xref: /openbmc/qemu/include/hw/xen/interface/io/blkif.h (revision ffe98631)
1 /******************************************************************************
2  * blkif.h
3  *
4  * Unified block-device I/O interface for Xen guest OSes.
5  *
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:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
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.
23  *
24  * Copyright (c) 2003-2004, Keir Fraser
25  * Copyright (c) 2012, Spectra Logic Corporation
26  */
27 
28 #ifndef __XEN_PUBLIC_IO_BLKIF_H__
29 #define __XEN_PUBLIC_IO_BLKIF_H__
30 
31 #include "ring.h"
32 #include "../grant_table.h"
33 
34 /*
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()).
39  *
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()).
44  */
45 
46 #ifndef blkif_vdev_t
47 #define blkif_vdev_t   uint16_t
48 #endif
49 #define blkif_sector_t uint64_t
50 
51 /*
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.
58  *
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.
63  *
64  * Any specified default value is in effect if the corresponding XenBus node
65  * is not present in the XenStore.
66  *
67  * XenStore nodes in sections marked "PRIVATE" are solely for use by the
68  * driver side whose XenBus tree contains them.
69  *
70  * XenStore nodes marked "DEPRECATED" in their notes section should only be
71  * used to provide interoperability with legacy implementations.
72  *
73  * See the XenBus state transition diagram below for details on when XenBus
74  * nodes must be published and when they can be queried.
75  *
76  *****************************************************************************
77  *                            Backend XenBus Nodes
78  *****************************************************************************
79  *
80  *------------------ Backend Device Identification (PRIVATE) ------------------
81  *
82  * mode
83  *      Values:         "r" (read only), "w" (writable)
84  *
85  *      The read or write access permissions to the backing store to be
86  *      granted to the frontend.
87  *
88  * params
89  *      Values:         string
90  *
91  *      A free formatted string providing sufficient information for the
92  *      hotplug script to attach the device and provide a suitable
93  *      handler (ie: a block device) for blkback to use.
94  *
95  * physical-device
96  *      Values:         "MAJOR:MINOR"
97  *      Notes: 11
98  *
99  *      MAJOR and MINOR are the major number and minor number of the
100  *      backing device respectively.
101  *
102  * physical-device-path
103  *      Values:         path string
104  *
105  *      A string that contains the absolute path to the disk image. On
106  *      NetBSD and Linux this is always a block device, while on FreeBSD
107  *      it can be either a block device or a regular file.
108  *
109  * type
110  *      Values:         "file", "phy", "tap"
111  *
112  *      The type of the backing device/object.
113  *
114  *
115  * direct-io-safe
116  *      Values:         0/1 (boolean)
117  *      Default Value:  0
118  *
119  *      The underlying storage is not affected by the direct IO memory
120  *      lifetime bug.  See:
121  *        http://lists.xen.org/archives/html/xen-devel/2012-12/msg01154.html
122  *
123  *      Therefore this option gives the backend permission to use
124  *      O_DIRECT, notwithstanding that bug.
125  *
126  *      That is, if this option is enabled, use of O_DIRECT is safe,
127  *      in circumstances where we would normally have avoided it as a
128  *      workaround for that bug.  This option is not relevant for all
129  *      backends, and even not necessarily supported for those for
130  *      which it is relevant.  A backend which knows that it is not
131  *      affected by the bug can ignore this option.
132  *
133  *      This option doesn't require a backend to use O_DIRECT, so it
134  *      should not be used to try to control the caching behaviour.
135  *
136  *--------------------------------- Features ---------------------------------
137  *
138  * feature-barrier
139  *      Values:         0/1 (boolean)
140  *      Default Value:  0
141  *
142  *      A value of "1" indicates that the backend can process requests
143  *      containing the BLKIF_OP_WRITE_BARRIER request opcode.  Requests
144  *      of this type may still be returned at any time with the
145  *      BLKIF_RSP_EOPNOTSUPP result code.
146  *
147  * feature-flush-cache
148  *      Values:         0/1 (boolean)
149  *      Default Value:  0
150  *
151  *      A value of "1" indicates that the backend can process requests
152  *      containing the BLKIF_OP_FLUSH_DISKCACHE request opcode.  Requests
153  *      of this type may still be returned at any time with the
154  *      BLKIF_RSP_EOPNOTSUPP result code.
155  *
156  * feature-discard
157  *      Values:         0/1 (boolean)
158  *      Default Value:  0
159  *
160  *      A value of "1" indicates that the backend can process requests
161  *      containing the BLKIF_OP_DISCARD request opcode.  Requests
162  *      of this type may still be returned at any time with the
163  *      BLKIF_RSP_EOPNOTSUPP result code.
164  *
165  * feature-persistent
166  *      Values:         0/1 (boolean)
167  *      Default Value:  0
168  *      Notes: 7
169  *
170  *      A value of "1" indicates that the backend can keep the grants used
171  *      by the frontend driver mapped, so the same set of grants should be
172  *      used in all transactions. The maximum number of grants the backend
173  *      can map persistently depends on the implementation, but ideally it
174  *      should be RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST. Using this
175  *      feature the backend doesn't need to unmap each grant, preventing
176  *      costly TLB flushes. The backend driver should only map grants
177  *      persistently if the frontend supports it. If a backend driver chooses
178  *      to use the persistent protocol when the frontend doesn't support it,
179  *      it will probably hit the maximum number of persistently mapped grants
180  *      (due to the fact that the frontend won't be reusing the same grants),
181  *      and fall back to non-persistent mode. Backend implementations may
182  *      shrink or expand the number of persistently mapped grants without
183  *      notifying the frontend depending on memory constraints (this might
184  *      cause a performance degradation).
185  *
186  *      If a backend driver wants to limit the maximum number of persistently
187  *      mapped grants to a value less than RING_SIZE *
188  *      BLKIF_MAX_SEGMENTS_PER_REQUEST a LRU strategy should be used to
189  *      discard the grants that are less commonly used. Using a LRU in the
190  *      backend driver paired with a LIFO queue in the frontend will
191  *      allow us to have better performance in this scenario.
192  *
193  *----------------------- Request Transport Parameters ------------------------
194  *
195  * max-ring-page-order
196  *      Values:         <uint32_t>
197  *      Default Value:  0
198  *      Notes:          1, 3
199  *
200  *      The maximum supported size of the request ring buffer in units of
201  *      lb(machine pages). (e.g. 0 == 1 page,  1 = 2 pages, 2 == 4 pages,
202  *      etc.).
203  *
204  * max-ring-pages
205  *      Values:         <uint32_t>
206  *      Default Value:  1
207  *      Notes:          DEPRECATED, 2, 3
208  *
209  *      The maximum supported size of the request ring buffer in units of
210  *      machine pages.  The value must be a power of 2.
211  *
212  *------------------------- Backend Device Properties -------------------------
213  *
214  * discard-enable
215  *      Values:         0/1 (boolean)
216  *      Default Value:  1
217  *
218  *      This optional property, set by the toolstack, instructs the backend
219  *      to offer (or not to offer) discard to the frontend. If the property
220  *      is missing the backend should offer discard if the backing storage
221  *      actually supports it.
222  *
223  * discard-alignment
224  *      Values:         <uint32_t>
225  *      Default Value:  0
226  *      Notes:          4, 5
227  *
228  *      The offset, in bytes from the beginning of the virtual block device,
229  *      to the first, addressable, discard extent on the underlying device.
230  *
231  * discard-granularity
232  *      Values:         <uint32_t>
233  *      Default Value:  <"sector-size">
234  *      Notes:          4
235  *
236  *      The size, in bytes, of the individually addressable discard extents
237  *      of the underlying device.
238  *
239  * discard-secure
240  *      Values:         0/1 (boolean)
241  *      Default Value:  0
242  *      Notes:          10
243  *
244  *      A value of "1" indicates that the backend can process BLKIF_OP_DISCARD
245  *      requests with the BLKIF_DISCARD_SECURE flag set.
246  *
247  * info
248  *      Values:         <uint32_t> (bitmap)
249  *
250  *      A collection of bit flags describing attributes of the backing
251  *      device.  The VDISK_* macros define the meaning of each bit
252  *      location.
253  *
254  * sector-size
255  *      Values:         <uint32_t>
256  *
257  *      The logical block size, in bytes, of the underlying storage. This
258  *      must be a power of two with a minimum value of 512.
259  *
260  *      NOTE: Because of implementation bugs in some frontends this must be
261  *            set to 512, unless the frontend advertizes a non-zero value
262  *            in its "feature-large-sector-size" xenbus node. (See below).
263  *
264  * physical-sector-size
265  *      Values:         <uint32_t>
266  *      Default Value:  <"sector-size">
267  *
268  *      The physical block size, in bytes, of the backend storage. This
269  *      must be an integer multiple of "sector-size".
270  *
271  * sectors
272  *      Values:         <uint64_t>
273  *
274  *      The size of the backend device, expressed in units of "sector-size".
275  *      The product of "sector-size" and "sectors" must also be an integer
276  *      multiple of "physical-sector-size", if that node is present.
277  *
278  *****************************************************************************
279  *                            Frontend XenBus Nodes
280  *****************************************************************************
281  *
282  *----------------------- Request Transport Parameters -----------------------
283  *
284  * event-channel
285  *      Values:         <uint32_t>
286  *
287  *      The identifier of the Xen event channel used to signal activity
288  *      in the ring buffer.
289  *
290  * ring-ref
291  *      Values:         <uint32_t>
292  *      Notes:          6
293  *
294  *      The Xen grant reference granting permission for the backend to map
295  *      the sole page in a single page sized ring buffer.
296  *
297  * ring-ref%u
298  *      Values:         <uint32_t>
299  *      Notes:          6
300  *
301  *      For a frontend providing a multi-page ring, a "number of ring pages"
302  *      sized list of nodes, each containing a Xen grant reference granting
303  *      permission for the backend to map the page of the ring located
304  *      at page index "%u".  Page indexes are zero based.
305  *
306  * protocol
307  *      Values:         string (XEN_IO_PROTO_ABI_*)
308  *      Default Value:  XEN_IO_PROTO_ABI_NATIVE
309  *
310  *      The machine ABI rules governing the format of all ring request and
311  *      response structures.
312  *
313  * ring-page-order
314  *      Values:         <uint32_t>
315  *      Default Value:  0
316  *      Maximum Value:  MAX(ffs(max-ring-pages) - 1, max-ring-page-order)
317  *      Notes:          1, 3
318  *
319  *      The size of the frontend allocated request ring buffer in units
320  *      of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
321  *      etc.).
322  *
323  * num-ring-pages
324  *      Values:         <uint32_t>
325  *      Default Value:  1
326  *      Maximum Value:  MAX(max-ring-pages,(0x1 << max-ring-page-order))
327  *      Notes:          DEPRECATED, 2, 3
328  *
329  *      The size of the frontend allocated request ring buffer in units of
330  *      machine pages.  The value must be a power of 2.
331  *
332  *--------------------------------- Features ---------------------------------
333  *
334  * feature-persistent
335  *      Values:         0/1 (boolean)
336  *      Default Value:  0
337  *      Notes: 7, 8, 9
338  *
339  *      A value of "1" indicates that the frontend will reuse the same grants
340  *      for all transactions, allowing the backend to map them with write
341  *      access (even when it should be read-only). If the frontend hits the
342  *      maximum number of allowed persistently mapped grants, it can fallback
343  *      to non persistent mode. This will cause a performance degradation,
344  *      since the backend driver will still try to map those grants
345  *      persistently. Since the persistent grants protocol is compatible with
346  *      the previous protocol, a frontend driver can choose to work in
347  *      persistent mode even when the backend doesn't support it.
348  *
349  *      It is recommended that the frontend driver stores the persistently
350  *      mapped grants in a LIFO queue, so a subset of all persistently mapped
351  *      grants gets used commonly. This is done in case the backend driver
352  *      decides to limit the maximum number of persistently mapped grants
353  *      to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
354  *
355  * feature-large-sector-size
356  *      Values:         0/1 (boolean)
357  *      Default Value:  0
358  *
359  *      A value of "1" indicates that the frontend will correctly supply and
360  *      interpret all sector-based quantities in terms of the "sector-size"
361  *      value supplied in the backend info, whatever that may be set to.
362  *      If this node is not present or its value is "0" then it is assumed
363  *      that the frontend requires that the logical block size is 512 as it
364  *      is hardcoded (which is the case in some frontend implementations).
365  *
366  *------------------------- Virtual Device Properties -------------------------
367  *
368  * device-type
369  *      Values:         "disk", "cdrom", "floppy", etc.
370  *
371  * virtual-device
372  *      Values:         <uint32_t>
373  *
374  *      A value indicating the physical device to virtualize within the
375  *      frontend's domain.  (e.g. "The first ATA disk", "The third SCSI
376  *      disk", etc.)
377  *
378  *      See docs/misc/vbd-interface.txt for details on the format of this
379  *      value.
380  *
381  * Notes
382  * -----
383  * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer
384  *     PV drivers.
385  * (2) Multi-page ring buffer scheme first used in some RedHat distributions
386  *     including a distribution deployed on certain nodes of the Amazon
387  *     EC2 cluster.
388  * (3) Support for multi-page ring buffers was implemented independently,
389  *     in slightly different forms, by both Citrix and RedHat/Amazon.
390  *     For full interoperability, block front and backends should publish
391  *     identical ring parameters, adjusted for unit differences, to the
392  *     XenStore nodes used in both schemes.
393  * (4) Devices that support discard functionality may internally allocate space
394  *     (discardable extents) in units that are larger than the exported logical
395  *     block size. If the backing device has such discardable extents the
396  *     backend should provide both discard-granularity and discard-alignment.
397  *     Providing just one of the two may be considered an error by the frontend.
398  *     Backends supporting discard should include discard-granularity and
399  *     discard-alignment even if it supports discarding individual sectors.
400  *     Frontends should assume discard-alignment == 0 and discard-granularity
401  *     == sector size if these keys are missing.
402  * (5) The discard-alignment parameter allows a physical device to be
403  *     partitioned into virtual devices that do not necessarily begin or
404  *     end on a discardable extent boundary.
405  * (6) When there is only a single page allocated to the request ring,
406  *     'ring-ref' is used to communicate the grant reference for this
407  *     page to the backend.  When using a multi-page ring, the 'ring-ref'
408  *     node is not created.  Instead 'ring-ref0' - 'ring-refN' are used.
409  * (7) When using persistent grants data has to be copied from/to the page
410  *     where the grant is currently mapped. The overhead of doing this copy
411  *     however doesn't suppress the speed improvement of not having to unmap
412  *     the grants.
413  * (8) The frontend driver has to allow the backend driver to map all grants
414  *     with write access, even when they should be mapped read-only, since
415  *     further requests may reuse these grants and require write permissions.
416  * (9) Linux implementation doesn't have a limit on the maximum number of
417  *     grants that can be persistently mapped in the frontend driver, but
418  *     due to the frontent driver implementation it should never be bigger
419  *     than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
420  *(10) The discard-secure property may be present and will be set to 1 if the
421  *     backing device supports secure discard.
422  *(11) Only used by Linux and NetBSD.
423  */
424 
425 /*
426  * Multiple hardware queues/rings:
427  * If supported, the backend will write the key "multi-queue-max-queues" to
428  * the directory for that vbd, and set its value to the maximum supported
429  * number of queues.
430  * Frontends that are aware of this feature and wish to use it can write the
431  * key "multi-queue-num-queues" with the number they wish to use, which must be
432  * greater than zero, and no more than the value reported by the backend in
433  * "multi-queue-max-queues".
434  *
435  * For frontends requesting just one queue, the usual event-channel and
436  * ring-ref keys are written as before, simplifying the backend processing
437  * to avoid distinguishing between a frontend that doesn't understand the
438  * multi-queue feature, and one that does, but requested only one queue.
439  *
440  * Frontends requesting two or more queues must not write the toplevel
441  * event-channel and ring-ref keys, instead writing those keys under sub-keys
442  * having the name "queue-N" where N is the integer ID of the queue/ring for
443  * which those keys belong. Queues are indexed from zero.
444  * For example, a frontend with two queues must write the following set of
445  * queue-related keys:
446  *
447  * /local/domain/1/device/vbd/0/multi-queue-num-queues = "2"
448  * /local/domain/1/device/vbd/0/queue-0 = ""
449  * /local/domain/1/device/vbd/0/queue-0/ring-ref = "<ring-ref#0>"
450  * /local/domain/1/device/vbd/0/queue-0/event-channel = "<evtchn#0>"
451  * /local/domain/1/device/vbd/0/queue-1 = ""
452  * /local/domain/1/device/vbd/0/queue-1/ring-ref = "<ring-ref#1>"
453  * /local/domain/1/device/vbd/0/queue-1/event-channel = "<evtchn#1>"
454  *
455  * It is also possible to use multiple queues/rings together with
456  * feature multi-page ring buffer.
457  * For example, a frontend requests two queues/rings and the size of each ring
458  * buffer is two pages must write the following set of related keys:
459  *
460  * /local/domain/1/device/vbd/0/multi-queue-num-queues = "2"
461  * /local/domain/1/device/vbd/0/ring-page-order = "1"
462  * /local/domain/1/device/vbd/0/queue-0 = ""
463  * /local/domain/1/device/vbd/0/queue-0/ring-ref0 = "<ring-ref#0>"
464  * /local/domain/1/device/vbd/0/queue-0/ring-ref1 = "<ring-ref#1>"
465  * /local/domain/1/device/vbd/0/queue-0/event-channel = "<evtchn#0>"
466  * /local/domain/1/device/vbd/0/queue-1 = ""
467  * /local/domain/1/device/vbd/0/queue-1/ring-ref0 = "<ring-ref#2>"
468  * /local/domain/1/device/vbd/0/queue-1/ring-ref1 = "<ring-ref#3>"
469  * /local/domain/1/device/vbd/0/queue-1/event-channel = "<evtchn#1>"
470  *
471  */
472 
473 /*
474  * STATE DIAGRAMS
475  *
476  *****************************************************************************
477  *                                   Startup                                 *
478  *****************************************************************************
479  *
480  * Tool stack creates front and back nodes with state XenbusStateInitialising.
481  *
482  * Front                                Back
483  * =================================    =====================================
484  * XenbusStateInitialising              XenbusStateInitialising
485  *  o Query virtual device               o Query backend device identification
486  *    properties.                          data.
487  *  o Setup OS device instance.          o Open and validate backend device.
488  *                                       o Publish backend features and
489  *                                         transport parameters.
490  *                                                      |
491  *                                                      |
492  *                                                      V
493  *                                      XenbusStateInitWait
494  *
495  * o Query backend features and
496  *   transport parameters.
497  * o Allocate and initialize the
498  *   request ring.
499  * o Publish transport parameters
500  *   that will be in effect during
501  *   this connection.
502  *              |
503  *              |
504  *              V
505  * XenbusStateInitialised
506  *
507  *                                       o Query frontend transport parameters.
508  *                                       o Connect to the request ring and
509  *                                         event channel.
510  *                                       o Publish backend device properties.
511  *                                                      |
512  *                                                      |
513  *                                                      V
514  *                                      XenbusStateConnected
515  *
516  *  o Query backend device properties.
517  *  o Finalize OS virtual device
518  *    instance.
519  *              |
520  *              |
521  *              V
522  * XenbusStateConnected
523  *
524  * Note: Drivers that do not support any optional features, or the negotiation
525  *       of transport parameters, can skip certain states in the state machine:
526  *
527  *       o A frontend may transition to XenbusStateInitialised without
528  *         waiting for the backend to enter XenbusStateInitWait.  In this
529  *         case, default transport parameters are in effect and any
530  *         transport parameters published by the frontend must contain
531  *         their default values.
532  *
533  *       o A backend may transition to XenbusStateInitialised, bypassing
534  *         XenbusStateInitWait, without waiting for the frontend to first
535  *         enter the XenbusStateInitialised state.  In this case, default
536  *         transport parameters are in effect and any transport parameters
537  *         published by the backend must contain their default values.
538  *
539  *       Drivers that support optional features and/or transport parameter
540  *       negotiation must tolerate these additional state transition paths.
541  *       In general this means performing the work of any skipped state
542  *       transition, if it has not already been performed, in addition to the
543  *       work associated with entry into the current state.
544  */
545 
546 /*
547  * REQUEST CODES.
548  */
549 #define BLKIF_OP_READ              0
550 #define BLKIF_OP_WRITE             1
551 /*
552  * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER
553  * operation code ("barrier request") must be completed prior to the
554  * execution of the barrier request.  All writes issued after the barrier
555  * request must not execute until after the completion of the barrier request.
556  *
557  * Optional.  See "feature-barrier" XenBus node documentation above.
558  */
559 #define BLKIF_OP_WRITE_BARRIER     2
560 /*
561  * Commit any uncommitted contents of the backing device's volatile cache
562  * to stable storage.
563  *
564  * Optional.  See "feature-flush-cache" XenBus node documentation above.
565  */
566 #define BLKIF_OP_FLUSH_DISKCACHE   3
567 /*
568  * Used in SLES sources for device specific command packet
569  * contained within the request. Reserved for that purpose.
570  */
571 #define BLKIF_OP_RESERVED_1        4
572 /*
573  * Indicate to the backend device that a region of storage is no longer in
574  * use, and may be discarded at any time without impact to the client.  If
575  * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the
576  * discarded region on the device must be rendered unrecoverable before the
577  * command returns.
578  *
579  * This operation is analogous to performing a trim (ATA) or unamp (SCSI),
580  * command on a native device.
581  *
582  * More information about trim/unmap operations can be found at:
583  * http://t13.org/Documents/UploadedDocuments/docs2008/
584  *     e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
585  * http://www.seagate.com/staticfiles/support/disc/manuals/
586  *     Interface%20manuals/100293068c.pdf
587  *
588  * Optional.  See "feature-discard", "discard-alignment",
589  * "discard-granularity", and "discard-secure" in the XenBus node
590  * documentation above.
591  */
592 #define BLKIF_OP_DISCARD           5
593 
594 /*
595  * Recognized if "feature-max-indirect-segments" in present in the backend
596  * xenbus info. The "feature-max-indirect-segments" node contains the maximum
597  * number of segments allowed by the backend per request. If the node is
598  * present, the frontend might use blkif_request_indirect structs in order to
599  * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The
600  * maximum number of indirect segments is fixed by the backend, but the
601  * frontend can issue requests with any number of indirect segments as long as
602  * it's less than the number provided by the backend. The indirect_grefs field
603  * in blkif_request_indirect should be filled by the frontend with the
604  * grant references of the pages that are holding the indirect segments.
605  * These pages are filled with an array of blkif_request_segment that hold the
606  * information about the segments. The number of indirect pages to use is
607  * determined by the number of segments an indirect request contains. Every
608  * indirect page can contain a maximum of
609  * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to
610  * calculate the number of indirect pages to use we have to do
611  * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))).
612  *
613  * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not*
614  * create the "feature-max-indirect-segments" node!
615  */
616 #define BLKIF_OP_INDIRECT          6
617 
618 /*
619  * Maximum scatter/gather segments per request.
620  * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE.
621  * NB. This could be 12 if the ring indexes weren't stored in the same page.
622  */
623 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
624 
625 /*
626  * Maximum number of indirect pages to use per request.
627  */
628 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8
629 
630 /*
631  * NB. 'first_sect' and 'last_sect' in blkif_request_segment, as well as
632  * 'sector_number' in blkif_request, blkif_request_discard and
633  * blkif_request_indirect are sector-based quantities. See the description
634  * of the "feature-large-sector-size" frontend xenbus node above for
635  * more information.
636  */
637 struct blkif_request_segment {
638     grant_ref_t gref;        /* reference to I/O buffer frame        */
639     /* @first_sect: first sector in frame to transfer (inclusive).   */
640     /* @last_sect: last sector in frame to transfer (inclusive).     */
641     uint8_t     first_sect, last_sect;
642 };
643 
644 /*
645  * Starting ring element for any I/O request.
646  */
647 struct blkif_request {
648     uint8_t        operation;    /* BLKIF_OP_???                         */
649     uint8_t        nr_segments;  /* number of segments                   */
650     blkif_vdev_t   handle;       /* only for read/write requests         */
651     uint64_t       id;           /* private guest value, echoed in resp  */
652     blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
653     struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
654 };
655 typedef struct blkif_request blkif_request_t;
656 
657 /*
658  * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD
659  * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request)
660  */
661 struct blkif_request_discard {
662     uint8_t        operation;    /* BLKIF_OP_DISCARD                     */
663     uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
664 #define BLKIF_DISCARD_SECURE (1<<0)  /* ignored if discard-secure=0      */
665     blkif_vdev_t   handle;       /* same as for read/write requests      */
666     uint64_t       id;           /* private guest value, echoed in resp  */
667     blkif_sector_t sector_number;/* start sector idx on disk             */
668     uint64_t       nr_sectors;   /* number of contiguous sectors to discard*/
669 };
670 typedef struct blkif_request_discard blkif_request_discard_t;
671 
672 struct blkif_request_indirect {
673     uint8_t        operation;    /* BLKIF_OP_INDIRECT                    */
674     uint8_t        indirect_op;  /* BLKIF_OP_{READ/WRITE}                */
675     uint16_t       nr_segments;  /* number of segments                   */
676     uint64_t       id;           /* private guest value, echoed in resp  */
677     blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
678     blkif_vdev_t   handle;       /* same as for read/write requests      */
679     grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
680 #ifdef __i386__
681     uint64_t       pad;          /* Make it 64 byte aligned on i386      */
682 #endif
683 };
684 typedef struct blkif_request_indirect blkif_request_indirect_t;
685 
686 struct blkif_response {
687     uint64_t        id;              /* copied from request */
688     uint8_t         operation;       /* copied from request */
689     int16_t         status;          /* BLKIF_RSP_???       */
690 };
691 typedef struct blkif_response blkif_response_t;
692 
693 /*
694  * STATUS RETURN CODES.
695  */
696  /* Operation not supported (only happens on barrier writes). */
697 #define BLKIF_RSP_EOPNOTSUPP  -2
698  /* Operation failed for some unspecified reason (-EIO). */
699 #define BLKIF_RSP_ERROR       -1
700  /* Operation completed successfully. */
701 #define BLKIF_RSP_OKAY         0
702 
703 /*
704  * Generate blkif ring structures and types.
705  */
706 DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response);
707 
708 #define VDISK_CDROM        0x1
709 #define VDISK_REMOVABLE    0x2
710 #define VDISK_READONLY     0x4
711 
712 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */
713