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 sector size, in bytes, of the backend device.
258  *
259  * physical-sector-size
260  *      Values:         <uint32_t>
261  *
262  *      The physical sector size, in bytes, of the backend device.
263  *
264  * sectors
265  *      Values:         <uint64_t>
266  *
267  *      The size of the backend device, expressed in units of its logical
268  *      sector size ("sector-size").
269  *
270  *****************************************************************************
271  *                            Frontend XenBus Nodes
272  *****************************************************************************
273  *
274  *----------------------- Request Transport Parameters -----------------------
275  *
276  * event-channel
277  *      Values:         <uint32_t>
278  *
279  *      The identifier of the Xen event channel used to signal activity
280  *      in the ring buffer.
281  *
282  * ring-ref
283  *      Values:         <uint32_t>
284  *      Notes:          6
285  *
286  *      The Xen grant reference granting permission for the backend to map
287  *      the sole page in a single page sized ring buffer.
288  *
289  * ring-ref%u
290  *      Values:         <uint32_t>
291  *      Notes:          6
292  *
293  *      For a frontend providing a multi-page ring, a "number of ring pages"
294  *      sized list of nodes, each containing a Xen grant reference granting
295  *      permission for the backend to map the page of the ring located
296  *      at page index "%u".  Page indexes are zero based.
297  *
298  * protocol
299  *      Values:         string (XEN_IO_PROTO_ABI_*)
300  *      Default Value:  XEN_IO_PROTO_ABI_NATIVE
301  *
302  *      The machine ABI rules governing the format of all ring request and
303  *      response structures.
304  *
305  * ring-page-order
306  *      Values:         <uint32_t>
307  *      Default Value:  0
308  *      Maximum Value:  MAX(ffs(max-ring-pages) - 1, max-ring-page-order)
309  *      Notes:          1, 3
310  *
311  *      The size of the frontend allocated request ring buffer in units
312  *      of lb(machine pages). (e.g. 0 == 1 page, 1 = 2 pages, 2 == 4 pages,
313  *      etc.).
314  *
315  * num-ring-pages
316  *      Values:         <uint32_t>
317  *      Default Value:  1
318  *      Maximum Value:  MAX(max-ring-pages,(0x1 << max-ring-page-order))
319  *      Notes:          DEPRECATED, 2, 3
320  *
321  *      The size of the frontend allocated request ring buffer in units of
322  *      machine pages.  The value must be a power of 2.
323  *
324  * feature-persistent
325  *      Values:         0/1 (boolean)
326  *      Default Value:  0
327  *      Notes: 7, 8, 9
328  *
329  *      A value of "1" indicates that the frontend will reuse the same grants
330  *      for all transactions, allowing the backend to map them with write
331  *      access (even when it should be read-only). If the frontend hits the
332  *      maximum number of allowed persistently mapped grants, it can fallback
333  *      to non persistent mode. This will cause a performance degradation,
334  *      since the the backend driver will still try to map those grants
335  *      persistently. Since the persistent grants protocol is compatible with
336  *      the previous protocol, a frontend driver can choose to work in
337  *      persistent mode even when the backend doesn't support it.
338  *
339  *      It is recommended that the frontend driver stores the persistently
340  *      mapped grants in a LIFO queue, so a subset of all persistently mapped
341  *      grants gets used commonly. This is done in case the backend driver
342  *      decides to limit the maximum number of persistently mapped grants
343  *      to a value less than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
344  *
345  *------------------------- Virtual Device Properties -------------------------
346  *
347  * device-type
348  *      Values:         "disk", "cdrom", "floppy", etc.
349  *
350  * virtual-device
351  *      Values:         <uint32_t>
352  *
353  *      A value indicating the physical device to virtualize within the
354  *      frontend's domain.  (e.g. "The first ATA disk", "The third SCSI
355  *      disk", etc.)
356  *
357  *      See docs/misc/vbd-interface.txt for details on the format of this
358  *      value.
359  *
360  * Notes
361  * -----
362  * (1) Multi-page ring buffer scheme first developed in the Citrix XenServer
363  *     PV drivers.
364  * (2) Multi-page ring buffer scheme first used in some RedHat distributions
365  *     including a distribution deployed on certain nodes of the Amazon
366  *     EC2 cluster.
367  * (3) Support for multi-page ring buffers was implemented independently,
368  *     in slightly different forms, by both Citrix and RedHat/Amazon.
369  *     For full interoperability, block front and backends should publish
370  *     identical ring parameters, adjusted for unit differences, to the
371  *     XenStore nodes used in both schemes.
372  * (4) Devices that support discard functionality may internally allocate space
373  *     (discardable extents) in units that are larger than the exported logical
374  *     block size. If the backing device has such discardable extents the
375  *     backend should provide both discard-granularity and discard-alignment.
376  *     Providing just one of the two may be considered an error by the frontend.
377  *     Backends supporting discard should include discard-granularity and
378  *     discard-alignment even if it supports discarding individual sectors.
379  *     Frontends should assume discard-alignment == 0 and discard-granularity
380  *     == sector size if these keys are missing.
381  * (5) The discard-alignment parameter allows a physical device to be
382  *     partitioned into virtual devices that do not necessarily begin or
383  *     end on a discardable extent boundary.
384  * (6) When there is only a single page allocated to the request ring,
385  *     'ring-ref' is used to communicate the grant reference for this
386  *     page to the backend.  When using a multi-page ring, the 'ring-ref'
387  *     node is not created.  Instead 'ring-ref0' - 'ring-refN' are used.
388  * (7) When using persistent grants data has to be copied from/to the page
389  *     where the grant is currently mapped. The overhead of doing this copy
390  *     however doesn't suppress the speed improvement of not having to unmap
391  *     the grants.
392  * (8) The frontend driver has to allow the backend driver to map all grants
393  *     with write access, even when they should be mapped read-only, since
394  *     further requests may reuse these grants and require write permissions.
395  * (9) Linux implementation doesn't have a limit on the maximum number of
396  *     grants that can be persistently mapped in the frontend driver, but
397  *     due to the frontent driver implementation it should never be bigger
398  *     than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST.
399  *(10) The discard-secure property may be present and will be set to 1 if the
400  *     backing device supports secure discard.
401  *(11) Only used by Linux and NetBSD.
402  */
403 
404 /*
405  * Multiple hardware queues/rings:
406  * If supported, the backend will write the key "multi-queue-max-queues" to
407  * the directory for that vbd, and set its value to the maximum supported
408  * number of queues.
409  * Frontends that are aware of this feature and wish to use it can write the
410  * key "multi-queue-num-queues" with the number they wish to use, which must be
411  * greater than zero, and no more than the value reported by the backend in
412  * "multi-queue-max-queues".
413  *
414  * For frontends requesting just one queue, the usual event-channel and
415  * ring-ref keys are written as before, simplifying the backend processing
416  * to avoid distinguishing between a frontend that doesn't understand the
417  * multi-queue feature, and one that does, but requested only one queue.
418  *
419  * Frontends requesting two or more queues must not write the toplevel
420  * event-channel and ring-ref keys, instead writing those keys under sub-keys
421  * having the name "queue-N" where N is the integer ID of the queue/ring for
422  * which those keys belong. Queues are indexed from zero.
423  * For example, a frontend with two queues must write the following set of
424  * queue-related keys:
425  *
426  * /local/domain/1/device/vbd/0/multi-queue-num-queues = "2"
427  * /local/domain/1/device/vbd/0/queue-0 = ""
428  * /local/domain/1/device/vbd/0/queue-0/ring-ref = "<ring-ref#0>"
429  * /local/domain/1/device/vbd/0/queue-0/event-channel = "<evtchn#0>"
430  * /local/domain/1/device/vbd/0/queue-1 = ""
431  * /local/domain/1/device/vbd/0/queue-1/ring-ref = "<ring-ref#1>"
432  * /local/domain/1/device/vbd/0/queue-1/event-channel = "<evtchn#1>"
433  *
434  * It is also possible to use multiple queues/rings together with
435  * feature multi-page ring buffer.
436  * For example, a frontend requests two queues/rings and the size of each ring
437  * buffer is two pages must write the following set of related keys:
438  *
439  * /local/domain/1/device/vbd/0/multi-queue-num-queues = "2"
440  * /local/domain/1/device/vbd/0/ring-page-order = "1"
441  * /local/domain/1/device/vbd/0/queue-0 = ""
442  * /local/domain/1/device/vbd/0/queue-0/ring-ref0 = "<ring-ref#0>"
443  * /local/domain/1/device/vbd/0/queue-0/ring-ref1 = "<ring-ref#1>"
444  * /local/domain/1/device/vbd/0/queue-0/event-channel = "<evtchn#0>"
445  * /local/domain/1/device/vbd/0/queue-1 = ""
446  * /local/domain/1/device/vbd/0/queue-1/ring-ref0 = "<ring-ref#2>"
447  * /local/domain/1/device/vbd/0/queue-1/ring-ref1 = "<ring-ref#3>"
448  * /local/domain/1/device/vbd/0/queue-1/event-channel = "<evtchn#1>"
449  *
450  */
451 
452 /*
453  * STATE DIAGRAMS
454  *
455  *****************************************************************************
456  *                                   Startup                                 *
457  *****************************************************************************
458  *
459  * Tool stack creates front and back nodes with state XenbusStateInitialising.
460  *
461  * Front                                Back
462  * =================================    =====================================
463  * XenbusStateInitialising              XenbusStateInitialising
464  *  o Query virtual device               o Query backend device identification
465  *    properties.                          data.
466  *  o Setup OS device instance.          o Open and validate backend device.
467  *                                       o Publish backend features and
468  *                                         transport parameters.
469  *                                                      |
470  *                                                      |
471  *                                                      V
472  *                                      XenbusStateInitWait
473  *
474  * o Query backend features and
475  *   transport parameters.
476  * o Allocate and initialize the
477  *   request ring.
478  * o Publish transport parameters
479  *   that will be in effect during
480  *   this connection.
481  *              |
482  *              |
483  *              V
484  * XenbusStateInitialised
485  *
486  *                                       o Query frontend transport parameters.
487  *                                       o Connect to the request ring and
488  *                                         event channel.
489  *                                       o Publish backend device properties.
490  *                                                      |
491  *                                                      |
492  *                                                      V
493  *                                      XenbusStateConnected
494  *
495  *  o Query backend device properties.
496  *  o Finalize OS virtual device
497  *    instance.
498  *              |
499  *              |
500  *              V
501  * XenbusStateConnected
502  *
503  * Note: Drivers that do not support any optional features, or the negotiation
504  *       of transport parameters, can skip certain states in the state machine:
505  *
506  *       o A frontend may transition to XenbusStateInitialised without
507  *         waiting for the backend to enter XenbusStateInitWait.  In this
508  *         case, default transport parameters are in effect and any
509  *         transport parameters published by the frontend must contain
510  *         their default values.
511  *
512  *       o A backend may transition to XenbusStateInitialised, bypassing
513  *         XenbusStateInitWait, without waiting for the frontend to first
514  *         enter the XenbusStateInitialised state.  In this case, default
515  *         transport parameters are in effect and any transport parameters
516  *         published by the backend must contain their default values.
517  *
518  *       Drivers that support optional features and/or transport parameter
519  *       negotiation must tolerate these additional state transition paths.
520  *       In general this means performing the work of any skipped state
521  *       transition, if it has not already been performed, in addition to the
522  *       work associated with entry into the current state.
523  */
524 
525 /*
526  * REQUEST CODES.
527  */
528 #define BLKIF_OP_READ              0
529 #define BLKIF_OP_WRITE             1
530 /*
531  * All writes issued prior to a request with the BLKIF_OP_WRITE_BARRIER
532  * operation code ("barrier request") must be completed prior to the
533  * execution of the barrier request.  All writes issued after the barrier
534  * request must not execute until after the completion of the barrier request.
535  *
536  * Optional.  See "feature-barrier" XenBus node documentation above.
537  */
538 #define BLKIF_OP_WRITE_BARRIER     2
539 /*
540  * Commit any uncommitted contents of the backing device's volatile cache
541  * to stable storage.
542  *
543  * Optional.  See "feature-flush-cache" XenBus node documentation above.
544  */
545 #define BLKIF_OP_FLUSH_DISKCACHE   3
546 /*
547  * Used in SLES sources for device specific command packet
548  * contained within the request. Reserved for that purpose.
549  */
550 #define BLKIF_OP_RESERVED_1        4
551 /*
552  * Indicate to the backend device that a region of storage is no longer in
553  * use, and may be discarded at any time without impact to the client.  If
554  * the BLKIF_DISCARD_SECURE flag is set on the request, all copies of the
555  * discarded region on the device must be rendered unrecoverable before the
556  * command returns.
557  *
558  * This operation is analogous to performing a trim (ATA) or unamp (SCSI),
559  * command on a native device.
560  *
561  * More information about trim/unmap operations can be found at:
562  * http://t13.org/Documents/UploadedDocuments/docs2008/
563  *     e07154r6-Data_Set_Management_Proposal_for_ATA-ACS2.doc
564  * http://www.seagate.com/staticfiles/support/disc/manuals/
565  *     Interface%20manuals/100293068c.pdf
566  *
567  * Optional.  See "feature-discard", "discard-alignment",
568  * "discard-granularity", and "discard-secure" in the XenBus node
569  * documentation above.
570  */
571 #define BLKIF_OP_DISCARD           5
572 
573 /*
574  * Recognized if "feature-max-indirect-segments" in present in the backend
575  * xenbus info. The "feature-max-indirect-segments" node contains the maximum
576  * number of segments allowed by the backend per request. If the node is
577  * present, the frontend might use blkif_request_indirect structs in order to
578  * issue requests with more than BLKIF_MAX_SEGMENTS_PER_REQUEST (11). The
579  * maximum number of indirect segments is fixed by the backend, but the
580  * frontend can issue requests with any number of indirect segments as long as
581  * it's less than the number provided by the backend. The indirect_grefs field
582  * in blkif_request_indirect should be filled by the frontend with the
583  * grant references of the pages that are holding the indirect segments.
584  * These pages are filled with an array of blkif_request_segment that hold the
585  * information about the segments. The number of indirect pages to use is
586  * determined by the number of segments an indirect request contains. Every
587  * indirect page can contain a maximum of
588  * (PAGE_SIZE / sizeof(struct blkif_request_segment)) segments, so to
589  * calculate the number of indirect pages to use we have to do
590  * ceil(indirect_segments / (PAGE_SIZE / sizeof(struct blkif_request_segment))).
591  *
592  * If a backend does not recognize BLKIF_OP_INDIRECT, it should *not*
593  * create the "feature-max-indirect-segments" node!
594  */
595 #define BLKIF_OP_INDIRECT          6
596 
597 /*
598  * Maximum scatter/gather segments per request.
599  * This is carefully chosen so that sizeof(blkif_ring_t) <= PAGE_SIZE.
600  * NB. This could be 12 if the ring indexes weren't stored in the same page.
601  */
602 #define BLKIF_MAX_SEGMENTS_PER_REQUEST 11
603 
604 /*
605  * Maximum number of indirect pages to use per request.
606  */
607 #define BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST 8
608 
609 /*
610  * NB. first_sect and last_sect in blkif_request_segment, as well as
611  * sector_number in blkif_request, are always expressed in 512-byte units.
612  * However they must be properly aligned to the real sector size of the
613  * physical disk, which is reported in the "physical-sector-size" node in
614  * the backend xenbus info. Also the xenbus "sectors" node is expressed in
615  * 512-byte units.
616  */
617 struct blkif_request_segment {
618     grant_ref_t gref;        /* reference to I/O buffer frame        */
619     /* @first_sect: first sector in frame to transfer (inclusive).   */
620     /* @last_sect: last sector in frame to transfer (inclusive).     */
621     uint8_t     first_sect, last_sect;
622 };
623 
624 /*
625  * Starting ring element for any I/O request.
626  */
627 struct blkif_request {
628     uint8_t        operation;    /* BLKIF_OP_???                         */
629     uint8_t        nr_segments;  /* number of segments                   */
630     blkif_vdev_t   handle;       /* only for read/write requests         */
631     uint64_t       id;           /* private guest value, echoed in resp  */
632     blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
633     struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
634 };
635 typedef struct blkif_request blkif_request_t;
636 
637 /*
638  * Cast to this structure when blkif_request.operation == BLKIF_OP_DISCARD
639  * sizeof(struct blkif_request_discard) <= sizeof(struct blkif_request)
640  */
641 struct blkif_request_discard {
642     uint8_t        operation;    /* BLKIF_OP_DISCARD                     */
643     uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
644 #define BLKIF_DISCARD_SECURE (1<<0)  /* ignored if discard-secure=0      */
645     blkif_vdev_t   handle;       /* same as for read/write requests      */
646     uint64_t       id;           /* private guest value, echoed in resp  */
647     blkif_sector_t sector_number;/* start sector idx on disk             */
648     uint64_t       nr_sectors;   /* number of contiguous sectors to discard*/
649 };
650 typedef struct blkif_request_discard blkif_request_discard_t;
651 
652 struct blkif_request_indirect {
653     uint8_t        operation;    /* BLKIF_OP_INDIRECT                    */
654     uint8_t        indirect_op;  /* BLKIF_OP_{READ/WRITE}                */
655     uint16_t       nr_segments;  /* number of segments                   */
656     uint64_t       id;           /* private guest value, echoed in resp  */
657     blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
658     blkif_vdev_t   handle;       /* same as for read/write requests      */
659     grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
660 #ifdef __i386__
661     uint64_t       pad;          /* Make it 64 byte aligned on i386      */
662 #endif
663 };
664 typedef struct blkif_request_indirect blkif_request_indirect_t;
665 
666 struct blkif_response {
667     uint64_t        id;              /* copied from request */
668     uint8_t         operation;       /* copied from request */
669     int16_t         status;          /* BLKIF_RSP_???       */
670 };
671 typedef struct blkif_response blkif_response_t;
672 
673 /*
674  * STATUS RETURN CODES.
675  */
676  /* Operation not supported (only happens on barrier writes). */
677 #define BLKIF_RSP_EOPNOTSUPP  -2
678  /* Operation failed for some unspecified reason (-EIO). */
679 #define BLKIF_RSP_ERROR       -1
680  /* Operation completed successfully. */
681 #define BLKIF_RSP_OKAY         0
682 
683 /*
684  * Generate blkif ring structures and types.
685  */
686 DEFINE_RING_TYPES(blkif, struct blkif_request, struct blkif_response);
687 
688 #define VDISK_CDROM        0x1
689 #define VDISK_REMOVABLE    0x2
690 #define VDISK_READONLY     0x4
691 
692 #endif /* __XEN_PUBLIC_IO_BLKIF_H__ */
693 
694 /*
695  * Local variables:
696  * mode: C
697  * c-file-style: "BSD"
698  * c-basic-offset: 4
699  * tab-width: 4
700  * indent-tabs-mode: nil
701  * End:
702  */
703