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_t
49 #define blkif_sector_t uint64_t
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 formatted 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 * Data used by the backend driver to locate and configure the backing
92 * device. The format and semantics of this data vary according to the
93 * backing device in use and are outside the scope of this specification.
96 * Values: "file", "phy", "tap"
98 * The type of the backing device/object.
102 * Values: 0/1 (boolean)
105 * The underlying storage is not affected by the direct IO memory
107 * http://lists.xen.org/archives/html/xen-devel/2012-12/msg01154.html
109 * Therefore this option gives the backend permission to use
110 * O_DIRECT, notwithstanding that bug.
112 * That is, if this option is enabled, use of O_DIRECT is safe,
113 * in circumstances where we would normally have avoided it as a
114 * workaround for that bug. This option is not relevant for all
115 * backends, and even not necessarily supported for those for
116 * which it is relevant. A backend which knows that it is not
117 * affected by the bug can ignore this option.
119 * This option doesn't require a backend to use O_DIRECT, so it
120 * should not be used to try to control the caching behaviour.
122 *--------------------------------- Features ---------------------------------
125 * Values: 0/1 (boolean)
128 * A value of "1" indicates that the backend can process requests
129 * containing the BLKIF_OP_WRITE_BARRIER request opcode. Requests
130 * of this type may still be returned at any time with the
131 * BLKIF_RSP_EOPNOTSUPP result code.
133 * feature-flush-cache
134 * Values: 0/1 (boolean)
137 * A value of "1" indicates that the backend can process requests
138 * containing the BLKIF_OP_FLUSH_DISKCACHE request opcode. Requests
139 * of this type may still be returned at any time with the
140 * BLKIF_RSP_EOPNOTSUPP result code.
143 * Values: 0/1 (boolean)
146 * A value of "1" indicates that the backend can process requests
147 * containing the BLKIF_OP_DISCARD request opcode. Requests
148 * of this type may still be returned at any time with the
149 * BLKIF_RSP_EOPNOTSUPP result code.
152 * Values: 0/1 (boolean)
156 * A value of "1" indicates that the backend can keep the grants used
157 * by the frontend driver mapped, so the same set of grants should be
158 * used in all transactions. The maximum number of grants the backend
159 * can map persistently depends on the implementation, but ideally it
160 * should be RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. Using this
161 * feature the backend doesn't need to unmap each grant, preventing
162 * costly TLB flushes. The backend driver should only map grants
163 * persistently if the frontend supports it. If a backend driver chooses
164 * to use the persistent protocol when the frontend doesn't support it,
165 * it will probably hit the maximum number of persistently mapped grants
166 * (due to the fact that the frontend won't be reusing the same grants),
167 * and fall back to non-persistent mode. Backend implementations may
168 * shrink or expand the number of persistently mapped grants without
169 * notifying the frontend depending on memory constraints (this might
170 * cause a performance degradation).
172 * If a backend driver wants to limit the maximum number of persistently
173 * mapped grants to a value less than RING_SIZE *
174 * BLKIF_MAX_SEGMENTS_PER_REQUEST a LRU strategy should be used to
175 * discard the grants that are less commonly used. Using a LRU in the
176 * backend driver paired with a LIFO queue in the frontend will
177 * allow us to have better performance in this scenario.
179 *----------------------- Request Transport Parameters ------------------------
181 * max-ring-page-order
186 * The maximum supported size of the request ring buffer in units of
187 * lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
193 * Notes: DEPRECATED, 2, 3
195 * The maximum supported size of the request ring buffer in units of
196 * machine pages. The value must be a power of 2.
198 *------------------------- Backend Device Properties -------------------------
201 * Values: 0/1 (boolean)
204 * This optional property, set by the toolstack, instructs the backend
205 * to offer discard to the frontend. If the property is missing the
206 * backend should offer discard if the backing storage actually supports
207 * it. This optional property, set by the toolstack, requests that the
208 * backend offer, or not offer, discard to the frontend.
215 * The offset, in bytes from the beginning of the virtual block device,
216 * to the first, addressable, discard extent on the underlying device.
218 * discard-granularity
220 * Default Value: <"sector-size">
223 * The size, in bytes, of the individually addressable discard extents
224 * of the underlying device.
227 * Values: 0/1 (boolean)
231 * A value of "1" indicates that the backend can process BLKIF_OP_DISCARD
232 * requests with the BLKIF_DISCARD_SECURE flag set.
235 * Values: <uint32_t> (bitmap)
237 * A collection of bit flags describing attributes of the backing
238 * device. The VDISK_* macros define the meaning of each bit
244 * The logical sector size, in bytes, of the backend device.
246 * physical-sector-size
249 * The physical sector size, in bytes, of the backend device.
254 * The size of the backend device, expressed in units of its logical
255 * sector size ("sector-size").
257 *****************************************************************************
258 * Frontend XenBus Nodes
259 *****************************************************************************
261 *----------------------- Request Transport Parameters -----------------------
266 * The identifier of the Xen event channel used to signal activity
267 * in the ring buffer.
273 * The Xen grant reference granting permission for the backend to map
274 * the sole page in a single page sized ring buffer.
280 * For a frontend providing a multi-page ring, a "number of ring pages"
281 * sized list of nodes, each containing a Xen grant reference granting
282 * permission for the backend to map the page of the ring located
283 * at page index "%u". Page indexes are zero based.
286 * Values: string (XEN_IO_PROTO_ABI_*)
287 * Default Value: XEN_IO_PROTO_ABI_NATIVE
289 * The machine ABI rules governing the format of all ring request and
290 * response structures.
295 * Maximum Value: MAX(ffs(max-ring-pages) - 1, max-ring-page-order)
298 * The size of the frontend allocated request ring buffer in units
299 * of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
305 * Maximum Value: MAX(max-ring-pages,(0x1 << max-ring-page-order))
306 * Notes: DEPRECATED, 2, 3
308 * The size of the frontend allocated request ring buffer in units of
309 * machine pages. The value must be a power of 2.
312 * Values: 0/1 (boolean)
316 * A value of "1" indicates that the frontend will reuse the same grants
317 * for all transactions, allowing the backend to map them with write
318 * access (even when it should be read-only). If the frontend hits the
319 * maximum number of allowed persistently mapped grants, it can fallback
320 * to non persistent mode. This will cause a performance degradation,
321 * since the the backend driver will still try to map those grants
322 * persistently. Since the persistent grants protocol is compatible with
323 * the previous protocol, a frontend driver can choose to work in
324 * persistent mode even when the backend doesn't support it.
326 * It is recommended that the frontend driver stores the persistently
327 * mapped grants in a LIFO queue, so a subset of all persistently mapped
328 * grants gets used commonly. This is done in case the backend driver
329 * decides to limit the maximum number of persistently mapped grants
330 * to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
332 *------------------------- Virtual Device Properties -------------------------
335 * Values: "disk", "cdrom", "floppy", etc.
340 * A value indicating the physical device to virtualize within the
341 * frontend's domain. (e.g. "The first ATA disk", "The third SCSI
344 * See docs/misc/vbd-interface.txt for details on the format of this
349 * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer
351 * (2) Multi-page ring buffer scheme first used in some RedHat distributions
352 * including a distribution deployed on certain nodes of the Amazon
354 * (3) Support for multi-page ring buffers was implemented independently,
355 * in slightly different forms, by both Citrix and RedHat/Amazon.
356 * For full interoperability, block front and backends should publish
357 * identical ring parameters, adjusted for unit differences, to the
358 * XenStore nodes used in both schemes.
359 * (4) Devices that support discard functionality may internally allocate space
360 * (discardable extents) in units that are larger than the exported logical
361 * block size. If the backing device has such discardable extents the
362 * backend should provide both discard-granularity and discard-alignment.
363 * Providing just one of the two may be considered an error by the frontend.
364 * Backends supporting discard should include discard-granularity and
365 * discard-alignment even if it supports discarding individual sectors.
366 * Frontends should assume discard-alignment == 0 and discard-granularity
367 * == sector size if these keys are missing.
368 * (5) The discard-alignment parameter allows a physical device to be
369 * partitioned into virtual devices that do not necessarily begin or
370 * end on a discardable extent boundary.
371 * (6) When there is only a single page allocated to the request ring,
372 * 'ring-ref' is used to communicate the grant reference for this
373 * page to the backend. When using a multi-page ring, the 'ring-ref'
374 * node is not created. Instead 'ring-ref0' - 'ring-refN' are used.
375 * (7) When using persistent grants data has to be copied from/to the page
376 * where the grant is currently mapped. The overhead of doing this copy
377 * however doesn't suppress the speed improvement of not having to unmap
379 * (8) The frontend driver has to allow the backend driver to map all grants
380 * with write access, even when they should be mapped read-only, since
381 * further requests may reuse these grants and require write permissions.
382 * (9) Linux implementation doesn't have a limit on the maximum number of
383 * grants that can be persistently mapped in the frontend driver, but
384 * due to the frontent driver implementation it should never be bigger
385 * than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
386 *(10) The discard-secure property may be present and will be set to 1 if the
387 * backing device supports secure discard.
393 *****************************************************************************
395 *****************************************************************************
397 * Tool stack creates front and back nodes with state XenbusStateInitialising.
400 * ================================= =====================================
401 * XenbusStateInitialising XenbusStateInitialising
402 * o Query virtual device o Query backend device identification
404 * o Setup OS device instance. o Open and validate backend device.
405 * o Publish backend features and
406 * transport parameters.
410 * XenbusStateInitWait
412 * o Query backend features and
413 * transport parameters.
414 * o Allocate and initialize the
416 * o Publish transport parameters
417 * that will be in effect during
422 * XenbusStateInitialised
424 * o Query frontend transport parameters.
425 * o Connect to the request ring and
427 * o Publish backend device properties.
431 * XenbusStateConnected
433 * o Query backend device properties.
434 * o Finalize OS virtual device
439 * XenbusStateConnected
441 * Note: Drivers that do not support any optional features, or the negotiation
442 * of transport parameters, can skip certain states in the state machine:
444 * o A frontend may transition to XenbusStateInitialised without
445 * waiting for the backend to enter XenbusStateInitWait. In this
446 * case, default transport parameters are in effect and any
447 * transport parameters published by the frontend must contain
448 * their default values.
450 * o A backend may transition to XenbusStateInitialised, bypassing
451 * XenbusStateInitWait, without waiting for the frontend to first
452 * enter the XenbusStateInitialised state. In this case, default
453 * transport parameters are in effect and any transport parameters
454 * published by the backend must contain their default values.
456 * Drivers that support optional features and/or transport parameter
457 * negotiation must tolerate these additional state transition paths.
458 * In general this means performing the work of any skipped state
459 * transition, if it has not already been performed, in addition to the
460 * work associated with entry into the current state.
466 #define BLKIF_OP_READ 0
467 #define BLKIF_OP_WRITE 1
469 * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER
470 * operation code ("barrier request") must be completed prior to the
471 * execution of the barrier request. All writes issued after the barrier
472 * request must not execute until after the completion of the barrier request.
474 * Optional. See "feature-barrier" XenBus node documentation above.
476 #define BLKIF_OP_WRITE_BARRIER 2
478 * Commit any uncommitted contents of the backing device's volatile cache
481 * Optional. See "feature-flush-cache" XenBus node documentation above.
483 #define BLKIF_OP_FLUSH_DISKCACHE 3
485 * Used in SLES sources for device specific command packet
486 * contained within the request. Reserved for that purpose.
488 #define BLKIF_OP_RESERVED_1 4
490 * Indicate to the backend device that a region of storage is no longer in
491 * use, and may be discarded at any time without impact to the client. If
492 * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the
493 * discarded region on the device must be rendered unrecoverable before the
496 * This operation is analogous to performing a trim (ATA) or unmap (SCSI),
497 * command on a native device.
499 * More information about trim/unmap operations can be found at:
500 * http://t13.org/Documents/UploadedDocuments/docs2008/
501 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
502 * http://www.seagate.com/staticfiles/support/disc/manuals/
503 * Interface%20manuals/100293068c.pdf
505 * Optional. See "feature-discard", "discard-alignment",
506 * "discard-granularity", and "discard-secure" in the XenBus node
507 * documentation above.
509 #define BLKIF_OP_DISCARD 5
512 * Recognized if "feature-max-indirect-segments" in present in the backend
513 * xenbus info. The "feature-max-indirect-segments" node contains the maximum
514 * number of segments allowed by the backend per request. If the node is
515 * present, the frontend might use blkif_request_indirect structs in order to
516 * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The
517 * maximum number of indirect segments is fixed by the backend, but the
518 * frontend can issue requests with any number of indirect segments as long as
519 * it's less than the number provided by the backend. The indirect_grefs field
520 * in blkif_request_indirect should be filled by the frontend with the
521 * grant references of the pages that are holding the indirect segments.
522 * These pages are filled with an array of blkif_request_segment that hold the
523 * information about the segments. The number of indirect pages to use is
524 * determined by the number of segments an indirect request contains. Every
525 * indirect page can contain a maximum of
526 * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to
527 * calculate the number of indirect pages to use we have to do
528 * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))).
530 * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not*
531 * create the "feature-max-indirect-segments" node!
533 #define BLKIF_OP_INDIRECT 6
536 * Maximum scatter/gather segments per request.
537 * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE.
538 * NB. This could be 12 if the ring indexes weren't stored in the same page.
540 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
543 * Maximum number of indirect pages to use per request.
545 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8
548 * NB. first_sect and last_sect in blkif_request_segment, as well as
549 * sector_number in blkif_request, are always expressed in 512-byte units.
550 * However they must be properly aligned to the real sector size of the
551 * physical disk, which is reported in the "physical-sector-size" node in
552 * the backend xenbus info. Also the xenbus "sectors" node is expressed in
555 struct blkif_request_segment {
556 grant_ref_t gref; /* reference to I/O buffer frame */
557 /* @first_sect: first sector in frame to transfer (inclusive). */
558 /* @last_sect: last sector in frame to transfer (inclusive). */
559 uint8_t first_sect, last_sect;
561 typedef struct blkif_request_segment blkif_request_segment_t;
564 * Starting ring element for any I/O request.
566 struct blkif_request {
567 uint8_t operation; /* BLKIF_OP_??? */
568 uint8_t nr_segments; /* number of segments */
569 blkif_vdev_t handle; /* only for read/write requests */
570 uint64_t id; /* private guest value, echoed in resp */
571 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
572 blkif_request_segment_t seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
574 typedef struct blkif_request blkif_request_t;
577 * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD
578 * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request)
580 struct blkif_request_discard {
581 uint8_t operation; /* BLKIF_OP_DISCARD */
582 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */
583 #define BLKIF_DISCARD_SECURE (1<<0) /* ignored if discard-secure=0 */
584 blkif_vdev_t handle; /* same as for read/write requests */
585 uint64_t id; /* private guest value, echoed in resp */
586 blkif_sector_t sector_number;/* start sector idx on disk */
587 uint64_t nr_sectors; /* number of contiguous sectors to discard*/
589 typedef struct blkif_request_discard blkif_request_discard_t;
591 struct blkif_request_indirect {
592 uint8_t operation; /* BLKIF_OP_INDIRECT */
593 uint8_t indirect_op; /* BLKIF_OP_{READ/WRITE} */
594 uint16_t nr_segments; /* number of segments */
595 uint64_t id; /* private guest value, echoed in resp */
596 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
597 blkif_vdev_t handle; /* same as for read/write requests */
598 grant_ref_t indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
600 uint64_t pad; /* Make it 64 byte aligned on i386 */
603 typedef struct blkif_request_indirect blkif_request_indirect_t;
605 struct blkif_response {
606 uint64_t id; /* copied from request */
607 uint8_t operation; /* copied from request */
608 int16_t status; /* BLKIF_RSP_??? */
610 typedef struct blkif_response blkif_response_t;
613 * STATUS RETURN CODES.
615 /* Operation not supported (only happens on barrier writes). */
616 #define BLKIF_RSP_EOPNOTSUPP -2
617 /* Operation failed for some unspecified reason (-EIO). */
618 #define BLKIF_RSP_ERROR -1
619 /* Operation completed successfully. */
620 #define BLKIF_RSP_OKAY 0
623 * Generate blkif ring structures and types.
625 DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response);
627 #define VDISK_CDROM 0x1
628 #define VDISK_REMOVABLE 0x2
629 #define VDISK_READONLY 0x4
631 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */
636 * c-file-style: "BSD"
639 * indent-tabs-mode: nil