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