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/blkdev.h> 36 #include <scsi/scsi.h> 37 #include <scsi/scsi_cmnd.h> 38 #include <scsi/scsi_host.h> 39 #include <scsi/scsi_device.h> 40 #include <scsi/scsi_tcq.h> 41 #include <scsi/scsi_eh.h> 42 #include <scsi/scsi_devinfo.h> 43 #include <scsi/scsi_dbg.h> 44 45 /* 46 * All wire protocol details (storage protocol between the guest and the host) 47 * are consolidated here. 48 * 49 * Begin protocol definitions. 50 */ 51 52 /* 53 * Version history: 54 * V1 Beta: 0.1 55 * V1 RC < 2008/1/31: 1.0 56 * V1 RC > 2008/1/31: 2.0 57 * Win7: 4.2 58 * Win8: 5.1 59 */ 60 61 62 #define VMSTOR_WIN7_MAJOR 4 63 #define VMSTOR_WIN7_MINOR 2 64 65 #define VMSTOR_WIN8_MAJOR 5 66 #define VMSTOR_WIN8_MINOR 1 67 68 69 /* Packet structure describing virtual storage requests. */ 70 enum vstor_packet_operation { 71 VSTOR_OPERATION_COMPLETE_IO = 1, 72 VSTOR_OPERATION_REMOVE_DEVICE = 2, 73 VSTOR_OPERATION_EXECUTE_SRB = 3, 74 VSTOR_OPERATION_RESET_LUN = 4, 75 VSTOR_OPERATION_RESET_ADAPTER = 5, 76 VSTOR_OPERATION_RESET_BUS = 6, 77 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7, 78 VSTOR_OPERATION_END_INITIALIZATION = 8, 79 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9, 80 VSTOR_OPERATION_QUERY_PROPERTIES = 10, 81 VSTOR_OPERATION_ENUMERATE_BUS = 11, 82 VSTOR_OPERATION_FCHBA_DATA = 12, 83 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13, 84 VSTOR_OPERATION_MAXIMUM = 13 85 }; 86 87 /* 88 * WWN packet for Fibre Channel HBA 89 */ 90 91 struct hv_fc_wwn_packet { 92 bool primary_active; 93 u8 reserved1; 94 u8 reserved2; 95 u8 primary_port_wwn[8]; 96 u8 primary_node_wwn[8]; 97 u8 secondary_port_wwn[8]; 98 u8 secondary_node_wwn[8]; 99 }; 100 101 102 103 /* 104 * SRB Flag Bits 105 */ 106 107 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002 108 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004 109 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008 110 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010 111 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020 112 #define SRB_FLAGS_DATA_IN 0x00000040 113 #define SRB_FLAGS_DATA_OUT 0x00000080 114 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000 115 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT) 116 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100 117 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200 118 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400 119 120 /* 121 * This flag indicates the request is part of the workflow for processing a D3. 122 */ 123 #define SRB_FLAGS_D3_PROCESSING 0x00000800 124 #define SRB_FLAGS_IS_ACTIVE 0x00010000 125 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000 126 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000 127 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000 128 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000 129 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000 130 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000 131 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000 132 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000 133 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000 134 135 136 /* 137 * Platform neutral description of a scsi request - 138 * this remains the same across the write regardless of 32/64 bit 139 * note: it's patterned off the SCSI_PASS_THROUGH structure 140 */ 141 #define STORVSC_MAX_CMD_LEN 0x10 142 143 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14 144 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12 145 146 #define STORVSC_SENSE_BUFFER_SIZE 0x14 147 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14 148 149 /* 150 * Sense buffer size changed in win8; have a run-time 151 * variable to track the size we should use. 152 */ 153 static int sense_buffer_size; 154 155 /* 156 * The size of the vmscsi_request has changed in win8. The 157 * additional size is because of new elements added to the 158 * structure. These elements are valid only when we are talking 159 * to a win8 host. 160 * Track the correction to size we need to apply. 161 */ 162 163 static int vmscsi_size_delta; 164 static int vmstor_current_major; 165 static int vmstor_current_minor; 166 167 struct vmscsi_win8_extension { 168 /* 169 * The following were added in Windows 8 170 */ 171 u16 reserve; 172 u8 queue_tag; 173 u8 queue_action; 174 u32 srb_flags; 175 u32 time_out_value; 176 u32 queue_sort_ey; 177 } __packed; 178 179 struct vmscsi_request { 180 u16 length; 181 u8 srb_status; 182 u8 scsi_status; 183 184 u8 port_number; 185 u8 path_id; 186 u8 target_id; 187 u8 lun; 188 189 u8 cdb_length; 190 u8 sense_info_length; 191 u8 data_in; 192 u8 reserved; 193 194 u32 data_transfer_length; 195 196 union { 197 u8 cdb[STORVSC_MAX_CMD_LEN]; 198 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE]; 199 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING]; 200 }; 201 /* 202 * The following was added in win8. 203 */ 204 struct vmscsi_win8_extension win8_extension; 205 206 } __attribute((packed)); 207 208 209 /* 210 * This structure is sent during the intialization phase to get the different 211 * properties of the channel. 212 */ 213 214 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1 215 216 struct vmstorage_channel_properties { 217 u32 reserved; 218 u16 max_channel_cnt; 219 u16 reserved1; 220 221 u32 flags; 222 u32 max_transfer_bytes; 223 224 u64 reserved2; 225 } __packed; 226 227 /* This structure is sent during the storage protocol negotiations. */ 228 struct vmstorage_protocol_version { 229 /* Major (MSW) and minor (LSW) version numbers. */ 230 u16 major_minor; 231 232 /* 233 * Revision number is auto-incremented whenever this file is changed 234 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not 235 * definitely indicate incompatibility--but it does indicate mismatched 236 * builds. 237 * This is only used on the windows side. Just set it to 0. 238 */ 239 u16 revision; 240 } __packed; 241 242 /* Channel Property Flags */ 243 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1 244 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2 245 246 struct vstor_packet { 247 /* Requested operation type */ 248 enum vstor_packet_operation operation; 249 250 /* Flags - see below for values */ 251 u32 flags; 252 253 /* Status of the request returned from the server side. */ 254 u32 status; 255 256 /* Data payload area */ 257 union { 258 /* 259 * Structure used to forward SCSI commands from the 260 * client to the server. 261 */ 262 struct vmscsi_request vm_srb; 263 264 /* Structure used to query channel properties. */ 265 struct vmstorage_channel_properties storage_channel_properties; 266 267 /* Used during version negotiations. */ 268 struct vmstorage_protocol_version version; 269 270 /* Fibre channel address packet */ 271 struct hv_fc_wwn_packet wwn_packet; 272 273 /* Number of sub-channels to create */ 274 u16 sub_channel_count; 275 276 /* This will be the maximum of the union members */ 277 u8 buffer[0x34]; 278 }; 279 } __packed; 280 281 /* 282 * Packet Flags: 283 * 284 * This flag indicates that the server should send back a completion for this 285 * packet. 286 */ 287 288 #define REQUEST_COMPLETION_FLAG 0x1 289 290 /* Matches Windows-end */ 291 enum storvsc_request_type { 292 WRITE_TYPE = 0, 293 READ_TYPE, 294 UNKNOWN_TYPE, 295 }; 296 297 /* 298 * SRB status codes and masks; a subset of the codes used here. 299 */ 300 301 #define SRB_STATUS_AUTOSENSE_VALID 0x80 302 #define SRB_STATUS_INVALID_LUN 0x20 303 #define SRB_STATUS_SUCCESS 0x01 304 #define SRB_STATUS_ABORTED 0x02 305 #define SRB_STATUS_ERROR 0x04 306 307 /* 308 * This is the end of Protocol specific defines. 309 */ 310 311 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE); 312 313 module_param(storvsc_ringbuffer_size, int, S_IRUGO); 314 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)"); 315 316 /* 317 * Timeout in seconds for all devices managed by this driver. 318 */ 319 static int storvsc_timeout = 180; 320 321 static int msft_blist_flags = BLIST_TRY_VPD_PAGES; 322 323 #define STORVSC_MAX_IO_REQUESTS 200 324 325 static void storvsc_on_channel_callback(void *context); 326 327 #define STORVSC_MAX_LUNS_PER_TARGET 255 328 #define STORVSC_MAX_TARGETS 2 329 #define STORVSC_MAX_CHANNELS 8 330 331 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255 332 #define STORVSC_FC_MAX_TARGETS 128 333 #define STORVSC_FC_MAX_CHANNELS 8 334 335 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64 336 #define STORVSC_IDE_MAX_TARGETS 1 337 #define STORVSC_IDE_MAX_CHANNELS 1 338 339 struct storvsc_cmd_request { 340 struct scsi_cmnd *cmd; 341 342 unsigned int bounce_sgl_count; 343 struct scatterlist *bounce_sgl; 344 345 struct hv_device *device; 346 347 /* Synchronize the request/response if needed */ 348 struct completion wait_event; 349 350 struct hv_multipage_buffer data_buffer; 351 struct vstor_packet vstor_packet; 352 }; 353 354 355 /* A storvsc device is a device object that contains a vmbus channel */ 356 struct storvsc_device { 357 struct hv_device *device; 358 359 bool destroy; 360 bool drain_notify; 361 bool open_sub_channel; 362 atomic_t num_outstanding_req; 363 struct Scsi_Host *host; 364 365 wait_queue_head_t waiting_to_drain; 366 367 /* 368 * Each unique Port/Path/Target represents 1 channel ie scsi 369 * controller. In reality, the pathid, targetid is always 0 370 * and the port is set by us 371 */ 372 unsigned int port_number; 373 unsigned char path_id; 374 unsigned char target_id; 375 376 /* Used for vsc/vsp channel reset process */ 377 struct storvsc_cmd_request init_request; 378 struct storvsc_cmd_request reset_request; 379 }; 380 381 struct hv_host_device { 382 struct hv_device *dev; 383 unsigned int port; 384 unsigned char path; 385 unsigned char target; 386 }; 387 388 struct storvsc_scan_work { 389 struct work_struct work; 390 struct Scsi_Host *host; 391 uint lun; 392 }; 393 394 static void storvsc_device_scan(struct work_struct *work) 395 { 396 struct storvsc_scan_work *wrk; 397 uint lun; 398 struct scsi_device *sdev; 399 400 wrk = container_of(work, struct storvsc_scan_work, work); 401 lun = wrk->lun; 402 403 sdev = scsi_device_lookup(wrk->host, 0, 0, lun); 404 if (!sdev) 405 goto done; 406 scsi_rescan_device(&sdev->sdev_gendev); 407 scsi_device_put(sdev); 408 409 done: 410 kfree(wrk); 411 } 412 413 static void storvsc_host_scan(struct work_struct *work) 414 { 415 struct storvsc_scan_work *wrk; 416 struct Scsi_Host *host; 417 struct scsi_device *sdev; 418 unsigned long flags; 419 420 wrk = container_of(work, struct storvsc_scan_work, work); 421 host = wrk->host; 422 423 /* 424 * Before scanning the host, first check to see if any of the 425 * currrently known devices have been hot removed. We issue a 426 * "unit ready" command against all currently known devices. 427 * This I/O will result in an error for devices that have been 428 * removed. As part of handling the I/O error, we remove the device. 429 * 430 * When a LUN is added or removed, the host sends us a signal to 431 * scan the host. Thus we are forced to discover the LUNs that 432 * may have been removed this way. 433 */ 434 mutex_lock(&host->scan_mutex); 435 spin_lock_irqsave(host->host_lock, flags); 436 list_for_each_entry(sdev, &host->__devices, siblings) { 437 spin_unlock_irqrestore(host->host_lock, flags); 438 scsi_test_unit_ready(sdev, 1, 1, NULL); 439 spin_lock_irqsave(host->host_lock, flags); 440 continue; 441 } 442 spin_unlock_irqrestore(host->host_lock, flags); 443 mutex_unlock(&host->scan_mutex); 444 /* 445 * Now scan the host to discover LUNs that may have been added. 446 */ 447 scsi_scan_host(host); 448 449 kfree(wrk); 450 } 451 452 static void storvsc_remove_lun(struct work_struct *work) 453 { 454 struct storvsc_scan_work *wrk; 455 struct scsi_device *sdev; 456 457 wrk = container_of(work, struct storvsc_scan_work, work); 458 if (!scsi_host_get(wrk->host)) 459 goto done; 460 461 sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun); 462 463 if (sdev) { 464 scsi_remove_device(sdev); 465 scsi_device_put(sdev); 466 } 467 scsi_host_put(wrk->host); 468 469 done: 470 kfree(wrk); 471 } 472 473 /* 474 * Major/minor macros. Minor version is in LSB, meaning that earlier flat 475 * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). 476 */ 477 478 static inline u16 storvsc_get_version(u8 major, u8 minor) 479 { 480 u16 version; 481 482 version = ((major << 8) | minor); 483 return version; 484 } 485 486 /* 487 * We can get incoming messages from the host that are not in response to 488 * messages that we have sent out. An example of this would be messages 489 * received by the guest to notify dynamic addition/removal of LUNs. To 490 * deal with potential race conditions where the driver may be in the 491 * midst of being unloaded when we might receive an unsolicited message 492 * from the host, we have implemented a mechanism to gurantee sequential 493 * consistency: 494 * 495 * 1) Once the device is marked as being destroyed, we will fail all 496 * outgoing messages. 497 * 2) We permit incoming messages when the device is being destroyed, 498 * only to properly account for messages already sent out. 499 */ 500 501 static inline struct storvsc_device *get_out_stor_device( 502 struct hv_device *device) 503 { 504 struct storvsc_device *stor_device; 505 506 stor_device = hv_get_drvdata(device); 507 508 if (stor_device && stor_device->destroy) 509 stor_device = NULL; 510 511 return stor_device; 512 } 513 514 515 static inline void storvsc_wait_to_drain(struct storvsc_device *dev) 516 { 517 dev->drain_notify = true; 518 wait_event(dev->waiting_to_drain, 519 atomic_read(&dev->num_outstanding_req) == 0); 520 dev->drain_notify = false; 521 } 522 523 static inline struct storvsc_device *get_in_stor_device( 524 struct hv_device *device) 525 { 526 struct storvsc_device *stor_device; 527 528 stor_device = hv_get_drvdata(device); 529 530 if (!stor_device) 531 goto get_in_err; 532 533 /* 534 * If the device is being destroyed; allow incoming 535 * traffic only to cleanup outstanding requests. 536 */ 537 538 if (stor_device->destroy && 539 (atomic_read(&stor_device->num_outstanding_req) == 0)) 540 stor_device = NULL; 541 542 get_in_err: 543 return stor_device; 544 545 } 546 547 static void destroy_bounce_buffer(struct scatterlist *sgl, 548 unsigned int sg_count) 549 { 550 int i; 551 struct page *page_buf; 552 553 for (i = 0; i < sg_count; i++) { 554 page_buf = sg_page((&sgl[i])); 555 if (page_buf != NULL) 556 __free_page(page_buf); 557 } 558 559 kfree(sgl); 560 } 561 562 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count) 563 { 564 int i; 565 566 /* No need to check */ 567 if (sg_count < 2) 568 return -1; 569 570 /* We have at least 2 sg entries */ 571 for (i = 0; i < sg_count; i++) { 572 if (i == 0) { 573 /* make sure 1st one does not have hole */ 574 if (sgl[i].offset + sgl[i].length != PAGE_SIZE) 575 return i; 576 } else if (i == sg_count - 1) { 577 /* make sure last one does not have hole */ 578 if (sgl[i].offset != 0) 579 return i; 580 } else { 581 /* make sure no hole in the middle */ 582 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0) 583 return i; 584 } 585 } 586 return -1; 587 } 588 589 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl, 590 unsigned int sg_count, 591 unsigned int len, 592 int write) 593 { 594 int i; 595 int num_pages; 596 struct scatterlist *bounce_sgl; 597 struct page *page_buf; 598 unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE); 599 600 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT; 601 602 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC); 603 if (!bounce_sgl) 604 return NULL; 605 606 sg_init_table(bounce_sgl, num_pages); 607 for (i = 0; i < num_pages; i++) { 608 page_buf = alloc_page(GFP_ATOMIC); 609 if (!page_buf) 610 goto cleanup; 611 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0); 612 } 613 614 return bounce_sgl; 615 616 cleanup: 617 destroy_bounce_buffer(bounce_sgl, num_pages); 618 return NULL; 619 } 620 621 /* Disgusting wrapper functions */ 622 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx) 623 { 624 void *addr = kmap_atomic(sg_page(sgl + idx)); 625 return (unsigned long)addr; 626 } 627 628 static inline void sg_kunmap_atomic(unsigned long addr) 629 { 630 kunmap_atomic((void *)addr); 631 } 632 633 634 /* Assume the original sgl has enough room */ 635 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl, 636 struct scatterlist *bounce_sgl, 637 unsigned int orig_sgl_count, 638 unsigned int bounce_sgl_count) 639 { 640 int i; 641 int j = 0; 642 unsigned long src, dest; 643 unsigned int srclen, destlen, copylen; 644 unsigned int total_copied = 0; 645 unsigned long bounce_addr = 0; 646 unsigned long dest_addr = 0; 647 unsigned long flags; 648 649 local_irq_save(flags); 650 651 for (i = 0; i < orig_sgl_count; i++) { 652 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; 653 dest = dest_addr; 654 destlen = orig_sgl[i].length; 655 656 if (bounce_addr == 0) 657 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 658 659 while (destlen) { 660 src = bounce_addr + bounce_sgl[j].offset; 661 srclen = bounce_sgl[j].length - bounce_sgl[j].offset; 662 663 copylen = min(srclen, destlen); 664 memcpy((void *)dest, (void *)src, copylen); 665 666 total_copied += copylen; 667 bounce_sgl[j].offset += copylen; 668 destlen -= copylen; 669 dest += copylen; 670 671 if (bounce_sgl[j].offset == bounce_sgl[j].length) { 672 /* full */ 673 sg_kunmap_atomic(bounce_addr); 674 j++; 675 676 /* 677 * It is possible that the number of elements 678 * in the bounce buffer may not be equal to 679 * the number of elements in the original 680 * scatter list. Handle this correctly. 681 */ 682 683 if (j == bounce_sgl_count) { 684 /* 685 * We are done; cleanup and return. 686 */ 687 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); 688 local_irq_restore(flags); 689 return total_copied; 690 } 691 692 /* if we need to use another bounce buffer */ 693 if (destlen || i != orig_sgl_count - 1) 694 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 695 } else if (destlen == 0 && i == orig_sgl_count - 1) { 696 /* unmap the last bounce that is < PAGE_SIZE */ 697 sg_kunmap_atomic(bounce_addr); 698 } 699 } 700 701 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); 702 } 703 704 local_irq_restore(flags); 705 706 return total_copied; 707 } 708 709 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */ 710 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl, 711 struct scatterlist *bounce_sgl, 712 unsigned int orig_sgl_count) 713 { 714 int i; 715 int j = 0; 716 unsigned long src, dest; 717 unsigned int srclen, destlen, copylen; 718 unsigned int total_copied = 0; 719 unsigned long bounce_addr = 0; 720 unsigned long src_addr = 0; 721 unsigned long flags; 722 723 local_irq_save(flags); 724 725 for (i = 0; i < orig_sgl_count; i++) { 726 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; 727 src = src_addr; 728 srclen = orig_sgl[i].length; 729 730 if (bounce_addr == 0) 731 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 732 733 while (srclen) { 734 /* assume bounce offset always == 0 */ 735 dest = bounce_addr + bounce_sgl[j].length; 736 destlen = PAGE_SIZE - bounce_sgl[j].length; 737 738 copylen = min(srclen, destlen); 739 memcpy((void *)dest, (void *)src, copylen); 740 741 total_copied += copylen; 742 bounce_sgl[j].length += copylen; 743 srclen -= copylen; 744 src += copylen; 745 746 if (bounce_sgl[j].length == PAGE_SIZE) { 747 /* full..move to next entry */ 748 sg_kunmap_atomic(bounce_addr); 749 j++; 750 751 /* if we need to use another bounce buffer */ 752 if (srclen || i != orig_sgl_count - 1) 753 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 754 755 } else if (srclen == 0 && i == orig_sgl_count - 1) { 756 /* unmap the last bounce that is < PAGE_SIZE */ 757 sg_kunmap_atomic(bounce_addr); 758 } 759 } 760 761 sg_kunmap_atomic(src_addr - orig_sgl[i].offset); 762 } 763 764 local_irq_restore(flags); 765 766 return total_copied; 767 } 768 769 static void handle_sc_creation(struct vmbus_channel *new_sc) 770 { 771 struct hv_device *device = new_sc->primary_channel->device_obj; 772 struct storvsc_device *stor_device; 773 struct vmstorage_channel_properties props; 774 775 stor_device = get_out_stor_device(device); 776 if (!stor_device) 777 return; 778 779 if (stor_device->open_sub_channel == false) 780 return; 781 782 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 783 784 vmbus_open(new_sc, 785 storvsc_ringbuffer_size, 786 storvsc_ringbuffer_size, 787 (void *)&props, 788 sizeof(struct vmstorage_channel_properties), 789 storvsc_on_channel_callback, new_sc); 790 } 791 792 static void handle_multichannel_storage(struct hv_device *device, int max_chns) 793 { 794 struct storvsc_device *stor_device; 795 int num_cpus = num_online_cpus(); 796 int num_sc; 797 struct storvsc_cmd_request *request; 798 struct vstor_packet *vstor_packet; 799 int ret, t; 800 801 num_sc = ((max_chns > num_cpus) ? num_cpus : max_chns); 802 stor_device = get_out_stor_device(device); 803 if (!stor_device) 804 return; 805 806 request = &stor_device->init_request; 807 vstor_packet = &request->vstor_packet; 808 809 stor_device->open_sub_channel = true; 810 /* 811 * Establish a handler for dealing with subchannels. 812 */ 813 vmbus_set_sc_create_callback(device->channel, handle_sc_creation); 814 815 /* 816 * Check to see if sub-channels have already been created. This 817 * can happen when this driver is re-loaded after unloading. 818 */ 819 820 if (vmbus_are_subchannels_present(device->channel)) 821 return; 822 823 stor_device->open_sub_channel = false; 824 /* 825 * Request the host to create sub-channels. 826 */ 827 memset(request, 0, sizeof(struct storvsc_cmd_request)); 828 init_completion(&request->wait_event); 829 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS; 830 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 831 vstor_packet->sub_channel_count = num_sc; 832 833 ret = vmbus_sendpacket(device->channel, vstor_packet, 834 (sizeof(struct vstor_packet) - 835 vmscsi_size_delta), 836 (unsigned long)request, 837 VM_PKT_DATA_INBAND, 838 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 839 840 if (ret != 0) 841 return; 842 843 t = wait_for_completion_timeout(&request->wait_event, 10*HZ); 844 if (t == 0) 845 return; 846 847 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 848 vstor_packet->status != 0) 849 return; 850 851 /* 852 * Now that we created the sub-channels, invoke the check; this 853 * may trigger the callback. 854 */ 855 stor_device->open_sub_channel = true; 856 vmbus_are_subchannels_present(device->channel); 857 } 858 859 static int storvsc_channel_init(struct hv_device *device) 860 { 861 struct storvsc_device *stor_device; 862 struct storvsc_cmd_request *request; 863 struct vstor_packet *vstor_packet; 864 int ret, t; 865 int max_chns; 866 bool process_sub_channels = false; 867 868 stor_device = get_out_stor_device(device); 869 if (!stor_device) 870 return -ENODEV; 871 872 request = &stor_device->init_request; 873 vstor_packet = &request->vstor_packet; 874 875 /* 876 * Now, initiate the vsc/vsp initialization protocol on the open 877 * channel 878 */ 879 memset(request, 0, sizeof(struct storvsc_cmd_request)); 880 init_completion(&request->wait_event); 881 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; 882 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 883 884 ret = vmbus_sendpacket(device->channel, vstor_packet, 885 (sizeof(struct vstor_packet) - 886 vmscsi_size_delta), 887 (unsigned long)request, 888 VM_PKT_DATA_INBAND, 889 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 890 if (ret != 0) 891 goto cleanup; 892 893 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 894 if (t == 0) { 895 ret = -ETIMEDOUT; 896 goto cleanup; 897 } 898 899 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 900 vstor_packet->status != 0) 901 goto cleanup; 902 903 904 /* reuse the packet for version range supported */ 905 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 906 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; 907 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 908 909 vstor_packet->version.major_minor = 910 storvsc_get_version(vmstor_current_major, vmstor_current_minor); 911 912 /* 913 * The revision number is only used in Windows; set it to 0. 914 */ 915 vstor_packet->version.revision = 0; 916 917 ret = vmbus_sendpacket(device->channel, vstor_packet, 918 (sizeof(struct vstor_packet) - 919 vmscsi_size_delta), 920 (unsigned long)request, 921 VM_PKT_DATA_INBAND, 922 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 923 if (ret != 0) 924 goto cleanup; 925 926 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 927 if (t == 0) { 928 ret = -ETIMEDOUT; 929 goto cleanup; 930 } 931 932 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 933 vstor_packet->status != 0) 934 goto cleanup; 935 936 937 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 938 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; 939 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 940 941 ret = vmbus_sendpacket(device->channel, vstor_packet, 942 (sizeof(struct vstor_packet) - 943 vmscsi_size_delta), 944 (unsigned long)request, 945 VM_PKT_DATA_INBAND, 946 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 947 948 if (ret != 0) 949 goto cleanup; 950 951 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 952 if (t == 0) { 953 ret = -ETIMEDOUT; 954 goto cleanup; 955 } 956 957 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 958 vstor_packet->status != 0) 959 goto cleanup; 960 961 /* 962 * Check to see if multi-channel support is there. 963 * Hosts that implement protocol version of 5.1 and above 964 * support multi-channel. 965 */ 966 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt; 967 if ((vmbus_proto_version != VERSION_WIN7) && 968 (vmbus_proto_version != VERSION_WS2008)) { 969 if (vstor_packet->storage_channel_properties.flags & 970 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL) 971 process_sub_channels = true; 972 } 973 974 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 975 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; 976 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 977 978 ret = vmbus_sendpacket(device->channel, vstor_packet, 979 (sizeof(struct vstor_packet) - 980 vmscsi_size_delta), 981 (unsigned long)request, 982 VM_PKT_DATA_INBAND, 983 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 984 985 if (ret != 0) 986 goto cleanup; 987 988 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 989 if (t == 0) { 990 ret = -ETIMEDOUT; 991 goto cleanup; 992 } 993 994 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 995 vstor_packet->status != 0) 996 goto cleanup; 997 998 if (process_sub_channels) 999 handle_multichannel_storage(device, max_chns); 1000 1001 1002 cleanup: 1003 return ret; 1004 } 1005 1006 static void storvsc_handle_error(struct vmscsi_request *vm_srb, 1007 struct scsi_cmnd *scmnd, 1008 struct Scsi_Host *host, 1009 u8 asc, u8 ascq) 1010 { 1011 struct storvsc_scan_work *wrk; 1012 void (*process_err_fn)(struct work_struct *work); 1013 bool do_work = false; 1014 1015 switch (vm_srb->srb_status) { 1016 case SRB_STATUS_ERROR: 1017 /* 1018 * If there is an error; offline the device since all 1019 * error recovery strategies would have already been 1020 * deployed on the host side. However, if the command 1021 * were a pass-through command deal with it appropriately. 1022 */ 1023 switch (scmnd->cmnd[0]) { 1024 case ATA_16: 1025 case ATA_12: 1026 set_host_byte(scmnd, DID_PASSTHROUGH); 1027 break; 1028 /* 1029 * On Some Windows hosts TEST_UNIT_READY command can return 1030 * SRB_STATUS_ERROR, let the upper level code deal with it 1031 * based on the sense information. 1032 */ 1033 case TEST_UNIT_READY: 1034 break; 1035 default: 1036 set_host_byte(scmnd, DID_TARGET_FAILURE); 1037 } 1038 break; 1039 case SRB_STATUS_INVALID_LUN: 1040 do_work = true; 1041 process_err_fn = storvsc_remove_lun; 1042 break; 1043 case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID): 1044 if ((asc == 0x2a) && (ascq == 0x9)) { 1045 do_work = true; 1046 process_err_fn = storvsc_device_scan; 1047 /* 1048 * Retry the I/O that trigerred this. 1049 */ 1050 set_host_byte(scmnd, DID_REQUEUE); 1051 } 1052 break; 1053 } 1054 1055 if (!do_work) 1056 return; 1057 1058 /* 1059 * We need to schedule work to process this error; schedule it. 1060 */ 1061 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1062 if (!wrk) { 1063 set_host_byte(scmnd, DID_TARGET_FAILURE); 1064 return; 1065 } 1066 1067 wrk->host = host; 1068 wrk->lun = vm_srb->lun; 1069 INIT_WORK(&wrk->work, process_err_fn); 1070 schedule_work(&wrk->work); 1071 } 1072 1073 1074 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request) 1075 { 1076 struct scsi_cmnd *scmnd = cmd_request->cmd; 1077 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1078 struct scsi_sense_hdr sense_hdr; 1079 struct vmscsi_request *vm_srb; 1080 struct Scsi_Host *host; 1081 struct storvsc_device *stor_dev; 1082 struct hv_device *dev = host_dev->dev; 1083 1084 stor_dev = get_in_stor_device(dev); 1085 host = stor_dev->host; 1086 1087 vm_srb = &cmd_request->vstor_packet.vm_srb; 1088 if (cmd_request->bounce_sgl_count) { 1089 if (vm_srb->data_in == READ_TYPE) 1090 copy_from_bounce_buffer(scsi_sglist(scmnd), 1091 cmd_request->bounce_sgl, 1092 scsi_sg_count(scmnd), 1093 cmd_request->bounce_sgl_count); 1094 destroy_bounce_buffer(cmd_request->bounce_sgl, 1095 cmd_request->bounce_sgl_count); 1096 } 1097 1098 scmnd->result = vm_srb->scsi_status; 1099 1100 if (scmnd->result) { 1101 if (scsi_normalize_sense(scmnd->sense_buffer, 1102 SCSI_SENSE_BUFFERSIZE, &sense_hdr)) 1103 scsi_print_sense_hdr(scmnd->device, "storvsc", 1104 &sense_hdr); 1105 } 1106 1107 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) 1108 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, 1109 sense_hdr.ascq); 1110 1111 scsi_set_resid(scmnd, 1112 cmd_request->data_buffer.len - 1113 vm_srb->data_transfer_length); 1114 1115 scmnd->scsi_done(scmnd); 1116 } 1117 1118 static void storvsc_on_io_completion(struct hv_device *device, 1119 struct vstor_packet *vstor_packet, 1120 struct storvsc_cmd_request *request) 1121 { 1122 struct storvsc_device *stor_device; 1123 struct vstor_packet *stor_pkt; 1124 1125 stor_device = hv_get_drvdata(device); 1126 stor_pkt = &request->vstor_packet; 1127 1128 /* 1129 * The current SCSI handling on the host side does 1130 * not correctly handle: 1131 * INQUIRY command with page code parameter set to 0x80 1132 * MODE_SENSE command with cmd[2] == 0x1c 1133 * 1134 * Setup srb and scsi status so this won't be fatal. 1135 * We do this so we can distinguish truly fatal failues 1136 * (srb status == 0x4) and off-line the device in that case. 1137 */ 1138 1139 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || 1140 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { 1141 vstor_packet->vm_srb.scsi_status = 0; 1142 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; 1143 } 1144 1145 1146 /* Copy over the status...etc */ 1147 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; 1148 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; 1149 stor_pkt->vm_srb.sense_info_length = 1150 vstor_packet->vm_srb.sense_info_length; 1151 1152 1153 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { 1154 /* CHECK_CONDITION */ 1155 if (vstor_packet->vm_srb.srb_status & 1156 SRB_STATUS_AUTOSENSE_VALID) { 1157 /* autosense data available */ 1158 1159 memcpy(request->cmd->sense_buffer, 1160 vstor_packet->vm_srb.sense_data, 1161 vstor_packet->vm_srb.sense_info_length); 1162 1163 } 1164 } 1165 1166 stor_pkt->vm_srb.data_transfer_length = 1167 vstor_packet->vm_srb.data_transfer_length; 1168 1169 storvsc_command_completion(request); 1170 1171 if (atomic_dec_and_test(&stor_device->num_outstanding_req) && 1172 stor_device->drain_notify) 1173 wake_up(&stor_device->waiting_to_drain); 1174 1175 1176 } 1177 1178 static void storvsc_on_receive(struct hv_device *device, 1179 struct vstor_packet *vstor_packet, 1180 struct storvsc_cmd_request *request) 1181 { 1182 struct storvsc_scan_work *work; 1183 struct storvsc_device *stor_device; 1184 1185 switch (vstor_packet->operation) { 1186 case VSTOR_OPERATION_COMPLETE_IO: 1187 storvsc_on_io_completion(device, vstor_packet, request); 1188 break; 1189 1190 case VSTOR_OPERATION_REMOVE_DEVICE: 1191 case VSTOR_OPERATION_ENUMERATE_BUS: 1192 stor_device = get_in_stor_device(device); 1193 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1194 if (!work) 1195 return; 1196 1197 INIT_WORK(&work->work, storvsc_host_scan); 1198 work->host = stor_device->host; 1199 schedule_work(&work->work); 1200 break; 1201 1202 default: 1203 break; 1204 } 1205 } 1206 1207 static void storvsc_on_channel_callback(void *context) 1208 { 1209 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1210 struct hv_device *device; 1211 struct storvsc_device *stor_device; 1212 u32 bytes_recvd; 1213 u64 request_id; 1214 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)]; 1215 struct storvsc_cmd_request *request; 1216 int ret; 1217 1218 if (channel->primary_channel != NULL) 1219 device = channel->primary_channel->device_obj; 1220 else 1221 device = channel->device_obj; 1222 1223 stor_device = get_in_stor_device(device); 1224 if (!stor_device) 1225 return; 1226 1227 do { 1228 ret = vmbus_recvpacket(channel, packet, 1229 ALIGN((sizeof(struct vstor_packet) - 1230 vmscsi_size_delta), 8), 1231 &bytes_recvd, &request_id); 1232 if (ret == 0 && bytes_recvd > 0) { 1233 1234 request = (struct storvsc_cmd_request *) 1235 (unsigned long)request_id; 1236 1237 if ((request == &stor_device->init_request) || 1238 (request == &stor_device->reset_request)) { 1239 1240 memcpy(&request->vstor_packet, packet, 1241 (sizeof(struct vstor_packet) - 1242 vmscsi_size_delta)); 1243 complete(&request->wait_event); 1244 } else { 1245 storvsc_on_receive(device, 1246 (struct vstor_packet *)packet, 1247 request); 1248 } 1249 } else { 1250 break; 1251 } 1252 } while (1); 1253 1254 return; 1255 } 1256 1257 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size) 1258 { 1259 struct vmstorage_channel_properties props; 1260 int ret; 1261 1262 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 1263 1264 ret = vmbus_open(device->channel, 1265 ring_size, 1266 ring_size, 1267 (void *)&props, 1268 sizeof(struct vmstorage_channel_properties), 1269 storvsc_on_channel_callback, device->channel); 1270 1271 if (ret != 0) 1272 return ret; 1273 1274 ret = storvsc_channel_init(device); 1275 1276 return ret; 1277 } 1278 1279 static int storvsc_dev_remove(struct hv_device *device) 1280 { 1281 struct storvsc_device *stor_device; 1282 unsigned long flags; 1283 1284 stor_device = hv_get_drvdata(device); 1285 1286 spin_lock_irqsave(&device->channel->inbound_lock, flags); 1287 stor_device->destroy = true; 1288 spin_unlock_irqrestore(&device->channel->inbound_lock, flags); 1289 1290 /* 1291 * At this point, all outbound traffic should be disable. We 1292 * only allow inbound traffic (responses) to proceed so that 1293 * outstanding requests can be completed. 1294 */ 1295 1296 storvsc_wait_to_drain(stor_device); 1297 1298 /* 1299 * Since we have already drained, we don't need to busy wait 1300 * as was done in final_release_stor_device() 1301 * Note that we cannot set the ext pointer to NULL until 1302 * we have drained - to drain the outgoing packets, we need to 1303 * allow incoming packets. 1304 */ 1305 spin_lock_irqsave(&device->channel->inbound_lock, flags); 1306 hv_set_drvdata(device, NULL); 1307 spin_unlock_irqrestore(&device->channel->inbound_lock, flags); 1308 1309 /* Close the channel */ 1310 vmbus_close(device->channel); 1311 1312 kfree(stor_device); 1313 return 0; 1314 } 1315 1316 static int storvsc_do_io(struct hv_device *device, 1317 struct storvsc_cmd_request *request) 1318 { 1319 struct storvsc_device *stor_device; 1320 struct vstor_packet *vstor_packet; 1321 struct vmbus_channel *outgoing_channel; 1322 int ret = 0; 1323 1324 vstor_packet = &request->vstor_packet; 1325 stor_device = get_out_stor_device(device); 1326 1327 if (!stor_device) 1328 return -ENODEV; 1329 1330 1331 request->device = device; 1332 /* 1333 * Select an an appropriate channel to send the request out. 1334 */ 1335 1336 outgoing_channel = vmbus_get_outgoing_channel(device->channel); 1337 1338 1339 vstor_packet->flags |= REQUEST_COMPLETION_FLAG; 1340 1341 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) - 1342 vmscsi_size_delta); 1343 1344 1345 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1346 1347 1348 vstor_packet->vm_srb.data_transfer_length = 1349 request->data_buffer.len; 1350 1351 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; 1352 1353 if (request->data_buffer.len) { 1354 ret = vmbus_sendpacket_multipagebuffer(outgoing_channel, 1355 &request->data_buffer, 1356 vstor_packet, 1357 (sizeof(struct vstor_packet) - 1358 vmscsi_size_delta), 1359 (unsigned long)request); 1360 } else { 1361 ret = vmbus_sendpacket(device->channel, vstor_packet, 1362 (sizeof(struct vstor_packet) - 1363 vmscsi_size_delta), 1364 (unsigned long)request, 1365 VM_PKT_DATA_INBAND, 1366 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1367 } 1368 1369 if (ret != 0) 1370 return ret; 1371 1372 atomic_inc(&stor_device->num_outstanding_req); 1373 1374 return ret; 1375 } 1376 1377 static int storvsc_device_configure(struct scsi_device *sdevice) 1378 { 1379 scsi_change_queue_depth(sdevice, STORVSC_MAX_IO_REQUESTS); 1380 1381 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE); 1382 1383 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY); 1384 1385 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ)); 1386 1387 sdevice->no_write_same = 1; 1388 1389 /* 1390 * Add blist flags to permit the reading of the VPD pages even when 1391 * the target may claim SPC-2 compliance. MSFT targets currently 1392 * claim SPC-2 compliance while they implement post SPC-2 features. 1393 * With this patch we can correctly handle WRITE_SAME_16 issues. 1394 */ 1395 sdevice->sdev_bflags |= msft_blist_flags; 1396 1397 /* 1398 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3 1399 * if the device is a MSFT virtual device. 1400 */ 1401 if (!strncmp(sdevice->vendor, "Msft", 4)) { 1402 switch (vmbus_proto_version) { 1403 case VERSION_WIN8: 1404 case VERSION_WIN8_1: 1405 sdevice->scsi_level = SCSI_SPC_3; 1406 break; 1407 } 1408 } 1409 1410 return 0; 1411 } 1412 1413 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev, 1414 sector_t capacity, int *info) 1415 { 1416 sector_t nsect = capacity; 1417 sector_t cylinders = nsect; 1418 int heads, sectors_pt; 1419 1420 /* 1421 * We are making up these values; let us keep it simple. 1422 */ 1423 heads = 0xff; 1424 sectors_pt = 0x3f; /* Sectors per track */ 1425 sector_div(cylinders, heads * sectors_pt); 1426 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect) 1427 cylinders = 0xffff; 1428 1429 info[0] = heads; 1430 info[1] = sectors_pt; 1431 info[2] = (int)cylinders; 1432 1433 return 0; 1434 } 1435 1436 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd) 1437 { 1438 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1439 struct hv_device *device = host_dev->dev; 1440 1441 struct storvsc_device *stor_device; 1442 struct storvsc_cmd_request *request; 1443 struct vstor_packet *vstor_packet; 1444 int ret, t; 1445 1446 1447 stor_device = get_out_stor_device(device); 1448 if (!stor_device) 1449 return FAILED; 1450 1451 request = &stor_device->reset_request; 1452 vstor_packet = &request->vstor_packet; 1453 1454 init_completion(&request->wait_event); 1455 1456 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS; 1457 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 1458 vstor_packet->vm_srb.path_id = stor_device->path_id; 1459 1460 ret = vmbus_sendpacket(device->channel, vstor_packet, 1461 (sizeof(struct vstor_packet) - 1462 vmscsi_size_delta), 1463 (unsigned long)&stor_device->reset_request, 1464 VM_PKT_DATA_INBAND, 1465 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1466 if (ret != 0) 1467 return FAILED; 1468 1469 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 1470 if (t == 0) 1471 return TIMEOUT_ERROR; 1472 1473 1474 /* 1475 * At this point, all outstanding requests in the adapter 1476 * should have been flushed out and return to us 1477 * There is a potential race here where the host may be in 1478 * the process of responding when we return from here. 1479 * Just wait for all in-transit packets to be accounted for 1480 * before we return from here. 1481 */ 1482 storvsc_wait_to_drain(stor_device); 1483 1484 return SUCCESS; 1485 } 1486 1487 /* 1488 * The host guarantees to respond to each command, although I/O latencies might 1489 * be unbounded on Azure. Reset the timer unconditionally to give the host a 1490 * chance to perform EH. 1491 */ 1492 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd) 1493 { 1494 return BLK_EH_RESET_TIMER; 1495 } 1496 1497 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) 1498 { 1499 bool allowed = true; 1500 u8 scsi_op = scmnd->cmnd[0]; 1501 1502 switch (scsi_op) { 1503 /* the host does not handle WRITE_SAME, log accident usage */ 1504 case WRITE_SAME: 1505 /* 1506 * smartd sends this command and the host does not handle 1507 * this. So, don't send it. 1508 */ 1509 case SET_WINDOW: 1510 scmnd->result = ILLEGAL_REQUEST << 16; 1511 allowed = false; 1512 break; 1513 default: 1514 break; 1515 } 1516 return allowed; 1517 } 1518 1519 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) 1520 { 1521 int ret; 1522 struct hv_host_device *host_dev = shost_priv(host); 1523 struct hv_device *dev = host_dev->dev; 1524 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd); 1525 int i; 1526 struct scatterlist *sgl; 1527 unsigned int sg_count = 0; 1528 struct vmscsi_request *vm_srb; 1529 1530 if (vmstor_current_major <= VMSTOR_WIN8_MAJOR) { 1531 /* 1532 * On legacy hosts filter unimplemented commands. 1533 * Future hosts are expected to correctly handle 1534 * unsupported commands. Furthermore, it is 1535 * possible that some of the currently 1536 * unsupported commands maybe supported in 1537 * future versions of the host. 1538 */ 1539 if (!storvsc_scsi_cmd_ok(scmnd)) { 1540 scmnd->scsi_done(scmnd); 1541 return 0; 1542 } 1543 } 1544 1545 /* Setup the cmd request */ 1546 cmd_request->cmd = scmnd; 1547 1548 vm_srb = &cmd_request->vstor_packet.vm_srb; 1549 vm_srb->win8_extension.time_out_value = 60; 1550 1551 vm_srb->win8_extension.srb_flags |= 1552 (SRB_FLAGS_QUEUE_ACTION_ENABLE | 1553 SRB_FLAGS_DISABLE_SYNCH_TRANSFER); 1554 1555 /* Build the SRB */ 1556 switch (scmnd->sc_data_direction) { 1557 case DMA_TO_DEVICE: 1558 vm_srb->data_in = WRITE_TYPE; 1559 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT; 1560 break; 1561 case DMA_FROM_DEVICE: 1562 vm_srb->data_in = READ_TYPE; 1563 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN; 1564 break; 1565 default: 1566 vm_srb->data_in = UNKNOWN_TYPE; 1567 vm_srb->win8_extension.srb_flags |= (SRB_FLAGS_DATA_IN | 1568 SRB_FLAGS_DATA_OUT); 1569 break; 1570 } 1571 1572 1573 vm_srb->port_number = host_dev->port; 1574 vm_srb->path_id = scmnd->device->channel; 1575 vm_srb->target_id = scmnd->device->id; 1576 vm_srb->lun = scmnd->device->lun; 1577 1578 vm_srb->cdb_length = scmnd->cmd_len; 1579 1580 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); 1581 1582 cmd_request->data_buffer.len = scsi_bufflen(scmnd); 1583 if (scsi_sg_count(scmnd)) { 1584 sgl = (struct scatterlist *)scsi_sglist(scmnd); 1585 sg_count = scsi_sg_count(scmnd); 1586 1587 /* check if we need to bounce the sgl */ 1588 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) { 1589 cmd_request->bounce_sgl = 1590 create_bounce_buffer(sgl, scsi_sg_count(scmnd), 1591 scsi_bufflen(scmnd), 1592 vm_srb->data_in); 1593 if (!cmd_request->bounce_sgl) 1594 return SCSI_MLQUEUE_HOST_BUSY; 1595 1596 cmd_request->bounce_sgl_count = 1597 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >> 1598 PAGE_SHIFT; 1599 1600 if (vm_srb->data_in == WRITE_TYPE) 1601 copy_to_bounce_buffer(sgl, 1602 cmd_request->bounce_sgl, 1603 scsi_sg_count(scmnd)); 1604 1605 sgl = cmd_request->bounce_sgl; 1606 sg_count = cmd_request->bounce_sgl_count; 1607 } 1608 1609 cmd_request->data_buffer.offset = sgl[0].offset; 1610 1611 for (i = 0; i < sg_count; i++) 1612 cmd_request->data_buffer.pfn_array[i] = 1613 page_to_pfn(sg_page((&sgl[i]))); 1614 1615 } else if (scsi_sglist(scmnd)) { 1616 cmd_request->data_buffer.offset = 1617 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1); 1618 cmd_request->data_buffer.pfn_array[0] = 1619 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT; 1620 } 1621 1622 /* Invokes the vsc to start an IO */ 1623 ret = storvsc_do_io(dev, cmd_request); 1624 1625 if (ret == -EAGAIN) { 1626 /* no more space */ 1627 1628 if (cmd_request->bounce_sgl_count) 1629 destroy_bounce_buffer(cmd_request->bounce_sgl, 1630 cmd_request->bounce_sgl_count); 1631 1632 return SCSI_MLQUEUE_DEVICE_BUSY; 1633 } 1634 1635 return 0; 1636 } 1637 1638 static struct scsi_host_template scsi_driver = { 1639 .module = THIS_MODULE, 1640 .name = "storvsc_host_t", 1641 .cmd_size = sizeof(struct storvsc_cmd_request), 1642 .bios_param = storvsc_get_chs, 1643 .queuecommand = storvsc_queuecommand, 1644 .eh_host_reset_handler = storvsc_host_reset_handler, 1645 .proc_name = "storvsc_host", 1646 .eh_timed_out = storvsc_eh_timed_out, 1647 .slave_configure = storvsc_device_configure, 1648 .cmd_per_lun = 255, 1649 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS, 1650 .this_id = -1, 1651 /* no use setting to 0 since ll_blk_rw reset it to 1 */ 1652 /* currently 32 */ 1653 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT, 1654 .use_clustering = DISABLE_CLUSTERING, 1655 /* Make sure we dont get a sg segment crosses a page boundary */ 1656 .dma_boundary = PAGE_SIZE-1, 1657 .no_write_same = 1, 1658 }; 1659 1660 enum { 1661 SCSI_GUID, 1662 IDE_GUID, 1663 SFC_GUID, 1664 }; 1665 1666 static const struct hv_vmbus_device_id id_table[] = { 1667 /* SCSI guid */ 1668 { HV_SCSI_GUID, 1669 .driver_data = SCSI_GUID 1670 }, 1671 /* IDE guid */ 1672 { HV_IDE_GUID, 1673 .driver_data = IDE_GUID 1674 }, 1675 /* Fibre Channel GUID */ 1676 { 1677 HV_SYNTHFC_GUID, 1678 .driver_data = SFC_GUID 1679 }, 1680 { }, 1681 }; 1682 1683 MODULE_DEVICE_TABLE(vmbus, id_table); 1684 1685 static int storvsc_probe(struct hv_device *device, 1686 const struct hv_vmbus_device_id *dev_id) 1687 { 1688 int ret; 1689 struct Scsi_Host *host; 1690 struct hv_host_device *host_dev; 1691 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); 1692 int target = 0; 1693 struct storvsc_device *stor_device; 1694 int max_luns_per_target; 1695 int max_targets; 1696 int max_channels; 1697 1698 /* 1699 * Based on the windows host we are running on, 1700 * set state to properly communicate with the host. 1701 */ 1702 1703 switch (vmbus_proto_version) { 1704 case VERSION_WS2008: 1705 case VERSION_WIN7: 1706 sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE; 1707 vmscsi_size_delta = sizeof(struct vmscsi_win8_extension); 1708 vmstor_current_major = VMSTOR_WIN7_MAJOR; 1709 vmstor_current_minor = VMSTOR_WIN7_MINOR; 1710 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1711 max_targets = STORVSC_IDE_MAX_TARGETS; 1712 max_channels = STORVSC_IDE_MAX_CHANNELS; 1713 break; 1714 default: 1715 sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE; 1716 vmscsi_size_delta = 0; 1717 vmstor_current_major = VMSTOR_WIN8_MAJOR; 1718 vmstor_current_minor = VMSTOR_WIN8_MINOR; 1719 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET; 1720 max_targets = STORVSC_MAX_TARGETS; 1721 max_channels = STORVSC_MAX_CHANNELS; 1722 break; 1723 } 1724 1725 if (dev_id->driver_data == SFC_GUID) 1726 scsi_driver.can_queue = (STORVSC_MAX_IO_REQUESTS * 1727 STORVSC_FC_MAX_TARGETS); 1728 host = scsi_host_alloc(&scsi_driver, 1729 sizeof(struct hv_host_device)); 1730 if (!host) 1731 return -ENOMEM; 1732 1733 host_dev = shost_priv(host); 1734 memset(host_dev, 0, sizeof(struct hv_host_device)); 1735 1736 host_dev->port = host->host_no; 1737 host_dev->dev = device; 1738 1739 1740 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); 1741 if (!stor_device) { 1742 ret = -ENOMEM; 1743 goto err_out0; 1744 } 1745 1746 stor_device->destroy = false; 1747 stor_device->open_sub_channel = false; 1748 init_waitqueue_head(&stor_device->waiting_to_drain); 1749 stor_device->device = device; 1750 stor_device->host = host; 1751 hv_set_drvdata(device, stor_device); 1752 1753 stor_device->port_number = host->host_no; 1754 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size); 1755 if (ret) 1756 goto err_out1; 1757 1758 host_dev->path = stor_device->path_id; 1759 host_dev->target = stor_device->target_id; 1760 1761 switch (dev_id->driver_data) { 1762 case SFC_GUID: 1763 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET; 1764 host->max_id = STORVSC_FC_MAX_TARGETS; 1765 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1; 1766 break; 1767 1768 case SCSI_GUID: 1769 host->max_lun = max_luns_per_target; 1770 host->max_id = max_targets; 1771 host->max_channel = max_channels - 1; 1772 break; 1773 1774 default: 1775 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1776 host->max_id = STORVSC_IDE_MAX_TARGETS; 1777 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1; 1778 break; 1779 } 1780 /* max cmd length */ 1781 host->max_cmd_len = STORVSC_MAX_CMD_LEN; 1782 1783 /* Register the HBA and start the scsi bus scan */ 1784 ret = scsi_add_host(host, &device->device); 1785 if (ret != 0) 1786 goto err_out2; 1787 1788 if (!dev_is_ide) { 1789 scsi_scan_host(host); 1790 } else { 1791 target = (device->dev_instance.b[5] << 8 | 1792 device->dev_instance.b[4]); 1793 ret = scsi_add_device(host, 0, target, 0); 1794 if (ret) { 1795 scsi_remove_host(host); 1796 goto err_out2; 1797 } 1798 } 1799 return 0; 1800 1801 err_out2: 1802 /* 1803 * Once we have connected with the host, we would need to 1804 * to invoke storvsc_dev_remove() to rollback this state and 1805 * this call also frees up the stor_device; hence the jump around 1806 * err_out1 label. 1807 */ 1808 storvsc_dev_remove(device); 1809 goto err_out0; 1810 1811 err_out1: 1812 kfree(stor_device); 1813 1814 err_out0: 1815 scsi_host_put(host); 1816 return ret; 1817 } 1818 1819 static int storvsc_remove(struct hv_device *dev) 1820 { 1821 struct storvsc_device *stor_device = hv_get_drvdata(dev); 1822 struct Scsi_Host *host = stor_device->host; 1823 1824 scsi_remove_host(host); 1825 storvsc_dev_remove(dev); 1826 scsi_host_put(host); 1827 1828 return 0; 1829 } 1830 1831 static struct hv_driver storvsc_drv = { 1832 .name = KBUILD_MODNAME, 1833 .id_table = id_table, 1834 .probe = storvsc_probe, 1835 .remove = storvsc_remove, 1836 }; 1837 1838 static int __init storvsc_drv_init(void) 1839 { 1840 u32 max_outstanding_req_per_channel; 1841 1842 /* 1843 * Divide the ring buffer data size (which is 1 page less 1844 * than the ring buffer size since that page is reserved for 1845 * the ring buffer indices) by the max request size (which is 1846 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) 1847 */ 1848 max_outstanding_req_per_channel = 1849 ((storvsc_ringbuffer_size - PAGE_SIZE) / 1850 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + 1851 sizeof(struct vstor_packet) + sizeof(u64) - 1852 vmscsi_size_delta, 1853 sizeof(u64))); 1854 1855 if (max_outstanding_req_per_channel < 1856 STORVSC_MAX_IO_REQUESTS) 1857 return -EINVAL; 1858 1859 return vmbus_driver_register(&storvsc_drv); 1860 } 1861 1862 static void __exit storvsc_drv_exit(void) 1863 { 1864 vmbus_driver_unregister(&storvsc_drv); 1865 } 1866 1867 MODULE_LICENSE("GPL"); 1868 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); 1869 module_init(storvsc_drv_init); 1870 module_exit(storvsc_drv_exit); 1871