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  *   K. Y. Srinivasan <kys@microsoft.com>
  21  */
  22 #include <linux/init.h>
  23 #include <linux/module.h>
  24 #include <linux/device.h>
  25 #include <linux/blkdev.h>
  26 #include <linux/major.h>
  27 #include <linux/delay.h>
  28 #include <linux/hdreg.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
  35 #include "hyperv.h"
  36 #include "hyperv_storage.h"
  37
  38
  39 #define BLKVSC_MINORS   64
  40
  41 enum blkvsc_device_type {
  42         UNKNOWN_DEV_TYPE,
  43         HARDDISK_TYPE,
  44         DVD_TYPE,
  45 };
  46
  47 enum blkvsc_op_type {
  48         DO_INQUIRY,
  49         DO_CAPACITY,
  50         DO_FLUSH,
  51 };
  52
  53 /*
  54  * This request ties the struct request and struct
  55  * blkvsc_request/hv_storvsc_request together A struct request may be
  56  * represented by 1 or more struct blkvsc_request
  57  */
  58 struct blkvsc_request_group {
  59         int outstanding;
  60         int status;
  61         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  62 };
  63
  64 struct blkvsc_request {
  65         /* blkvsc_request_group.blkvsc_req_list */
  66         struct list_head req_entry;
  67
  68         /* block_device_context.pending_list */
  69         struct list_head pend_entry;
  70
  71         /* This may be null if we generate a request internally */
  72         struct request *req;
  73
  74         struct block_device_context *dev;
  75
  76         /* The group this request is part of. Maybe null */
  77         struct blkvsc_request_group *group;
  78
  79         int write;
  80         sector_t sector_start;
  81         unsigned long sector_count;
  82
  83         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  84         unsigned char cmd_len;
  85         unsigned char cmnd[MAX_COMMAND_SIZE];
  86
  87         struct hv_storvsc_request request;
  88 };
  89
  90 /* Per device structure */
  91 struct block_device_context {
  92         /* point back to our device context */
  93         struct hv_device *device_ctx;
  94         struct kmem_cache *request_pool;
  95         spinlock_t lock;
  96         struct gendisk *gd;
  97         enum blkvsc_device_type device_type;
  98         struct list_head pending_list;
  99
 100         unsigned char device_id[64];
 101         unsigned int device_id_len;
 102         int num_outstanding_reqs;
 103         int shutting_down;
 104         unsigned int sector_size;
 105         sector_t capacity;
 106         unsigned int port;
 107         unsigned char path;
 108         unsigned char target;
 109         int users;
 110 };
 111
 112 static const char *drv_name = "blkvsc";
 113
 114 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 115 static const struct hv_guid dev_type = {
 116         .data = {
 117                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 118                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 119         }
 120 };
 121
 122 /*
 123  * There is a circular dependency involving blkvsc_request_completion()
 124  * and blkvsc_do_request().
 125  */
 126 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 127
 128 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 129
 130 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 131 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 132
 133 /*
 134  * There is a circular dependency involving blkvsc_probe()
 135  * and block_ops.
 136  */
 137 static int blkvsc_probe(struct hv_device *dev);
 138
 139 static int blkvsc_device_add(struct hv_device *device,
 140                                 void *additional_info)
 141 {
 142         struct storvsc_device_info *device_info;
 143         int ret = 0;
 144
 145         device_info = (struct storvsc_device_info *)additional_info;
 146
 147         device_info->ring_buffer_size = blkvsc_ringbuffer_size;
 148
 149         ret = storvsc_dev_add(device, additional_info);
 150         if (ret != 0)
 151                 return ret;
 152
 153         /*
 154          * We need to use the device instance guid to set the path and target
 155          * id. For IDE devices, the device instance id is formatted as
 156          * <bus id> * - <device id> - 8899 - 000000000000.
 157          */
 158         device_info->path_id = device->dev_instance.data[3] << 24 |
 159                              device->dev_instance.data[2] << 16 |
 160                              device->dev_instance.data[1] << 8  |
 161                              device->dev_instance.data[0];
 162
 163         device_info->target_id = device->dev_instance.data[5] << 8 |
 164                                device->dev_instance.data[4];
 165
 166         return ret;
 167 }
 168
 169 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 170                         void (*request_completion)(struct hv_storvsc_request *))
 171 {
 172         struct block_device_context *blkdev = blkvsc_req->dev;
 173         struct hv_storvsc_request *storvsc_req;
 174         struct vmscsi_request *vm_srb;
 175         int ret;
 176
 177
 178         storvsc_req = &blkvsc_req->request;
 179         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 180
 181         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 182
 183         storvsc_req->on_io_completion = request_completion;
 184         storvsc_req->context = blkvsc_req;
 185
 186         vm_srb->port_number = blkdev->port;
 187         vm_srb->path_id = blkdev->path;
 188         vm_srb->target_id = blkdev->target;
 189         vm_srb->lun = 0;         /* this is not really used at all */
 190
 191         vm_srb->cdb_length = blkvsc_req->cmd_len;
 192
 193         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 194
 195         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 196
 197         ret =  storvsc_do_io(blkdev->device_ctx,
 198                                            &blkvsc_req->request);
 199         if (ret == 0)
 200                 blkdev->num_outstanding_reqs++;
 201
 202         return ret;
 203 }
 204
 205
 206 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
 207 {
 208         struct block_device_context *blkdev = bdev->bd_disk->private_data;
 209         unsigned long flags;
 210
 211         spin_lock_irqsave(&blkdev->lock, flags);
 212
 213         blkdev->users++;
 214
 215         spin_unlock_irqrestore(&blkdev->lock, flags);
 216
 217         return 0;
 218 }
 219
 220
 221 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
 222 {
 223         sector_t nsect = get_capacity(bd->bd_disk);
 224         sector_t cylinders = nsect;
 225
 226         /*
 227          * We are making up these values; let us keep it simple.
 228          */
 229         hg->heads = 0xff;
 230         hg->sectors = 0x3f;
 231         sector_div(cylinders, hg->heads * hg->sectors);
 232         hg->cylinders = cylinders;
 233         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 234                 hg->cylinders = 0xffff;
 235         return 0;
 236
 237 }
 238
 239
 240 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 241 {
 242
 243         blkvsc_req->cmd_len = 16;
 244
 245         if (rq_data_dir(blkvsc_req->req)) {
 246                 blkvsc_req->write = 1;
 247                 blkvsc_req->cmnd[0] = WRITE_16;
 248         } else {
 249                 blkvsc_req->write = 0;
 250                 blkvsc_req->cmnd[0] = READ_16;
 251         }
 252
 253         blkvsc_req->cmnd[1] |=
 254         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 255
 256         *(unsigned long long *)&blkvsc_req->cmnd[2] =
 257         cpu_to_be64(blkvsc_req->sector_start);
 258         *(unsigned int *)&blkvsc_req->cmnd[10] =
 259         cpu_to_be32(blkvsc_req->sector_count);
 260 }
 261
 262
 263 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 264                         unsigned cmd, unsigned long arg)
 265 {
 266         struct block_device_context *blkdev = bd->bd_disk->private_data;
 267         int ret = 0;
 268
 269         switch (cmd) {
 270         case HDIO_GET_IDENTITY:
 271                 if (copy_to_user((void __user *)arg, blkdev->device_id,
 272                                  blkdev->device_id_len))
 273                         ret = -EFAULT;
 274                 break;
 275         default:
 276                 ret = -EINVAL;
 277                 break;
 278         }
 279
 280         return ret;
 281 }
 282
 283 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
 284 {
 285         struct blkvsc_request *blkvsc_req =
 286                         (struct blkvsc_request *)request->context;
 287         struct block_device_context *blkdev =
 288                         (struct block_device_context *)blkvsc_req->dev;
 289         struct scsi_sense_hdr sense_hdr;
 290         struct vmscsi_request *vm_srb;
 291         unsigned long flags;
 292
 293
 294         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 295
 296         spin_lock_irqsave(&blkdev->lock, flags);
 297         blkdev->num_outstanding_reqs--;
 298         spin_unlock_irqrestore(&blkdev->lock, flags);
 299
 300         if (vm_srb->scsi_status)
 301                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
 302                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 303                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
 304
 305         complete(&blkvsc_req->request.wait_event);
 306 }
 307
 308
 309 static int blkvsc_do_operation(struct block_device_context *blkdev,
 310                                 enum blkvsc_op_type op)
 311 {
 312         struct blkvsc_request *blkvsc_req;
 313         struct page *page_buf;
 314         unsigned char *buf;
 315         unsigned char device_type;
 316         struct scsi_sense_hdr sense_hdr;
 317         struct vmscsi_request *vm_srb;
 318         unsigned long flags;
 319
 320         int ret = 0;
 321
 322         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 323         if (!blkvsc_req)
 324                 return -ENOMEM;
 325
 326         page_buf = alloc_page(GFP_KERNEL);
 327         if (!page_buf) {
 328                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 329                 return -ENOMEM;
 330         }
 331
 332         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 333         init_completion(&blkvsc_req->request.wait_event);
 334         blkvsc_req->dev = blkdev;
 335         blkvsc_req->req = NULL;
 336         blkvsc_req->write = 0;
 337
 338         blkvsc_req->request.data_buffer.pfn_array[0] =
 339         page_to_pfn(page_buf);
 340         blkvsc_req->request.data_buffer.offset = 0;
 341
 342         switch (op) {
 343         case DO_INQUIRY:
 344                 blkvsc_req->cmnd[0] = INQUIRY;
 345                 blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 346                 blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 347                 blkvsc_req->cmnd[4] = 64;
 348                 blkvsc_req->cmd_len = 6;
 349                 blkvsc_req->request.data_buffer.len = 64;
 350                 break;
 351
 352         case DO_CAPACITY:
 353                 blkdev->sector_size = 0;
 354                 blkdev->capacity = 0;
 355
 356                 blkvsc_req->cmnd[0] = READ_CAPACITY;
 357                 blkvsc_req->cmd_len = 16;
 358                 blkvsc_req->request.data_buffer.len = 8;
 359                 break;
 360
 361         case DO_FLUSH:
 362                 blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 363                 blkvsc_req->cmd_len = 10;
 364                 blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 365                 blkvsc_req->request.data_buffer.len = 0;
 366                 break;
 367         default:
 368                 ret = -EINVAL;
 369                 goto cleanup;
 370         }
 371
 372         spin_lock_irqsave(&blkdev->lock, flags);
 373         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 374         spin_unlock_irqrestore(&blkdev->lock, flags);
 375
 376         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 377
 378         /* check error */
 379         if (vm_srb->scsi_status) {
 380                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 381                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 382
 383                 return 0;
 384         }
 385
 386         buf = kmap(page_buf);
 387
 388         switch (op) {
 389         case DO_INQUIRY:
 390                 device_type = buf[0] & 0x1F;
 391
 392                 if (device_type == 0x0)
 393                         blkdev->device_type = HARDDISK_TYPE;
 394                  else
 395                         blkdev->device_type = UNKNOWN_DEV_TYPE;
 396
 397                 blkdev->device_id_len = buf[7];
 398                 if (blkdev->device_id_len > 64)
 399                         blkdev->device_id_len = 64;
 400
 401                 memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 402                 break;
 403
 404         case DO_CAPACITY:
 405                 /* be to le */
 406                 blkdev->capacity =
 407                 ((buf[0] << 24) | (buf[1] << 16) |
 408                 (buf[2] << 8) | buf[3]) + 1;
 409
 410                 blkdev->sector_size =
 411                 (buf[4] << 24) | (buf[5] << 16) |
 412                 (buf[6] << 8) | buf[7];
 413                 break;
 414         default:
 415                 break;
 416
 417         }
 418
 419 cleanup:
 420
 421         kunmap(page_buf);
 422
 423         __free_page(page_buf);
 424
 425         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 426
 427         return ret;
 428 }
 429
 430
 431 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
 432 {
 433         struct blkvsc_request *pend_req, *tmp;
 434         struct blkvsc_request *comp_req, *tmp2;
 435         struct vmscsi_request *vm_srb;
 436
 437         int ret = 0;
 438
 439
 440         /* Flush the pending list first */
 441         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 442                                  pend_entry) {
 443                 /*
 444                  * The pend_req could be part of a partially completed
 445                  * request. If so, complete those req first until we
 446                  * hit the pend_req
 447                  */
 448                 list_for_each_entry_safe(comp_req, tmp2,
 449                                          &pend_req->group->blkvsc_req_list,
 450                                          req_entry) {
 451
 452                         if (comp_req == pend_req)
 453                                 break;
 454
 455                         list_del(&comp_req->req_entry);
 456
 457                         if (comp_req->req) {
 458                                 vm_srb =
 459                                 &comp_req->request.vstor_packet.
 460                                 vm_srb;
 461                                 ret = __blk_end_request(comp_req->req,
 462                                         (!vm_srb->scsi_status ? 0 : -EIO),
 463                                         comp_req->sector_count *
 464                                         blkdev->sector_size);
 465
 466                                 /* FIXME: shouldn't this do more than return? */
 467                                 if (ret)
 468                                         goto out;
 469                         }
 470
 471                         kmem_cache_free(blkdev->request_pool, comp_req);
 472                 }
 473
 474                 list_del(&pend_req->pend_entry);
 475
 476                 list_del(&pend_req->req_entry);
 477
 478                 if (comp_req->req) {
 479                         if (!__blk_end_request(pend_req->req, -EIO,
 480                                                pend_req->sector_count *
 481                                                blkdev->sector_size)) {
 482                                 /*
 483                                  * All the sectors have been xferred ie the
 484                                  * request is done
 485                                  */
 486                                 kmem_cache_free(blkdev->request_pool,
 487                                                 pend_req->group);
 488                         }
 489                 }
 490
 491                 kmem_cache_free(blkdev->request_pool, pend_req);
 492         }
 493
 494 out:
 495         return ret;
 496 }
 497
 498
 499 /*
 500  * blkvsc_remove() - Callback when our device is removed
 501  */
 502 static int blkvsc_remove(struct hv_device *dev)
 503 {
 504         struct block_device_context *blkdev = dev_get_drvdata(&dev->device);
 505         unsigned long flags;
 506
 507
 508         /* Get to a known state */
 509         spin_lock_irqsave(&blkdev->lock, flags);
 510
 511         blkdev->shutting_down = 1;
 512
 513         blk_stop_queue(blkdev->gd->queue);
 514
 515         blkvsc_cancel_pending_reqs(blkdev);
 516
 517         spin_unlock_irqrestore(&blkdev->lock, flags);
 518
 519         blkvsc_do_operation(blkdev, DO_FLUSH);
 520
 521         if (blkdev->users == 0) {
 522                 del_gendisk(blkdev->gd);
 523                 put_disk(blkdev->gd);
 524                 blk_cleanup_queue(blkdev->gd->queue);
 525
 526                 storvsc_dev_remove(blkdev->device_ctx);
 527
 528                 kmem_cache_destroy(blkdev->request_pool);
 529                 kfree(blkdev);
 530         }
 531
 532         return 0;
 533 }
 534
 535 static void blkvsc_shutdown(struct hv_device *dev)
 536 {
 537         struct block_device_context *blkdev = dev_get_drvdata(&dev->device);
 538         unsigned long flags;
 539
 540         if (!blkdev)
 541                 return;
 542
 543         spin_lock_irqsave(&blkdev->lock, flags);
 544
 545         blkdev->shutting_down = 1;
 546
 547         blk_stop_queue(blkdev->gd->queue);
 548
 549         blkvsc_cancel_pending_reqs(blkdev);
 550
 551         spin_unlock_irqrestore(&blkdev->lock, flags);
 552
 553         blkvsc_do_operation(blkdev, DO_FLUSH);
 554
 555         /*
 556          * Now wait for all outgoing I/O to be drained.
 557          */
 558         storvsc_wait_to_drain((struct storvsc_device *)dev->ext);
 559
 560 }
 561
 562 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
 563 {
 564         struct block_device_context *blkdev = disk->private_data;
 565         unsigned long flags;
 566
 567         spin_lock_irqsave(&blkdev->lock, flags);
 568
 569         if ((--blkdev->users == 0) && (blkdev->shutting_down)) {
 570                 blk_stop_queue(blkdev->gd->queue);
 571                 spin_unlock_irqrestore(&blkdev->lock, flags);
 572
 573                 blkvsc_do_operation(blkdev, DO_FLUSH);
 574                 del_gendisk(blkdev->gd);
 575                 put_disk(blkdev->gd);
 576                 blk_cleanup_queue(blkdev->gd->queue);
 577
 578                 storvsc_dev_remove(blkdev->device_ctx);
 579
 580                 kmem_cache_destroy(blkdev->request_pool);
 581                 kfree(blkdev);
 582         } else
 583                 spin_unlock_irqrestore(&blkdev->lock, flags);
 584
 585         return 0;
 586 }
 587
 588
 589 /*
 590  * We break the request into 1 or more blkvsc_requests and submit
 591  * them.  If we cant submit them all, we put them on the
 592  * pending_list. The blkvsc_request() will work on the pending_list.
 593  */
 594 static int blkvsc_do_request(struct block_device_context *blkdev,
 595                              struct request *req)
 596 {
 597         struct bio *bio = NULL;
 598         struct bio_vec *bvec = NULL;
 599         struct bio_vec *prev_bvec = NULL;
 600         struct blkvsc_request *blkvsc_req = NULL;
 601         struct blkvsc_request *tmp;
 602         int databuf_idx = 0;
 603         int seg_idx = 0;
 604         sector_t start_sector;
 605         unsigned long num_sectors = 0;
 606         int ret = 0;
 607         int pending = 0;
 608         struct blkvsc_request_group *group = NULL;
 609
 610         /* Create a group to tie req to list of blkvsc_reqs */
 611         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 612         if (!group)
 613                 return -ENOMEM;
 614
 615         INIT_LIST_HEAD(&group->blkvsc_req_list);
 616         group->outstanding = group->status = 0;
 617
 618         start_sector = blk_rq_pos(req);
 619
 620         /* foreach bio in the request */
 621         if (req->bio) {
 622                 for (bio = req->bio; bio; bio = bio->bi_next) {
 623                         /*
 624                          * Map this bio into an existing or new storvsc request
 625                          */
 626                         bio_for_each_segment(bvec, bio, seg_idx) {
 627                                 /* Get a new storvsc request */
 628                                 /* 1st-time */
 629                                 if ((!blkvsc_req) ||
 630                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 631                                     /* hole at the begin of page */
 632                                     || (bvec->bv_offset != 0) ||
 633                                     /* hold at the end of page */
 634                                     (prev_bvec &&
 635                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 636                                         /* submit the prev one */
 637                                         if (blkvsc_req) {
 638                                                 blkvsc_req->sector_start =
 639                                                 start_sector;
 640                                                 sector_div(
 641                                                 blkvsc_req->sector_start,
 642                                                 (blkdev->sector_size >> 9));
 643
 644                                                 blkvsc_req->sector_count =
 645                                                 num_sectors /
 646                                                 (blkdev->sector_size >> 9);
 647                                                 blkvsc_init_rw(blkvsc_req);
 648                                         }
 649
 650                                         /*
 651                                          * Create new blkvsc_req to represent
 652                                          * the current bvec
 653                                          */
 654                                         blkvsc_req =
 655                                         kmem_cache_zalloc(
 656                                         blkdev->request_pool, GFP_ATOMIC);
 657                                         if (!blkvsc_req) {
 658                                                 /* free up everything */
 659                                                 list_for_each_entry_safe(
 660                                                         blkvsc_req, tmp,
 661                                                         &group->blkvsc_req_list,
 662                                                         req_entry) {
 663                                                         list_del(
 664                                                         &blkvsc_req->req_entry);
 665                                                         kmem_cache_free(
 666                                                         blkdev->request_pool,
 667                                                         blkvsc_req);
 668                                                 }
 669
 670                                                 kmem_cache_free(
 671                                                 blkdev->request_pool, group);
 672                                                 return -ENOMEM;
 673                                         }
 674
 675                                         memset(blkvsc_req, 0,
 676                                                sizeof(struct blkvsc_request));
 677
 678                                         blkvsc_req->dev = blkdev;
 679                                         blkvsc_req->req = req;
 680                                         blkvsc_req->request.
 681                                         data_buffer.offset
 682                                         = bvec->bv_offset;
 683                                         blkvsc_req->request.
 684                                         data_buffer.len = 0;
 685
 686                                         /* Add to the group */
 687                                         blkvsc_req->group = group;
 688                                         blkvsc_req->group->outstanding++;
 689                                         list_add_tail(&blkvsc_req->req_entry,
 690                                         &blkvsc_req->group->blkvsc_req_list);
 691
 692                                         start_sector += num_sectors;
 693                                         num_sectors = 0;
 694                                         databuf_idx = 0;
 695                                 }
 696
 697                                 /*
 698                                  * Add the curr bvec/segment to the curr
 699                                  * blkvsc_req
 700                                  */
 701                                 blkvsc_req->request.data_buffer.
 702                                         pfn_array[databuf_idx]
 703                                                 = page_to_pfn(bvec->bv_page);
 704                                 blkvsc_req->request.data_buffer.len
 705                                         += bvec->bv_len;
 706
 707                                 prev_bvec = bvec;
 708
 709                                 databuf_idx++;
 710                                 num_sectors += bvec->bv_len >> 9;
 711
 712                         } /* bio_for_each_segment */
 713
 714                 } /* rq_for_each_bio */
 715         }
 716
 717         /* Handle the last one */
 718         if (blkvsc_req) {
 719                 blkvsc_req->sector_start = start_sector;
 720                 sector_div(blkvsc_req->sector_start,
 721                            (blkdev->sector_size >> 9));
 722
 723                 blkvsc_req->sector_count = num_sectors /
 724                                            (blkdev->sector_size >> 9);
 725
 726                 blkvsc_init_rw(blkvsc_req);
 727         }
 728
 729         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
 730                 if (pending) {
 731
 732                         list_add_tail(&blkvsc_req->pend_entry,
 733                                       &blkdev->pending_list);
 734                 } else {
 735                         ret = blkvsc_submit_request(blkvsc_req,
 736                                                     blkvsc_request_completion);
 737                         if (ret == -1) {
 738                                 pending = 1;
 739                                 list_add_tail(&blkvsc_req->pend_entry,
 740                                               &blkdev->pending_list);
 741                         }
 742
 743                 }
 744         }
 745
 746         return pending;
 747 }
 748
 749 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
 750 {
 751         struct blkvsc_request *pend_req, *tmp;
 752         int ret = 0;
 753
 754         /* Flush the pending list first */
 755         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 756                                  pend_entry) {
 757
 758                 ret = blkvsc_submit_request(pend_req,
 759                                             blkvsc_request_completion);
 760                 if (ret != 0)
 761                         break;
 762                 else
 763                         list_del(&pend_req->pend_entry);
 764         }
 765
 766         return ret;
 767 }
 768
 769
 770 static void blkvsc_request(struct request_queue *queue)
 771 {
 772         struct block_device_context *blkdev = NULL;
 773         struct request *req;
 774         int ret = 0;
 775
 776         while ((req = blk_peek_request(queue)) != NULL) {
 777
 778                 blkdev = req->rq_disk->private_data;
 779                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS) {
 780                         __blk_end_request_cur(req, 0);
 781                         continue;
 782                 }
 783
 784                 ret = blkvsc_do_pending_reqs(blkdev);
 785
 786                 if (ret != 0) {
 787                         blk_stop_queue(queue);
 788                         break;
 789                 }
 790
 791                 blk_start_request(req);
 792
 793                 ret = blkvsc_do_request(blkdev, req);
 794                 if (ret > 0) {
 795                         blk_stop_queue(queue);
 796                         break;
 797                 } else if (ret < 0) {
 798                         blk_requeue_request(queue, req);
 799                         blk_stop_queue(queue);
 800                         break;
 801                 }
 802         }
 803 }
 804
 805
 806
 807 /* The one and only one */
 808 static  struct hv_driver blkvsc_drv = {
 809         .probe =  blkvsc_probe,
 810         .remove =  blkvsc_remove,
 811         .shutdown = blkvsc_shutdown,
 812 };
 813
 814 static const struct block_device_operations block_ops = {
 815         .owner = THIS_MODULE,
 816         .open = blkvsc_open,
 817         .release = blkvsc_release,
 818         .getgeo = blkvsc_getgeo,
 819         .ioctl  = blkvsc_ioctl,
 820 };
 821
 822 /*
 823  * blkvsc_drv_init -  BlkVsc driver initialization.
 824  */
 825 static int blkvsc_drv_init(void)
 826 {
 827         struct hv_driver *drv = &blkvsc_drv;
 828         int ret;
 829
 830         BUILD_BUG_ON(sizeof(sector_t) != 8);
 831
 832         memcpy(&drv->dev_type, &dev_type, sizeof(struct hv_guid));
 833         drv->driver.name = drv_name;
 834
 835         /* The driver belongs to vmbus */
 836         ret = vmbus_child_driver_register(&drv->driver);
 837
 838         return ret;
 839 }
 840
 841
 842 static void blkvsc_drv_exit(void)
 843 {
 844
 845         vmbus_child_driver_unregister(&blkvsc_drv.driver);
 846 }
 847
 848 /*
 849  * blkvsc_probe - Add a new device for this driver
 850  */
 851 static int blkvsc_probe(struct hv_device *dev)
 852 {
 853         struct block_device_context *blkdev = NULL;
 854         struct storvsc_device_info device_info;
 855         struct storvsc_major_info major_info;
 856         int ret = 0;
 857
 858         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 859         if (!blkdev) {
 860                 ret = -ENOMEM;
 861                 goto cleanup;
 862         }
 863
 864         INIT_LIST_HEAD(&blkdev->pending_list);
 865
 866         /* Initialize what we can here */
 867         spin_lock_init(&blkdev->lock);
 868
 869
 870         blkdev->request_pool = kmem_cache_create(dev_name(&dev->device),
 871                                         sizeof(struct blkvsc_request), 0,
 872                                         SLAB_HWCACHE_ALIGN, NULL);
 873         if (!blkdev->request_pool) {
 874                 ret = -ENOMEM;
 875                 goto cleanup;
 876         }
 877
 878
 879         ret = blkvsc_device_add(dev, &device_info);
 880         if (ret != 0)
 881                 goto cleanup;
 882
 883         blkdev->device_ctx = dev;
 884         /* this identified the device 0 or 1 */
 885         blkdev->target = device_info.target_id;
 886         /* this identified the ide ctrl 0 or 1 */
 887         blkdev->path = device_info.path_id;
 888
 889         dev_set_drvdata(&dev->device, blkdev);
 890
 891         ret = storvsc_get_major_info(&device_info, &major_info);
 892
 893         if (ret)
 894                 goto cleanup;
 895
 896         if (major_info.do_register) {
 897                 ret = register_blkdev(major_info.major, major_info.devname);
 898
 899                 if (ret != 0) {
 900                         DPRINT_ERR(BLKVSC_DRV,
 901                                    "register_blkdev() failed! ret %d", ret);
 902                         goto remove;
 903                 }
 904         }
 905
 906         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!",
 907                         major_info.major);
 908
 909         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 910         if (!blkdev->gd) {
 911                 ret = -1;
 912                 goto cleanup;
 913         }
 914
 915         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 916
 917         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 918         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 919         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 920         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 921         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 922
 923         blkdev->gd->major = major_info.major;
 924         if (major_info.index == 1 || major_info.index == 3)
 925                 blkdev->gd->first_minor = BLKVSC_MINORS;
 926         else
 927                 blkdev->gd->first_minor = 0;
 928         blkdev->gd->fops = &block_ops;
 929         blkdev->gd->private_data = blkdev;
 930         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 931         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + major_info.index);
 932
 933         blkvsc_do_operation(blkdev, DO_INQUIRY);
 934         blkvsc_do_operation(blkdev, DO_CAPACITY);
 935
 936         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 937         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 938         /* go! */
 939         add_disk(blkdev->gd);
 940
 941         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 942                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 943                     blkdev->sector_size);
 944
 945         return ret;
 946
 947 remove:
 948         storvsc_dev_remove(dev);
 949
 950 cleanup:
 951         if (blkdev) {
 952                 if (blkdev->request_pool) {
 953                         kmem_cache_destroy(blkdev->request_pool);
 954                         blkdev->request_pool = NULL;
 955                 }
 956                 kfree(blkdev);
 957                 blkdev = NULL;
 958         }
 959
 960         return ret;
 961 }
 962
 963 static void blkvsc_request_completion(struct hv_storvsc_request *request)
 964 {
 965         struct blkvsc_request *blkvsc_req =
 966                         (struct blkvsc_request *)request->context;
 967         struct block_device_context *blkdev =
 968                         (struct block_device_context *)blkvsc_req->dev;
 969         unsigned long flags;
 970         struct blkvsc_request *comp_req, *tmp;
 971         struct vmscsi_request *vm_srb;
 972
 973
 974         spin_lock_irqsave(&blkdev->lock, flags);
 975
 976         blkdev->num_outstanding_reqs--;
 977         blkvsc_req->group->outstanding--;
 978
 979         /*
 980          * Only start processing when all the blkvsc_reqs are
 981          * completed. This guarantees no out-of-order blkvsc_req
 982          * completion when calling end_that_request_first()
 983          */
 984         if (blkvsc_req->group->outstanding == 0) {
 985                 list_for_each_entry_safe(comp_req, tmp,
 986                                          &blkvsc_req->group->blkvsc_req_list,
 987                                          req_entry) {
 988
 989                         list_del(&comp_req->req_entry);
 990
 991                         vm_srb =
 992                         &comp_req->request.vstor_packet.vm_srb;
 993                         if (!__blk_end_request(comp_req->req,
 994                                 (!vm_srb->scsi_status ? 0 : -EIO),
 995                                 comp_req->sector_count * blkdev->sector_size)) {
 996                                 /*
 997                                  * All the sectors have been xferred ie the
 998                                  * request is done
 999                                  */
1000                                 kmem_cache_free(blkdev->request_pool,
1001                                                 comp_req->group);
1002                         }
1003
1004                         kmem_cache_free(blkdev->request_pool, comp_req);
1005                 }
1006
1007                 if (!blkdev->shutting_down) {
1008                         blkvsc_do_pending_reqs(blkdev);
1009                         blk_start_queue(blkdev->gd->queue);
1010                         blkvsc_request(blkdev->gd->queue);
1011                 }
1012         }
1013
1014         spin_unlock_irqrestore(&blkdev->lock, flags);
1015 }
1016
1017 static void __exit blkvsc_exit(void)
1018 {
1019         blkvsc_drv_exit();
1020 }
1021
1022 MODULE_LICENSE("GPL");
1023 MODULE_VERSION(HV_DRV_VERSION);
1024 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1025 module_init(blkvsc_drv_init);
1026 module_exit(blkvsc_exit);