drivers/scsi/storvsc_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
  23 #include <linux/kernel.h>
  24 #include <linux/wait.h>
  25 #include <linux/sched.h>
  26 #include <linux/completion.h>
  27 #include <linux/string.h>
  28 #include <linux/mm.h>
  29 #include <linux/delay.h>
  30 #include <linux/init.h>
  31 #include <linux/slab.h>
  32 #include <linux/module.h>
  33 #include <linux/device.h>
  34 #include <linux/hyperv.h>
  35 #include <linux/mempool.h>
  36 #include <linux/blkdev.h>
  37 #include <scsi/scsi.h>
  38 #include <scsi/scsi_cmnd.h>
  39 #include <scsi/scsi_host.h>
  40 #include <scsi/scsi_device.h>
  41 #include <scsi/scsi_tcq.h>
  42 #include <scsi/scsi_eh.h>
  43 #include <scsi/scsi_devinfo.h>
  44 #include <scsi/scsi_dbg.h>
  45
  46 /*
  47  * All wire protocol details (storage protocol between the guest and the host)
  48  * are consolidated here.
  49  *
  50  * Begin protocol definitions.
  51  */
  52
  53 /*
  54  * Version history:
  55  * V1 Beta: 0.1
  56  * V1 RC < 2008/1/31: 1.0
  57  * V1 RC > 2008/1/31:  2.0
  58  * Win7: 4.2
  59  */
  60
  61 #define VMSTOR_CURRENT_MAJOR  4
  62 #define VMSTOR_CURRENT_MINOR  2
  63
  64
  65 /*  Packet structure describing virtual storage requests. */
  66 enum vstor_packet_operation {
  67         VSTOR_OPERATION_COMPLETE_IO             = 1,
  68         VSTOR_OPERATION_REMOVE_DEVICE           = 2,
  69         VSTOR_OPERATION_EXECUTE_SRB             = 3,
  70         VSTOR_OPERATION_RESET_LUN               = 4,
  71         VSTOR_OPERATION_RESET_ADAPTER           = 5,
  72         VSTOR_OPERATION_RESET_BUS               = 6,
  73         VSTOR_OPERATION_BEGIN_INITIALIZATION    = 7,
  74         VSTOR_OPERATION_END_INITIALIZATION      = 8,
  75         VSTOR_OPERATION_QUERY_PROTOCOL_VERSION  = 9,
  76         VSTOR_OPERATION_QUERY_PROPERTIES        = 10,
  77         VSTOR_OPERATION_ENUMERATE_BUS           = 11,
  78         VSTOR_OPERATION_MAXIMUM                 = 11
  79 };
  80
  81 /*
  82  * Platform neutral description of a scsi request -
  83  * this remains the same across the write regardless of 32/64 bit
  84  * note: it's patterned off the SCSI_PASS_THROUGH structure
  85  */
  86 #define STORVSC_MAX_CMD_LEN                     0x10
  87 #define STORVSC_SENSE_BUFFER_SIZE               0x12
  88 #define STORVSC_MAX_BUF_LEN_WITH_PADDING        0x14
  89
  90 struct vmscsi_request {
  91         u16 length;
  92         u8 srb_status;
  93         u8 scsi_status;
  94
  95         u8  port_number;
  96         u8  path_id;
  97         u8  target_id;
  98         u8  lun;
  99
 100         u8  cdb_length;
 101         u8  sense_info_length;
 102         u8  data_in;
 103         u8  reserved;
 104
 105         u32 data_transfer_length;
 106
 107         union {
 108                 u8 cdb[STORVSC_MAX_CMD_LEN];
 109                 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE];
 110                 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING];
 111         };
 112 } __attribute((packed));
 113
 114
 115 /*
 116  * This structure is sent during the intialization phase to get the different
 117  * properties of the channel.
 118  */
 119 struct vmstorage_channel_properties {
 120         u16 protocol_version;
 121         u8  path_id;
 122         u8 target_id;
 123
 124         /* Note: port number is only really known on the client side */
 125         u32  port_number;
 126         u32  flags;
 127         u32   max_transfer_bytes;
 128
 129         /*
 130          * This id is unique for each channel and will correspond with
 131          * vendor specific data in the inquiry data.
 132          */
 133
 134         u64  unique_id;
 135 } __packed;
 136
 137 /*  This structure is sent during the storage protocol negotiations. */
 138 struct vmstorage_protocol_version {
 139         /* Major (MSW) and minor (LSW) version numbers. */
 140         u16 major_minor;
 141
 142         /*
 143          * Revision number is auto-incremented whenever this file is changed
 144          * (See FILL_VMSTOR_REVISION macro above).  Mismatch does not
 145          * definitely indicate incompatibility--but it does indicate mismatched
 146          * builds.
 147          * This is only used on the windows side. Just set it to 0.
 148          */
 149         u16 revision;
 150 } __packed;
 151
 152 /* Channel Property Flags */
 153 #define STORAGE_CHANNEL_REMOVABLE_FLAG          0x1
 154 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG       0x2
 155
 156 struct vstor_packet {
 157         /* Requested operation type */
 158         enum vstor_packet_operation operation;
 159
 160         /*  Flags - see below for values */
 161         u32 flags;
 162
 163         /* Status of the request returned from the server side. */
 164         u32 status;
 165
 166         /* Data payload area */
 167         union {
 168                 /*
 169                  * Structure used to forward SCSI commands from the
 170                  * client to the server.
 171                  */
 172                 struct vmscsi_request vm_srb;
 173
 174                 /* Structure used to query channel properties. */
 175                 struct vmstorage_channel_properties storage_channel_properties;
 176
 177                 /* Used during version negotiations. */
 178                 struct vmstorage_protocol_version version;
 179         };
 180 } __packed;
 181
 182 /*
 183  * Packet Flags:
 184  *
 185  * This flag indicates that the server should send back a completion for this
 186  * packet.
 187  */
 188
 189 #define REQUEST_COMPLETION_FLAG 0x1
 190
 191 /* Matches Windows-end */
 192 enum storvsc_request_type {
 193         WRITE_TYPE = 0,
 194         READ_TYPE,
 195         UNKNOWN_TYPE,
 196 };
 197
 198 /*
 199  * SRB status codes and masks; a subset of the codes used here.
 200  */
 201
 202 #define SRB_STATUS_AUTOSENSE_VALID      0x80
 203 #define SRB_STATUS_INVALID_LUN  0x20
 204 #define SRB_STATUS_SUCCESS      0x01
 205 #define SRB_STATUS_ABORTED      0x02
 206 #define SRB_STATUS_ERROR        0x04
 207
 208 /*
 209  * This is the end of Protocol specific defines.
 210  */
 211
 212
 213 /*
 214  * We setup a mempool to allocate request structures for this driver
 215  * on a per-lun basis. The following define specifies the number of
 216  * elements in the pool.
 217  */
 218
 219 #define STORVSC_MIN_BUF_NR                              64
 220 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE);
 221
 222 module_param(storvsc_ringbuffer_size, int, S_IRUGO);
 223 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)");
 224
 225 #define STORVSC_MAX_IO_REQUESTS                         128
 226
 227 /*
 228  * In Hyper-V, each port/path/target maps to 1 scsi host adapter.  In
 229  * reality, the path/target is not used (ie always set to 0) so our
 230  * scsi host adapter essentially has 1 bus with 1 target that contains
 231  * up to 256 luns.
 232  */
 233 #define STORVSC_MAX_LUNS_PER_TARGET                     64
 234 #define STORVSC_MAX_TARGETS                             1
 235 #define STORVSC_MAX_CHANNELS                            1
 236
 237
 238
 239 struct storvsc_cmd_request {
 240         struct list_head entry;
 241         struct scsi_cmnd *cmd;
 242
 243         unsigned int bounce_sgl_count;
 244         struct scatterlist *bounce_sgl;
 245
 246         struct hv_device *device;
 247
 248         /* Synchronize the request/response if needed */
 249         struct completion wait_event;
 250
 251         unsigned char *sense_buffer;
 252         struct hv_multipage_buffer data_buffer;
 253         struct vstor_packet vstor_packet;
 254 };
 255
 256
 257 /* A storvsc device is a device object that contains a vmbus channel */
 258 struct storvsc_device {
 259         struct hv_device *device;
 260
 261         bool     destroy;
 262         bool     drain_notify;
 263         atomic_t num_outstanding_req;
 264         struct Scsi_Host *host;
 265
 266         wait_queue_head_t waiting_to_drain;
 267
 268         /*
 269          * Each unique Port/Path/Target represents 1 channel ie scsi
 270          * controller. In reality, the pathid, targetid is always 0
 271          * and the port is set by us
 272          */
 273         unsigned int port_number;
 274         unsigned char path_id;
 275         unsigned char target_id;
 276
 277         /* Used for vsc/vsp channel reset process */
 278         struct storvsc_cmd_request init_request;
 279         struct storvsc_cmd_request reset_request;
 280 };
 281
 282 struct stor_mem_pools {
 283         struct kmem_cache *request_pool;
 284         mempool_t *request_mempool;
 285 };
 286
 287 struct hv_host_device {
 288         struct hv_device *dev;
 289         unsigned int port;
 290         unsigned char path;
 291         unsigned char target;
 292 };
 293
 294 struct storvsc_scan_work {
 295         struct work_struct work;
 296         struct Scsi_Host *host;
 297         uint lun;
 298 };
 299
 300 static void storvsc_device_scan(struct work_struct *work)
 301 {
 302         struct storvsc_scan_work *wrk;
 303         uint lun;
 304         struct scsi_device *sdev;
 305
 306         wrk = container_of(work, struct storvsc_scan_work, work);
 307         lun = wrk->lun;
 308
 309         sdev = scsi_device_lookup(wrk->host, 0, 0, lun);
 310         if (!sdev)
 311                 goto done;
 312         scsi_rescan_device(&sdev->sdev_gendev);
 313         scsi_device_put(sdev);
 314
 315 done:
 316         kfree(wrk);
 317 }
 318
 319 static void storvsc_bus_scan(struct work_struct *work)
 320 {
 321         struct storvsc_scan_work *wrk;
 322         int id, order_id;
 323
 324         wrk = container_of(work, struct storvsc_scan_work, work);
 325         for (id = 0; id < wrk->host->max_id; ++id) {
 326                 if (wrk->host->reverse_ordering)
 327                         order_id = wrk->host->max_id - id - 1;
 328                 else
 329                         order_id = id;
 330
 331                 scsi_scan_target(&wrk->host->shost_gendev, 0,
 332                                 order_id, SCAN_WILD_CARD, 1);
 333         }
 334         kfree(wrk);
 335 }
 336
 337 static void storvsc_remove_lun(struct work_struct *work)
 338 {
 339         struct storvsc_scan_work *wrk;
 340         struct scsi_device *sdev;
 341
 342         wrk = container_of(work, struct storvsc_scan_work, work);
 343         if (!scsi_host_get(wrk->host))
 344                 goto done;
 345
 346         sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun);
 347
 348         if (sdev) {
 349                 scsi_remove_device(sdev);
 350                 scsi_device_put(sdev);
 351         }
 352         scsi_host_put(wrk->host);
 353
 354 done:
 355         kfree(wrk);
 356 }
 357
 358 /*
 359  * Major/minor macros.  Minor version is in LSB, meaning that earlier flat
 360  * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1).
 361  */
 362
 363 static inline u16 storvsc_get_version(u8 major, u8 minor)
 364 {
 365         u16 version;
 366
 367         version = ((major << 8) | minor);
 368         return version;
 369 }
 370
 371 /*
 372  * We can get incoming messages from the host that are not in response to
 373  * messages that we have sent out. An example of this would be messages
 374  * received by the guest to notify dynamic addition/removal of LUNs. To
 375  * deal with potential race conditions where the driver may be in the
 376  * midst of being unloaded when we might receive an unsolicited message
 377  * from the host, we have implemented a mechanism to gurantee sequential
 378  * consistency:
 379  *
 380  * 1) Once the device is marked as being destroyed, we will fail all
 381  *    outgoing messages.
 382  * 2) We permit incoming messages when the device is being destroyed,
 383  *    only to properly account for messages already sent out.
 384  */
 385
 386 static inline struct storvsc_device *get_out_stor_device(
 387                                         struct hv_device *device)
 388 {
 389         struct storvsc_device *stor_device;
 390
 391         stor_device = hv_get_drvdata(device);
 392
 393         if (stor_device && stor_device->destroy)
 394                 stor_device = NULL;
 395
 396         return stor_device;
 397 }
 398
 399
 400 static inline void storvsc_wait_to_drain(struct storvsc_device *dev)
 401 {
 402         dev->drain_notify = true;
 403         wait_event(dev->waiting_to_drain,
 404                    atomic_read(&dev->num_outstanding_req) == 0);
 405         dev->drain_notify = false;
 406 }
 407
 408 static inline struct storvsc_device *get_in_stor_device(
 409                                         struct hv_device *device)
 410 {
 411         struct storvsc_device *stor_device;
 412
 413         stor_device = hv_get_drvdata(device);
 414
 415         if (!stor_device)
 416                 goto get_in_err;
 417
 418         /*
 419          * If the device is being destroyed; allow incoming
 420          * traffic only to cleanup outstanding requests.
 421          */
 422
 423         if (stor_device->destroy  &&
 424                 (atomic_read(&stor_device->num_outstanding_req) == 0))
 425                 stor_device = NULL;
 426
 427 get_in_err:
 428         return stor_device;
 429
 430 }
 431
 432 static void destroy_bounce_buffer(struct scatterlist *sgl,
 433                                   unsigned int sg_count)
 434 {
 435         int i;
 436         struct page *page_buf;
 437
 438         for (i = 0; i < sg_count; i++) {
 439                 page_buf = sg_page((&sgl[i]));
 440                 if (page_buf != NULL)
 441                         __free_page(page_buf);
 442         }
 443
 444         kfree(sgl);
 445 }
 446
 447 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count)
 448 {
 449         int i;
 450
 451         /* No need to check */
 452         if (sg_count < 2)
 453                 return -1;
 454
 455         /* We have at least 2 sg entries */
 456         for (i = 0; i < sg_count; i++) {
 457                 if (i == 0) {
 458                         /* make sure 1st one does not have hole */
 459                         if (sgl[i].offset + sgl[i].length != PAGE_SIZE)
 460                                 return i;
 461                 } else if (i == sg_count - 1) {
 462                         /* make sure last one does not have hole */
 463                         if (sgl[i].offset != 0)
 464                                 return i;
 465                 } else {
 466                         /* make sure no hole in the middle */
 467                         if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0)
 468                                 return i;
 469                 }
 470         }
 471         return -1;
 472 }
 473
 474 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl,
 475                                                 unsigned int sg_count,
 476                                                 unsigned int len,
 477                                                 int write)
 478 {
 479         int i;
 480         int num_pages;
 481         struct scatterlist *bounce_sgl;
 482         struct page *page_buf;
 483         unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE);
 484
 485         num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT;
 486
 487         bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC);
 488         if (!bounce_sgl)
 489                 return NULL;
 490
 491         sg_init_table(bounce_sgl, num_pages);
 492         for (i = 0; i < num_pages; i++) {
 493                 page_buf = alloc_page(GFP_ATOMIC);
 494                 if (!page_buf)
 495                         goto cleanup;
 496                 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0);
 497         }
 498
 499         return bounce_sgl;
 500
 501 cleanup:
 502         destroy_bounce_buffer(bounce_sgl, num_pages);
 503         return NULL;
 504 }
 505
 506 /* Disgusting wrapper functions */
 507 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx)
 508 {
 509         void *addr = kmap_atomic(sg_page(sgl + idx));
 510         return (unsigned long)addr;
 511 }
 512
 513 static inline void sg_kunmap_atomic(unsigned long addr)
 514 {
 515         kunmap_atomic((void *)addr);
 516 }
 517
 518
 519 /* Assume the original sgl has enough room */
 520 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl,
 521                                             struct scatterlist *bounce_sgl,
 522                                             unsigned int orig_sgl_count,
 523                                             unsigned int bounce_sgl_count)
 524 {
 525         int i;
 526         int j = 0;
 527         unsigned long src, dest;
 528         unsigned int srclen, destlen, copylen;
 529         unsigned int total_copied = 0;
 530         unsigned long bounce_addr = 0;
 531         unsigned long dest_addr = 0;
 532         unsigned long flags;
 533
 534         local_irq_save(flags);
 535
 536         for (i = 0; i < orig_sgl_count; i++) {
 537                 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 538                 dest = dest_addr;
 539                 destlen = orig_sgl[i].length;
 540
 541                 if (bounce_addr == 0)
 542                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 543
 544                 while (destlen) {
 545                         src = bounce_addr + bounce_sgl[j].offset;
 546                         srclen = bounce_sgl[j].length - bounce_sgl[j].offset;
 547
 548                         copylen = min(srclen, destlen);
 549                         memcpy((void *)dest, (void *)src, copylen);
 550
 551                         total_copied += copylen;
 552                         bounce_sgl[j].offset += copylen;
 553                         destlen -= copylen;
 554                         dest += copylen;
 555
 556                         if (bounce_sgl[j].offset == bounce_sgl[j].length) {
 557                                 /* full */
 558                                 sg_kunmap_atomic(bounce_addr);
 559                                 j++;
 560
 561                                 /*
 562                                  * It is possible that the number of elements
 563                                  * in the bounce buffer may not be equal to
 564                                  * the number of elements in the original
 565                                  * scatter list. Handle this correctly.
 566                                  */
 567
 568                                 if (j == bounce_sgl_count) {
 569                                         /*
 570                                          * We are done; cleanup and return.
 571                                          */
 572                                         sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 573                                         local_irq_restore(flags);
 574                                         return total_copied;
 575                                 }
 576
 577                                 /* if we need to use another bounce buffer */
 578                                 if (destlen || i != orig_sgl_count - 1)
 579                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 580                         } else if (destlen == 0 && i == orig_sgl_count - 1) {
 581                                 /* unmap the last bounce that is < PAGE_SIZE */
 582                                 sg_kunmap_atomic(bounce_addr);
 583                         }
 584                 }
 585
 586                 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset);
 587         }
 588
 589         local_irq_restore(flags);
 590
 591         return total_copied;
 592 }
 593
 594 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */
 595 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl,
 596                                           struct scatterlist *bounce_sgl,
 597                                           unsigned int orig_sgl_count)
 598 {
 599         int i;
 600         int j = 0;
 601         unsigned long src, dest;
 602         unsigned int srclen, destlen, copylen;
 603         unsigned int total_copied = 0;
 604         unsigned long bounce_addr = 0;
 605         unsigned long src_addr = 0;
 606         unsigned long flags;
 607
 608         local_irq_save(flags);
 609
 610         for (i = 0; i < orig_sgl_count; i++) {
 611                 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset;
 612                 src = src_addr;
 613                 srclen = orig_sgl[i].length;
 614
 615                 if (bounce_addr == 0)
 616                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 617
 618                 while (srclen) {
 619                         /* assume bounce offset always == 0 */
 620                         dest = bounce_addr + bounce_sgl[j].length;
 621                         destlen = PAGE_SIZE - bounce_sgl[j].length;
 622
 623                         copylen = min(srclen, destlen);
 624                         memcpy((void *)dest, (void *)src, copylen);
 625
 626                         total_copied += copylen;
 627                         bounce_sgl[j].length += copylen;
 628                         srclen -= copylen;
 629                         src += copylen;
 630
 631                         if (bounce_sgl[j].length == PAGE_SIZE) {
 632                                 /* full..move to next entry */
 633                                 sg_kunmap_atomic(bounce_addr);
 634                                 j++;
 635
 636                                 /* if we need to use another bounce buffer */
 637                                 if (srclen || i != orig_sgl_count - 1)
 638                                         bounce_addr = sg_kmap_atomic(bounce_sgl,j);
 639
 640                         } else if (srclen == 0 && i == orig_sgl_count - 1) {
 641                                 /* unmap the last bounce that is < PAGE_SIZE */
 642                                 sg_kunmap_atomic(bounce_addr);
 643                         }
 644                 }
 645
 646                 sg_kunmap_atomic(src_addr - orig_sgl[i].offset);
 647         }
 648
 649         local_irq_restore(flags);
 650
 651         return total_copied;
 652 }
 653
 654 static int storvsc_channel_init(struct hv_device *device)
 655 {
 656         struct storvsc_device *stor_device;
 657         struct storvsc_cmd_request *request;
 658         struct vstor_packet *vstor_packet;
 659         int ret, t;
 660
 661         stor_device = get_out_stor_device(device);
 662         if (!stor_device)
 663                 return -ENODEV;
 664
 665         request = &stor_device->init_request;
 666         vstor_packet = &request->vstor_packet;
 667
 668         /*
 669          * Now, initiate the vsc/vsp initialization protocol on the open
 670          * channel
 671          */
 672         memset(request, 0, sizeof(struct storvsc_cmd_request));
 673         init_completion(&request->wait_event);
 674         vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION;
 675         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 676
 677         ret = vmbus_sendpacket(device->channel, vstor_packet,
 678                                sizeof(struct vstor_packet),
 679                                (unsigned long)request,
 680                                VM_PKT_DATA_INBAND,
 681                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 682         if (ret != 0)
 683                 goto cleanup;
 684
 685         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 686         if (t == 0) {
 687                 ret = -ETIMEDOUT;
 688                 goto cleanup;
 689         }
 690
 691         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 692             vstor_packet->status != 0)
 693                 goto cleanup;
 694
 695
 696         /* reuse the packet for version range supported */
 697         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 698         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION;
 699         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 700
 701         vstor_packet->version.major_minor =
 702                 storvsc_get_version(VMSTOR_CURRENT_MAJOR, VMSTOR_CURRENT_MINOR);
 703
 704         /*
 705          * The revision number is only used in Windows; set it to 0.
 706          */
 707         vstor_packet->version.revision = 0;
 708
 709         ret = vmbus_sendpacket(device->channel, vstor_packet,
 710                                sizeof(struct vstor_packet),
 711                                (unsigned long)request,
 712                                VM_PKT_DATA_INBAND,
 713                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 714         if (ret != 0)
 715                 goto cleanup;
 716
 717         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 718         if (t == 0) {
 719                 ret = -ETIMEDOUT;
 720                 goto cleanup;
 721         }
 722
 723         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 724             vstor_packet->status != 0)
 725                 goto cleanup;
 726
 727
 728         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 729         vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES;
 730         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 731         vstor_packet->storage_channel_properties.port_number =
 732                                         stor_device->port_number;
 733
 734         ret = vmbus_sendpacket(device->channel, vstor_packet,
 735                                sizeof(struct vstor_packet),
 736                                (unsigned long)request,
 737                                VM_PKT_DATA_INBAND,
 738                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 739
 740         if (ret != 0)
 741                 goto cleanup;
 742
 743         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 744         if (t == 0) {
 745                 ret = -ETIMEDOUT;
 746                 goto cleanup;
 747         }
 748
 749         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 750             vstor_packet->status != 0)
 751                 goto cleanup;
 752
 753         stor_device->path_id = vstor_packet->storage_channel_properties.path_id;
 754         stor_device->target_id
 755                 = vstor_packet->storage_channel_properties.target_id;
 756
 757         memset(vstor_packet, 0, sizeof(struct vstor_packet));
 758         vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION;
 759         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
 760
 761         ret = vmbus_sendpacket(device->channel, vstor_packet,
 762                                sizeof(struct vstor_packet),
 763                                (unsigned long)request,
 764                                VM_PKT_DATA_INBAND,
 765                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 766
 767         if (ret != 0)
 768                 goto cleanup;
 769
 770         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
 771         if (t == 0) {
 772                 ret = -ETIMEDOUT;
 773                 goto cleanup;
 774         }
 775
 776         if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO ||
 777             vstor_packet->status != 0)
 778                 goto cleanup;
 779
 780
 781 cleanup:
 782         return ret;
 783 }
 784
 785 static void storvsc_handle_error(struct vmscsi_request *vm_srb,
 786                                 struct scsi_cmnd *scmnd,
 787                                 struct Scsi_Host *host,
 788                                 u8 asc, u8 ascq)
 789 {
 790         struct storvsc_scan_work *wrk;
 791         void (*process_err_fn)(struct work_struct *work);
 792         bool do_work = false;
 793
 794         switch (vm_srb->srb_status) {
 795         case SRB_STATUS_ERROR:
 796                 /*
 797                  * If there is an error; offline the device since all
 798                  * error recovery strategies would have already been
 799                  * deployed on the host side. However, if the command
 800                  * were a pass-through command deal with it appropriately.
 801                  */
 802                 switch (scmnd->cmnd[0]) {
 803                 case ATA_16:
 804                 case ATA_12:
 805                         set_host_byte(scmnd, DID_PASSTHROUGH);
 806                         break;
 807                 /*
 808                  * On Some Windows hosts TEST_UNIT_READY command can return
 809                  * SRB_STATUS_ERROR, let the upper level code deal with it
 810                  * based on the sense information.
 811                  */
 812                 case TEST_UNIT_READY:
 813                         break;
 814                 default:
 815                         set_host_byte(scmnd, DID_TARGET_FAILURE);
 816                 }
 817                 break;
 818         case SRB_STATUS_INVALID_LUN:
 819                 do_work = true;
 820                 process_err_fn = storvsc_remove_lun;
 821                 break;
 822         case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID):
 823                 if ((asc == 0x2a) && (ascq == 0x9)) {
 824                         do_work = true;
 825                         process_err_fn = storvsc_device_scan;
 826                         /*
 827                          * Retry the I/O that trigerred this.
 828                          */
 829                         set_host_byte(scmnd, DID_REQUEUE);
 830                 }
 831                 break;
 832         }
 833
 834         if (!do_work)
 835                 return;
 836
 837         /*
 838          * We need to schedule work to process this error; schedule it.
 839          */
 840         wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
 841         if (!wrk) {
 842                 set_host_byte(scmnd, DID_TARGET_FAILURE);
 843                 return;
 844         }
 845
 846         wrk->host = host;
 847         wrk->lun = vm_srb->lun;
 848         INIT_WORK(&wrk->work, process_err_fn);
 849         schedule_work(&wrk->work);
 850 }
 851
 852
 853 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request)
 854 {
 855         struct scsi_cmnd *scmnd = cmd_request->cmd;
 856         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
 857         void (*scsi_done_fn)(struct scsi_cmnd *);
 858         struct scsi_sense_hdr sense_hdr;
 859         struct vmscsi_request *vm_srb;
 860         struct stor_mem_pools *memp = scmnd->device->hostdata;
 861         struct Scsi_Host *host;
 862         struct storvsc_device *stor_dev;
 863         struct hv_device *dev = host_dev->dev;
 864
 865         stor_dev = get_in_stor_device(dev);
 866         host = stor_dev->host;
 867
 868         vm_srb = &cmd_request->vstor_packet.vm_srb;
 869         if (cmd_request->bounce_sgl_count) {
 870                 if (vm_srb->data_in == READ_TYPE)
 871                         copy_from_bounce_buffer(scsi_sglist(scmnd),
 872                                         cmd_request->bounce_sgl,
 873                                         scsi_sg_count(scmnd),
 874                                         cmd_request->bounce_sgl_count);
 875                 destroy_bounce_buffer(cmd_request->bounce_sgl,
 876                                         cmd_request->bounce_sgl_count);
 877         }
 878
 879         scmnd->result = vm_srb->scsi_status;
 880
 881         if (scmnd->result) {
 882                 if (scsi_normalize_sense(scmnd->sense_buffer,
 883                                 SCSI_SENSE_BUFFERSIZE, &sense_hdr))
 884                         scsi_print_sense_hdr("storvsc", &sense_hdr);
 885         }
 886
 887         if (vm_srb->srb_status != SRB_STATUS_SUCCESS)
 888                 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc,
 889                                          sense_hdr.ascq);
 890
 891         scsi_set_resid(scmnd,
 892                 cmd_request->data_buffer.len -
 893                 vm_srb->data_transfer_length);
 894
 895         scsi_done_fn = scmnd->scsi_done;
 896
 897         scmnd->host_scribble = NULL;
 898         scmnd->scsi_done = NULL;
 899
 900         scsi_done_fn(scmnd);
 901
 902         mempool_free(cmd_request, memp->request_mempool);
 903 }
 904
 905 static void storvsc_on_io_completion(struct hv_device *device,
 906                                   struct vstor_packet *vstor_packet,
 907                                   struct storvsc_cmd_request *request)
 908 {
 909         struct storvsc_device *stor_device;
 910         struct vstor_packet *stor_pkt;
 911
 912         stor_device = hv_get_drvdata(device);
 913         stor_pkt = &request->vstor_packet;
 914
 915         /*
 916          * The current SCSI handling on the host side does
 917          * not correctly handle:
 918          * INQUIRY command with page code parameter set to 0x80
 919          * MODE_SENSE command with cmd[2] == 0x1c
 920          *
 921          * Setup srb and scsi status so this won't be fatal.
 922          * We do this so we can distinguish truly fatal failues
 923          * (srb status == 0x4) and off-line the device in that case.
 924          */
 925
 926         if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) ||
 927            (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) {
 928                 vstor_packet->vm_srb.scsi_status = 0;
 929                 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS;
 930         }
 931
 932
 933         /* Copy over the status...etc */
 934         stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status;
 935         stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status;
 936         stor_pkt->vm_srb.sense_info_length =
 937         vstor_packet->vm_srb.sense_info_length;
 938
 939         if (vstor_packet->vm_srb.scsi_status != 0 ||
 940                 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS){
 941                 dev_warn(&device->device,
 942                          "cmd 0x%x scsi status 0x%x srb status 0x%x\n",
 943                          stor_pkt->vm_srb.cdb[0],
 944                          vstor_packet->vm_srb.scsi_status,
 945                          vstor_packet->vm_srb.srb_status);
 946         }
 947
 948         if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) {
 949                 /* CHECK_CONDITION */
 950                 if (vstor_packet->vm_srb.srb_status &
 951                         SRB_STATUS_AUTOSENSE_VALID) {
 952                         /* autosense data available */
 953                         dev_warn(&device->device,
 954                                  "stor pkt %p autosense data valid - len %d\n",
 955                                  request,
 956                                  vstor_packet->vm_srb.sense_info_length);
 957
 958                         memcpy(request->sense_buffer,
 959                                vstor_packet->vm_srb.sense_data,
 960                                vstor_packet->vm_srb.sense_info_length);
 961
 962                 }
 963         }
 964
 965         stor_pkt->vm_srb.data_transfer_length =
 966         vstor_packet->vm_srb.data_transfer_length;
 967
 968         storvsc_command_completion(request);
 969
 970         if (atomic_dec_and_test(&stor_device->num_outstanding_req) &&
 971                 stor_device->drain_notify)
 972                 wake_up(&stor_device->waiting_to_drain);
 973
 974
 975 }
 976
 977 static void storvsc_on_receive(struct hv_device *device,
 978                              struct vstor_packet *vstor_packet,
 979                              struct storvsc_cmd_request *request)
 980 {
 981         struct storvsc_scan_work *work;
 982         struct storvsc_device *stor_device;
 983
 984         switch (vstor_packet->operation) {
 985         case VSTOR_OPERATION_COMPLETE_IO:
 986                 storvsc_on_io_completion(device, vstor_packet, request);
 987                 break;
 988
 989         case VSTOR_OPERATION_REMOVE_DEVICE:
 990         case VSTOR_OPERATION_ENUMERATE_BUS:
 991                 stor_device = get_in_stor_device(device);
 992                 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC);
 993                 if (!work)
 994                         return;
 995
 996                 INIT_WORK(&work->work, storvsc_bus_scan);
 997                 work->host = stor_device->host;
 998                 schedule_work(&work->work);
 999                 break;
1000
1001         default:
1002                 break;
1003         }
1004 }
1005
1006 static void storvsc_on_channel_callback(void *context)
1007 {
1008         struct hv_device *device = (struct hv_device *)context;
1009         struct storvsc_device *stor_device;
1010         u32 bytes_recvd;
1011         u64 request_id;
1012         unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)];
1013         struct storvsc_cmd_request *request;
1014         int ret;
1015
1016
1017         stor_device = get_in_stor_device(device);
1018         if (!stor_device)
1019                 return;
1020
1021         do {
1022                 ret = vmbus_recvpacket(device->channel, packet,
1023                                        ALIGN(sizeof(struct vstor_packet), 8),
1024                                        &bytes_recvd, &request_id);
1025                 if (ret == 0 && bytes_recvd > 0) {
1026
1027                         request = (struct storvsc_cmd_request *)
1028                                         (unsigned long)request_id;
1029
1030                         if ((request == &stor_device->init_request) ||
1031                             (request == &stor_device->reset_request)) {
1032
1033                                 memcpy(&request->vstor_packet, packet,
1034                                        sizeof(struct vstor_packet));
1035                                 complete(&request->wait_event);
1036                         } else {
1037                                 storvsc_on_receive(device,
1038                                                 (struct vstor_packet *)packet,
1039                                                 request);
1040                         }
1041                 } else {
1042                         break;
1043                 }
1044         } while (1);
1045
1046         return;
1047 }
1048
1049 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size)
1050 {
1051         struct vmstorage_channel_properties props;
1052         int ret;
1053
1054         memset(&props, 0, sizeof(struct vmstorage_channel_properties));
1055
1056         ret = vmbus_open(device->channel,
1057                          ring_size,
1058                          ring_size,
1059                          (void *)&props,
1060                          sizeof(struct vmstorage_channel_properties),
1061                          storvsc_on_channel_callback, device);
1062
1063         if (ret != 0)
1064                 return ret;
1065
1066         ret = storvsc_channel_init(device);
1067
1068         return ret;
1069 }
1070
1071 static int storvsc_dev_remove(struct hv_device *device)
1072 {
1073         struct storvsc_device *stor_device;
1074         unsigned long flags;
1075
1076         stor_device = hv_get_drvdata(device);
1077
1078         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1079         stor_device->destroy = true;
1080         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1081
1082         /*
1083          * At this point, all outbound traffic should be disable. We
1084          * only allow inbound traffic (responses) to proceed so that
1085          * outstanding requests can be completed.
1086          */
1087
1088         storvsc_wait_to_drain(stor_device);
1089
1090         /*
1091          * Since we have already drained, we don't need to busy wait
1092          * as was done in final_release_stor_device()
1093          * Note that we cannot set the ext pointer to NULL until
1094          * we have drained - to drain the outgoing packets, we need to
1095          * allow incoming packets.
1096          */
1097         spin_lock_irqsave(&device->channel->inbound_lock, flags);
1098         hv_set_drvdata(device, NULL);
1099         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
1100
1101         /* Close the channel */
1102         vmbus_close(device->channel);
1103
1104         kfree(stor_device);
1105         return 0;
1106 }
1107
1108 static int storvsc_do_io(struct hv_device *device,
1109                               struct storvsc_cmd_request *request)
1110 {
1111         struct storvsc_device *stor_device;
1112         struct vstor_packet *vstor_packet;
1113         int ret = 0;
1114
1115         vstor_packet = &request->vstor_packet;
1116         stor_device = get_out_stor_device(device);
1117
1118         if (!stor_device)
1119                 return -ENODEV;
1120
1121
1122         request->device  = device;
1123
1124
1125         vstor_packet->flags |= REQUEST_COMPLETION_FLAG;
1126
1127         vstor_packet->vm_srb.length = sizeof(struct vmscsi_request);
1128
1129
1130         vstor_packet->vm_srb.sense_info_length = STORVSC_SENSE_BUFFER_SIZE;
1131
1132
1133         vstor_packet->vm_srb.data_transfer_length =
1134         request->data_buffer.len;
1135
1136         vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB;
1137
1138         if (request->data_buffer.len) {
1139                 ret = vmbus_sendpacket_multipagebuffer(device->channel,
1140                                 &request->data_buffer,
1141                                 vstor_packet,
1142                                 sizeof(struct vstor_packet),
1143                                 (unsigned long)request);
1144         } else {
1145                 ret = vmbus_sendpacket(device->channel, vstor_packet,
1146                                sizeof(struct vstor_packet),
1147                                (unsigned long)request,
1148                                VM_PKT_DATA_INBAND,
1149                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1150         }
1151
1152         if (ret != 0)
1153                 return ret;
1154
1155         atomic_inc(&stor_device->num_outstanding_req);
1156
1157         return ret;
1158 }
1159
1160 static int storvsc_device_alloc(struct scsi_device *sdevice)
1161 {
1162         struct stor_mem_pools *memp;
1163         int number = STORVSC_MIN_BUF_NR;
1164
1165         memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL);
1166         if (!memp)
1167                 return -ENOMEM;
1168
1169         memp->request_pool =
1170                 kmem_cache_create(dev_name(&sdevice->sdev_dev),
1171                                 sizeof(struct storvsc_cmd_request), 0,
1172                                 SLAB_HWCACHE_ALIGN, NULL);
1173
1174         if (!memp->request_pool)
1175                 goto err0;
1176
1177         memp->request_mempool = mempool_create(number, mempool_alloc_slab,
1178                                                 mempool_free_slab,
1179                                                 memp->request_pool);
1180
1181         if (!memp->request_mempool)
1182                 goto err1;
1183
1184         sdevice->hostdata = memp;
1185
1186         return 0;
1187
1188 err1:
1189         kmem_cache_destroy(memp->request_pool);
1190
1191 err0:
1192         kfree(memp);
1193         return -ENOMEM;
1194 }
1195
1196 static void storvsc_device_destroy(struct scsi_device *sdevice)
1197 {
1198         struct stor_mem_pools *memp = sdevice->hostdata;
1199
1200         if (!memp)
1201                 return;
1202
1203         mempool_destroy(memp->request_mempool);
1204         kmem_cache_destroy(memp->request_pool);
1205         kfree(memp);
1206         sdevice->hostdata = NULL;
1207 }
1208
1209 static int storvsc_device_configure(struct scsi_device *sdevice)
1210 {
1211         scsi_adjust_queue_depth(sdevice, MSG_SIMPLE_TAG,
1212                                 STORVSC_MAX_IO_REQUESTS);
1213
1214         blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE);
1215
1216         blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY);
1217
1218         sdevice->no_write_same = 1;
1219
1220         return 0;
1221 }
1222
1223 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev,
1224                            sector_t capacity, int *info)
1225 {
1226         sector_t nsect = capacity;
1227         sector_t cylinders = nsect;
1228         int heads, sectors_pt;
1229
1230         /*
1231          * We are making up these values; let us keep it simple.
1232          */
1233         heads = 0xff;
1234         sectors_pt = 0x3f;      /* Sectors per track */
1235         sector_div(cylinders, heads * sectors_pt);
1236         if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect)
1237                 cylinders = 0xffff;
1238
1239         info[0] = heads;
1240         info[1] = sectors_pt;
1241         info[2] = (int)cylinders;
1242
1243         return 0;
1244 }
1245
1246 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd)
1247 {
1248         struct hv_host_device *host_dev = shost_priv(scmnd->device->host);
1249         struct hv_device *device = host_dev->dev;
1250
1251         struct storvsc_device *stor_device;
1252         struct storvsc_cmd_request *request;
1253         struct vstor_packet *vstor_packet;
1254         int ret, t;
1255
1256
1257         stor_device = get_out_stor_device(device);
1258         if (!stor_device)
1259                 return FAILED;
1260
1261         request = &stor_device->reset_request;
1262         vstor_packet = &request->vstor_packet;
1263
1264         init_completion(&request->wait_event);
1265
1266         vstor_packet->operation = VSTOR_OPERATION_RESET_BUS;
1267         vstor_packet->flags = REQUEST_COMPLETION_FLAG;
1268         vstor_packet->vm_srb.path_id = stor_device->path_id;
1269
1270         ret = vmbus_sendpacket(device->channel, vstor_packet,
1271                                sizeof(struct vstor_packet),
1272                                (unsigned long)&stor_device->reset_request,
1273                                VM_PKT_DATA_INBAND,
1274                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1275         if (ret != 0)
1276                 return FAILED;
1277
1278         t = wait_for_completion_timeout(&request->wait_event, 5*HZ);
1279         if (t == 0)
1280                 return TIMEOUT_ERROR;
1281
1282
1283         /*
1284          * At this point, all outstanding requests in the adapter
1285          * should have been flushed out and return to us
1286          * There is a potential race here where the host may be in
1287          * the process of responding when we return from here.
1288          * Just wait for all in-transit packets to be accounted for
1289          * before we return from here.
1290          */
1291         storvsc_wait_to_drain(stor_device);
1292
1293         return SUCCESS;
1294 }
1295
1296 /*
1297  * The host guarantees to respond to each command, although I/O latencies might
1298  * be unbounded on Azure.  Reset the timer unconditionally to give the host a
1299  * chance to perform EH.
1300  */
1301 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd)
1302 {
1303         return BLK_EH_RESET_TIMER;
1304 }
1305
1306 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd)
1307 {
1308         bool allowed = true;
1309         u8 scsi_op = scmnd->cmnd[0];
1310
1311         switch (scsi_op) {
1312         /* the host does not handle WRITE_SAME, log accident usage */
1313         case WRITE_SAME:
1314         /*
1315          * smartd sends this command and the host does not handle
1316          * this. So, don't send it.
1317          */
1318         case SET_WINDOW:
1319                 scmnd->result = ILLEGAL_REQUEST << 16;
1320                 allowed = false;
1321                 break;
1322         default:
1323                 break;
1324         }
1325         return allowed;
1326 }
1327
1328 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
1329 {
1330         int ret;
1331         struct hv_host_device *host_dev = shost_priv(host);
1332         struct hv_device *dev = host_dev->dev;
1333         struct storvsc_cmd_request *cmd_request;
1334         unsigned int request_size = 0;
1335         int i;
1336         struct scatterlist *sgl;
1337         unsigned int sg_count = 0;
1338         struct vmscsi_request *vm_srb;
1339         struct stor_mem_pools *memp = scmnd->device->hostdata;
1340
1341         if (!storvsc_scsi_cmd_ok(scmnd)) {
1342                 scmnd->scsi_done(scmnd);
1343                 return 0;
1344         }
1345
1346         request_size = sizeof(struct storvsc_cmd_request);
1347
1348         cmd_request = mempool_alloc(memp->request_mempool,
1349                                        GFP_ATOMIC);
1350
1351         /*
1352          * We might be invoked in an interrupt context; hence
1353          * mempool_alloc() can fail.
1354          */
1355         if (!cmd_request)
1356                 return SCSI_MLQUEUE_DEVICE_BUSY;
1357
1358         memset(cmd_request, 0, sizeof(struct storvsc_cmd_request));
1359
1360         /* Setup the cmd request */
1361         cmd_request->cmd = scmnd;
1362
1363         scmnd->host_scribble = (unsigned char *)cmd_request;
1364
1365         vm_srb = &cmd_request->vstor_packet.vm_srb;
1366
1367
1368         /* Build the SRB */
1369         switch (scmnd->sc_data_direction) {
1370         case DMA_TO_DEVICE:
1371                 vm_srb->data_in = WRITE_TYPE;
1372                 break;
1373         case DMA_FROM_DEVICE:
1374                 vm_srb->data_in = READ_TYPE;
1375                 break;
1376         default:
1377                 vm_srb->data_in = UNKNOWN_TYPE;
1378                 break;
1379         }
1380
1381
1382         vm_srb->port_number = host_dev->port;
1383         vm_srb->path_id = scmnd->device->channel;
1384         vm_srb->target_id = scmnd->device->id;
1385         vm_srb->lun = scmnd->device->lun;
1386
1387         vm_srb->cdb_length = scmnd->cmd_len;
1388
1389         memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length);
1390
1391         cmd_request->sense_buffer = scmnd->sense_buffer;
1392
1393
1394         cmd_request->data_buffer.len = scsi_bufflen(scmnd);
1395         if (scsi_sg_count(scmnd)) {
1396                 sgl = (struct scatterlist *)scsi_sglist(scmnd);
1397                 sg_count = scsi_sg_count(scmnd);
1398
1399                 /* check if we need to bounce the sgl */
1400                 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) {
1401                         cmd_request->bounce_sgl =
1402                                 create_bounce_buffer(sgl, scsi_sg_count(scmnd),
1403                                                      scsi_bufflen(scmnd),
1404                                                      vm_srb->data_in);
1405                         if (!cmd_request->bounce_sgl) {
1406                                 ret = SCSI_MLQUEUE_HOST_BUSY;
1407                                 goto queue_error;
1408                         }
1409
1410                         cmd_request->bounce_sgl_count =
1411                                 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >>
1412                                         PAGE_SHIFT;
1413
1414                         if (vm_srb->data_in == WRITE_TYPE)
1415                                 copy_to_bounce_buffer(sgl,
1416                                         cmd_request->bounce_sgl,
1417                                         scsi_sg_count(scmnd));
1418
1419                         sgl = cmd_request->bounce_sgl;
1420                         sg_count = cmd_request->bounce_sgl_count;
1421                 }
1422
1423                 cmd_request->data_buffer.offset = sgl[0].offset;
1424
1425                 for (i = 0; i < sg_count; i++)
1426                         cmd_request->data_buffer.pfn_array[i] =
1427                                 page_to_pfn(sg_page((&sgl[i])));
1428
1429         } else if (scsi_sglist(scmnd)) {
1430                 cmd_request->data_buffer.offset =
1431                         virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1);
1432                 cmd_request->data_buffer.pfn_array[0] =
1433                         virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT;
1434         }
1435
1436         /* Invokes the vsc to start an IO */
1437         ret = storvsc_do_io(dev, cmd_request);
1438
1439         if (ret == -EAGAIN) {
1440                 /* no more space */
1441
1442                 if (cmd_request->bounce_sgl_count) {
1443                         destroy_bounce_buffer(cmd_request->bounce_sgl,
1444                                         cmd_request->bounce_sgl_count);
1445
1446                         ret = SCSI_MLQUEUE_DEVICE_BUSY;
1447                         goto queue_error;
1448                 }
1449         }
1450
1451         return 0;
1452
1453 queue_error:
1454         mempool_free(cmd_request, memp->request_mempool);
1455         scmnd->host_scribble = NULL;
1456         return ret;
1457 }
1458
1459 static struct scsi_host_template scsi_driver = {
1460         .module =               THIS_MODULE,
1461         .name =                 "storvsc_host_t",
1462         .bios_param =           storvsc_get_chs,
1463         .queuecommand =         storvsc_queuecommand,
1464         .eh_host_reset_handler =        storvsc_host_reset_handler,
1465         .eh_timed_out =         storvsc_eh_timed_out,
1466         .slave_alloc =          storvsc_device_alloc,
1467         .slave_destroy =        storvsc_device_destroy,
1468         .slave_configure =      storvsc_device_configure,
1469         .cmd_per_lun =          1,
1470         /* 64 max_queue * 1 target */
1471         .can_queue =            STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS,
1472         .this_id =              -1,
1473         /* no use setting to 0 since ll_blk_rw reset it to 1 */
1474         /* currently 32 */
1475         .sg_tablesize =         MAX_MULTIPAGE_BUFFER_COUNT,
1476         .use_clustering =       DISABLE_CLUSTERING,
1477         /* Make sure we dont get a sg segment crosses a page boundary */
1478         .dma_boundary =         PAGE_SIZE-1,
1479         .no_write_same =        1,
1480 };
1481
1482 enum {
1483         SCSI_GUID,
1484         IDE_GUID,
1485 };
1486
1487 static const struct hv_vmbus_device_id id_table[] = {
1488         /* SCSI guid */
1489         { HV_SCSI_GUID,
1490           .driver_data = SCSI_GUID
1491         },
1492         /* IDE guid */
1493         { HV_IDE_GUID,
1494           .driver_data = IDE_GUID
1495         },
1496         { },
1497 };
1498
1499 MODULE_DEVICE_TABLE(vmbus, id_table);
1500
1501 static int storvsc_probe(struct hv_device *device,
1502                         const struct hv_vmbus_device_id *dev_id)
1503 {
1504         int ret;
1505         struct Scsi_Host *host;
1506         struct hv_host_device *host_dev;
1507         bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false);
1508         int target = 0;
1509         struct storvsc_device *stor_device;
1510
1511         host = scsi_host_alloc(&scsi_driver,
1512                                sizeof(struct hv_host_device));
1513         if (!host)
1514                 return -ENOMEM;
1515
1516         host_dev = shost_priv(host);
1517         memset(host_dev, 0, sizeof(struct hv_host_device));
1518
1519         host_dev->port = host->host_no;
1520         host_dev->dev = device;
1521
1522
1523         stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL);
1524         if (!stor_device) {
1525                 ret = -ENOMEM;
1526                 goto err_out0;
1527         }
1528
1529         stor_device->destroy = false;
1530         init_waitqueue_head(&stor_device->waiting_to_drain);
1531         stor_device->device = device;
1532         stor_device->host = host;
1533         hv_set_drvdata(device, stor_device);
1534
1535         stor_device->port_number = host->host_no;
1536         ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size);
1537         if (ret)
1538                 goto err_out1;
1539
1540         host_dev->path = stor_device->path_id;
1541         host_dev->target = stor_device->target_id;
1542
1543         /* max # of devices per target */
1544         host->max_lun = STORVSC_MAX_LUNS_PER_TARGET;
1545         /* max # of targets per channel */
1546         host->max_id = STORVSC_MAX_TARGETS;
1547         /* max # of channels */
1548         host->max_channel = STORVSC_MAX_CHANNELS - 1;
1549         /* max cmd length */
1550         host->max_cmd_len = STORVSC_MAX_CMD_LEN;
1551
1552         /* Register the HBA and start the scsi bus scan */
1553         ret = scsi_add_host(host, &device->device);
1554         if (ret != 0)
1555                 goto err_out2;
1556
1557         if (!dev_is_ide) {
1558                 scsi_scan_host(host);
1559         } else {
1560                 target = (device->dev_instance.b[5] << 8 |
1561                          device->dev_instance.b[4]);
1562                 ret = scsi_add_device(host, 0, target, 0);
1563                 if (ret) {
1564                         scsi_remove_host(host);
1565                         goto err_out2;
1566                 }
1567         }
1568         return 0;
1569
1570 err_out2:
1571         /*
1572          * Once we have connected with the host, we would need to
1573          * to invoke storvsc_dev_remove() to rollback this state and
1574          * this call also frees up the stor_device; hence the jump around
1575          * err_out1 label.
1576          */
1577         storvsc_dev_remove(device);
1578         goto err_out0;
1579
1580 err_out1:
1581         kfree(stor_device);
1582
1583 err_out0:
1584         scsi_host_put(host);
1585         return ret;
1586 }
1587
1588 static int storvsc_remove(struct hv_device *dev)
1589 {
1590         struct storvsc_device *stor_device = hv_get_drvdata(dev);
1591         struct Scsi_Host *host = stor_device->host;
1592
1593         scsi_remove_host(host);
1594         storvsc_dev_remove(dev);
1595         scsi_host_put(host);
1596
1597         return 0;
1598 }
1599
1600 static struct hv_driver storvsc_drv = {
1601         .name = KBUILD_MODNAME,
1602         .id_table = id_table,
1603         .probe = storvsc_probe,
1604         .remove = storvsc_remove,
1605 };
1606
1607 static int __init storvsc_drv_init(void)
1608 {
1609         u32 max_outstanding_req_per_channel;
1610
1611         /*
1612          * Divide the ring buffer data size (which is 1 page less
1613          * than the ring buffer size since that page is reserved for
1614          * the ring buffer indices) by the max request size (which is
1615          * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64)
1616          */
1617         max_outstanding_req_per_channel =
1618                 ((storvsc_ringbuffer_size - PAGE_SIZE) /
1619                 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET +
1620                 sizeof(struct vstor_packet) + sizeof(u64),
1621                 sizeof(u64)));
1622
1623         if (max_outstanding_req_per_channel <
1624             STORVSC_MAX_IO_REQUESTS)
1625                 return -EINVAL;
1626
1627         return vmbus_driver_register(&storvsc_drv);
1628 }
1629
1630 static void __exit storvsc_drv_exit(void)
1631 {
1632         vmbus_driver_unregister(&storvsc_drv);
1633 }
1634
1635 MODULE_LICENSE("GPL");
1636 MODULE_VERSION(HV_DRV_VERSION);
1637 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver");
1638 module_init(storvsc_drv_init);
1639 module_exit(storvsc_drv_exit);