1 /****************************************************************************** 2 * blkif.h 3 * 4 * Unified block-device I/O interface for Xen guest OSes. 5 * 6 * SPDX-License-Identifier: MIT 7 * 8 * Copyright (c) 2003-2004, Keir Fraser 9 * Copyright (c) 2012, Spectra Logic Corporation 10 */ 11 12 #ifndef __XEN_PUBLIC_IO_BLKIF_H__ 13 #define __XEN_PUBLIC_IO_BLKIF_H__ 14 15 #include "ring.h" 16 #include "../grant_table.h" 17 18 /* 19 * Front->back notifications: When enqueuing a new request, sending a 20 * notification can be made conditional on req_event (i.e., the generic 21 * hold-off mechanism provided by the ring macros). Backends must set 22 * req_event appropriately (e.g., using RING_FINAL_CHECK_FOR_REQUESTS()). 23 * 24 * Back->front notifications: When enqueuing a new response, sending a 25 * notification can be made conditional on rsp_event (i.e., the generic 26 * hold-off mechanism provided by the ring macros). Frontends must set 27 * rsp_event appropriately (e.g., using RING_FINAL_CHECK_FOR_RESPONSES()). 28 */ 29 30 #ifndef blkif_vdev_t 31 #define blkif_vdev_t UINT16 32 #endif 33 #define blkif_sector_t UINT64 34 35 /* 36 * Feature and Parameter Negotiation 37 * ================================= 38 * The two halves of a Xen block driver utilize nodes within the XenStore to 39 * communicate capabilities and to negotiate operating parameters. This 40 * section enumerates these nodes which reside in the respective front and 41 * backend portions of the XenStore, following the XenBus convention. 42 * 43 * All data in the XenStore is stored as strings. Nodes specifying numeric 44 * values are encoded in decimal. Integer value ranges listed below are 45 * expressed as fixed sized integer types capable of storing the conversion 46 * of a properly formatted node string, without loss of information. 47 * 48 * Any specified default value is in effect if the corresponding XenBus node 49 * is not present in the XenStore. 50 * 51 * XenStore nodes in sections marked "PRIVATE" are solely for use by the 52 * driver side whose XenBus tree contains them. 53 * 54 * XenStore nodes marked "DEPRECATED" in their notes section should only be 55 * used to provide interoperability with legacy implementations. 56 * 57 * See the XenBus state transition diagram below for details on when XenBus 58 * nodes must be published and when they can be queried. 59 * 60 ***************************************************************************** 61 * Backend XenBus Nodes 62 ***************************************************************************** 63 * 64 *------------------ Backend Device Identification (PRIVATE) ------------------ 65 * 66 * mode 67 * Values: "r" (read only), "w" (writable) 68 * 69 * The read or write access permissions to the backing store to be 70 * granted to the frontend. 71 * 72 * params 73 * Values: string 74 * 75 * A free formatted string providing sufficient information for the 76 * backend driver to open the backing device. (e.g. the path to the 77 * file or block device representing the backing store.) 78 * 79 * type 80 * Values: "file", "phy", "tap" 81 * 82 * The type of the backing device/object. 83 * 84 *--------------------------------- Features --------------------------------- 85 * 86 * feature-barrier 87 * Values: 0/1 (boolean) 88 * Default Value: 0 89 * 90 * A value of "1" indicates that the backend can process requests 91 * containing the BLKIF_OP_WRITE_BARRIER request opcode. Requests 92 * of this type may still be returned at any time with the 93 * BLKIF_RSP_EOPNOTSUPP result code. 94 * 95 * feature-flush-cache 96 * Values: 0/1 (boolean) 97 * Default Value: 0 98 * 99 * A value of "1" indicates that the backend can process requests 100 * containing the BLKIF_OP_FLUSH_DISKCACHE request opcode. Requests 101 * of this type may still be returned at any time with the 102 * BLKIF_RSP_EOPNOTSUPP result code. 103 * 104 * feature-discard 105 * Values: 0/1 (boolean) 106 * Default Value: 0 107 * 108 * A value of "1" indicates that the backend can process requests 109 * containing the BLKIF_OP_DISCARD request opcode. Requests 110 * of this type may still be returned at any time with the 111 * BLKIF_RSP_EOPNOTSUPP result code. 112 * 113 * feature-persistent 114 * Values: 0/1 (boolean) 115 * Default Value: 0 116 * Notes: 7 117 * 118 * A value of "1" indicates that the backend can keep the grants used 119 * by the frontend driver mapped, so the same set of grants should be 120 * used in all transactions. The maximum number of grants the backend 121 * can map persistently depends on the implementation, but ideally it 122 * should be RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. Using this 123 * feature the backend doesn't need to unmap each grant, preventing 124 * costly TLB flushes. The backend driver should only map grants 125 * persistently if the frontend supports it. If a backend driver chooses 126 * to use the persistent protocol when the frontend doesn't support it, 127 * it will probably hit the maximum number of persistently mapped grants 128 * (due to the fact that the frontend won't be reusing the same grants), 129 * and fall back to non-persistent mode. Backend implementations may 130 * shrink or expand the number of persistently mapped grants without 131 * notifying the frontend depending on memory constraints (this might 132 * cause a performance degradation). 133 * 134 * If a backend driver wants to limit the maximum number of persistently 135 * mapped grants to a value less than RING_SIZE * 136 * BLKIF_MAX_SEGMENTS_PER_REQUEST a LRU strategy should be used to 137 * discard the grants that are less commonly used. Using a LRU in the 138 * backend driver paired with a LIFO queue in the frontend will 139 * allow us to have better performance in this scenario. 140 * 141 *----------------------- Request Transport Parameters ------------------------ 142 * 143 * max-ring-page-order 144 * Values: <UINT32> 145 * Default Value: 0 146 * Notes: 1, 3 147 * 148 * The maximum supported size of the request ring buffer in units of 149 * lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages, 150 * etc.). 151 * 152 * max-ring-pages 153 * Values: <UINT32> 154 * Default Value: 1 155 * Notes: DEPRECATED, 2, 3 156 * 157 * The maximum supported size of the request ring buffer in units of 158 * machine pages. The value must be a power of 2. 159 * 160 *------------------------- Backend Device Properties ------------------------- 161 * 162 * discard-alignment 163 * Values: <UINT32> 164 * Default Value: 0 165 * Notes: 4, 5 166 * 167 * The offset, in bytes from the beginning of the virtual block device, 168 * to the first, addressable, discard extent on the underlying device. 169 * 170 * discard-granularity 171 * Values: <UINT32> 172 * Default Value: <"sector-size"> 173 * Notes: 4 174 * 175 * The size, in bytes, of the individually addressable discard extents 176 * of the underlying device. 177 * 178 * discard-secure 179 * Values: 0/1 (boolean) 180 * Default Value: 0 181 * Notes: 10 182 * 183 * A value of "1" indicates that the backend can process BLKIF_OP_DISCARD 184 * requests with the BLKIF_DISCARD_SECURE flag set. 185 * 186 * info 187 * Values: <UINT32> (bitmap) 188 * 189 * A collection of bit flags describing attributes of the backing 190 * device. The VDISK_* macros define the meaning of each bit 191 * location. 192 * 193 * sector-size 194 * Values: <UINT32> 195 * 196 * The logical sector size, in bytes, of the backend device. 197 * 198 * physical-sector-size 199 * Values: <UINT32> 200 * 201 * The physical sector size, in bytes, of the backend device. 202 * 203 * sectors 204 * Values: <UINT64> 205 * 206 * The size of the backend device, expressed in units of its logical 207 * sector size ("sector-size"). 208 * 209 ***************************************************************************** 210 * Frontend XenBus Nodes 211 ***************************************************************************** 212 * 213 *----------------------- Request Transport Parameters ----------------------- 214 * 215 * event-channel 216 * Values: <UINT32> 217 * 218 * The identifier of the Xen event channel used to signal activity 219 * in the ring buffer. 220 * 221 * ring-ref 222 * Values: <UINT32> 223 * Notes: 6 224 * 225 * The Xen grant reference granting permission for the backend to map 226 * the sole page in a single page sized ring buffer. 227 * 228 * ring-ref%u 229 * Values: <UINT32> 230 * Notes: 6 231 * 232 * For a frontend providing a multi-page ring, a "number of ring pages" 233 * sized list of nodes, each containing a Xen grant reference granting 234 * permission for the backend to map the page of the ring located 235 * at page index "%u". Page indexes are zero based. 236 * 237 * protocol 238 * Values: string (XEN_IO_PROTO_ABI_*) 239 * Default Value: XEN_IO_PROTO_ABI_NATIVE 240 * 241 * The machine ABI rules governing the format of all ring request and 242 * response structures. 243 * 244 * ring-page-order 245 * Values: <UINT32> 246 * Default Value: 0 247 * Maximum Value: MAX(ffs(max-ring-pages) - 1, max-ring-page-order) 248 * Notes: 1, 3 249 * 250 * The size of the frontend allocated request ring buffer in units 251 * of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages, 252 * etc.). 253 * 254 * num-ring-pages 255 * Values: <UINT32> 256 * Default Value: 1 257 * Maximum Value: MAX(max-ring-pages,(0x1 << max-ring-page-order)) 258 * Notes: DEPRECATED, 2, 3 259 * 260 * The size of the frontend allocated request ring buffer in units of 261 * machine pages. The value must be a power of 2. 262 * 263 * feature-persistent 264 * Values: 0/1 (boolean) 265 * Default Value: 0 266 * Notes: 7, 8, 9 267 * 268 * A value of "1" indicates that the frontend will reuse the same grants 269 * for all transactions, allowing the backend to map them with write 270 * access (even when it should be read-only). If the frontend hits the 271 * maximum number of allowed persistently mapped grants, it can fallback 272 * to non persistent mode. This will cause a performance degradation, 273 * since the backend driver will still try to map those grants 274 * persistently. Since the persistent grants protocol is compatible with 275 * the previous protocol, a frontend driver can choose to work in 276 * persistent mode even when the backend doesn't support it. 277 * 278 * It is recommended that the frontend driver stores the persistently 279 * mapped grants in a LIFO queue, so a subset of all persistently mapped 280 * grants gets used commonly. This is done in case the backend driver 281 * decides to limit the maximum number of persistently mapped grants 282 * to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. 283 * 284 *------------------------- Virtual Device Properties ------------------------- 285 * 286 * device-type 287 * Values: "disk", "cdrom", "floppy", etc. 288 * 289 * virtual-device 290 * Values: <UINT32> 291 * 292 * A value indicating the physical device to virtualize within the 293 * frontend's domain. (e.g. "The first ATA disk", "The third SCSI 294 * disk", etc.) 295 * 296 * See docs/misc/vbd-interface.txt for details on the format of this 297 * value. 298 * 299 * Notes 300 * ----- 301 * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer 302 * PV drivers. 303 * (2) Multi-page ring buffer scheme first used in some RedHat distributions 304 * including a distribution deployed on certain nodes of the Amazon 305 * EC2 cluster. 306 * (3) Support for multi-page ring buffers was implemented independently, 307 * in slightly different forms, by both Citrix and RedHat/Amazon. 308 * For full interoperability, block front and backends should publish 309 * identical ring parameters, adjusted for unit differences, to the 310 * XenStore nodes used in both schemes. 311 * (4) Devices that support discard functionality may internally allocate space 312 * (discardable extents) in units that are larger than the exported logical 313 * block size. If the backing device has such discardable extents the 314 * backend should provide both discard-granularity and discard-alignment. 315 * Providing just one of the two may be considered an error by the frontend. 316 * Backends supporting discard should include discard-granularity and 317 * discard-alignment even if it supports discarding individual sectors. 318 * Frontends should assume discard-alignment == 0 and discard-granularity 319 * == sector size if these keys are missing. 320 * (5) The discard-alignment parameter allows a physical device to be 321 * partitioned into virtual devices that do not necessarily begin or 322 * end on a discardable extent boundary. 323 * (6) When there is only a single page allocated to the request ring, 324 * 'ring-ref' is used to communicate the grant reference for this 325 * page to the backend. When using a multi-page ring, the 'ring-ref' 326 * node is not created. Instead 'ring-ref0' - 'ring-refN' are used. 327 * (7) When using persistent grants data has to be copied from/to the page 328 * where the grant is currently mapped. The overhead of doing this copy 329 * however doesn't suppress the speed improvement of not having to unmap 330 * the grants. 331 * (8) The frontend driver has to allow the backend driver to map all grants 332 * with write access, even when they should be mapped read-only, since 333 * further requests may reuse these grants and require write permissions. 334 * (9) Linux implementation doesn't have a limit on the maximum number of 335 * grants that can be persistently mapped in the frontend driver, but 336 * due to the frontent driver implementation it should never be bigger 337 * than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. 338 *(10) The discard-secure property may be present and will be set to 1 if the 339 * backing device supports secure discard. 340 */ 341 342 /* 343 * STATE DIAGRAMS 344 * 345 ***************************************************************************** 346 * Startup * 347 ***************************************************************************** 348 * 349 * Tool stack creates front and back nodes with state XenbusStateInitialising. 350 * 351 * Front Back 352 * ================================= ===================================== 353 * XenbusStateInitialising XenbusStateInitialising 354 * o Query virtual device o Query backend device identification 355 * properties. data. 356 * o Setup OS device instance. o Open and validate backend device. 357 * o Publish backend features and 358 * transport parameters. 359 * | 360 * | 361 * V 362 * XenbusStateInitWait 363 * 364 * o Query backend features and 365 * transport parameters. 366 * o Allocate and initialize the 367 * request ring. 368 * o Publish transport parameters 369 * that will be in effect during 370 * this connection. 371 * | 372 * | 373 * V 374 * XenbusStateInitialised 375 * 376 * o Query frontend transport parameters. 377 * o Connect to the request ring and 378 * event channel. 379 * o Publish backend device properties. 380 * | 381 * | 382 * V 383 * XenbusStateConnected 384 * 385 * o Query backend device properties. 386 * o Finalize OS virtual device 387 * instance. 388 * | 389 * | 390 * V 391 * XenbusStateConnected 392 * 393 * Note: Drivers that do not support any optional features, or the negotiation 394 * of transport parameters, can skip certain states in the state machine: 395 * 396 * o A frontend may transition to XenbusStateInitialised without 397 * waiting for the backend to enter XenbusStateInitWait. In this 398 * case, default transport parameters are in effect and any 399 * transport parameters published by the frontend must contain 400 * their default values. 401 * 402 * o A backend may transition to XenbusStateInitialised, bypassing 403 * XenbusStateInitWait, without waiting for the frontend to first 404 * enter the XenbusStateInitialised state. In this case, default 405 * transport parameters are in effect and any transport parameters 406 * published by the backend must contain their default values. 407 * 408 * Drivers that support optional features and/or transport parameter 409 * negotiation must tolerate these additional state transition paths. 410 * In general this means performing the work of any skipped state 411 * transition, if it has not already been performed, in addition to the 412 * work associated with entry into the current state. 413 */ 414 415 /* 416 * REQUEST CODES. 417 */ 418 #define BLKIF_OP_READ 0 419 #define BLKIF_OP_WRITE 1 420 /* 421 * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER 422 * operation code ("barrier request") must be completed prior to the 423 * execution of the barrier request. All writes issued after the barrier 424 * request must not execute until after the completion of the barrier request. 425 * 426 * Optional. See "feature-barrier" XenBus node documentation above. 427 */ 428 #define BLKIF_OP_WRITE_BARRIER 2 429 /* 430 * Commit any uncommitted contents of the backing device's volatile cache 431 * to stable storage. 432 * 433 * Optional. See "feature-flush-cache" XenBus node documentation above. 434 */ 435 #define BLKIF_OP_FLUSH_DISKCACHE 3 436 /* 437 * Used in SLES sources for device specific command packet 438 * contained within the request. Reserved for that purpose. 439 */ 440 #define BLKIF_OP_RESERVED_1 4 441 /* 442 * Indicate to the backend device that a region of storage is no longer in 443 * use, and may be discarded at any time without impact to the client. If 444 * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the 445 * discarded region on the device must be rendered unrecoverable before the 446 * command returns. 447 * 448 * This operation is analogous to performing a trim (ATA) or unamp (SCSI), 449 * command on a native device. 450 * 451 * More information about trim/unmap operations can be found at: 452 * http://t13.org/Documents/UploadedDocuments/docs2008/ 453 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc 454 * http://www.seagate.com/staticfiles/support/disc/manuals/ 455 * Interface%20manuals/100293068c.pdf 456 * 457 * Optional. See "feature-discard", "discard-alignment", 458 * "discard-granularity", and "discard-secure" in the XenBus node 459 * documentation above. 460 */ 461 #define BLKIF_OP_DISCARD 5 462 463 /* 464 * Recognized if "feature-max-indirect-segments" in present in the backend 465 * xenbus info. The "feature-max-indirect-segments" node contains the maximum 466 * number of segments allowed by the backend per request. If the node is 467 * present, the frontend might use blkif_request_indirect structs in order to 468 * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The 469 * maximum number of indirect segments is fixed by the backend, but the 470 * frontend can issue requests with any number of indirect segments as long as 471 * it's less than the number provided by the backend. The indirect_grefs field 472 * in blkif_request_indirect should be filled by the frontend with the 473 * grant references of the pages that are holding the indirect segments. 474 * These pages are filled with an array of blkif_request_segment that hold the 475 * information about the segments. The number of indirect pages to use is 476 * determined by the number of segments an indirect request contains. Every 477 * indirect page can contain a maximum of 478 * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to 479 * calculate the number of indirect pages to use we have to do 480 * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))). 481 * 482 * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not* 483 * create the "feature-max-indirect-segments" node! 484 */ 485 #define BLKIF_OP_INDIRECT 6 486 487 /* 488 * Maximum scatter/gather segments per request. 489 * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE. 490 * NB. This could be 12 if the ring indexes weren't stored in the same page. 491 */ 492 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11 493 494 /* 495 * Maximum number of indirect pages to use per request. 496 */ 497 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8 498 499 /* 500 * NB. first_sect and last_sect in blkif_request_segment, as well as 501 * sector_number in blkif_request, are always expressed in 512-byte units. 502 * However they must be properly aligned to the real sector size of the 503 * physical disk, which is reported in the "physical-sector-size" node in 504 * the backend xenbus info. Also the xenbus "sectors" node is expressed in 505 * 512-byte units. 506 */ 507 struct blkif_request_segment { 508 grant_ref_t gref; /* reference to I/O buffer frame */ 509 /* @first_sect: first sector in frame to transfer (inclusive). */ 510 /* @last_sect: last sector in frame to transfer (inclusive). */ 511 UINT8 first_sect, last_sect; 512 }; 513 514 /* 515 * Starting ring element for any I/O request. 516 */ 517 #if defined(MDE_CPU_IA32) 518 // 519 // pack(4) is necessary when these structs are compiled for Ia32. 520 // Without it, the struct will have a different alignment than the one 521 // a backend expect for a 32bit guest. 522 // 523 #pragma pack(4) 524 #endif 525 struct blkif_request { 526 UINT8 operation; /* BLKIF_OP_??? */ 527 UINT8 nr_segments; /* number of segments */ 528 blkif_vdev_t handle; /* only for read/write requests */ 529 UINT64 id; /* private guest value, echoed in resp */ 530 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 531 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 532 }; 533 typedef struct blkif_request blkif_request_t; 534 535 /* 536 * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD 537 * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request) 538 */ 539 struct blkif_request_discard { 540 UINT8 operation; /* BLKIF_OP_DISCARD */ 541 UINT8 flag; /* BLKIF_DISCARD_SECURE or zero */ 542 #define BLKIF_DISCARD_SECURE (1<<0) /* ignored if discard-secure=0 */ 543 blkif_vdev_t handle; /* same as for read/write requests */ 544 UINT64 id; /* private guest value, echoed in resp */ 545 blkif_sector_t sector_number;/* start sector idx on disk */ 546 UINT64 nr_sectors; /* number of contiguous sectors to discard*/ 547 }; 548 typedef struct blkif_request_discard blkif_request_discard_t; 549 550 struct blkif_request_indirect { 551 UINT8 operation; /* BLKIF_OP_INDIRECT */ 552 UINT8 indirect_op; /* BLKIF_OP_{READ/WRITE} */ 553 UINT16 nr_segments; /* number of segments */ 554 UINT64 id; /* private guest value, echoed in resp */ 555 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */ 556 blkif_vdev_t handle; /* same as for read/write requests */ 557 grant_ref_t indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST]; 558 #ifdef MDE_CPU_IA32 559 UINT64 pad; /* Make it 64 byte aligned on i386 */ 560 #endif 561 }; 562 typedef struct blkif_request_indirect blkif_request_indirect_t; 563 564 struct blkif_response { 565 UINT64 id; /* copied from request */ 566 UINT8 operation; /* copied from request */ 567 INT16 status; /* BLKIF_RSP_??? */ 568 }; 569 typedef struct blkif_response blkif_response_t; 570 #if defined(MDE_CPU_IA32) 571 #pragma pack() 572 #endif 573 574 /* 575 * STATUS RETURN CODES. 576 */ 577 /* Operation not supported (only happens on barrier writes). */ 578 #define BLKIF_RSP_EOPNOTSUPP -2 579 /* Operation failed for some unspecified reason (-EIO). */ 580 #define BLKIF_RSP_ERROR -1 581 /* Operation completed successfully. */ 582 #define BLKIF_RSP_OKAY 0 583 584 /* 585 * Generate blkif ring structures and types. 586 */ 587 DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response); 588 589 #define VDISK_CDROM 0x1 590 #define VDISK_REMOVABLE 0x2 591 #define VDISK_READONLY 0x4 592 593 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */ 594 595 /* 596 * Local variables: 597 * mode: C 598 * c-file-style: "BSD" 599 * c-basic-offset: 4 600 * tab-width: 4 601 * indent-tabs-mode: nil 602 * End: 603 */ 604