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 * Win8: 5.1 60 */ 61 62 63 #define VMSTOR_WIN7_MAJOR 4 64 #define VMSTOR_WIN7_MINOR 2 65 66 #define VMSTOR_WIN8_MAJOR 5 67 #define VMSTOR_WIN8_MINOR 1 68 69 70 /* Packet structure describing virtual storage requests. */ 71 enum vstor_packet_operation { 72 VSTOR_OPERATION_COMPLETE_IO = 1, 73 VSTOR_OPERATION_REMOVE_DEVICE = 2, 74 VSTOR_OPERATION_EXECUTE_SRB = 3, 75 VSTOR_OPERATION_RESET_LUN = 4, 76 VSTOR_OPERATION_RESET_ADAPTER = 5, 77 VSTOR_OPERATION_RESET_BUS = 6, 78 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7, 79 VSTOR_OPERATION_END_INITIALIZATION = 8, 80 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9, 81 VSTOR_OPERATION_QUERY_PROPERTIES = 10, 82 VSTOR_OPERATION_ENUMERATE_BUS = 11, 83 VSTOR_OPERATION_FCHBA_DATA = 12, 84 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13, 85 VSTOR_OPERATION_MAXIMUM = 13 86 }; 87 88 /* 89 * WWN packet for Fibre Channel HBA 90 */ 91 92 struct hv_fc_wwn_packet { 93 bool primary_active; 94 u8 reserved1; 95 u8 reserved2; 96 u8 primary_port_wwn[8]; 97 u8 primary_node_wwn[8]; 98 u8 secondary_port_wwn[8]; 99 u8 secondary_node_wwn[8]; 100 }; 101 102 103 104 /* 105 * SRB Flag Bits 106 */ 107 108 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002 109 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004 110 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008 111 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010 112 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020 113 #define SRB_FLAGS_DATA_IN 0x00000040 114 #define SRB_FLAGS_DATA_OUT 0x00000080 115 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000 116 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT) 117 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100 118 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200 119 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400 120 121 /* 122 * This flag indicates the request is part of the workflow for processing a D3. 123 */ 124 #define SRB_FLAGS_D3_PROCESSING 0x00000800 125 #define SRB_FLAGS_IS_ACTIVE 0x00010000 126 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000 127 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000 128 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000 129 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000 130 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000 131 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000 132 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000 133 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000 134 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000 135 136 137 /* 138 * Platform neutral description of a scsi request - 139 * this remains the same across the write regardless of 32/64 bit 140 * note: it's patterned off the SCSI_PASS_THROUGH structure 141 */ 142 #define STORVSC_MAX_CMD_LEN 0x10 143 144 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14 145 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12 146 147 #define STORVSC_SENSE_BUFFER_SIZE 0x14 148 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14 149 150 /* 151 * Sense buffer size changed in win8; have a run-time 152 * variable to track the size we should use. 153 */ 154 static int sense_buffer_size; 155 156 /* 157 * The size of the vmscsi_request has changed in win8. The 158 * additional size is because of new elements added to the 159 * structure. These elements are valid only when we are talking 160 * to a win8 host. 161 * Track the correction to size we need to apply. 162 */ 163 164 static int vmscsi_size_delta; 165 static int vmstor_current_major; 166 static int vmstor_current_minor; 167 168 struct vmscsi_win8_extension { 169 /* 170 * The following were added in Windows 8 171 */ 172 u16 reserve; 173 u8 queue_tag; 174 u8 queue_action; 175 u32 srb_flags; 176 u32 time_out_value; 177 u32 queue_sort_ey; 178 } __packed; 179 180 struct vmscsi_request { 181 u16 length; 182 u8 srb_status; 183 u8 scsi_status; 184 185 u8 port_number; 186 u8 path_id; 187 u8 target_id; 188 u8 lun; 189 190 u8 cdb_length; 191 u8 sense_info_length; 192 u8 data_in; 193 u8 reserved; 194 195 u32 data_transfer_length; 196 197 union { 198 u8 cdb[STORVSC_MAX_CMD_LEN]; 199 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE]; 200 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING]; 201 }; 202 /* 203 * The following was added in win8. 204 */ 205 struct vmscsi_win8_extension win8_extension; 206 207 } __attribute((packed)); 208 209 210 /* 211 * This structure is sent during the intialization phase to get the different 212 * properties of the channel. 213 */ 214 215 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1 216 217 struct vmstorage_channel_properties { 218 u32 reserved; 219 u16 max_channel_cnt; 220 u16 reserved1; 221 222 u32 flags; 223 u32 max_transfer_bytes; 224 225 u64 reserved2; 226 } __packed; 227 228 /* This structure is sent during the storage protocol negotiations. */ 229 struct vmstorage_protocol_version { 230 /* Major (MSW) and minor (LSW) version numbers. */ 231 u16 major_minor; 232 233 /* 234 * Revision number is auto-incremented whenever this file is changed 235 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not 236 * definitely indicate incompatibility--but it does indicate mismatched 237 * builds. 238 * This is only used on the windows side. Just set it to 0. 239 */ 240 u16 revision; 241 } __packed; 242 243 /* Channel Property Flags */ 244 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1 245 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2 246 247 struct vstor_packet { 248 /* Requested operation type */ 249 enum vstor_packet_operation operation; 250 251 /* Flags - see below for values */ 252 u32 flags; 253 254 /* Status of the request returned from the server side. */ 255 u32 status; 256 257 /* Data payload area */ 258 union { 259 /* 260 * Structure used to forward SCSI commands from the 261 * client to the server. 262 */ 263 struct vmscsi_request vm_srb; 264 265 /* Structure used to query channel properties. */ 266 struct vmstorage_channel_properties storage_channel_properties; 267 268 /* Used during version negotiations. */ 269 struct vmstorage_protocol_version version; 270 271 /* Fibre channel address packet */ 272 struct hv_fc_wwn_packet wwn_packet; 273 274 /* Number of sub-channels to create */ 275 u16 sub_channel_count; 276 277 /* This will be the maximum of the union members */ 278 u8 buffer[0x34]; 279 }; 280 } __packed; 281 282 /* 283 * Packet Flags: 284 * 285 * This flag indicates that the server should send back a completion for this 286 * packet. 287 */ 288 289 #define REQUEST_COMPLETION_FLAG 0x1 290 291 /* Matches Windows-end */ 292 enum storvsc_request_type { 293 WRITE_TYPE = 0, 294 READ_TYPE, 295 UNKNOWN_TYPE, 296 }; 297 298 /* 299 * SRB status codes and masks; a subset of the codes used here. 300 */ 301 302 #define SRB_STATUS_AUTOSENSE_VALID 0x80 303 #define SRB_STATUS_INVALID_LUN 0x20 304 #define SRB_STATUS_SUCCESS 0x01 305 #define SRB_STATUS_ABORTED 0x02 306 #define SRB_STATUS_ERROR 0x04 307 308 /* 309 * This is the end of Protocol specific defines. 310 */ 311 312 313 /* 314 * We setup a mempool to allocate request structures for this driver 315 * on a per-lun basis. The following define specifies the number of 316 * elements in the pool. 317 */ 318 319 #define STORVSC_MIN_BUF_NR 64 320 static int storvsc_ringbuffer_size = (20 * PAGE_SIZE); 321 322 module_param(storvsc_ringbuffer_size, int, S_IRUGO); 323 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)"); 324 325 /* 326 * Timeout in seconds for all devices managed by this driver. 327 */ 328 static int storvsc_timeout = 180; 329 330 static int msft_blist_flags = BLIST_TRY_VPD_PAGES; 331 332 #define STORVSC_MAX_IO_REQUESTS 200 333 334 static void storvsc_on_channel_callback(void *context); 335 336 #define STORVSC_MAX_LUNS_PER_TARGET 255 337 #define STORVSC_MAX_TARGETS 2 338 #define STORVSC_MAX_CHANNELS 8 339 340 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255 341 #define STORVSC_FC_MAX_TARGETS 128 342 #define STORVSC_FC_MAX_CHANNELS 8 343 344 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64 345 #define STORVSC_IDE_MAX_TARGETS 1 346 #define STORVSC_IDE_MAX_CHANNELS 1 347 348 struct storvsc_cmd_request { 349 struct list_head entry; 350 struct scsi_cmnd *cmd; 351 352 unsigned int bounce_sgl_count; 353 struct scatterlist *bounce_sgl; 354 355 struct hv_device *device; 356 357 /* Synchronize the request/response if needed */ 358 struct completion wait_event; 359 360 unsigned char *sense_buffer; 361 struct hv_multipage_buffer data_buffer; 362 struct vstor_packet vstor_packet; 363 }; 364 365 366 /* A storvsc device is a device object that contains a vmbus channel */ 367 struct storvsc_device { 368 struct hv_device *device; 369 370 bool destroy; 371 bool drain_notify; 372 bool open_sub_channel; 373 atomic_t num_outstanding_req; 374 struct Scsi_Host *host; 375 376 wait_queue_head_t waiting_to_drain; 377 378 /* 379 * Each unique Port/Path/Target represents 1 channel ie scsi 380 * controller. In reality, the pathid, targetid is always 0 381 * and the port is set by us 382 */ 383 unsigned int port_number; 384 unsigned char path_id; 385 unsigned char target_id; 386 387 /* Used for vsc/vsp channel reset process */ 388 struct storvsc_cmd_request init_request; 389 struct storvsc_cmd_request reset_request; 390 }; 391 392 struct stor_mem_pools { 393 struct kmem_cache *request_pool; 394 mempool_t *request_mempool; 395 }; 396 397 struct hv_host_device { 398 struct hv_device *dev; 399 unsigned int port; 400 unsigned char path; 401 unsigned char target; 402 }; 403 404 struct storvsc_scan_work { 405 struct work_struct work; 406 struct Scsi_Host *host; 407 uint lun; 408 }; 409 410 static void storvsc_device_scan(struct work_struct *work) 411 { 412 struct storvsc_scan_work *wrk; 413 uint lun; 414 struct scsi_device *sdev; 415 416 wrk = container_of(work, struct storvsc_scan_work, work); 417 lun = wrk->lun; 418 419 sdev = scsi_device_lookup(wrk->host, 0, 0, lun); 420 if (!sdev) 421 goto done; 422 scsi_rescan_device(&sdev->sdev_gendev); 423 scsi_device_put(sdev); 424 425 done: 426 kfree(wrk); 427 } 428 429 static void storvsc_bus_scan(struct work_struct *work) 430 { 431 struct storvsc_scan_work *wrk; 432 int id, order_id; 433 434 wrk = container_of(work, struct storvsc_scan_work, work); 435 for (id = 0; id < wrk->host->max_id; ++id) { 436 if (wrk->host->reverse_ordering) 437 order_id = wrk->host->max_id - id - 1; 438 else 439 order_id = id; 440 441 scsi_scan_target(&wrk->host->shost_gendev, 0, 442 order_id, SCAN_WILD_CARD, 1); 443 } 444 kfree(wrk); 445 } 446 447 static void storvsc_remove_lun(struct work_struct *work) 448 { 449 struct storvsc_scan_work *wrk; 450 struct scsi_device *sdev; 451 452 wrk = container_of(work, struct storvsc_scan_work, work); 453 if (!scsi_host_get(wrk->host)) 454 goto done; 455 456 sdev = scsi_device_lookup(wrk->host, 0, 0, wrk->lun); 457 458 if (sdev) { 459 scsi_remove_device(sdev); 460 scsi_device_put(sdev); 461 } 462 scsi_host_put(wrk->host); 463 464 done: 465 kfree(wrk); 466 } 467 468 /* 469 * Major/minor macros. Minor version is in LSB, meaning that earlier flat 470 * version numbers will be interpreted as "0.x" (i.e., 1 becomes 0.1). 471 */ 472 473 static inline u16 storvsc_get_version(u8 major, u8 minor) 474 { 475 u16 version; 476 477 version = ((major << 8) | minor); 478 return version; 479 } 480 481 /* 482 * We can get incoming messages from the host that are not in response to 483 * messages that we have sent out. An example of this would be messages 484 * received by the guest to notify dynamic addition/removal of LUNs. To 485 * deal with potential race conditions where the driver may be in the 486 * midst of being unloaded when we might receive an unsolicited message 487 * from the host, we have implemented a mechanism to gurantee sequential 488 * consistency: 489 * 490 * 1) Once the device is marked as being destroyed, we will fail all 491 * outgoing messages. 492 * 2) We permit incoming messages when the device is being destroyed, 493 * only to properly account for messages already sent out. 494 */ 495 496 static inline struct storvsc_device *get_out_stor_device( 497 struct hv_device *device) 498 { 499 struct storvsc_device *stor_device; 500 501 stor_device = hv_get_drvdata(device); 502 503 if (stor_device && stor_device->destroy) 504 stor_device = NULL; 505 506 return stor_device; 507 } 508 509 510 static inline void storvsc_wait_to_drain(struct storvsc_device *dev) 511 { 512 dev->drain_notify = true; 513 wait_event(dev->waiting_to_drain, 514 atomic_read(&dev->num_outstanding_req) == 0); 515 dev->drain_notify = false; 516 } 517 518 static inline struct storvsc_device *get_in_stor_device( 519 struct hv_device *device) 520 { 521 struct storvsc_device *stor_device; 522 523 stor_device = hv_get_drvdata(device); 524 525 if (!stor_device) 526 goto get_in_err; 527 528 /* 529 * If the device is being destroyed; allow incoming 530 * traffic only to cleanup outstanding requests. 531 */ 532 533 if (stor_device->destroy && 534 (atomic_read(&stor_device->num_outstanding_req) == 0)) 535 stor_device = NULL; 536 537 get_in_err: 538 return stor_device; 539 540 } 541 542 static void destroy_bounce_buffer(struct scatterlist *sgl, 543 unsigned int sg_count) 544 { 545 int i; 546 struct page *page_buf; 547 548 for (i = 0; i < sg_count; i++) { 549 page_buf = sg_page((&sgl[i])); 550 if (page_buf != NULL) 551 __free_page(page_buf); 552 } 553 554 kfree(sgl); 555 } 556 557 static int do_bounce_buffer(struct scatterlist *sgl, unsigned int sg_count) 558 { 559 int i; 560 561 /* No need to check */ 562 if (sg_count < 2) 563 return -1; 564 565 /* We have at least 2 sg entries */ 566 for (i = 0; i < sg_count; i++) { 567 if (i == 0) { 568 /* make sure 1st one does not have hole */ 569 if (sgl[i].offset + sgl[i].length != PAGE_SIZE) 570 return i; 571 } else if (i == sg_count - 1) { 572 /* make sure last one does not have hole */ 573 if (sgl[i].offset != 0) 574 return i; 575 } else { 576 /* make sure no hole in the middle */ 577 if (sgl[i].length != PAGE_SIZE || sgl[i].offset != 0) 578 return i; 579 } 580 } 581 return -1; 582 } 583 584 static struct scatterlist *create_bounce_buffer(struct scatterlist *sgl, 585 unsigned int sg_count, 586 unsigned int len, 587 int write) 588 { 589 int i; 590 int num_pages; 591 struct scatterlist *bounce_sgl; 592 struct page *page_buf; 593 unsigned int buf_len = ((write == WRITE_TYPE) ? 0 : PAGE_SIZE); 594 595 num_pages = ALIGN(len, PAGE_SIZE) >> PAGE_SHIFT; 596 597 bounce_sgl = kcalloc(num_pages, sizeof(struct scatterlist), GFP_ATOMIC); 598 if (!bounce_sgl) 599 return NULL; 600 601 sg_init_table(bounce_sgl, num_pages); 602 for (i = 0; i < num_pages; i++) { 603 page_buf = alloc_page(GFP_ATOMIC); 604 if (!page_buf) 605 goto cleanup; 606 sg_set_page(&bounce_sgl[i], page_buf, buf_len, 0); 607 } 608 609 return bounce_sgl; 610 611 cleanup: 612 destroy_bounce_buffer(bounce_sgl, num_pages); 613 return NULL; 614 } 615 616 /* Disgusting wrapper functions */ 617 static inline unsigned long sg_kmap_atomic(struct scatterlist *sgl, int idx) 618 { 619 void *addr = kmap_atomic(sg_page(sgl + idx)); 620 return (unsigned long)addr; 621 } 622 623 static inline void sg_kunmap_atomic(unsigned long addr) 624 { 625 kunmap_atomic((void *)addr); 626 } 627 628 629 /* Assume the original sgl has enough room */ 630 static unsigned int copy_from_bounce_buffer(struct scatterlist *orig_sgl, 631 struct scatterlist *bounce_sgl, 632 unsigned int orig_sgl_count, 633 unsigned int bounce_sgl_count) 634 { 635 int i; 636 int j = 0; 637 unsigned long src, dest; 638 unsigned int srclen, destlen, copylen; 639 unsigned int total_copied = 0; 640 unsigned long bounce_addr = 0; 641 unsigned long dest_addr = 0; 642 unsigned long flags; 643 644 local_irq_save(flags); 645 646 for (i = 0; i < orig_sgl_count; i++) { 647 dest_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; 648 dest = dest_addr; 649 destlen = orig_sgl[i].length; 650 651 if (bounce_addr == 0) 652 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 653 654 while (destlen) { 655 src = bounce_addr + bounce_sgl[j].offset; 656 srclen = bounce_sgl[j].length - bounce_sgl[j].offset; 657 658 copylen = min(srclen, destlen); 659 memcpy((void *)dest, (void *)src, copylen); 660 661 total_copied += copylen; 662 bounce_sgl[j].offset += copylen; 663 destlen -= copylen; 664 dest += copylen; 665 666 if (bounce_sgl[j].offset == bounce_sgl[j].length) { 667 /* full */ 668 sg_kunmap_atomic(bounce_addr); 669 j++; 670 671 /* 672 * It is possible that the number of elements 673 * in the bounce buffer may not be equal to 674 * the number of elements in the original 675 * scatter list. Handle this correctly. 676 */ 677 678 if (j == bounce_sgl_count) { 679 /* 680 * We are done; cleanup and return. 681 */ 682 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); 683 local_irq_restore(flags); 684 return total_copied; 685 } 686 687 /* if we need to use another bounce buffer */ 688 if (destlen || i != orig_sgl_count - 1) 689 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 690 } else if (destlen == 0 && i == orig_sgl_count - 1) { 691 /* unmap the last bounce that is < PAGE_SIZE */ 692 sg_kunmap_atomic(bounce_addr); 693 } 694 } 695 696 sg_kunmap_atomic(dest_addr - orig_sgl[i].offset); 697 } 698 699 local_irq_restore(flags); 700 701 return total_copied; 702 } 703 704 /* Assume the bounce_sgl has enough room ie using the create_bounce_buffer() */ 705 static unsigned int copy_to_bounce_buffer(struct scatterlist *orig_sgl, 706 struct scatterlist *bounce_sgl, 707 unsigned int orig_sgl_count) 708 { 709 int i; 710 int j = 0; 711 unsigned long src, dest; 712 unsigned int srclen, destlen, copylen; 713 unsigned int total_copied = 0; 714 unsigned long bounce_addr = 0; 715 unsigned long src_addr = 0; 716 unsigned long flags; 717 718 local_irq_save(flags); 719 720 for (i = 0; i < orig_sgl_count; i++) { 721 src_addr = sg_kmap_atomic(orig_sgl,i) + orig_sgl[i].offset; 722 src = src_addr; 723 srclen = orig_sgl[i].length; 724 725 if (bounce_addr == 0) 726 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 727 728 while (srclen) { 729 /* assume bounce offset always == 0 */ 730 dest = bounce_addr + bounce_sgl[j].length; 731 destlen = PAGE_SIZE - bounce_sgl[j].length; 732 733 copylen = min(srclen, destlen); 734 memcpy((void *)dest, (void *)src, copylen); 735 736 total_copied += copylen; 737 bounce_sgl[j].length += copylen; 738 srclen -= copylen; 739 src += copylen; 740 741 if (bounce_sgl[j].length == PAGE_SIZE) { 742 /* full..move to next entry */ 743 sg_kunmap_atomic(bounce_addr); 744 j++; 745 746 /* if we need to use another bounce buffer */ 747 if (srclen || i != orig_sgl_count - 1) 748 bounce_addr = sg_kmap_atomic(bounce_sgl,j); 749 750 } else if (srclen == 0 && i == orig_sgl_count - 1) { 751 /* unmap the last bounce that is < PAGE_SIZE */ 752 sg_kunmap_atomic(bounce_addr); 753 } 754 } 755 756 sg_kunmap_atomic(src_addr - orig_sgl[i].offset); 757 } 758 759 local_irq_restore(flags); 760 761 return total_copied; 762 } 763 764 static void handle_sc_creation(struct vmbus_channel *new_sc) 765 { 766 struct hv_device *device = new_sc->primary_channel->device_obj; 767 struct storvsc_device *stor_device; 768 struct vmstorage_channel_properties props; 769 770 stor_device = get_out_stor_device(device); 771 if (!stor_device) 772 return; 773 774 if (stor_device->open_sub_channel == false) 775 return; 776 777 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 778 779 vmbus_open(new_sc, 780 storvsc_ringbuffer_size, 781 storvsc_ringbuffer_size, 782 (void *)&props, 783 sizeof(struct vmstorage_channel_properties), 784 storvsc_on_channel_callback, new_sc); 785 } 786 787 static void handle_multichannel_storage(struct hv_device *device, int max_chns) 788 { 789 struct storvsc_device *stor_device; 790 int num_cpus = num_online_cpus(); 791 int num_sc; 792 struct storvsc_cmd_request *request; 793 struct vstor_packet *vstor_packet; 794 int ret, t; 795 796 num_sc = ((max_chns > num_cpus) ? num_cpus : max_chns); 797 stor_device = get_out_stor_device(device); 798 if (!stor_device) 799 return; 800 801 request = &stor_device->init_request; 802 vstor_packet = &request->vstor_packet; 803 804 stor_device->open_sub_channel = true; 805 /* 806 * Establish a handler for dealing with subchannels. 807 */ 808 vmbus_set_sc_create_callback(device->channel, handle_sc_creation); 809 810 /* 811 * Check to see if sub-channels have already been created. This 812 * can happen when this driver is re-loaded after unloading. 813 */ 814 815 if (vmbus_are_subchannels_present(device->channel)) 816 return; 817 818 stor_device->open_sub_channel = false; 819 /* 820 * Request the host to create sub-channels. 821 */ 822 memset(request, 0, sizeof(struct storvsc_cmd_request)); 823 init_completion(&request->wait_event); 824 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS; 825 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 826 vstor_packet->sub_channel_count = num_sc; 827 828 ret = vmbus_sendpacket(device->channel, vstor_packet, 829 (sizeof(struct vstor_packet) - 830 vmscsi_size_delta), 831 (unsigned long)request, 832 VM_PKT_DATA_INBAND, 833 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 834 835 if (ret != 0) 836 return; 837 838 t = wait_for_completion_timeout(&request->wait_event, 10*HZ); 839 if (t == 0) 840 return; 841 842 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 843 vstor_packet->status != 0) 844 return; 845 846 /* 847 * Now that we created the sub-channels, invoke the check; this 848 * may trigger the callback. 849 */ 850 stor_device->open_sub_channel = true; 851 vmbus_are_subchannels_present(device->channel); 852 } 853 854 static int storvsc_channel_init(struct hv_device *device) 855 { 856 struct storvsc_device *stor_device; 857 struct storvsc_cmd_request *request; 858 struct vstor_packet *vstor_packet; 859 int ret, t; 860 int max_chns; 861 bool process_sub_channels = false; 862 863 stor_device = get_out_stor_device(device); 864 if (!stor_device) 865 return -ENODEV; 866 867 request = &stor_device->init_request; 868 vstor_packet = &request->vstor_packet; 869 870 /* 871 * Now, initiate the vsc/vsp initialization protocol on the open 872 * channel 873 */ 874 memset(request, 0, sizeof(struct storvsc_cmd_request)); 875 init_completion(&request->wait_event); 876 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; 877 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 878 879 ret = vmbus_sendpacket(device->channel, vstor_packet, 880 (sizeof(struct vstor_packet) - 881 vmscsi_size_delta), 882 (unsigned long)request, 883 VM_PKT_DATA_INBAND, 884 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 885 if (ret != 0) 886 goto cleanup; 887 888 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 889 if (t == 0) { 890 ret = -ETIMEDOUT; 891 goto cleanup; 892 } 893 894 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 895 vstor_packet->status != 0) 896 goto cleanup; 897 898 899 /* reuse the packet for version range supported */ 900 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 901 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; 902 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 903 904 vstor_packet->version.major_minor = 905 storvsc_get_version(vmstor_current_major, vmstor_current_minor); 906 907 /* 908 * The revision number is only used in Windows; set it to 0. 909 */ 910 vstor_packet->version.revision = 0; 911 912 ret = vmbus_sendpacket(device->channel, vstor_packet, 913 (sizeof(struct vstor_packet) - 914 vmscsi_size_delta), 915 (unsigned long)request, 916 VM_PKT_DATA_INBAND, 917 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 918 if (ret != 0) 919 goto cleanup; 920 921 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 922 if (t == 0) { 923 ret = -ETIMEDOUT; 924 goto cleanup; 925 } 926 927 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 928 vstor_packet->status != 0) 929 goto cleanup; 930 931 932 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 933 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; 934 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 935 936 ret = vmbus_sendpacket(device->channel, vstor_packet, 937 (sizeof(struct vstor_packet) - 938 vmscsi_size_delta), 939 (unsigned long)request, 940 VM_PKT_DATA_INBAND, 941 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 942 943 if (ret != 0) 944 goto cleanup; 945 946 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 947 if (t == 0) { 948 ret = -ETIMEDOUT; 949 goto cleanup; 950 } 951 952 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 953 vstor_packet->status != 0) 954 goto cleanup; 955 956 /* 957 * Check to see if multi-channel support is there. 958 * Hosts that implement protocol version of 5.1 and above 959 * support multi-channel. 960 */ 961 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt; 962 if ((vmbus_proto_version != VERSION_WIN7) && 963 (vmbus_proto_version != VERSION_WS2008)) { 964 if (vstor_packet->storage_channel_properties.flags & 965 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL) 966 process_sub_channels = true; 967 } 968 969 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 970 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; 971 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 972 973 ret = vmbus_sendpacket(device->channel, vstor_packet, 974 (sizeof(struct vstor_packet) - 975 vmscsi_size_delta), 976 (unsigned long)request, 977 VM_PKT_DATA_INBAND, 978 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 979 980 if (ret != 0) 981 goto cleanup; 982 983 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 984 if (t == 0) { 985 ret = -ETIMEDOUT; 986 goto cleanup; 987 } 988 989 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 990 vstor_packet->status != 0) 991 goto cleanup; 992 993 if (process_sub_channels) 994 handle_multichannel_storage(device, max_chns); 995 996 997 cleanup: 998 return ret; 999 } 1000 1001 static void storvsc_handle_error(struct vmscsi_request *vm_srb, 1002 struct scsi_cmnd *scmnd, 1003 struct Scsi_Host *host, 1004 u8 asc, u8 ascq) 1005 { 1006 struct storvsc_scan_work *wrk; 1007 void (*process_err_fn)(struct work_struct *work); 1008 bool do_work = false; 1009 1010 switch (vm_srb->srb_status) { 1011 case SRB_STATUS_ERROR: 1012 /* 1013 * If there is an error; offline the device since all 1014 * error recovery strategies would have already been 1015 * deployed on the host side. However, if the command 1016 * were a pass-through command deal with it appropriately. 1017 */ 1018 switch (scmnd->cmnd[0]) { 1019 case ATA_16: 1020 case ATA_12: 1021 set_host_byte(scmnd, DID_PASSTHROUGH); 1022 break; 1023 /* 1024 * On Some Windows hosts TEST_UNIT_READY command can return 1025 * SRB_STATUS_ERROR, let the upper level code deal with it 1026 * based on the sense information. 1027 */ 1028 case TEST_UNIT_READY: 1029 break; 1030 default: 1031 set_host_byte(scmnd, DID_TARGET_FAILURE); 1032 } 1033 break; 1034 case SRB_STATUS_INVALID_LUN: 1035 do_work = true; 1036 process_err_fn = storvsc_remove_lun; 1037 break; 1038 case (SRB_STATUS_ABORTED | SRB_STATUS_AUTOSENSE_VALID): 1039 if ((asc == 0x2a) && (ascq == 0x9)) { 1040 do_work = true; 1041 process_err_fn = storvsc_device_scan; 1042 /* 1043 * Retry the I/O that trigerred this. 1044 */ 1045 set_host_byte(scmnd, DID_REQUEUE); 1046 } 1047 break; 1048 } 1049 1050 if (!do_work) 1051 return; 1052 1053 /* 1054 * We need to schedule work to process this error; schedule it. 1055 */ 1056 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1057 if (!wrk) { 1058 set_host_byte(scmnd, DID_TARGET_FAILURE); 1059 return; 1060 } 1061 1062 wrk->host = host; 1063 wrk->lun = vm_srb->lun; 1064 INIT_WORK(&wrk->work, process_err_fn); 1065 schedule_work(&wrk->work); 1066 } 1067 1068 1069 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request) 1070 { 1071 struct scsi_cmnd *scmnd = cmd_request->cmd; 1072 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1073 void (*scsi_done_fn)(struct scsi_cmnd *); 1074 struct scsi_sense_hdr sense_hdr; 1075 struct vmscsi_request *vm_srb; 1076 struct stor_mem_pools *memp = scmnd->device->hostdata; 1077 struct Scsi_Host *host; 1078 struct storvsc_device *stor_dev; 1079 struct hv_device *dev = host_dev->dev; 1080 1081 stor_dev = get_in_stor_device(dev); 1082 host = stor_dev->host; 1083 1084 vm_srb = &cmd_request->vstor_packet.vm_srb; 1085 if (cmd_request->bounce_sgl_count) { 1086 if (vm_srb->data_in == READ_TYPE) 1087 copy_from_bounce_buffer(scsi_sglist(scmnd), 1088 cmd_request->bounce_sgl, 1089 scsi_sg_count(scmnd), 1090 cmd_request->bounce_sgl_count); 1091 destroy_bounce_buffer(cmd_request->bounce_sgl, 1092 cmd_request->bounce_sgl_count); 1093 } 1094 1095 scmnd->result = vm_srb->scsi_status; 1096 1097 if (scmnd->result) { 1098 if (scsi_normalize_sense(scmnd->sense_buffer, 1099 SCSI_SENSE_BUFFERSIZE, &sense_hdr)) 1100 scsi_print_sense_hdr(scmnd->device, "storvsc", 1101 &sense_hdr); 1102 } 1103 1104 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) 1105 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, 1106 sense_hdr.ascq); 1107 1108 scsi_set_resid(scmnd, 1109 cmd_request->data_buffer.len - 1110 vm_srb->data_transfer_length); 1111 1112 scsi_done_fn = scmnd->scsi_done; 1113 1114 scmnd->host_scribble = NULL; 1115 scmnd->scsi_done = NULL; 1116 1117 scsi_done_fn(scmnd); 1118 1119 mempool_free(cmd_request, memp->request_mempool); 1120 } 1121 1122 static void storvsc_on_io_completion(struct hv_device *device, 1123 struct vstor_packet *vstor_packet, 1124 struct storvsc_cmd_request *request) 1125 { 1126 struct storvsc_device *stor_device; 1127 struct vstor_packet *stor_pkt; 1128 1129 stor_device = hv_get_drvdata(device); 1130 stor_pkt = &request->vstor_packet; 1131 1132 /* 1133 * The current SCSI handling on the host side does 1134 * not correctly handle: 1135 * INQUIRY command with page code parameter set to 0x80 1136 * MODE_SENSE command with cmd[2] == 0x1c 1137 * 1138 * Setup srb and scsi status so this won't be fatal. 1139 * We do this so we can distinguish truly fatal failues 1140 * (srb status == 0x4) and off-line the device in that case. 1141 */ 1142 1143 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || 1144 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { 1145 vstor_packet->vm_srb.scsi_status = 0; 1146 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; 1147 } 1148 1149 1150 /* Copy over the status...etc */ 1151 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; 1152 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; 1153 stor_pkt->vm_srb.sense_info_length = 1154 vstor_packet->vm_srb.sense_info_length; 1155 1156 1157 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { 1158 /* CHECK_CONDITION */ 1159 if (vstor_packet->vm_srb.srb_status & 1160 SRB_STATUS_AUTOSENSE_VALID) { 1161 /* autosense data available */ 1162 1163 memcpy(request->sense_buffer, 1164 vstor_packet->vm_srb.sense_data, 1165 vstor_packet->vm_srb.sense_info_length); 1166 1167 } 1168 } 1169 1170 stor_pkt->vm_srb.data_transfer_length = 1171 vstor_packet->vm_srb.data_transfer_length; 1172 1173 storvsc_command_completion(request); 1174 1175 if (atomic_dec_and_test(&stor_device->num_outstanding_req) && 1176 stor_device->drain_notify) 1177 wake_up(&stor_device->waiting_to_drain); 1178 1179 1180 } 1181 1182 static void storvsc_on_receive(struct hv_device *device, 1183 struct vstor_packet *vstor_packet, 1184 struct storvsc_cmd_request *request) 1185 { 1186 struct storvsc_scan_work *work; 1187 struct storvsc_device *stor_device; 1188 1189 switch (vstor_packet->operation) { 1190 case VSTOR_OPERATION_COMPLETE_IO: 1191 storvsc_on_io_completion(device, vstor_packet, request); 1192 break; 1193 1194 case VSTOR_OPERATION_REMOVE_DEVICE: 1195 case VSTOR_OPERATION_ENUMERATE_BUS: 1196 stor_device = get_in_stor_device(device); 1197 work = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 1198 if (!work) 1199 return; 1200 1201 INIT_WORK(&work->work, storvsc_bus_scan); 1202 work->host = stor_device->host; 1203 schedule_work(&work->work); 1204 break; 1205 1206 default: 1207 break; 1208 } 1209 } 1210 1211 static void storvsc_on_channel_callback(void *context) 1212 { 1213 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1214 struct hv_device *device; 1215 struct storvsc_device *stor_device; 1216 u32 bytes_recvd; 1217 u64 request_id; 1218 unsigned char packet[ALIGN(sizeof(struct vstor_packet), 8)]; 1219 struct storvsc_cmd_request *request; 1220 int ret; 1221 1222 if (channel->primary_channel != NULL) 1223 device = channel->primary_channel->device_obj; 1224 else 1225 device = channel->device_obj; 1226 1227 stor_device = get_in_stor_device(device); 1228 if (!stor_device) 1229 return; 1230 1231 do { 1232 ret = vmbus_recvpacket(channel, packet, 1233 ALIGN((sizeof(struct vstor_packet) - 1234 vmscsi_size_delta), 8), 1235 &bytes_recvd, &request_id); 1236 if (ret == 0 && bytes_recvd > 0) { 1237 1238 request = (struct storvsc_cmd_request *) 1239 (unsigned long)request_id; 1240 1241 if ((request == &stor_device->init_request) || 1242 (request == &stor_device->reset_request)) { 1243 1244 memcpy(&request->vstor_packet, packet, 1245 (sizeof(struct vstor_packet) - 1246 vmscsi_size_delta)); 1247 complete(&request->wait_event); 1248 } else { 1249 storvsc_on_receive(device, 1250 (struct vstor_packet *)packet, 1251 request); 1252 } 1253 } else { 1254 break; 1255 } 1256 } while (1); 1257 1258 return; 1259 } 1260 1261 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size) 1262 { 1263 struct vmstorage_channel_properties props; 1264 int ret; 1265 1266 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 1267 1268 ret = vmbus_open(device->channel, 1269 ring_size, 1270 ring_size, 1271 (void *)&props, 1272 sizeof(struct vmstorage_channel_properties), 1273 storvsc_on_channel_callback, device->channel); 1274 1275 if (ret != 0) 1276 return ret; 1277 1278 ret = storvsc_channel_init(device); 1279 1280 return ret; 1281 } 1282 1283 static int storvsc_dev_remove(struct hv_device *device) 1284 { 1285 struct storvsc_device *stor_device; 1286 unsigned long flags; 1287 1288 stor_device = hv_get_drvdata(device); 1289 1290 spin_lock_irqsave(&device->channel->inbound_lock, flags); 1291 stor_device->destroy = true; 1292 spin_unlock_irqrestore(&device->channel->inbound_lock, flags); 1293 1294 /* 1295 * At this point, all outbound traffic should be disable. We 1296 * only allow inbound traffic (responses) to proceed so that 1297 * outstanding requests can be completed. 1298 */ 1299 1300 storvsc_wait_to_drain(stor_device); 1301 1302 /* 1303 * Since we have already drained, we don't need to busy wait 1304 * as was done in final_release_stor_device() 1305 * Note that we cannot set the ext pointer to NULL until 1306 * we have drained - to drain the outgoing packets, we need to 1307 * allow incoming packets. 1308 */ 1309 spin_lock_irqsave(&device->channel->inbound_lock, flags); 1310 hv_set_drvdata(device, NULL); 1311 spin_unlock_irqrestore(&device->channel->inbound_lock, flags); 1312 1313 /* Close the channel */ 1314 vmbus_close(device->channel); 1315 1316 kfree(stor_device); 1317 return 0; 1318 } 1319 1320 static int storvsc_do_io(struct hv_device *device, 1321 struct storvsc_cmd_request *request) 1322 { 1323 struct storvsc_device *stor_device; 1324 struct vstor_packet *vstor_packet; 1325 struct vmbus_channel *outgoing_channel; 1326 int ret = 0; 1327 1328 vstor_packet = &request->vstor_packet; 1329 stor_device = get_out_stor_device(device); 1330 1331 if (!stor_device) 1332 return -ENODEV; 1333 1334 1335 request->device = device; 1336 /* 1337 * Select an an appropriate channel to send the request out. 1338 */ 1339 1340 outgoing_channel = vmbus_get_outgoing_channel(device->channel); 1341 1342 1343 vstor_packet->flags |= REQUEST_COMPLETION_FLAG; 1344 1345 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) - 1346 vmscsi_size_delta); 1347 1348 1349 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1350 1351 1352 vstor_packet->vm_srb.data_transfer_length = 1353 request->data_buffer.len; 1354 1355 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; 1356 1357 if (request->data_buffer.len) { 1358 ret = vmbus_sendpacket_multipagebuffer(outgoing_channel, 1359 &request->data_buffer, 1360 vstor_packet, 1361 (sizeof(struct vstor_packet) - 1362 vmscsi_size_delta), 1363 (unsigned long)request); 1364 } else { 1365 ret = vmbus_sendpacket(device->channel, vstor_packet, 1366 (sizeof(struct vstor_packet) - 1367 vmscsi_size_delta), 1368 (unsigned long)request, 1369 VM_PKT_DATA_INBAND, 1370 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1371 } 1372 1373 if (ret != 0) 1374 return ret; 1375 1376 atomic_inc(&stor_device->num_outstanding_req); 1377 1378 return ret; 1379 } 1380 1381 static int storvsc_device_alloc(struct scsi_device *sdevice) 1382 { 1383 struct stor_mem_pools *memp; 1384 int number = STORVSC_MIN_BUF_NR; 1385 1386 memp = kzalloc(sizeof(struct stor_mem_pools), GFP_KERNEL); 1387 if (!memp) 1388 return -ENOMEM; 1389 1390 memp->request_pool = 1391 kmem_cache_create(dev_name(&sdevice->sdev_dev), 1392 sizeof(struct storvsc_cmd_request), 0, 1393 SLAB_HWCACHE_ALIGN, NULL); 1394 1395 if (!memp->request_pool) 1396 goto err0; 1397 1398 memp->request_mempool = mempool_create(number, mempool_alloc_slab, 1399 mempool_free_slab, 1400 memp->request_pool); 1401 1402 if (!memp->request_mempool) 1403 goto err1; 1404 1405 sdevice->hostdata = memp; 1406 1407 return 0; 1408 1409 err1: 1410 kmem_cache_destroy(memp->request_pool); 1411 1412 err0: 1413 kfree(memp); 1414 return -ENOMEM; 1415 } 1416 1417 static void storvsc_device_destroy(struct scsi_device *sdevice) 1418 { 1419 struct stor_mem_pools *memp = sdevice->hostdata; 1420 1421 if (!memp) 1422 return; 1423 1424 mempool_destroy(memp->request_mempool); 1425 kmem_cache_destroy(memp->request_pool); 1426 kfree(memp); 1427 sdevice->hostdata = NULL; 1428 } 1429 1430 static int storvsc_device_configure(struct scsi_device *sdevice) 1431 { 1432 scsi_change_queue_depth(sdevice, STORVSC_MAX_IO_REQUESTS); 1433 1434 blk_queue_max_segment_size(sdevice->request_queue, PAGE_SIZE); 1435 1436 blk_queue_bounce_limit(sdevice->request_queue, BLK_BOUNCE_ANY); 1437 1438 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ)); 1439 1440 sdevice->no_write_same = 1; 1441 1442 /* 1443 * Add blist flags to permit the reading of the VPD pages even when 1444 * the target may claim SPC-2 compliance. MSFT targets currently 1445 * claim SPC-2 compliance while they implement post SPC-2 features. 1446 * With this patch we can correctly handle WRITE_SAME_16 issues. 1447 */ 1448 sdevice->sdev_bflags |= msft_blist_flags; 1449 1450 return 0; 1451 } 1452 1453 static int storvsc_get_chs(struct scsi_device *sdev, struct block_device * bdev, 1454 sector_t capacity, int *info) 1455 { 1456 sector_t nsect = capacity; 1457 sector_t cylinders = nsect; 1458 int heads, sectors_pt; 1459 1460 /* 1461 * We are making up these values; let us keep it simple. 1462 */ 1463 heads = 0xff; 1464 sectors_pt = 0x3f; /* Sectors per track */ 1465 sector_div(cylinders, heads * sectors_pt); 1466 if ((sector_t)(cylinders + 1) * heads * sectors_pt < nsect) 1467 cylinders = 0xffff; 1468 1469 info[0] = heads; 1470 info[1] = sectors_pt; 1471 info[2] = (int)cylinders; 1472 1473 return 0; 1474 } 1475 1476 static int storvsc_host_reset_handler(struct scsi_cmnd *scmnd) 1477 { 1478 struct hv_host_device *host_dev = shost_priv(scmnd->device->host); 1479 struct hv_device *device = host_dev->dev; 1480 1481 struct storvsc_device *stor_device; 1482 struct storvsc_cmd_request *request; 1483 struct vstor_packet *vstor_packet; 1484 int ret, t; 1485 1486 1487 stor_device = get_out_stor_device(device); 1488 if (!stor_device) 1489 return FAILED; 1490 1491 request = &stor_device->reset_request; 1492 vstor_packet = &request->vstor_packet; 1493 1494 init_completion(&request->wait_event); 1495 1496 vstor_packet->operation = VSTOR_OPERATION_RESET_BUS; 1497 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 1498 vstor_packet->vm_srb.path_id = stor_device->path_id; 1499 1500 ret = vmbus_sendpacket(device->channel, vstor_packet, 1501 (sizeof(struct vstor_packet) - 1502 vmscsi_size_delta), 1503 (unsigned long)&stor_device->reset_request, 1504 VM_PKT_DATA_INBAND, 1505 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1506 if (ret != 0) 1507 return FAILED; 1508 1509 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 1510 if (t == 0) 1511 return TIMEOUT_ERROR; 1512 1513 1514 /* 1515 * At this point, all outstanding requests in the adapter 1516 * should have been flushed out and return to us 1517 * There is a potential race here where the host may be in 1518 * the process of responding when we return from here. 1519 * Just wait for all in-transit packets to be accounted for 1520 * before we return from here. 1521 */ 1522 storvsc_wait_to_drain(stor_device); 1523 1524 return SUCCESS; 1525 } 1526 1527 /* 1528 * The host guarantees to respond to each command, although I/O latencies might 1529 * be unbounded on Azure. Reset the timer unconditionally to give the host a 1530 * chance to perform EH. 1531 */ 1532 static enum blk_eh_timer_return storvsc_eh_timed_out(struct scsi_cmnd *scmnd) 1533 { 1534 return BLK_EH_RESET_TIMER; 1535 } 1536 1537 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) 1538 { 1539 bool allowed = true; 1540 u8 scsi_op = scmnd->cmnd[0]; 1541 1542 switch (scsi_op) { 1543 /* the host does not handle WRITE_SAME, log accident usage */ 1544 case WRITE_SAME: 1545 /* 1546 * smartd sends this command and the host does not handle 1547 * this. So, don't send it. 1548 */ 1549 case SET_WINDOW: 1550 scmnd->result = ILLEGAL_REQUEST << 16; 1551 allowed = false; 1552 break; 1553 default: 1554 break; 1555 } 1556 return allowed; 1557 } 1558 1559 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) 1560 { 1561 int ret; 1562 struct hv_host_device *host_dev = shost_priv(host); 1563 struct hv_device *dev = host_dev->dev; 1564 struct storvsc_cmd_request *cmd_request; 1565 unsigned int request_size = 0; 1566 int i; 1567 struct scatterlist *sgl; 1568 unsigned int sg_count = 0; 1569 struct vmscsi_request *vm_srb; 1570 struct stor_mem_pools *memp = scmnd->device->hostdata; 1571 1572 if (vmstor_current_major <= VMSTOR_WIN8_MAJOR) { 1573 /* 1574 * On legacy hosts filter unimplemented commands. 1575 * Future hosts are expected to correctly handle 1576 * unsupported commands. Furthermore, it is 1577 * possible that some of the currently 1578 * unsupported commands maybe supported in 1579 * future versions of the host. 1580 */ 1581 if (!storvsc_scsi_cmd_ok(scmnd)) { 1582 scmnd->scsi_done(scmnd); 1583 return 0; 1584 } 1585 } 1586 1587 request_size = sizeof(struct storvsc_cmd_request); 1588 1589 cmd_request = mempool_alloc(memp->request_mempool, 1590 GFP_ATOMIC); 1591 1592 /* 1593 * We might be invoked in an interrupt context; hence 1594 * mempool_alloc() can fail. 1595 */ 1596 if (!cmd_request) 1597 return SCSI_MLQUEUE_DEVICE_BUSY; 1598 1599 memset(cmd_request, 0, sizeof(struct storvsc_cmd_request)); 1600 1601 /* Setup the cmd request */ 1602 cmd_request->cmd = scmnd; 1603 1604 scmnd->host_scribble = (unsigned char *)cmd_request; 1605 1606 vm_srb = &cmd_request->vstor_packet.vm_srb; 1607 vm_srb->win8_extension.time_out_value = 60; 1608 1609 vm_srb->win8_extension.srb_flags |= 1610 (SRB_FLAGS_QUEUE_ACTION_ENABLE | 1611 SRB_FLAGS_DISABLE_SYNCH_TRANSFER); 1612 1613 /* Build the SRB */ 1614 switch (scmnd->sc_data_direction) { 1615 case DMA_TO_DEVICE: 1616 vm_srb->data_in = WRITE_TYPE; 1617 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT; 1618 break; 1619 case DMA_FROM_DEVICE: 1620 vm_srb->data_in = READ_TYPE; 1621 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN; 1622 break; 1623 default: 1624 vm_srb->data_in = UNKNOWN_TYPE; 1625 vm_srb->win8_extension.srb_flags |= (SRB_FLAGS_DATA_IN | 1626 SRB_FLAGS_DATA_OUT); 1627 break; 1628 } 1629 1630 1631 vm_srb->port_number = host_dev->port; 1632 vm_srb->path_id = scmnd->device->channel; 1633 vm_srb->target_id = scmnd->device->id; 1634 vm_srb->lun = scmnd->device->lun; 1635 1636 vm_srb->cdb_length = scmnd->cmd_len; 1637 1638 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); 1639 1640 cmd_request->sense_buffer = scmnd->sense_buffer; 1641 1642 1643 cmd_request->data_buffer.len = scsi_bufflen(scmnd); 1644 if (scsi_sg_count(scmnd)) { 1645 sgl = (struct scatterlist *)scsi_sglist(scmnd); 1646 sg_count = scsi_sg_count(scmnd); 1647 1648 /* check if we need to bounce the sgl */ 1649 if (do_bounce_buffer(sgl, scsi_sg_count(scmnd)) != -1) { 1650 cmd_request->bounce_sgl = 1651 create_bounce_buffer(sgl, scsi_sg_count(scmnd), 1652 scsi_bufflen(scmnd), 1653 vm_srb->data_in); 1654 if (!cmd_request->bounce_sgl) { 1655 ret = SCSI_MLQUEUE_HOST_BUSY; 1656 goto queue_error; 1657 } 1658 1659 cmd_request->bounce_sgl_count = 1660 ALIGN(scsi_bufflen(scmnd), PAGE_SIZE) >> 1661 PAGE_SHIFT; 1662 1663 if (vm_srb->data_in == WRITE_TYPE) 1664 copy_to_bounce_buffer(sgl, 1665 cmd_request->bounce_sgl, 1666 scsi_sg_count(scmnd)); 1667 1668 sgl = cmd_request->bounce_sgl; 1669 sg_count = cmd_request->bounce_sgl_count; 1670 } 1671 1672 cmd_request->data_buffer.offset = sgl[0].offset; 1673 1674 for (i = 0; i < sg_count; i++) 1675 cmd_request->data_buffer.pfn_array[i] = 1676 page_to_pfn(sg_page((&sgl[i]))); 1677 1678 } else if (scsi_sglist(scmnd)) { 1679 cmd_request->data_buffer.offset = 1680 virt_to_phys(scsi_sglist(scmnd)) & (PAGE_SIZE-1); 1681 cmd_request->data_buffer.pfn_array[0] = 1682 virt_to_phys(scsi_sglist(scmnd)) >> PAGE_SHIFT; 1683 } 1684 1685 /* Invokes the vsc to start an IO */ 1686 ret = storvsc_do_io(dev, cmd_request); 1687 1688 if (ret == -EAGAIN) { 1689 /* no more space */ 1690 1691 if (cmd_request->bounce_sgl_count) 1692 destroy_bounce_buffer(cmd_request->bounce_sgl, 1693 cmd_request->bounce_sgl_count); 1694 1695 ret = SCSI_MLQUEUE_DEVICE_BUSY; 1696 goto queue_error; 1697 } 1698 1699 return 0; 1700 1701 queue_error: 1702 mempool_free(cmd_request, memp->request_mempool); 1703 scmnd->host_scribble = NULL; 1704 return ret; 1705 } 1706 1707 static struct scsi_host_template scsi_driver = { 1708 .module = THIS_MODULE, 1709 .name = "storvsc_host_t", 1710 .bios_param = storvsc_get_chs, 1711 .queuecommand = storvsc_queuecommand, 1712 .eh_host_reset_handler = storvsc_host_reset_handler, 1713 .eh_timed_out = storvsc_eh_timed_out, 1714 .slave_alloc = storvsc_device_alloc, 1715 .slave_destroy = storvsc_device_destroy, 1716 .slave_configure = storvsc_device_configure, 1717 .cmd_per_lun = 255, 1718 .can_queue = STORVSC_MAX_IO_REQUESTS*STORVSC_MAX_TARGETS, 1719 .this_id = -1, 1720 /* no use setting to 0 since ll_blk_rw reset it to 1 */ 1721 /* currently 32 */ 1722 .sg_tablesize = MAX_MULTIPAGE_BUFFER_COUNT, 1723 .use_clustering = DISABLE_CLUSTERING, 1724 /* Make sure we dont get a sg segment crosses a page boundary */ 1725 .dma_boundary = PAGE_SIZE-1, 1726 .no_write_same = 1, 1727 }; 1728 1729 enum { 1730 SCSI_GUID, 1731 IDE_GUID, 1732 SFC_GUID, 1733 }; 1734 1735 static const struct hv_vmbus_device_id id_table[] = { 1736 /* SCSI guid */ 1737 { HV_SCSI_GUID, 1738 .driver_data = SCSI_GUID 1739 }, 1740 /* IDE guid */ 1741 { HV_IDE_GUID, 1742 .driver_data = IDE_GUID 1743 }, 1744 /* Fibre Channel GUID */ 1745 { 1746 HV_SYNTHFC_GUID, 1747 .driver_data = SFC_GUID 1748 }, 1749 { }, 1750 }; 1751 1752 MODULE_DEVICE_TABLE(vmbus, id_table); 1753 1754 static int storvsc_probe(struct hv_device *device, 1755 const struct hv_vmbus_device_id *dev_id) 1756 { 1757 int ret; 1758 struct Scsi_Host *host; 1759 struct hv_host_device *host_dev; 1760 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); 1761 int target = 0; 1762 struct storvsc_device *stor_device; 1763 1764 /* 1765 * Based on the windows host we are running on, 1766 * set state to properly communicate with the host. 1767 */ 1768 1769 switch (vmbus_proto_version) { 1770 case VERSION_WS2008: 1771 case VERSION_WIN7: 1772 sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE; 1773 vmscsi_size_delta = sizeof(struct vmscsi_win8_extension); 1774 vmstor_current_major = VMSTOR_WIN7_MAJOR; 1775 vmstor_current_minor = VMSTOR_WIN7_MINOR; 1776 break; 1777 default: 1778 sense_buffer_size = POST_WIN7_STORVSC_SENSE_BUFFER_SIZE; 1779 vmscsi_size_delta = 0; 1780 vmstor_current_major = VMSTOR_WIN8_MAJOR; 1781 vmstor_current_minor = VMSTOR_WIN8_MINOR; 1782 break; 1783 } 1784 1785 if (dev_id->driver_data == SFC_GUID) 1786 scsi_driver.can_queue = (STORVSC_MAX_IO_REQUESTS * 1787 STORVSC_FC_MAX_TARGETS); 1788 host = scsi_host_alloc(&scsi_driver, 1789 sizeof(struct hv_host_device)); 1790 if (!host) 1791 return -ENOMEM; 1792 1793 host_dev = shost_priv(host); 1794 memset(host_dev, 0, sizeof(struct hv_host_device)); 1795 1796 host_dev->port = host->host_no; 1797 host_dev->dev = device; 1798 1799 1800 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); 1801 if (!stor_device) { 1802 ret = -ENOMEM; 1803 goto err_out0; 1804 } 1805 1806 stor_device->destroy = false; 1807 stor_device->open_sub_channel = false; 1808 init_waitqueue_head(&stor_device->waiting_to_drain); 1809 stor_device->device = device; 1810 stor_device->host = host; 1811 hv_set_drvdata(device, stor_device); 1812 1813 stor_device->port_number = host->host_no; 1814 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size); 1815 if (ret) 1816 goto err_out1; 1817 1818 host_dev->path = stor_device->path_id; 1819 host_dev->target = stor_device->target_id; 1820 1821 switch (dev_id->driver_data) { 1822 case SFC_GUID: 1823 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET; 1824 host->max_id = STORVSC_FC_MAX_TARGETS; 1825 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1; 1826 break; 1827 1828 case SCSI_GUID: 1829 host->max_lun = STORVSC_MAX_LUNS_PER_TARGET; 1830 host->max_id = STORVSC_MAX_TARGETS; 1831 host->max_channel = STORVSC_MAX_CHANNELS - 1; 1832 break; 1833 1834 default: 1835 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1836 host->max_id = STORVSC_IDE_MAX_TARGETS; 1837 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1; 1838 break; 1839 } 1840 /* max cmd length */ 1841 host->max_cmd_len = STORVSC_MAX_CMD_LEN; 1842 1843 /* Register the HBA and start the scsi bus scan */ 1844 ret = scsi_add_host(host, &device->device); 1845 if (ret != 0) 1846 goto err_out2; 1847 1848 if (!dev_is_ide) { 1849 scsi_scan_host(host); 1850 } else { 1851 target = (device->dev_instance.b[5] << 8 | 1852 device->dev_instance.b[4]); 1853 ret = scsi_add_device(host, 0, target, 0); 1854 if (ret) { 1855 scsi_remove_host(host); 1856 goto err_out2; 1857 } 1858 } 1859 return 0; 1860 1861 err_out2: 1862 /* 1863 * Once we have connected with the host, we would need to 1864 * to invoke storvsc_dev_remove() to rollback this state and 1865 * this call also frees up the stor_device; hence the jump around 1866 * err_out1 label. 1867 */ 1868 storvsc_dev_remove(device); 1869 goto err_out0; 1870 1871 err_out1: 1872 kfree(stor_device); 1873 1874 err_out0: 1875 scsi_host_put(host); 1876 return ret; 1877 } 1878 1879 static int storvsc_remove(struct hv_device *dev) 1880 { 1881 struct storvsc_device *stor_device = hv_get_drvdata(dev); 1882 struct Scsi_Host *host = stor_device->host; 1883 1884 scsi_remove_host(host); 1885 storvsc_dev_remove(dev); 1886 scsi_host_put(host); 1887 1888 return 0; 1889 } 1890 1891 static struct hv_driver storvsc_drv = { 1892 .name = KBUILD_MODNAME, 1893 .id_table = id_table, 1894 .probe = storvsc_probe, 1895 .remove = storvsc_remove, 1896 }; 1897 1898 static int __init storvsc_drv_init(void) 1899 { 1900 u32 max_outstanding_req_per_channel; 1901 1902 /* 1903 * Divide the ring buffer data size (which is 1 page less 1904 * than the ring buffer size since that page is reserved for 1905 * the ring buffer indices) by the max request size (which is 1906 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) 1907 */ 1908 max_outstanding_req_per_channel = 1909 ((storvsc_ringbuffer_size - PAGE_SIZE) / 1910 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + 1911 sizeof(struct vstor_packet) + sizeof(u64) - 1912 vmscsi_size_delta, 1913 sizeof(u64))); 1914 1915 if (max_outstanding_req_per_channel < 1916 STORVSC_MAX_IO_REQUESTS) 1917 return -EINVAL; 1918 1919 return vmbus_driver_register(&storvsc_drv); 1920 } 1921 1922 static void __exit storvsc_drv_exit(void) 1923 { 1924 vmbus_driver_unregister(&storvsc_drv); 1925 } 1926 1927 MODULE_LICENSE("GPL"); 1928 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); 1929 module_init(storvsc_drv_init); 1930 module_exit(storvsc_drv_exit); 1931