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

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  *
  21  * 4/3/2011: K. Y. Srinivasan - Significant restructuring and cleanup.
  22  */
  23 #include <linux/init.h>
  24 #include <linux/module.h>
  25 #include <linux/device.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/major.h>
  28 #include <linux/delay.h>
  29 #include <linux/hdreg.h>
  30 #include <linux/mutex.h>
  31 #include <linux/slab.h>
  32 #include <scsi/scsi.h>
  33 #include <scsi/scsi_cmnd.h>
  34 #include <scsi/scsi_eh.h>
  35 #include <scsi/scsi_dbg.h>
  36 #include "hv_api.h"
  37 #include "logging.h"
  38 #include "version_info.h"
  39 #include "vmbus.h"
  40 #include "storvsc_api.h"
  41
  42
  43 #define BLKVSC_MINORS   64
  44
  45 enum blkvsc_device_type {
  46         UNKNOWN_DEV_TYPE,
  47         HARDDISK_TYPE,
  48         DVD_TYPE,
  49 };
  50
  51 enum blkvsc_op_type {
  52         DO_INQUIRY,
  53         DO_CAPACITY,
  54         DO_FLUSH,
  55 };
  56
  57 /*
  58  * This request ties the struct request and struct
  59  * blkvsc_request/hv_storvsc_request together A struct request may be
  60  * represented by 1 or more struct blkvsc_request
  61  */
  62 struct blkvsc_request_group {
  63         int outstanding;
  64         int status;
  65         struct list_head blkvsc_req_list;       /* list of blkvsc_requests */
  66 };
  67
  68 struct blkvsc_request {
  69         /* blkvsc_request_group.blkvsc_req_list */
  70         struct list_head req_entry;
  71
  72         /* block_device_context.pending_list */
  73         struct list_head pend_entry;
  74
  75         /* This may be null if we generate a request internally */
  76         struct request *req;
  77
  78         struct block_device_context *dev;
  79
  80         /* The group this request is part of. Maybe null */
  81         struct blkvsc_request_group *group;
  82
  83         int write;
  84         sector_t sector_start;
  85         unsigned long sector_count;
  86
  87         unsigned char sense_buffer[SCSI_SENSE_BUFFERSIZE];
  88         unsigned char cmd_len;
  89         unsigned char cmnd[MAX_COMMAND_SIZE];
  90
  91         struct hv_storvsc_request request;
  92 };
  93
  94 /* Per device structure */
  95 struct block_device_context {
  96         /* point back to our device context */
  97         struct hv_device *device_ctx;
  98         struct kmem_cache *request_pool;
  99         spinlock_t lock;
 100         struct gendisk *gd;
 101         enum blkvsc_device_type device_type;
 102         struct list_head pending_list;
 103
 104         unsigned char device_id[64];
 105         unsigned int device_id_len;
 106         int num_outstanding_reqs;
 107         int shutting_down;
 108         unsigned int sector_size;
 109         sector_t capacity;
 110         unsigned int port;
 111         unsigned char path;
 112         unsigned char target;
 113         int users;
 114 };
 115
 116 static DEFINE_MUTEX(blkvsc_mutex);
 117
 118 static const char *g_blk_driver_name = "blkvsc";
 119
 120 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 121 static const struct hv_guid g_blk_device_type = {
 122         .data = {
 123                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 124                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 125         }
 126 };
 127
 128 /*
 129  * There is a circular dependency involving blkvsc_request_completion()
 130  * and blkvsc_do_request().
 131  */
 132 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 133
 134 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 135
 136 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 137 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 138
 139 /*
 140  * There is a circular dependency involving blkvsc_probe()
 141  * and block_ops.
 142  */
 143 static int blkvsc_probe(struct device *dev);
 144
 145 static int blk_vsc_on_device_add(struct hv_device *device,
 146                                 void *additional_info)
 147 {
 148         struct storvsc_device_info *device_info;
 149         int ret = 0;
 150
 151         device_info = (struct storvsc_device_info *)additional_info;
 152
 153         ret = stor_vsc_on_device_add(device, additional_info);
 154         if (ret != 0)
 155                 return ret;
 156
 157         /*
 158          * We need to use the device instance guid to set the path and target
 159          * id. For IDE devices, the device instance id is formatted as
 160          * <bus id> * - <device id> - 8899 - 000000000000.
 161          */
 162         device_info->path_id = device->dev_instance.data[3] << 24 |
 163                              device->dev_instance.data[2] << 16 |
 164                              device->dev_instance.data[1] << 8  |
 165                              device->dev_instance.data[0];
 166
 167         device_info->target_id = device->dev_instance.data[5] << 8 |
 168                                device->dev_instance.data[4];
 169
 170         return ret;
 171 }
 172
 173
 174 static int blk_vsc_initialize(struct hv_driver *driver)
 175 {
 176         struct storvsc_driver_object *stor_driver;
 177         int ret = 0;
 178
 179         stor_driver = hvdr_to_stordr(driver);
 180
 181         /* Make sure we are at least 2 pages since 1 page is used for control */
 182
 183         driver->name = g_blk_driver_name;
 184         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 185
 186
 187         /*
 188          * Divide the ring buffer data size (which is 1 page less than the ring
 189          * buffer size since that page is reserved for the ring buffer indices)
 190          * by the max request size (which is
 191          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 192          */
 193         stor_driver->max_outstanding_req_per_channel =
 194                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 195                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 196                            sizeof(struct vstor_packet) + sizeof(u64),
 197                            sizeof(u64)));
 198
 199         DPRINT_INFO(BLKVSC, "max io outstd %u",
 200                     stor_driver->max_outstanding_req_per_channel);
 201
 202         /* Setup the dispatch table */
 203         stor_driver->base.dev_add = blk_vsc_on_device_add;
 204         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 205         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 206         stor_driver->on_io_request = stor_vsc_on_io_request;
 207
 208         return ret;
 209 }
 210
 211
 212 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 213                         void (*request_completion)(struct hv_storvsc_request *))
 214 {
 215         struct block_device_context *blkdev = blkvsc_req->dev;
 216         struct hv_device *device_ctx = blkdev->device_ctx;
 217         struct hv_driver *drv =
 218                         drv_to_hv_drv(device_ctx->device.driver);
 219         struct storvsc_driver_object *storvsc_drv_obj =
 220                         drv->priv;
 221         struct hv_storvsc_request *storvsc_req;
 222         struct vmscsi_request *vm_srb;
 223         int ret;
 224
 225
 226         storvsc_req = &blkvsc_req->request;
 227         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 228
 229         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 230
 231         storvsc_req->on_io_completion = request_completion;
 232         storvsc_req->context = blkvsc_req;
 233
 234         vm_srb->port_number = blkdev->port;
 235         vm_srb->path_id = blkdev->path;
 236         vm_srb->target_id = blkdev->target;
 237         vm_srb->lun = 0;         /* this is not really used at all */
 238
 239         vm_srb->cdb_length = blkvsc_req->cmd_len;
 240
 241         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 242
 243         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 244
 245         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 246                                            &blkvsc_req->request);
 247         if (ret == 0)
 248                 blkdev->num_outstanding_reqs++;
 249
 250         return ret;
 251 }
 252
 253
 254 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
 255 {
 256         struct block_device_context *blkdev = bdev->bd_disk->private_data;
 257
 258         spin_lock(&blkdev->lock);
 259
 260         blkdev->users++;
 261
 262         spin_unlock(&blkdev->lock);
 263
 264         return 0;
 265 }
 266
 267
 268 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
 269 {
 270         sector_t nsect = get_capacity(bd->bd_disk);
 271         sector_t cylinders = nsect;
 272
 273         /*
 274          * We are making up these values; let us keep it simple.
 275          */
 276         hg->heads = 0xff;
 277         hg->sectors = 0x3f;
 278         sector_div(cylinders, hg->heads * hg->sectors);
 279         hg->cylinders = cylinders;
 280         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 281                 hg->cylinders = 0xffff;
 282         return 0;
 283
 284 }
 285
 286
 287 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 288 {
 289
 290         blkvsc_req->cmd_len = 16;
 291
 292         if (rq_data_dir(blkvsc_req->req)) {
 293                 blkvsc_req->write = 1;
 294                 blkvsc_req->cmnd[0] = WRITE_16;
 295         } else {
 296                 blkvsc_req->write = 0;
 297                 blkvsc_req->cmnd[0] = READ_16;
 298         }
 299
 300         blkvsc_req->cmnd[1] |=
 301         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 302
 303         *(unsigned long long *)&blkvsc_req->cmnd[2] =
 304         cpu_to_be64(blkvsc_req->sector_start);
 305         *(unsigned int *)&blkvsc_req->cmnd[10] =
 306         cpu_to_be32(blkvsc_req->sector_count);
 307 }
 308
 309
 310 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 311                         unsigned cmd, unsigned long arg)
 312 {
 313         struct block_device_context *blkdev = bd->bd_disk->private_data;
 314         int ret = 0;
 315
 316         switch (cmd) {
 317         case HDIO_GET_IDENTITY:
 318                 if (copy_to_user((void __user *)arg, blkdev->device_id,
 319                                  blkdev->device_id_len))
 320                         ret = -EFAULT;
 321                 break;
 322         default:
 323                 ret = -EINVAL;
 324                 break;
 325         }
 326
 327         return ret;
 328 }
 329
 330 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
 331 {
 332         struct blkvsc_request *blkvsc_req =
 333                         (struct blkvsc_request *)request->context;
 334         struct block_device_context *blkdev =
 335                         (struct block_device_context *)blkvsc_req->dev;
 336         struct scsi_sense_hdr sense_hdr;
 337         struct vmscsi_request *vm_srb;
 338
 339
 340         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 341         blkdev->num_outstanding_reqs--;
 342
 343         if (vm_srb->scsi_status)
 344                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
 345                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 346                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
 347
 348         complete(&blkvsc_req->request.wait_event);
 349 }
 350
 351
 352 static int blkvsc_do_operation(struct block_device_context *blkdev,
 353                                 enum blkvsc_op_type op)
 354 {
 355         struct blkvsc_request *blkvsc_req;
 356         struct page *page_buf;
 357         unsigned char *buf;
 358         unsigned char device_type;
 359         struct scsi_sense_hdr sense_hdr;
 360         struct vmscsi_request *vm_srb;
 361
 362         int ret = 0;
 363
 364         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 365         if (!blkvsc_req)
 366                 return -ENOMEM;
 367
 368         page_buf = alloc_page(GFP_KERNEL);
 369         if (!page_buf) {
 370                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 371                 return -ENOMEM;
 372         }
 373
 374         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 375         init_completion(&blkvsc_req->request.wait_event);
 376         blkvsc_req->dev = blkdev;
 377         blkvsc_req->req = NULL;
 378         blkvsc_req->write = 0;
 379
 380         blkvsc_req->request.data_buffer.pfn_array[0] =
 381         page_to_pfn(page_buf);
 382         blkvsc_req->request.data_buffer.offset = 0;
 383
 384         switch (op) {
 385         case DO_INQUIRY:
 386                 blkvsc_req->cmnd[0] = INQUIRY;
 387                 blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 388                 blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 389                 blkvsc_req->cmnd[4] = 64;
 390                 blkvsc_req->cmd_len = 6;
 391                 blkvsc_req->request.data_buffer.len = 64;
 392                 break;
 393
 394         case DO_CAPACITY:
 395                 blkdev->sector_size = 0;
 396                 blkdev->capacity = 0;
 397
 398                 blkvsc_req->cmnd[0] = READ_CAPACITY;
 399                 blkvsc_req->cmd_len = 16;
 400                 blkvsc_req->request.data_buffer.len = 8;
 401                 break;
 402
 403         case DO_FLUSH:
 404                 blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 405                 blkvsc_req->cmd_len = 10;
 406                 blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 407                 blkvsc_req->request.data_buffer.len = 0;
 408                 break;
 409         default:
 410                 ret = -EINVAL;
 411                 goto cleanup;
 412         }
 413
 414         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 415
 416         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 417
 418         /* check error */
 419         if (vm_srb->scsi_status) {
 420                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 421                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 422
 423                 return 0;
 424         }
 425
 426         buf = kmap(page_buf);
 427
 428         switch (op) {
 429         case DO_INQUIRY:
 430                 device_type = buf[0] & 0x1F;
 431
 432                 if (device_type == 0x0)
 433                         blkdev->device_type = HARDDISK_TYPE;
 434                  else
 435                         blkdev->device_type = UNKNOWN_DEV_TYPE;
 436
 437                 blkdev->device_id_len = buf[7];
 438                 if (blkdev->device_id_len > 64)
 439                         blkdev->device_id_len = 64;
 440
 441                 memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 442                 break;
 443
 444         case DO_CAPACITY:
 445                 /* be to le */
 446                 blkdev->capacity =
 447                 ((buf[0] << 24) | (buf[1] << 16) |
 448                 (buf[2] << 8) | buf[3]) + 1;
 449
 450                 blkdev->sector_size =
 451                 (buf[4] << 24) | (buf[5] << 16) |
 452                 (buf[6] << 8) | buf[7];
 453                 break;
 454         default:
 455                 break;
 456
 457         }
 458
 459 cleanup:
 460
 461         kunmap(page_buf);
 462
 463         __free_page(page_buf);
 464
 465         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 466
 467         return ret;
 468 }
 469
 470
 471 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
 472 {
 473         struct blkvsc_request *pend_req, *tmp;
 474         struct blkvsc_request *comp_req, *tmp2;
 475         struct vmscsi_request *vm_srb;
 476
 477         int ret = 0;
 478
 479
 480         /* Flush the pending list first */
 481         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 482                                  pend_entry) {
 483                 /*
 484                  * The pend_req could be part of a partially completed
 485                  * request. If so, complete those req first until we
 486                  * hit the pend_req
 487                  */
 488                 list_for_each_entry_safe(comp_req, tmp2,
 489                                          &pend_req->group->blkvsc_req_list,
 490                                          req_entry) {
 491
 492                         if (comp_req == pend_req)
 493                                 break;
 494
 495                         list_del(&comp_req->req_entry);
 496
 497                         if (comp_req->req) {
 498                                 vm_srb =
 499                                 &comp_req->request.vstor_packet.
 500                                 vm_srb;
 501                                 ret = __blk_end_request(comp_req->req,
 502                                         (!vm_srb->scsi_status ? 0 : -EIO),
 503                                         comp_req->sector_count *
 504                                         blkdev->sector_size);
 505
 506                                 /* FIXME: shouldn't this do more than return? */
 507                                 if (ret)
 508                                         goto out;
 509                         }
 510
 511                         kmem_cache_free(blkdev->request_pool, comp_req);
 512                 }
 513
 514                 list_del(&pend_req->pend_entry);
 515
 516                 list_del(&pend_req->req_entry);
 517
 518                 if (comp_req->req) {
 519                         if (!__blk_end_request(pend_req->req, -EIO,
 520                                                pend_req->sector_count *
 521                                                blkdev->sector_size)) {
 522                                 /*
 523                                  * All the sectors have been xferred ie the
 524                                  * request is done
 525                                  */
 526                                 kmem_cache_free(blkdev->request_pool,
 527                                                 pend_req->group);
 528                         }
 529                 }
 530
 531                 kmem_cache_free(blkdev->request_pool, pend_req);
 532         }
 533
 534 out:
 535         return ret;
 536 }
 537
 538
 539 /*
 540  * blkvsc_remove() - Callback when our device is removed
 541  */
 542 static int blkvsc_remove(struct device *device)
 543 {
 544         struct hv_driver *drv =
 545                                 drv_to_hv_drv(device->driver);
 546         struct storvsc_driver_object *storvsc_drv_obj =
 547                                 drv->priv;
 548         struct hv_device *device_obj = device_to_hv_device(device);
 549         struct block_device_context *blkdev = dev_get_drvdata(device);
 550         unsigned long flags;
 551
 552         /*
 553          * Call to the vsc driver to let it know that the device is being
 554          * removed
 555          */
 556         storvsc_drv_obj->base.dev_rm(device_obj);
 557
 558         /* Get to a known state */
 559         spin_lock_irqsave(&blkdev->lock, flags);
 560
 561         blkdev->shutting_down = 1;
 562
 563         blk_stop_queue(blkdev->gd->queue);
 564
 565         spin_unlock_irqrestore(&blkdev->lock, flags);
 566
 567         while (blkdev->num_outstanding_reqs) {
 568                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 569                             blkdev->num_outstanding_reqs);
 570                 udelay(100);
 571         }
 572
 573         blkvsc_do_operation(blkdev, DO_FLUSH);
 574
 575         spin_lock_irqsave(&blkdev->lock, flags);
 576
 577         blkvsc_cancel_pending_reqs(blkdev);
 578
 579         spin_unlock_irqrestore(&blkdev->lock, flags);
 580
 581         blk_cleanup_queue(blkdev->gd->queue);
 582
 583         del_gendisk(blkdev->gd);
 584
 585         kmem_cache_destroy(blkdev->request_pool);
 586
 587         kfree(blkdev);
 588
 589         return 0;
 590
 591 }
 592
 593 static void blkvsc_shutdown(struct device *device)
 594 {
 595         struct block_device_context *blkdev = dev_get_drvdata(device);
 596         unsigned long flags;
 597
 598         if (!blkdev)
 599                 return;
 600
 601         spin_lock_irqsave(&blkdev->lock, flags);
 602
 603         blkdev->shutting_down = 1;
 604
 605         blk_stop_queue(blkdev->gd->queue);
 606
 607         spin_unlock_irqrestore(&blkdev->lock, flags);
 608
 609         while (blkdev->num_outstanding_reqs) {
 610                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 611                             blkdev->num_outstanding_reqs);
 612                 udelay(100);
 613         }
 614
 615         blkvsc_do_operation(blkdev, DO_FLUSH);
 616
 617         spin_lock_irqsave(&blkdev->lock, flags);
 618
 619         blkvsc_cancel_pending_reqs(blkdev);
 620
 621         spin_unlock_irqrestore(&blkdev->lock, flags);
 622 }
 623
 624 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
 625 {
 626         struct block_device_context *blkdev = disk->private_data;
 627
 628         mutex_lock(&blkvsc_mutex);
 629         spin_lock(&blkdev->lock);
 630         if (blkdev->users == 1) {
 631                 spin_unlock(&blkdev->lock);
 632                 blkvsc_do_operation(blkdev, DO_FLUSH);
 633                 spin_lock(&blkdev->lock);
 634         }
 635
 636         blkdev->users--;
 637
 638         spin_unlock(&blkdev->lock);
 639         mutex_unlock(&blkvsc_mutex);
 640         return 0;
 641 }
 642
 643
 644 /*
 645  * We break the request into 1 or more blkvsc_requests and submit
 646  * them.  If we cant submit them all, we put them on the
 647  * pending_list. The blkvsc_request() will work on the pending_list.
 648  */
 649 static int blkvsc_do_request(struct block_device_context *blkdev,
 650                              struct request *req)
 651 {
 652         struct bio *bio = NULL;
 653         struct bio_vec *bvec = NULL;
 654         struct bio_vec *prev_bvec = NULL;
 655         struct blkvsc_request *blkvsc_req = NULL;
 656         struct blkvsc_request *tmp;
 657         int databuf_idx = 0;
 658         int seg_idx = 0;
 659         sector_t start_sector;
 660         unsigned long num_sectors = 0;
 661         int ret = 0;
 662         int pending = 0;
 663         struct blkvsc_request_group *group = NULL;
 664
 665         /* Create a group to tie req to list of blkvsc_reqs */
 666         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 667         if (!group)
 668                 return -ENOMEM;
 669
 670         INIT_LIST_HEAD(&group->blkvsc_req_list);
 671         group->outstanding = group->status = 0;
 672
 673         start_sector = blk_rq_pos(req);
 674
 675         /* foreach bio in the request */
 676         if (req->bio) {
 677                 for (bio = req->bio; bio; bio = bio->bi_next) {
 678                         /*
 679                          * Map this bio into an existing or new storvsc request
 680                          */
 681                         bio_for_each_segment(bvec, bio, seg_idx) {
 682                                 /* Get a new storvsc request */
 683                                 /* 1st-time */
 684                                 if ((!blkvsc_req) ||
 685                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 686                                     /* hole at the begin of page */
 687                                     || (bvec->bv_offset != 0) ||
 688                                     /* hold at the end of page */
 689                                     (prev_bvec &&
 690                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 691                                         /* submit the prev one */
 692                                         if (blkvsc_req) {
 693                                                 blkvsc_req->sector_start =
 694                                                 start_sector;
 695                                                 sector_div(
 696                                                 blkvsc_req->sector_start,
 697                                                 (blkdev->sector_size >> 9));
 698
 699                                                 blkvsc_req->sector_count =
 700                                                 num_sectors /
 701                                                 (blkdev->sector_size >> 9);
 702                                                 blkvsc_init_rw(blkvsc_req);
 703                                         }
 704
 705                                         /*
 706                                          * Create new blkvsc_req to represent
 707                                          * the current bvec
 708                                          */
 709                                         blkvsc_req =
 710                                         kmem_cache_zalloc(
 711                                         blkdev->request_pool, GFP_ATOMIC);
 712                                         if (!blkvsc_req) {
 713                                                 /* free up everything */
 714                                                 list_for_each_entry_safe(
 715                                                         blkvsc_req, tmp,
 716                                                         &group->blkvsc_req_list,
 717                                                         req_entry) {
 718                                                         list_del(
 719                                                         &blkvsc_req->req_entry);
 720                                                         kmem_cache_free(
 721                                                         blkdev->request_pool,
 722                                                         blkvsc_req);
 723                                                 }
 724
 725                                                 kmem_cache_free(
 726                                                 blkdev->request_pool, group);
 727                                                 return -ENOMEM;
 728                                         }
 729
 730                                         memset(blkvsc_req, 0,
 731                                                sizeof(struct blkvsc_request));
 732
 733                                         blkvsc_req->dev = blkdev;
 734                                         blkvsc_req->req = req;
 735                                         blkvsc_req->request.
 736                                         data_buffer.offset
 737                                         = bvec->bv_offset;
 738                                         blkvsc_req->request.
 739                                         data_buffer.len = 0;
 740
 741                                         /* Add to the group */
 742                                         blkvsc_req->group = group;
 743                                         blkvsc_req->group->outstanding++;
 744                                         list_add_tail(&blkvsc_req->req_entry,
 745                                         &blkvsc_req->group->blkvsc_req_list);
 746
 747                                         start_sector += num_sectors;
 748                                         num_sectors = 0;
 749                                         databuf_idx = 0;
 750                                 }
 751
 752                                 /*
 753                                  * Add the curr bvec/segment to the curr
 754                                  * blkvsc_req
 755                                  */
 756                                 blkvsc_req->request.data_buffer.
 757                                         pfn_array[databuf_idx]
 758                                                 = page_to_pfn(bvec->bv_page);
 759                                 blkvsc_req->request.data_buffer.len
 760                                         += bvec->bv_len;
 761
 762                                 prev_bvec = bvec;
 763
 764                                 databuf_idx++;
 765                                 num_sectors += bvec->bv_len >> 9;
 766
 767                         } /* bio_for_each_segment */
 768
 769                 } /* rq_for_each_bio */
 770         }
 771
 772         /* Handle the last one */
 773         if (blkvsc_req) {
 774                 blkvsc_req->sector_start = start_sector;
 775                 sector_div(blkvsc_req->sector_start,
 776                            (blkdev->sector_size >> 9));
 777
 778                 blkvsc_req->sector_count = num_sectors /
 779                                            (blkdev->sector_size >> 9);
 780
 781                 blkvsc_init_rw(blkvsc_req);
 782         }
 783
 784         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
 785                 if (pending) {
 786
 787                         list_add_tail(&blkvsc_req->pend_entry,
 788                                       &blkdev->pending_list);
 789                 } else {
 790                         ret = blkvsc_submit_request(blkvsc_req,
 791                                                     blkvsc_request_completion);
 792                         if (ret == -1) {
 793                                 pending = 1;
 794                                 list_add_tail(&blkvsc_req->pend_entry,
 795                                               &blkdev->pending_list);
 796                         }
 797
 798                 }
 799         }
 800
 801         return pending;
 802 }
 803
 804 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
 805 {
 806         struct blkvsc_request *pend_req, *tmp;
 807         int ret = 0;
 808
 809         /* Flush the pending list first */
 810         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 811                                  pend_entry) {
 812
 813                 ret = blkvsc_submit_request(pend_req,
 814                                             blkvsc_request_completion);
 815                 if (ret != 0)
 816                         break;
 817                 else
 818                         list_del(&pend_req->pend_entry);
 819         }
 820
 821         return ret;
 822 }
 823
 824
 825 static void blkvsc_request(struct request_queue *queue)
 826 {
 827         struct block_device_context *blkdev = NULL;
 828         struct request *req;
 829         int ret = 0;
 830
 831         while ((req = blk_peek_request(queue)) != NULL) {
 832
 833                 blkdev = req->rq_disk->private_data;
 834                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS) {
 835                         __blk_end_request_cur(req, 0);
 836                         continue;
 837                 }
 838
 839                 ret = blkvsc_do_pending_reqs(blkdev);
 840
 841                 if (ret != 0) {
 842                         blk_stop_queue(queue);
 843                         break;
 844                 }
 845
 846                 blk_start_request(req);
 847
 848                 ret = blkvsc_do_request(blkdev, req);
 849                 if (ret > 0) {
 850                         blk_stop_queue(queue);
 851                         break;
 852                 } else if (ret < 0) {
 853                         blk_requeue_request(queue, req);
 854                         blk_stop_queue(queue);
 855                         break;
 856                 }
 857         }
 858 }
 859
 860
 861
 862 /* The one and only one */
 863 static  struct storvsc_driver_object g_blkvsc_drv;
 864
 865 static const struct block_device_operations block_ops = {
 866         .owner = THIS_MODULE,
 867         .open = blkvsc_open,
 868         .release = blkvsc_release,
 869         .getgeo = blkvsc_getgeo,
 870         .ioctl  = blkvsc_ioctl,
 871 };
 872
 873 /*
 874  * blkvsc_drv_init -  BlkVsc driver initialization.
 875  */
 876 static int blkvsc_drv_init(void)
 877 {
 878         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 879         struct hv_driver *drv = &g_blkvsc_drv.base;
 880         int ret;
 881
 882         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 883
 884         drv->priv = storvsc_drv_obj;
 885
 886         /* Callback to client driver to complete the initialization */
 887         blk_vsc_initialize(&storvsc_drv_obj->base);
 888
 889         drv->driver.name = storvsc_drv_obj->base.name;
 890
 891         drv->driver.probe = blkvsc_probe;
 892         drv->driver.remove = blkvsc_remove;
 893         drv->driver.shutdown = blkvsc_shutdown;
 894
 895         /* The driver belongs to vmbus */
 896         ret = vmbus_child_driver_register(&drv->driver);
 897
 898         return ret;
 899 }
 900
 901 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 902 {
 903         struct device **curr = (struct device **)data;
 904         *curr = dev;
 905         return 1; /* stop iterating */
 906 }
 907
 908 static void blkvsc_drv_exit(void)
 909 {
 910         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 911         struct hv_driver *drv = &g_blkvsc_drv.base;
 912         struct device *current_dev;
 913         int ret;
 914
 915         while (1) {
 916                 current_dev = NULL;
 917
 918                 /* Get the device */
 919                 ret = driver_for_each_device(&drv->driver, NULL,
 920                                              (void *) &current_dev,
 921                                              blkvsc_drv_exit_cb);
 922
 923                 if (ret)
 924                         DPRINT_WARN(BLKVSC_DRV,
 925                                     "driver_for_each_device returned %d", ret);
 926
 927
 928                 if (current_dev == NULL)
 929                         break;
 930
 931                 /* Initiate removal from the top-down */
 932                 device_unregister(current_dev);
 933         }
 934
 935         if (storvsc_drv_obj->base.cleanup)
 936                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 937
 938         vmbus_child_driver_unregister(&drv->driver);
 939
 940         return;
 941 }
 942
 943 /*
 944  * blkvsc_probe - Add a new device for this driver
 945  */
 946 static int blkvsc_probe(struct device *device)
 947 {
 948         struct hv_driver *drv =
 949                                 drv_to_hv_drv(device->driver);
 950         struct storvsc_driver_object *storvsc_drv_obj =
 951                                 drv->priv;
 952         struct hv_device *device_obj = device_to_hv_device(device);
 953
 954         struct block_device_context *blkdev = NULL;
 955         struct storvsc_device_info device_info;
 956         struct storvsc_major_info major_info;
 957         int ret = 0;
 958
 959
 960         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 961         if (!blkdev) {
 962                 ret = -ENOMEM;
 963                 goto cleanup;
 964         }
 965
 966         INIT_LIST_HEAD(&blkdev->pending_list);
 967
 968         /* Initialize what we can here */
 969         spin_lock_init(&blkdev->lock);
 970
 971
 972         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 973                                         sizeof(struct blkvsc_request), 0,
 974                                         SLAB_HWCACHE_ALIGN, NULL);
 975         if (!blkdev->request_pool) {
 976                 ret = -ENOMEM;
 977                 goto cleanup;
 978         }
 979
 980
 981         /* Call to the vsc driver to add the device */
 982         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 983         if (ret != 0)
 984                 goto cleanup;
 985
 986         blkdev->device_ctx = device_obj;
 987         /* this identified the device 0 or 1 */
 988         blkdev->target = device_info.target_id;
 989         /* this identified the ide ctrl 0 or 1 */
 990         blkdev->path = device_info.path_id;
 991
 992         dev_set_drvdata(device, blkdev);
 993
 994         ret = stor_vsc_get_major_info(&device_info, &major_info);
 995
 996         if (ret)
 997                 goto cleanup;
 998
 999         if (major_info.do_register) {
1000                 ret = register_blkdev(major_info.major, major_info.devname);
1001
1002                 if (ret != 0) {
1003                         DPRINT_ERR(BLKVSC_DRV,
1004                                    "register_blkdev() failed! ret %d", ret);
1005                         goto remove;
1006                 }
1007         }
1008
1009         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!",
1010                         major_info.major);
1011
1012         blkdev->gd = alloc_disk(BLKVSC_MINORS);
1013         if (!blkdev->gd) {
1014                 ret = -1;
1015                 goto cleanup;
1016         }
1017
1018         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
1019
1020         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
1021         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
1022         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
1023         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
1024         blk_queue_dma_alignment(blkdev->gd->queue, 511);
1025
1026         blkdev->gd->major = major_info.major;
1027         if (major_info.index == 1 || major_info.index == 3)
1028                 blkdev->gd->first_minor = BLKVSC_MINORS;
1029         else
1030                 blkdev->gd->first_minor = 0;
1031         blkdev->gd->fops = &block_ops;
1032         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
1033         blkdev->gd->private_data = blkdev;
1034         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
1035         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + major_info.index);
1036
1037         blkvsc_do_operation(blkdev, DO_INQUIRY);
1038         blkvsc_do_operation(blkdev, DO_CAPACITY);
1039
1040         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
1041         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
1042         /* go! */
1043         add_disk(blkdev->gd);
1044
1045         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
1046                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
1047                     blkdev->sector_size);
1048
1049         return ret;
1050
1051 remove:
1052         storvsc_drv_obj->base.dev_rm(device_obj);
1053
1054 cleanup:
1055         if (blkdev) {
1056                 if (blkdev->request_pool) {
1057                         kmem_cache_destroy(blkdev->request_pool);
1058                         blkdev->request_pool = NULL;
1059                 }
1060                 kfree(blkdev);
1061                 blkdev = NULL;
1062         }
1063
1064         return ret;
1065 }
1066
1067 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1068 {
1069         struct blkvsc_request *blkvsc_req =
1070                         (struct blkvsc_request *)request->context;
1071         struct block_device_context *blkdev =
1072                         (struct block_device_context *)blkvsc_req->dev;
1073         unsigned long flags;
1074         struct blkvsc_request *comp_req, *tmp;
1075         struct vmscsi_request *vm_srb;
1076
1077
1078         spin_lock_irqsave(&blkdev->lock, flags);
1079
1080         blkdev->num_outstanding_reqs--;
1081         blkvsc_req->group->outstanding--;
1082
1083         /*
1084          * Only start processing when all the blkvsc_reqs are
1085          * completed. This guarantees no out-of-order blkvsc_req
1086          * completion when calling end_that_request_first()
1087          */
1088         if (blkvsc_req->group->outstanding == 0) {
1089                 list_for_each_entry_safe(comp_req, tmp,
1090                                          &blkvsc_req->group->blkvsc_req_list,
1091                                          req_entry) {
1092
1093                         list_del(&comp_req->req_entry);
1094
1095                         vm_srb =
1096                         &comp_req->request.vstor_packet.vm_srb;
1097                         if (!__blk_end_request(comp_req->req,
1098                                 (!vm_srb->scsi_status ? 0 : -EIO),
1099                                 comp_req->sector_count * blkdev->sector_size)) {
1100                                 /*
1101                                  * All the sectors have been xferred ie the
1102                                  * request is done
1103                                  */
1104                                 kmem_cache_free(blkdev->request_pool,
1105                                                 comp_req->group);
1106                         }
1107
1108                         kmem_cache_free(blkdev->request_pool, comp_req);
1109                 }
1110
1111                 if (!blkdev->shutting_down) {
1112                         blkvsc_do_pending_reqs(blkdev);
1113                         blk_start_queue(blkdev->gd->queue);
1114                         blkvsc_request(blkdev->gd->queue);
1115                 }
1116         }
1117
1118         spin_unlock_irqrestore(&blkdev->lock, flags);
1119 }
1120
1121 static int __init blkvsc_init(void)
1122 {
1123         int ret;
1124
1125         BUILD_BUG_ON(sizeof(sector_t) != 8);
1126
1127         ret = blkvsc_drv_init();
1128
1129         return ret;
1130 }
1131
1132 static void __exit blkvsc_exit(void)
1133 {
1134         blkvsc_drv_exit();
1135 }
1136
1137 MODULE_LICENSE("GPL");
1138 MODULE_VERSION(HV_DRV_VERSION);
1139 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1140 module_init(blkvsc_init);
1141 module_exit(blkvsc_exit);