drivers/staging/hv/blkvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/device.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/major.h>
  26 #include <linux/delay.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/mutex.h>
  29 #include <linux/slab.h>
  30 #include <scsi/scsi.h>
  31 #include <scsi/scsi_cmnd.h>
  32 #include <scsi/scsi_eh.h>
  33 #include <scsi/scsi_dbg.h>
  34 #include "hv_api.h"
  35 #include "logging.h"
  36 #include "version_info.h"
  37 #include "vmbus.h"
  38 #include "storvsc_api.h"
  39
  40
  41 #define BLKVSC_MINORS   64
  42
  43 enum blkvsc_device_type {
  44         UNKNOWN_DEV_TYPE,
  45         HARDDISK_TYPE,
  46         DVD_TYPE,
  47 };
  48
  49 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 DEFINE_MUTEX(blkvsc_mutex);
 115
 116 static const char *g_blk_driver_name = "blkvsc";
 117
 118 /* {32412632-86cb-44a2-9b5c-50d1417354f5} */
 119 static const struct hv_guid g_blk_device_type = {
 120         .data = {
 121                 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
 122                 0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5
 123         }
 124 };
 125
 126 /*
 127  * There is a circular dependency involving blkvsc_request_completion()
 128  * and blkvsc_do_request().
 129  */
 130 static void blkvsc_request_completion(struct hv_storvsc_request *request);
 131
 132 static int blkvsc_ringbuffer_size = BLKVSC_RING_BUFFER_SIZE;
 133
 134 module_param(blkvsc_ringbuffer_size, int, S_IRUGO);
 135 MODULE_PARM_DESC(ring_size, "Ring buffer size (in bytes)");
 136
 137 /*
 138  * There is a circular dependency involving blkvsc_probe()
 139  * and block_ops.
 140  */
 141 static int blkvsc_probe(struct device *dev);
 142
 143 static int blk_vsc_on_device_add(struct hv_device *device,
 144                                 void *additional_info)
 145 {
 146         struct storvsc_device_info *device_info;
 147         int ret = 0;
 148
 149         device_info = (struct storvsc_device_info *)additional_info;
 150
 151         ret = stor_vsc_on_device_add(device, additional_info);
 152         if (ret != 0)
 153                 return ret;
 154
 155         /*
 156          * We need to use the device instance guid to set the path and target
 157          * id. For IDE devices, the device instance id is formatted as
 158          * <bus id> * - <device id> - 8899 - 000000000000.
 159          */
 160         device_info->path_id = device->dev_instance.data[3] << 24 |
 161                              device->dev_instance.data[2] << 16 |
 162                              device->dev_instance.data[1] << 8  |
 163                              device->dev_instance.data[0];
 164
 165         device_info->target_id = device->dev_instance.data[5] << 8 |
 166                                device->dev_instance.data[4];
 167
 168         return ret;
 169 }
 170
 171
 172 static int blk_vsc_initialize(struct hv_driver *driver)
 173 {
 174         struct storvsc_driver_object *stor_driver;
 175         int ret = 0;
 176
 177         stor_driver = hvdr_to_stordr(driver);
 178
 179         /* Make sure we are at least 2 pages since 1 page is used for control */
 180
 181         driver->name = g_blk_driver_name;
 182         memcpy(&driver->dev_type, &g_blk_device_type, sizeof(struct hv_guid));
 183
 184
 185         /*
 186          * Divide the ring buffer data size (which is 1 page less than the ring
 187          * buffer size since that page is reserved for the ring buffer indices)
 188          * by the max request size (which is
 189          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
 190          */
 191         stor_driver->max_outstanding_req_per_channel =
 192                 ((stor_driver->ring_buffer_size - PAGE_SIZE) /
 193                   ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
 194                            sizeof(struct vstor_packet) + sizeof(u64),
 195                            sizeof(u64)));
 196
 197         DPRINT_INFO(BLKVSC, "max io outstd %u",
 198                     stor_driver->max_outstanding_req_per_channel);
 199
 200         /* Setup the dispatch table */
 201         stor_driver->base.dev_add = blk_vsc_on_device_add;
 202         stor_driver->base.dev_rm = stor_vsc_on_device_remove;
 203         stor_driver->base.cleanup = stor_vsc_on_cleanup;
 204         stor_driver->on_io_request = stor_vsc_on_io_request;
 205
 206         return ret;
 207 }
 208
 209
 210 static int blkvsc_submit_request(struct blkvsc_request *blkvsc_req,
 211                         void (*request_completion)(struct hv_storvsc_request *))
 212 {
 213         struct block_device_context *blkdev = blkvsc_req->dev;
 214         struct hv_device *device_ctx = blkdev->device_ctx;
 215         struct hv_driver *drv =
 216                         drv_to_hv_drv(device_ctx->device.driver);
 217         struct storvsc_driver_object *storvsc_drv_obj =
 218                         drv->priv;
 219         struct hv_storvsc_request *storvsc_req;
 220         struct vmscsi_request *vm_srb;
 221         int ret;
 222
 223
 224         storvsc_req = &blkvsc_req->request;
 225         vm_srb = &storvsc_req->vstor_packet.vm_srb;
 226
 227         vm_srb->data_in = blkvsc_req->write ? WRITE_TYPE : READ_TYPE;
 228
 229         storvsc_req->on_io_completion = request_completion;
 230         storvsc_req->context = blkvsc_req;
 231
 232         vm_srb->port_number = blkdev->port;
 233         vm_srb->path_id = blkdev->path;
 234         vm_srb->target_id = blkdev->target;
 235         vm_srb->lun = 0;         /* this is not really used at all */
 236
 237         vm_srb->cdb_length = blkvsc_req->cmd_len;
 238
 239         memcpy(vm_srb->cdb, blkvsc_req->cmnd, vm_srb->cdb_length);
 240
 241         storvsc_req->sense_buffer = blkvsc_req->sense_buffer;
 242
 243         ret = storvsc_drv_obj->on_io_request(blkdev->device_ctx,
 244                                            &blkvsc_req->request);
 245         if (ret == 0)
 246                 blkdev->num_outstanding_reqs++;
 247
 248         return ret;
 249 }
 250
 251
 252 static int blkvsc_open(struct block_device *bdev, fmode_t mode)
 253 {
 254         struct block_device_context *blkdev = bdev->bd_disk->private_data;
 255
 256
 257         mutex_lock(&blkvsc_mutex);
 258         spin_lock(&blkdev->lock);
 259
 260         if (!blkdev->users && blkdev->device_type == DVD_TYPE) {
 261                 spin_unlock(&blkdev->lock);
 262                 check_disk_change(bdev);
 263                 spin_lock(&blkdev->lock);
 264         }
 265
 266         blkdev->users++;
 267
 268         spin_unlock(&blkdev->lock);
 269         mutex_unlock(&blkvsc_mutex);
 270         return 0;
 271 }
 272
 273
 274 static int blkvsc_getgeo(struct block_device *bd, struct hd_geometry *hg)
 275 {
 276         sector_t nsect = get_capacity(bd->bd_disk);
 277         sector_t cylinders = nsect;
 278
 279         /*
 280          * We are making up these values; let us keep it simple.
 281          */
 282         hg->heads = 0xff;
 283         hg->sectors = 0x3f;
 284         sector_div(cylinders, hg->heads * hg->sectors);
 285         hg->cylinders = cylinders;
 286         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 287                 hg->cylinders = 0xffff;
 288         return 0;
 289
 290 }
 291
 292
 293 static void blkvsc_init_rw(struct blkvsc_request *blkvsc_req)
 294 {
 295
 296         blkvsc_req->cmd_len = 16;
 297
 298         if (rq_data_dir(blkvsc_req->req)) {
 299                 blkvsc_req->write = 1;
 300                 blkvsc_req->cmnd[0] = WRITE_16;
 301         } else {
 302                 blkvsc_req->write = 0;
 303                 blkvsc_req->cmnd[0] = READ_16;
 304         }
 305
 306         blkvsc_req->cmnd[1] |=
 307         (blkvsc_req->req->cmd_flags & REQ_FUA) ? 0x8 : 0;
 308
 309         *(unsigned long long *)&blkvsc_req->cmnd[2] =
 310         cpu_to_be64(blkvsc_req->sector_start);
 311         *(unsigned int *)&blkvsc_req->cmnd[10] =
 312         cpu_to_be32(blkvsc_req->sector_count);
 313 }
 314
 315
 316 static int blkvsc_ioctl(struct block_device *bd, fmode_t mode,
 317                         unsigned cmd, unsigned long arg)
 318 {
 319         struct block_device_context *blkdev = bd->bd_disk->private_data;
 320         int ret = 0;
 321
 322         switch (cmd) {
 323         case HDIO_GET_IDENTITY:
 324                 if (copy_to_user((void __user *)arg, blkdev->device_id,
 325                                  blkdev->device_id_len))
 326                         ret = -EFAULT;
 327                 break;
 328         default:
 329                 ret = -EINVAL;
 330                 break;
 331         }
 332
 333         return ret;
 334 }
 335
 336 static void blkvsc_cmd_completion(struct hv_storvsc_request *request)
 337 {
 338         struct blkvsc_request *blkvsc_req =
 339                         (struct blkvsc_request *)request->context;
 340         struct block_device_context *blkdev =
 341                         (struct block_device_context *)blkvsc_req->dev;
 342         struct scsi_sense_hdr sense_hdr;
 343         struct vmscsi_request *vm_srb;
 344
 345
 346         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 347         blkdev->num_outstanding_reqs--;
 348
 349         if (vm_srb->scsi_status)
 350                 if (scsi_normalize_sense(blkvsc_req->sense_buffer,
 351                                          SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 352                         scsi_print_sense_hdr("blkvsc", &sense_hdr);
 353
 354         complete(&blkvsc_req->request.wait_event);
 355 }
 356
 357
 358 static int blkvsc_do_operation(struct block_device_context *blkdev,
 359                                 enum blkvsc_op_type op)
 360 {
 361         struct blkvsc_request *blkvsc_req;
 362         struct page *page_buf;
 363         unsigned char *buf;
 364         unsigned char device_type;
 365         struct scsi_sense_hdr sense_hdr;
 366         struct vmscsi_request *vm_srb;
 367
 368         int ret = 0;
 369
 370         blkvsc_req = kmem_cache_zalloc(blkdev->request_pool, GFP_KERNEL);
 371         if (!blkvsc_req)
 372                 return -ENOMEM;
 373
 374         page_buf = alloc_page(GFP_KERNEL);
 375         if (!page_buf) {
 376                 kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 377                 return -ENOMEM;
 378         }
 379
 380         vm_srb = &blkvsc_req->request.vstor_packet.vm_srb;
 381         init_completion(&blkvsc_req->request.wait_event);
 382         blkvsc_req->dev = blkdev;
 383         blkvsc_req->req = NULL;
 384         blkvsc_req->write = 0;
 385
 386         blkvsc_req->request.data_buffer.pfn_array[0] =
 387         page_to_pfn(page_buf);
 388         blkvsc_req->request.data_buffer.offset = 0;
 389
 390         switch (op) {
 391         case DO_INQUIRY:
 392                 blkvsc_req->cmnd[0] = INQUIRY;
 393                 blkvsc_req->cmnd[1] = 0x1;              /* Get product data */
 394                 blkvsc_req->cmnd[2] = 0x83;             /* mode page 83 */
 395                 blkvsc_req->cmnd[4] = 64;
 396                 blkvsc_req->cmd_len = 6;
 397                 blkvsc_req->request.data_buffer.len = 64;
 398                 break;
 399
 400         case DO_CAPACITY:
 401                 blkdev->sector_size = 0;
 402                 blkdev->capacity = 0;
 403
 404                 blkvsc_req->cmnd[0] = READ_CAPACITY;
 405                 blkvsc_req->cmd_len = 16;
 406                 blkvsc_req->request.data_buffer.len = 8;
 407                 break;
 408
 409         case DO_FLUSH:
 410                 blkvsc_req->cmnd[0] = SYNCHRONIZE_CACHE;
 411                 blkvsc_req->cmd_len = 10;
 412                 blkvsc_req->request.data_buffer.pfn_array[0] = 0;
 413                 blkvsc_req->request.data_buffer.len = 0;
 414                 break;
 415         default:
 416                 ret = -EINVAL;
 417                 goto cleanup;
 418         }
 419
 420         blkvsc_submit_request(blkvsc_req, blkvsc_cmd_completion);
 421
 422         wait_for_completion_interruptible(&blkvsc_req->request.wait_event);
 423
 424         /* check error */
 425         if (vm_srb->scsi_status) {
 426                 scsi_normalize_sense(blkvsc_req->sense_buffer,
 427                                      SCSI_SENSE_BUFFERSIZE, &sense_hdr);
 428
 429                 return 0;
 430         }
 431
 432         buf = kmap(page_buf);
 433
 434         switch (op) {
 435         case DO_INQUIRY:
 436                 device_type = buf[0] & 0x1F;
 437
 438                 if (device_type == 0x0)
 439                         blkdev->device_type = HARDDISK_TYPE;
 440                  else
 441                         blkdev->device_type = UNKNOWN_DEV_TYPE;
 442
 443                 blkdev->device_id_len = buf[7];
 444                 if (blkdev->device_id_len > 64)
 445                         blkdev->device_id_len = 64;
 446
 447                 memcpy(blkdev->device_id, &buf[8], blkdev->device_id_len);
 448                 break;
 449
 450         case DO_CAPACITY:
 451                 /* be to le */
 452                 blkdev->capacity =
 453                 ((buf[0] << 24) | (buf[1] << 16) |
 454                 (buf[2] << 8) | buf[3]) + 1;
 455
 456                 blkdev->sector_size =
 457                 (buf[4] << 24) | (buf[5] << 16) |
 458                 (buf[6] << 8) | buf[7];
 459                 break;
 460         default:
 461                 break;
 462
 463         }
 464
 465 cleanup:
 466
 467         kunmap(page_buf);
 468
 469         __free_page(page_buf);
 470
 471         kmem_cache_free(blkvsc_req->dev->request_pool, blkvsc_req);
 472
 473         return ret;
 474 }
 475
 476
 477 static int blkvsc_cancel_pending_reqs(struct block_device_context *blkdev)
 478 {
 479         struct blkvsc_request *pend_req, *tmp;
 480         struct blkvsc_request *comp_req, *tmp2;
 481         struct vmscsi_request *vm_srb;
 482
 483         int ret = 0;
 484
 485
 486         /* Flush the pending list first */
 487         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 488                                  pend_entry) {
 489                 /*
 490                  * The pend_req could be part of a partially completed
 491                  * request. If so, complete those req first until we
 492                  * hit the pend_req
 493                  */
 494                 list_for_each_entry_safe(comp_req, tmp2,
 495                                          &pend_req->group->blkvsc_req_list,
 496                                          req_entry) {
 497
 498                         if (comp_req == pend_req)
 499                                 break;
 500
 501                         list_del(&comp_req->req_entry);
 502
 503                         if (comp_req->req) {
 504                                 vm_srb =
 505                                 &comp_req->request.vstor_packet.
 506                                 vm_srb;
 507                                 ret = __blk_end_request(comp_req->req,
 508                                         (!vm_srb->scsi_status ? 0 : -EIO),
 509                                         comp_req->sector_count *
 510                                         blkdev->sector_size);
 511
 512                                 /* FIXME: shouldn't this do more than return? */
 513                                 if (ret)
 514                                         goto out;
 515                         }
 516
 517                         kmem_cache_free(blkdev->request_pool, comp_req);
 518                 }
 519
 520                 list_del(&pend_req->pend_entry);
 521
 522                 list_del(&pend_req->req_entry);
 523
 524                 if (comp_req->req) {
 525                         if (!__blk_end_request(pend_req->req, -EIO,
 526                                                pend_req->sector_count *
 527                                                blkdev->sector_size)) {
 528                                 /*
 529                                  * All the sectors have been xferred ie the
 530                                  * request is done
 531                                  */
 532                                 kmem_cache_free(blkdev->request_pool,
 533                                                 pend_req->group);
 534                         }
 535                 }
 536
 537                 kmem_cache_free(blkdev->request_pool, pend_req);
 538         }
 539
 540 out:
 541         return ret;
 542 }
 543
 544
 545 /*
 546  * blkvsc_remove() - Callback when our device is removed
 547  */
 548 static int blkvsc_remove(struct device *device)
 549 {
 550         struct hv_driver *drv =
 551                                 drv_to_hv_drv(device->driver);
 552         struct storvsc_driver_object *storvsc_drv_obj =
 553                                 drv->priv;
 554         struct hv_device *device_obj = device_to_hv_device(device);
 555         struct block_device_context *blkdev = dev_get_drvdata(device);
 556         unsigned long flags;
 557
 558         /*
 559          * Call to the vsc driver to let it know that the device is being
 560          * removed
 561          */
 562         storvsc_drv_obj->base.dev_rm(device_obj);
 563
 564         /* Get to a known state */
 565         spin_lock_irqsave(&blkdev->lock, flags);
 566
 567         blkdev->shutting_down = 1;
 568
 569         blk_stop_queue(blkdev->gd->queue);
 570
 571         spin_unlock_irqrestore(&blkdev->lock, flags);
 572
 573         while (blkdev->num_outstanding_reqs) {
 574                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 575                             blkdev->num_outstanding_reqs);
 576                 udelay(100);
 577         }
 578
 579         blkvsc_do_operation(blkdev, DO_FLUSH);
 580
 581         spin_lock_irqsave(&blkdev->lock, flags);
 582
 583         blkvsc_cancel_pending_reqs(blkdev);
 584
 585         spin_unlock_irqrestore(&blkdev->lock, flags);
 586
 587         blk_cleanup_queue(blkdev->gd->queue);
 588
 589         del_gendisk(blkdev->gd);
 590
 591         kmem_cache_destroy(blkdev->request_pool);
 592
 593         kfree(blkdev);
 594
 595         return 0;
 596
 597 }
 598
 599 static void blkvsc_shutdown(struct device *device)
 600 {
 601         struct block_device_context *blkdev = dev_get_drvdata(device);
 602         unsigned long flags;
 603
 604         if (!blkdev)
 605                 return;
 606
 607         spin_lock_irqsave(&blkdev->lock, flags);
 608
 609         blkdev->shutting_down = 1;
 610
 611         blk_stop_queue(blkdev->gd->queue);
 612
 613         spin_unlock_irqrestore(&blkdev->lock, flags);
 614
 615         while (blkdev->num_outstanding_reqs) {
 616                 DPRINT_INFO(STORVSC, "waiting for %d requests to complete...",
 617                             blkdev->num_outstanding_reqs);
 618                 udelay(100);
 619         }
 620
 621         blkvsc_do_operation(blkdev, DO_FLUSH);
 622
 623         spin_lock_irqsave(&blkdev->lock, flags);
 624
 625         blkvsc_cancel_pending_reqs(blkdev);
 626
 627         spin_unlock_irqrestore(&blkdev->lock, flags);
 628 }
 629
 630 static int blkvsc_release(struct gendisk *disk, fmode_t mode)
 631 {
 632         struct block_device_context *blkdev = disk->private_data;
 633
 634         mutex_lock(&blkvsc_mutex);
 635         spin_lock(&blkdev->lock);
 636         if (blkdev->users == 1) {
 637                 spin_unlock(&blkdev->lock);
 638                 blkvsc_do_operation(blkdev, DO_FLUSH);
 639                 spin_lock(&blkdev->lock);
 640         }
 641
 642         blkdev->users--;
 643
 644         spin_unlock(&blkdev->lock);
 645         mutex_unlock(&blkvsc_mutex);
 646         return 0;
 647 }
 648
 649
 650 /*
 651  * We break the request into 1 or more blkvsc_requests and submit
 652  * them.  If we cant submit them all, we put them on the
 653  * pending_list. The blkvsc_request() will work on the pending_list.
 654  */
 655 static int blkvsc_do_request(struct block_device_context *blkdev,
 656                              struct request *req)
 657 {
 658         struct bio *bio = NULL;
 659         struct bio_vec *bvec = NULL;
 660         struct bio_vec *prev_bvec = NULL;
 661         struct blkvsc_request *blkvsc_req = NULL;
 662         struct blkvsc_request *tmp;
 663         int databuf_idx = 0;
 664         int seg_idx = 0;
 665         sector_t start_sector;
 666         unsigned long num_sectors = 0;
 667         int ret = 0;
 668         int pending = 0;
 669         struct blkvsc_request_group *group = NULL;
 670
 671         /* Create a group to tie req to list of blkvsc_reqs */
 672         group = kmem_cache_zalloc(blkdev->request_pool, GFP_ATOMIC);
 673         if (!group)
 674                 return -ENOMEM;
 675
 676         INIT_LIST_HEAD(&group->blkvsc_req_list);
 677         group->outstanding = group->status = 0;
 678
 679         start_sector = blk_rq_pos(req);
 680
 681         /* foreach bio in the request */
 682         if (req->bio) {
 683                 for (bio = req->bio; bio; bio = bio->bi_next) {
 684                         /*
 685                          * Map this bio into an existing or new storvsc request
 686                          */
 687                         bio_for_each_segment(bvec, bio, seg_idx) {
 688                                 /* Get a new storvsc request */
 689                                 /* 1st-time */
 690                                 if ((!blkvsc_req) ||
 691                                     (databuf_idx >= MAX_MULTIPAGE_BUFFER_COUNT)
 692                                     /* hole at the begin of page */
 693                                     || (bvec->bv_offset != 0) ||
 694                                     /* hold at the end of page */
 695                                     (prev_bvec &&
 696                                      (prev_bvec->bv_len != PAGE_SIZE))) {
 697                                         /* submit the prev one */
 698                                         if (blkvsc_req) {
 699                                                 blkvsc_req->sector_start =
 700                                                 start_sector;
 701                                                 sector_div(
 702                                                 blkvsc_req->sector_start,
 703                                                 (blkdev->sector_size >> 9));
 704
 705                                                 blkvsc_req->sector_count =
 706                                                 num_sectors /
 707                                                 (blkdev->sector_size >> 9);
 708                                                 blkvsc_init_rw(blkvsc_req);
 709                                         }
 710
 711                                         /*
 712                                          * Create new blkvsc_req to represent
 713                                          * the current bvec
 714                                          */
 715                                         blkvsc_req =
 716                                         kmem_cache_zalloc(
 717                                         blkdev->request_pool, GFP_ATOMIC);
 718                                         if (!blkvsc_req) {
 719                                                 /* free up everything */
 720                                                 list_for_each_entry_safe(
 721                                                         blkvsc_req, tmp,
 722                                                         &group->blkvsc_req_list,
 723                                                         req_entry) {
 724                                                         list_del(
 725                                                         &blkvsc_req->req_entry);
 726                                                         kmem_cache_free(
 727                                                         blkdev->request_pool,
 728                                                         blkvsc_req);
 729                                                 }
 730
 731                                                 kmem_cache_free(
 732                                                 blkdev->request_pool, group);
 733                                                 return -ENOMEM;
 734                                         }
 735
 736                                         memset(blkvsc_req, 0,
 737                                                sizeof(struct blkvsc_request));
 738
 739                                         blkvsc_req->dev = blkdev;
 740                                         blkvsc_req->req = req;
 741                                         blkvsc_req->request.
 742                                         data_buffer.offset
 743                                         = bvec->bv_offset;
 744                                         blkvsc_req->request.
 745                                         data_buffer.len = 0;
 746
 747                                         /* Add to the group */
 748                                         blkvsc_req->group = group;
 749                                         blkvsc_req->group->outstanding++;
 750                                         list_add_tail(&blkvsc_req->req_entry,
 751                                         &blkvsc_req->group->blkvsc_req_list);
 752
 753                                         start_sector += num_sectors;
 754                                         num_sectors = 0;
 755                                         databuf_idx = 0;
 756                                 }
 757
 758                                 /*
 759                                  * Add the curr bvec/segment to the curr
 760                                  * blkvsc_req
 761                                  */
 762                                 blkvsc_req->request.data_buffer.
 763                                         pfn_array[databuf_idx]
 764                                                 = page_to_pfn(bvec->bv_page);
 765                                 blkvsc_req->request.data_buffer.len
 766                                         += bvec->bv_len;
 767
 768                                 prev_bvec = bvec;
 769
 770                                 databuf_idx++;
 771                                 num_sectors += bvec->bv_len >> 9;
 772
 773                         } /* bio_for_each_segment */
 774
 775                 } /* rq_for_each_bio */
 776         }
 777
 778         /* Handle the last one */
 779         if (blkvsc_req) {
 780                 blkvsc_req->sector_start = start_sector;
 781                 sector_div(blkvsc_req->sector_start,
 782                            (blkdev->sector_size >> 9));
 783
 784                 blkvsc_req->sector_count = num_sectors /
 785                                            (blkdev->sector_size >> 9);
 786
 787                 blkvsc_init_rw(blkvsc_req);
 788         }
 789
 790         list_for_each_entry(blkvsc_req, &group->blkvsc_req_list, req_entry) {
 791                 if (pending) {
 792
 793                         list_add_tail(&blkvsc_req->pend_entry,
 794                                       &blkdev->pending_list);
 795                 } else {
 796                         ret = blkvsc_submit_request(blkvsc_req,
 797                                                     blkvsc_request_completion);
 798                         if (ret == -1) {
 799                                 pending = 1;
 800                                 list_add_tail(&blkvsc_req->pend_entry,
 801                                               &blkdev->pending_list);
 802                         }
 803
 804                 }
 805         }
 806
 807         return pending;
 808 }
 809
 810 static int blkvsc_do_pending_reqs(struct block_device_context *blkdev)
 811 {
 812         struct blkvsc_request *pend_req, *tmp;
 813         int ret = 0;
 814
 815         /* Flush the pending list first */
 816         list_for_each_entry_safe(pend_req, tmp, &blkdev->pending_list,
 817                                  pend_entry) {
 818
 819                 ret = blkvsc_submit_request(pend_req,
 820                                             blkvsc_request_completion);
 821                 if (ret != 0)
 822                         break;
 823                 else
 824                         list_del(&pend_req->pend_entry);
 825         }
 826
 827         return ret;
 828 }
 829
 830
 831 static void blkvsc_request(struct request_queue *queue)
 832 {
 833         struct block_device_context *blkdev = NULL;
 834         struct request *req;
 835         int ret = 0;
 836
 837         while ((req = blk_peek_request(queue)) != NULL) {
 838
 839                 blkdev = req->rq_disk->private_data;
 840                 if (blkdev->shutting_down || req->cmd_type != REQ_TYPE_FS) {
 841                         __blk_end_request_cur(req, 0);
 842                         continue;
 843                 }
 844
 845                 ret = blkvsc_do_pending_reqs(blkdev);
 846
 847                 if (ret != 0) {
 848                         blk_stop_queue(queue);
 849                         break;
 850                 }
 851
 852                 blk_start_request(req);
 853
 854                 ret = blkvsc_do_request(blkdev, req);
 855                 if (ret > 0) {
 856                         blk_stop_queue(queue);
 857                         break;
 858                 } else if (ret < 0) {
 859                         blk_requeue_request(queue, req);
 860                         blk_stop_queue(queue);
 861                         break;
 862                 }
 863         }
 864 }
 865
 866
 867
 868 /* The one and only one */
 869 static  struct storvsc_driver_object g_blkvsc_drv;
 870
 871 static const struct block_device_operations block_ops = {
 872         .owner = THIS_MODULE,
 873         .open = blkvsc_open,
 874         .release = blkvsc_release,
 875         .getgeo = blkvsc_getgeo,
 876         .ioctl  = blkvsc_ioctl,
 877 };
 878
 879 /*
 880  * blkvsc_drv_init -  BlkVsc driver initialization.
 881  */
 882 static int blkvsc_drv_init(void)
 883 {
 884         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 885         struct hv_driver *drv = &g_blkvsc_drv.base;
 886         int ret;
 887
 888         storvsc_drv_obj->ring_buffer_size = blkvsc_ringbuffer_size;
 889
 890         drv->priv = storvsc_drv_obj;
 891
 892         /* Callback to client driver to complete the initialization */
 893         blk_vsc_initialize(&storvsc_drv_obj->base);
 894
 895         drv->driver.name = storvsc_drv_obj->base.name;
 896
 897         drv->driver.probe = blkvsc_probe;
 898         drv->driver.remove = blkvsc_remove;
 899         drv->driver.shutdown = blkvsc_shutdown;
 900
 901         /* The driver belongs to vmbus */
 902         ret = vmbus_child_driver_register(&drv->driver);
 903
 904         return ret;
 905 }
 906
 907 static int blkvsc_drv_exit_cb(struct device *dev, void *data)
 908 {
 909         struct device **curr = (struct device **)data;
 910         *curr = dev;
 911         return 1; /* stop iterating */
 912 }
 913
 914 static void blkvsc_drv_exit(void)
 915 {
 916         struct storvsc_driver_object *storvsc_drv_obj = &g_blkvsc_drv;
 917         struct hv_driver *drv = &g_blkvsc_drv.base;
 918         struct device *current_dev;
 919         int ret;
 920
 921         while (1) {
 922                 current_dev = NULL;
 923
 924                 /* Get the device */
 925                 ret = driver_for_each_device(&drv->driver, NULL,
 926                                              (void *) &current_dev,
 927                                              blkvsc_drv_exit_cb);
 928
 929                 if (ret)
 930                         DPRINT_WARN(BLKVSC_DRV,
 931                                     "driver_for_each_device returned %d", ret);
 932
 933
 934                 if (current_dev == NULL)
 935                         break;
 936
 937                 /* Initiate removal from the top-down */
 938                 device_unregister(current_dev);
 939         }
 940
 941         if (storvsc_drv_obj->base.cleanup)
 942                 storvsc_drv_obj->base.cleanup(&storvsc_drv_obj->base);
 943
 944         vmbus_child_driver_unregister(&drv->driver);
 945
 946         return;
 947 }
 948
 949 /*
 950  * blkvsc_probe - Add a new device for this driver
 951  */
 952 static int blkvsc_probe(struct device *device)
 953 {
 954         struct hv_driver *drv =
 955                                 drv_to_hv_drv(device->driver);
 956         struct storvsc_driver_object *storvsc_drv_obj =
 957                                 drv->priv;
 958         struct hv_device *device_obj = device_to_hv_device(device);
 959
 960         struct block_device_context *blkdev = NULL;
 961         struct storvsc_device_info device_info;
 962         int major = 0;
 963         int devnum = 0;
 964         int ret = 0;
 965         static int ide0_registered;
 966         static int ide1_registered;
 967
 968
 969         if (!storvsc_drv_obj->base.dev_add) {
 970                 DPRINT_ERR(BLKVSC_DRV, "OnDeviceAdd() not set");
 971                 ret = -1;
 972                 goto Cleanup;
 973         }
 974
 975         blkdev = kzalloc(sizeof(struct block_device_context), GFP_KERNEL);
 976         if (!blkdev) {
 977                 ret = -ENOMEM;
 978                 goto Cleanup;
 979         }
 980
 981         INIT_LIST_HEAD(&blkdev->pending_list);
 982
 983         /* Initialize what we can here */
 984         spin_lock_init(&blkdev->lock);
 985
 986
 987         blkdev->request_pool = kmem_cache_create(dev_name(&device_obj->device),
 988                                         sizeof(struct blkvsc_request), 0,
 989                                         SLAB_HWCACHE_ALIGN, NULL);
 990         if (!blkdev->request_pool) {
 991                 ret = -ENOMEM;
 992                 goto Cleanup;
 993         }
 994
 995
 996         /* Call to the vsc driver to add the device */
 997         ret = storvsc_drv_obj->base.dev_add(device_obj, &device_info);
 998         if (ret != 0) {
 999                 DPRINT_ERR(BLKVSC_DRV, "unable to add blkvsc device");
1000                 goto Cleanup;
1001         }
1002
1003         blkdev->device_ctx = device_obj;
1004         /* this identified the device 0 or 1 */
1005         blkdev->target = device_info.target_id;
1006         /* this identified the ide ctrl 0 or 1 */
1007         blkdev->path = device_info.path_id;
1008
1009         dev_set_drvdata(device, blkdev);
1010
1011         /* Calculate the major and device num */
1012         if (blkdev->path == 0) {
1013                 major = IDE0_MAJOR;
1014                 devnum = blkdev->path + blkdev->target;         /* 0 or 1 */
1015
1016                 if (!ide0_registered) {
1017                         ret = register_blkdev(major, "ide");
1018                         if (ret != 0) {
1019                                 DPRINT_ERR(BLKVSC_DRV,
1020                                            "register_blkdev() failed! ret %d",
1021                                            ret);
1022                                 goto Remove;
1023                         }
1024
1025                         ide0_registered = 1;
1026                 }
1027         } else if (blkdev->path == 1) {
1028                 major = IDE1_MAJOR;
1029                 devnum = blkdev->path + blkdev->target + 1; /* 2 or 3 */
1030
1031                 if (!ide1_registered) {
1032                         ret = register_blkdev(major, "ide");
1033                         if (ret != 0) {
1034                                 DPRINT_ERR(BLKVSC_DRV,
1035                                            "register_blkdev() failed! ret %d",
1036                                            ret);
1037                                 goto Remove;
1038                         }
1039
1040                         ide1_registered = 1;
1041                 }
1042         } else {
1043                 DPRINT_ERR(BLKVSC_DRV, "invalid pathid");
1044                 ret = -1;
1045                 goto Cleanup;
1046         }
1047
1048         DPRINT_INFO(BLKVSC_DRV, "blkvsc registered for major %d!!", major);
1049
1050         blkdev->gd = alloc_disk(BLKVSC_MINORS);
1051         if (!blkdev->gd) {
1052                 DPRINT_ERR(BLKVSC_DRV, "register_blkdev() failed! ret %d", ret);
1053                 ret = -1;
1054                 goto Cleanup;
1055         }
1056
1057         blkdev->gd->queue = blk_init_queue(blkvsc_request, &blkdev->lock);
1058
1059         blk_queue_max_segment_size(blkdev->gd->queue, PAGE_SIZE);
1060         blk_queue_max_segments(blkdev->gd->queue, MAX_MULTIPAGE_BUFFER_COUNT);
1061         blk_queue_segment_boundary(blkdev->gd->queue, PAGE_SIZE-1);
1062         blk_queue_bounce_limit(blkdev->gd->queue, BLK_BOUNCE_ANY);
1063         blk_queue_dma_alignment(blkdev->gd->queue, 511);
1064
1065         blkdev->gd->major = major;
1066         if (devnum == 1 || devnum == 3)
1067                 blkdev->gd->first_minor = BLKVSC_MINORS;
1068         else
1069                 blkdev->gd->first_minor = 0;
1070         blkdev->gd->fops = &block_ops;
1071         blkdev->gd->events = DISK_EVENT_MEDIA_CHANGE;
1072         blkdev->gd->private_data = blkdev;
1073         blkdev->gd->driverfs_dev = &(blkdev->device_ctx->device);
1074         sprintf(blkdev->gd->disk_name, "hd%c", 'a' + devnum);
1075
1076         blkvsc_do_operation(blkdev, DO_INQUIRY);
1077         blkvsc_do_operation(blkdev, DO_CAPACITY);
1078
1079         set_capacity(blkdev->gd, blkdev->capacity * (blkdev->sector_size/512));
1080         blk_queue_logical_block_size(blkdev->gd->queue, blkdev->sector_size);
1081         /* go! */
1082         add_disk(blkdev->gd);
1083
1084         DPRINT_INFO(BLKVSC_DRV, "%s added!! capacity %lu sector_size %d",
1085                     blkdev->gd->disk_name, (unsigned long)blkdev->capacity,
1086                     blkdev->sector_size);
1087
1088         return ret;
1089
1090 Remove:
1091         storvsc_drv_obj->base.dev_rm(device_obj);
1092
1093 Cleanup:
1094         if (blkdev) {
1095                 if (blkdev->request_pool) {
1096                         kmem_cache_destroy(blkdev->request_pool);
1097                         blkdev->request_pool = NULL;
1098                 }
1099                 kfree(blkdev);
1100                 blkdev = NULL;
1101         }
1102
1103         return ret;
1104 }
1105
1106 static void blkvsc_request_completion(struct hv_storvsc_request *request)
1107 {
1108         struct blkvsc_request *blkvsc_req =
1109                         (struct blkvsc_request *)request->context;
1110         struct block_device_context *blkdev =
1111                         (struct block_device_context *)blkvsc_req->dev;
1112         unsigned long flags;
1113         struct blkvsc_request *comp_req, *tmp;
1114         struct vmscsi_request *vm_srb;
1115
1116
1117         spin_lock_irqsave(&blkdev->lock, flags);
1118
1119         blkdev->num_outstanding_reqs--;
1120         blkvsc_req->group->outstanding--;
1121
1122         /*
1123          * Only start processing when all the blkvsc_reqs are
1124          * completed. This guarantees no out-of-order blkvsc_req
1125          * completion when calling end_that_request_first()
1126          */
1127         if (blkvsc_req->group->outstanding == 0) {
1128                 list_for_each_entry_safe(comp_req, tmp,
1129                                          &blkvsc_req->group->blkvsc_req_list,
1130                                          req_entry) {
1131
1132                         list_del(&comp_req->req_entry);
1133
1134                         vm_srb =
1135                         &comp_req->request.vstor_packet.vm_srb;
1136                         if (!__blk_end_request(comp_req->req,
1137                                 (!vm_srb->scsi_status ? 0 : -EIO),
1138                                 comp_req->sector_count * blkdev->sector_size)) {
1139                                 /*
1140                                  * All the sectors have been xferred ie the
1141                                  * request is done
1142                                  */
1143                                 kmem_cache_free(blkdev->request_pool,
1144                                                 comp_req->group);
1145                         }
1146
1147                         kmem_cache_free(blkdev->request_pool, comp_req);
1148                 }
1149
1150                 if (!blkdev->shutting_down) {
1151                         blkvsc_do_pending_reqs(blkdev);
1152                         blk_start_queue(blkdev->gd->queue);
1153                         blkvsc_request(blkdev->gd->queue);
1154                 }
1155         }
1156
1157         spin_unlock_irqrestore(&blkdev->lock, flags);
1158 }
1159
1160 static int __init blkvsc_init(void)
1161 {
1162         int ret;
1163
1164         BUILD_BUG_ON(sizeof(sector_t) != 8);
1165
1166         ret = blkvsc_drv_init();
1167
1168         return ret;
1169 }
1170
1171 static void __exit blkvsc_exit(void)
1172 {
1173         blkvsc_drv_exit();
1174 }
1175
1176 MODULE_LICENSE("GPL");
1177 MODULE_VERSION(HV_DRV_VERSION);
1178 MODULE_DESCRIPTION("Microsoft Hyper-V virtual block driver");
1179 module_init(blkvsc_init);
1180 module_exit(blkvsc_exit);