1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2009, Microsoft Corporation. 4 * 5 * Authors: 6 * Haiyang Zhang <haiyangz@microsoft.com> 7 * Hank Janssen <hjanssen@microsoft.com> 8 * K. Y. Srinivasan <kys@microsoft.com> 9 */ 10 11 #include <linux/kernel.h> 12 #include <linux/wait.h> 13 #include <linux/sched.h> 14 #include <linux/completion.h> 15 #include <linux/string.h> 16 #include <linux/mm.h> 17 #include <linux/delay.h> 18 #include <linux/init.h> 19 #include <linux/slab.h> 20 #include <linux/module.h> 21 #include <linux/device.h> 22 #include <linux/hyperv.h> 23 #include <linux/blkdev.h> 24 #include <scsi/scsi.h> 25 #include <scsi/scsi_cmnd.h> 26 #include <scsi/scsi_host.h> 27 #include <scsi/scsi_device.h> 28 #include <scsi/scsi_tcq.h> 29 #include <scsi/scsi_eh.h> 30 #include <scsi/scsi_devinfo.h> 31 #include <scsi/scsi_dbg.h> 32 #include <scsi/scsi_transport_fc.h> 33 #include <scsi/scsi_transport.h> 34 35 /* 36 * All wire protocol details (storage protocol between the guest and the host) 37 * are consolidated here. 38 * 39 * Begin protocol definitions. 40 */ 41 42 /* 43 * Version history: 44 * V1 Beta: 0.1 45 * V1 RC < 2008/1/31: 1.0 46 * V1 RC > 2008/1/31: 2.0 47 * Win7: 4.2 48 * Win8: 5.1 49 * Win8.1: 6.0 50 * Win10: 6.2 51 */ 52 53 #define VMSTOR_PROTO_VERSION(MAJOR_, MINOR_) ((((MAJOR_) & 0xff) << 8) | \ 54 (((MINOR_) & 0xff))) 55 56 #define VMSTOR_PROTO_VERSION_WIN6 VMSTOR_PROTO_VERSION(2, 0) 57 #define VMSTOR_PROTO_VERSION_WIN7 VMSTOR_PROTO_VERSION(4, 2) 58 #define VMSTOR_PROTO_VERSION_WIN8 VMSTOR_PROTO_VERSION(5, 1) 59 #define VMSTOR_PROTO_VERSION_WIN8_1 VMSTOR_PROTO_VERSION(6, 0) 60 #define VMSTOR_PROTO_VERSION_WIN10 VMSTOR_PROTO_VERSION(6, 2) 61 62 /* Packet structure describing virtual storage requests. */ 63 enum vstor_packet_operation { 64 VSTOR_OPERATION_COMPLETE_IO = 1, 65 VSTOR_OPERATION_REMOVE_DEVICE = 2, 66 VSTOR_OPERATION_EXECUTE_SRB = 3, 67 VSTOR_OPERATION_RESET_LUN = 4, 68 VSTOR_OPERATION_RESET_ADAPTER = 5, 69 VSTOR_OPERATION_RESET_BUS = 6, 70 VSTOR_OPERATION_BEGIN_INITIALIZATION = 7, 71 VSTOR_OPERATION_END_INITIALIZATION = 8, 72 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION = 9, 73 VSTOR_OPERATION_QUERY_PROPERTIES = 10, 74 VSTOR_OPERATION_ENUMERATE_BUS = 11, 75 VSTOR_OPERATION_FCHBA_DATA = 12, 76 VSTOR_OPERATION_CREATE_SUB_CHANNELS = 13, 77 VSTOR_OPERATION_MAXIMUM = 13 78 }; 79 80 /* 81 * WWN packet for Fibre Channel HBA 82 */ 83 84 struct hv_fc_wwn_packet { 85 u8 primary_active; 86 u8 reserved1[3]; 87 u8 primary_port_wwn[8]; 88 u8 primary_node_wwn[8]; 89 u8 secondary_port_wwn[8]; 90 u8 secondary_node_wwn[8]; 91 }; 92 93 94 95 /* 96 * SRB Flag Bits 97 */ 98 99 #define SRB_FLAGS_QUEUE_ACTION_ENABLE 0x00000002 100 #define SRB_FLAGS_DISABLE_DISCONNECT 0x00000004 101 #define SRB_FLAGS_DISABLE_SYNCH_TRANSFER 0x00000008 102 #define SRB_FLAGS_BYPASS_FROZEN_QUEUE 0x00000010 103 #define SRB_FLAGS_DISABLE_AUTOSENSE 0x00000020 104 #define SRB_FLAGS_DATA_IN 0x00000040 105 #define SRB_FLAGS_DATA_OUT 0x00000080 106 #define SRB_FLAGS_NO_DATA_TRANSFER 0x00000000 107 #define SRB_FLAGS_UNSPECIFIED_DIRECTION (SRB_FLAGS_DATA_IN | SRB_FLAGS_DATA_OUT) 108 #define SRB_FLAGS_NO_QUEUE_FREEZE 0x00000100 109 #define SRB_FLAGS_ADAPTER_CACHE_ENABLE 0x00000200 110 #define SRB_FLAGS_FREE_SENSE_BUFFER 0x00000400 111 112 /* 113 * This flag indicates the request is part of the workflow for processing a D3. 114 */ 115 #define SRB_FLAGS_D3_PROCESSING 0x00000800 116 #define SRB_FLAGS_IS_ACTIVE 0x00010000 117 #define SRB_FLAGS_ALLOCATED_FROM_ZONE 0x00020000 118 #define SRB_FLAGS_SGLIST_FROM_POOL 0x00040000 119 #define SRB_FLAGS_BYPASS_LOCKED_QUEUE 0x00080000 120 #define SRB_FLAGS_NO_KEEP_AWAKE 0x00100000 121 #define SRB_FLAGS_PORT_DRIVER_ALLOCSENSE 0x00200000 122 #define SRB_FLAGS_PORT_DRIVER_SENSEHASPORT 0x00400000 123 #define SRB_FLAGS_DONT_START_NEXT_PACKET 0x00800000 124 #define SRB_FLAGS_PORT_DRIVER_RESERVED 0x0F000000 125 #define SRB_FLAGS_CLASS_DRIVER_RESERVED 0xF0000000 126 127 #define SP_UNTAGGED ((unsigned char) ~0) 128 #define SRB_SIMPLE_TAG_REQUEST 0x20 129 130 /* 131 * Platform neutral description of a scsi request - 132 * this remains the same across the write regardless of 32/64 bit 133 * note: it's patterned off the SCSI_PASS_THROUGH structure 134 */ 135 #define STORVSC_MAX_CMD_LEN 0x10 136 137 #define POST_WIN7_STORVSC_SENSE_BUFFER_SIZE 0x14 138 #define PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE 0x12 139 140 #define STORVSC_SENSE_BUFFER_SIZE 0x14 141 #define STORVSC_MAX_BUF_LEN_WITH_PADDING 0x14 142 143 /* 144 * Sense buffer size changed in win8; have a run-time 145 * variable to track the size we should use. This value will 146 * likely change during protocol negotiation but it is valid 147 * to start by assuming pre-Win8. 148 */ 149 static int sense_buffer_size = PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE; 150 151 /* 152 * The storage protocol version is determined during the 153 * initial exchange with the host. It will indicate which 154 * storage functionality is available in the host. 155 */ 156 static int vmstor_proto_version; 157 158 #define STORVSC_LOGGING_NONE 0 159 #define STORVSC_LOGGING_ERROR 1 160 #define STORVSC_LOGGING_WARN 2 161 162 static int logging_level = STORVSC_LOGGING_ERROR; 163 module_param(logging_level, int, S_IRUGO|S_IWUSR); 164 MODULE_PARM_DESC(logging_level, 165 "Logging level, 0 - None, 1 - Error (default), 2 - Warning."); 166 167 static inline bool do_logging(int level) 168 { 169 return logging_level >= level; 170 } 171 172 #define storvsc_log(dev, level, fmt, ...) \ 173 do { \ 174 if (do_logging(level)) \ 175 dev_warn(&(dev)->device, fmt, ##__VA_ARGS__); \ 176 } while (0) 177 178 struct vmscsi_win8_extension { 179 /* 180 * The following were added in Windows 8 181 */ 182 u16 reserve; 183 u8 queue_tag; 184 u8 queue_action; 185 u32 srb_flags; 186 u32 time_out_value; 187 u32 queue_sort_ey; 188 } __packed; 189 190 struct vmscsi_request { 191 u16 length; 192 u8 srb_status; 193 u8 scsi_status; 194 195 u8 port_number; 196 u8 path_id; 197 u8 target_id; 198 u8 lun; 199 200 u8 cdb_length; 201 u8 sense_info_length; 202 u8 data_in; 203 u8 reserved; 204 205 u32 data_transfer_length; 206 207 union { 208 u8 cdb[STORVSC_MAX_CMD_LEN]; 209 u8 sense_data[STORVSC_SENSE_BUFFER_SIZE]; 210 u8 reserved_array[STORVSC_MAX_BUF_LEN_WITH_PADDING]; 211 }; 212 /* 213 * The following was added in win8. 214 */ 215 struct vmscsi_win8_extension win8_extension; 216 217 } __attribute((packed)); 218 219 220 /* 221 * The size of the vmscsi_request has changed in win8. The 222 * additional size is because of new elements added to the 223 * structure. These elements are valid only when we are talking 224 * to a win8 host. 225 * Track the correction to size we need to apply. This value 226 * will likely change during protocol negotiation but it is 227 * valid to start by assuming pre-Win8. 228 */ 229 static int vmscsi_size_delta = sizeof(struct vmscsi_win8_extension); 230 231 /* 232 * The list of storage protocols in order of preference. 233 */ 234 struct vmstor_protocol { 235 int protocol_version; 236 int sense_buffer_size; 237 int vmscsi_size_delta; 238 }; 239 240 241 static const struct vmstor_protocol vmstor_protocols[] = { 242 { 243 VMSTOR_PROTO_VERSION_WIN10, 244 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 245 0 246 }, 247 { 248 VMSTOR_PROTO_VERSION_WIN8_1, 249 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 250 0 251 }, 252 { 253 VMSTOR_PROTO_VERSION_WIN8, 254 POST_WIN7_STORVSC_SENSE_BUFFER_SIZE, 255 0 256 }, 257 { 258 VMSTOR_PROTO_VERSION_WIN7, 259 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE, 260 sizeof(struct vmscsi_win8_extension), 261 }, 262 { 263 VMSTOR_PROTO_VERSION_WIN6, 264 PRE_WIN8_STORVSC_SENSE_BUFFER_SIZE, 265 sizeof(struct vmscsi_win8_extension), 266 } 267 }; 268 269 270 /* 271 * This structure is sent during the initialization phase to get the different 272 * properties of the channel. 273 */ 274 275 #define STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL 0x1 276 277 struct vmstorage_channel_properties { 278 u32 reserved; 279 u16 max_channel_cnt; 280 u16 reserved1; 281 282 u32 flags; 283 u32 max_transfer_bytes; 284 285 u64 reserved2; 286 } __packed; 287 288 /* This structure is sent during the storage protocol negotiations. */ 289 struct vmstorage_protocol_version { 290 /* Major (MSW) and minor (LSW) version numbers. */ 291 u16 major_minor; 292 293 /* 294 * Revision number is auto-incremented whenever this file is changed 295 * (See FILL_VMSTOR_REVISION macro above). Mismatch does not 296 * definitely indicate incompatibility--but it does indicate mismatched 297 * builds. 298 * This is only used on the windows side. Just set it to 0. 299 */ 300 u16 revision; 301 } __packed; 302 303 /* Channel Property Flags */ 304 #define STORAGE_CHANNEL_REMOVABLE_FLAG 0x1 305 #define STORAGE_CHANNEL_EMULATED_IDE_FLAG 0x2 306 307 struct vstor_packet { 308 /* Requested operation type */ 309 enum vstor_packet_operation operation; 310 311 /* Flags - see below for values */ 312 u32 flags; 313 314 /* Status of the request returned from the server side. */ 315 u32 status; 316 317 /* Data payload area */ 318 union { 319 /* 320 * Structure used to forward SCSI commands from the 321 * client to the server. 322 */ 323 struct vmscsi_request vm_srb; 324 325 /* Structure used to query channel properties. */ 326 struct vmstorage_channel_properties storage_channel_properties; 327 328 /* Used during version negotiations. */ 329 struct vmstorage_protocol_version version; 330 331 /* Fibre channel address packet */ 332 struct hv_fc_wwn_packet wwn_packet; 333 334 /* Number of sub-channels to create */ 335 u16 sub_channel_count; 336 337 /* This will be the maximum of the union members */ 338 u8 buffer[0x34]; 339 }; 340 } __packed; 341 342 /* 343 * Packet Flags: 344 * 345 * This flag indicates that the server should send back a completion for this 346 * packet. 347 */ 348 349 #define REQUEST_COMPLETION_FLAG 0x1 350 351 /* Matches Windows-end */ 352 enum storvsc_request_type { 353 WRITE_TYPE = 0, 354 READ_TYPE, 355 UNKNOWN_TYPE, 356 }; 357 358 /* 359 * SRB status codes and masks; a subset of the codes used here. 360 */ 361 362 #define SRB_STATUS_AUTOSENSE_VALID 0x80 363 #define SRB_STATUS_QUEUE_FROZEN 0x40 364 #define SRB_STATUS_INVALID_LUN 0x20 365 #define SRB_STATUS_SUCCESS 0x01 366 #define SRB_STATUS_ABORTED 0x02 367 #define SRB_STATUS_ERROR 0x04 368 #define SRB_STATUS_DATA_OVERRUN 0x12 369 370 #define SRB_STATUS(status) \ 371 (status & ~(SRB_STATUS_AUTOSENSE_VALID | SRB_STATUS_QUEUE_FROZEN)) 372 /* 373 * This is the end of Protocol specific defines. 374 */ 375 376 static int storvsc_ringbuffer_size = (128 * 1024); 377 static u32 max_outstanding_req_per_channel; 378 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth); 379 380 static int storvsc_vcpus_per_sub_channel = 4; 381 382 module_param(storvsc_ringbuffer_size, int, S_IRUGO); 383 MODULE_PARM_DESC(storvsc_ringbuffer_size, "Ring buffer size (bytes)"); 384 385 module_param(storvsc_vcpus_per_sub_channel, int, S_IRUGO); 386 MODULE_PARM_DESC(storvsc_vcpus_per_sub_channel, "Ratio of VCPUs to subchannels"); 387 388 static int ring_avail_percent_lowater = 10; 389 module_param(ring_avail_percent_lowater, int, S_IRUGO); 390 MODULE_PARM_DESC(ring_avail_percent_lowater, 391 "Select a channel if available ring size > this in percent"); 392 393 /* 394 * Timeout in seconds for all devices managed by this driver. 395 */ 396 static int storvsc_timeout = 180; 397 398 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 399 static struct scsi_transport_template *fc_transport_template; 400 #endif 401 402 static void storvsc_on_channel_callback(void *context); 403 404 #define STORVSC_MAX_LUNS_PER_TARGET 255 405 #define STORVSC_MAX_TARGETS 2 406 #define STORVSC_MAX_CHANNELS 8 407 408 #define STORVSC_FC_MAX_LUNS_PER_TARGET 255 409 #define STORVSC_FC_MAX_TARGETS 128 410 #define STORVSC_FC_MAX_CHANNELS 8 411 412 #define STORVSC_IDE_MAX_LUNS_PER_TARGET 64 413 #define STORVSC_IDE_MAX_TARGETS 1 414 #define STORVSC_IDE_MAX_CHANNELS 1 415 416 struct storvsc_cmd_request { 417 struct scsi_cmnd *cmd; 418 419 struct hv_device *device; 420 421 /* Synchronize the request/response if needed */ 422 struct completion wait_event; 423 424 struct vmbus_channel_packet_multipage_buffer mpb; 425 struct vmbus_packet_mpb_array *payload; 426 u32 payload_sz; 427 428 struct vstor_packet vstor_packet; 429 }; 430 431 432 /* A storvsc device is a device object that contains a vmbus channel */ 433 struct storvsc_device { 434 struct hv_device *device; 435 436 bool destroy; 437 bool drain_notify; 438 atomic_t num_outstanding_req; 439 struct Scsi_Host *host; 440 441 wait_queue_head_t waiting_to_drain; 442 443 /* 444 * Each unique Port/Path/Target represents 1 channel ie scsi 445 * controller. In reality, the pathid, targetid is always 0 446 * and the port is set by us 447 */ 448 unsigned int port_number; 449 unsigned char path_id; 450 unsigned char target_id; 451 452 /* 453 * Max I/O, the device can support. 454 */ 455 u32 max_transfer_bytes; 456 /* 457 * Number of sub-channels we will open. 458 */ 459 u16 num_sc; 460 struct vmbus_channel **stor_chns; 461 /* 462 * Mask of CPUs bound to subchannels. 463 */ 464 struct cpumask alloced_cpus; 465 /* Used for vsc/vsp channel reset process */ 466 struct storvsc_cmd_request init_request; 467 struct storvsc_cmd_request reset_request; 468 /* 469 * Currently active port and node names for FC devices. 470 */ 471 u64 node_name; 472 u64 port_name; 473 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 474 struct fc_rport *rport; 475 #endif 476 }; 477 478 struct hv_host_device { 479 struct hv_device *dev; 480 unsigned int port; 481 unsigned char path; 482 unsigned char target; 483 struct workqueue_struct *handle_error_wq; 484 struct work_struct host_scan_work; 485 struct Scsi_Host *host; 486 }; 487 488 struct storvsc_scan_work { 489 struct work_struct work; 490 struct Scsi_Host *host; 491 u8 lun; 492 u8 tgt_id; 493 }; 494 495 static void storvsc_device_scan(struct work_struct *work) 496 { 497 struct storvsc_scan_work *wrk; 498 struct scsi_device *sdev; 499 500 wrk = container_of(work, struct storvsc_scan_work, work); 501 502 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 503 if (!sdev) 504 goto done; 505 scsi_rescan_device(&sdev->sdev_gendev); 506 scsi_device_put(sdev); 507 508 done: 509 kfree(wrk); 510 } 511 512 static void storvsc_host_scan(struct work_struct *work) 513 { 514 struct Scsi_Host *host; 515 struct scsi_device *sdev; 516 struct hv_host_device *host_device = 517 container_of(work, struct hv_host_device, host_scan_work); 518 519 host = host_device->host; 520 /* 521 * Before scanning the host, first check to see if any of the 522 * currrently known devices have been hot removed. We issue a 523 * "unit ready" command against all currently known devices. 524 * This I/O will result in an error for devices that have been 525 * removed. As part of handling the I/O error, we remove the device. 526 * 527 * When a LUN is added or removed, the host sends us a signal to 528 * scan the host. Thus we are forced to discover the LUNs that 529 * may have been removed this way. 530 */ 531 mutex_lock(&host->scan_mutex); 532 shost_for_each_device(sdev, host) 533 scsi_test_unit_ready(sdev, 1, 1, NULL); 534 mutex_unlock(&host->scan_mutex); 535 /* 536 * Now scan the host to discover LUNs that may have been added. 537 */ 538 scsi_scan_host(host); 539 } 540 541 static void storvsc_remove_lun(struct work_struct *work) 542 { 543 struct storvsc_scan_work *wrk; 544 struct scsi_device *sdev; 545 546 wrk = container_of(work, struct storvsc_scan_work, work); 547 if (!scsi_host_get(wrk->host)) 548 goto done; 549 550 sdev = scsi_device_lookup(wrk->host, 0, wrk->tgt_id, wrk->lun); 551 552 if (sdev) { 553 scsi_remove_device(sdev); 554 scsi_device_put(sdev); 555 } 556 scsi_host_put(wrk->host); 557 558 done: 559 kfree(wrk); 560 } 561 562 563 /* 564 * We can get incoming messages from the host that are not in response to 565 * messages that we have sent out. An example of this would be messages 566 * received by the guest to notify dynamic addition/removal of LUNs. To 567 * deal with potential race conditions where the driver may be in the 568 * midst of being unloaded when we might receive an unsolicited message 569 * from the host, we have implemented a mechanism to gurantee sequential 570 * consistency: 571 * 572 * 1) Once the device is marked as being destroyed, we will fail all 573 * outgoing messages. 574 * 2) We permit incoming messages when the device is being destroyed, 575 * only to properly account for messages already sent out. 576 */ 577 578 static inline struct storvsc_device *get_out_stor_device( 579 struct hv_device *device) 580 { 581 struct storvsc_device *stor_device; 582 583 stor_device = hv_get_drvdata(device); 584 585 if (stor_device && stor_device->destroy) 586 stor_device = NULL; 587 588 return stor_device; 589 } 590 591 592 static inline void storvsc_wait_to_drain(struct storvsc_device *dev) 593 { 594 dev->drain_notify = true; 595 wait_event(dev->waiting_to_drain, 596 atomic_read(&dev->num_outstanding_req) == 0); 597 dev->drain_notify = false; 598 } 599 600 static inline struct storvsc_device *get_in_stor_device( 601 struct hv_device *device) 602 { 603 struct storvsc_device *stor_device; 604 605 stor_device = hv_get_drvdata(device); 606 607 if (!stor_device) 608 goto get_in_err; 609 610 /* 611 * If the device is being destroyed; allow incoming 612 * traffic only to cleanup outstanding requests. 613 */ 614 615 if (stor_device->destroy && 616 (atomic_read(&stor_device->num_outstanding_req) == 0)) 617 stor_device = NULL; 618 619 get_in_err: 620 return stor_device; 621 622 } 623 624 static void handle_sc_creation(struct vmbus_channel *new_sc) 625 { 626 struct hv_device *device = new_sc->primary_channel->device_obj; 627 struct device *dev = &device->device; 628 struct storvsc_device *stor_device; 629 struct vmstorage_channel_properties props; 630 int ret; 631 632 stor_device = get_out_stor_device(device); 633 if (!stor_device) 634 return; 635 636 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 637 638 ret = vmbus_open(new_sc, 639 storvsc_ringbuffer_size, 640 storvsc_ringbuffer_size, 641 (void *)&props, 642 sizeof(struct vmstorage_channel_properties), 643 storvsc_on_channel_callback, new_sc); 644 645 /* In case vmbus_open() fails, we don't use the sub-channel. */ 646 if (ret != 0) { 647 dev_err(dev, "Failed to open sub-channel: err=%d\n", ret); 648 return; 649 } 650 651 /* Add the sub-channel to the array of available channels. */ 652 stor_device->stor_chns[new_sc->target_cpu] = new_sc; 653 cpumask_set_cpu(new_sc->target_cpu, &stor_device->alloced_cpus); 654 } 655 656 static void handle_multichannel_storage(struct hv_device *device, int max_chns) 657 { 658 struct device *dev = &device->device; 659 struct storvsc_device *stor_device; 660 int num_sc; 661 struct storvsc_cmd_request *request; 662 struct vstor_packet *vstor_packet; 663 int ret, t; 664 665 /* 666 * If the number of CPUs is artificially restricted, such as 667 * with maxcpus=1 on the kernel boot line, Hyper-V could offer 668 * sub-channels >= the number of CPUs. These sub-channels 669 * should not be created. The primary channel is already created 670 * and assigned to one CPU, so check against # CPUs - 1. 671 */ 672 num_sc = min((int)(num_online_cpus() - 1), max_chns); 673 if (!num_sc) 674 return; 675 676 stor_device = get_out_stor_device(device); 677 if (!stor_device) 678 return; 679 680 stor_device->num_sc = num_sc; 681 request = &stor_device->init_request; 682 vstor_packet = &request->vstor_packet; 683 684 /* 685 * Establish a handler for dealing with subchannels. 686 */ 687 vmbus_set_sc_create_callback(device->channel, handle_sc_creation); 688 689 /* 690 * Request the host to create sub-channels. 691 */ 692 memset(request, 0, sizeof(struct storvsc_cmd_request)); 693 init_completion(&request->wait_event); 694 vstor_packet->operation = VSTOR_OPERATION_CREATE_SUB_CHANNELS; 695 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 696 vstor_packet->sub_channel_count = num_sc; 697 698 ret = vmbus_sendpacket(device->channel, vstor_packet, 699 (sizeof(struct vstor_packet) - 700 vmscsi_size_delta), 701 (unsigned long)request, 702 VM_PKT_DATA_INBAND, 703 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 704 705 if (ret != 0) { 706 dev_err(dev, "Failed to create sub-channel: err=%d\n", ret); 707 return; 708 } 709 710 t = wait_for_completion_timeout(&request->wait_event, 10*HZ); 711 if (t == 0) { 712 dev_err(dev, "Failed to create sub-channel: timed out\n"); 713 return; 714 } 715 716 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 717 vstor_packet->status != 0) { 718 dev_err(dev, "Failed to create sub-channel: op=%d, sts=%d\n", 719 vstor_packet->operation, vstor_packet->status); 720 return; 721 } 722 723 /* 724 * We need to do nothing here, because vmbus_process_offer() 725 * invokes channel->sc_creation_callback, which will open and use 726 * the sub-channel(s). 727 */ 728 } 729 730 static void cache_wwn(struct storvsc_device *stor_device, 731 struct vstor_packet *vstor_packet) 732 { 733 /* 734 * Cache the currently active port and node ww names. 735 */ 736 if (vstor_packet->wwn_packet.primary_active) { 737 stor_device->node_name = 738 wwn_to_u64(vstor_packet->wwn_packet.primary_node_wwn); 739 stor_device->port_name = 740 wwn_to_u64(vstor_packet->wwn_packet.primary_port_wwn); 741 } else { 742 stor_device->node_name = 743 wwn_to_u64(vstor_packet->wwn_packet.secondary_node_wwn); 744 stor_device->port_name = 745 wwn_to_u64(vstor_packet->wwn_packet.secondary_port_wwn); 746 } 747 } 748 749 750 static int storvsc_execute_vstor_op(struct hv_device *device, 751 struct storvsc_cmd_request *request, 752 bool status_check) 753 { 754 struct vstor_packet *vstor_packet; 755 int ret, t; 756 757 vstor_packet = &request->vstor_packet; 758 759 init_completion(&request->wait_event); 760 vstor_packet->flags = REQUEST_COMPLETION_FLAG; 761 762 ret = vmbus_sendpacket(device->channel, vstor_packet, 763 (sizeof(struct vstor_packet) - 764 vmscsi_size_delta), 765 (unsigned long)request, 766 VM_PKT_DATA_INBAND, 767 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 768 if (ret != 0) 769 return ret; 770 771 t = wait_for_completion_timeout(&request->wait_event, 5*HZ); 772 if (t == 0) 773 return -ETIMEDOUT; 774 775 if (!status_check) 776 return ret; 777 778 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO || 779 vstor_packet->status != 0) 780 return -EINVAL; 781 782 return ret; 783 } 784 785 static int storvsc_channel_init(struct hv_device *device, bool is_fc) 786 { 787 struct storvsc_device *stor_device; 788 struct storvsc_cmd_request *request; 789 struct vstor_packet *vstor_packet; 790 int ret, i; 791 int max_chns; 792 bool process_sub_channels = false; 793 794 stor_device = get_out_stor_device(device); 795 if (!stor_device) 796 return -ENODEV; 797 798 request = &stor_device->init_request; 799 vstor_packet = &request->vstor_packet; 800 801 /* 802 * Now, initiate the vsc/vsp initialization protocol on the open 803 * channel 804 */ 805 memset(request, 0, sizeof(struct storvsc_cmd_request)); 806 vstor_packet->operation = VSTOR_OPERATION_BEGIN_INITIALIZATION; 807 ret = storvsc_execute_vstor_op(device, request, true); 808 if (ret) 809 return ret; 810 /* 811 * Query host supported protocol version. 812 */ 813 814 for (i = 0; i < ARRAY_SIZE(vmstor_protocols); i++) { 815 /* reuse the packet for version range supported */ 816 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 817 vstor_packet->operation = 818 VSTOR_OPERATION_QUERY_PROTOCOL_VERSION; 819 820 vstor_packet->version.major_minor = 821 vmstor_protocols[i].protocol_version; 822 823 /* 824 * The revision number is only used in Windows; set it to 0. 825 */ 826 vstor_packet->version.revision = 0; 827 ret = storvsc_execute_vstor_op(device, request, false); 828 if (ret != 0) 829 return ret; 830 831 if (vstor_packet->operation != VSTOR_OPERATION_COMPLETE_IO) 832 return -EINVAL; 833 834 if (vstor_packet->status == 0) { 835 vmstor_proto_version = 836 vmstor_protocols[i].protocol_version; 837 838 sense_buffer_size = 839 vmstor_protocols[i].sense_buffer_size; 840 841 vmscsi_size_delta = 842 vmstor_protocols[i].vmscsi_size_delta; 843 844 break; 845 } 846 } 847 848 if (vstor_packet->status != 0) 849 return -EINVAL; 850 851 852 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 853 vstor_packet->operation = VSTOR_OPERATION_QUERY_PROPERTIES; 854 ret = storvsc_execute_vstor_op(device, request, true); 855 if (ret != 0) 856 return ret; 857 858 /* 859 * Check to see if multi-channel support is there. 860 * Hosts that implement protocol version of 5.1 and above 861 * support multi-channel. 862 */ 863 max_chns = vstor_packet->storage_channel_properties.max_channel_cnt; 864 865 /* 866 * Allocate state to manage the sub-channels. 867 * We allocate an array based on the numbers of possible CPUs 868 * (Hyper-V does not support cpu online/offline). 869 * This Array will be sparseley populated with unique 870 * channels - primary + sub-channels. 871 * We will however populate all the slots to evenly distribute 872 * the load. 873 */ 874 stor_device->stor_chns = kcalloc(num_possible_cpus(), sizeof(void *), 875 GFP_KERNEL); 876 if (stor_device->stor_chns == NULL) 877 return -ENOMEM; 878 879 stor_device->stor_chns[device->channel->target_cpu] = device->channel; 880 cpumask_set_cpu(device->channel->target_cpu, 881 &stor_device->alloced_cpus); 882 883 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN8) { 884 if (vstor_packet->storage_channel_properties.flags & 885 STORAGE_CHANNEL_SUPPORTS_MULTI_CHANNEL) 886 process_sub_channels = true; 887 } 888 stor_device->max_transfer_bytes = 889 vstor_packet->storage_channel_properties.max_transfer_bytes; 890 891 if (!is_fc) 892 goto done; 893 894 /* 895 * For FC devices retrieve FC HBA data. 896 */ 897 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 898 vstor_packet->operation = VSTOR_OPERATION_FCHBA_DATA; 899 ret = storvsc_execute_vstor_op(device, request, true); 900 if (ret != 0) 901 return ret; 902 903 /* 904 * Cache the currently active port and node ww names. 905 */ 906 cache_wwn(stor_device, vstor_packet); 907 908 done: 909 910 memset(vstor_packet, 0, sizeof(struct vstor_packet)); 911 vstor_packet->operation = VSTOR_OPERATION_END_INITIALIZATION; 912 ret = storvsc_execute_vstor_op(device, request, true); 913 if (ret != 0) 914 return ret; 915 916 if (process_sub_channels) 917 handle_multichannel_storage(device, max_chns); 918 919 return ret; 920 } 921 922 static void storvsc_handle_error(struct vmscsi_request *vm_srb, 923 struct scsi_cmnd *scmnd, 924 struct Scsi_Host *host, 925 u8 asc, u8 ascq) 926 { 927 struct storvsc_scan_work *wrk; 928 void (*process_err_fn)(struct work_struct *work); 929 struct hv_host_device *host_dev = shost_priv(host); 930 bool do_work = false; 931 932 switch (SRB_STATUS(vm_srb->srb_status)) { 933 case SRB_STATUS_ERROR: 934 /* 935 * Let upper layer deal with error when 936 * sense message is present. 937 */ 938 939 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID) 940 break; 941 /* 942 * If there is an error; offline the device since all 943 * error recovery strategies would have already been 944 * deployed on the host side. However, if the command 945 * were a pass-through command deal with it appropriately. 946 */ 947 switch (scmnd->cmnd[0]) { 948 case ATA_16: 949 case ATA_12: 950 set_host_byte(scmnd, DID_PASSTHROUGH); 951 break; 952 /* 953 * On Some Windows hosts TEST_UNIT_READY command can return 954 * SRB_STATUS_ERROR, let the upper level code deal with it 955 * based on the sense information. 956 */ 957 case TEST_UNIT_READY: 958 break; 959 default: 960 set_host_byte(scmnd, DID_ERROR); 961 } 962 break; 963 case SRB_STATUS_INVALID_LUN: 964 set_host_byte(scmnd, DID_NO_CONNECT); 965 do_work = true; 966 process_err_fn = storvsc_remove_lun; 967 break; 968 case SRB_STATUS_ABORTED: 969 if (vm_srb->srb_status & SRB_STATUS_AUTOSENSE_VALID && 970 (asc == 0x2a) && (ascq == 0x9)) { 971 do_work = true; 972 process_err_fn = storvsc_device_scan; 973 /* 974 * Retry the I/O that trigerred this. 975 */ 976 set_host_byte(scmnd, DID_REQUEUE); 977 } 978 break; 979 } 980 981 if (!do_work) 982 return; 983 984 /* 985 * We need to schedule work to process this error; schedule it. 986 */ 987 wrk = kmalloc(sizeof(struct storvsc_scan_work), GFP_ATOMIC); 988 if (!wrk) { 989 set_host_byte(scmnd, DID_TARGET_FAILURE); 990 return; 991 } 992 993 wrk->host = host; 994 wrk->lun = vm_srb->lun; 995 wrk->tgt_id = vm_srb->target_id; 996 INIT_WORK(&wrk->work, process_err_fn); 997 queue_work(host_dev->handle_error_wq, &wrk->work); 998 } 999 1000 1001 static void storvsc_command_completion(struct storvsc_cmd_request *cmd_request, 1002 struct storvsc_device *stor_dev) 1003 { 1004 struct scsi_cmnd *scmnd = cmd_request->cmd; 1005 struct scsi_sense_hdr sense_hdr; 1006 struct vmscsi_request *vm_srb; 1007 u32 data_transfer_length; 1008 struct Scsi_Host *host; 1009 u32 payload_sz = cmd_request->payload_sz; 1010 void *payload = cmd_request->payload; 1011 1012 host = stor_dev->host; 1013 1014 vm_srb = &cmd_request->vstor_packet.vm_srb; 1015 data_transfer_length = vm_srb->data_transfer_length; 1016 1017 scmnd->result = vm_srb->scsi_status; 1018 1019 if (scmnd->result) { 1020 if (scsi_normalize_sense(scmnd->sense_buffer, 1021 SCSI_SENSE_BUFFERSIZE, &sense_hdr) && 1022 !(sense_hdr.sense_key == NOT_READY && 1023 sense_hdr.asc == 0x03A) && 1024 do_logging(STORVSC_LOGGING_ERROR)) 1025 scsi_print_sense_hdr(scmnd->device, "storvsc", 1026 &sense_hdr); 1027 } 1028 1029 if (vm_srb->srb_status != SRB_STATUS_SUCCESS) { 1030 storvsc_handle_error(vm_srb, scmnd, host, sense_hdr.asc, 1031 sense_hdr.ascq); 1032 /* 1033 * The Windows driver set data_transfer_length on 1034 * SRB_STATUS_DATA_OVERRUN. On other errors, this value 1035 * is untouched. In these cases we set it to 0. 1036 */ 1037 if (vm_srb->srb_status != SRB_STATUS_DATA_OVERRUN) 1038 data_transfer_length = 0; 1039 } 1040 1041 scsi_set_resid(scmnd, 1042 cmd_request->payload->range.len - data_transfer_length); 1043 1044 scmnd->scsi_done(scmnd); 1045 1046 if (payload_sz > 1047 sizeof(struct vmbus_channel_packet_multipage_buffer)) 1048 kfree(payload); 1049 } 1050 1051 static void storvsc_on_io_completion(struct storvsc_device *stor_device, 1052 struct vstor_packet *vstor_packet, 1053 struct storvsc_cmd_request *request) 1054 { 1055 struct vstor_packet *stor_pkt; 1056 struct hv_device *device = stor_device->device; 1057 1058 stor_pkt = &request->vstor_packet; 1059 1060 /* 1061 * The current SCSI handling on the host side does 1062 * not correctly handle: 1063 * INQUIRY command with page code parameter set to 0x80 1064 * MODE_SENSE command with cmd[2] == 0x1c 1065 * 1066 * Setup srb and scsi status so this won't be fatal. 1067 * We do this so we can distinguish truly fatal failues 1068 * (srb status == 0x4) and off-line the device in that case. 1069 */ 1070 1071 if ((stor_pkt->vm_srb.cdb[0] == INQUIRY) || 1072 (stor_pkt->vm_srb.cdb[0] == MODE_SENSE)) { 1073 vstor_packet->vm_srb.scsi_status = 0; 1074 vstor_packet->vm_srb.srb_status = SRB_STATUS_SUCCESS; 1075 } 1076 1077 1078 /* Copy over the status...etc */ 1079 stor_pkt->vm_srb.scsi_status = vstor_packet->vm_srb.scsi_status; 1080 stor_pkt->vm_srb.srb_status = vstor_packet->vm_srb.srb_status; 1081 stor_pkt->vm_srb.sense_info_length = 1082 vstor_packet->vm_srb.sense_info_length; 1083 1084 if (vstor_packet->vm_srb.scsi_status != 0 || 1085 vstor_packet->vm_srb.srb_status != SRB_STATUS_SUCCESS) 1086 storvsc_log(device, STORVSC_LOGGING_WARN, 1087 "cmd 0x%x scsi status 0x%x srb status 0x%x\n", 1088 stor_pkt->vm_srb.cdb[0], 1089 vstor_packet->vm_srb.scsi_status, 1090 vstor_packet->vm_srb.srb_status); 1091 1092 if ((vstor_packet->vm_srb.scsi_status & 0xFF) == 0x02) { 1093 /* CHECK_CONDITION */ 1094 if (vstor_packet->vm_srb.srb_status & 1095 SRB_STATUS_AUTOSENSE_VALID) { 1096 /* autosense data available */ 1097 1098 storvsc_log(device, STORVSC_LOGGING_WARN, 1099 "stor pkt %p autosense data valid - len %d\n", 1100 request, vstor_packet->vm_srb.sense_info_length); 1101 1102 memcpy(request->cmd->sense_buffer, 1103 vstor_packet->vm_srb.sense_data, 1104 vstor_packet->vm_srb.sense_info_length); 1105 1106 } 1107 } 1108 1109 stor_pkt->vm_srb.data_transfer_length = 1110 vstor_packet->vm_srb.data_transfer_length; 1111 1112 storvsc_command_completion(request, stor_device); 1113 1114 if (atomic_dec_and_test(&stor_device->num_outstanding_req) && 1115 stor_device->drain_notify) 1116 wake_up(&stor_device->waiting_to_drain); 1117 1118 1119 } 1120 1121 static void storvsc_on_receive(struct storvsc_device *stor_device, 1122 struct vstor_packet *vstor_packet, 1123 struct storvsc_cmd_request *request) 1124 { 1125 struct hv_host_device *host_dev; 1126 switch (vstor_packet->operation) { 1127 case VSTOR_OPERATION_COMPLETE_IO: 1128 storvsc_on_io_completion(stor_device, vstor_packet, request); 1129 break; 1130 1131 case VSTOR_OPERATION_REMOVE_DEVICE: 1132 case VSTOR_OPERATION_ENUMERATE_BUS: 1133 host_dev = shost_priv(stor_device->host); 1134 queue_work( 1135 host_dev->handle_error_wq, &host_dev->host_scan_work); 1136 break; 1137 1138 case VSTOR_OPERATION_FCHBA_DATA: 1139 cache_wwn(stor_device, vstor_packet); 1140 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1141 fc_host_node_name(stor_device->host) = stor_device->node_name; 1142 fc_host_port_name(stor_device->host) = stor_device->port_name; 1143 #endif 1144 break; 1145 default: 1146 break; 1147 } 1148 } 1149 1150 static void storvsc_on_channel_callback(void *context) 1151 { 1152 struct vmbus_channel *channel = (struct vmbus_channel *)context; 1153 const struct vmpacket_descriptor *desc; 1154 struct hv_device *device; 1155 struct storvsc_device *stor_device; 1156 1157 if (channel->primary_channel != NULL) 1158 device = channel->primary_channel->device_obj; 1159 else 1160 device = channel->device_obj; 1161 1162 stor_device = get_in_stor_device(device); 1163 if (!stor_device) 1164 return; 1165 1166 foreach_vmbus_pkt(desc, channel) { 1167 void *packet = hv_pkt_data(desc); 1168 struct storvsc_cmd_request *request; 1169 1170 request = (struct storvsc_cmd_request *) 1171 ((unsigned long)desc->trans_id); 1172 1173 if (request == &stor_device->init_request || 1174 request == &stor_device->reset_request) { 1175 memcpy(&request->vstor_packet, packet, 1176 (sizeof(struct vstor_packet) - vmscsi_size_delta)); 1177 complete(&request->wait_event); 1178 } else { 1179 storvsc_on_receive(stor_device, packet, request); 1180 } 1181 } 1182 } 1183 1184 static int storvsc_connect_to_vsp(struct hv_device *device, u32 ring_size, 1185 bool is_fc) 1186 { 1187 struct vmstorage_channel_properties props; 1188 int ret; 1189 1190 memset(&props, 0, sizeof(struct vmstorage_channel_properties)); 1191 1192 ret = vmbus_open(device->channel, 1193 ring_size, 1194 ring_size, 1195 (void *)&props, 1196 sizeof(struct vmstorage_channel_properties), 1197 storvsc_on_channel_callback, device->channel); 1198 1199 if (ret != 0) 1200 return ret; 1201 1202 ret = storvsc_channel_init(device, is_fc); 1203 1204 return ret; 1205 } 1206 1207 static int storvsc_dev_remove(struct hv_device *device) 1208 { 1209 struct storvsc_device *stor_device; 1210 1211 stor_device = hv_get_drvdata(device); 1212 1213 stor_device->destroy = true; 1214 1215 /* Make sure flag is set before waiting */ 1216 wmb(); 1217 1218 /* 1219 * At this point, all outbound traffic should be disable. We 1220 * only allow inbound traffic (responses) to proceed so that 1221 * outstanding requests can be completed. 1222 */ 1223 1224 storvsc_wait_to_drain(stor_device); 1225 1226 /* 1227 * Since we have already drained, we don't need to busy wait 1228 * as was done in final_release_stor_device() 1229 * Note that we cannot set the ext pointer to NULL until 1230 * we have drained - to drain the outgoing packets, we need to 1231 * allow incoming packets. 1232 */ 1233 hv_set_drvdata(device, NULL); 1234 1235 /* Close the channel */ 1236 vmbus_close(device->channel); 1237 1238 kfree(stor_device->stor_chns); 1239 kfree(stor_device); 1240 return 0; 1241 } 1242 1243 static struct vmbus_channel *get_og_chn(struct storvsc_device *stor_device, 1244 u16 q_num) 1245 { 1246 u16 slot = 0; 1247 u16 hash_qnum; 1248 const struct cpumask *node_mask; 1249 int num_channels, tgt_cpu; 1250 1251 if (stor_device->num_sc == 0) 1252 return stor_device->device->channel; 1253 1254 /* 1255 * Our channel array is sparsley populated and we 1256 * initiated I/O on a processor/hw-q that does not 1257 * currently have a designated channel. Fix this. 1258 * The strategy is simple: 1259 * I. Ensure NUMA locality 1260 * II. Distribute evenly (best effort) 1261 * III. Mapping is persistent. 1262 */ 1263 1264 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1265 1266 num_channels = 0; 1267 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1268 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1269 num_channels++; 1270 } 1271 if (num_channels == 0) 1272 return stor_device->device->channel; 1273 1274 hash_qnum = q_num; 1275 while (hash_qnum >= num_channels) 1276 hash_qnum -= num_channels; 1277 1278 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1279 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1280 continue; 1281 if (slot == hash_qnum) 1282 break; 1283 slot++; 1284 } 1285 1286 stor_device->stor_chns[q_num] = stor_device->stor_chns[tgt_cpu]; 1287 1288 return stor_device->stor_chns[q_num]; 1289 } 1290 1291 1292 static int storvsc_do_io(struct hv_device *device, 1293 struct storvsc_cmd_request *request, u16 q_num) 1294 { 1295 struct storvsc_device *stor_device; 1296 struct vstor_packet *vstor_packet; 1297 struct vmbus_channel *outgoing_channel, *channel; 1298 int ret = 0; 1299 const struct cpumask *node_mask; 1300 int tgt_cpu; 1301 1302 vstor_packet = &request->vstor_packet; 1303 stor_device = get_out_stor_device(device); 1304 1305 if (!stor_device) 1306 return -ENODEV; 1307 1308 1309 request->device = device; 1310 /* 1311 * Select an an appropriate channel to send the request out. 1312 */ 1313 if (stor_device->stor_chns[q_num] != NULL) { 1314 outgoing_channel = stor_device->stor_chns[q_num]; 1315 if (outgoing_channel->target_cpu == q_num) { 1316 /* 1317 * Ideally, we want to pick a different channel if 1318 * available on the same NUMA node. 1319 */ 1320 node_mask = cpumask_of_node(cpu_to_node(q_num)); 1321 for_each_cpu_wrap(tgt_cpu, 1322 &stor_device->alloced_cpus, q_num + 1) { 1323 if (!cpumask_test_cpu(tgt_cpu, node_mask)) 1324 continue; 1325 if (tgt_cpu == q_num) 1326 continue; 1327 channel = stor_device->stor_chns[tgt_cpu]; 1328 if (hv_get_avail_to_write_percent( 1329 &channel->outbound) 1330 > ring_avail_percent_lowater) { 1331 outgoing_channel = channel; 1332 goto found_channel; 1333 } 1334 } 1335 1336 /* 1337 * All the other channels on the same NUMA node are 1338 * busy. Try to use the channel on the current CPU 1339 */ 1340 if (hv_get_avail_to_write_percent( 1341 &outgoing_channel->outbound) 1342 > ring_avail_percent_lowater) 1343 goto found_channel; 1344 1345 /* 1346 * If we reach here, all the channels on the current 1347 * NUMA node are busy. Try to find a channel in 1348 * other NUMA nodes 1349 */ 1350 for_each_cpu(tgt_cpu, &stor_device->alloced_cpus) { 1351 if (cpumask_test_cpu(tgt_cpu, node_mask)) 1352 continue; 1353 channel = stor_device->stor_chns[tgt_cpu]; 1354 if (hv_get_avail_to_write_percent( 1355 &channel->outbound) 1356 > ring_avail_percent_lowater) { 1357 outgoing_channel = channel; 1358 goto found_channel; 1359 } 1360 } 1361 } 1362 } else { 1363 outgoing_channel = get_og_chn(stor_device, q_num); 1364 } 1365 1366 found_channel: 1367 vstor_packet->flags |= REQUEST_COMPLETION_FLAG; 1368 1369 vstor_packet->vm_srb.length = (sizeof(struct vmscsi_request) - 1370 vmscsi_size_delta); 1371 1372 1373 vstor_packet->vm_srb.sense_info_length = sense_buffer_size; 1374 1375 1376 vstor_packet->vm_srb.data_transfer_length = 1377 request->payload->range.len; 1378 1379 vstor_packet->operation = VSTOR_OPERATION_EXECUTE_SRB; 1380 1381 if (request->payload->range.len) { 1382 1383 ret = vmbus_sendpacket_mpb_desc(outgoing_channel, 1384 request->payload, request->payload_sz, 1385 vstor_packet, 1386 (sizeof(struct vstor_packet) - 1387 vmscsi_size_delta), 1388 (unsigned long)request); 1389 } else { 1390 ret = vmbus_sendpacket(outgoing_channel, vstor_packet, 1391 (sizeof(struct vstor_packet) - 1392 vmscsi_size_delta), 1393 (unsigned long)request, 1394 VM_PKT_DATA_INBAND, 1395 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); 1396 } 1397 1398 if (ret != 0) 1399 return ret; 1400 1401 atomic_inc(&stor_device->num_outstanding_req); 1402 1403 return ret; 1404 } 1405 1406 static int storvsc_device_alloc(struct scsi_device *sdevice) 1407 { 1408 /* 1409 * Set blist flag to permit the reading of the VPD pages even when 1410 * the target may claim SPC-2 compliance. MSFT targets currently 1411 * claim SPC-2 compliance while they implement post SPC-2 features. 1412 * With this flag we can correctly handle WRITE_SAME_16 issues. 1413 * 1414 * Hypervisor reports SCSI_UNKNOWN type for DVD ROM device but 1415 * still supports REPORT LUN. 1416 */ 1417 sdevice->sdev_bflags = BLIST_REPORTLUN2 | BLIST_TRY_VPD_PAGES; 1418 1419 return 0; 1420 } 1421 1422 static int storvsc_device_configure(struct scsi_device *sdevice) 1423 { 1424 blk_queue_rq_timeout(sdevice->request_queue, (storvsc_timeout * HZ)); 1425 1426 sdevice->no_write_same = 1; 1427 1428 /* 1429 * If the host is WIN8 or WIN8 R2, claim conformance to SPC-3 1430 * if the device is a MSFT virtual device. If the host is 1431 * WIN10 or newer, allow write_same. 1432 */ 1433 if (!strncmp(sdevice->vendor, "Msft", 4)) { 1434 switch (vmstor_proto_version) { 1435 case VMSTOR_PROTO_VERSION_WIN8: 1436 case VMSTOR_PROTO_VERSION_WIN8_1: 1437 sdevice->scsi_level = SCSI_SPC_3; 1438 break; 1439 } 1440 1441 if (vmstor_proto_version >= VMSTOR_PROTO_VERSION_WIN10) 1442 sdevice->no_write_same = 0; 1443 } 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 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1530 if (scmnd->device->host->transportt == fc_transport_template) 1531 return fc_eh_timed_out(scmnd); 1532 #endif 1533 return BLK_EH_RESET_TIMER; 1534 } 1535 1536 static bool storvsc_scsi_cmd_ok(struct scsi_cmnd *scmnd) 1537 { 1538 bool allowed = true; 1539 u8 scsi_op = scmnd->cmnd[0]; 1540 1541 switch (scsi_op) { 1542 /* the host does not handle WRITE_SAME, log accident usage */ 1543 case WRITE_SAME: 1544 /* 1545 * smartd sends this command and the host does not handle 1546 * this. So, don't send it. 1547 */ 1548 case SET_WINDOW: 1549 scmnd->result = ILLEGAL_REQUEST << 16; 1550 allowed = false; 1551 break; 1552 default: 1553 break; 1554 } 1555 return allowed; 1556 } 1557 1558 static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) 1559 { 1560 int ret; 1561 struct hv_host_device *host_dev = shost_priv(host); 1562 struct hv_device *dev = host_dev->dev; 1563 struct storvsc_cmd_request *cmd_request = scsi_cmd_priv(scmnd); 1564 int i; 1565 struct scatterlist *sgl; 1566 unsigned int sg_count = 0; 1567 struct vmscsi_request *vm_srb; 1568 struct scatterlist *cur_sgl; 1569 struct vmbus_packet_mpb_array *payload; 1570 u32 payload_sz; 1571 u32 length; 1572 1573 if (vmstor_proto_version <= VMSTOR_PROTO_VERSION_WIN8) { 1574 /* 1575 * On legacy hosts filter unimplemented commands. 1576 * Future hosts are expected to correctly handle 1577 * unsupported commands. Furthermore, it is 1578 * possible that some of the currently 1579 * unsupported commands maybe supported in 1580 * future versions of the host. 1581 */ 1582 if (!storvsc_scsi_cmd_ok(scmnd)) { 1583 scmnd->scsi_done(scmnd); 1584 return 0; 1585 } 1586 } 1587 1588 /* Setup the cmd request */ 1589 cmd_request->cmd = scmnd; 1590 1591 vm_srb = &cmd_request->vstor_packet.vm_srb; 1592 vm_srb->win8_extension.time_out_value = 60; 1593 1594 vm_srb->win8_extension.srb_flags |= 1595 SRB_FLAGS_DISABLE_SYNCH_TRANSFER; 1596 1597 if (scmnd->device->tagged_supported) { 1598 vm_srb->win8_extension.srb_flags |= 1599 (SRB_FLAGS_QUEUE_ACTION_ENABLE | SRB_FLAGS_NO_QUEUE_FREEZE); 1600 vm_srb->win8_extension.queue_tag = SP_UNTAGGED; 1601 vm_srb->win8_extension.queue_action = SRB_SIMPLE_TAG_REQUEST; 1602 } 1603 1604 /* Build the SRB */ 1605 switch (scmnd->sc_data_direction) { 1606 case DMA_TO_DEVICE: 1607 vm_srb->data_in = WRITE_TYPE; 1608 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_OUT; 1609 break; 1610 case DMA_FROM_DEVICE: 1611 vm_srb->data_in = READ_TYPE; 1612 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_DATA_IN; 1613 break; 1614 case DMA_NONE: 1615 vm_srb->data_in = UNKNOWN_TYPE; 1616 vm_srb->win8_extension.srb_flags |= SRB_FLAGS_NO_DATA_TRANSFER; 1617 break; 1618 default: 1619 /* 1620 * This is DMA_BIDIRECTIONAL or something else we are never 1621 * supposed to see here. 1622 */ 1623 WARN(1, "Unexpected data direction: %d\n", 1624 scmnd->sc_data_direction); 1625 return -EINVAL; 1626 } 1627 1628 1629 vm_srb->port_number = host_dev->port; 1630 vm_srb->path_id = scmnd->device->channel; 1631 vm_srb->target_id = scmnd->device->id; 1632 vm_srb->lun = scmnd->device->lun; 1633 1634 vm_srb->cdb_length = scmnd->cmd_len; 1635 1636 memcpy(vm_srb->cdb, scmnd->cmnd, vm_srb->cdb_length); 1637 1638 sgl = (struct scatterlist *)scsi_sglist(scmnd); 1639 sg_count = scsi_sg_count(scmnd); 1640 1641 length = scsi_bufflen(scmnd); 1642 payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb; 1643 payload_sz = sizeof(cmd_request->mpb); 1644 1645 if (sg_count) { 1646 if (sg_count > MAX_PAGE_BUFFER_COUNT) { 1647 1648 payload_sz = (sg_count * sizeof(u64) + 1649 sizeof(struct vmbus_packet_mpb_array)); 1650 payload = kzalloc(payload_sz, GFP_ATOMIC); 1651 if (!payload) 1652 return SCSI_MLQUEUE_DEVICE_BUSY; 1653 } 1654 1655 payload->range.len = length; 1656 payload->range.offset = sgl[0].offset; 1657 1658 cur_sgl = sgl; 1659 for (i = 0; i < sg_count; i++) { 1660 payload->range.pfn_array[i] = 1661 page_to_pfn(sg_page((cur_sgl))); 1662 cur_sgl = sg_next(cur_sgl); 1663 } 1664 } 1665 1666 cmd_request->payload = payload; 1667 cmd_request->payload_sz = payload_sz; 1668 1669 /* Invokes the vsc to start an IO */ 1670 ret = storvsc_do_io(dev, cmd_request, get_cpu()); 1671 put_cpu(); 1672 1673 if (ret == -EAGAIN) { 1674 if (payload_sz > sizeof(cmd_request->mpb)) 1675 kfree(payload); 1676 /* no more space */ 1677 return SCSI_MLQUEUE_DEVICE_BUSY; 1678 } 1679 1680 return 0; 1681 } 1682 1683 static struct scsi_host_template scsi_driver = { 1684 .module = THIS_MODULE, 1685 .name = "storvsc_host_t", 1686 .cmd_size = sizeof(struct storvsc_cmd_request), 1687 .bios_param = storvsc_get_chs, 1688 .queuecommand = storvsc_queuecommand, 1689 .eh_host_reset_handler = storvsc_host_reset_handler, 1690 .proc_name = "storvsc_host", 1691 .eh_timed_out = storvsc_eh_timed_out, 1692 .slave_alloc = storvsc_device_alloc, 1693 .slave_configure = storvsc_device_configure, 1694 .cmd_per_lun = 2048, 1695 .this_id = -1, 1696 /* Make sure we dont get a sg segment crosses a page boundary */ 1697 .dma_boundary = PAGE_SIZE-1, 1698 /* Ensure there are no gaps in presented sgls */ 1699 .virt_boundary_mask = PAGE_SIZE-1, 1700 .no_write_same = 1, 1701 .track_queue_depth = 1, 1702 .change_queue_depth = storvsc_change_queue_depth, 1703 }; 1704 1705 enum { 1706 SCSI_GUID, 1707 IDE_GUID, 1708 SFC_GUID, 1709 }; 1710 1711 static const struct hv_vmbus_device_id id_table[] = { 1712 /* SCSI guid */ 1713 { HV_SCSI_GUID, 1714 .driver_data = SCSI_GUID 1715 }, 1716 /* IDE guid */ 1717 { HV_IDE_GUID, 1718 .driver_data = IDE_GUID 1719 }, 1720 /* Fibre Channel GUID */ 1721 { 1722 HV_SYNTHFC_GUID, 1723 .driver_data = SFC_GUID 1724 }, 1725 { }, 1726 }; 1727 1728 MODULE_DEVICE_TABLE(vmbus, id_table); 1729 1730 static int storvsc_probe(struct hv_device *device, 1731 const struct hv_vmbus_device_id *dev_id) 1732 { 1733 int ret; 1734 int num_cpus = num_online_cpus(); 1735 struct Scsi_Host *host; 1736 struct hv_host_device *host_dev; 1737 bool dev_is_ide = ((dev_id->driver_data == IDE_GUID) ? true : false); 1738 bool is_fc = ((dev_id->driver_data == SFC_GUID) ? true : false); 1739 int target = 0; 1740 struct storvsc_device *stor_device; 1741 int max_luns_per_target; 1742 int max_targets; 1743 int max_channels; 1744 int max_sub_channels = 0; 1745 1746 /* 1747 * Based on the windows host we are running on, 1748 * set state to properly communicate with the host. 1749 */ 1750 1751 if (vmbus_proto_version < VERSION_WIN8) { 1752 max_luns_per_target = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1753 max_targets = STORVSC_IDE_MAX_TARGETS; 1754 max_channels = STORVSC_IDE_MAX_CHANNELS; 1755 } else { 1756 max_luns_per_target = STORVSC_MAX_LUNS_PER_TARGET; 1757 max_targets = STORVSC_MAX_TARGETS; 1758 max_channels = STORVSC_MAX_CHANNELS; 1759 /* 1760 * On Windows8 and above, we support sub-channels for storage 1761 * on SCSI and FC controllers. 1762 * The number of sub-channels offerred is based on the number of 1763 * VCPUs in the guest. 1764 */ 1765 if (!dev_is_ide) 1766 max_sub_channels = 1767 (num_cpus - 1) / storvsc_vcpus_per_sub_channel; 1768 } 1769 1770 scsi_driver.can_queue = max_outstanding_req_per_channel * 1771 (max_sub_channels + 1) * 1772 (100 - ring_avail_percent_lowater) / 100; 1773 1774 host = scsi_host_alloc(&scsi_driver, 1775 sizeof(struct hv_host_device)); 1776 if (!host) 1777 return -ENOMEM; 1778 1779 host_dev = shost_priv(host); 1780 memset(host_dev, 0, sizeof(struct hv_host_device)); 1781 1782 host_dev->port = host->host_no; 1783 host_dev->dev = device; 1784 host_dev->host = host; 1785 1786 1787 stor_device = kzalloc(sizeof(struct storvsc_device), GFP_KERNEL); 1788 if (!stor_device) { 1789 ret = -ENOMEM; 1790 goto err_out0; 1791 } 1792 1793 stor_device->destroy = false; 1794 init_waitqueue_head(&stor_device->waiting_to_drain); 1795 stor_device->device = device; 1796 stor_device->host = host; 1797 hv_set_drvdata(device, stor_device); 1798 1799 stor_device->port_number = host->host_no; 1800 ret = storvsc_connect_to_vsp(device, storvsc_ringbuffer_size, is_fc); 1801 if (ret) 1802 goto err_out1; 1803 1804 host_dev->path = stor_device->path_id; 1805 host_dev->target = stor_device->target_id; 1806 1807 switch (dev_id->driver_data) { 1808 case SFC_GUID: 1809 host->max_lun = STORVSC_FC_MAX_LUNS_PER_TARGET; 1810 host->max_id = STORVSC_FC_MAX_TARGETS; 1811 host->max_channel = STORVSC_FC_MAX_CHANNELS - 1; 1812 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1813 host->transportt = fc_transport_template; 1814 #endif 1815 break; 1816 1817 case SCSI_GUID: 1818 host->max_lun = max_luns_per_target; 1819 host->max_id = max_targets; 1820 host->max_channel = max_channels - 1; 1821 break; 1822 1823 default: 1824 host->max_lun = STORVSC_IDE_MAX_LUNS_PER_TARGET; 1825 host->max_id = STORVSC_IDE_MAX_TARGETS; 1826 host->max_channel = STORVSC_IDE_MAX_CHANNELS - 1; 1827 break; 1828 } 1829 /* max cmd length */ 1830 host->max_cmd_len = STORVSC_MAX_CMD_LEN; 1831 1832 /* 1833 * set the table size based on the info we got 1834 * from the host. 1835 */ 1836 host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); 1837 /* 1838 * Set the number of HW queues we are supporting. 1839 */ 1840 host->nr_hw_queues = num_present_cpus(); 1841 1842 /* 1843 * Set the error handler work queue. 1844 */ 1845 host_dev->handle_error_wq = 1846 alloc_ordered_workqueue("storvsc_error_wq_%d", 1847 WQ_MEM_RECLAIM, 1848 host->host_no); 1849 if (!host_dev->handle_error_wq) 1850 goto err_out2; 1851 INIT_WORK(&host_dev->host_scan_work, storvsc_host_scan); 1852 /* Register the HBA and start the scsi bus scan */ 1853 ret = scsi_add_host(host, &device->device); 1854 if (ret != 0) 1855 goto err_out3; 1856 1857 if (!dev_is_ide) { 1858 scsi_scan_host(host); 1859 } else { 1860 target = (device->dev_instance.b[5] << 8 | 1861 device->dev_instance.b[4]); 1862 ret = scsi_add_device(host, 0, target, 0); 1863 if (ret) 1864 goto err_out4; 1865 } 1866 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1867 if (host->transportt == fc_transport_template) { 1868 struct fc_rport_identifiers ids = { 1869 .roles = FC_PORT_ROLE_FCP_DUMMY_INITIATOR, 1870 }; 1871 1872 fc_host_node_name(host) = stor_device->node_name; 1873 fc_host_port_name(host) = stor_device->port_name; 1874 stor_device->rport = fc_remote_port_add(host, 0, &ids); 1875 if (!stor_device->rport) { 1876 ret = -ENOMEM; 1877 goto err_out4; 1878 } 1879 } 1880 #endif 1881 return 0; 1882 1883 err_out4: 1884 scsi_remove_host(host); 1885 1886 err_out3: 1887 destroy_workqueue(host_dev->handle_error_wq); 1888 1889 err_out2: 1890 /* 1891 * Once we have connected with the host, we would need to 1892 * to invoke storvsc_dev_remove() to rollback this state and 1893 * this call also frees up the stor_device; hence the jump around 1894 * err_out1 label. 1895 */ 1896 storvsc_dev_remove(device); 1897 goto err_out0; 1898 1899 err_out1: 1900 kfree(stor_device->stor_chns); 1901 kfree(stor_device); 1902 1903 err_out0: 1904 scsi_host_put(host); 1905 return ret; 1906 } 1907 1908 /* Change a scsi target's queue depth */ 1909 static int storvsc_change_queue_depth(struct scsi_device *sdev, int queue_depth) 1910 { 1911 if (queue_depth > scsi_driver.can_queue) 1912 queue_depth = scsi_driver.can_queue; 1913 1914 return scsi_change_queue_depth(sdev, queue_depth); 1915 } 1916 1917 static int storvsc_remove(struct hv_device *dev) 1918 { 1919 struct storvsc_device *stor_device = hv_get_drvdata(dev); 1920 struct Scsi_Host *host = stor_device->host; 1921 struct hv_host_device *host_dev = shost_priv(host); 1922 1923 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1924 if (host->transportt == fc_transport_template) { 1925 fc_remote_port_delete(stor_device->rport); 1926 fc_remove_host(host); 1927 } 1928 #endif 1929 destroy_workqueue(host_dev->handle_error_wq); 1930 scsi_remove_host(host); 1931 storvsc_dev_remove(dev); 1932 scsi_host_put(host); 1933 1934 return 0; 1935 } 1936 1937 static struct hv_driver storvsc_drv = { 1938 .name = KBUILD_MODNAME, 1939 .id_table = id_table, 1940 .probe = storvsc_probe, 1941 .remove = storvsc_remove, 1942 .driver = { 1943 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 1944 }, 1945 }; 1946 1947 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1948 static struct fc_function_template fc_transport_functions = { 1949 .show_host_node_name = 1, 1950 .show_host_port_name = 1, 1951 }; 1952 #endif 1953 1954 static int __init storvsc_drv_init(void) 1955 { 1956 int ret; 1957 1958 /* 1959 * Divide the ring buffer data size (which is 1 page less 1960 * than the ring buffer size since that page is reserved for 1961 * the ring buffer indices) by the max request size (which is 1962 * vmbus_channel_packet_multipage_buffer + struct vstor_packet + u64) 1963 */ 1964 max_outstanding_req_per_channel = 1965 ((storvsc_ringbuffer_size - PAGE_SIZE) / 1966 ALIGN(MAX_MULTIPAGE_BUFFER_PACKET + 1967 sizeof(struct vstor_packet) + sizeof(u64) - 1968 vmscsi_size_delta, 1969 sizeof(u64))); 1970 1971 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1972 fc_transport_template = fc_attach_transport(&fc_transport_functions); 1973 if (!fc_transport_template) 1974 return -ENODEV; 1975 #endif 1976 1977 ret = vmbus_driver_register(&storvsc_drv); 1978 1979 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1980 if (ret) 1981 fc_release_transport(fc_transport_template); 1982 #endif 1983 1984 return ret; 1985 } 1986 1987 static void __exit storvsc_drv_exit(void) 1988 { 1989 vmbus_driver_unregister(&storvsc_drv); 1990 #if IS_ENABLED(CONFIG_SCSI_FC_ATTRS) 1991 fc_release_transport(fc_transport_template); 1992 #endif 1993 } 1994 1995 MODULE_LICENSE("GPL"); 1996 MODULE_DESCRIPTION("Microsoft Hyper-V virtual storage driver"); 1997 module_init(storvsc_drv_init); 1998 module_exit(storvsc_drv_exit); 1999