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 #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 enum blkvsc_op_type {
  50         DO_INQUIRY,
  51         DO_CAPACITY,
  52         DO_FLUSH,
  53 };
  54
  55 /*
  56  * This request ties the struct request and struct
  57  * blkvsc_request/hv_storvsc_request together A struct request may be
  58  * represented by 1 or more struct blkvsc_request
  59  */
  60 struct blkvsc_request_group {
  61         int outstanding;
  62         int status;
  63         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  64 };
  65
  66 struct blkvsc_request {
  67         /* blkvsc_request_group.blkvsc_req_list */
  68         struct list_head req_entry;
  69
  70         /* block_device_context.pending_list */
  71         struct list_head pend_entry;
  72
  73         /* This may be null if we generate a request internally */
  74         struct request *req;
  75
  76         struct block_device_context *dev;
  77
  78         /* The group this request is part of. Maybe null */
  79         struct blkvsc_request_group *group;
  80
  81         int write;
  82         sector_t sector_start;
  83         unsigned long sector_count;
  84
  85         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  86         unsigned char cmd_len;
  87         unsigned char cmnd[MAX_COMMAND_SIZE];
  88
  89         struct hv_storvsc_request request;
  90 };
  91
  92 /* Per device structure */
  93 struct block_device_context {
  94         /* point back to our device context */
  95         struct hv_device *device_ctx;
  96         struct kmem_cache *request_pool;
  97         spinlock_t lock;
  98         struct gendisk *gd;
  99         enum blkvsc_device_type device_type;
 100         struct list_head pending_list;
 101
 102         unsigned char device_id[64];
 103         unsigned int device_id_len;
 104         int num_outstanding_reqs;
 105         int shutting_down;
 106         unsigned int sector_size;
 107         sector_t capacity;
 108         unsigned int port;
 109         unsigned char path;
 110         unsigned char target;
 111         int users;
 112 };
 113
 114 static const char *drv_name = "blkvsc";
 115
 116 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 117 static const struct hv_guid dev_type = {
 118         .data = {
 119                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 120                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 121         }
 122 };
 123
 124 /*
 125  * There is a circular dependency involving blkvsc_request_completion()
 126  * and blkvsc_do_request().
 127  */
 128 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 129
 130 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 131
 132 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 133 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 134
 135 /*
 136  * There is a circular dependency involving blkvsc_probe()
 137  * and block_ops.
 138  */
 139 static int blkvsc_probe(struct hv_device *dev);
 140
 141 static int blkvsc_device_add(struct hv_device *device,
 142                                 void *additional_info)
 143 {
 144         struct storvsc_device_info *device_info;
 145         int ret = 0;
 146
 147         device_info = (struct storvsc_device_info *)additional_info;
 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         blk_cleanup_queue(blkdev->gd->queue);
 522
 523         /*
 524          * Call to the vsc driver to let it know that the device is being
 525          * removed
 526          */
 527         storvsc_dev_remove(dev);
 528
 529         del_gendisk(blkdev->gd);
 530
 531         kmem_cache_destroy(blkdev->request_pool);
 532
 533         kfree(blkdev);
 534
 535         return 0;
 536
 537 }
 538
 539 static void blkvsc_shutdown(struct hv_device *dev)
 540 {
 541         struct block_device_context *blkdev = dev_get_drvdata(&dev->device);
 542         unsigned long flags;
 543
 544         if (!blkdev)
 545                 return;
 546
 547         spin_lock_irqsave(&blkdev->lock, flags);
 548
 549         blkdev->shutting_down = 1;
 550
 551         blk_stop_queue(blkdev->gd->queue);
 552
 553         blkvsc_cancel_pending_reqs(blkdev);
 554
 555         spin_unlock_irqrestore(&blkdev->lock, flags);
 556
 557         blkvsc_do_operation(blkdev, DO_FLUSH);
 558
 559         /*
 560          * Now wait for all outgoing I/O to be drained.
 561          */
 562         storvsc_wait_to_drain((struct storvsc_device *)dev->ext);
 563
 564 }
 565
 566 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
 567 {
 568         struct block_device_context *blkdev = disk->private_data;
 569         unsigned long flags;
 570
 571         if (blkdev->users == 1) {
 572                 blkvsc_do_operation(blkdev, DO_FLUSH);
 573         }
 574
 575         spin_lock_irqsave(&blkdev->lock, flags);
 576         blkdev->users--;
 577         spin_unlock_irqrestore(&blkdev->lock, flags);
 578
 579         return 0;
 580 }
 581
 582
 583 /*
 584  * We break the request into 1 or more blkvsc_requests and submit
 585  * them.  If we cant submit them all, we put them on the
 586  * pending_list. The blkvsc_request() will work on the pending_list.
 587  */
 588 static int blkvsc_do_request(struct block_device_context *blkdev,
 589                              struct request *req)
 590 {
 591         struct bio *bio = NULL;
 592         struct bio_vec *bvec = NULL;
 593         struct bio_vec *prev_bvec = NULL;
 594         struct blkvsc_request *blkvsc_req = NULL;
 595         struct blkvsc_request *tmp;
 596         int databuf_idx = 0;
 597         int seg_idx = 0;
 598         sector_t start_sector;
 599         unsigned long num_sectors = 0;
 600         int ret = 0;
 601         int pending = 0;
 602         struct blkvsc_request_group *group = NULL;
 603
 604         /* Create a group to tie req to list of blkvsc_reqs */
 605         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 606         if (!group)
 607                 return -ENOMEM;
 608
 609         INIT_LIST_HEAD(&group->blkvsc_req_list);
 610         group->outstanding = group->status = 0;
 611
 612         start_sector = blk_rq_pos(req);
 613
 614         /* foreach bio in the request */
 615         if (req->bio) {
 616                 for (bio = req->bio; bio; bio = bio->bi_next) {
 617                         /*
 618                          * Map this bio into an existing or new storvsc request
 619                          */
 620                         bio_for_each_segment(bvec, bio, seg_idx) {
 621                                 /* Get a new storvsc request */
 622                                 /* 1st-time */
 623                                 if ((!blkvsc_req) ||
 624                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 625                                     /* hole at the begin of page */
 626                                     || (bvec->bv_offset != 0) ||
 627                                     /* hold at the end of page */
 628                                     (prev_bvec &&
 629                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 630                                         /* submit the prev one */
 631                                         if (blkvsc_req) {
 632                                                 blkvsc_req->sector_start =
 633                                                 start_sector;
 634                                                 sector_div(
 635                                                 blkvsc_req->sector_start,
 636                                                 (blkdev->sector_size >> 9));
 637
 638                                                 blkvsc_req->sector_count =
 639                                                 num_sectors /
 640                                                 (blkdev->sector_size >> 9);
 641                                                 blkvsc_init_rw(blkvsc_req);
 642                                         }
 643
 644                                         /*
 645                                          * Create new blkvsc_req to represent
 646                                          * the current bvec
 647                                          */
 648                                         blkvsc_req =
 649                                         kmem_cache_zalloc(
 650                                         blkdev->request_pool, GFP_ATOMIC);
 651                                         if (!blkvsc_req) {
 652                                                 /* free up everything */
 653                                                 list_for_each_entry_safe(
 654                                                         blkvsc_req, tmp,
 655                                                         &group->blkvsc_req_list,
 656                                                         req_entry) {
 657                                                         list_del(
 658                                                         &blkvsc_req->req_entry);
 659                                                         kmem_cache_free(
 660                                                         blkdev->request_pool,
 661                                                         blkvsc_req);
 662                                                 }
 663
 664                                                 kmem_cache_free(
 665                                                 blkdev->request_pool, group);
 666                                                 return -ENOMEM;
 667                                         }
 668
 669                                         memset(blkvsc_req, 0,
 670                                                sizeof(struct blkvsc_request));
 671
 672                                         blkvsc_req->dev = blkdev;
 673                                         blkvsc_req->req = req;
 674                                         blkvsc_req->request.
 675                                         data_buffer.offset
 676                                         = bvec->bv_offset;
 677                                         blkvsc_req->request.
 678                                         data_buffer.len = 0;
 679
 680                                         /* Add to the group */
 681                                         blkvsc_req->group = group;
 682                                         blkvsc_req->group->outstanding++;
 683                                         list_add_tail(&blkvsc_req->req_entry,
 684                                         &blkvsc_req->group->blkvsc_req_list);
 685
 686                                         start_sector += num_sectors;
 687                                         num_sectors = 0;
 688                                         databuf_idx = 0;
 689                                 }
 690
 691                                 /*
 692                                  * Add the curr bvec/segment to the curr
 693                                  * blkvsc_req
 694                                  */
 695                                 blkvsc_req->request.data_buffer.
 696                                         pfn_array[databuf_idx]
 697                                                 = page_to_pfn(bvec->bv_page);
 698                                 blkvsc_req->request.data_buffer.len
 699                                         += bvec->bv_len;
 700
 701                                 prev_bvec = bvec;
 702
 703                                 databuf_idx++;
 704                                 num_sectors += bvec->bv_len >> 9;
 705
 706                         } /* bio_for_each_segment */
 707
 708                 } /* rq_for_each_bio */
 709         }
 710
 711         /* Handle the last one */
 712         if (blkvsc_req) {
 713                 blkvsc_req->sector_start = start_sector;
 714                 sector_div(blkvsc_req->sector_start,
 715                            (blkdev->sector_size >> 9));
 716
 717                 blkvsc_req->sector_count = num_sectors /
 718                                            (blkdev->sector_size >> 9);
 719
 720                 blkvsc_init_rw(blkvsc_req);
 721         }
 722
 723         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
 724                 if (pending) {
 725
 726                         list_add_tail(&blkvsc_req->pend_entry,
 727                                       &blkdev->pending_list);
 728                 } else {
 729                         ret = blkvsc_submit_request(blkvsc_req,
 730                                                     blkvsc_request_completion);
 731                         if (ret == -1) {
 732                                 pending = 1;
 733                                 list_add_tail(&blkvsc_req->pend_entry,
 734                                               &blkdev->pending_list);
 735                         }
 736
 737                 }
 738         }
 739
 740         return pending;
 741 }
 742
 743 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
 744 {
 745         struct blkvsc_request *pend_req, *tmp;
 746         int ret = 0;
 747
 748         /* Flush the pending list first */
 749         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 750                                  pend_entry) {
 751
 752                 ret = blkvsc_submit_request(pend_req,
 753                                             blkvsc_request_completion);
 754                 if (ret != 0)
 755                         break;
 756                 else
 757                         list_del(&pend_req->pend_entry);
 758         }
 759
 760         return ret;
 761 }
 762
 763
 764 static void blkvsc_request(struct request_queue *queue)
 765 {
 766         struct block_device_context *blkdev = NULL;
 767         struct request *req;
 768         int ret = 0;
 769
 770         while ((req = blk_peek_request(queue)) != NULL) {
 771
 772                 blkdev = req->rq_disk->private_data;
 773                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS) {
 774                         __blk_end_request_cur(req, 0);
 775                         continue;
 776                 }
 777
 778                 ret = blkvsc_do_pending_reqs(blkdev);
 779
 780                 if (ret != 0) {
 781                         blk_stop_queue(queue);
 782                         break;
 783                 }
 784
 785                 blk_start_request(req);
 786
 787                 ret = blkvsc_do_request(blkdev, req);
 788                 if (ret > 0) {
 789                         blk_stop_queue(queue);
 790                         break;
 791                 } else if (ret < 0) {
 792                         blk_requeue_request(queue, req);
 793                         blk_stop_queue(queue);
 794                         break;
 795                 }
 796         }
 797 }
 798
 799
 800
 801 /* The one and only one */
 802 static  struct storvsc_driver blkvsc_drv = {
 803         .base.probe =  blkvsc_probe,
 804         .base.remove =  blkvsc_remove,
 805         .base.shutdown = blkvsc_shutdown,
 806 };
 807
 808 static const struct block_device_operations block_ops = {
 809         .owner = THIS_MODULE,
 810         .open = blkvsc_open,
 811         .release = blkvsc_release,
 812         .getgeo = blkvsc_getgeo,
 813         .ioctl  = blkvsc_ioctl,
 814 };
 815
 816 /*
 817  * blkvsc_drv_init -  BlkVsc driver initialization.
 818  */
 819 static int blkvsc_drv_init(void)
 820 {
 821         struct storvsc_driver *storvsc_drv = &blkvsc_drv;
 822         struct hv_driver *drv = &blkvsc_drv.base;
 823         int ret;
 824
 825         BUILD_BUG_ON(sizeof(sector_t) != 8);
 826
 827         storvsc_drv->ring_buffer_size = blkvsc_ringbuffer_size;
 828
 829         memcpy(&drv->dev_type, &dev_type, sizeof(struct hv_guid));
 830         drv->name = drv_name;
 831         drv->driver.name = drv_name;
 832
 833         /* The driver belongs to vmbus */
 834         ret = vmbus_child_driver_register(&drv->driver);
 835
 836         return ret;
 837 }
 838
 839 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 840 {
 841         struct device **curr = (struct device **)data;
 842         *curr = dev;
 843         return 1; /* stop iterating */
 844 }
 845
 846 static void blkvsc_drv_exit(void)
 847 {
 848         struct hv_driver *drv = &blkvsc_drv.base;
 849         struct device *current_dev;
 850         int ret;
 851
 852         while (1) {
 853                 current_dev = NULL;
 854
 855                 /* Get the device */
 856                 ret = driver_for_each_device(&drv->driver, NULL,
 857                                              (void *) &current_dev,
 858                                              blkvsc_drv_exit_cb);
 859
 860                 if (ret)
 861                         DPRINT_WARN(BLKVSC_DRV,
 862                                     "driver_for_each_device returned %d", ret);
 863
 864
 865                 if (current_dev == NULL)
 866                         break;
 867
 868                 /* Initiate removal from the top-down */
 869                 device_unregister(current_dev);
 870         }
 871
 872         vmbus_child_driver_unregister(&drv->driver);
 873
 874         return;
 875 }
 876
 877 /*
 878  * blkvsc_probe - Add a new device for this driver
 879  */
 880 static int blkvsc_probe(struct hv_device *dev)
 881 {
 882         struct block_device_context *blkdev = NULL;
 883         struct storvsc_device_info device_info;
 884         struct storvsc_major_info major_info;
 885         int ret = 0;
 886
 887         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 888         if (!blkdev) {
 889                 ret = -ENOMEM;
 890                 goto cleanup;
 891         }
 892
 893         INIT_LIST_HEAD(&blkdev->pending_list);
 894
 895         /* Initialize what we can here */
 896         spin_lock_init(&blkdev->lock);
 897
 898
 899         blkdev->request_pool = kmem_cache_create(dev_name(&dev->device),
 900                                         sizeof(struct blkvsc_request), 0,
 901                                         SLAB_HWCACHE_ALIGN, NULL);
 902         if (!blkdev->request_pool) {
 903                 ret = -ENOMEM;
 904                 goto cleanup;
 905         }
 906
 907
 908         ret = blkvsc_device_add(dev, &device_info);
 909         if (ret != 0)
 910                 goto cleanup;
 911
 912         blkdev->device_ctx = dev;
 913         /* this identified the device 0 or 1 */
 914         blkdev->target = device_info.target_id;
 915         /* this identified the ide ctrl 0 or 1 */
 916         blkdev->path = device_info.path_id;
 917
 918         dev_set_drvdata(&dev->device, blkdev);
 919
 920         ret = storvsc_get_major_info(&device_info, &major_info);
 921
 922         if (ret)
 923                 goto cleanup;
 924
 925         if (major_info.do_register) {
 926                 ret = register_blkdev(major_info.major, major_info.devname);
 927
 928                 if (ret != 0) {
 929                         DPRINT_ERR(BLKVSC_DRV,
 930                                    "register_blkdev() failed! ret %d", ret);
 931                         goto remove;
 932                 }
 933         }
 934
 935         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!",
 936                         major_info.major);
 937
 938         blkdev->gd = alloc_disk(BLKVSC_MINORS);
 939         if (!blkdev->gd) {
 940                 ret = -1;
 941                 goto cleanup;
 942         }
 943
 944         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
 945
 946         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
 947         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
 948         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
 949         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
 950         blk_queue_dma_alignment(blkdev->gd->queue, 511);
 951
 952         blkdev->gd->major = major_info.major;
 953         if (major_info.index == 1 || major_info.index == 3)
 954                 blkdev->gd->first_minor = BLKVSC_MINORS;
 955         else
 956                 blkdev->gd->first_minor = 0;
 957         blkdev->gd->fops = &block_ops;
 958         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
 959         blkdev->gd->private_data = blkdev;
 960         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
 961         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + major_info.index);
 962
 963         blkvsc_do_operation(blkdev, DO_INQUIRY);
 964         blkvsc_do_operation(blkdev, DO_CAPACITY);
 965
 966         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
 967         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
 968         /* go! */
 969         add_disk(blkdev->gd);
 970
 971         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
 972                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
 973                     blkdev->sector_size);
 974
 975         return ret;
 976
 977 remove:
 978         storvsc_dev_remove(dev);
 979
 980 cleanup:
 981         if (blkdev) {
 982                 if (blkdev->request_pool) {
 983                         kmem_cache_destroy(blkdev->request_pool);
 984                         blkdev->request_pool = NULL;
 985                 }
 986                 kfree(blkdev);
 987                 blkdev = NULL;
 988         }
 989
 990         return ret;
 991 }
 992
 993 static void blkvsc_request_completion(struct hv_storvsc_request *request)
 994 {
 995         struct blkvsc_request *blkvsc_req =
 996                         (struct blkvsc_request *)request->context;
 997         struct block_device_context *blkdev =
 998                         (struct block_device_context *)blkvsc_req->dev;
 999         unsigned long flags;
1000         struct blkvsc_request *comp_req, *tmp;
1001         struct vmscsi_request *vm_srb;
1002
1003
1004         spin_lock_irqsave(&blkdev->lock, flags);
1005
1006         blkdev->num_outstanding_reqs--;
1007         blkvsc_req->group->outstanding--;
1008
1009         /*
1010          * Only start processing when all the blkvsc_reqs are
1011          * completed. This guarantees no out-of-order blkvsc_req
1012          * completion when calling end_that_request_first()
1013          */
1014         if (blkvsc_req->group->outstanding == 0) {
1015                 list_for_each_entry_safe(comp_req, tmp,
1016                                          &blkvsc_req->group->blkvsc_req_list,
1017                                          req_entry) {
1018
1019                         list_del(&comp_req->req_entry);
1020
1021                         vm_srb =
1022                         &comp_req->request.vstor_packet.vm_srb;
1023                         if (!__blk_end_request(comp_req->req,
1024                                 (!vm_srb->scsi_status ? 0 : -EIO),
1025                                 comp_req->sector_count * blkdev->sector_size)) {
1026                                 /*
1027                                  * All the sectors have been xferred ie the
1028                                  * request is done
1029                                  */
1030                                 kmem_cache_free(blkdev->request_pool,
1031                                                 comp_req->group);
1032                         }
1033
1034                         kmem_cache_free(blkdev->request_pool, comp_req);
1035                 }
1036
1037                 if (!blkdev->shutting_down) {
1038                         blkvsc_do_pending_reqs(blkdev);
1039                         blk_start_queue(blkdev->gd->queue);
1040                         blkvsc_request(blkdev->gd->queue);
1041                 }
1042         }
1043
1044         spin_unlock_irqrestore(&blkdev->lock, flags);
1045 }
1046
1047 static int __init blkvsc_init(void)
1048 {
1049         int ret;
1050
1051         ret = blkvsc_drv_init();
1052
1053         return ret;
1054 }
1055
1056 static void __exit blkvsc_exit(void)
1057 {
1058         blkvsc_drv_exit();
1059 }
1060
1061 MODULE_LICENSE("GPL");
1062 MODULE_VERSION(HV_DRV_VERSION);
1063 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1064 module_init(blkvsc_init);
1065 module_exit(blkvsc_exit);