1 /******************************************************************************
4 * Unified block-device I/O interface for Xen guest OSes.
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
24 * Copyright (c) 2003-2004, Keir Fraser
25 * Copyright (c) 2012, Spectra Logic Corporation
28 #ifndef __XEN_PUBLIC_IO_BLKIF_H__
29 #define __XEN_PUBLIC_IO_BLKIF_H__
32 #include "../grant_table.h"
35 * Front->back notifications: When enqueuing a new request, sending a
36 * notification can be made conditional on req_event (i.e., the generic
37 * hold-off mechanism provided by the ring macros). Backends must set
38 * req_event appropriately (e.g., using RING_FINAL_CHECK_FOR_REQUESTS()).
40 * Back->front notifications: When enqueuing a new response, sending a
41 * notification can be made conditional on rsp_event (i.e., the generic
42 * hold-off mechanism provided by the ring macros). Frontends must set
43 * rsp_event appropriately (e.g., using RING_FINAL_CHECK_FOR_RESPONSES()).
47 #define blkif_vdev_t UINT16
49 #define blkif_sector_t UINT64
52 * Feature and Parameter Negotiation
53 * =================================
54 * The two halves of a Xen block driver utilize nodes within the XenStore to
55 * communicate capabilities and to negotiate operating parameters. This
56 * section enumerates these nodes which reside in the respective front and
57 * backend portions of the XenStore, following the XenBus convention.
59 * All data in the XenStore is stored as strings. Nodes specifying numeric
60 * values are encoded in decimal. Integer value ranges listed below are
61 * expressed as fixed sized integer types capable of storing the conversion
62 * of a properly formated node string, without loss of information.
64 * Any specified default value is in effect if the corresponding XenBus node
65 * is not present in the XenStore.
67 * XenStore nodes in sections marked "PRIVATE" are solely for use by the
68 * driver side whose XenBus tree contains them.
70 * XenStore nodes marked "DEPRECATED" in their notes section should only be
71 * used to provide interoperability with legacy implementations.
73 * See the XenBus state transition diagram below for details on when XenBus
74 * nodes must be published and when they can be queried.
76 *****************************************************************************
77 * Backend XenBus Nodes
78 *****************************************************************************
80 *------------------ Backend Device Identification (PRIVATE) ------------------
83 * Values: "r" (read only), "w" (writable)
85 * The read or write access permissions to the backing store to be
86 * granted to the frontend.
91 * A free formatted string providing sufficient information for the
92 * backend driver to open the backing device. (e.g. the path to the
93 * file or block device representing the backing store.)
96 * Values: "file", "phy", "tap"
98 * The type of the backing device/object.
100 *--------------------------------- Features ---------------------------------
103 * Values: 0/1 (boolean)
106 * A value of "1" indicates that the backend can process requests
107 * containing the BLKIF_OP_WRITE_BARRIER request opcode. Requests
108 * of this type may still be returned at any time with the
109 * BLKIF_RSP_EOPNOTSUPP result code.
111 * feature-flush-cache
112 * Values: 0/1 (boolean)
115 * A value of "1" indicates that the backend can process requests
116 * containing the BLKIF_OP_FLUSH_DISKCACHE request opcode. Requests
117 * of this type may still be returned at any time with the
118 * BLKIF_RSP_EOPNOTSUPP result code.
121 * Values: 0/1 (boolean)
124 * A value of "1" indicates that the backend can process requests
125 * containing the BLKIF_OP_DISCARD request opcode. Requests
126 * of this type may still be returned at any time with the
127 * BLKIF_RSP_EOPNOTSUPP result code.
130 * Values: 0/1 (boolean)
134 * A value of "1" indicates that the backend can keep the grants used
135 * by the frontend driver mapped, so the same set of grants should be
136 * used in all transactions. The maximum number of grants the backend
137 * can map persistently depends on the implementation, but ideally it
138 * should be RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. Using this
139 * feature the backend doesn't need to unmap each grant, preventing
140 * costly TLB flushes. The backend driver should only map grants
141 * persistently if the frontend supports it. If a backend driver chooses
142 * to use the persistent protocol when the frontend doesn't support it,
143 * it will probably hit the maximum number of persistently mapped grants
144 * (due to the fact that the frontend won't be reusing the same grants),
145 * and fall back to non-persistent mode. Backend implementations may
146 * shrink or expand the number of persistently mapped grants without
147 * notifying the frontend depending on memory constraints (this might
148 * cause a performance degradation).
150 * If a backend driver wants to limit the maximum number of persistently
151 * mapped grants to a value less than RING_SIZE *
152 * BLKIF_MAX_SEGMENTS_PER_REQUEST a LRU strategy should be used to
153 * discard the grants that are less commonly used. Using a LRU in the
154 * backend driver paired with a LIFO queue in the frontend will
155 * allow us to have better performance in this scenario.
157 *----------------------- Request Transport Parameters ------------------------
159 * max-ring-page-order
164 * The maximum supported size of the request ring buffer in units of
165 * lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
171 * Notes: DEPRECATED, 2, 3
173 * The maximum supported size of the request ring buffer in units of
174 * machine pages. The value must be a power of 2.
176 *------------------------- Backend Device Properties -------------------------
183 * The offset, in bytes from the beginning of the virtual block device,
184 * to the first, addressable, discard extent on the underlying device.
186 * discard-granularity
188 * Default Value: <"sector-size">
191 * The size, in bytes, of the individually addressable discard extents
192 * of the underlying device.
195 * Values: 0/1 (boolean)
199 * A value of "1" indicates that the backend can process BLKIF_OP_DISCARD
200 * requests with the BLKIF_DISCARD_SECURE flag set.
203 * Values: <UINT32> (bitmap)
205 * A collection of bit flags describing attributes of the backing
206 * device. The VDISK_* macros define the meaning of each bit
212 * The logical sector size, in bytes, of the backend device.
214 * physical-sector-size
217 * The physical sector size, in bytes, of the backend device.
222 * The size of the backend device, expressed in units of its logical
223 * sector size ("sector-size").
225 *****************************************************************************
226 * Frontend XenBus Nodes
227 *****************************************************************************
229 *----------------------- Request Transport Parameters -----------------------
234 * The identifier of the Xen event channel used to signal activity
235 * in the ring buffer.
241 * The Xen grant reference granting permission for the backend to map
242 * the sole page in a single page sized ring buffer.
248 * For a frontend providing a multi-page ring, a "number of ring pages"
249 * sized list of nodes, each containing a Xen grant reference granting
250 * permission for the backend to map the page of the ring located
251 * at page index "%u". Page indexes are zero based.
254 * Values: string (XEN_IO_PROTO_ABI_*)
255 * Default Value: XEN_IO_PROTO_ABI_NATIVE
257 * The machine ABI rules governing the format of all ring request and
258 * response structures.
263 * Maximum Value: MAX(ffs(max-ring-pages) - 1, max-ring-page-order)
266 * The size of the frontend allocated request ring buffer in units
267 * of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
273 * Maximum Value: MAX(max-ring-pages,(0x1 << max-ring-page-order))
274 * Notes: DEPRECATED, 2, 3
276 * The size of the frontend allocated request ring buffer in units of
277 * machine pages. The value must be a power of 2.
280 * Values: 0/1 (boolean)
284 * A value of "1" indicates that the frontend will reuse the same grants
285 * for all transactions, allowing the backend to map them with write
286 * access (even when it should be read-only). If the frontend hits the
287 * maximum number of allowed persistently mapped grants, it can fallback
288 * to non persistent mode. This will cause a performance degradation,
289 * since the the backend driver will still try to map those grants
290 * persistently. Since the persistent grants protocol is compatible with
291 * the previous protocol, a frontend driver can choose to work in
292 * persistent mode even when the backend doesn't support it.
294 * It is recommended that the frontend driver stores the persistently
295 * mapped grants in a LIFO queue, so a subset of all persistently mapped
296 * grants gets used commonly. This is done in case the backend driver
297 * decides to limit the maximum number of persistently mapped grants
298 * to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
300 *------------------------- Virtual Device Properties -------------------------
303 * Values: "disk", "cdrom", "floppy", etc.
308 * A value indicating the physical device to virtualize within the
309 * frontend's domain. (e.g. "The first ATA disk", "The third SCSI
312 * See docs/misc/vbd-interface.txt for details on the format of this
317 * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer
319 * (2) Multi-page ring buffer scheme first used in some RedHat distributions
320 * including a distribution deployed on certain nodes of the Amazon
322 * (3) Support for multi-page ring buffers was implemented independently,
323 * in slightly different forms, by both Citrix and RedHat/Amazon.
324 * For full interoperability, block front and backends should publish
325 * identical ring parameters, adjusted for unit differences, to the
326 * XenStore nodes used in both schemes.
327 * (4) Devices that support discard functionality may internally allocate space
328 * (discardable extents) in units that are larger than the exported logical
329 * block size. If the backing device has such discardable extents the
330 * backend should provide both discard-granularity and discard-alignment.
331 * Providing just one of the two may be considered an error by the frontend.
332 * Backends supporting discard should include discard-granularity and
333 * discard-alignment even if it supports discarding individual sectors.
334 * Frontends should assume discard-alignment == 0 and discard-granularity
335 * == sector size if these keys are missing.
336 * (5) The discard-alignment parameter allows a physical device to be
337 * partitioned into virtual devices that do not necessarily begin or
338 * end on a discardable extent boundary.
339 * (6) When there is only a single page allocated to the request ring,
340 * 'ring-ref' is used to communicate the grant reference for this
341 * page to the backend. When using a multi-page ring, the 'ring-ref'
342 * node is not created. Instead 'ring-ref0' - 'ring-refN' are used.
343 * (7) When using persistent grants data has to be copied from/to the page
344 * where the grant is currently mapped. The overhead of doing this copy
345 * however doesn't suppress the speed improvement of not having to unmap
347 * (8) The frontend driver has to allow the backend driver to map all grants
348 * with write access, even when they should be mapped read-only, since
349 * further requests may reuse these grants and require write permissions.
350 * (9) Linux implementation doesn't have a limit on the maximum number of
351 * grants that can be persistently mapped in the frontend driver, but
352 * due to the frontent driver implementation it should never be bigger
353 * than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
354 *(10) The discard-secure property may be present and will be set to 1 if the
355 * backing device supports secure discard.
361 *****************************************************************************
363 *****************************************************************************
365 * Tool stack creates front and back nodes with state XenbusStateInitialising.
368 * ================================= =====================================
369 * XenbusStateInitialising XenbusStateInitialising
370 * o Query virtual device o Query backend device identification
372 * o Setup OS device instance. o Open and validate backend device.
373 * o Publish backend features and
374 * transport parameters.
378 * XenbusStateInitWait
380 * o Query backend features and
381 * transport parameters.
382 * o Allocate and initialize the
384 * o Publish transport parameters
385 * that will be in effect during
390 * XenbusStateInitialised
392 * o Query frontend transport parameters.
393 * o Connect to the request ring and
395 * o Publish backend device properties.
399 * XenbusStateConnected
401 * o Query backend device properties.
402 * o Finalize OS virtual device
407 * XenbusStateConnected
409 * Note: Drivers that do not support any optional features, or the negotiation
410 * of transport parameters, can skip certain states in the state machine:
412 * o A frontend may transition to XenbusStateInitialised without
413 * waiting for the backend to enter XenbusStateInitWait. In this
414 * case, default transport parameters are in effect and any
415 * transport parameters published by the frontend must contain
416 * their default values.
418 * o A backend may transition to XenbusStateInitialised, bypassing
419 * XenbusStateInitWait, without waiting for the frontend to first
420 * enter the XenbusStateInitialised state. In this case, default
421 * transport parameters are in effect and any transport parameters
422 * published by the backend must contain their default values.
424 * Drivers that support optional features and/or transport parameter
425 * negotiation must tolerate these additional state transition paths.
426 * In general this means performing the work of any skipped state
427 * transition, if it has not already been performed, in addition to the
428 * work associated with entry into the current state.
434 #define BLKIF_OP_READ 0
435 #define BLKIF_OP_WRITE 1
437 * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER
438 * operation code ("barrier request") must be completed prior to the
439 * execution of the barrier request. All writes issued after the barrier
440 * request must not execute until after the completion of the barrier request.
442 * Optional. See "feature-barrier" XenBus node documentation above.
444 #define BLKIF_OP_WRITE_BARRIER 2
446 * Commit any uncommitted contents of the backing device's volatile cache
449 * Optional. See "feature-flush-cache" XenBus node documentation above.
451 #define BLKIF_OP_FLUSH_DISKCACHE 3
453 * Used in SLES sources for device specific command packet
454 * contained within the request. Reserved for that purpose.
456 #define BLKIF_OP_RESERVED_1 4
458 * Indicate to the backend device that a region of storage is no longer in
459 * use, and may be discarded at any time without impact to the client. If
460 * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the
461 * discarded region on the device must be rendered unrecoverable before the
464 * This operation is analogous to performing a trim (ATA) or unamp (SCSI),
465 * command on a native device.
467 * More information about trim/unmap operations can be found at:
468 * http://t13.org/Documents/UploadedDocuments/docs2008/
469 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
470 * http://www.seagate.com/staticfiles/support/disc/manuals/
471 * Interface%20manuals/100293068c.pdf
473 * Optional. See "feature-discard", "discard-alignment",
474 * "discard-granularity", and "discard-secure" in the XenBus node
475 * documentation above.
477 #define BLKIF_OP_DISCARD 5
480 * Recognized if "feature-max-indirect-segments" in present in the backend
481 * xenbus info. The "feature-max-indirect-segments" node contains the maximum
482 * number of segments allowed by the backend per request. If the node is
483 * present, the frontend might use blkif_request_indirect structs in order to
484 * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The
485 * maximum number of indirect segments is fixed by the backend, but the
486 * frontend can issue requests with any number of indirect segments as long as
487 * it's less than the number provided by the backend. The indirect_grefs field
488 * in blkif_request_indirect should be filled by the frontend with the
489 * grant references of the pages that are holding the indirect segments.
490 * These pages are filled with an array of blkif_request_segment that hold the
491 * information about the segments. The number of indirect pages to use is
492 * determined by the number of segments an indirect request contains. Every
493 * indirect page can contain a maximum of
494 * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to
495 * calculate the number of indirect pages to use we have to do
496 * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))).
498 * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not*
499 * create the "feature-max-indirect-segments" node!
501 #define BLKIF_OP_INDIRECT 6
504 * Maximum scatter/gather segments per request.
505 * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE.
506 * NB. This could be 12 if the ring indexes weren't stored in the same page.
508 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
511 * Maximum number of indirect pages to use per request.
513 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8
516 * NB. first_sect and last_sect in blkif_request_segment, as well as
517 * sector_number in blkif_request, are always expressed in 512-byte units.
518 * However they must be properly aligned to the real sector size of the
519 * physical disk, which is reported in the "physical-sector-size" node in
520 * the backend xenbus info. Also the xenbus "sectors" node is expressed in
523 struct blkif_request_segment
{
524 grant_ref_t gref
; /* reference to I/O buffer frame */
525 /* @first_sect: first sector in frame to transfer (inclusive). */
526 /* @last_sect: last sector in frame to transfer (inclusive). */
527 UINT8 first_sect
, last_sect
;
531 * Starting ring element for any I/O request.
533 #if defined(MDE_CPU_IA32)
535 // pack(4) is necessary when these structs are compiled for Ia32.
536 // Without it, the struct will have a different alignment than the one
537 // a backend expect for a 32bit guest.
541 struct blkif_request
{
542 UINT8 operation
; /* BLKIF_OP_??? */
543 UINT8 nr_segments
; /* number of segments */
544 blkif_vdev_t handle
; /* only for read/write requests */
545 UINT64 id
; /* private guest value, echoed in resp */
546 blkif_sector_t sector_number
;/* start sector idx on disk (r/w only) */
547 struct blkif_request_segment seg
[BLKIF_MAX_SEGMENTS_PER_REQUEST
];
549 typedef struct blkif_request blkif_request_t
;
552 * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD
553 * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request)
555 struct blkif_request_discard
{
556 UINT8 operation
; /* BLKIF_OP_DISCARD */
557 UINT8 flag
; /* BLKIF_DISCARD_SECURE or zero */
558 #define BLKIF_DISCARD_SECURE (1<<0) /* ignored if discard-secure=0 */
559 blkif_vdev_t handle
; /* same as for read/write requests */
560 UINT64 id
; /* private guest value, echoed in resp */
561 blkif_sector_t sector_number
;/* start sector idx on disk */
562 UINT64 nr_sectors
; /* number of contiguous sectors to discard*/
564 typedef struct blkif_request_discard blkif_request_discard_t
;
566 struct blkif_request_indirect
{
567 UINT8 operation
; /* BLKIF_OP_INDIRECT */
568 UINT8 indirect_op
; /* BLKIF_OP_{READ/WRITE} */
569 UINT16 nr_segments
; /* number of segments */
570 UINT64 id
; /* private guest value, echoed in resp */
571 blkif_sector_t sector_number
;/* start sector idx on disk (r/w only) */
572 blkif_vdev_t handle
; /* same as for read/write requests */
573 grant_ref_t indirect_grefs
[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST
];
575 UINT64 pad
; /* Make it 64 byte aligned on i386 */
578 typedef struct blkif_request_indirect blkif_request_indirect_t
;
580 struct blkif_response
{
581 UINT64 id
; /* copied from request */
582 UINT8 operation
; /* copied from request */
583 INT16 status
; /* BLKIF_RSP_??? */
585 typedef struct blkif_response blkif_response_t
;
586 #if defined(MDE_CPU_IA32)
591 * STATUS RETURN CODES.
593 /* Operation not supported (only happens on barrier writes). */
594 #define BLKIF_RSP_EOPNOTSUPP -2
595 /* Operation failed for some unspecified reason (-EIO). */
596 #define BLKIF_RSP_ERROR -1
597 /* Operation completed successfully. */
598 #define BLKIF_RSP_OKAY 0
601 * Generate blkif ring structures and types.
603 DEFINE_RING_TYPES(blkif
, struct blkif_request
, struct blkif_response
);
605 #define VDISK_CDROM 0x1
606 #define VDISK_REMOVABLE 0x2
607 #define VDISK_READONLY 0x4
609 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */
614 * c-file-style: "BSD"
617 * indent-tabs-mode: nil