2 * QEMU System Emulator block driver
4 * Copyright (c) 2003 Fabrice Bellard
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 deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * 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
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 #ifndef BLOCK_INT_COMMON_H
25 #define BLOCK_INT_COMMON_H
27 #include "block/aio.h"
28 #include "block/block-common.h"
29 #include "block/block-global-state.h"
30 #include "block/snapshot.h"
33 #include "qemu/stats64.h"
35 #define BLOCK_FLAG_LAZY_REFCOUNTS 8
37 #define BLOCK_OPT_SIZE "size"
38 #define BLOCK_OPT_ENCRYPT "encryption"
39 #define BLOCK_OPT_ENCRYPT_FORMAT "encrypt.format"
40 #define BLOCK_OPT_COMPAT6 "compat6"
41 #define BLOCK_OPT_HWVERSION "hwversion"
42 #define BLOCK_OPT_BACKING_FILE "backing_file"
43 #define BLOCK_OPT_BACKING_FMT "backing_fmt"
44 #define BLOCK_OPT_CLUSTER_SIZE "cluster_size"
45 #define BLOCK_OPT_TABLE_SIZE "table_size"
46 #define BLOCK_OPT_PREALLOC "preallocation"
47 #define BLOCK_OPT_SUBFMT "subformat"
48 #define BLOCK_OPT_COMPAT_LEVEL "compat"
49 #define BLOCK_OPT_LAZY_REFCOUNTS "lazy_refcounts"
50 #define BLOCK_OPT_ADAPTER_TYPE "adapter_type"
51 #define BLOCK_OPT_REDUNDANCY "redundancy"
52 #define BLOCK_OPT_NOCOW "nocow"
53 #define BLOCK_OPT_EXTENT_SIZE_HINT "extent_size_hint"
54 #define BLOCK_OPT_OBJECT_SIZE "object_size"
55 #define BLOCK_OPT_REFCOUNT_BITS "refcount_bits"
56 #define BLOCK_OPT_DATA_FILE "data_file"
57 #define BLOCK_OPT_DATA_FILE_RAW "data_file_raw"
58 #define BLOCK_OPT_COMPRESSION_TYPE "compression_type"
59 #define BLOCK_OPT_EXTL2 "extended_l2"
61 #define BLOCK_PROBE_BUF_SIZE 512
63 enum BdrvTrackedRequestType
{
67 BDRV_TRACKED_TRUNCATE
,
71 * That is not quite good that BdrvTrackedRequest structure is public,
72 * as block/io.c is very careful about incoming offset/bytes being
73 * correct. Be sure to assert bdrv_check_request() succeeded after any
74 * modification of BdrvTrackedRequest object out of block/io.c
76 typedef struct BdrvTrackedRequest
{
80 enum BdrvTrackedRequestType type
;
83 int64_t overlap_offset
;
84 int64_t overlap_bytes
;
86 QLIST_ENTRY(BdrvTrackedRequest
) list
;
87 Coroutine
*co
; /* owner, used for deadlock detection */
88 CoQueue wait_queue
; /* coroutines blocked on this request */
90 struct BdrvTrackedRequest
*waiting_for
;
96 * These fields are initialized when this object is created,
97 * and are never changed afterwards.
100 const char *format_name
;
104 * Set to true if the BlockDriver is a block filter. Block filters pass
105 * certain callbacks that refer to data (see block.c) to their bs->file
106 * or bs->backing (whichever one exists) if the driver doesn't implement
107 * them. Drivers that do not wish to forward must implement them and return
109 * Note that filters are not allowed to modify data.
111 * Filters generally cannot have more than a single filtered child,
112 * because the data they present must at all times be the same as
113 * that on their filtered child. That would be impossible to
114 * achieve for multiple filtered children.
115 * (And this filtered child must then be bs->file or bs->backing.)
119 * Only make sense for filter drivers, for others must be false.
120 * If true, filtered child is bs->backing. Otherwise it's bs->file.
121 * Two internal filters use bs->backing as filtered child and has this
122 * field set to true: mirror_top and commit_top. There also two such test
123 * filters in tests/unit/test-bdrv-graph-mod.c.
125 * Never create any more such filters!
127 * TODO: imagine how to deprecate this behavior and make all filters work
128 * similarly using bs->file as filtered child.
130 bool filtered_child_is_backing
;
133 * Set to true if the BlockDriver is a format driver. Format nodes
134 * generally do not expect their children to be other format nodes
135 * (except for backing files), and so format probing is disabled
141 * Drivers not implementing bdrv_parse_filename nor bdrv_open should have
142 * this field set to true, except ones that are defined only by their
144 * An example of the last type will be the quorum block driver.
146 bool bdrv_needs_filename
;
149 * Set if a driver can support backing files. This also implies the
150 * following semantics:
152 * - Return status 0 of .bdrv_co_block_status means that corresponding
153 * blocks are not allocated in this layer of backing-chain
154 * - For such (unallocated) blocks, read will:
155 * - fill buffer with zeros if there is no backing file
156 * - read from the backing file otherwise, where the block layer
157 * takes care of reading zeros beyond EOF if backing file is short
159 bool supports_backing
;
161 bool has_variable_length
;
164 * Drivers setting this field must be able to work with just a plain
165 * filename with '<protocol_name>:' as a prefix, and no other options.
166 * Options may be extracted from the filename by implementing
167 * bdrv_parse_filename.
169 const char *protocol_name
;
171 /* List of options for creating images, terminated by name == NULL */
172 QemuOptsList
*create_opts
;
174 /* List of options for image amend */
175 QemuOptsList
*amend_opts
;
178 * If this driver supports reopening images this contains a
179 * NULL-terminated list of the runtime options that can be
180 * modified. If an option in this list is unspecified during
181 * reopen then it _must_ be reset to its default value or return
184 const char *const *mutable_opts
;
187 * Pointer to a NULL-terminated array of names of strong options
188 * that can be specified for bdrv_open(). A strong option is one
189 * that changes the data of a BDS.
190 * If this pointer is NULL, the array is considered empty.
191 * "filename" and "driver" are always considered strong.
193 const char *const *strong_runtime_opts
;
197 * Global state (GS) API. These functions run under the BQL.
199 * See include/block/block-global-state.h for more information about
204 * This function is invoked under BQL before .bdrv_co_amend()
205 * (which in contrast does not necessarily run under the BQL)
206 * to allow driver-specific initialization code that requires
207 * the BQL, like setting up specific permission flags.
209 int (*bdrv_amend_pre_run
)(BlockDriverState
*bs
, Error
**errp
);
211 * This function is invoked under BQL after .bdrv_co_amend()
212 * to allow cleaning up what was done in .bdrv_amend_pre_run().
214 void (*bdrv_amend_clean
)(BlockDriverState
*bs
);
217 * Return true if @to_replace can be replaced by a BDS with the
218 * same data as @bs without it affecting @bs's behavior (that is,
219 * without it being visible to @bs's parents).
221 bool (*bdrv_recurse_can_replace
)(BlockDriverState
*bs
,
222 BlockDriverState
*to_replace
);
224 int (*bdrv_probe_device
)(const char *filename
);
227 * Any driver implementing this callback is expected to be able to handle
228 * NULL file names in its .bdrv_open() implementation.
230 void (*bdrv_parse_filename
)(const char *filename
, QDict
*options
,
233 /* For handling image reopen for split or non-split files. */
234 int (*bdrv_reopen_prepare
)(BDRVReopenState
*reopen_state
,
235 BlockReopenQueue
*queue
, Error
**errp
);
236 void (*bdrv_reopen_commit
)(BDRVReopenState
*reopen_state
);
237 void (*bdrv_reopen_commit_post
)(BDRVReopenState
*reopen_state
);
238 void (*bdrv_reopen_abort
)(BDRVReopenState
*reopen_state
);
239 void (*bdrv_join_options
)(QDict
*options
, QDict
*old_options
);
241 int (*bdrv_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
244 /* Protocol drivers should implement this instead of bdrv_open */
245 int (*bdrv_file_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
247 void (*bdrv_close
)(BlockDriverState
*bs
);
249 int coroutine_fn (*bdrv_co_create
)(BlockdevCreateOptions
*opts
,
251 int coroutine_fn (*bdrv_co_create_opts
)(BlockDriver
*drv
,
252 const char *filename
,
256 int (*bdrv_amend_options
)(BlockDriverState
*bs
,
258 BlockDriverAmendStatusCB
*status_cb
,
263 int (*bdrv_make_empty
)(BlockDriverState
*bs
);
266 * Refreshes the bs->exact_filename field. If that is impossible,
267 * bs->exact_filename has to be left empty.
269 void (*bdrv_refresh_filename
)(BlockDriverState
*bs
);
272 * Gathers the open options for all children into @target.
273 * A simple format driver (without backing file support) might
274 * implement this function like this:
276 * QINCREF(bs->file->bs->full_open_options);
277 * qdict_put(target, "file", bs->file->bs->full_open_options);
279 * If not specified, the generic implementation will simply put
280 * all children's options under their respective name.
282 * @backing_overridden is true when bs->backing seems not to be
283 * the child that would result from opening bs->backing_file.
284 * Therefore, if it is true, the backing child's options should be
285 * gathered; otherwise, there is no need since the backing child
286 * is the one implied by the image header.
288 * Note that ideally this function would not be needed. Every
289 * block driver which implements it is probably doing something
290 * shady regarding its runtime option structure.
292 void (*bdrv_gather_child_options
)(BlockDriverState
*bs
, QDict
*target
,
293 bool backing_overridden
);
296 * Returns an allocated string which is the directory name of this BDS: It
297 * will be used to make relative filenames absolute by prepending this
298 * function's return value to them.
300 char *(*bdrv_dirname
)(BlockDriverState
*bs
, Error
**errp
);
303 * This informs the driver that we are no longer interested in the result
304 * of in-flight requests, so don't waste the time if possible.
306 * One example usage is to avoid waiting for an nbd target node reconnect
307 * timeout during job-cancel with force=true.
309 void (*bdrv_cancel_in_flight
)(BlockDriverState
*bs
);
311 int (*bdrv_inactivate
)(BlockDriverState
*bs
);
313 int (*bdrv_snapshot_create
)(BlockDriverState
*bs
,
314 QEMUSnapshotInfo
*sn_info
);
315 int (*bdrv_snapshot_goto
)(BlockDriverState
*bs
,
316 const char *snapshot_id
);
317 int (*bdrv_snapshot_delete
)(BlockDriverState
*bs
,
318 const char *snapshot_id
,
321 int (*bdrv_snapshot_list
)(BlockDriverState
*bs
,
322 QEMUSnapshotInfo
**psn_info
);
323 int (*bdrv_snapshot_load_tmp
)(BlockDriverState
*bs
,
324 const char *snapshot_id
,
328 int (*bdrv_change_backing_file
)(BlockDriverState
*bs
,
329 const char *backing_file
, const char *backing_fmt
);
331 /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
332 int (*bdrv_debug_breakpoint
)(BlockDriverState
*bs
, const char *event
,
334 int (*bdrv_debug_remove_breakpoint
)(BlockDriverState
*bs
,
336 int (*bdrv_debug_resume
)(BlockDriverState
*bs
, const char *tag
);
337 bool (*bdrv_debug_is_suspended
)(BlockDriverState
*bs
, const char *tag
);
339 void (*bdrv_refresh_limits
)(BlockDriverState
*bs
, Error
**errp
);
342 * Returns 1 if newly created images are guaranteed to contain only
343 * zeros, 0 otherwise.
345 int (*bdrv_has_zero_init
)(BlockDriverState
*bs
);
348 * Remove fd handlers, timers, and other event loop callbacks so the event
349 * loop is no longer in use. Called with no in-flight requests and in
350 * depth-first traversal order with parents before child nodes.
352 void (*bdrv_detach_aio_context
)(BlockDriverState
*bs
);
355 * Add fd handlers, timers, and other event loop callbacks so I/O requests
356 * can be processed again. Called with no in-flight requests and in
357 * depth-first traversal order with child nodes before parent nodes.
359 void (*bdrv_attach_aio_context
)(BlockDriverState
*bs
,
360 AioContext
*new_context
);
363 * Try to get @bs's logical and physical block size.
364 * On success, store them in @bsz and return zero.
365 * On failure, return negative errno.
367 int (*bdrv_probe_blocksizes
)(BlockDriverState
*bs
, BlockSizes
*bsz
);
369 * Try to get @bs's geometry (cyls, heads, sectors)
370 * On success, store them in @geo and return 0.
371 * On failure return -errno.
372 * Only drivers that want to override guest geometry implement this
373 * callback; see hd_geometry_guess().
375 int (*bdrv_probe_geometry
)(BlockDriverState
*bs
, HDGeometry
*geo
);
377 void (*bdrv_add_child
)(BlockDriverState
*parent
, BlockDriverState
*child
,
379 void (*bdrv_del_child
)(BlockDriverState
*parent
, BdrvChild
*child
,
383 * Informs the block driver that a permission change is intended. The
384 * driver checks whether the change is permissible and may take other
385 * preparations for the change (e.g. get file system locks). This operation
386 * is always followed either by a call to either .bdrv_set_perm or
387 * .bdrv_abort_perm_update.
389 * Checks whether the requested set of cumulative permissions in @perm
390 * can be granted for accessing @bs and whether no other users are using
391 * permissions other than those given in @shared (both arguments take
392 * BLK_PERM_* bitmasks).
394 * If both conditions are met, 0 is returned. Otherwise, -errno is returned
395 * and errp is set to an error describing the conflict.
397 int (*bdrv_check_perm
)(BlockDriverState
*bs
, uint64_t perm
,
398 uint64_t shared
, Error
**errp
);
401 * Called to inform the driver that the set of cumulative set of used
402 * permissions for @bs has changed to @perm, and the set of sharable
403 * permission to @shared. The driver can use this to propagate changes to
404 * its children (i.e. request permissions only if a parent actually needs
407 * This function is only invoked after bdrv_check_perm(), so block drivers
408 * may rely on preparations made in their .bdrv_check_perm implementation.
410 void (*bdrv_set_perm
)(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
);
413 * Called to inform the driver that after a previous bdrv_check_perm()
414 * call, the permission update is not performed and any preparations made
415 * for it (e.g. taken file locks) need to be undone.
417 * This function can be called even for nodes that never saw a
418 * bdrv_check_perm() call. It is a no-op then.
420 void (*bdrv_abort_perm_update
)(BlockDriverState
*bs
);
423 * Returns in @nperm and @nshared the permissions that the driver for @bs
424 * needs on its child @c, based on the cumulative permissions requested by
425 * the parents in @parent_perm and @parent_shared.
427 * If @c is NULL, return the permissions for attaching a new child for the
428 * given @child_class and @role.
430 * If @reopen_queue is non-NULL, don't return the currently needed
431 * permissions, but those that will be needed after applying the
434 void (*bdrv_child_perm
)(BlockDriverState
*bs
, BdrvChild
*c
,
436 BlockReopenQueue
*reopen_queue
,
437 uint64_t parent_perm
, uint64_t parent_shared
,
438 uint64_t *nperm
, uint64_t *nshared
);
441 * Register/unregister a buffer for I/O. For example, when the driver is
442 * interested to know the memory areas that will later be used in iovs, so
443 * that it can do IOMMU mapping with VFIO etc., in order to get better
444 * performance. In the case of VFIO drivers, this callback is used to do
445 * DMA mapping for hot buffers.
447 * Returns: true on success, false on failure
449 bool (*bdrv_register_buf
)(BlockDriverState
*bs
, void *host
, size_t size
,
451 void (*bdrv_unregister_buf
)(BlockDriverState
*bs
, void *host
, size_t size
);
454 * This field is modified only under the BQL, and is part of
457 QLIST_ENTRY(BlockDriver
) list
;
460 * I/O API functions. These functions are thread-safe.
462 * See include/block/block-io.h for more information about
466 int (*bdrv_probe
)(const uint8_t *buf
, int buf_size
, const char *filename
);
468 int coroutine_fn (*bdrv_co_amend
)(BlockDriverState
*bs
,
469 BlockdevAmendOptions
*opts
,
474 BlockAIOCB
*(*bdrv_aio_preadv
)(BlockDriverState
*bs
,
475 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
476 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
477 BlockAIOCB
*(*bdrv_aio_pwritev
)(BlockDriverState
*bs
,
478 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
479 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
480 BlockAIOCB
* GRAPH_RDLOCK_PTR (*bdrv_aio_flush
)(
481 BlockDriverState
*bs
, BlockCompletionFunc
*cb
, void *opaque
);
483 BlockAIOCB
* GRAPH_RDLOCK_PTR (*bdrv_aio_pdiscard
)(
484 BlockDriverState
*bs
, int64_t offset
, int bytes
,
485 BlockCompletionFunc
*cb
, void *opaque
);
487 int coroutine_fn (*bdrv_co_readv
)(BlockDriverState
*bs
,
488 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
);
491 * @offset: position in bytes to read at
492 * @bytes: number of bytes to read
493 * @qiov: the buffers to fill with read data
494 * @flags: currently unused, always 0
496 * @offset and @bytes will be a multiple of 'request_alignment',
497 * but the length of individual @qiov elements does not have to
500 * @bytes will always equal the total size of @qiov, and will be
501 * no larger than 'max_transfer'.
503 * The buffer in @qiov may point directly to guest memory.
505 int coroutine_fn (*bdrv_co_preadv
)(BlockDriverState
*bs
,
506 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
507 BdrvRequestFlags flags
);
509 int coroutine_fn (*bdrv_co_preadv_part
)(BlockDriverState
*bs
,
510 int64_t offset
, int64_t bytes
,
511 QEMUIOVector
*qiov
, size_t qiov_offset
,
512 BdrvRequestFlags flags
);
514 int coroutine_fn (*bdrv_co_writev
)(BlockDriverState
*bs
,
515 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
518 * @offset: position in bytes to write at
519 * @bytes: number of bytes to write
520 * @qiov: the buffers containing data to write
521 * @flags: zero or more bits allowed by 'supported_write_flags'
523 * @offset and @bytes will be a multiple of 'request_alignment',
524 * but the length of individual @qiov elements does not have to
527 * @bytes will always equal the total size of @qiov, and will be
528 * no larger than 'max_transfer'.
530 * The buffer in @qiov may point directly to guest memory.
532 int coroutine_fn (*bdrv_co_pwritev
)(BlockDriverState
*bs
,
533 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
534 BdrvRequestFlags flags
);
535 int coroutine_fn (*bdrv_co_pwritev_part
)(BlockDriverState
*bs
,
536 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
537 BdrvRequestFlags flags
);
540 * Efficiently zero a region of the disk image. Typically an image format
541 * would use a compact metadata representation to implement this. This
542 * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
543 * will be called instead.
545 int coroutine_fn (*bdrv_co_pwrite_zeroes
)(BlockDriverState
*bs
,
546 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
);
548 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard
)(
549 BlockDriverState
*bs
, int64_t offset
, int64_t bytes
);
552 * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
553 * and invoke bdrv_co_copy_range_from(child, ...), or invoke
554 * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
556 * See the comment of bdrv_co_copy_range for the parameter and return value
559 int coroutine_fn (*bdrv_co_copy_range_from
)(BlockDriverState
*bs
,
565 BdrvRequestFlags read_flags
,
566 BdrvRequestFlags write_flags
);
569 * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
570 * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
571 * operation if @bs is the leaf and @src has the same BlockDriver. Return
572 * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
574 * See the comment of bdrv_co_copy_range for the parameter and return value
577 int coroutine_fn (*bdrv_co_copy_range_to
)(BlockDriverState
*bs
,
583 BdrvRequestFlags read_flags
,
584 BdrvRequestFlags write_flags
);
587 * Building block for bdrv_block_status[_above] and
588 * bdrv_is_allocated[_above]. The driver should answer only
589 * according to the current layer, and should only need to set
590 * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
591 * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
592 * layer, the result should be 0 (and not BDRV_BLOCK_ZERO). See
593 * block.h for the overall meaning of the bits. As a hint, the
594 * flag want_zero is true if the caller cares more about precise
595 * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
596 * overall allocation (favor larger *pnum, perhaps by reporting
597 * _DATA instead of _ZERO). The block layer guarantees input
598 * clamped to bdrv_getlength() and aligned to request_alignment,
599 * as well as non-NULL pnum, map, and file; in turn, the driver
600 * must return an error or set pnum to an aligned non-zero value.
602 * Note that @bytes is just a hint on how big of a region the
603 * caller wants to inspect. It is not a limit on *pnum.
604 * Implementations are free to return larger values of *pnum if
605 * doing so does not incur a performance penalty.
607 * block/io.c's bdrv_co_block_status() will utilize an unclamped
608 * *pnum value for the block-status cache on protocol nodes, prior
609 * to clamping *pnum for return to its caller.
611 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_block_status
)(
612 BlockDriverState
*bs
,
613 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
614 int64_t *map
, BlockDriverState
**file
);
617 * Snapshot-access API.
619 * Block-driver may provide snapshot-access API: special functions to access
620 * some internal "snapshot". The functions are similar with normal
621 * read/block_status/discard handler, but don't have any specific handling
622 * in generic block-layer: no serializing, no alignment, no tracked
623 * requests. So, block-driver that realizes these APIs is fully responsible
624 * for synchronization between snapshot-access API and normal IO requests.
626 * TODO: To be able to support qcow2's internal snapshots, this API will
627 * need to be extended to:
628 * - be able to select a specific snapshot
629 * - receive the snapshot's actual length (which may differ from bs's
632 int coroutine_fn (*bdrv_co_preadv_snapshot
)(BlockDriverState
*bs
,
633 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
);
634 int coroutine_fn (*bdrv_co_snapshot_block_status
)(BlockDriverState
*bs
,
635 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
636 int64_t *map
, BlockDriverState
**file
);
638 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard_snapshot
)(
639 BlockDriverState
*bs
, int64_t offset
, int64_t bytes
);
642 * Invalidate any cached meta-data.
644 void coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_invalidate_cache
)(
645 BlockDriverState
*bs
, Error
**errp
);
648 * Flushes all data for all layers by calling bdrv_co_flush for underlying
649 * layers, if needed. This function is needed for deterministic
650 * synchronization of the flush finishing callback.
652 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_flush
)(BlockDriverState
*bs
);
654 /* Delete a created file. */
655 int coroutine_fn (*bdrv_co_delete_file
)(BlockDriverState
*bs
,
659 * Flushes all data that was already written to the OS all the way down to
660 * the disk (for example file-posix.c calls fsync()).
662 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_disk
)(
663 BlockDriverState
*bs
);
666 * Flushes all internal caches to the OS. The data may still sit in a
667 * writeback cache of the host OS, but it will survive a crash of the qemu
670 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_os
)(
671 BlockDriverState
*bs
);
674 * Truncate @bs to @offset bytes using the given @prealloc mode
675 * when growing. Modes other than PREALLOC_MODE_OFF should be
676 * rejected when shrinking @bs.
678 * If @exact is true, @bs must be resized to exactly @offset.
679 * Otherwise, it is sufficient for @bs (if it is a host block
680 * device and thus there is no way to resize it) to be at least
681 * @offset bytes in length.
683 * If @exact is true and this function fails but would succeed
684 * with @exact = false, it should return -ENOTSUP.
686 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_truncate
)(
687 BlockDriverState
*bs
, int64_t offset
, bool exact
,
688 PreallocMode prealloc
, BdrvRequestFlags flags
, Error
**errp
);
690 int64_t coroutine_fn (*bdrv_co_getlength
)(BlockDriverState
*bs
);
691 int64_t coroutine_fn (*bdrv_co_get_allocated_file_size
)(
692 BlockDriverState
*bs
);
694 BlockMeasureInfo
*(*bdrv_measure
)(QemuOpts
*opts
, BlockDriverState
*in_bs
,
697 int coroutine_fn (*bdrv_co_pwritev_compressed
)(BlockDriverState
*bs
,
698 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
);
699 int coroutine_fn (*bdrv_co_pwritev_compressed_part
)(BlockDriverState
*bs
,
700 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
703 int coroutine_fn (*bdrv_co_get_info
)(BlockDriverState
*bs
,
704 BlockDriverInfo
*bdi
);
706 ImageInfoSpecific
*(*bdrv_get_specific_info
)(BlockDriverState
*bs
,
708 BlockStatsSpecific
*(*bdrv_get_specific_stats
)(BlockDriverState
*bs
);
710 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_save_vmstate
)(
711 BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
);
713 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_load_vmstate
)(
714 BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
);
716 /* removable device specific */
717 bool coroutine_fn (*bdrv_co_is_inserted
)(BlockDriverState
*bs
);
718 void coroutine_fn (*bdrv_co_eject
)(BlockDriverState
*bs
, bool eject_flag
);
719 void coroutine_fn (*bdrv_co_lock_medium
)(BlockDriverState
*bs
, bool locked
);
721 /* to control generic scsi devices */
722 BlockAIOCB
*coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_aio_ioctl
)(
723 BlockDriverState
*bs
, unsigned long int req
, void *buf
,
724 BlockCompletionFunc
*cb
, void *opaque
);
726 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_ioctl
)(
727 BlockDriverState
*bs
, unsigned long int req
, void *buf
);
730 * Returns 0 for completed check, -errno for internal errors.
731 * The check results are stored in result.
733 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_check
)(
734 BlockDriverState
*bs
, BdrvCheckResult
*result
, BdrvCheckMode fix
);
736 void coroutine_fn (*bdrv_co_debug_event
)(BlockDriverState
*bs
,
737 BlkdebugEvent event
);
739 /* io queue for linux-aio */
740 void coroutine_fn (*bdrv_co_io_plug
)(BlockDriverState
*bs
);
741 void coroutine_fn (*bdrv_co_io_unplug
)(BlockDriverState
*bs
);
744 * bdrv_drain_begin is called if implemented in the beginning of a
745 * drain operation to drain and stop any internal sources of requests in
747 * bdrv_drain_end is called if implemented at the end of the drain.
749 * They should be used by the driver to e.g. manage scheduled I/O
750 * requests, or toggle an internal state. After the end of the drain new
751 * requests will continue normally.
753 * Implementations of both functions must not call aio_poll().
755 void (*bdrv_drain_begin
)(BlockDriverState
*bs
);
756 void (*bdrv_drain_end
)(BlockDriverState
*bs
);
758 bool (*bdrv_supports_persistent_dirty_bitmap
)(BlockDriverState
*bs
);
759 bool coroutine_fn (*bdrv_co_can_store_new_dirty_bitmap
)(
760 BlockDriverState
*bs
, const char *name
, uint32_t granularity
,
762 int coroutine_fn (*bdrv_co_remove_persistent_dirty_bitmap
)(
763 BlockDriverState
*bs
, const char *name
, Error
**errp
);
766 static inline bool block_driver_can_compress(BlockDriver
*drv
)
768 return drv
->bdrv_co_pwritev_compressed
||
769 drv
->bdrv_co_pwritev_compressed_part
;
772 typedef struct BlockLimits
{
774 * Alignment requirement, in bytes, for offset/length of I/O
775 * requests. Must be a power of 2 less than INT_MAX; defaults to
776 * 1 for drivers with modern byte interfaces, and to 512
779 uint32_t request_alignment
;
782 * Maximum number of bytes that can be discarded at once. Must be multiple
783 * of pdiscard_alignment, but need not be power of 2. May be 0 if no
784 * inherent 64-bit limit.
786 int64_t max_pdiscard
;
789 * Optimal alignment for discard requests in bytes. A power of 2
790 * is best but not mandatory. Must be a multiple of
791 * bl.request_alignment, and must be less than max_pdiscard if
792 * that is set. May be 0 if bl.request_alignment is good enough
794 uint32_t pdiscard_alignment
;
797 * Maximum number of bytes that can zeroized at once. Must be multiple of
798 * pwrite_zeroes_alignment. 0 means no limit.
800 int64_t max_pwrite_zeroes
;
803 * Optimal alignment for write zeroes requests in bytes. A power
804 * of 2 is best but not mandatory. Must be a multiple of
805 * bl.request_alignment, and must be less than max_pwrite_zeroes
806 * if that is set. May be 0 if bl.request_alignment is good
809 uint32_t pwrite_zeroes_alignment
;
812 * Optimal transfer length in bytes. A power of 2 is best but not
813 * mandatory. Must be a multiple of bl.request_alignment, or 0 if
816 uint32_t opt_transfer
;
819 * Maximal transfer length in bytes. Need not be power of 2, but
820 * must be multiple of opt_transfer and bl.request_alignment, or 0
821 * for no 32-bit limit. For now, anything larger than INT_MAX is
824 uint32_t max_transfer
;
827 * Maximal hardware transfer length in bytes. Applies whenever
828 * transfers to the device bypass the kernel I/O scheduler, for
829 * example with SG_IO. If larger than max_transfer or if zero,
830 * blk_get_max_hw_transfer will fall back to max_transfer.
832 uint64_t max_hw_transfer
;
835 * Maximal number of scatter/gather elements allowed by the hardware.
836 * Applies whenever transfers to the device bypass the kernel I/O
837 * scheduler, for example with SG_IO. If larger than max_iov
838 * or if zero, blk_get_max_hw_iov will fall back to max_iov.
843 /* memory alignment, in bytes so that no bounce buffer is needed */
844 size_t min_mem_alignment
;
846 /* memory alignment, in bytes, for bounce buffer */
847 size_t opt_mem_alignment
;
849 /* maximum number of iovec elements */
853 typedef struct BdrvOpBlocker BdrvOpBlocker
;
855 typedef struct BdrvAioNotifier
{
856 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
857 void (*detach_aio_context
)(void *opaque
);
862 QLIST_ENTRY(BdrvAioNotifier
) list
;
865 struct BdrvChildClass
{
867 * If true, bdrv_replace_node() doesn't change the node this BdrvChild
873 * If true, the parent is a BlockDriverState and bdrv_next_all_states()
874 * will return it. This information is used for drain_all, where every node
875 * will be drained separately, so the drain only needs to be propagated to
881 * Global state (GS) API. These functions run under the BQL.
883 * See include/block/block-global-state.h for more information about
886 void (*inherit_options
)(BdrvChildRole role
, bool parent_is_format
,
887 int *child_flags
, QDict
*child_options
,
888 int parent_flags
, QDict
*parent_options
);
889 void (*change_media
)(BdrvChild
*child
, bool load
);
892 * Returns a malloced string that describes the parent of the child for a
893 * human reader. This could be a node-name, BlockBackend name, qdev ID or
894 * QOM path of the device owning the BlockBackend, job type and ID etc. The
895 * caller is responsible for freeing the memory.
897 char *(*get_parent_desc
)(BdrvChild
*child
);
900 * Notifies the parent that the child has been activated/inactivated (e.g.
901 * when migration is completing) and it can start/stop requesting
902 * permissions and doing I/O on it.
904 void (*activate
)(BdrvChild
*child
, Error
**errp
);
905 int (*inactivate
)(BdrvChild
*child
);
907 void GRAPH_WRLOCK_PTR (*attach
)(BdrvChild
*child
);
908 void GRAPH_WRLOCK_PTR (*detach
)(BdrvChild
*child
);
911 * Notifies the parent that the filename of its child has changed (e.g.
912 * because the direct child was removed from the backing chain), so that it
913 * can update its reference.
915 int (*update_filename
)(BdrvChild
*child
, BlockDriverState
*new_base
,
916 const char *filename
, Error
**errp
);
918 bool (*change_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
,
919 GHashTable
*visited
, Transaction
*tran
,
923 * I/O API functions. These functions are thread-safe.
925 * See include/block/block-io.h for more information about
929 void (*resize
)(BdrvChild
*child
);
932 * Returns a name that is supposedly more useful for human users than the
933 * node name for identifying the node in question (in particular, a BB
934 * name), or NULL if the parent can't provide a better name.
936 const char *(*get_name
)(BdrvChild
*child
);
938 AioContext
*(*get_parent_aio_context
)(BdrvChild
*child
);
941 * If this pair of functions is implemented, the parent doesn't issue new
942 * requests after returning from .drained_begin() until .drained_end() is
945 * These functions must not change the graph (and therefore also must not
946 * call aio_poll(), which could change the graph indirectly).
948 * Note that this can be nested. If drained_begin() was called twice, new
949 * I/O is allowed only after drained_end() was called twice, too.
951 void (*drained_begin
)(BdrvChild
*child
);
952 void (*drained_end
)(BdrvChild
*child
);
955 * Returns whether the parent has pending requests for the child. This
956 * callback is polled after .drained_begin() has been called until all
957 * activity on the child has stopped.
959 bool (*drained_poll
)(BdrvChild
*child
);
962 extern const BdrvChildClass child_of_bds
;
965 BlockDriverState
*bs
;
967 const BdrvChildClass
*klass
;
972 * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
977 * Permissions that can still be granted to other users of @bs while this
978 * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
980 uint64_t shared_perm
;
983 * This link is frozen: the child can neither be replaced nor
984 * detached from the parent.
989 * True if the parent of this child has been drained by this BdrvChild
990 * (through klass->drained_*).
992 * It is generally true if bs->quiesce_counter > 0. It may differ while the
993 * child is entering or leaving a drained section.
995 bool quiesced_parent
;
997 QLIST_ENTRY(BdrvChild
) next
;
998 QLIST_ENTRY(BdrvChild
) next_parent
;
1002 * Allows bdrv_co_block_status() to cache one data region for a
1005 * @valid: Whether the cache is valid (should be accessed with atomic
1006 * functions so this can be reset by RCU readers)
1007 * @data_start: Offset where we know (or strongly assume) is data
1008 * @data_end: Offset where the data region ends (which is not necessarily
1009 * the start of a zeroed region)
1011 typedef struct BdrvBlockStatusCache
{
1012 struct rcu_head rcu
;
1017 } BdrvBlockStatusCache
;
1019 struct BlockDriverState
{
1021 * Protected by big QEMU lock or read-only after opening. No special
1022 * locking needed during I/O...
1024 int open_flags
; /* flags used to open the file, re-used for re-open */
1025 bool encrypted
; /* if true, the media is encrypted */
1026 bool sg
; /* if true, the device is a /dev/sg* */
1027 bool probed
; /* if true, format was probed rather than specified */
1028 bool force_share
; /* if true, always allow all shared permissions */
1029 bool implicit
; /* if true, this filter node was automatically inserted */
1031 BlockDriver
*drv
; /* NULL means no media */
1034 AioContext
*aio_context
; /* event loop used for fd handlers, timers, etc */
1036 * long-running tasks intended to always use the same AioContext as this
1037 * BDS may register themselves in this list to be notified of changes
1038 * regarding this BDS's context
1040 QLIST_HEAD(, BdrvAioNotifier
) aio_notifiers
;
1041 bool walking_aio_notifiers
; /* to make removal during iteration safe */
1043 char filename
[PATH_MAX
];
1045 * If not empty, this image is a diff in relation to backing_file.
1046 * Note that this is the name given in the image header and
1047 * therefore may or may not be equal to .backing->bs->filename.
1048 * If this field contains a relative path, it is to be resolved
1049 * relatively to the overlay's location.
1051 char backing_file
[PATH_MAX
];
1053 * The backing filename indicated by the image header. Contrary
1054 * to backing_file, if we ever open this file, auto_backing_file
1055 * is replaced by the resulting BDS's filename (i.e. after a
1056 * bdrv_refresh_filename() run).
1058 char auto_backing_file
[PATH_MAX
];
1059 char backing_format
[16]; /* if non-zero and backing_file exists */
1061 QDict
*full_open_options
;
1062 char exact_filename
[PATH_MAX
];
1068 * Flags honored during pread
1070 BdrvRequestFlags supported_read_flags
;
1072 * Flags honored during pwrite (so far: BDRV_REQ_FUA,
1073 * BDRV_REQ_WRITE_UNCHANGED).
1074 * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
1075 * writes will be issued as normal writes without the flag set.
1076 * This is important to note for drivers that do not explicitly
1077 * request a WRITE permission for their children and instead take
1078 * the same permissions as their parent did (this is commonly what
1079 * block filters do). Such drivers have to be aware that the
1080 * parent may have taken a WRITE_UNCHANGED permission only and is
1081 * issuing such requests. Drivers either must make sure that
1082 * these requests do not result in plain WRITE accesses (usually
1083 * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1084 * every incoming write request as-is, including potentially that
1085 * flag), or they have to explicitly take the WRITE permission for
1088 BdrvRequestFlags supported_write_flags
;
1090 * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1091 * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1093 BdrvRequestFlags supported_zero_flags
;
1095 * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1097 * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1098 * that any added space reads as all zeros. If this can't be guaranteed,
1099 * the operation must fail.
1101 BdrvRequestFlags supported_truncate_flags
;
1103 /* the following member gives a name to every node on the bs graph. */
1105 /* element of the list of named nodes building the graph */
1106 QTAILQ_ENTRY(BlockDriverState
) node_list
;
1107 /* element of the list of all BlockDriverStates (all_bdrv_states) */
1108 QTAILQ_ENTRY(BlockDriverState
) bs_list
;
1109 /* element of the list of monitor-owned BDS */
1110 QTAILQ_ENTRY(BlockDriverState
) monitor_list
;
1113 /* operation blockers. Protected by BQL. */
1114 QLIST_HEAD(, BdrvOpBlocker
) op_blockers
[BLOCK_OP_TYPE_MAX
];
1117 * The node that this node inherited default options from (and a reopen on
1118 * which can affect this node by changing these defaults). This is always a
1119 * parent node of this node.
1121 BlockDriverState
*inherits_from
;
1124 * @backing and @file are some of @children or NULL. All these three fields
1125 * (@file, @backing and @children) are modified only in
1126 * bdrv_child_cb_attach() and bdrv_child_cb_detach().
1128 * See also comment in include/block/block.h, to learn how backing and file
1129 * are connected with BdrvChildRole.
1131 QLIST_HEAD(, BdrvChild
) children
;
1135 QLIST_HEAD(, BdrvChild
) parents
;
1138 QDict
*explicit_options
;
1139 BlockdevDetectZeroesOptions detect_zeroes
;
1141 /* The error object in use for blocking operations on backing_hd */
1142 Error
*backing_blocker
;
1144 /* Protected by AioContext lock */
1147 * If we are reading a disk image, give its size in sectors.
1148 * Generally read-only; it is written to by load_snapshot and
1149 * save_snaphost, but the block layer is quiescent during those.
1151 int64_t total_sectors
;
1153 /* threshold limit for writes, in bytes. "High water mark". */
1154 uint64_t write_threshold_offset
;
1157 * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1158 * Reading from the list can be done with either the BQL or the
1159 * dirty_bitmap_mutex. Modifying a bitmap only requires
1160 * dirty_bitmap_mutex.
1162 QemuMutex dirty_bitmap_mutex
;
1163 QLIST_HEAD(, BdrvDirtyBitmap
) dirty_bitmaps
;
1165 /* Offset after the highest byte written to */
1166 Stat64 wr_highest_offset
;
1169 * If true, copy read backing sectors into image. Can be >1 if more
1170 * than one client has requested copy-on-read. Accessed with atomic
1176 * number of in-flight requests; overall and serialising.
1177 * Accessed with atomic ops.
1179 unsigned int in_flight
;
1180 unsigned int serialising_in_flight
;
1183 * counter for nested bdrv_io_plug.
1184 * Accessed with atomic ops.
1186 unsigned io_plugged
;
1188 /* do we need to tell the quest if we have a volatile write cache? */
1189 int enable_write_cache
;
1191 /* Accessed with atomic ops. */
1192 int quiesce_counter
;
1194 unsigned int write_gen
; /* Current data generation */
1196 /* Protected by reqs_lock. */
1198 QLIST_HEAD(, BdrvTrackedRequest
) tracked_requests
;
1199 CoQueue flush_queue
; /* Serializing flush queue */
1200 bool active_flush_req
; /* Flush request in flight? */
1202 /* Only read/written by whoever has set active_flush_req to true. */
1203 unsigned int flushed_gen
; /* Flushed write generation */
1205 /* BdrvChild links to this node may never be frozen */
1208 /* Lock for block-status cache RCU writers */
1209 CoMutex bsc_modify_lock
;
1210 /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1211 BdrvBlockStatusCache
*block_status_cache
;
1214 struct BlockBackendRootState
{
1216 BlockdevDetectZeroesOptions detect_zeroes
;
1219 typedef enum BlockMirrorBackingMode
{
1221 * Reuse the existing backing chain from the source for the target.
1222 * - sync=full: Set backing BDS to NULL.
1223 * - sync=top: Use source's backing BDS.
1224 * - sync=none: Use source as the backing BDS.
1226 MIRROR_SOURCE_BACKING_CHAIN
,
1228 /* Open the target's backing chain completely anew */
1229 MIRROR_OPEN_BACKING_CHAIN
,
1231 /* Do not change the target's backing BDS after job completion */
1232 MIRROR_LEAVE_BACKING_CHAIN
,
1233 } BlockMirrorBackingMode
;
1237 * Essential block drivers which must always be statically linked into qemu, and
1238 * which therefore can be accessed without using bdrv_find_format()
1240 extern BlockDriver bdrv_file
;
1241 extern BlockDriver bdrv_raw
;
1242 extern BlockDriver bdrv_qcow2
;
1244 extern unsigned int bdrv_drain_all_count
;
1245 extern QemuOptsList bdrv_create_opts_simple
;
1248 * Common functions that are neither I/O nor Global State.
1250 * See include/block/block-commmon.h for more information about
1254 static inline BlockDriverState
*child_bs(BdrvChild
*child
)
1256 return child
? child
->bs
: NULL
;
1259 int bdrv_check_request(int64_t offset
, int64_t bytes
, Error
**errp
);
1260 char *create_tmp_file(Error
**errp
);
1261 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
1265 int bdrv_check_qiov_request(int64_t offset
, int64_t bytes
,
1266 QEMUIOVector
*qiov
, size_t qiov_offset
,
1270 int is_windows_drive(const char *filename
);
1273 #endif /* BLOCK_INT_COMMON_H */