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
;
99 const char *format_name
;
103 * Set to true if the BlockDriver is a block filter. Block filters pass
104 * certain callbacks that refer to data (see block.c) to their bs->file
105 * or bs->backing (whichever one exists) if the driver doesn't implement
106 * them. Drivers that do not wish to forward must implement them and return
108 * Note that filters are not allowed to modify data.
110 * Filters generally cannot have more than a single filtered child,
111 * because the data they present must at all times be the same as
112 * that on their filtered child. That would be impossible to
113 * achieve for multiple filtered children.
114 * (And this filtered child must then be bs->file or bs->backing.)
118 * Set to true if the BlockDriver is a format driver. Format nodes
119 * generally do not expect their children to be other format nodes
120 * (except for backing files), and so format probing is disabled
126 * This function is invoked under BQL before .bdrv_co_amend()
127 * (which in contrast does not necessarily run under the BQL)
128 * to allow driver-specific initialization code that requires
129 * the BQL, like setting up specific permission flags.
131 int (*bdrv_amend_pre_run
)(BlockDriverState
*bs
, Error
**errp
);
133 * This function is invoked under BQL after .bdrv_co_amend()
134 * to allow cleaning up what was done in .bdrv_amend_pre_run().
136 void (*bdrv_amend_clean
)(BlockDriverState
*bs
);
139 * Return true if @to_replace can be replaced by a BDS with the
140 * same data as @bs without it affecting @bs's behavior (that is,
141 * without it being visible to @bs's parents).
143 bool (*bdrv_recurse_can_replace
)(BlockDriverState
*bs
,
144 BlockDriverState
*to_replace
);
146 int (*bdrv_probe
)(const uint8_t *buf
, int buf_size
, const char *filename
);
147 int (*bdrv_probe_device
)(const char *filename
);
150 * Any driver implementing this callback is expected to be able to handle
151 * NULL file names in its .bdrv_open() implementation.
153 void (*bdrv_parse_filename
)(const char *filename
, QDict
*options
,
156 * Drivers not implementing bdrv_parse_filename nor bdrv_open should have
157 * this field set to true, except ones that are defined only by their
159 * An example of the last type will be the quorum block driver.
161 bool bdrv_needs_filename
;
164 * Set if a driver can support backing files. This also implies the
165 * following semantics:
167 * - Return status 0 of .bdrv_co_block_status means that corresponding
168 * blocks are not allocated in this layer of backing-chain
169 * - For such (unallocated) blocks, read will:
170 * - fill buffer with zeros if there is no backing file
171 * - read from the backing file otherwise, where the block layer
172 * takes care of reading zeros beyond EOF if backing file is short
174 bool supports_backing
;
176 /* For handling image reopen for split or non-split files */
177 int (*bdrv_reopen_prepare
)(BDRVReopenState
*reopen_state
,
178 BlockReopenQueue
*queue
, Error
**errp
);
179 void (*bdrv_reopen_commit
)(BDRVReopenState
*reopen_state
);
180 void (*bdrv_reopen_commit_post
)(BDRVReopenState
*reopen_state
);
181 void (*bdrv_reopen_abort
)(BDRVReopenState
*reopen_state
);
182 void (*bdrv_join_options
)(QDict
*options
, QDict
*old_options
);
184 int (*bdrv_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
187 /* Protocol drivers should implement this instead of bdrv_open */
188 int (*bdrv_file_open
)(BlockDriverState
*bs
, QDict
*options
, int flags
,
190 void (*bdrv_close
)(BlockDriverState
*bs
);
193 int coroutine_fn (*bdrv_co_create
)(BlockdevCreateOptions
*opts
,
195 int coroutine_fn (*bdrv_co_create_opts
)(BlockDriver
*drv
,
196 const char *filename
,
200 int coroutine_fn (*bdrv_co_amend
)(BlockDriverState
*bs
,
201 BlockdevAmendOptions
*opts
,
205 int (*bdrv_amend_options
)(BlockDriverState
*bs
,
207 BlockDriverAmendStatusCB
*status_cb
,
212 int (*bdrv_make_empty
)(BlockDriverState
*bs
);
215 * Refreshes the bs->exact_filename field. If that is impossible,
216 * bs->exact_filename has to be left empty.
218 void (*bdrv_refresh_filename
)(BlockDriverState
*bs
);
221 * Gathers the open options for all children into @target.
222 * A simple format driver (without backing file support) might
223 * implement this function like this:
225 * QINCREF(bs->file->bs->full_open_options);
226 * qdict_put(target, "file", bs->file->bs->full_open_options);
228 * If not specified, the generic implementation will simply put
229 * all children's options under their respective name.
231 * @backing_overridden is true when bs->backing seems not to be
232 * the child that would result from opening bs->backing_file.
233 * Therefore, if it is true, the backing child's options should be
234 * gathered; otherwise, there is no need since the backing child
235 * is the one implied by the image header.
237 * Note that ideally this function would not be needed. Every
238 * block driver which implements it is probably doing something
239 * shady regarding its runtime option structure.
241 void (*bdrv_gather_child_options
)(BlockDriverState
*bs
, QDict
*target
,
242 bool backing_overridden
);
245 * Returns an allocated string which is the directory name of this BDS: It
246 * will be used to make relative filenames absolute by prepending this
247 * function's return value to them.
249 char *(*bdrv_dirname
)(BlockDriverState
*bs
, Error
**errp
);
252 BlockAIOCB
*(*bdrv_aio_preadv
)(BlockDriverState
*bs
,
253 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
254 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
255 BlockAIOCB
*(*bdrv_aio_pwritev
)(BlockDriverState
*bs
,
256 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
257 BdrvRequestFlags flags
, BlockCompletionFunc
*cb
, void *opaque
);
258 BlockAIOCB
*(*bdrv_aio_flush
)(BlockDriverState
*bs
,
259 BlockCompletionFunc
*cb
, void *opaque
);
260 BlockAIOCB
*(*bdrv_aio_pdiscard
)(BlockDriverState
*bs
,
261 int64_t offset
, int bytes
,
262 BlockCompletionFunc
*cb
, void *opaque
);
264 int coroutine_fn (*bdrv_co_readv
)(BlockDriverState
*bs
,
265 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
);
268 * @offset: position in bytes to read at
269 * @bytes: number of bytes to read
270 * @qiov: the buffers to fill with read data
271 * @flags: currently unused, always 0
273 * @offset and @bytes will be a multiple of 'request_alignment',
274 * but the length of individual @qiov elements does not have to
277 * @bytes will always equal the total size of @qiov, and will be
278 * no larger than 'max_transfer'.
280 * The buffer in @qiov may point directly to guest memory.
282 int coroutine_fn (*bdrv_co_preadv
)(BlockDriverState
*bs
,
283 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
284 BdrvRequestFlags flags
);
286 int coroutine_fn (*bdrv_co_preadv_part
)(BlockDriverState
*bs
,
287 int64_t offset
, int64_t bytes
,
288 QEMUIOVector
*qiov
, size_t qiov_offset
,
289 BdrvRequestFlags flags
);
291 int coroutine_fn (*bdrv_co_writev
)(BlockDriverState
*bs
,
292 int64_t sector_num
, int nb_sectors
, QEMUIOVector
*qiov
,
295 * @offset: position in bytes to write at
296 * @bytes: number of bytes to write
297 * @qiov: the buffers containing data to write
298 * @flags: zero or more bits allowed by 'supported_write_flags'
300 * @offset and @bytes will be a multiple of 'request_alignment',
301 * but the length of individual @qiov elements does not have to
304 * @bytes will always equal the total size of @qiov, and will be
305 * no larger than 'max_transfer'.
307 * The buffer in @qiov may point directly to guest memory.
309 int coroutine_fn (*bdrv_co_pwritev
)(BlockDriverState
*bs
,
310 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
311 BdrvRequestFlags flags
);
312 int coroutine_fn (*bdrv_co_pwritev_part
)(BlockDriverState
*bs
,
313 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
, size_t qiov_offset
,
314 BdrvRequestFlags flags
);
317 * Efficiently zero a region of the disk image. Typically an image format
318 * would use a compact metadata representation to implement this. This
319 * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
320 * will be called instead.
322 int coroutine_fn (*bdrv_co_pwrite_zeroes
)(BlockDriverState
*bs
,
323 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
);
324 int coroutine_fn (*bdrv_co_pdiscard
)(BlockDriverState
*bs
,
325 int64_t offset
, int64_t bytes
);
328 * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
329 * and invoke bdrv_co_copy_range_from(child, ...), or invoke
330 * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
332 * See the comment of bdrv_co_copy_range for the parameter and return value
335 int coroutine_fn (*bdrv_co_copy_range_from
)(BlockDriverState
*bs
,
341 BdrvRequestFlags read_flags
,
342 BdrvRequestFlags write_flags
);
345 * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
346 * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
347 * operation if @bs is the leaf and @src has the same BlockDriver. Return
348 * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
350 * See the comment of bdrv_co_copy_range for the parameter and return value
353 int coroutine_fn (*bdrv_co_copy_range_to
)(BlockDriverState
*bs
,
359 BdrvRequestFlags read_flags
,
360 BdrvRequestFlags write_flags
);
363 * Building block for bdrv_block_status[_above] and
364 * bdrv_is_allocated[_above]. The driver should answer only
365 * according to the current layer, and should only need to set
366 * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
367 * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
368 * layer, the result should be 0 (and not BDRV_BLOCK_ZERO). See
369 * block.h for the overall meaning of the bits. As a hint, the
370 * flag want_zero is true if the caller cares more about precise
371 * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
372 * overall allocation (favor larger *pnum, perhaps by reporting
373 * _DATA instead of _ZERO). The block layer guarantees input
374 * clamped to bdrv_getlength() and aligned to request_alignment,
375 * as well as non-NULL pnum, map, and file; in turn, the driver
376 * must return an error or set pnum to an aligned non-zero value.
378 * Note that @bytes is just a hint on how big of a region the
379 * caller wants to inspect. It is not a limit on *pnum.
380 * Implementations are free to return larger values of *pnum if
381 * doing so does not incur a performance penalty.
383 * block/io.c's bdrv_co_block_status() will utilize an unclamped
384 * *pnum value for the block-status cache on protocol nodes, prior
385 * to clamping *pnum for return to its caller.
387 int coroutine_fn (*bdrv_co_block_status
)(BlockDriverState
*bs
,
388 bool want_zero
, int64_t offset
, int64_t bytes
, int64_t *pnum
,
389 int64_t *map
, BlockDriverState
**file
);
392 * This informs the driver that we are no longer interested in the result
393 * of in-flight requests, so don't waste the time if possible.
395 * One example usage is to avoid waiting for an nbd target node reconnect
396 * timeout during job-cancel with force=true.
398 void (*bdrv_cancel_in_flight
)(BlockDriverState
*bs
);
401 * Invalidate any cached meta-data.
403 void coroutine_fn (*bdrv_co_invalidate_cache
)(BlockDriverState
*bs
,
405 int (*bdrv_inactivate
)(BlockDriverState
*bs
);
408 * Flushes all data for all layers by calling bdrv_co_flush for underlying
409 * layers, if needed. This function is needed for deterministic
410 * synchronization of the flush finishing callback.
412 int coroutine_fn (*bdrv_co_flush
)(BlockDriverState
*bs
);
414 /* Delete a created file. */
415 int coroutine_fn (*bdrv_co_delete_file
)(BlockDriverState
*bs
,
419 * Flushes all data that was already written to the OS all the way down to
420 * the disk (for example file-posix.c calls fsync()).
422 int coroutine_fn (*bdrv_co_flush_to_disk
)(BlockDriverState
*bs
);
425 * Flushes all internal caches to the OS. The data may still sit in a
426 * writeback cache of the host OS, but it will survive a crash of the qemu
429 int coroutine_fn (*bdrv_co_flush_to_os
)(BlockDriverState
*bs
);
432 * Drivers setting this field must be able to work with just a plain
433 * filename with '<protocol_name>:' as a prefix, and no other options.
434 * Options may be extracted from the filename by implementing
435 * bdrv_parse_filename.
437 const char *protocol_name
;
440 * Truncate @bs to @offset bytes using the given @prealloc mode
441 * when growing. Modes other than PREALLOC_MODE_OFF should be
442 * rejected when shrinking @bs.
444 * If @exact is true, @bs must be resized to exactly @offset.
445 * Otherwise, it is sufficient for @bs (if it is a host block
446 * device and thus there is no way to resize it) to be at least
447 * @offset bytes in length.
449 * If @exact is true and this function fails but would succeed
450 * with @exact = false, it should return -ENOTSUP.
452 int coroutine_fn (*bdrv_co_truncate
)(BlockDriverState
*bs
, int64_t offset
,
453 bool exact
, PreallocMode prealloc
,
454 BdrvRequestFlags flags
, Error
**errp
);
455 int64_t (*bdrv_getlength
)(BlockDriverState
*bs
);
456 bool has_variable_length
;
457 int64_t (*bdrv_get_allocated_file_size
)(BlockDriverState
*bs
);
458 BlockMeasureInfo
*(*bdrv_measure
)(QemuOpts
*opts
, BlockDriverState
*in_bs
,
461 int coroutine_fn (*bdrv_co_pwritev_compressed
)(BlockDriverState
*bs
,
462 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
);
463 int coroutine_fn (*bdrv_co_pwritev_compressed_part
)(BlockDriverState
*bs
,
464 int64_t offset
, int64_t bytes
, QEMUIOVector
*qiov
,
467 int (*bdrv_snapshot_create
)(BlockDriverState
*bs
,
468 QEMUSnapshotInfo
*sn_info
);
469 int (*bdrv_snapshot_goto
)(BlockDriverState
*bs
,
470 const char *snapshot_id
);
471 int (*bdrv_snapshot_delete
)(BlockDriverState
*bs
,
472 const char *snapshot_id
,
475 int (*bdrv_snapshot_list
)(BlockDriverState
*bs
,
476 QEMUSnapshotInfo
**psn_info
);
477 int (*bdrv_snapshot_load_tmp
)(BlockDriverState
*bs
,
478 const char *snapshot_id
,
481 int (*bdrv_get_info
)(BlockDriverState
*bs
, BlockDriverInfo
*bdi
);
483 ImageInfoSpecific
*(*bdrv_get_specific_info
)(BlockDriverState
*bs
,
485 BlockStatsSpecific
*(*bdrv_get_specific_stats
)(BlockDriverState
*bs
);
487 int coroutine_fn (*bdrv_save_vmstate
)(BlockDriverState
*bs
,
490 int coroutine_fn (*bdrv_load_vmstate
)(BlockDriverState
*bs
,
494 int (*bdrv_change_backing_file
)(BlockDriverState
*bs
,
495 const char *backing_file
, const char *backing_fmt
);
497 /* removable device specific */
498 bool (*bdrv_is_inserted
)(BlockDriverState
*bs
);
499 void (*bdrv_eject
)(BlockDriverState
*bs
, bool eject_flag
);
500 void (*bdrv_lock_medium
)(BlockDriverState
*bs
, bool locked
);
502 /* to control generic scsi devices */
503 BlockAIOCB
*(*bdrv_aio_ioctl
)(BlockDriverState
*bs
,
504 unsigned long int req
, void *buf
,
505 BlockCompletionFunc
*cb
, void *opaque
);
506 int coroutine_fn (*bdrv_co_ioctl
)(BlockDriverState
*bs
,
507 unsigned long int req
, void *buf
);
509 /* List of options for creating images, terminated by name == NULL */
510 QemuOptsList
*create_opts
;
512 /* List of options for image amend */
513 QemuOptsList
*amend_opts
;
516 * If this driver supports reopening images this contains a
517 * NULL-terminated list of the runtime options that can be
518 * modified. If an option in this list is unspecified during
519 * reopen then it _must_ be reset to its default value or return
522 const char *const *mutable_opts
;
525 * Returns 0 for completed check, -errno for internal errors.
526 * The check results are stored in result.
528 int coroutine_fn (*bdrv_co_check
)(BlockDriverState
*bs
,
529 BdrvCheckResult
*result
,
532 void (*bdrv_debug_event
)(BlockDriverState
*bs
, BlkdebugEvent event
);
534 /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
535 int (*bdrv_debug_breakpoint
)(BlockDriverState
*bs
, const char *event
,
537 int (*bdrv_debug_remove_breakpoint
)(BlockDriverState
*bs
,
539 int (*bdrv_debug_resume
)(BlockDriverState
*bs
, const char *tag
);
540 bool (*bdrv_debug_is_suspended
)(BlockDriverState
*bs
, const char *tag
);
542 void (*bdrv_refresh_limits
)(BlockDriverState
*bs
, Error
**errp
);
545 * Returns 1 if newly created images are guaranteed to contain only
546 * zeros, 0 otherwise.
548 int (*bdrv_has_zero_init
)(BlockDriverState
*bs
);
551 * Remove fd handlers, timers, and other event loop callbacks so the event
552 * loop is no longer in use. Called with no in-flight requests and in
553 * depth-first traversal order with parents before child nodes.
555 void (*bdrv_detach_aio_context
)(BlockDriverState
*bs
);
558 * Add fd handlers, timers, and other event loop callbacks so I/O requests
559 * can be processed again. Called with no in-flight requests and in
560 * depth-first traversal order with child nodes before parent nodes.
562 void (*bdrv_attach_aio_context
)(BlockDriverState
*bs
,
563 AioContext
*new_context
);
565 /* io queue for linux-aio */
566 void (*bdrv_io_plug
)(BlockDriverState
*bs
);
567 void (*bdrv_io_unplug
)(BlockDriverState
*bs
);
570 * Try to get @bs's logical and physical block size.
571 * On success, store them in @bsz and return zero.
572 * On failure, return negative errno.
574 int (*bdrv_probe_blocksizes
)(BlockDriverState
*bs
, BlockSizes
*bsz
);
576 * Try to get @bs's geometry (cyls, heads, sectors)
577 * On success, store them in @geo and return 0.
578 * On failure return -errno.
579 * Only drivers that want to override guest geometry implement this
580 * callback; see hd_geometry_guess().
582 int (*bdrv_probe_geometry
)(BlockDriverState
*bs
, HDGeometry
*geo
);
585 * bdrv_co_drain_begin is called if implemented in the beginning of a
586 * drain operation to drain and stop any internal sources of requests in
588 * bdrv_co_drain_end is called if implemented at the end of the drain.
590 * They should be used by the driver to e.g. manage scheduled I/O
591 * requests, or toggle an internal state. After the end of the drain new
592 * requests will continue normally.
594 void coroutine_fn (*bdrv_co_drain_begin
)(BlockDriverState
*bs
);
595 void coroutine_fn (*bdrv_co_drain_end
)(BlockDriverState
*bs
);
597 void (*bdrv_add_child
)(BlockDriverState
*parent
, BlockDriverState
*child
,
599 void (*bdrv_del_child
)(BlockDriverState
*parent
, BdrvChild
*child
,
603 * Informs the block driver that a permission change is intended. The
604 * driver checks whether the change is permissible and may take other
605 * preparations for the change (e.g. get file system locks). This operation
606 * is always followed either by a call to either .bdrv_set_perm or
607 * .bdrv_abort_perm_update.
609 * Checks whether the requested set of cumulative permissions in @perm
610 * can be granted for accessing @bs and whether no other users are using
611 * permissions other than those given in @shared (both arguments take
612 * BLK_PERM_* bitmasks).
614 * If both conditions are met, 0 is returned. Otherwise, -errno is returned
615 * and errp is set to an error describing the conflict.
617 int (*bdrv_check_perm
)(BlockDriverState
*bs
, uint64_t perm
,
618 uint64_t shared
, Error
**errp
);
621 * Called to inform the driver that the set of cumulative set of used
622 * permissions for @bs has changed to @perm, and the set of sharable
623 * permission to @shared. The driver can use this to propagate changes to
624 * its children (i.e. request permissions only if a parent actually needs
627 * This function is only invoked after bdrv_check_perm(), so block drivers
628 * may rely on preparations made in their .bdrv_check_perm implementation.
630 void (*bdrv_set_perm
)(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
);
633 * Called to inform the driver that after a previous bdrv_check_perm()
634 * call, the permission update is not performed and any preparations made
635 * for it (e.g. taken file locks) need to be undone.
637 * This function can be called even for nodes that never saw a
638 * bdrv_check_perm() call. It is a no-op then.
640 void (*bdrv_abort_perm_update
)(BlockDriverState
*bs
);
643 * Returns in @nperm and @nshared the permissions that the driver for @bs
644 * needs on its child @c, based on the cumulative permissions requested by
645 * the parents in @parent_perm and @parent_shared.
647 * If @c is NULL, return the permissions for attaching a new child for the
648 * given @child_class and @role.
650 * If @reopen_queue is non-NULL, don't return the currently needed
651 * permissions, but those that will be needed after applying the
654 void (*bdrv_child_perm
)(BlockDriverState
*bs
, BdrvChild
*c
,
656 BlockReopenQueue
*reopen_queue
,
657 uint64_t parent_perm
, uint64_t parent_shared
,
658 uint64_t *nperm
, uint64_t *nshared
);
660 bool (*bdrv_supports_persistent_dirty_bitmap
)(BlockDriverState
*bs
);
661 bool (*bdrv_co_can_store_new_dirty_bitmap
)(BlockDriverState
*bs
,
663 uint32_t granularity
,
665 int (*bdrv_co_remove_persistent_dirty_bitmap
)(BlockDriverState
*bs
,
670 * Register/unregister a buffer for I/O. For example, when the driver is
671 * interested to know the memory areas that will later be used in iovs, so
672 * that it can do IOMMU mapping with VFIO etc., in order to get better
673 * performance. In the case of VFIO drivers, this callback is used to do
674 * DMA mapping for hot buffers.
676 void (*bdrv_register_buf
)(BlockDriverState
*bs
, void *host
, size_t size
);
677 void (*bdrv_unregister_buf
)(BlockDriverState
*bs
, void *host
);
678 QLIST_ENTRY(BlockDriver
) list
;
681 * Pointer to a NULL-terminated array of names of strong options
682 * that can be specified for bdrv_open(). A strong option is one
683 * that changes the data of a BDS.
684 * If this pointer is NULL, the array is considered empty.
685 * "filename" and "driver" are always considered strong.
687 const char *const *strong_runtime_opts
;
690 static inline bool block_driver_can_compress(BlockDriver
*drv
)
692 return drv
->bdrv_co_pwritev_compressed
||
693 drv
->bdrv_co_pwritev_compressed_part
;
696 typedef struct BlockLimits
{
698 * Alignment requirement, in bytes, for offset/length of I/O
699 * requests. Must be a power of 2 less than INT_MAX; defaults to
700 * 1 for drivers with modern byte interfaces, and to 512
703 uint32_t request_alignment
;
706 * Maximum number of bytes that can be discarded at once. Must be multiple
707 * of pdiscard_alignment, but need not be power of 2. May be 0 if no
708 * inherent 64-bit limit.
710 int64_t max_pdiscard
;
713 * Optimal alignment for discard requests in bytes. A power of 2
714 * is best but not mandatory. Must be a multiple of
715 * bl.request_alignment, and must be less than max_pdiscard if
716 * that is set. May be 0 if bl.request_alignment is good enough
718 uint32_t pdiscard_alignment
;
721 * Maximum number of bytes that can zeroized at once. Must be multiple of
722 * pwrite_zeroes_alignment. 0 means no limit.
724 int64_t max_pwrite_zeroes
;
727 * Optimal alignment for write zeroes requests in bytes. A power
728 * of 2 is best but not mandatory. Must be a multiple of
729 * bl.request_alignment, and must be less than max_pwrite_zeroes
730 * if that is set. May be 0 if bl.request_alignment is good
733 uint32_t pwrite_zeroes_alignment
;
736 * Optimal transfer length in bytes. A power of 2 is best but not
737 * mandatory. Must be a multiple of bl.request_alignment, or 0 if
740 uint32_t opt_transfer
;
743 * Maximal transfer length in bytes. Need not be power of 2, but
744 * must be multiple of opt_transfer and bl.request_alignment, or 0
745 * for no 32-bit limit. For now, anything larger than INT_MAX is
748 uint32_t max_transfer
;
751 * Maximal hardware transfer length in bytes. Applies whenever
752 * transfers to the device bypass the kernel I/O scheduler, for
753 * example with SG_IO. If larger than max_transfer or if zero,
754 * blk_get_max_hw_transfer will fall back to max_transfer.
756 uint64_t max_hw_transfer
;
759 * Maximal number of scatter/gather elements allowed by the hardware.
760 * Applies whenever transfers to the device bypass the kernel I/O
761 * scheduler, for example with SG_IO. If larger than max_iov
762 * or if zero, blk_get_max_hw_iov will fall back to max_iov.
767 /* memory alignment, in bytes so that no bounce buffer is needed */
768 size_t min_mem_alignment
;
770 /* memory alignment, in bytes, for bounce buffer */
771 size_t opt_mem_alignment
;
773 /* maximum number of iovec elements */
777 typedef struct BdrvOpBlocker BdrvOpBlocker
;
779 typedef struct BdrvAioNotifier
{
780 void (*attached_aio_context
)(AioContext
*new_context
, void *opaque
);
781 void (*detach_aio_context
)(void *opaque
);
786 QLIST_ENTRY(BdrvAioNotifier
) list
;
789 struct BdrvChildClass
{
791 * If true, bdrv_replace_node() doesn't change the node this BdrvChild
797 * If true, the parent is a BlockDriverState and bdrv_next_all_states()
798 * will return it. This information is used for drain_all, where every node
799 * will be drained separately, so the drain only needs to be propagated to
804 void (*inherit_options
)(BdrvChildRole role
, bool parent_is_format
,
805 int *child_flags
, QDict
*child_options
,
806 int parent_flags
, QDict
*parent_options
);
808 void (*change_media
)(BdrvChild
*child
, bool load
);
809 void (*resize
)(BdrvChild
*child
);
812 * Returns a name that is supposedly more useful for human users than the
813 * node name for identifying the node in question (in particular, a BB
814 * name), or NULL if the parent can't provide a better name.
816 const char *(*get_name
)(BdrvChild
*child
);
819 * Returns a malloced string that describes the parent of the child for a
820 * human reader. This could be a node-name, BlockBackend name, qdev ID or
821 * QOM path of the device owning the BlockBackend, job type and ID etc. The
822 * caller is responsible for freeing the memory.
824 char *(*get_parent_desc
)(BdrvChild
*child
);
827 * If this pair of functions is implemented, the parent doesn't issue new
828 * requests after returning from .drained_begin() until .drained_end() is
831 * These functions must not change the graph (and therefore also must not
832 * call aio_poll(), which could change the graph indirectly).
834 * If drained_end() schedules background operations, it must atomically
835 * increment *drained_end_counter for each such operation and atomically
836 * decrement it once the operation has settled.
838 * Note that this can be nested. If drained_begin() was called twice, new
839 * I/O is allowed only after drained_end() was called twice, too.
841 void (*drained_begin
)(BdrvChild
*child
);
842 void (*drained_end
)(BdrvChild
*child
, int *drained_end_counter
);
845 * Returns whether the parent has pending requests for the child. This
846 * callback is polled after .drained_begin() has been called until all
847 * activity on the child has stopped.
849 bool (*drained_poll
)(BdrvChild
*child
);
852 * Notifies the parent that the child has been activated/inactivated (e.g.
853 * when migration is completing) and it can start/stop requesting
854 * permissions and doing I/O on it.
856 void (*activate
)(BdrvChild
*child
, Error
**errp
);
857 int (*inactivate
)(BdrvChild
*child
);
859 void (*attach
)(BdrvChild
*child
);
860 void (*detach
)(BdrvChild
*child
);
863 * Notifies the parent that the filename of its child has changed (e.g.
864 * because the direct child was removed from the backing chain), so that it
865 * can update its reference.
867 int (*update_filename
)(BdrvChild
*child
, BlockDriverState
*new_base
,
868 const char *filename
, Error
**errp
);
870 bool (*can_set_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
,
871 GSList
**ignore
, Error
**errp
);
872 void (*set_aio_ctx
)(BdrvChild
*child
, AioContext
*ctx
, GSList
**ignore
);
874 AioContext
*(*get_parent_aio_context
)(BdrvChild
*child
);
877 extern const BdrvChildClass child_of_bds
;
880 BlockDriverState
*bs
;
882 const BdrvChildClass
*klass
;
887 * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
892 * Permissions that can still be granted to other users of @bs while this
893 * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
895 uint64_t shared_perm
;
898 * This link is frozen: the child can neither be replaced nor
899 * detached from the parent.
904 * How many times the parent of this child has been drained
905 * (through klass->drained_*).
906 * Usually, this is equal to bs->quiesce_counter (potentially
907 * reduced by bdrv_drain_all_count). It may differ while the
908 * child is entering or leaving a drained section.
910 int parent_quiesce_counter
;
912 QLIST_ENTRY(BdrvChild
) next
;
913 QLIST_ENTRY(BdrvChild
) next_parent
;
917 * Allows bdrv_co_block_status() to cache one data region for a
920 * @valid: Whether the cache is valid (should be accessed with atomic
921 * functions so this can be reset by RCU readers)
922 * @data_start: Offset where we know (or strongly assume) is data
923 * @data_end: Offset where the data region ends (which is not necessarily
924 * the start of a zeroed region)
926 typedef struct BdrvBlockStatusCache
{
932 } BdrvBlockStatusCache
;
934 struct BlockDriverState
{
936 * Protected by big QEMU lock or read-only after opening. No special
937 * locking needed during I/O...
939 int open_flags
; /* flags used to open the file, re-used for re-open */
940 bool encrypted
; /* if true, the media is encrypted */
941 bool sg
; /* if true, the device is a /dev/sg* */
942 bool probed
; /* if true, format was probed rather than specified */
943 bool force_share
; /* if true, always allow all shared permissions */
944 bool implicit
; /* if true, this filter node was automatically inserted */
946 BlockDriver
*drv
; /* NULL means no media */
949 AioContext
*aio_context
; /* event loop used for fd handlers, timers, etc */
951 * long-running tasks intended to always use the same AioContext as this
952 * BDS may register themselves in this list to be notified of changes
953 * regarding this BDS's context
955 QLIST_HEAD(, BdrvAioNotifier
) aio_notifiers
;
956 bool walking_aio_notifiers
; /* to make removal during iteration safe */
958 char filename
[PATH_MAX
];
960 * If not empty, this image is a diff in relation to backing_file.
961 * Note that this is the name given in the image header and
962 * therefore may or may not be equal to .backing->bs->filename.
963 * If this field contains a relative path, it is to be resolved
964 * relatively to the overlay's location.
966 char backing_file
[PATH_MAX
];
968 * The backing filename indicated by the image header. Contrary
969 * to backing_file, if we ever open this file, auto_backing_file
970 * is replaced by the resulting BDS's filename (i.e. after a
971 * bdrv_refresh_filename() run).
973 char auto_backing_file
[PATH_MAX
];
974 char backing_format
[16]; /* if non-zero and backing_file exists */
976 QDict
*full_open_options
;
977 char exact_filename
[PATH_MAX
];
986 * Flags honored during pread
988 unsigned int supported_read_flags
;
990 * Flags honored during pwrite (so far: BDRV_REQ_FUA,
991 * BDRV_REQ_WRITE_UNCHANGED).
992 * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
993 * writes will be issued as normal writes without the flag set.
994 * This is important to note for drivers that do not explicitly
995 * request a WRITE permission for their children and instead take
996 * the same permissions as their parent did (this is commonly what
997 * block filters do). Such drivers have to be aware that the
998 * parent may have taken a WRITE_UNCHANGED permission only and is
999 * issuing such requests. Drivers either must make sure that
1000 * these requests do not result in plain WRITE accesses (usually
1001 * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1002 * every incoming write request as-is, including potentially that
1003 * flag), or they have to explicitly take the WRITE permission for
1006 unsigned int supported_write_flags
;
1008 * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1009 * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1011 unsigned int supported_zero_flags
;
1013 * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1015 * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1016 * that any added space reads as all zeros. If this can't be guaranteed,
1017 * the operation must fail.
1019 unsigned int supported_truncate_flags
;
1021 /* the following member gives a name to every node on the bs graph. */
1023 /* element of the list of named nodes building the graph */
1024 QTAILQ_ENTRY(BlockDriverState
) node_list
;
1025 /* element of the list of all BlockDriverStates (all_bdrv_states) */
1026 QTAILQ_ENTRY(BlockDriverState
) bs_list
;
1027 /* element of the list of monitor-owned BDS */
1028 QTAILQ_ENTRY(BlockDriverState
) monitor_list
;
1031 /* operation blockers. Protected by BQL. */
1032 QLIST_HEAD(, BdrvOpBlocker
) op_blockers
[BLOCK_OP_TYPE_MAX
];
1035 * The node that this node inherited default options from (and a reopen on
1036 * which can affect this node by changing these defaults). This is always a
1037 * parent node of this node.
1039 BlockDriverState
*inherits_from
;
1040 QLIST_HEAD(, BdrvChild
) children
;
1041 QLIST_HEAD(, BdrvChild
) parents
;
1044 QDict
*explicit_options
;
1045 BlockdevDetectZeroesOptions detect_zeroes
;
1047 /* The error object in use for blocking operations on backing_hd */
1048 Error
*backing_blocker
;
1050 /* Protected by AioContext lock */
1053 * If we are reading a disk image, give its size in sectors.
1054 * Generally read-only; it is written to by load_snapshot and
1055 * save_snaphost, but the block layer is quiescent during those.
1057 int64_t total_sectors
;
1059 /* threshold limit for writes, in bytes. "High water mark". */
1060 uint64_t write_threshold_offset
;
1063 * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1064 * Reading from the list can be done with either the BQL or the
1065 * dirty_bitmap_mutex. Modifying a bitmap only requires
1066 * dirty_bitmap_mutex.
1068 QemuMutex dirty_bitmap_mutex
;
1069 QLIST_HEAD(, BdrvDirtyBitmap
) dirty_bitmaps
;
1071 /* Offset after the highest byte written to */
1072 Stat64 wr_highest_offset
;
1075 * If true, copy read backing sectors into image. Can be >1 if more
1076 * than one client has requested copy-on-read. Accessed with atomic
1082 * number of in-flight requests; overall and serialising.
1083 * Accessed with atomic ops.
1085 unsigned int in_flight
;
1086 unsigned int serialising_in_flight
;
1089 * counter for nested bdrv_io_plug.
1090 * Accessed with atomic ops.
1092 unsigned io_plugged
;
1094 /* do we need to tell the quest if we have a volatile write cache? */
1095 int enable_write_cache
;
1097 /* Accessed with atomic ops. */
1098 int quiesce_counter
;
1099 int recursive_quiesce_counter
;
1101 unsigned int write_gen
; /* Current data generation */
1103 /* Protected by reqs_lock. */
1105 QLIST_HEAD(, BdrvTrackedRequest
) tracked_requests
;
1106 CoQueue flush_queue
; /* Serializing flush queue */
1107 bool active_flush_req
; /* Flush request in flight? */
1109 /* Only read/written by whoever has set active_flush_req to true. */
1110 unsigned int flushed_gen
; /* Flushed write generation */
1112 /* BdrvChild links to this node may never be frozen */
1115 /* Lock for block-status cache RCU writers */
1116 CoMutex bsc_modify_lock
;
1117 /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1118 BdrvBlockStatusCache
*block_status_cache
;
1121 struct BlockBackendRootState
{
1123 BlockdevDetectZeroesOptions detect_zeroes
;
1126 typedef enum BlockMirrorBackingMode
{
1128 * Reuse the existing backing chain from the source for the target.
1129 * - sync=full: Set backing BDS to NULL.
1130 * - sync=top: Use source's backing BDS.
1131 * - sync=none: Use source as the backing BDS.
1133 MIRROR_SOURCE_BACKING_CHAIN
,
1135 /* Open the target's backing chain completely anew */
1136 MIRROR_OPEN_BACKING_CHAIN
,
1138 /* Do not change the target's backing BDS after job completion */
1139 MIRROR_LEAVE_BACKING_CHAIN
,
1140 } BlockMirrorBackingMode
;
1144 * Essential block drivers which must always be statically linked into qemu, and
1145 * which therefore can be accessed without using bdrv_find_format()
1147 extern BlockDriver bdrv_file
;
1148 extern BlockDriver bdrv_raw
;
1149 extern BlockDriver bdrv_qcow2
;
1151 extern unsigned int bdrv_drain_all_count
;
1152 extern QemuOptsList bdrv_create_opts_simple
;
1155 * Common functions that are neither I/O nor Global State.
1157 * See include/block/block-commmon.h for more information about
1161 static inline BlockDriverState
*child_bs(BdrvChild
*child
)
1163 return child
? child
->bs
: NULL
;
1166 int bdrv_check_request(int64_t offset
, int64_t bytes
, Error
**errp
);
1167 int get_tmp_filename(char *filename
, int size
);
1168 void bdrv_parse_filename_strip_prefix(const char *filename
, const char *prefix
,
1172 int bdrv_check_qiov_request(int64_t offset
, int64_t bytes
,
1173 QEMUIOVector
*qiov
, size_t qiov_offset
,
1177 int is_windows_drive(const char *filename
);
1180 #endif /* BLOCK_INT_COMMON_H */