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 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: "MAJOR:MINOR"
98 * MAJOR and MINOR are the major number and minor number of the
99 * backing device respectively.
102 * Values: "file", "phy", "tap"
104 * The type of the backing device/object.
108 * Values: 0/1 (boolean)
111 * The underlying storage is not affected by the direct IO memory
113 * http://lists.xen.org/archives/html/xen-devel/2012-12/msg01154.html
115 * Therefore this option gives the backend permission to use
116 * O_DIRECT, notwithstanding that bug.
118 * That is, if this option is enabled, use of O_DIRECT is safe,
119 * in circumstances where we would normally have avoided it as a
120 * workaround for that bug. This option is not relevant for all
121 * backends, and even not necessarily supported for those for
122 * which it is relevant. A backend which knows that it is not
123 * affected by the bug can ignore this option.
125 * This option doesn't require a backend to use O_DIRECT, so it
126 * should not be used to try to control the caching behaviour.
128 *--------------------------------- Features ---------------------------------
131 * Values: 0/1 (boolean)
134 * A value of "1" indicates that the backend can process requests
135 * containing the BLKIF_OP_WRITE_BARRIER request opcode. Requests
136 * of this type may still be returned at any time with the
137 * BLKIF_RSP_EOPNOTSUPP result code.
139 * feature-flush-cache
140 * Values: 0/1 (boolean)
143 * A value of "1" indicates that the backend can process requests
144 * containing the BLKIF_OP_FLUSH_DISKCACHE request opcode. Requests
145 * of this type may still be returned at any time with the
146 * BLKIF_RSP_EOPNOTSUPP result code.
149 * Values: 0/1 (boolean)
152 * A value of "1" indicates that the backend can process requests
153 * containing the BLKIF_OP_DISCARD request opcode. Requests
154 * of this type may still be returned at any time with the
155 * BLKIF_RSP_EOPNOTSUPP result code.
158 * Values: 0/1 (boolean)
162 * A value of "1" indicates that the backend can keep the grants used
163 * by the frontend driver mapped, so the same set of grants should be
164 * used in all transactions. The maximum number of grants the backend
165 * can map persistently depends on the implementation, but ideally it
166 * should be RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. Using this
167 * feature the backend doesn't need to unmap each grant, preventing
168 * costly TLB flushes. The backend driver should only map grants
169 * persistently if the frontend supports it. If a backend driver chooses
170 * to use the persistent protocol when the frontend doesn't support it,
171 * it will probably hit the maximum number of persistently mapped grants
172 * (due to the fact that the frontend won't be reusing the same grants),
173 * and fall back to non-persistent mode. Backend implementations may
174 * shrink or expand the number of persistently mapped grants without
175 * notifying the frontend depending on memory constraints (this might
176 * cause a performance degradation).
178 * If a backend driver wants to limit the maximum number of persistently
179 * mapped grants to a value less than RING_SIZE *
180 * BLKIF_MAX_SEGMENTS_PER_REQUEST a LRU strategy should be used to
181 * discard the grants that are less commonly used. Using a LRU in the
182 * backend driver paired with a LIFO queue in the frontend will
183 * allow us to have better performance in this scenario.
185 *----------------------- Request Transport Parameters ------------------------
187 * max-ring-page-order
192 * The maximum supported size of the request ring buffer in units of
193 * lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
199 * Notes: DEPRECATED, 2, 3
201 * The maximum supported size of the request ring buffer in units of
202 * machine pages. The value must be a power of 2.
204 *------------------------- Backend Device Properties -------------------------
207 * Values: 0/1 (boolean)
210 * This optional property, set by the toolstack, instructs the backend
211 * to offer discard to the frontend. If the property is missing the
212 * backend should offer discard if the backing storage actually supports
213 * it. This optional property, set by the toolstack, requests that the
214 * backend offer, or not offer, discard to the frontend.
221 * The offset, in bytes from the beginning of the virtual block device,
222 * to the first, addressable, discard extent on the underlying device.
224 * discard-granularity
226 * Default Value: <"sector-size">
229 * The size, in bytes, of the individually addressable discard extents
230 * of the underlying device.
233 * Values: 0/1 (boolean)
237 * A value of "1" indicates that the backend can process BLKIF_OP_DISCARD
238 * requests with the BLKIF_DISCARD_SECURE flag set.
241 * Values: <uint32_t> (bitmap)
243 * A collection of bit flags describing attributes of the backing
244 * device. The VDISK_* macros define the meaning of each bit
250 * The logical sector size, in bytes, of the backend device.
252 * physical-sector-size
255 * The physical sector size, in bytes, of the backend device.
260 * The size of the backend device, expressed in units of its logical
261 * sector size ("sector-size").
263 *****************************************************************************
264 * Frontend XenBus Nodes
265 *****************************************************************************
267 *----------------------- Request Transport Parameters -----------------------
272 * The identifier of the Xen event channel used to signal activity
273 * in the ring buffer.
279 * The Xen grant reference granting permission for the backend to map
280 * the sole page in a single page sized ring buffer.
286 * For a frontend providing a multi-page ring, a "number of ring pages"
287 * sized list of nodes, each containing a Xen grant reference granting
288 * permission for the backend to map the page of the ring located
289 * at page index "%u". Page indexes are zero based.
292 * Values: string (XEN_IO_PROTO_ABI_*)
293 * Default Value: XEN_IO_PROTO_ABI_NATIVE
295 * The machine ABI rules governing the format of all ring request and
296 * response structures.
301 * Maximum Value: MAX(ffs(max-ring-pages) - 1, max-ring-page-order)
304 * The size of the frontend allocated request ring buffer in units
305 * of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
311 * Maximum Value: MAX(max-ring-pages,(0x1 << max-ring-page-order))
312 * Notes: DEPRECATED, 2, 3
314 * The size of the frontend allocated request ring buffer in units of
315 * machine pages. The value must be a power of 2.
318 * Values: 0/1 (boolean)
322 * A value of "1" indicates that the frontend will reuse the same grants
323 * for all transactions, allowing the backend to map them with write
324 * access (even when it should be read-only). If the frontend hits the
325 * maximum number of allowed persistently mapped grants, it can fallback
326 * to non persistent mode. This will cause a performance degradation,
327 * since the backend driver will still try to map those grants
328 * persistently. Since the persistent grants protocol is compatible with
329 * the previous protocol, a frontend driver can choose to work in
330 * persistent mode even when the backend doesn't support it.
332 * It is recommended that the frontend driver stores the persistently
333 * mapped grants in a LIFO queue, so a subset of all persistently mapped
334 * grants gets used commonly. This is done in case the backend driver
335 * decides to limit the maximum number of persistently mapped grants
336 * to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
338 *------------------------- Virtual Device Properties -------------------------
341 * Values: "disk", "cdrom", "floppy", etc.
346 * A value indicating the physical device to virtualize within the
347 * frontend's domain. (e.g. "The first ATA disk", "The third SCSI
350 * See docs/misc/vbd-interface.txt for details on the format of this
355 * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer
357 * (2) Multi-page ring buffer scheme first used in some RedHat distributions
358 * including a distribution deployed on certain nodes of the Amazon
360 * (3) Support for multi-page ring buffers was implemented independently,
361 * in slightly different forms, by both Citrix and RedHat/Amazon.
362 * For full interoperability, block front and backends should publish
363 * identical ring parameters, adjusted for unit differences, to the
364 * XenStore nodes used in both schemes.
365 * (4) Devices that support discard functionality may internally allocate space
366 * (discardable extents) in units that are larger than the exported logical
367 * block size. If the backing device has such discardable extents the
368 * backend should provide both discard-granularity and discard-alignment.
369 * Providing just one of the two may be considered an error by the frontend.
370 * Backends supporting discard should include discard-granularity and
371 * discard-alignment even if it supports discarding individual sectors.
372 * Frontends should assume discard-alignment == 0 and discard-granularity
373 * == sector size if these keys are missing.
374 * (5) The discard-alignment parameter allows a physical device to be
375 * partitioned into virtual devices that do not necessarily begin or
376 * end on a discardable extent boundary.
377 * (6) When there is only a single page allocated to the request ring,
378 * 'ring-ref' is used to communicate the grant reference for this
379 * page to the backend. When using a multi-page ring, the 'ring-ref'
380 * node is not created. Instead 'ring-ref0' - 'ring-refN' are used.
381 * (7) When using persistent grants data has to be copied from/to the page
382 * where the grant is currently mapped. The overhead of doing this copy
383 * however doesn't suppress the speed improvement of not having to unmap
385 * (8) The frontend driver has to allow the backend driver to map all grants
386 * with write access, even when they should be mapped read-only, since
387 * further requests may reuse these grants and require write permissions.
388 * (9) Linux implementation doesn't have a limit on the maximum number of
389 * grants that can be persistently mapped in the frontend driver, but
390 * due to the frontent driver implementation it should never be bigger
391 * than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
392 *(10) The discard-secure property may be present and will be set to 1 if the
393 * backing device supports secure discard.
399 *****************************************************************************
401 *****************************************************************************
403 * Tool stack creates front and back nodes with state XenbusStateInitialising.
406 * ================================= =====================================
407 * XenbusStateInitialising XenbusStateInitialising
408 * o Query virtual device o Query backend device identification
410 * o Setup OS device instance. o Open and validate backend device.
411 * o Publish backend features and
412 * transport parameters.
416 * XenbusStateInitWait
418 * o Query backend features and
419 * transport parameters.
420 * o Allocate and initialize the
422 * o Publish transport parameters
423 * that will be in effect during
428 * XenbusStateInitialised
430 * o Query frontend transport parameters.
431 * o Connect to the request ring and
433 * o Publish backend device properties.
437 * XenbusStateConnected
439 * o Query backend device properties.
440 * o Finalize OS virtual device
445 * XenbusStateConnected
447 * Note: Drivers that do not support any optional features, or the negotiation
448 * of transport parameters, can skip certain states in the state machine:
450 * o A frontend may transition to XenbusStateInitialised without
451 * waiting for the backend to enter XenbusStateInitWait. In this
452 * case, default transport parameters are in effect and any
453 * transport parameters published by the frontend must contain
454 * their default values.
456 * o A backend may transition to XenbusStateInitialised, bypassing
457 * XenbusStateInitWait, without waiting for the frontend to first
458 * enter the XenbusStateInitialised state. In this case, default
459 * transport parameters are in effect and any transport parameters
460 * published by the backend must contain their default values.
462 * Drivers that support optional features and/or transport parameter
463 * negotiation must tolerate these additional state transition paths.
464 * In general this means performing the work of any skipped state
465 * transition, if it has not already been performed, in addition to the
466 * work associated with entry into the current state.
472 #define BLKIF_OP_READ 0
473 #define BLKIF_OP_WRITE 1
475 * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER
476 * operation code ("barrier request") must be completed prior to the
477 * execution of the barrier request. All writes issued after the barrier
478 * request must not execute until after the completion of the barrier request.
480 * Optional. See "feature-barrier" XenBus node documentation above.
482 #define BLKIF_OP_WRITE_BARRIER 2
484 * Commit any uncommitted contents of the backing device's volatile cache
487 * Optional. See "feature-flush-cache" XenBus node documentation above.
489 #define BLKIF_OP_FLUSH_DISKCACHE 3
491 * Used in SLES sources for device specific command packet
492 * contained within the request. Reserved for that purpose.
494 #define BLKIF_OP_RESERVED_1 4
496 * Indicate to the backend device that a region of storage is no longer in
497 * use, and may be discarded at any time without impact to the client. If
498 * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the
499 * discarded region on the device must be rendered unrecoverable before the
502 * This operation is analogous to performing a trim (ATA) or unamp (SCSI),
503 * command on a native device.
505 * More information about trim/unmap operations can be found at:
506 * http://t13.org/Documents/UploadedDocuments/docs2008/
507 * e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
508 * http://www.seagate.com/staticfiles/support/disc/manuals/
509 * Interface%20manuals/100293068c.pdf
511 * Optional. See "feature-discard", "discard-alignment",
512 * "discard-granularity", and "discard-secure" in the XenBus node
513 * documentation above.
515 #define BLKIF_OP_DISCARD 5
518 * Recognized if "feature-max-indirect-segments" in present in the backend
519 * xenbus info. The "feature-max-indirect-segments" node contains the maximum
520 * number of segments allowed by the backend per request. If the node is
521 * present, the frontend might use blkif_request_indirect structs in order to
522 * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The
523 * maximum number of indirect segments is fixed by the backend, but the
524 * frontend can issue requests with any number of indirect segments as long as
525 * it's less than the number provided by the backend. The indirect_grefs field
526 * in blkif_request_indirect should be filled by the frontend with the
527 * grant references of the pages that are holding the indirect segments.
528 * These pages are filled with an array of blkif_request_segment that hold the
529 * information about the segments. The number of indirect pages to use is
530 * determined by the number of segments an indirect request contains. Every
531 * indirect page can contain a maximum of
532 * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to
533 * calculate the number of indirect pages to use we have to do
534 * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))).
536 * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not*
537 * create the "feature-max-indirect-segments" node!
539 #define BLKIF_OP_INDIRECT 6
542 * Maximum scatter/gather segments per request.
543 * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE.
544 * NB. This could be 12 if the ring indexes weren't stored in the same page.
546 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
549 * Maximum number of indirect pages to use per request.
551 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8
554 * NB. first_sect and last_sect in blkif_request_segment, as well as
555 * sector_number in blkif_request, are always expressed in 512-byte units.
556 * However they must be properly aligned to the real sector size of the
557 * physical disk, which is reported in the "physical-sector-size" node in
558 * the backend xenbus info. Also the xenbus "sectors" node is expressed in
561 struct blkif_request_segment {
562 grant_ref_t gref; /* reference to I/O buffer frame */
563 /* @first_sect: first sector in frame to transfer (inclusive). */
564 /* @last_sect: last sector in frame to transfer (inclusive). */
565 uint8_t first_sect, last_sect;
569 * Starting ring element for any I/O request.
571 struct blkif_request {
572 uint8_t operation; /* BLKIF_OP_??? */
573 uint8_t nr_segments; /* number of segments */
574 blkif_vdev_t handle; /* only for read/write requests */
575 uint64_t id; /* private guest value, echoed in resp */
576 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
577 struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
579 typedef struct blkif_request blkif_request_t;
582 * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD
583 * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request)
585 struct blkif_request_discard {
586 uint8_t operation; /* BLKIF_OP_DISCARD */
587 uint8_t flag; /* BLKIF_DISCARD_SECURE or zero */
588 #define BLKIF_DISCARD_SECURE (1<<0) /* ignored if discard-secure=0 */
589 blkif_vdev_t handle; /* same as for read/write requests */
590 uint64_t id; /* private guest value, echoed in resp */
591 blkif_sector_t sector_number;/* start sector idx on disk */
592 uint64_t nr_sectors; /* number of contiguous sectors to discard*/
594 typedef struct blkif_request_discard blkif_request_discard_t;
596 struct blkif_request_indirect {
597 uint8_t operation; /* BLKIF_OP_INDIRECT */
598 uint8_t indirect_op; /* BLKIF_OP_{READ/WRITE} */
599 uint16_t nr_segments; /* number of segments */
600 uint64_t id; /* private guest value, echoed in resp */
601 blkif_sector_t sector_number;/* start sector idx on disk (r/w only) */
602 blkif_vdev_t handle; /* same as for read/write requests */
603 grant_ref_t indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
605 uint64_t pad; /* Make it 64 byte aligned on i386 */
608 typedef struct blkif_request_indirect blkif_request_indirect_t;
610 struct blkif_response {
611 uint64_t id; /* copied from request */
612 uint8_t operation; /* copied from request */
613 int16_t status; /* BLKIF_RSP_??? */
615 typedef struct blkif_response blkif_response_t;
618 * STATUS RETURN CODES.
620 /* Operation not supported (only happens on barrier writes). */
621 #define BLKIF_RSP_EOPNOTSUPP -2
622 /* Operation failed for some unspecified reason (-EIO). */
623 #define BLKIF_RSP_ERROR -1
624 /* Operation completed successfully. */
625 #define BLKIF_RSP_OKAY 0
628 * Generate blkif ring structures and types.
630 DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response);
632 #define VDISK_CDROM 0x1
633 #define VDISK_REMOVABLE 0x2
634 #define VDISK_READONLY 0x4
636 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */
641 * c-file-style: "BSD"
644 * indent-tabs-mode: nil