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/accounting.h"
28 #include "block/block.h"
29 #include "block/aio-wait.h"
30 #include "qemu/queue.h"
31 #include "qemu/coroutine.h"
32 #include "qemu/stats64.h"
33 #include "qemu/timer.h"
34 #include "qemu/hbitmap.h"
35 #include "block/snapshot.h"
36 #include "qemu/throttle.h"
39 #define BLOCK_FLAG_LAZY_REFCOUNTS 8
41 #define BLOCK_OPT_SIZE "size"
42 #define BLOCK_OPT_ENCRYPT "encryption"
43 #define BLOCK_OPT_ENCRYPT_FORMAT "encrypt.format"
44 #define BLOCK_OPT_COMPAT6 "compat6"
45 #define BLOCK_OPT_HWVERSION "hwversion"
46 #define BLOCK_OPT_BACKING_FILE "backing_file"
47 #define BLOCK_OPT_BACKING_FMT "backing_fmt"
48 #define BLOCK_OPT_CLUSTER_SIZE "cluster_size"
49 #define BLOCK_OPT_TABLE_SIZE "table_size"
50 #define BLOCK_OPT_PREALLOC "preallocation"
51 #define BLOCK_OPT_SUBFMT "subformat"
52 #define BLOCK_OPT_COMPAT_LEVEL "compat"
53 #define BLOCK_OPT_LAZY_REFCOUNTS "lazy_refcounts"
54 #define BLOCK_OPT_ADAPTER_TYPE "adapter_type"
55 #define BLOCK_OPT_REDUNDANCY "redundancy"
56 #define BLOCK_OPT_NOCOW "nocow"
57 #define BLOCK_OPT_EXTENT_SIZE_HINT "extent_size_hint"
58 #define BLOCK_OPT_OBJECT_SIZE "object_size"
59 #define BLOCK_OPT_REFCOUNT_BITS "refcount_bits"
60 #define BLOCK_OPT_DATA_FILE "data_file"
61 #define BLOCK_OPT_DATA_FILE_RAW "data_file_raw"
62 #define BLOCK_OPT_COMPRESSION_TYPE "compression_type"
63 #define BLOCK_OPT_EXTL2 "extended_l2"
65 #define BLOCK_PROBE_BUF_SIZE 512
67 enum BdrvTrackedRequestType
{
71 BDRV_TRACKED_TRUNCATE
,
75 * That is not quite good that BdrvTrackedRequest structure is public,
76 * as block/io.c is very careful about incoming offset/bytes being
77 * correct. Be sure to assert bdrv_check_request() succeeded after any
78 * modification of BdrvTrackedRequest object out of block/io.c
80 typedef struct BdrvTrackedRequest
{
84 enum BdrvTrackedRequestType type
;
87 int64_t overlap_offset
;
88 int64_t overlap_bytes
;
90 QLIST_ENTRY(BdrvTrackedRequest
) list
;
91 Coroutine
*co
; /* owner, used for deadlock detection */
92 CoQueue wait_queue
; /* coroutines blocked on this request */
94 struct BdrvTrackedRequest
*waiting_for
;
100 * These fields are initialized when this object is created,
101 * and are never changed afterwards.
104 const char *format_name
;
108 * Set to true if the BlockDriver is a block filter. Block filters pass
109 * certain callbacks that refer to data (see block.c) to their bs->file
110 * or bs->backing (whichever one exists) if the driver doesn't implement
111 * them. Drivers that do not wish to forward must implement them and return
113 * Note that filters are not allowed to modify data.
115 * Filters generally cannot have more than a single filtered child,
116 * because the data they present must at all times be the same as
117 * that on their filtered child. That would be impossible to
118 * achieve for multiple filtered children.
119 * (And this filtered child must then be bs->file or bs->backing.)
123 * Only make sense for filter drivers, for others must be false.
124 * If true, filtered child is bs->backing. Otherwise it's bs->file.
125 * Two internal filters use bs->backing as filtered child and has this
126 * field set to true: mirror_top and commit_top. There also two such test
127 * filters in tests/unit/test-bdrv-graph-mod.c.
129 * Never create any more such filters!
131 * TODO: imagine how to deprecate this behavior and make all filters work
132 * similarly using bs->file as filtered child.
134 bool filtered_child_is_backing
;
137 * Set to true if the BlockDriver is a format driver. Format nodes
138 * generally do not expect their children to be other format nodes
139 * (except for backing files), and so format probing is disabled
145 * Drivers not implementing bdrv_parse_filename nor bdrv_open should have
146 * this field set to true, except ones that are defined only by their
148 * An example of the last type will be the quorum block driver.
150 bool bdrv_needs_filename
;
153 * Set if a driver can support backing files. This also implies the
154 * following semantics:
156 * - Return status 0 of .bdrv_co_block_status means that corresponding
157 * blocks are not allocated in this layer of backing-chain
158 * - For such (unallocated) blocks, read will:
159 * - fill buffer with zeros if there is no backing file
160 * - read from the backing file otherwise, where the block layer
161 * takes care of reading zeros beyond EOF if backing file is short
163 bool supports_backing
;
165 bool has_variable_length
;
168 * Drivers setting this field must be able to work with just a plain
169 * filename with '<protocol_name>:' as a prefix, and no other options.
170 * Options may be extracted from the filename by implementing
171 * bdrv_parse_filename.
173 const char *protocol_name
;
175 /* List of options for creating images, terminated by name == NULL */
176 QemuOptsList
*create_opts
;
178 /* List of options for image amend */
179 QemuOptsList
*amend_opts
;
182 * If this driver supports reopening images this contains a
183 * NULL-terminated list of the runtime options that can be
184 * modified. If an option in this list is unspecified during
185 * reopen then it _must_ be reset to its default value or return
188 const char *const *mutable_opts
;
191 * Pointer to a NULL-terminated array of names of strong options
192 * that can be specified for bdrv_open(). A strong option is one
193 * that changes the data of a BDS.
194 * If this pointer is NULL, the array is considered empty.
195 * "filename" and "driver" are always considered strong.
197 const char *const *strong_runtime_opts
;
201 * Global state (GS) API. These functions run under the BQL.
203 * See include/block/block-global-state.h for more information about
208 * This function is invoked under BQL before .bdrv_co_amend()
209 * (which in contrast does not necessarily run under the BQL)
210 * to allow driver-specific initialization code that requires
211 * the BQL, like setting up specific permission flags.
213 int (*bdrv_amend_pre_run
)(BlockDriverState
*bs
, Error
**errp
);
215 * This function is invoked under BQL after .bdrv_co_amend()
216 * to allow cleaning up what was done in .bdrv_amend_pre_run().
218 void (*bdrv_amend_clean
)(BlockDriverState
*bs
);
221 * Return true if @to_replace can be replaced by a BDS with the
222 * same data as @bs without it affecting @bs's behavior (that is,
223 * without it being visible to @bs's parents).
225 bool (*bdrv_recurse_can_replace
)(BlockDriverState
*bs
,
226 BlockDriverState
*to_replace
);
228 int (*bdrv_probe_device
)(const char *filename
);
231 * Any driver implementing this callback is expected to be able to handle
232 * NULL file names in its .bdrv_open() implementation.
234 void (*bdrv_parse_filename
)(const char *filename
, QDict
*options
,
237 /* For handling image reopen for split or non-split files. */
238 int (*bdrv_reopen_prepare
)(BDRVReopenState
*reopen_state
,
239 BlockReopenQueue
*queue
, Error
**errp
);
240 void (*bdrv_reopen_commit
)(BDRVReopenState
*reopen_state
);
241 void (*bdrv_reopen_commit_post
)(BDRVReopenState
*reopen_state
);
242 void (*bdrv_reopen_abort
)(BDRVReopenState
*reopen_state
);
243 void (*bdrv_join_options
)(QDict
*options
, QDict
*old_options
);
245 int (*bdrv_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
248 /* Protocol drivers should implement this instead of bdrv_open */
249 int (*bdrv_file_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
251 void (*bdrv_close
)(BlockDriverState
*bs
);
253 int coroutine_fn (*bdrv_co_create
)(BlockdevCreateOptions
*opts
,
255 int coroutine_fn (*bdrv_co_create_opts
)(BlockDriver
*drv
,
256 const char *filename
,
260 int (*bdrv_amend_options
)(BlockDriverState
*bs
,
262 BlockDriverAmendStatusCB
*status_cb
,
267 int (*bdrv_make_empty
)(BlockDriverState
*bs
);
270 * Refreshes the bs->exact_filename field. If that is impossible,
271 * bs->exact_filename has to be left empty.
273 void (*bdrv_refresh_filename
)(BlockDriverState
*bs
);
276 * Gathers the open options for all children into @target.
277 * A simple format driver (without backing file support) might
278 * implement this function like this:
280 * QINCREF(bs->file->bs->full_open_options);
281 * qdict_put(target, "file", bs->file->bs->full_open_options);
283 * If not specified, the generic implementation will simply put
284 * all children's options under their respective name.
286 * @backing_overridden is true when bs->backing seems not to be
287 * the child that would result from opening bs->backing_file.
288 * Therefore, if it is true, the backing child's options should be
289 * gathered; otherwise, there is no need since the backing child
290 * is the one implied by the image header.
292 * Note that ideally this function would not be needed. Every
293 * block driver which implements it is probably doing something
294 * shady regarding its runtime option structure.
296 void (*bdrv_gather_child_options
)(BlockDriverState
*bs
, QDict
*target
,
297 bool backing_overridden
);
300 * Returns an allocated string which is the directory name of this BDS: It
301 * will be used to make relative filenames absolute by prepending this
302 * function's return value to them.
304 char *(*bdrv_dirname
)(BlockDriverState
*bs
, Error
**errp
);
307 * This informs the driver that we are no longer interested in the result
308 * of in-flight requests, so don't waste the time if possible.
310 * One example usage is to avoid waiting for an nbd target node reconnect
311 * timeout during job-cancel with force=true.
313 void (*bdrv_cancel_in_flight
)(BlockDriverState
*bs
);
315 int (*bdrv_inactivate
)(BlockDriverState
*bs
);
317 int (*bdrv_snapshot_create
)(BlockDriverState
*bs
,
318 QEMUSnapshotInfo
*sn_info
);
319 int (*bdrv_snapshot_goto
)(BlockDriverState
*bs
,
320 const char *snapshot_id
);
321 int (*bdrv_snapshot_delete
)(BlockDriverState
*bs
,
322 const char *snapshot_id
,
325 int (*bdrv_snapshot_list
)(BlockDriverState
*bs
,
326 QEMUSnapshotInfo
**psn_info
);
327 int (*bdrv_snapshot_load_tmp
)(BlockDriverState
*bs
,
328 const char *snapshot_id
,
332 int (*bdrv_change_backing_file
)(BlockDriverState
*bs
,
333 const char *backing_file
, const char *backing_fmt
);
335 /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
336 int (*bdrv_debug_breakpoint
)(BlockDriverState
*bs
, const char *event
,
338 int (*bdrv_debug_remove_breakpoint
)(BlockDriverState
*bs
,
340 int (*bdrv_debug_resume
)(BlockDriverState
*bs
, const char *tag
);
341 bool (*bdrv_debug_is_suspended
)(BlockDriverState
*bs
, const char *tag
);
343 void (*bdrv_refresh_limits
)(BlockDriverState
*bs
, Error
**errp
);
346 * Returns 1 if newly created images are guaranteed to contain only
347 * zeros, 0 otherwise.
349 int (*bdrv_has_zero_init
)(BlockDriverState
*bs
);
352 * Remove fd handlers, timers, and other event loop callbacks so the event
353 * loop is no longer in use. Called with no in-flight requests and in
354 * depth-first traversal order with parents before child nodes.
356 void (*bdrv_detach_aio_context
)(BlockDriverState
*bs
);
359 * Add fd handlers, timers, and other event loop callbacks so I/O requests
360 * can be processed again. Called with no in-flight requests and in
361 * depth-first traversal order with child nodes before parent nodes.
363 void (*bdrv_attach_aio_context
)(BlockDriverState
*bs
,
364 AioContext
*new_context
);
367 * Try to get @bs's logical and physical block size.
368 * On success, store them in @bsz and return zero.
369 * On failure, return negative errno.
371 int (*bdrv_probe_blocksizes
)(BlockDriverState
*bs
, BlockSizes
*bsz
);
373 * Try to get @bs's geometry (cyls, heads, sectors)
374 * On success, store them in @geo and return 0.
375 * On failure return -errno.
376 * Only drivers that want to override guest geometry implement this
377 * callback; see hd_geometry_guess().
379 int (*bdrv_probe_geometry
)(BlockDriverState
*bs
, HDGeometry
*geo
);
381 void (*bdrv_add_child
)(BlockDriverState
*parent
, BlockDriverState
*child
,
383 void (*bdrv_del_child
)(BlockDriverState
*parent
, BdrvChild
*child
,
387 * Informs the block driver that a permission change is intended. The
388 * driver checks whether the change is permissible and may take other
389 * preparations for the change (e.g. get file system locks). This operation
390 * is always followed either by a call to either .bdrv_set_perm or
391 * .bdrv_abort_perm_update.
393 * Checks whether the requested set of cumulative permissions in @perm
394 * can be granted for accessing @bs and whether no other users are using
395 * permissions other than those given in @shared (both arguments take
396 * BLK_PERM_* bitmasks).
398 * If both conditions are met, 0 is returned. Otherwise, -errno is returned
399 * and errp is set to an error describing the conflict.
401 int (*bdrv_check_perm
)(BlockDriverState
*bs
, uint64_t perm
,
402 uint64_t shared
, Error
**errp
);
405 * Called to inform the driver that the set of cumulative set of used
406 * permissions for @bs has changed to @perm, and the set of sharable
407 * permission to @shared. The driver can use this to propagate changes to
408 * its children (i.e. request permissions only if a parent actually needs
411 * This function is only invoked after bdrv_check_perm(), so block drivers
412 * may rely on preparations made in their .bdrv_check_perm implementation.
414 void (*bdrv_set_perm
)(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
);
417 * Called to inform the driver that after a previous bdrv_check_perm()
418 * call, the permission update is not performed and any preparations made
419 * for it (e.g. taken file locks) need to be undone.
421 * This function can be called even for nodes that never saw a
422 * bdrv_check_perm() call. It is a no-op then.
424 void (*bdrv_abort_perm_update
)(BlockDriverState
*bs
);
427 * Returns in @nperm and @nshared the permissions that the driver for @bs
428 * needs on its child @c, based on the cumulative permissions requested by
429 * the parents in @parent_perm and @parent_shared.
431 * If @c is NULL, return the permissions for attaching a new child for the
432 * given @child_class and @role.
434 * If @reopen_queue is non-NULL, don't return the currently needed
435 * permissions, but those that will be needed after applying the
438 void (*bdrv_child_perm
)(BlockDriverState
*bs
, BdrvChild
*c
,
440 BlockReopenQueue
*reopen_queue
,
441 uint64_t parent_perm
, uint64_t parent_shared
,
442 uint64_t *nperm
, uint64_t *nshared
);
445 * Register/unregister a buffer for I/O. For example, when the driver is
446 * interested to know the memory areas that will later be used in iovs, so
447 * that it can do IOMMU mapping with VFIO etc., in order to get better
448 * performance. In the case of VFIO drivers, this callback is used to do
449 * DMA mapping for hot buffers.
451 * Returns: true on success, false on failure
453 bool (*bdrv_register_buf
)(BlockDriverState
*bs
, void *host
, size_t size
,
455 void (*bdrv_unregister_buf
)(BlockDriverState
*bs
, void *host
, size_t size
);
458 * This field is modified only under the BQL, and is part of
461 QLIST_ENTRY(BlockDriver
) list
;
464 * I/O API functions. These functions are thread-safe.
466 * See include/block/block-io.h for more information about
470 int (*bdrv_probe
)(const uint8_t *buf
, int buf_size
, const char *filename
);
472 int coroutine_fn (*bdrv_co_amend
)(BlockDriverState
*bs
,
473 BlockdevAmendOptions
*opts
,
478 BlockAIOCB
*(*bdrv_aio_preadv
)(BlockDriverState
*bs
,
479 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
480 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
481 BlockAIOCB
*(*bdrv_aio_pwritev
)(BlockDriverState
*bs
,
482 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
483 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
484 BlockAIOCB
*(*bdrv_aio_flush
)(BlockDriverState
*bs
,
485 BlockCompletionFunc
*cb
, void *opaque
);
486 BlockAIOCB
*(*bdrv_aio_pdiscard
)(BlockDriverState
*bs
,
487 int64_t offset
, int bytes
,
488 BlockCompletionFunc
*cb
, void *opaque
);
490 int coroutine_fn (*bdrv_co_readv
)(BlockDriverState
*bs
,
491 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
);
494 * @offset: position in bytes to read at
495 * @bytes: number of bytes to read
496 * @qiov: the buffers to fill with read data
497 * @flags: currently unused, always 0
499 * @offset and @bytes will be a multiple of 'request_alignment',
500 * but the length of individual @qiov elements does not have to
503 * @bytes will always equal the total size of @qiov, and will be
504 * no larger than 'max_transfer'.
506 * The buffer in @qiov may point directly to guest memory.
508 int coroutine_fn (*bdrv_co_preadv
)(BlockDriverState
*bs
,
509 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
510 BdrvRequestFlags flags
);
512 int coroutine_fn (*bdrv_co_preadv_part
)(BlockDriverState
*bs
,
513 int64_t offset
, int64_t bytes
,
514 QEMUIOVector
*qiov
, size_t qiov_offset
,
515 BdrvRequestFlags flags
);
517 int coroutine_fn (*bdrv_co_writev
)(BlockDriverState
*bs
,
518 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
521 * @offset: position in bytes to write at
522 * @bytes: number of bytes to write
523 * @qiov: the buffers containing data to write
524 * @flags: zero or more bits allowed by 'supported_write_flags'
526 * @offset and @bytes will be a multiple of 'request_alignment',
527 * but the length of individual @qiov elements does not have to
530 * @bytes will always equal the total size of @qiov, and will be
531 * no larger than 'max_transfer'.
533 * The buffer in @qiov may point directly to guest memory.
535 int coroutine_fn (*bdrv_co_pwritev
)(BlockDriverState
*bs
,
536 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
537 BdrvRequestFlags flags
);
538 int coroutine_fn (*bdrv_co_pwritev_part
)(BlockDriverState
*bs
,
539 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
540 BdrvRequestFlags flags
);
543 * Efficiently zero a region of the disk image. Typically an image format
544 * would use a compact metadata representation to implement this. This
545 * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
546 * will be called instead.
548 int coroutine_fn (*bdrv_co_pwrite_zeroes
)(BlockDriverState
*bs
,
549 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
);
550 int coroutine_fn (*bdrv_co_pdiscard
)(BlockDriverState
*bs
,
551 int64_t offset
, int64_t bytes
);
554 * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
555 * and invoke bdrv_co_copy_range_from(child, ...), or invoke
556 * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
558 * See the comment of bdrv_co_copy_range for the parameter and return value
561 int coroutine_fn (*bdrv_co_copy_range_from
)(BlockDriverState
*bs
,
567 BdrvRequestFlags read_flags
,
568 BdrvRequestFlags write_flags
);
571 * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
572 * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
573 * operation if @bs is the leaf and @src has the same BlockDriver. Return
574 * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
576 * See the comment of bdrv_co_copy_range for the parameter and return value
579 int coroutine_fn (*bdrv_co_copy_range_to
)(BlockDriverState
*bs
,
585 BdrvRequestFlags read_flags
,
586 BdrvRequestFlags write_flags
);
589 * Building block for bdrv_block_status[_above] and
590 * bdrv_is_allocated[_above]. The driver should answer only
591 * according to the current layer, and should only need to set
592 * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
593 * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
594 * layer, the result should be 0 (and not BDRV_BLOCK_ZERO). See
595 * block.h for the overall meaning of the bits. As a hint, the
596 * flag want_zero is true if the caller cares more about precise
597 * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
598 * overall allocation (favor larger *pnum, perhaps by reporting
599 * _DATA instead of _ZERO). The block layer guarantees input
600 * clamped to bdrv_getlength() and aligned to request_alignment,
601 * as well as non-NULL pnum, map, and file; in turn, the driver
602 * must return an error or set pnum to an aligned non-zero value.
604 * Note that @bytes is just a hint on how big of a region the
605 * caller wants to inspect. It is not a limit on *pnum.
606 * Implementations are free to return larger values of *pnum if
607 * doing so does not incur a performance penalty.
609 * block/io.c's bdrv_co_block_status() will utilize an unclamped
610 * *pnum value for the block-status cache on protocol nodes, prior
611 * to clamping *pnum for return to its caller.
613 int coroutine_fn (*bdrv_co_block_status
)(BlockDriverState
*bs
,
614 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
615 int64_t *map
, BlockDriverState
**file
);
618 * Snapshot-access API.
620 * Block-driver may provide snapshot-access API: special functions to access
621 * some internal "snapshot". The functions are similar with normal
622 * read/block_status/discard handler, but don't have any specific handling
623 * in generic block-layer: no serializing, no alignment, no tracked
624 * requests. So, block-driver that realizes these APIs is fully responsible
625 * for synchronization between snapshot-access API and normal IO requests.
627 * TODO: To be able to support qcow2's internal snapshots, this API will
628 * need to be extended to:
629 * - be able to select a specific snapshot
630 * - receive the snapshot's actual length (which may differ from bs's
633 int coroutine_fn (*bdrv_co_preadv_snapshot
)(BlockDriverState
*bs
,
634 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
);
635 int coroutine_fn (*bdrv_co_snapshot_block_status
)(BlockDriverState
*bs
,
636 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
637 int64_t *map
, BlockDriverState
**file
);
638 int coroutine_fn (*bdrv_co_pdiscard_snapshot
)(BlockDriverState
*bs
,
639 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 (*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 (*bdrv_co_flush_to_disk
)(BlockDriverState
*bs
);
665 * Flushes all internal caches to the OS. The data may still sit in a
666 * writeback cache of the host OS, but it will survive a crash of the qemu
669 int coroutine_fn (*bdrv_co_flush_to_os
)(BlockDriverState
*bs
);
672 * Truncate @bs to @offset bytes using the given @prealloc mode
673 * when growing. Modes other than PREALLOC_MODE_OFF should be
674 * rejected when shrinking @bs.
676 * If @exact is true, @bs must be resized to exactly @offset.
677 * Otherwise, it is sufficient for @bs (if it is a host block
678 * device and thus there is no way to resize it) to be at least
679 * @offset bytes in length.
681 * If @exact is true and this function fails but would succeed
682 * with @exact = false, it should return -ENOTSUP.
684 int coroutine_fn (*bdrv_co_truncate
)(BlockDriverState
*bs
, int64_t offset
,
685 bool exact
, PreallocMode prealloc
,
686 BdrvRequestFlags flags
, Error
**errp
);
687 int64_t (*bdrv_getlength
)(BlockDriverState
*bs
);
688 int64_t (*bdrv_get_allocated_file_size
)(BlockDriverState
*bs
);
689 BlockMeasureInfo
*(*bdrv_measure
)(QemuOpts
*opts
, BlockDriverState
*in_bs
,
692 int coroutine_fn (*bdrv_co_pwritev_compressed
)(BlockDriverState
*bs
,
693 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
);
694 int coroutine_fn (*bdrv_co_pwritev_compressed_part
)(BlockDriverState
*bs
,
695 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
698 int (*bdrv_get_info
)(BlockDriverState
*bs
, BlockDriverInfo
*bdi
);
700 ImageInfoSpecific
*(*bdrv_get_specific_info
)(BlockDriverState
*bs
,
702 BlockStatsSpecific
*(*bdrv_get_specific_stats
)(BlockDriverState
*bs
);
704 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_save_vmstate
)(
705 BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
);
707 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_load_vmstate
)(
708 BlockDriverState
*bs
, QEMUIOVector
*qiov
, int64_t pos
);
710 /* removable device specific */
711 bool (*bdrv_is_inserted
)(BlockDriverState
*bs
);
712 void (*bdrv_eject
)(BlockDriverState
*bs
, bool eject_flag
);
713 void (*bdrv_lock_medium
)(BlockDriverState
*bs
, bool locked
);
715 /* to control generic scsi devices */
716 BlockAIOCB
*(*bdrv_aio_ioctl
)(BlockDriverState
*bs
,
717 unsigned long int req
, void *buf
,
718 BlockCompletionFunc
*cb
, void *opaque
);
719 int coroutine_fn (*bdrv_co_ioctl
)(BlockDriverState
*bs
,
720 unsigned long int req
, void *buf
);
723 * Returns 0 for completed check, -errno for internal errors.
724 * The check results are stored in result.
726 int coroutine_fn
GRAPH_RDLOCK_PTR (*bdrv_co_check
)(
727 BlockDriverState
*bs
, BdrvCheckResult
*result
, BdrvCheckMode fix
);
729 void (*bdrv_debug_event
)(BlockDriverState
*bs
, BlkdebugEvent event
);
731 /* io queue for linux-aio */
732 void (*bdrv_io_plug
)(BlockDriverState
*bs
);
733 void (*bdrv_io_unplug
)(BlockDriverState
*bs
);
736 * bdrv_drain_begin is called if implemented in the beginning of a
737 * drain operation to drain and stop any internal sources of requests in
739 * bdrv_drain_end is called if implemented at the end of the drain.
741 * They should be used by the driver to e.g. manage scheduled I/O
742 * requests, or toggle an internal state. After the end of the drain new
743 * requests will continue normally.
745 * Implementations of both functions must not call aio_poll().
747 void (*bdrv_drain_begin
)(BlockDriverState
*bs
);
748 void (*bdrv_drain_end
)(BlockDriverState
*bs
);
750 bool (*bdrv_supports_persistent_dirty_bitmap
)(BlockDriverState
*bs
);
751 bool coroutine_fn (*bdrv_co_can_store_new_dirty_bitmap
)(
752 BlockDriverState
*bs
, const char *name
, uint32_t granularity
,
754 int coroutine_fn (*bdrv_co_remove_persistent_dirty_bitmap
)(
755 BlockDriverState
*bs
, const char *name
, Error
**errp
);
758 static inline bool block_driver_can_compress(BlockDriver
*drv
)
760 return drv
->bdrv_co_pwritev_compressed
||
761 drv
->bdrv_co_pwritev_compressed_part
;
764 typedef struct BlockLimits
{
766 * Alignment requirement, in bytes, for offset/length of I/O
767 * requests. Must be a power of 2 less than INT_MAX; defaults to
768 * 1 for drivers with modern byte interfaces, and to 512
771 uint32_t request_alignment
;
774 * Maximum number of bytes that can be discarded at once. Must be multiple
775 * of pdiscard_alignment, but need not be power of 2. May be 0 if no
776 * inherent 64-bit limit.
778 int64_t max_pdiscard
;
781 * Optimal alignment for discard requests in bytes. A power of 2
782 * is best but not mandatory. Must be a multiple of
783 * bl.request_alignment, and must be less than max_pdiscard if
784 * that is set. May be 0 if bl.request_alignment is good enough
786 uint32_t pdiscard_alignment
;
789 * Maximum number of bytes that can zeroized at once. Must be multiple of
790 * pwrite_zeroes_alignment. 0 means no limit.
792 int64_t max_pwrite_zeroes
;
795 * Optimal alignment for write zeroes requests in bytes. A power
796 * of 2 is best but not mandatory. Must be a multiple of
797 * bl.request_alignment, and must be less than max_pwrite_zeroes
798 * if that is set. May be 0 if bl.request_alignment is good
801 uint32_t pwrite_zeroes_alignment
;
804 * Optimal transfer length in bytes. A power of 2 is best but not
805 * mandatory. Must be a multiple of bl.request_alignment, or 0 if
808 uint32_t opt_transfer
;
811 * Maximal transfer length in bytes. Need not be power of 2, but
812 * must be multiple of opt_transfer and bl.request_alignment, or 0
813 * for no 32-bit limit. For now, anything larger than INT_MAX is
816 uint32_t max_transfer
;
819 * Maximal hardware transfer length in bytes. Applies whenever
820 * transfers to the device bypass the kernel I/O scheduler, for
821 * example with SG_IO. If larger than max_transfer or if zero,
822 * blk_get_max_hw_transfer will fall back to max_transfer.
824 uint64_t max_hw_transfer
;
827 * Maximal number of scatter/gather elements allowed by the hardware.
828 * Applies whenever transfers to the device bypass the kernel I/O
829 * scheduler, for example with SG_IO. If larger than max_iov
830 * or if zero, blk_get_max_hw_iov will fall back to max_iov.
835 /* memory alignment, in bytes so that no bounce buffer is needed */
836 size_t min_mem_alignment
;
838 /* memory alignment, in bytes, for bounce buffer */
839 size_t opt_mem_alignment
;
841 /* maximum number of iovec elements */
845 typedef struct BdrvOpBlocker BdrvOpBlocker
;
847 typedef struct BdrvAioNotifier
{
848 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
849 void (*detach_aio_context
)(void *opaque
);
854 QLIST_ENTRY(BdrvAioNotifier
) list
;
857 struct BdrvChildClass
{
859 * If true, bdrv_replace_node() doesn't change the node this BdrvChild
865 * If true, the parent is a BlockDriverState and bdrv_next_all_states()
866 * will return it. This information is used for drain_all, where every node
867 * will be drained separately, so the drain only needs to be propagated to
873 * Global state (GS) API. These functions run under the BQL.
875 * See include/block/block-global-state.h for more information about
878 void (*inherit_options
)(BdrvChildRole role
, bool parent_is_format
,
879 int *child_flags
, QDict
*child_options
,
880 int parent_flags
, QDict
*parent_options
);
881 void (*change_media
)(BdrvChild
*child
, bool load
);
884 * Returns a malloced string that describes the parent of the child for a
885 * human reader. This could be a node-name, BlockBackend name, qdev ID or
886 * QOM path of the device owning the BlockBackend, job type and ID etc. The
887 * caller is responsible for freeing the memory.
889 char *(*get_parent_desc
)(BdrvChild
*child
);
892 * Notifies the parent that the child has been activated/inactivated (e.g.
893 * when migration is completing) and it can start/stop requesting
894 * permissions and doing I/O on it.
896 void (*activate
)(BdrvChild
*child
, Error
**errp
);
897 int (*inactivate
)(BdrvChild
*child
);
899 void GRAPH_WRLOCK_PTR (*attach
)(BdrvChild
*child
);
900 void GRAPH_WRLOCK_PTR (*detach
)(BdrvChild
*child
);
903 * Notifies the parent that the filename of its child has changed (e.g.
904 * because the direct child was removed from the backing chain), so that it
905 * can update its reference.
907 int (*update_filename
)(BdrvChild
*child
, BlockDriverState
*new_base
,
908 const char *filename
, Error
**errp
);
910 bool (*change_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
,
911 GHashTable
*visited
, Transaction
*tran
,
915 * I/O API functions. These functions are thread-safe.
917 * See include/block/block-io.h for more information about
921 void (*resize
)(BdrvChild
*child
);
924 * Returns a name that is supposedly more useful for human users than the
925 * node name for identifying the node in question (in particular, a BB
926 * name), or NULL if the parent can't provide a better name.
928 const char *(*get_name
)(BdrvChild
*child
);
930 AioContext
*(*get_parent_aio_context
)(BdrvChild
*child
);
933 * If this pair of functions is implemented, the parent doesn't issue new
934 * requests after returning from .drained_begin() until .drained_end() is
937 * These functions must not change the graph (and therefore also must not
938 * call aio_poll(), which could change the graph indirectly).
940 * Note that this can be nested. If drained_begin() was called twice, new
941 * I/O is allowed only after drained_end() was called twice, too.
943 void (*drained_begin
)(BdrvChild
*child
);
944 void (*drained_end
)(BdrvChild
*child
);
947 * Returns whether the parent has pending requests for the child. This
948 * callback is polled after .drained_begin() has been called until all
949 * activity on the child has stopped.
951 bool (*drained_poll
)(BdrvChild
*child
);
954 extern const BdrvChildClass child_of_bds
;
957 BlockDriverState
*bs
;
959 const BdrvChildClass
*klass
;
964 * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
969 * Permissions that can still be granted to other users of @bs while this
970 * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
972 uint64_t shared_perm
;
975 * This link is frozen: the child can neither be replaced nor
976 * detached from the parent.
981 * True if the parent of this child has been drained by this BdrvChild
982 * (through klass->drained_*).
984 * It is generally true if bs->quiesce_counter > 0. It may differ while the
985 * child is entering or leaving a drained section.
987 bool quiesced_parent
;
989 QLIST_ENTRY(BdrvChild
) next
;
990 QLIST_ENTRY(BdrvChild
) next_parent
;
994 * Allows bdrv_co_block_status() to cache one data region for a
997 * @valid: Whether the cache is valid (should be accessed with atomic
998 * functions so this can be reset by RCU readers)
999 * @data_start: Offset where we know (or strongly assume) is data
1000 * @data_end: Offset where the data region ends (which is not necessarily
1001 * the start of a zeroed region)
1003 typedef struct BdrvBlockStatusCache
{
1004 struct rcu_head rcu
;
1009 } BdrvBlockStatusCache
;
1011 struct BlockDriverState
{
1013 * Protected by big QEMU lock or read-only after opening. No special
1014 * locking needed during I/O...
1016 int open_flags
; /* flags used to open the file, re-used for re-open */
1017 bool encrypted
; /* if true, the media is encrypted */
1018 bool sg
; /* if true, the device is a /dev/sg* */
1019 bool probed
; /* if true, format was probed rather than specified */
1020 bool force_share
; /* if true, always allow all shared permissions */
1021 bool implicit
; /* if true, this filter node was automatically inserted */
1023 BlockDriver
*drv
; /* NULL means no media */
1026 AioContext
*aio_context
; /* event loop used for fd handlers, timers, etc */
1028 * long-running tasks intended to always use the same AioContext as this
1029 * BDS may register themselves in this list to be notified of changes
1030 * regarding this BDS's context
1032 QLIST_HEAD(, BdrvAioNotifier
) aio_notifiers
;
1033 bool walking_aio_notifiers
; /* to make removal during iteration safe */
1035 char filename
[PATH_MAX
];
1037 * If not empty, this image is a diff in relation to backing_file.
1038 * Note that this is the name given in the image header and
1039 * therefore may or may not be equal to .backing->bs->filename.
1040 * If this field contains a relative path, it is to be resolved
1041 * relatively to the overlay's location.
1043 char backing_file
[PATH_MAX
];
1045 * The backing filename indicated by the image header. Contrary
1046 * to backing_file, if we ever open this file, auto_backing_file
1047 * is replaced by the resulting BDS's filename (i.e. after a
1048 * bdrv_refresh_filename() run).
1050 char auto_backing_file
[PATH_MAX
];
1051 char backing_format
[16]; /* if non-zero and backing_file exists */
1053 QDict
*full_open_options
;
1054 char exact_filename
[PATH_MAX
];
1060 * Flags honored during pread
1062 BdrvRequestFlags supported_read_flags
;
1064 * Flags honored during pwrite (so far: BDRV_REQ_FUA,
1065 * BDRV_REQ_WRITE_UNCHANGED).
1066 * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
1067 * writes will be issued as normal writes without the flag set.
1068 * This is important to note for drivers that do not explicitly
1069 * request a WRITE permission for their children and instead take
1070 * the same permissions as their parent did (this is commonly what
1071 * block filters do). Such drivers have to be aware that the
1072 * parent may have taken a WRITE_UNCHANGED permission only and is
1073 * issuing such requests. Drivers either must make sure that
1074 * these requests do not result in plain WRITE accesses (usually
1075 * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1076 * every incoming write request as-is, including potentially that
1077 * flag), or they have to explicitly take the WRITE permission for
1080 BdrvRequestFlags supported_write_flags
;
1082 * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1083 * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1085 BdrvRequestFlags supported_zero_flags
;
1087 * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1089 * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1090 * that any added space reads as all zeros. If this can't be guaranteed,
1091 * the operation must fail.
1093 BdrvRequestFlags supported_truncate_flags
;
1095 /* the following member gives a name to every node on the bs graph. */
1097 /* element of the list of named nodes building the graph */
1098 QTAILQ_ENTRY(BlockDriverState
) node_list
;
1099 /* element of the list of all BlockDriverStates (all_bdrv_states) */
1100 QTAILQ_ENTRY(BlockDriverState
) bs_list
;
1101 /* element of the list of monitor-owned BDS */
1102 QTAILQ_ENTRY(BlockDriverState
) monitor_list
;
1105 /* operation blockers. Protected by BQL. */
1106 QLIST_HEAD(, BdrvOpBlocker
) op_blockers
[BLOCK_OP_TYPE_MAX
];
1109 * The node that this node inherited default options from (and a reopen on
1110 * which can affect this node by changing these defaults). This is always a
1111 * parent node of this node.
1113 BlockDriverState
*inherits_from
;
1116 * @backing and @file are some of @children or NULL. All these three fields
1117 * (@file, @backing and @children) are modified only in
1118 * bdrv_child_cb_attach() and bdrv_child_cb_detach().
1120 * See also comment in include/block/block.h, to learn how backing and file
1121 * are connected with BdrvChildRole.
1123 QLIST_HEAD(, BdrvChild
) children
;
1127 QLIST_HEAD(, BdrvChild
) parents
;
1130 QDict
*explicit_options
;
1131 BlockdevDetectZeroesOptions detect_zeroes
;
1133 /* The error object in use for blocking operations on backing_hd */
1134 Error
*backing_blocker
;
1136 /* Protected by AioContext lock */
1139 * If we are reading a disk image, give its size in sectors.
1140 * Generally read-only; it is written to by load_snapshot and
1141 * save_snaphost, but the block layer is quiescent during those.
1143 int64_t total_sectors
;
1145 /* threshold limit for writes, in bytes. "High water mark". */
1146 uint64_t write_threshold_offset
;
1149 * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1150 * Reading from the list can be done with either the BQL or the
1151 * dirty_bitmap_mutex. Modifying a bitmap only requires
1152 * dirty_bitmap_mutex.
1154 QemuMutex dirty_bitmap_mutex
;
1155 QLIST_HEAD(, BdrvDirtyBitmap
) dirty_bitmaps
;
1157 /* Offset after the highest byte written to */
1158 Stat64 wr_highest_offset
;
1161 * If true, copy read backing sectors into image. Can be >1 if more
1162 * than one client has requested copy-on-read. Accessed with atomic
1168 * number of in-flight requests; overall and serialising.
1169 * Accessed with atomic ops.
1171 unsigned int in_flight
;
1172 unsigned int serialising_in_flight
;
1175 * counter for nested bdrv_io_plug.
1176 * Accessed with atomic ops.
1178 unsigned io_plugged
;
1180 /* do we need to tell the quest if we have a volatile write cache? */
1181 int enable_write_cache
;
1183 /* Accessed with atomic ops. */
1184 int quiesce_counter
;
1186 unsigned int write_gen
; /* Current data generation */
1188 /* Protected by reqs_lock. */
1190 QLIST_HEAD(, BdrvTrackedRequest
) tracked_requests
;
1191 CoQueue flush_queue
; /* Serializing flush queue */
1192 bool active_flush_req
; /* Flush request in flight? */
1194 /* Only read/written by whoever has set active_flush_req to true. */
1195 unsigned int flushed_gen
; /* Flushed write generation */
1197 /* BdrvChild links to this node may never be frozen */
1200 /* Lock for block-status cache RCU writers */
1201 CoMutex bsc_modify_lock
;
1202 /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1203 BdrvBlockStatusCache
*block_status_cache
;
1206 struct BlockBackendRootState
{
1208 BlockdevDetectZeroesOptions detect_zeroes
;
1211 typedef enum BlockMirrorBackingMode
{
1213 * Reuse the existing backing chain from the source for the target.
1214 * - sync=full: Set backing BDS to NULL.
1215 * - sync=top: Use source's backing BDS.
1216 * - sync=none: Use source as the backing BDS.
1218 MIRROR_SOURCE_BACKING_CHAIN
,
1220 /* Open the target's backing chain completely anew */
1221 MIRROR_OPEN_BACKING_CHAIN
,
1223 /* Do not change the target's backing BDS after job completion */
1224 MIRROR_LEAVE_BACKING_CHAIN
,
1225 } BlockMirrorBackingMode
;
1229 * Essential block drivers which must always be statically linked into qemu, and
1230 * which therefore can be accessed without using bdrv_find_format()
1232 extern BlockDriver bdrv_file
;
1233 extern BlockDriver bdrv_raw
;
1234 extern BlockDriver bdrv_qcow2
;
1236 extern unsigned int bdrv_drain_all_count
;
1237 extern QemuOptsList bdrv_create_opts_simple
;
1240 * Common functions that are neither I/O nor Global State.
1242 * See include/block/block-commmon.h for more information about
1246 static inline BlockDriverState
*child_bs(BdrvChild
*child
)
1248 return child
? child
->bs
: NULL
;
1251 int bdrv_check_request(int64_t offset
, int64_t bytes
, Error
**errp
);
1252 char *create_tmp_file(Error
**errp
);
1253 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
1257 int bdrv_check_qiov_request(int64_t offset
, int64_t bytes
,
1258 QEMUIOVector
*qiov
, size_t qiov_offset
,
1262 int is_windows_drive(const char *filename
);
1265 #endif /* BLOCK_INT_COMMON_H */