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 * Set to true if the BlockDriver is a format driver. Format nodes
124 * generally do not expect their children to be other format nodes
125 * (except for backing files), and so format probing is disabled
131 * Drivers not implementing bdrv_parse_filename nor bdrv_open should have
132 * this field set to true, except ones that are defined only by their
134 * An example of the last type will be the quorum block driver.
136 bool bdrv_needs_filename
;
139 * Set if a driver can support backing files. This also implies the
140 * following semantics:
142 * - Return status 0 of .bdrv_co_block_status means that corresponding
143 * blocks are not allocated in this layer of backing-chain
144 * - For such (unallocated) blocks, read will:
145 * - fill buffer with zeros if there is no backing file
146 * - read from the backing file otherwise, where the block layer
147 * takes care of reading zeros beyond EOF if backing file is short
149 bool supports_backing
;
151 bool has_variable_length
;
154 * Drivers setting this field must be able to work with just a plain
155 * filename with '<protocol_name>:' as a prefix, and no other options.
156 * Options may be extracted from the filename by implementing
157 * bdrv_parse_filename.
159 const char *protocol_name
;
161 /* List of options for creating images, terminated by name == NULL */
162 QemuOptsList
*create_opts
;
164 /* List of options for image amend */
165 QemuOptsList
*amend_opts
;
168 * If this driver supports reopening images this contains a
169 * NULL-terminated list of the runtime options that can be
170 * modified. If an option in this list is unspecified during
171 * reopen then it _must_ be reset to its default value or return
174 const char *const *mutable_opts
;
177 * Pointer to a NULL-terminated array of names of strong options
178 * that can be specified for bdrv_open(). A strong option is one
179 * that changes the data of a BDS.
180 * If this pointer is NULL, the array is considered empty.
181 * "filename" and "driver" are always considered strong.
183 const char *const *strong_runtime_opts
;
187 * Global state (GS) API. These functions run under the BQL.
189 * See include/block/block-global-state.h for more information about
194 * This function is invoked under BQL before .bdrv_co_amend()
195 * (which in contrast does not necessarily run under the BQL)
196 * to allow driver-specific initialization code that requires
197 * the BQL, like setting up specific permission flags.
199 int (*bdrv_amend_pre_run
)(BlockDriverState
*bs
, Error
**errp
);
201 * This function is invoked under BQL after .bdrv_co_amend()
202 * to allow cleaning up what was done in .bdrv_amend_pre_run().
204 void (*bdrv_amend_clean
)(BlockDriverState
*bs
);
207 * Return true if @to_replace can be replaced by a BDS with the
208 * same data as @bs without it affecting @bs's behavior (that is,
209 * without it being visible to @bs's parents).
211 bool (*bdrv_recurse_can_replace
)(BlockDriverState
*bs
,
212 BlockDriverState
*to_replace
);
214 int (*bdrv_probe_device
)(const char *filename
);
217 * Any driver implementing this callback is expected to be able to handle
218 * NULL file names in its .bdrv_open() implementation.
220 void (*bdrv_parse_filename
)(const char *filename
, QDict
*options
,
223 /* For handling image reopen for split or non-split files. */
224 int (*bdrv_reopen_prepare
)(BDRVReopenState
*reopen_state
,
225 BlockReopenQueue
*queue
, Error
**errp
);
226 void (*bdrv_reopen_commit
)(BDRVReopenState
*reopen_state
);
227 void (*bdrv_reopen_commit_post
)(BDRVReopenState
*reopen_state
);
228 void (*bdrv_reopen_abort
)(BDRVReopenState
*reopen_state
);
229 void (*bdrv_join_options
)(QDict
*options
, QDict
*old_options
);
231 int (*bdrv_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
234 /* Protocol drivers should implement this instead of bdrv_open */
235 int (*bdrv_file_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
237 void (*bdrv_close
)(BlockDriverState
*bs
);
239 int coroutine_fn (*bdrv_co_create
)(BlockdevCreateOptions
*opts
,
241 int coroutine_fn (*bdrv_co_create_opts
)(BlockDriver
*drv
,
242 const char *filename
,
246 int (*bdrv_amend_options
)(BlockDriverState
*bs
,
248 BlockDriverAmendStatusCB
*status_cb
,
253 int (*bdrv_make_empty
)(BlockDriverState
*bs
);
256 * Refreshes the bs->exact_filename field. If that is impossible,
257 * bs->exact_filename has to be left empty.
259 void (*bdrv_refresh_filename
)(BlockDriverState
*bs
);
262 * Gathers the open options for all children into @target.
263 * A simple format driver (without backing file support) might
264 * implement this function like this:
266 * QINCREF(bs->file->bs->full_open_options);
267 * qdict_put(target, "file", bs->file->bs->full_open_options);
269 * If not specified, the generic implementation will simply put
270 * all children's options under their respective name.
272 * @backing_overridden is true when bs->backing seems not to be
273 * the child that would result from opening bs->backing_file.
274 * Therefore, if it is true, the backing child's options should be
275 * gathered; otherwise, there is no need since the backing child
276 * is the one implied by the image header.
278 * Note that ideally this function would not be needed. Every
279 * block driver which implements it is probably doing something
280 * shady regarding its runtime option structure.
282 void (*bdrv_gather_child_options
)(BlockDriverState
*bs
, QDict
*target
,
283 bool backing_overridden
);
286 * Returns an allocated string which is the directory name of this BDS: It
287 * will be used to make relative filenames absolute by prepending this
288 * function's return value to them.
290 char *(*bdrv_dirname
)(BlockDriverState
*bs
, Error
**errp
);
293 * This informs the driver that we are no longer interested in the result
294 * of in-flight requests, so don't waste the time if possible.
296 * One example usage is to avoid waiting for an nbd target node reconnect
297 * timeout during job-cancel with force=true.
299 void (*bdrv_cancel_in_flight
)(BlockDriverState
*bs
);
301 int (*bdrv_inactivate
)(BlockDriverState
*bs
);
303 int (*bdrv_snapshot_create
)(BlockDriverState
*bs
,
304 QEMUSnapshotInfo
*sn_info
);
305 int (*bdrv_snapshot_goto
)(BlockDriverState
*bs
,
306 const char *snapshot_id
);
307 int (*bdrv_snapshot_delete
)(BlockDriverState
*bs
,
308 const char *snapshot_id
,
311 int (*bdrv_snapshot_list
)(BlockDriverState
*bs
,
312 QEMUSnapshotInfo
**psn_info
);
313 int (*bdrv_snapshot_load_tmp
)(BlockDriverState
*bs
,
314 const char *snapshot_id
,
318 int (*bdrv_change_backing_file
)(BlockDriverState
*bs
,
319 const char *backing_file
, const char *backing_fmt
);
321 /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
322 int (*bdrv_debug_breakpoint
)(BlockDriverState
*bs
, const char *event
,
324 int (*bdrv_debug_remove_breakpoint
)(BlockDriverState
*bs
,
326 int (*bdrv_debug_resume
)(BlockDriverState
*bs
, const char *tag
);
327 bool (*bdrv_debug_is_suspended
)(BlockDriverState
*bs
, const char *tag
);
329 void (*bdrv_refresh_limits
)(BlockDriverState
*bs
, Error
**errp
);
332 * Returns 1 if newly created images are guaranteed to contain only
333 * zeros, 0 otherwise.
335 int (*bdrv_has_zero_init
)(BlockDriverState
*bs
);
338 * Remove fd handlers, timers, and other event loop callbacks so the event
339 * loop is no longer in use. Called with no in-flight requests and in
340 * depth-first traversal order with parents before child nodes.
342 void (*bdrv_detach_aio_context
)(BlockDriverState
*bs
);
345 * Add fd handlers, timers, and other event loop callbacks so I/O requests
346 * can be processed again. Called with no in-flight requests and in
347 * depth-first traversal order with child nodes before parent nodes.
349 void (*bdrv_attach_aio_context
)(BlockDriverState
*bs
,
350 AioContext
*new_context
);
353 * Try to get @bs's logical and physical block size.
354 * On success, store them in @bsz and return zero.
355 * On failure, return negative errno.
357 int (*bdrv_probe_blocksizes
)(BlockDriverState
*bs
, BlockSizes
*bsz
);
359 * Try to get @bs's geometry (cyls, heads, sectors)
360 * On success, store them in @geo and return 0.
361 * On failure return -errno.
362 * Only drivers that want to override guest geometry implement this
363 * callback; see hd_geometry_guess().
365 int (*bdrv_probe_geometry
)(BlockDriverState
*bs
, HDGeometry
*geo
);
367 void (*bdrv_add_child
)(BlockDriverState
*parent
, BlockDriverState
*child
,
369 void (*bdrv_del_child
)(BlockDriverState
*parent
, BdrvChild
*child
,
373 * Informs the block driver that a permission change is intended. The
374 * driver checks whether the change is permissible and may take other
375 * preparations for the change (e.g. get file system locks). This operation
376 * is always followed either by a call to either .bdrv_set_perm or
377 * .bdrv_abort_perm_update.
379 * Checks whether the requested set of cumulative permissions in @perm
380 * can be granted for accessing @bs and whether no other users are using
381 * permissions other than those given in @shared (both arguments take
382 * BLK_PERM_* bitmasks).
384 * If both conditions are met, 0 is returned. Otherwise, -errno is returned
385 * and errp is set to an error describing the conflict.
387 int (*bdrv_check_perm
)(BlockDriverState
*bs
, uint64_t perm
,
388 uint64_t shared
, Error
**errp
);
391 * Called to inform the driver that the set of cumulative set of used
392 * permissions for @bs has changed to @perm, and the set of sharable
393 * permission to @shared. The driver can use this to propagate changes to
394 * its children (i.e. request permissions only if a parent actually needs
397 * This function is only invoked after bdrv_check_perm(), so block drivers
398 * may rely on preparations made in their .bdrv_check_perm implementation.
400 void (*bdrv_set_perm
)(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
);
403 * Called to inform the driver that after a previous bdrv_check_perm()
404 * call, the permission update is not performed and any preparations made
405 * for it (e.g. taken file locks) need to be undone.
407 * This function can be called even for nodes that never saw a
408 * bdrv_check_perm() call. It is a no-op then.
410 void (*bdrv_abort_perm_update
)(BlockDriverState
*bs
);
413 * Returns in @nperm and @nshared the permissions that the driver for @bs
414 * needs on its child @c, based on the cumulative permissions requested by
415 * the parents in @parent_perm and @parent_shared.
417 * If @c is NULL, return the permissions for attaching a new child for the
418 * given @child_class and @role.
420 * If @reopen_queue is non-NULL, don't return the currently needed
421 * permissions, but those that will be needed after applying the
424 void (*bdrv_child_perm
)(BlockDriverState
*bs
, BdrvChild
*c
,
426 BlockReopenQueue
*reopen_queue
,
427 uint64_t parent_perm
, uint64_t parent_shared
,
428 uint64_t *nperm
, uint64_t *nshared
);
431 * Register/unregister a buffer for I/O. For example, when the driver is
432 * interested to know the memory areas that will later be used in iovs, so
433 * that it can do IOMMU mapping with VFIO etc., in order to get better
434 * performance. In the case of VFIO drivers, this callback is used to do
435 * DMA mapping for hot buffers.
437 * Returns: true on success, false on failure
439 bool (*bdrv_register_buf
)(BlockDriverState
*bs
, void *host
, size_t size
,
441 void (*bdrv_unregister_buf
)(BlockDriverState
*bs
, void *host
, size_t size
);
444 * This field is modified only under the BQL, and is part of
447 QLIST_ENTRY(BlockDriver
) list
;
450 * I/O API functions. These functions are thread-safe.
452 * See include/block/block-io.h for more information about
456 int (*bdrv_probe
)(const uint8_t *buf
, int buf_size
, const char *filename
);
458 int coroutine_fn (*bdrv_co_amend
)(BlockDriverState
*bs
,
459 BlockdevAmendOptions
*opts
,
464 BlockAIOCB
*(*bdrv_aio_preadv
)(BlockDriverState
*bs
,
465 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
466 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
467 BlockAIOCB
*(*bdrv_aio_pwritev
)(BlockDriverState
*bs
,
468 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
469 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
470 BlockAIOCB
*(*bdrv_aio_flush
)(BlockDriverState
*bs
,
471 BlockCompletionFunc
*cb
, void *opaque
);
472 BlockAIOCB
*(*bdrv_aio_pdiscard
)(BlockDriverState
*bs
,
473 int64_t offset
, int bytes
,
474 BlockCompletionFunc
*cb
, void *opaque
);
476 int coroutine_fn (*bdrv_co_readv
)(BlockDriverState
*bs
,
477 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
);
480 * @offset: position in bytes to read at
481 * @bytes: number of bytes to read
482 * @qiov: the buffers to fill with read data
483 * @flags: currently unused, always 0
485 * @offset and @bytes will be a multiple of 'request_alignment',
486 * but the length of individual @qiov elements does not have to
489 * @bytes will always equal the total size of @qiov, and will be
490 * no larger than 'max_transfer'.
492 * The buffer in @qiov may point directly to guest memory.
494 int coroutine_fn (*bdrv_co_preadv
)(BlockDriverState
*bs
,
495 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
496 BdrvRequestFlags flags
);
498 int coroutine_fn (*bdrv_co_preadv_part
)(BlockDriverState
*bs
,
499 int64_t offset
, int64_t bytes
,
500 QEMUIOVector
*qiov
, size_t qiov_offset
,
501 BdrvRequestFlags flags
);
503 int coroutine_fn (*bdrv_co_writev
)(BlockDriverState
*bs
,
504 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
507 * @offset: position in bytes to write at
508 * @bytes: number of bytes to write
509 * @qiov: the buffers containing data to write
510 * @flags: zero or more bits allowed by 'supported_write_flags'
512 * @offset and @bytes will be a multiple of 'request_alignment',
513 * but the length of individual @qiov elements does not have to
516 * @bytes will always equal the total size of @qiov, and will be
517 * no larger than 'max_transfer'.
519 * The buffer in @qiov may point directly to guest memory.
521 int coroutine_fn (*bdrv_co_pwritev
)(BlockDriverState
*bs
,
522 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
523 BdrvRequestFlags flags
);
524 int coroutine_fn (*bdrv_co_pwritev_part
)(BlockDriverState
*bs
,
525 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
526 BdrvRequestFlags flags
);
529 * Efficiently zero a region of the disk image. Typically an image format
530 * would use a compact metadata representation to implement this. This
531 * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
532 * will be called instead.
534 int coroutine_fn (*bdrv_co_pwrite_zeroes
)(BlockDriverState
*bs
,
535 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
);
536 int coroutine_fn (*bdrv_co_pdiscard
)(BlockDriverState
*bs
,
537 int64_t offset
, int64_t bytes
);
540 * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
541 * and invoke bdrv_co_copy_range_from(child, ...), or invoke
542 * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
544 * See the comment of bdrv_co_copy_range for the parameter and return value
547 int coroutine_fn (*bdrv_co_copy_range_from
)(BlockDriverState
*bs
,
553 BdrvRequestFlags read_flags
,
554 BdrvRequestFlags write_flags
);
557 * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
558 * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
559 * operation if @bs is the leaf and @src has the same BlockDriver. Return
560 * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
562 * See the comment of bdrv_co_copy_range for the parameter and return value
565 int coroutine_fn (*bdrv_co_copy_range_to
)(BlockDriverState
*bs
,
571 BdrvRequestFlags read_flags
,
572 BdrvRequestFlags write_flags
);
575 * Building block for bdrv_block_status[_above] and
576 * bdrv_is_allocated[_above]. The driver should answer only
577 * according to the current layer, and should only need to set
578 * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
579 * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
580 * layer, the result should be 0 (and not BDRV_BLOCK_ZERO). See
581 * block.h for the overall meaning of the bits. As a hint, the
582 * flag want_zero is true if the caller cares more about precise
583 * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
584 * overall allocation (favor larger *pnum, perhaps by reporting
585 * _DATA instead of _ZERO). The block layer guarantees input
586 * clamped to bdrv_getlength() and aligned to request_alignment,
587 * as well as non-NULL pnum, map, and file; in turn, the driver
588 * must return an error or set pnum to an aligned non-zero value.
590 * Note that @bytes is just a hint on how big of a region the
591 * caller wants to inspect. It is not a limit on *pnum.
592 * Implementations are free to return larger values of *pnum if
593 * doing so does not incur a performance penalty.
595 * block/io.c's bdrv_co_block_status() will utilize an unclamped
596 * *pnum value for the block-status cache on protocol nodes, prior
597 * to clamping *pnum for return to its caller.
599 int coroutine_fn (*bdrv_co_block_status
)(BlockDriverState
*bs
,
600 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
601 int64_t *map
, BlockDriverState
**file
);
604 * Snapshot-access API.
606 * Block-driver may provide snapshot-access API: special functions to access
607 * some internal "snapshot". The functions are similar with normal
608 * read/block_status/discard handler, but don't have any specific handling
609 * in generic block-layer: no serializing, no alignment, no tracked
610 * requests. So, block-driver that realizes these APIs is fully responsible
611 * for synchronization between snapshot-access API and normal IO requests.
613 * TODO: To be able to support qcow2's internal snapshots, this API will
614 * need to be extended to:
615 * - be able to select a specific snapshot
616 * - receive the snapshot's actual length (which may differ from bs's
619 int coroutine_fn (*bdrv_co_preadv_snapshot
)(BlockDriverState
*bs
,
620 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
);
621 int coroutine_fn (*bdrv_co_snapshot_block_status
)(BlockDriverState
*bs
,
622 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
623 int64_t *map
, BlockDriverState
**file
);
624 int coroutine_fn (*bdrv_co_pdiscard_snapshot
)(BlockDriverState
*bs
,
625 int64_t offset
, int64_t bytes
);
628 * Invalidate any cached meta-data.
630 void coroutine_fn (*bdrv_co_invalidate_cache
)(BlockDriverState
*bs
,
634 * Flushes all data for all layers by calling bdrv_co_flush for underlying
635 * layers, if needed. This function is needed for deterministic
636 * synchronization of the flush finishing callback.
638 int coroutine_fn (*bdrv_co_flush
)(BlockDriverState
*bs
);
640 /* Delete a created file. */
641 int coroutine_fn (*bdrv_co_delete_file
)(BlockDriverState
*bs
,
645 * Flushes all data that was already written to the OS all the way down to
646 * the disk (for example file-posix.c calls fsync()).
648 int coroutine_fn (*bdrv_co_flush_to_disk
)(BlockDriverState
*bs
);
651 * Flushes all internal caches to the OS. The data may still sit in a
652 * writeback cache of the host OS, but it will survive a crash of the qemu
655 int coroutine_fn (*bdrv_co_flush_to_os
)(BlockDriverState
*bs
);
658 * Truncate @bs to @offset bytes using the given @prealloc mode
659 * when growing. Modes other than PREALLOC_MODE_OFF should be
660 * rejected when shrinking @bs.
662 * If @exact is true, @bs must be resized to exactly @offset.
663 * Otherwise, it is sufficient for @bs (if it is a host block
664 * device and thus there is no way to resize it) to be at least
665 * @offset bytes in length.
667 * If @exact is true and this function fails but would succeed
668 * with @exact = false, it should return -ENOTSUP.
670 int coroutine_fn (*bdrv_co_truncate
)(BlockDriverState
*bs
, int64_t offset
,
671 bool exact
, PreallocMode prealloc
,
672 BdrvRequestFlags flags
, Error
**errp
);
673 int64_t (*bdrv_getlength
)(BlockDriverState
*bs
);
674 int64_t (*bdrv_get_allocated_file_size
)(BlockDriverState
*bs
);
675 BlockMeasureInfo
*(*bdrv_measure
)(QemuOpts
*opts
, BlockDriverState
*in_bs
,
678 int coroutine_fn (*bdrv_co_pwritev_compressed
)(BlockDriverState
*bs
,
679 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
);
680 int coroutine_fn (*bdrv_co_pwritev_compressed_part
)(BlockDriverState
*bs
,
681 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
684 int (*bdrv_get_info
)(BlockDriverState
*bs
, BlockDriverInfo
*bdi
);
686 ImageInfoSpecific
*(*bdrv_get_specific_info
)(BlockDriverState
*bs
,
688 BlockStatsSpecific
*(*bdrv_get_specific_stats
)(BlockDriverState
*bs
);
690 int coroutine_fn (*bdrv_save_vmstate
)(BlockDriverState
*bs
,
693 int coroutine_fn (*bdrv_load_vmstate
)(BlockDriverState
*bs
,
697 /* removable device specific */
698 bool (*bdrv_is_inserted
)(BlockDriverState
*bs
);
699 void (*bdrv_eject
)(BlockDriverState
*bs
, bool eject_flag
);
700 void (*bdrv_lock_medium
)(BlockDriverState
*bs
, bool locked
);
702 /* to control generic scsi devices */
703 BlockAIOCB
*(*bdrv_aio_ioctl
)(BlockDriverState
*bs
,
704 unsigned long int req
, void *buf
,
705 BlockCompletionFunc
*cb
, void *opaque
);
706 int coroutine_fn (*bdrv_co_ioctl
)(BlockDriverState
*bs
,
707 unsigned long int req
, void *buf
);
710 * Returns 0 for completed check, -errno for internal errors.
711 * The check results are stored in result.
713 int coroutine_fn (*bdrv_co_check
)(BlockDriverState
*bs
,
714 BdrvCheckResult
*result
,
717 void (*bdrv_debug_event
)(BlockDriverState
*bs
, BlkdebugEvent event
);
719 /* io queue for linux-aio */
720 void (*bdrv_io_plug
)(BlockDriverState
*bs
);
721 void (*bdrv_io_unplug
)(BlockDriverState
*bs
);
724 * bdrv_co_drain_begin is called if implemented in the beginning of a
725 * drain operation to drain and stop any internal sources of requests in
727 * bdrv_co_drain_end is called if implemented at the end of the drain.
729 * They should be used by the driver to e.g. manage scheduled I/O
730 * requests, or toggle an internal state. After the end of the drain new
731 * requests will continue normally.
733 void coroutine_fn (*bdrv_co_drain_begin
)(BlockDriverState
*bs
);
734 void coroutine_fn (*bdrv_co_drain_end
)(BlockDriverState
*bs
);
736 bool (*bdrv_supports_persistent_dirty_bitmap
)(BlockDriverState
*bs
);
737 bool (*bdrv_co_can_store_new_dirty_bitmap
)(BlockDriverState
*bs
,
739 uint32_t granularity
,
741 int (*bdrv_co_remove_persistent_dirty_bitmap
)(BlockDriverState
*bs
,
746 static inline bool block_driver_can_compress(BlockDriver
*drv
)
748 return drv
->bdrv_co_pwritev_compressed
||
749 drv
->bdrv_co_pwritev_compressed_part
;
752 typedef struct BlockLimits
{
754 * Alignment requirement, in bytes, for offset/length of I/O
755 * requests. Must be a power of 2 less than INT_MAX; defaults to
756 * 1 for drivers with modern byte interfaces, and to 512
759 uint32_t request_alignment
;
762 * Maximum number of bytes that can be discarded at once. Must be multiple
763 * of pdiscard_alignment, but need not be power of 2. May be 0 if no
764 * inherent 64-bit limit.
766 int64_t max_pdiscard
;
769 * Optimal alignment for discard requests in bytes. A power of 2
770 * is best but not mandatory. Must be a multiple of
771 * bl.request_alignment, and must be less than max_pdiscard if
772 * that is set. May be 0 if bl.request_alignment is good enough
774 uint32_t pdiscard_alignment
;
777 * Maximum number of bytes that can zeroized at once. Must be multiple of
778 * pwrite_zeroes_alignment. 0 means no limit.
780 int64_t max_pwrite_zeroes
;
783 * Optimal alignment for write zeroes requests in bytes. A power
784 * of 2 is best but not mandatory. Must be a multiple of
785 * bl.request_alignment, and must be less than max_pwrite_zeroes
786 * if that is set. May be 0 if bl.request_alignment is good
789 uint32_t pwrite_zeroes_alignment
;
792 * Optimal transfer length in bytes. A power of 2 is best but not
793 * mandatory. Must be a multiple of bl.request_alignment, or 0 if
796 uint32_t opt_transfer
;
799 * Maximal transfer length in bytes. Need not be power of 2, but
800 * must be multiple of opt_transfer and bl.request_alignment, or 0
801 * for no 32-bit limit. For now, anything larger than INT_MAX is
804 uint32_t max_transfer
;
807 * Maximal hardware transfer length in bytes. Applies whenever
808 * transfers to the device bypass the kernel I/O scheduler, for
809 * example with SG_IO. If larger than max_transfer or if zero,
810 * blk_get_max_hw_transfer will fall back to max_transfer.
812 uint64_t max_hw_transfer
;
815 * Maximal number of scatter/gather elements allowed by the hardware.
816 * Applies whenever transfers to the device bypass the kernel I/O
817 * scheduler, for example with SG_IO. If larger than max_iov
818 * or if zero, blk_get_max_hw_iov will fall back to max_iov.
823 /* memory alignment, in bytes so that no bounce buffer is needed */
824 size_t min_mem_alignment
;
826 /* memory alignment, in bytes, for bounce buffer */
827 size_t opt_mem_alignment
;
829 /* maximum number of iovec elements */
833 typedef struct BdrvOpBlocker BdrvOpBlocker
;
835 typedef struct BdrvAioNotifier
{
836 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
837 void (*detach_aio_context
)(void *opaque
);
842 QLIST_ENTRY(BdrvAioNotifier
) list
;
845 struct BdrvChildClass
{
847 * If true, bdrv_replace_node() doesn't change the node this BdrvChild
853 * If true, the parent is a BlockDriverState and bdrv_next_all_states()
854 * will return it. This information is used for drain_all, where every node
855 * will be drained separately, so the drain only needs to be propagated to
861 * Global state (GS) API. These functions run under the BQL.
863 * See include/block/block-global-state.h for more information about
866 void (*inherit_options
)(BdrvChildRole role
, bool parent_is_format
,
867 int *child_flags
, QDict
*child_options
,
868 int parent_flags
, QDict
*parent_options
);
869 void (*change_media
)(BdrvChild
*child
, bool load
);
872 * Returns a malloced string that describes the parent of the child for a
873 * human reader. This could be a node-name, BlockBackend name, qdev ID or
874 * QOM path of the device owning the BlockBackend, job type and ID etc. The
875 * caller is responsible for freeing the memory.
877 char *(*get_parent_desc
)(BdrvChild
*child
);
880 * Notifies the parent that the child has been activated/inactivated (e.g.
881 * when migration is completing) and it can start/stop requesting
882 * permissions and doing I/O on it.
884 void (*activate
)(BdrvChild
*child
, Error
**errp
);
885 int (*inactivate
)(BdrvChild
*child
);
887 void (*attach
)(BdrvChild
*child
);
888 void (*detach
)(BdrvChild
*child
);
891 * Notifies the parent that the filename of its child has changed (e.g.
892 * because the direct child was removed from the backing chain), so that it
893 * can update its reference.
895 int (*update_filename
)(BdrvChild
*child
, BlockDriverState
*new_base
,
896 const char *filename
, Error
**errp
);
898 bool (*can_set_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
,
899 GSList
**ignore
, Error
**errp
);
900 void (*set_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
, GSList
**ignore
);
902 AioContext
*(*get_parent_aio_context
)(BdrvChild
*child
);
905 * I/O API functions. These functions are thread-safe.
907 * See include/block/block-io.h for more information about
911 void (*resize
)(BdrvChild
*child
);
914 * Returns a name that is supposedly more useful for human users than the
915 * node name for identifying the node in question (in particular, a BB
916 * name), or NULL if the parent can't provide a better name.
918 const char *(*get_name
)(BdrvChild
*child
);
921 * If this pair of functions is implemented, the parent doesn't issue new
922 * requests after returning from .drained_begin() until .drained_end() is
925 * These functions must not change the graph (and therefore also must not
926 * call aio_poll(), which could change the graph indirectly).
928 * If drained_end() schedules background operations, it must atomically
929 * increment *drained_end_counter for each such operation and atomically
930 * decrement it once the operation has settled.
932 * Note that this can be nested. If drained_begin() was called twice, new
933 * I/O is allowed only after drained_end() was called twice, too.
935 void (*drained_begin
)(BdrvChild
*child
);
936 void (*drained_end
)(BdrvChild
*child
, int *drained_end_counter
);
939 * Returns whether the parent has pending requests for the child. This
940 * callback is polled after .drained_begin() has been called until all
941 * activity on the child has stopped.
943 bool (*drained_poll
)(BdrvChild
*child
);
946 extern const BdrvChildClass child_of_bds
;
949 BlockDriverState
*bs
;
951 const BdrvChildClass
*klass
;
956 * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
961 * Permissions that can still be granted to other users of @bs while this
962 * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
964 uint64_t shared_perm
;
967 * This link is frozen: the child can neither be replaced nor
968 * detached from the parent.
973 * How many times the parent of this child has been drained
974 * (through klass->drained_*).
975 * Usually, this is equal to bs->quiesce_counter (potentially
976 * reduced by bdrv_drain_all_count). It may differ while the
977 * child is entering or leaving a drained section.
979 int parent_quiesce_counter
;
981 QLIST_ENTRY(BdrvChild
) next
;
982 QLIST_ENTRY(BdrvChild
) next_parent
;
986 * Allows bdrv_co_block_status() to cache one data region for a
989 * @valid: Whether the cache is valid (should be accessed with atomic
990 * functions so this can be reset by RCU readers)
991 * @data_start: Offset where we know (or strongly assume) is data
992 * @data_end: Offset where the data region ends (which is not necessarily
993 * the start of a zeroed region)
995 typedef struct BdrvBlockStatusCache
{
1001 } BdrvBlockStatusCache
;
1003 struct BlockDriverState
{
1005 * Protected by big QEMU lock or read-only after opening. No special
1006 * locking needed during I/O...
1008 int open_flags
; /* flags used to open the file, re-used for re-open */
1009 bool encrypted
; /* if true, the media is encrypted */
1010 bool sg
; /* if true, the device is a /dev/sg* */
1011 bool probed
; /* if true, format was probed rather than specified */
1012 bool force_share
; /* if true, always allow all shared permissions */
1013 bool implicit
; /* if true, this filter node was automatically inserted */
1015 BlockDriver
*drv
; /* NULL means no media */
1018 AioContext
*aio_context
; /* event loop used for fd handlers, timers, etc */
1020 * long-running tasks intended to always use the same AioContext as this
1021 * BDS may register themselves in this list to be notified of changes
1022 * regarding this BDS's context
1024 QLIST_HEAD(, BdrvAioNotifier
) aio_notifiers
;
1025 bool walking_aio_notifiers
; /* to make removal during iteration safe */
1027 char filename
[PATH_MAX
];
1029 * If not empty, this image is a diff in relation to backing_file.
1030 * Note that this is the name given in the image header and
1031 * therefore may or may not be equal to .backing->bs->filename.
1032 * If this field contains a relative path, it is to be resolved
1033 * relatively to the overlay's location.
1035 char backing_file
[PATH_MAX
];
1037 * The backing filename indicated by the image header. Contrary
1038 * to backing_file, if we ever open this file, auto_backing_file
1039 * is replaced by the resulting BDS's filename (i.e. after a
1040 * bdrv_refresh_filename() run).
1042 char auto_backing_file
[PATH_MAX
];
1043 char backing_format
[16]; /* if non-zero and backing_file exists */
1045 QDict
*full_open_options
;
1046 char exact_filename
[PATH_MAX
];
1055 * Flags honored during pread
1057 BdrvRequestFlags supported_read_flags
;
1059 * Flags honored during pwrite (so far: BDRV_REQ_FUA,
1060 * BDRV_REQ_WRITE_UNCHANGED).
1061 * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
1062 * writes will be issued as normal writes without the flag set.
1063 * This is important to note for drivers that do not explicitly
1064 * request a WRITE permission for their children and instead take
1065 * the same permissions as their parent did (this is commonly what
1066 * block filters do). Such drivers have to be aware that the
1067 * parent may have taken a WRITE_UNCHANGED permission only and is
1068 * issuing such requests. Drivers either must make sure that
1069 * these requests do not result in plain WRITE accesses (usually
1070 * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1071 * every incoming write request as-is, including potentially that
1072 * flag), or they have to explicitly take the WRITE permission for
1075 BdrvRequestFlags supported_write_flags
;
1077 * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1078 * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1080 BdrvRequestFlags supported_zero_flags
;
1082 * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1084 * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1085 * that any added space reads as all zeros. If this can't be guaranteed,
1086 * the operation must fail.
1088 BdrvRequestFlags supported_truncate_flags
;
1090 /* the following member gives a name to every node on the bs graph. */
1092 /* element of the list of named nodes building the graph */
1093 QTAILQ_ENTRY(BlockDriverState
) node_list
;
1094 /* element of the list of all BlockDriverStates (all_bdrv_states) */
1095 QTAILQ_ENTRY(BlockDriverState
) bs_list
;
1096 /* element of the list of monitor-owned BDS */
1097 QTAILQ_ENTRY(BlockDriverState
) monitor_list
;
1100 /* operation blockers. Protected by BQL. */
1101 QLIST_HEAD(, BdrvOpBlocker
) op_blockers
[BLOCK_OP_TYPE_MAX
];
1104 * The node that this node inherited default options from (and a reopen on
1105 * which can affect this node by changing these defaults). This is always a
1106 * parent node of this node.
1108 BlockDriverState
*inherits_from
;
1109 QLIST_HEAD(, BdrvChild
) children
;
1110 QLIST_HEAD(, BdrvChild
) parents
;
1113 QDict
*explicit_options
;
1114 BlockdevDetectZeroesOptions detect_zeroes
;
1116 /* The error object in use for blocking operations on backing_hd */
1117 Error
*backing_blocker
;
1119 /* Protected by AioContext lock */
1122 * If we are reading a disk image, give its size in sectors.
1123 * Generally read-only; it is written to by load_snapshot and
1124 * save_snaphost, but the block layer is quiescent during those.
1126 int64_t total_sectors
;
1128 /* threshold limit for writes, in bytes. "High water mark". */
1129 uint64_t write_threshold_offset
;
1132 * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1133 * Reading from the list can be done with either the BQL or the
1134 * dirty_bitmap_mutex. Modifying a bitmap only requires
1135 * dirty_bitmap_mutex.
1137 QemuMutex dirty_bitmap_mutex
;
1138 QLIST_HEAD(, BdrvDirtyBitmap
) dirty_bitmaps
;
1140 /* Offset after the highest byte written to */
1141 Stat64 wr_highest_offset
;
1144 * If true, copy read backing sectors into image. Can be >1 if more
1145 * than one client has requested copy-on-read. Accessed with atomic
1151 * number of in-flight requests; overall and serialising.
1152 * Accessed with atomic ops.
1154 unsigned int in_flight
;
1155 unsigned int serialising_in_flight
;
1158 * counter for nested bdrv_io_plug.
1159 * Accessed with atomic ops.
1161 unsigned io_plugged
;
1163 /* do we need to tell the quest if we have a volatile write cache? */
1164 int enable_write_cache
;
1166 /* Accessed with atomic ops. */
1167 int quiesce_counter
;
1168 int recursive_quiesce_counter
;
1170 unsigned int write_gen
; /* Current data generation */
1172 /* Protected by reqs_lock. */
1174 QLIST_HEAD(, BdrvTrackedRequest
) tracked_requests
;
1175 CoQueue flush_queue
; /* Serializing flush queue */
1176 bool active_flush_req
; /* Flush request in flight? */
1178 /* Only read/written by whoever has set active_flush_req to true. */
1179 unsigned int flushed_gen
; /* Flushed write generation */
1181 /* BdrvChild links to this node may never be frozen */
1184 /* Lock for block-status cache RCU writers */
1185 CoMutex bsc_modify_lock
;
1186 /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1187 BdrvBlockStatusCache
*block_status_cache
;
1190 struct BlockBackendRootState
{
1192 BlockdevDetectZeroesOptions detect_zeroes
;
1195 typedef enum BlockMirrorBackingMode
{
1197 * Reuse the existing backing chain from the source for the target.
1198 * - sync=full: Set backing BDS to NULL.
1199 * - sync=top: Use source's backing BDS.
1200 * - sync=none: Use source as the backing BDS.
1202 MIRROR_SOURCE_BACKING_CHAIN
,
1204 /* Open the target's backing chain completely anew */
1205 MIRROR_OPEN_BACKING_CHAIN
,
1207 /* Do not change the target's backing BDS after job completion */
1208 MIRROR_LEAVE_BACKING_CHAIN
,
1209 } BlockMirrorBackingMode
;
1213 * Essential block drivers which must always be statically linked into qemu, and
1214 * which therefore can be accessed without using bdrv_find_format()
1216 extern BlockDriver bdrv_file
;
1217 extern BlockDriver bdrv_raw
;
1218 extern BlockDriver bdrv_qcow2
;
1220 extern unsigned int bdrv_drain_all_count
;
1221 extern QemuOptsList bdrv_create_opts_simple
;
1224 * Common functions that are neither I/O nor Global State.
1226 * See include/block/block-commmon.h for more information about
1230 static inline BlockDriverState
*child_bs(BdrvChild
*child
)
1232 return child
? child
->bs
: NULL
;
1235 int bdrv_check_request(int64_t offset
, int64_t bytes
, Error
**errp
);
1236 int get_tmp_filename(char *filename
, int size
);
1237 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
1241 int bdrv_check_qiov_request(int64_t offset
, int64_t bytes
,
1242 QEMUIOVector
*qiov
, size_t qiov_offset
,
1246 int is_windows_drive(const char *filename
);
1249 #endif /* BLOCK_INT_COMMON_H */