;
 727         }
 728
 729         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 730         init_completion(&blkvsc_req->request.wait_event);
 731         blkvsc_req->dev = blkdev;
 732         blkvsc_req->req = NULL;
 733         blkvsc_req->write = 0;
 734
 735         blkvsc_req->request.data_buffer.pfn_array[0] =
 736         page_to_pfn(page_buf);
 737         blkvsc_req->request.data_buffer.offset = 0;
 738         blkvsc_req->request.data_buffer.len = 8;
 739
 740         blkvsc_req->cmnd[0] = READ_CAPACITY;
 741         blkvsc_req->cmd_len = 16;
 742
 743         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 744
 745         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 746                    blkvsc_req);
 747
 748         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 749
 750         /* check error */
 751         if (vm_srb->scsi_status) {
 752                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 753                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 754
 755                 if (sense_hdr.asc == 0x3A) {
 756                         /* Medium not present */
 757                         blkdev->media_not_present = 1;
 758                 }
 759                 return 0;
 760         }
 761         buf = kmap(page_buf);
 762
 763         /* be to le */
 764         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 765                             (buf[2] << 8) | buf[3]) + 1;
 766         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 767                               (buf[6] << 8) | buf[7];
 768
 769         kunmap(page_buf);
 770
 771         __free_page(page_buf);
 772
 773         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 774
 775         return 0;
 776 }
 777
 778
 779 static int blkvsc_do_read_capacity16(struct block_device_context *blkdev)
 780 {
 781         struct blkvsc_request *blkvsc_req;
 782         struct page *page_buf;
 783         unsigned char *buf;
 784         struct scsi_sense_hdr sense_hdr;
 785         struct vmscsi_request *vm_srb;
 786
 787         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_read_capacity16()\n");
 788
 789         blkdev->sector_size = 0;
 790         blkdev->capacity = 0;
 791         blkdev->media_not_present = 0; /* assume a disk is present */
 792
 793         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 794         if (!blkvsc_req)
 795                 return -ENOMEM;
 796
 797         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 798         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 799         page_buf = alloc_page(GFP_KERNEL);
 800         if (!page_buf) {
 801                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 802                 return -ENOMEM;
 803         }
 804
 805         init_completion(&blkvsc_req->request.wait_event);
 806         blkvsc_req->dev = blkdev;
 807         blkvsc_req->req = NULL;
 808         blkvsc_req->write = 0;
 809
 810         blkvsc_req->request.data_buffer.pfn_array[0] =
 811         page_to_pfn(page_buf);
 812         blkvsc_req->request.data_buffer.offset = 0;
 813         blkvsc_req->request.data_buffer.len = 12;
 814
 815         blkvsc_req->cmnd[0] = 0x9E; /* READ_CAPACITY16; */
 816         blkvsc_req->cmd_len = 16;
 817
 818         /*
 819          * Set this here since the completion routine may be invoked
 820          * and completed before we return
 821          */
 822
 823         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 824
 825         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 826                    blkvsc_req);
 827
 828         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 829
 830         /* check error */
 831         if (vm_srb->scsi_status) {
 832                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 833                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 834                 if (sense_hdr.asc == 0x3A) {
 835                         /* Medium not present */
 836                         blkdev->media_not_present = 1;
 837                 }
 838                 return 0;
 839         }
 840         buf = kmap(page_buf);
 841
 842         /* be to le */
 843         blkdev->capacity = be64_to_cpu(*(unsigned long long *) &buf[0]) + 1;
 844         blkdev->sector_size = be32_to_cpu(*(unsigned int *)&buf[8]);
 845
 846 #if 0
 847         blkdev->capacity = ((buf[0] << 24) | (buf[1] << 16) |
 848                             (buf[2] << 8) | buf[3]) + 1;
 849         blkdev->sector_size = (buf[4] << 24) | (buf[5] << 16) |
 850                               (buf[6] << 8) | buf[7];
 851 #endif
 852
 853         kunmap(page_buf);
 854
 855         __free_page(page_buf);
 856
 857         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 858
 859         return 0;
 860 }
 861
 862 static int blkvsc_revalidate_disk(struct gendisk *gd)
 863 {
 864         struct block_device_context *blkdev = gd->private_data;
 865
 866         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
 867
 868         if (blkdev->device_type == DVD_TYPE) {
 869                 blkvsc_do_read_capacity(blkdev);
 870                 set_capacity(blkdev->gd, blkdev->capacity *
 871                             (blkdev->sector_size/512));
 872                 blk_queue_logical_block_size(gd->queue, blkdev->sector_size);
 873         }
 874         return 0;
 875 }
 876
 877
 878 /* Do a scsi INQUIRY cmd here to get the device type (ie disk or dvd) */
 879 static int blkvsc_do_inquiry(struct block_device_context *blkdev)
 880 {
 881         struct blkvsc_request *blkvsc_req;
 882         struct page *page_buf;
 883         unsigned char *buf;
 884         unsigned char device_type;
 885
 886         DPRINT_DBG(BLKVSC_DRV, "blkvsc_do_inquiry()\n");
 887
 888         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 889         if (!blkvsc_req)
 890                 return -ENOMEM;
 891
 892         memset(blkvsc_req, 0, sizeof(struct blkvsc_request));
 893         page_buf = alloc_page(GFP_KERNEL);
 894         if (!page_buf) {
 895                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 896                 return -ENOMEM;
 897         }
 898
 899         init_completion(&blkvsc_req->request.wait_event);
 900         blkvsc_req->dev = blkdev;
 901         blkvsc_req->req = NULL;
 902         blkvsc_req->write = 0;
 903
 904         blkvsc_req->request.data_buffer.pfn_array[0] =
 905         page_to_pfn(page_buf);
 906         blkvsc_req->request.data_buffer.offset = 0;
 907         blkvsc_req->request.data_buffer.len = 64;
 908
 909         blkvsc_req->cmnd[0] = INQUIRY;
 910         blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 911         blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 912         blkvsc_req->cmnd[4] = 64;
 913         blkvsc_req->cmd_len = 6;
 914
 915         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 916
 917         DPRINT_DBG(BLKVSC_DRV, "waiting %p to complete\n",
 918                    blkvsc_req);
 919
 920         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 921
 922         buf = kmap(page_buf);
 923
 924         /* print_hex_dump_bytes("", DUMP_PREFIX_NONE, buf, 64); */
 925         /* be to le */
 926         device_type = buf[0] & 0x1F;
 927
 928         if (device_type == 0x0) {
 929                 blkdev->device_type = HARDDISK_TYPE;
 930         } else if (device_type == 0x5) {
 931                 blkdev->device_type = DVD_TYPE;
 932         } else {
 933                 /* TODO: this is currently unsupported device type */
 934                 blkdev->device_type = UNKNOWN_DEV_TYPE;
 935         }
 936
 937         DPRINT_DBG(BLKVSC_DRV, "device type %d\n", device_type);
 938
 939         blkdev->device_id_len = buf[7];
 940         if (blkdev->device_id_len > 64)
 941                 blkdev->device_id_len = 64;
 942
 943         memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 944         /* printk_hex_dump_bytes("", DUMP_PREFIX_NONE, blkdev->device_id,
 945          * blkdev->device_id_len); */
 946
 947         kunmap(page_buf);
 948
 949         __free_page(page_buf);
 950
 951         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 952
 953         return 0;
 954 }
 955
 956
 957 /*
 958  * We break the request into 1 or more blkvsc_requests and submit
 959  * them.  If we cant submit them all, we put them on the
 960  * pending_list. The blkvsc_request() will work on the pending_list.
 961  */
 962 static int blkvsc_do_request(struct block_device_context *blkdev,
 963                              struct request *req)
 964 {
 965         struct bio *bio = NULL;
 966         struct bio_vec *bvec = NULL;
 967         struct bio_vec *prev_bvec = NULL;
 968         struct blkvsc_request *blkvsc_req = NULL;
 969         struct blkvsc_request *tmp;
 970         int databuf_idx = 0;
 971         int seg_idx = 0;
 972         sector_t start_sector;
 973         unsigned long num_sectors = 0;
 974         int ret = 0;
 975         int pending = 0;
 976         struct blkvsc_request_group *group = NULL;
 977
 978         DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p sect %lu\n", blkdev, req,
 979                   (unsigned long)blk_rq_pos(req));
 980
 981         /* Create a group to tie req to list of blkvsc_reqs */
 982         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 983         if (!group)
 984                 return -ENOMEM;
 985
 986         INIT_LIST_HEAD(&group->blkvsc_req_list);
 987         group->outstanding = group->status = 0;
 988
 989         start_sector = blk_rq_pos(req);
 990
 991         /* foreach bio in the request */
 992         if (req->bio) {
 993                 for (bio = req->bio; bio; bio = bio->bi_next) {
 994                         /*
 995                          * Map this bio into an existing or new storvsc request
 996                          */
 997                         bio_for_each_segment(bvec, bio, seg_idx) {
 998                                 DPRINT_DBG(BLKVSC_DRV, "bio_for_each_segment() "
 999                                            "- req %p bio %p bvec %p seg_idx %d "
1000                                            "databuf_idx %d\n", req, bio, bvec,
1001                                            seg_idx, databuf_idx);
1002
1003                                 /* Get a new storvsc request */
1004                                 /* 1st-time */
1005                                 if ((!blkvsc_req) ||
1006                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
1007                                     /* hole at the begin of page */
1008                                     || (bvec->bv_offset != 0) ||
1009                                     /* hold at the end of page */
1010                                     (prev_bvec &&
1011                                      (prev_bvec->bv_len != PAGE_SIZE))) {
1012                                         /* submit the prev one */
1013                                         if (blkvsc_req) {
1014                                                 blkvsc_req->sector_start =
1015                                                 start_sector;
1016                                                 sector_div(
1017                                                 blkvsc_req->sector_start,
1018                                                 (blkdev->sector_size >> 9));
1019
1020                                                 blkvsc_req->sector_count =
1021                                                 num_sectors /
1022                                                 (blkdev->sector_size >> 9);
1023                                                 blkvsc_init_rw(blkvsc_req);
1024                                         }
1025
1026                                         /*
1027                                          * Create new blkvsc_req to represent
1028                                          * the current bvec
1029                                          */
1030                                         blkvsc_req =
1031                                         kmem_cache_zalloc(
1032                                         blkdev->request_pool, GFP_ATOMIC);
1033                                         if (!blkvsc_req) {
1034                                                 /* free up everything */
1035                                                 list_for_each_entry_safe(
1036                                                         blkvsc_req, tmp,
1037                                                         &group->blkvsc_req_list,
1038                                                         req_entry) {
1039                                                         list_del(
1040                                                         &blkvsc_req->req_entry);
1041                                                         kmem_cache_free(
1042                                                         blkdev->request_pool,
1043                                                         blkvsc_req);
1044                                                 }
1045
1046                                                 kmem_cache_free(
1047                                                 blkdev->request_pool, group);
1048                                                 return -ENOMEM;
1049                                         }
1050
1051                                         memset(blkvsc_req, 0,
1052                                                sizeof(struct blkvsc_request));
1053
1054                                         blkvsc_req->dev = blkdev;
1055                                         blkvsc_req->req = req;
1056                                         blkvsc_req->request.
1057                                         data_buffer.offset
1058                                         = bvec->bv_offset;
1059                                         blkvsc_req->request.
1060                                         data_buffer.len = 0;
1061
1062                                         /* Add to the group */
1063                                         blkvsc_req->group = group;
1064                                         blkvsc_req->group->outstanding++;
1065                                         list_add_tail(&blkvsc_req->req_entry,
1066                                         &blkvsc_req->group->blkvsc_req_list);
1067
1068                                         start_sector += num_sectors;
1069                                         num_sectors = 0;
1070                                         databuf_idx = 0;
1071                                 }
1072
1073                                 /*
1074                                  * Add the curr bvec/segment to the curr
1075                                  * blkvsc_req
1076                                  */
1077                                 blkvsc_req->request.data_buffer.
1078                                         pfn_array[databuf_idx]
1079                                                 = page_to_pfn(bvec->bv_page);
1080                                 blkvsc_req->request.data_buffer.len
1081                                         += bvec->bv_len;
1082
1083                                 prev_bvec = bvec;
1084
1085                                 databuf_idx++;
1086                                 num_sectors += bvec->bv_len >> 9;
1087
1088                         } /* bio_for_each_segment */
1089
1090                 } /* rq_for_each_bio */
1091         }
1092
1093         /* Handle the last one */
1094         if (blkvsc_req) {
1095                 DPRINT_DBG(BLKVSC_DRV, "blkdev %p req %p group %p count %d\n",
1096                            blkdev, req, blkvsc_req->group,
1097                            blkvsc_req->group->outstanding);
1098
1099                 blkvsc_req->sector_start = start_sector;
1100                 sector_div(blkvsc_req->sector_start,
1101                            (blkdev->sector_size >> 9));
1102
1103                 blkvsc_req->sector_count = num_sectors /
1104                                            (blkdev->sector_size >> 9);
1105
1106                 blkvsc_init_rw(blkvsc_req);
1107         }
1108
1109         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
1110                 if (pending) {
1111                         DPRINT_DBG(BLKVSC_DRV, "adding blkvsc_req to "
1112                                    "pending_list - blkvsc_req %p start_sect %lu"
1113                                    " sect_count %ld (%lu %ld)\n", blkvsc_req,
1114                                    (unsigned long)blkvsc_req->sector_start,
1115                                    blkvsc_req->sector_count,
1116                                    (unsigned long)start_sector,
1117                                    (unsigned long)num_sectors);
1118
1119                         list_add_tail(&blkvsc_req->pend_entry,
1120                                       &blkdev->pending_list);
1121                 } else {
1122                         ret = blkvsc_submit_request(blkvsc_req,
1123                                                     blkvsc_request_completion);
1124                         if (ret == -1) {
1125                                 pending = 1;
1126                                 list_add_tail(&blkvsc_req->pend_entry,
1127                                               &blkdev->pending_list);
1128                         }
1129
1130                         DPRINT_DBG(BLKVSC_DRV, "submitted blkvsc_req %p "
1131                                    "start_sect %lu sect_count %ld (%lu %ld) "
1132                                    "ret %d\n", blkvsc_req,
1133                                    (unsigned long)blkvsc_req->sector_start,
1134                                    blkvsc_req->sector_count,
1135                                    (unsigned long)start_sector,
1136                                    num_sectors, ret);
1137                 }
1138         }
1139
1140         return pending;
1141 }
1142
1143 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
1144 {
1145         struct blkvsc_request *pend_req, *tmp;
1146         int ret = 0;
1147
1148         /* Flush the pending list first */
1149         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
1150                                  pend_entry) {
1151                 DPRINT_DBG(BLKVSC_DRV, "working off pending_list - %p\n",
1152                            pend_req);
1153
1154                 ret = blkvsc_submit_request(pend_req,
1155                                             blkvsc_request_completion);
1156                 if (ret != 0)
1157                         break;
1158                 else
1159                         list_del(&pend_req->pend_entry);
1160         }
1161
1162         return ret;
1163 }
1164
1165
1166 static void blkvsc_request(struct request_queue *queue)
1167 {
1168         struct block_device_context *blkdev = NULL;
1169         struct request *req;
1170         int ret = 0;
1171
1172         DPRINT_DBG(BLKVSC_DRV, "- enter\n");
1173         while ((req = blk_peek_request(queue)) != NULL) {
1174                 DPRINT_DBG(BLKVSC_DRV, "- req %p\n", req);
1175
1176                 blkdev = req->rq_disk->private_data;
1177                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS ||
1178                     blkdev->media_not_present) {
1179                         __blk_end_request_cur(req, 0);
1180                         continue;
1181                 }
1182
1183                 ret = blkvsc_do_pending_reqs(blkdev);
1184
1185                 if (ret != 0) {
1186                         DPRINT_DBG(BLKVSC_DRV,
1187                                    "- stop queue - pending_list not empty\n");
1188                         blk_stop_queue(queue);
1189                         break;
1190                 }
1191
1192                 blk_start_request(req);
1193
1194                 ret = blkvsc_do_request(blkdev, req);
1195                 if (ret > 0) {
1196                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no room\n");
1197                         blk_stop_queue(queue);
1198                         break;
1199                 } else if (ret < 0) {
1200                         DPRINT_DBG(BLKVSC_DRV, "- stop queue - no mem\n");
1201                         blk_requeue_request(queue, req);
1202                         blk_stop_queue(queue);
1203                         break;
1204                 }
1205         }
1206 }
1207
1208 /* Static decl */
1209 static int blkvsc_probe(struct device *dev);
1210
1211 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
1212 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
1213 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
1214
1215 /* The one and only one */
1216 static  struct storvsc_driver_object g_blkvsc_drv;
1217
1218 static const struct block_device_operations block_ops = {
1219         .owner = THIS_MODULE,
1220         .open = blkvsc_open,
1221         .release = blkvsc_release,
1222         .revalidate_disk = blkvsc_revalidate_disk,
1223         .getgeo = blkvsc_getgeo,
1224         .ioctl  = blkvsc_ioctl,
1225 };
1226
1227 /*
1228  * blkvsc_drv_init -  BlkVsc driver initialization.
1229  */
1230 static int blkvsc_drv_init(void)
1231 {
1232         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
1233         struct hv_driver *drv = &g_blkvsc_drv.base;
1234         int ret;
1235
1236         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
1237
1238         drv->priv = storvsc_drv_obj;
1239
1240         /* Callback to client driver to complete the initialization */
1241         blk_vsc_initialize(&storvsc_drv_obj->base);
1242
1243         drv->driver.name = storvsc_drv_obj->base.name;
1244
1245         drv->driver.probe = blkvsc_probe;
1246         drv->driver.remove = blkvsc_remove;
1247         drv->driver.shutdown = blkvsc_shutdown;
1248
1249         /* The driver belongs to vmbus */
1250         ret = vmbus_child_driver_register(&drv->driver);
1251
1252         return ret;
1253 }
1254
1255 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
1256 {
1257         struct device **curr = (struct device **)data;
1258         *curr = dev;
1259         return 1; /* stop iterating */
1260 }
1261
1262 static void blkvsc_drv_exit(void)
1263 {
1264         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
1265         struct hv_driver *drv = &g_blkvsc_drv.base;
1266         struct device *current_dev;
1267         int ret;
1268
1269         while (1) {
1270                 current_dev = NULL;
1271
1272                 /* Get the device */
1273                 ret = driver_for_each_device(&drv->driver, NULL,
1274                                              (void *) &current_dev,
1275                                              blkvsc_drv_exit_cb);
1276
1277                 if (ret)
1278                         DPRINT_WARN(BLKVSC_DRV,
1279                                     "driver_for_each_device returned %d", ret);
1280
1281
1282                 if (current_dev == NULL)
1283                         break;
1284
1285                 /* Initiate removal from the top-down */
1286                 device_unregister(current_dev);
1287         }
1288
1289         if (storvsc_drv_obj->base.cleanup)
1290                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
1291
1292         vmbus_child_driver_unregister(&drv->driver);
1293
1294         return;
1295 }
1296
1297 /*
1298  * blkvsc_probe - Add a new device for this driver
1299  */
1300 static int blkvsc_probe(struct device *device)
1301 {
1302         struct hv_driver *drv =
1303                                 drv_to_hv_drv(device->driver);
1304         struct storvsc_driver_object *storvsc_drv_obj =
1305                                 drv->priv;
1306         struct hv_device *device_obj = device_to_hv_device(device);
1307
1308         struct block_device_context *blkdev = NULL;
1309         struct storvsc_device_info device_info;
1310         int major = 0;
1311         int devnum = 0;
1312         int ret = 0;
1313         static int ide0_registered;
1314         static int ide1_registered;
1315
1316         DPRINT_DBG(BLKVSC_DRV, "blkvsc_probe - enter");
1317
1318         if (!storvsc_drv_obj->base.dev_add) {
1319                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
1320                 ret = -1;
1321                 goto Cleanup;
1322         }
1323
1324         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
1325         if (!blkdev) {
1326                 ret = -ENOMEM;
1327                 goto Cleanup;
1328         }
1329
1330         INIT_LIST_HEAD(&blkdev->pending_list);
1331
1332         /* Initialize what we can here */
1333         spin_lock_init(&blkdev->lock);
1334
1335         /* ASSERT(sizeof(struct blkvsc_request_group) <= */
1336         /*      sizeof(struct blkvsc_request)); */
1337
1338         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
1339                                         sizeof(struct blkvsc_request), 0,
1340                                         SLAB_HWCACHE_ALIGN, NULL);
1341         if (!blkdev->request_pool) {
1342                 ret = -ENOMEM;
1343                 goto Cleanup;
1344         }
1345
1346
1347         /* Call to the vsc driver to add the device */
1348         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
1349         if (ret != 0) {
1350                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
1351                 goto Cleanup;
1352         }
1353
1354         blkdev->device_ctx = device_obj;
1355         /* this identified the device 0 or 1 */
1356         blkdev->target = device_info.target_id;
1357         /* this identified the ide ctrl 0 or 1 */
1358         blkdev->path = device_info.path_id;
1359
1360         dev_set_drvdata(device, blkdev);
1361
1362         /* Calculate the major and device num */
1363         if (blkdev->path == 0) {
1364                 major = IDE0_MAJOR;
1365                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
1366
1367                 if (!ide0_registered) {
1368                         ret = register_blkdev(major, "ide");
1369                         if (ret != 0) {
1370                                 DPRINT_ERR(BLKVSC_DRV,
1371                                            "register_blkdev() failed! ret %d",
1372                                            ret);
1373                                 goto Remove;
1374                         }
1375
1376                         ide0_registered = 1;
1377                 }
1378         } else if (blkdev->path == 1) {
1379                 major = IDE1_MAJOR;
1380                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
1381
1382                 if (!ide1_registered) {
1383                         ret = register_blkdev(major, "ide");
1384                         if (ret != 0) {
1385                                 DPRINT_ERR(BLKVSC_DRV,
1386                                            "register_blkdev() failed! ret %d",
1387                                            ret);
1388                                 goto Remove;
1389                         }
1390
1391                         ide1_registered = 1;
1392                 }
1393         } else {
1394                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
1395                 ret = -1;
1396                 goto Cleanup;
1397         }
1398
1399         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
1400
1401         blkdev->gd = alloc_disk(BLKVSC_MINORS);
1402         if (!blkdev->gd) {
1403                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
1404                 ret = -1;
1405                 goto Cleanup;
1406         }
1407
1408         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
1409
1410         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
1411         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
1412         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
1413         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
1414         blk_queue_dma_alignment(blkdev->gd->queue, 511);
1415
1416         blkdev->gd->major = major;
1417         if (devnum == 1 || devnum == 3)
1418                 blkdev->gd->first_minor = BLKVSC_MINORS;
1419         else
1420                 blkdev->gd->first_minor = 0;
1421         blkdev->gd->fops = &block_ops;
1422         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
1423         blkdev->gd->private_data = blkdev;
1424         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
1425         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
1426
1427         blkvsc_do_inquiry(blkdev);
1428         if (blkdev->device_type == DVD_TYPE) {
1429                 set_disk_ro(blkdev->gd, 1);
1430                 blkdev->gd->flags |= GENHD_FL_REMOVABLE;
1431                 blkvsc_do_read_capacity(blkdev);
1432         } else {
1433                 blkvsc_do_read_capacity16(blkdev);
1434         }
1435
1436         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
1437         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
1438         /* go! */
1439         add_disk(blkdev->gd);
1440
1441         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
1442                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
1443                     blkdev->sector_size);
1444
1445         return ret;
1446
1447 Remove:
1448         storvsc_drv_obj->base.dev_rm(device_obj);
1449
1450 Cleanup:
1451         if (blkdev) {
1452                 if (blkdev->request_pool) {
1453                         kmem_cache_destroy(blkdev->request_pool);
1454                         blkdev->request_pool = NULL;
1455                 }
1456                 kfree(blkdev);
1457                 blkdev = NULL;
1458         }
1459
1460         return ret;
1461 }
1462
1463 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1464 {
1465         struct blkvsc_request *blkvsc_req =
1466                         (struct blkvsc_request *)request->context;
1467         struct block_device_context *blkdev =
1468                         (struct block_device_context *)blkvsc_req->dev;
1469         unsigned long flags;
1470         struct blkvsc_request *comp_req, *tmp;
1471         struct vmscsi_request *vm_srb;
1472
1473         /* ASSERT(blkvsc_req->group); */
1474
1475         DPRINT_DBG(BLKVSC_DRV, "blkdev %p blkvsc_req %p group %p type %s "
1476                    "sect_start %lu sect_count %ld len %d group outstd %d "
1477                    "total outstd %d\n",
1478                    blkdev, blkvsc_req, blkvsc_req->group,
1479                    (blkvsc_req->write) ? "WRITE" : "READ",
1480                    (unsigned long)blkvsc_req->sector_start,
1481                    blkvsc_req->sector_count,
1482                    blkvsc_req->request.data_buffer.len,
1483                    blkvsc_req->group->outstanding,
1484                    blkdev->num_outstanding_reqs);
1485
1486         spin_lock_irqsave(&blkdev->lock, flags);
1487
1488         blkdev->num_outstanding_reqs--;
1489         blkvsc_req->group->outstanding--;
1490
1491         /*
1492          * Only start processing when all the blkvsc_reqs are
1493          * completed. This guarantees no out-of-order blkvsc_req
1494          * completion when calling end_that_request_first()
1495          */
1496         if (blkvsc_req->group->outstanding == 0) {
1497                 list_for_each_entry_safe(comp_req, tmp,
1498                                          &blkvsc_req->group->blkvsc_req_list,
1499                                          req_entry) {
1500                         DPRINT_DBG(BLKVSC_DRV, "completing blkvsc_req %p "
1501                                    "sect_start %lu sect_count %ld\n",
1502                                    comp_req,
1503                                    (unsigned long)comp_req->sector_start,
1504                                    comp_req->sector_count);
1505
1506                         list_del(&comp_req->req_entry);
1507
1508                         vm_srb =
1509                         &comp_req->request.vstor_packet.vm_srb;
1510                         if (!__blk_end_request(comp_req->req,
1511                                 (!vm_srb->scsi_status ? 0 : -EIO),
1512                                 comp_req->sector_count * blkdev->sector_size)) {
1513                                 /*
1514                                  * All the sectors have been xferred ie the
1515                                  * request is done
1516                                  */
1517                                 DPRINT_DBG(BLKVSC_DRV, "req %p COMPLETED\n",
1518                                            comp_req->req);
1519                                 kmem_cache_free(blkdev->request_pool,
1520                                                 comp_req->group);
1521                         }
1522
1523                         kmem_cache_free(blkdev->request_pool, comp_req);
1524                 }
1525
1526                 if (!blkdev->shutting_down) {
1527                         blkvsc_do_pending_reqs(blkdev);
1528                         blk_start_queue(blkdev->gd->queue);
1529                         blkvsc_request(blkdev->gd->queue);
1530                 }
1531         }
1532
1533         spin_unlock_irqrestore(&blkdev->lock, flags);
1534 }
1535
1536 static int __init blkvsc_init(void)
1537 {
1538         int ret;
1539
1540         BUILD_BUG_ON(sizeof(sector_t) != 8);
1541
1542         DPRINT_INFO(BLKVSC_DRV, "Blkvsc initializing....");
1543
1544         ret = blkvsc_drv_init();
1545
1546         return ret;
1547 }
1548
1549 static void __exit blkvsc_exit(void)
1550 {
1551         blkvsc_drv_exit();
1552 }
1553
1554 MODULE_LICENSE("GPL");
1555 MODULE_VERSION(HV_DRV_VERSION);
1556 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1557 module_init(blkvsc_init);
1558 module_exit(blkvsc_exit);