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