xref: /openbmc/qemu/include/block/block_int.h (revision 30b6852c)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to 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:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
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
22  * THE SOFTWARE.
23  */
24 #ifndef BLOCK_INT_H
25 #define BLOCK_INT_H
26 
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"
37 #include "qemu/rcu.h"
38 
39 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
40 
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"
64 
65 #define BLOCK_PROBE_BUF_SIZE        512
66 
67 enum BdrvTrackedRequestType {
68     BDRV_TRACKED_READ,
69     BDRV_TRACKED_WRITE,
70     BDRV_TRACKED_DISCARD,
71     BDRV_TRACKED_TRUNCATE,
72 };
73 
74 /*
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
79  */
80 typedef struct BdrvTrackedRequest {
81     BlockDriverState *bs;
82     int64_t offset;
83     int64_t bytes;
84     enum BdrvTrackedRequestType type;
85 
86     bool serialising;
87     int64_t overlap_offset;
88     int64_t overlap_bytes;
89 
90     QLIST_ENTRY(BdrvTrackedRequest) list;
91     Coroutine *co; /* owner, used for deadlock detection */
92     CoQueue wait_queue; /* coroutines blocked on this request */
93 
94     struct BdrvTrackedRequest *waiting_for;
95 } BdrvTrackedRequest;
96 
97 int bdrv_check_qiov_request(int64_t offset, int64_t bytes,
98                             QEMUIOVector *qiov, size_t qiov_offset,
99                             Error **errp);
100 int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp);
101 
102 struct BlockDriver {
103     const char *format_name;
104     int instance_size;
105 
106     /* set to true if the BlockDriver is a block filter. Block filters pass
107      * certain callbacks that refer to data (see block.c) to their bs->file
108      * or bs->backing (whichever one exists) if the driver doesn't implement
109      * them. Drivers that do not wish to forward must implement them and return
110      * -ENOTSUP.
111      * Note that filters are not allowed to modify data.
112      *
113      * Filters generally cannot have more than a single filtered child,
114      * because the data they present must at all times be the same as
115      * that on their filtered child.  That would be impossible to
116      * achieve for multiple filtered children.
117      * (And this filtered child must then be bs->file or bs->backing.)
118      */
119     bool is_filter;
120     /*
121      * Set to true if the BlockDriver is a format driver.  Format nodes
122      * generally do not expect their children to be other format nodes
123      * (except for backing files), and so format probing is disabled
124      * on those children.
125      */
126     bool is_format;
127     /*
128      * Return true if @to_replace can be replaced by a BDS with the
129      * same data as @bs without it affecting @bs's behavior (that is,
130      * without it being visible to @bs's parents).
131      */
132     bool (*bdrv_recurse_can_replace)(BlockDriverState *bs,
133                                      BlockDriverState *to_replace);
134 
135     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
136     int (*bdrv_probe_device)(const char *filename);
137 
138     /* Any driver implementing this callback is expected to be able to handle
139      * NULL file names in its .bdrv_open() implementation */
140     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
141     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
142      * this field set to true, except ones that are defined only by their
143      * child's bs.
144      * An example of the last type will be the quorum block driver.
145      */
146     bool bdrv_needs_filename;
147 
148     /*
149      * Set if a driver can support backing files. This also implies the
150      * following semantics:
151      *
152      *  - Return status 0 of .bdrv_co_block_status means that corresponding
153      *    blocks are not allocated in this layer of backing-chain
154      *  - For such (unallocated) blocks, read will:
155      *    - fill buffer with zeros if there is no backing file
156      *    - read from the backing file otherwise, where the block layer
157      *      takes care of reading zeros beyond EOF if backing file is short
158      */
159     bool supports_backing;
160 
161     /* For handling image reopen for split or non-split files */
162     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
163                                BlockReopenQueue *queue, Error **errp);
164     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
165     void (*bdrv_reopen_commit_post)(BDRVReopenState *reopen_state);
166     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
167     void (*bdrv_join_options)(QDict *options, QDict *old_options);
168 
169     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
170                      Error **errp);
171 
172     /* Protocol drivers should implement this instead of bdrv_open */
173     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
174                           Error **errp);
175     void (*bdrv_close)(BlockDriverState *bs);
176 
177 
178     int coroutine_fn (*bdrv_co_create)(BlockdevCreateOptions *opts,
179                                        Error **errp);
180     int coroutine_fn (*bdrv_co_create_opts)(BlockDriver *drv,
181                                             const char *filename,
182                                             QemuOpts *opts,
183                                             Error **errp);
184 
185     int coroutine_fn (*bdrv_co_amend)(BlockDriverState *bs,
186                                       BlockdevAmendOptions *opts,
187                                       bool force,
188                                       Error **errp);
189 
190     int (*bdrv_amend_options)(BlockDriverState *bs,
191                               QemuOpts *opts,
192                               BlockDriverAmendStatusCB *status_cb,
193                               void *cb_opaque,
194                               bool force,
195                               Error **errp);
196 
197     int (*bdrv_make_empty)(BlockDriverState *bs);
198 
199     /*
200      * Refreshes the bs->exact_filename field. If that is impossible,
201      * bs->exact_filename has to be left empty.
202      */
203     void (*bdrv_refresh_filename)(BlockDriverState *bs);
204 
205     /*
206      * Gathers the open options for all children into @target.
207      * A simple format driver (without backing file support) might
208      * implement this function like this:
209      *
210      *     QINCREF(bs->file->bs->full_open_options);
211      *     qdict_put(target, "file", bs->file->bs->full_open_options);
212      *
213      * If not specified, the generic implementation will simply put
214      * all children's options under their respective name.
215      *
216      * @backing_overridden is true when bs->backing seems not to be
217      * the child that would result from opening bs->backing_file.
218      * Therefore, if it is true, the backing child's options should be
219      * gathered; otherwise, there is no need since the backing child
220      * is the one implied by the image header.
221      *
222      * Note that ideally this function would not be needed.  Every
223      * block driver which implements it is probably doing something
224      * shady regarding its runtime option structure.
225      */
226     void (*bdrv_gather_child_options)(BlockDriverState *bs, QDict *target,
227                                       bool backing_overridden);
228 
229     /*
230      * Returns an allocated string which is the directory name of this BDS: It
231      * will be used to make relative filenames absolute by prepending this
232      * function's return value to them.
233      */
234     char *(*bdrv_dirname)(BlockDriverState *bs, Error **errp);
235 
236     /* aio */
237     BlockAIOCB *(*bdrv_aio_preadv)(BlockDriverState *bs,
238         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
239         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
240     BlockAIOCB *(*bdrv_aio_pwritev)(BlockDriverState *bs,
241         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
242         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
243     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
244         BlockCompletionFunc *cb, void *opaque);
245     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
246         int64_t offset, int bytes,
247         BlockCompletionFunc *cb, void *opaque);
248 
249     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
250         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
251 
252     /**
253      * @offset: position in bytes to read at
254      * @bytes: number of bytes to read
255      * @qiov: the buffers to fill with read data
256      * @flags: currently unused, always 0
257      *
258      * @offset and @bytes will be a multiple of 'request_alignment',
259      * but the length of individual @qiov elements does not have to
260      * be a multiple.
261      *
262      * @bytes will always equal the total size of @qiov, and will be
263      * no larger than 'max_transfer'.
264      *
265      * The buffer in @qiov may point directly to guest memory.
266      */
267     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
268         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
269         BdrvRequestFlags flags);
270     int coroutine_fn (*bdrv_co_preadv_part)(BlockDriverState *bs,
271         int64_t offset, int64_t bytes,
272         QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags);
273     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
274         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
275     /**
276      * @offset: position in bytes to write at
277      * @bytes: number of bytes to write
278      * @qiov: the buffers containing data to write
279      * @flags: zero or more bits allowed by 'supported_write_flags'
280      *
281      * @offset and @bytes will be a multiple of 'request_alignment',
282      * but the length of individual @qiov elements does not have to
283      * be a multiple.
284      *
285      * @bytes will always equal the total size of @qiov, and will be
286      * no larger than 'max_transfer'.
287      *
288      * The buffer in @qiov may point directly to guest memory.
289      */
290     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
291         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
292         BdrvRequestFlags flags);
293     int coroutine_fn (*bdrv_co_pwritev_part)(BlockDriverState *bs,
294         int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset,
295         BdrvRequestFlags flags);
296 
297     /*
298      * Efficiently zero a region of the disk image.  Typically an image format
299      * would use a compact metadata representation to implement this.  This
300      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
301      * will be called instead.
302      */
303     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
304         int64_t offset, int64_t bytes, BdrvRequestFlags flags);
305     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
306         int64_t offset, int64_t bytes);
307 
308     /* Map [offset, offset + nbytes) range onto a child of @bs to copy from,
309      * and invoke bdrv_co_copy_range_from(child, ...), or invoke
310      * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
311      *
312      * See the comment of bdrv_co_copy_range for the parameter and return value
313      * semantics.
314      */
315     int coroutine_fn (*bdrv_co_copy_range_from)(BlockDriverState *bs,
316                                                 BdrvChild *src,
317                                                 int64_t offset,
318                                                 BdrvChild *dst,
319                                                 int64_t dst_offset,
320                                                 int64_t bytes,
321                                                 BdrvRequestFlags read_flags,
322                                                 BdrvRequestFlags write_flags);
323 
324     /* Map [offset, offset + nbytes) range onto a child of bs to copy data to,
325      * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
326      * operation if @bs is the leaf and @src has the same BlockDriver.  Return
327      * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
328      *
329      * See the comment of bdrv_co_copy_range for the parameter and return value
330      * semantics.
331      */
332     int coroutine_fn (*bdrv_co_copy_range_to)(BlockDriverState *bs,
333                                               BdrvChild *src,
334                                               int64_t src_offset,
335                                               BdrvChild *dst,
336                                               int64_t dst_offset,
337                                               int64_t bytes,
338                                               BdrvRequestFlags read_flags,
339                                               BdrvRequestFlags write_flags);
340 
341     /*
342      * Building block for bdrv_block_status[_above] and
343      * bdrv_is_allocated[_above].  The driver should answer only
344      * according to the current layer, and should only need to set
345      * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
346      * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
347      * layer, the result should be 0 (and not BDRV_BLOCK_ZERO).  See
348      * block.h for the overall meaning of the bits.  As a hint, the
349      * flag want_zero is true if the caller cares more about precise
350      * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
351      * overall allocation (favor larger *pnum, perhaps by reporting
352      * _DATA instead of _ZERO).  The block layer guarantees input
353      * clamped to bdrv_getlength() and aligned to request_alignment,
354      * as well as non-NULL pnum, map, and file; in turn, the driver
355      * must return an error or set pnum to an aligned non-zero value.
356      *
357      * Note that @bytes is just a hint on how big of a region the
358      * caller wants to inspect.  It is not a limit on *pnum.
359      * Implementations are free to return larger values of *pnum if
360      * doing so does not incur a performance penalty.
361      *
362      * block/io.c's bdrv_co_block_status() will utilize an unclamped
363      * *pnum value for the block-status cache on protocol nodes, prior
364      * to clamping *pnum for return to its caller.
365      */
366     int coroutine_fn (*bdrv_co_block_status)(BlockDriverState *bs,
367         bool want_zero, int64_t offset, int64_t bytes, int64_t *pnum,
368         int64_t *map, BlockDriverState **file);
369 
370     /*
371      * This informs the driver that we are no longer interested in the result
372      * of in-flight requests, so don't waste the time if possible.
373      *
374      * One example usage is to avoid waiting for an nbd target node reconnect
375      * timeout during job-cancel with force=true.
376      */
377     void (*bdrv_cancel_in_flight)(BlockDriverState *bs);
378 
379     /*
380      * Invalidate any cached meta-data.
381      */
382     void coroutine_fn (*bdrv_co_invalidate_cache)(BlockDriverState *bs,
383                                                   Error **errp);
384     int (*bdrv_inactivate)(BlockDriverState *bs);
385 
386     /*
387      * Flushes all data for all layers by calling bdrv_co_flush for underlying
388      * layers, if needed. This function is needed for deterministic
389      * synchronization of the flush finishing callback.
390      */
391     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
392 
393     /* Delete a created file. */
394     int coroutine_fn (*bdrv_co_delete_file)(BlockDriverState *bs,
395                                             Error **errp);
396 
397     /*
398      * Flushes all data that was already written to the OS all the way down to
399      * the disk (for example file-posix.c calls fsync()).
400      */
401     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
402 
403     /*
404      * Flushes all internal caches to the OS. The data may still sit in a
405      * writeback cache of the host OS, but it will survive a crash of the qemu
406      * process.
407      */
408     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
409 
410     /*
411      * Drivers setting this field must be able to work with just a plain
412      * filename with '<protocol_name>:' as a prefix, and no other options.
413      * Options may be extracted from the filename by implementing
414      * bdrv_parse_filename.
415      */
416     const char *protocol_name;
417 
418     /*
419      * Truncate @bs to @offset bytes using the given @prealloc mode
420      * when growing.  Modes other than PREALLOC_MODE_OFF should be
421      * rejected when shrinking @bs.
422      *
423      * If @exact is true, @bs must be resized to exactly @offset.
424      * Otherwise, it is sufficient for @bs (if it is a host block
425      * device and thus there is no way to resize it) to be at least
426      * @offset bytes in length.
427      *
428      * If @exact is true and this function fails but would succeed
429      * with @exact = false, it should return -ENOTSUP.
430      */
431     int coroutine_fn (*bdrv_co_truncate)(BlockDriverState *bs, int64_t offset,
432                                          bool exact, PreallocMode prealloc,
433                                          BdrvRequestFlags flags, Error **errp);
434 
435     int64_t (*bdrv_getlength)(BlockDriverState *bs);
436     bool has_variable_length;
437     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
438     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
439                                       Error **errp);
440 
441     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
442         int64_t offset, int64_t bytes, QEMUIOVector *qiov);
443     int coroutine_fn (*bdrv_co_pwritev_compressed_part)(BlockDriverState *bs,
444         int64_t offset, int64_t bytes, QEMUIOVector *qiov, size_t qiov_offset);
445 
446     int (*bdrv_snapshot_create)(BlockDriverState *bs,
447                                 QEMUSnapshotInfo *sn_info);
448     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
449                               const char *snapshot_id);
450     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
451                                 const char *snapshot_id,
452                                 const char *name,
453                                 Error **errp);
454     int (*bdrv_snapshot_list)(BlockDriverState *bs,
455                               QEMUSnapshotInfo **psn_info);
456     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
457                                   const char *snapshot_id,
458                                   const char *name,
459                                   Error **errp);
460     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
461     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs,
462                                                  Error **errp);
463     BlockStatsSpecific *(*bdrv_get_specific_stats)(BlockDriverState *bs);
464 
465     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
466                                           QEMUIOVector *qiov,
467                                           int64_t pos);
468     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
469                                           QEMUIOVector *qiov,
470                                           int64_t pos);
471 
472     int (*bdrv_change_backing_file)(BlockDriverState *bs,
473         const char *backing_file, const char *backing_fmt);
474 
475     /* removable device specific */
476     bool (*bdrv_is_inserted)(BlockDriverState *bs);
477     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
478     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
479 
480     /* to control generic scsi devices */
481     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
482         unsigned long int req, void *buf,
483         BlockCompletionFunc *cb, void *opaque);
484     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
485                                       unsigned long int req, void *buf);
486 
487     /* List of options for creating images, terminated by name == NULL */
488     QemuOptsList *create_opts;
489 
490     /* List of options for image amend */
491     QemuOptsList *amend_opts;
492 
493     /*
494      * If this driver supports reopening images this contains a
495      * NULL-terminated list of the runtime options that can be
496      * modified. If an option in this list is unspecified during
497      * reopen then it _must_ be reset to its default value or return
498      * an error.
499      */
500     const char *const *mutable_opts;
501 
502     /*
503      * Returns 0 for completed check, -errno for internal errors.
504      * The check results are stored in result.
505      */
506     int coroutine_fn (*bdrv_co_check)(BlockDriverState *bs,
507                                       BdrvCheckResult *result,
508                                       BdrvCheckMode fix);
509 
510     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
511 
512     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
513     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
514         const char *tag);
515     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
516         const char *tag);
517     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
518     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
519 
520     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
521 
522     /*
523      * Returns 1 if newly created images are guaranteed to contain only
524      * zeros, 0 otherwise.
525      */
526     int (*bdrv_has_zero_init)(BlockDriverState *bs);
527 
528     /* Remove fd handlers, timers, and other event loop callbacks so the event
529      * loop is no longer in use.  Called with no in-flight requests and in
530      * depth-first traversal order with parents before child nodes.
531      */
532     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
533 
534     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
535      * can be processed again.  Called with no in-flight requests and in
536      * depth-first traversal order with child nodes before parent nodes.
537      */
538     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
539                                     AioContext *new_context);
540 
541     /* io queue for linux-aio */
542     void (*bdrv_io_plug)(BlockDriverState *bs);
543     void (*bdrv_io_unplug)(BlockDriverState *bs);
544 
545     /**
546      * Try to get @bs's logical and physical block size.
547      * On success, store them in @bsz and return zero.
548      * On failure, return negative errno.
549      */
550     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
551     /**
552      * Try to get @bs's geometry (cyls, heads, sectors)
553      * On success, store them in @geo and return 0.
554      * On failure return -errno.
555      * Only drivers that want to override guest geometry implement this
556      * callback; see hd_geometry_guess().
557      */
558     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
559 
560     /**
561      * bdrv_co_drain_begin is called if implemented in the beginning of a
562      * drain operation to drain and stop any internal sources of requests in
563      * the driver.
564      * bdrv_co_drain_end is called if implemented at the end of the drain.
565      *
566      * They should be used by the driver to e.g. manage scheduled I/O
567      * requests, or toggle an internal state. After the end of the drain new
568      * requests will continue normally.
569      */
570     void coroutine_fn (*bdrv_co_drain_begin)(BlockDriverState *bs);
571     void coroutine_fn (*bdrv_co_drain_end)(BlockDriverState *bs);
572 
573     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
574                            Error **errp);
575     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
576                            Error **errp);
577 
578     /**
579      * Informs the block driver that a permission change is intended. The
580      * driver checks whether the change is permissible and may take other
581      * preparations for the change (e.g. get file system locks). This operation
582      * is always followed either by a call to either .bdrv_set_perm or
583      * .bdrv_abort_perm_update.
584      *
585      * Checks whether the requested set of cumulative permissions in @perm
586      * can be granted for accessing @bs and whether no other users are using
587      * permissions other than those given in @shared (both arguments take
588      * BLK_PERM_* bitmasks).
589      *
590      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
591      * and errp is set to an error describing the conflict.
592      */
593     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
594                            uint64_t shared, Error **errp);
595 
596     /**
597      * Called to inform the driver that the set of cumulative set of used
598      * permissions for @bs has changed to @perm, and the set of sharable
599      * permission to @shared. The driver can use this to propagate changes to
600      * its children (i.e. request permissions only if a parent actually needs
601      * them).
602      *
603      * This function is only invoked after bdrv_check_perm(), so block drivers
604      * may rely on preparations made in their .bdrv_check_perm implementation.
605      */
606     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
607 
608     /*
609      * Called to inform the driver that after a previous bdrv_check_perm()
610      * call, the permission update is not performed and any preparations made
611      * for it (e.g. taken file locks) need to be undone.
612      *
613      * This function can be called even for nodes that never saw a
614      * bdrv_check_perm() call. It is a no-op then.
615      */
616     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
617 
618     /**
619      * Returns in @nperm and @nshared the permissions that the driver for @bs
620      * needs on its child @c, based on the cumulative permissions requested by
621      * the parents in @parent_perm and @parent_shared.
622      *
623      * If @c is NULL, return the permissions for attaching a new child for the
624      * given @child_class and @role.
625      *
626      * If @reopen_queue is non-NULL, don't return the currently needed
627      * permissions, but those that will be needed after applying the
628      * @reopen_queue.
629      */
630      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
631                              BdrvChildRole role,
632                              BlockReopenQueue *reopen_queue,
633                              uint64_t parent_perm, uint64_t parent_shared,
634                              uint64_t *nperm, uint64_t *nshared);
635 
636     bool (*bdrv_supports_persistent_dirty_bitmap)(BlockDriverState *bs);
637     bool (*bdrv_co_can_store_new_dirty_bitmap)(BlockDriverState *bs,
638                                                const char *name,
639                                                uint32_t granularity,
640                                                Error **errp);
641     int (*bdrv_co_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
642                                                   const char *name,
643                                                   Error **errp);
644 
645     /**
646      * Register/unregister a buffer for I/O. For example, when the driver is
647      * interested to know the memory areas that will later be used in iovs, so
648      * that it can do IOMMU mapping with VFIO etc., in order to get better
649      * performance. In the case of VFIO drivers, this callback is used to do
650      * DMA mapping for hot buffers.
651      */
652     void (*bdrv_register_buf)(BlockDriverState *bs, void *host, size_t size);
653     void (*bdrv_unregister_buf)(BlockDriverState *bs, void *host);
654     QLIST_ENTRY(BlockDriver) list;
655 
656     /* Pointer to a NULL-terminated array of names of strong options
657      * that can be specified for bdrv_open(). A strong option is one
658      * that changes the data of a BDS.
659      * If this pointer is NULL, the array is considered empty.
660      * "filename" and "driver" are always considered strong. */
661     const char *const *strong_runtime_opts;
662 };
663 
664 static inline bool block_driver_can_compress(BlockDriver *drv)
665 {
666     return drv->bdrv_co_pwritev_compressed ||
667            drv->bdrv_co_pwritev_compressed_part;
668 }
669 
670 typedef struct BlockLimits {
671     /* Alignment requirement, in bytes, for offset/length of I/O
672      * requests. Must be a power of 2 less than INT_MAX; defaults to
673      * 1 for drivers with modern byte interfaces, and to 512
674      * otherwise. */
675     uint32_t request_alignment;
676 
677     /*
678      * Maximum number of bytes that can be discarded at once. Must be multiple
679      * of pdiscard_alignment, but need not be power of 2. May be 0 if no
680      * inherent 64-bit limit.
681      */
682     int64_t max_pdiscard;
683 
684     /* Optimal alignment for discard requests in bytes. A power of 2
685      * is best but not mandatory.  Must be a multiple of
686      * bl.request_alignment, and must be less than max_pdiscard if
687      * that is set. May be 0 if bl.request_alignment is good enough */
688     uint32_t pdiscard_alignment;
689 
690     /*
691      * Maximum number of bytes that can zeroized at once. Must be multiple of
692      * pwrite_zeroes_alignment. 0 means no limit.
693      */
694     int64_t max_pwrite_zeroes;
695 
696     /* Optimal alignment for write zeroes requests in bytes. A power
697      * of 2 is best but not mandatory.  Must be a multiple of
698      * bl.request_alignment, and must be less than max_pwrite_zeroes
699      * if that is set. May be 0 if bl.request_alignment is good
700      * enough */
701     uint32_t pwrite_zeroes_alignment;
702 
703     /* Optimal transfer length in bytes.  A power of 2 is best but not
704      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
705      * no preferred size */
706     uint32_t opt_transfer;
707 
708     /* Maximal transfer length in bytes.  Need not be power of 2, but
709      * must be multiple of opt_transfer and bl.request_alignment, or 0
710      * for no 32-bit limit.  For now, anything larger than INT_MAX is
711      * clamped down. */
712     uint32_t max_transfer;
713 
714     /* Maximal hardware transfer length in bytes.  Applies whenever
715      * transfers to the device bypass the kernel I/O scheduler, for
716      * example with SG_IO.  If larger than max_transfer or if zero,
717      * blk_get_max_hw_transfer will fall back to max_transfer.
718      */
719     uint64_t max_hw_transfer;
720 
721     /* Maximal number of scatter/gather elements allowed by the hardware.
722      * Applies whenever transfers to the device bypass the kernel I/O
723      * scheduler, for example with SG_IO.  If larger than max_iov
724      * or if zero, blk_get_max_hw_iov will fall back to max_iov.
725      */
726     int max_hw_iov;
727 
728     /* memory alignment, in bytes so that no bounce buffer is needed */
729     size_t min_mem_alignment;
730 
731     /* memory alignment, in bytes, for bounce buffer */
732     size_t opt_mem_alignment;
733 
734     /* maximum number of iovec elements */
735     int max_iov;
736 } BlockLimits;
737 
738 typedef struct BdrvOpBlocker BdrvOpBlocker;
739 
740 typedef struct BdrvAioNotifier {
741     void (*attached_aio_context)(AioContext *new_context, void *opaque);
742     void (*detach_aio_context)(void *opaque);
743 
744     void *opaque;
745     bool deleted;
746 
747     QLIST_ENTRY(BdrvAioNotifier) list;
748 } BdrvAioNotifier;
749 
750 struct BdrvChildClass {
751     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
752      * points to. */
753     bool stay_at_node;
754 
755     /* If true, the parent is a BlockDriverState and bdrv_next_all_states()
756      * will return it. This information is used for drain_all, where every node
757      * will be drained separately, so the drain only needs to be propagated to
758      * non-BDS parents. */
759     bool parent_is_bds;
760 
761     void (*inherit_options)(BdrvChildRole role, bool parent_is_format,
762                             int *child_flags, QDict *child_options,
763                             int parent_flags, QDict *parent_options);
764 
765     void (*change_media)(BdrvChild *child, bool load);
766     void (*resize)(BdrvChild *child);
767 
768     /* Returns a name that is supposedly more useful for human users than the
769      * node name for identifying the node in question (in particular, a BB
770      * name), or NULL if the parent can't provide a better name. */
771     const char *(*get_name)(BdrvChild *child);
772 
773     /* Returns a malloced string that describes the parent of the child for a
774      * human reader. This could be a node-name, BlockBackend name, qdev ID or
775      * QOM path of the device owning the BlockBackend, job type and ID etc. The
776      * caller is responsible for freeing the memory. */
777     char *(*get_parent_desc)(BdrvChild *child);
778 
779     /*
780      * If this pair of functions is implemented, the parent doesn't issue new
781      * requests after returning from .drained_begin() until .drained_end() is
782      * called.
783      *
784      * These functions must not change the graph (and therefore also must not
785      * call aio_poll(), which could change the graph indirectly).
786      *
787      * If drained_end() schedules background operations, it must atomically
788      * increment *drained_end_counter for each such operation and atomically
789      * decrement it once the operation has settled.
790      *
791      * Note that this can be nested. If drained_begin() was called twice, new
792      * I/O is allowed only after drained_end() was called twice, too.
793      */
794     void (*drained_begin)(BdrvChild *child);
795     void (*drained_end)(BdrvChild *child, int *drained_end_counter);
796 
797     /*
798      * Returns whether the parent has pending requests for the child. This
799      * callback is polled after .drained_begin() has been called until all
800      * activity on the child has stopped.
801      */
802     bool (*drained_poll)(BdrvChild *child);
803 
804     /* Notifies the parent that the child has been activated/inactivated (e.g.
805      * when migration is completing) and it can start/stop requesting
806      * permissions and doing I/O on it. */
807     void (*activate)(BdrvChild *child, Error **errp);
808     int (*inactivate)(BdrvChild *child);
809 
810     void (*attach)(BdrvChild *child);
811     void (*detach)(BdrvChild *child);
812 
813     /* Notifies the parent that the filename of its child has changed (e.g.
814      * because the direct child was removed from the backing chain), so that it
815      * can update its reference. */
816     int (*update_filename)(BdrvChild *child, BlockDriverState *new_base,
817                            const char *filename, Error **errp);
818 
819     bool (*can_set_aio_ctx)(BdrvChild *child, AioContext *ctx,
820                             GSList **ignore, Error **errp);
821     void (*set_aio_ctx)(BdrvChild *child, AioContext *ctx, GSList **ignore);
822 
823     AioContext *(*get_parent_aio_context)(BdrvChild *child);
824 };
825 
826 extern const BdrvChildClass child_of_bds;
827 
828 struct BdrvChild {
829     BlockDriverState *bs;
830     char *name;
831     const BdrvChildClass *klass;
832     BdrvChildRole role;
833     void *opaque;
834 
835     /**
836      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
837      */
838     uint64_t perm;
839 
840     /**
841      * Permissions that can still be granted to other users of @bs while this
842      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
843      */
844     uint64_t shared_perm;
845 
846     /*
847      * This link is frozen: the child can neither be replaced nor
848      * detached from the parent.
849      */
850     bool frozen;
851 
852     /*
853      * How many times the parent of this child has been drained
854      * (through klass->drained_*).
855      * Usually, this is equal to bs->quiesce_counter (potentially
856      * reduced by bdrv_drain_all_count).  It may differ while the
857      * child is entering or leaving a drained section.
858      */
859     int parent_quiesce_counter;
860 
861     QLIST_ENTRY(BdrvChild) next;
862     QLIST_ENTRY(BdrvChild) next_parent;
863 };
864 
865 /*
866  * Allows bdrv_co_block_status() to cache one data region for a
867  * protocol node.
868  *
869  * @valid: Whether the cache is valid (should be accessed with atomic
870  *         functions so this can be reset by RCU readers)
871  * @data_start: Offset where we know (or strongly assume) is data
872  * @data_end: Offset where the data region ends (which is not necessarily
873  *            the start of a zeroed region)
874  */
875 typedef struct BdrvBlockStatusCache {
876     struct rcu_head rcu;
877 
878     bool valid;
879     int64_t data_start;
880     int64_t data_end;
881 } BdrvBlockStatusCache;
882 
883 struct BlockDriverState {
884     /* Protected by big QEMU lock or read-only after opening.  No special
885      * locking needed during I/O...
886      */
887     int open_flags; /* flags used to open the file, re-used for re-open */
888     bool encrypted; /* if true, the media is encrypted */
889     bool sg;        /* if true, the device is a /dev/sg* */
890     bool probed;    /* if true, format was probed rather than specified */
891     bool force_share; /* if true, always allow all shared permissions */
892     bool implicit;  /* if true, this filter node was automatically inserted */
893 
894     BlockDriver *drv; /* NULL means no media */
895     void *opaque;
896 
897     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
898     /* long-running tasks intended to always use the same AioContext as this
899      * BDS may register themselves in this list to be notified of changes
900      * regarding this BDS's context */
901     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
902     bool walking_aio_notifiers; /* to make removal during iteration safe */
903 
904     char filename[PATH_MAX];
905     /*
906      * If not empty, this image is a diff in relation to backing_file.
907      * Note that this is the name given in the image header and
908      * therefore may or may not be equal to .backing->bs->filename.
909      * If this field contains a relative path, it is to be resolved
910      * relatively to the overlay's location.
911      */
912     char backing_file[PATH_MAX];
913     /*
914      * The backing filename indicated by the image header.  Contrary
915      * to backing_file, if we ever open this file, auto_backing_file
916      * is replaced by the resulting BDS's filename (i.e. after a
917      * bdrv_refresh_filename() run).
918      */
919     char auto_backing_file[PATH_MAX];
920     char backing_format[16]; /* if non-zero and backing_file exists */
921 
922     QDict *full_open_options;
923     char exact_filename[PATH_MAX];
924 
925     BdrvChild *backing;
926     BdrvChild *file;
927 
928     /* I/O Limits */
929     BlockLimits bl;
930 
931     /*
932      * Flags honored during pread
933      */
934     unsigned int supported_read_flags;
935     /* Flags honored during pwrite (so far: BDRV_REQ_FUA,
936      * BDRV_REQ_WRITE_UNCHANGED).
937      * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
938      * writes will be issued as normal writes without the flag set.
939      * This is important to note for drivers that do not explicitly
940      * request a WRITE permission for their children and instead take
941      * the same permissions as their parent did (this is commonly what
942      * block filters do).  Such drivers have to be aware that the
943      * parent may have taken a WRITE_UNCHANGED permission only and is
944      * issuing such requests.  Drivers either must make sure that
945      * these requests do not result in plain WRITE accesses (usually
946      * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
947      * every incoming write request as-is, including potentially that
948      * flag), or they have to explicitly take the WRITE permission for
949      * their children. */
950     unsigned int supported_write_flags;
951     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
952      * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED) */
953     unsigned int supported_zero_flags;
954     /*
955      * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
956      *
957      * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
958      * that any added space reads as all zeros. If this can't be guaranteed,
959      * the operation must fail.
960      */
961     unsigned int supported_truncate_flags;
962 
963     /* the following member gives a name to every node on the bs graph. */
964     char node_name[32];
965     /* element of the list of named nodes building the graph */
966     QTAILQ_ENTRY(BlockDriverState) node_list;
967     /* element of the list of all BlockDriverStates (all_bdrv_states) */
968     QTAILQ_ENTRY(BlockDriverState) bs_list;
969     /* element of the list of monitor-owned BDS */
970     QTAILQ_ENTRY(BlockDriverState) monitor_list;
971     int refcnt;
972 
973     /* operation blockers */
974     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
975 
976     /* The node that this node inherited default options from (and a reopen on
977      * which can affect this node by changing these defaults). This is always a
978      * parent node of this node. */
979     BlockDriverState *inherits_from;
980     QLIST_HEAD(, BdrvChild) children;
981     QLIST_HEAD(, BdrvChild) parents;
982 
983     QDict *options;
984     QDict *explicit_options;
985     BlockdevDetectZeroesOptions detect_zeroes;
986 
987     /* The error object in use for blocking operations on backing_hd */
988     Error *backing_blocker;
989 
990     /* Protected by AioContext lock */
991 
992     /* If we are reading a disk image, give its size in sectors.
993      * Generally read-only; it is written to by load_snapshot and
994      * save_snaphost, but the block layer is quiescent during those.
995      */
996     int64_t total_sectors;
997 
998     /* threshold limit for writes, in bytes. "High water mark". */
999     uint64_t write_threshold_offset;
1000 
1001     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1002      * Reading from the list can be done with either the BQL or the
1003      * dirty_bitmap_mutex.  Modifying a bitmap only requires
1004      * dirty_bitmap_mutex.  */
1005     QemuMutex dirty_bitmap_mutex;
1006     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
1007 
1008     /* Offset after the highest byte written to */
1009     Stat64 wr_highest_offset;
1010 
1011     /* If true, copy read backing sectors into image.  Can be >1 if more
1012      * than one client has requested copy-on-read.  Accessed with atomic
1013      * ops.
1014      */
1015     int copy_on_read;
1016 
1017     /* number of in-flight requests; overall and serialising.
1018      * Accessed with atomic ops.
1019      */
1020     unsigned int in_flight;
1021     unsigned int serialising_in_flight;
1022 
1023     /* counter for nested bdrv_io_plug.
1024      * Accessed with atomic ops.
1025     */
1026     unsigned io_plugged;
1027 
1028     /* do we need to tell the quest if we have a volatile write cache? */
1029     int enable_write_cache;
1030 
1031     /* Accessed with atomic ops.  */
1032     int quiesce_counter;
1033     int recursive_quiesce_counter;
1034 
1035     unsigned int write_gen;               /* Current data generation */
1036 
1037     /* Protected by reqs_lock.  */
1038     CoMutex reqs_lock;
1039     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
1040     CoQueue flush_queue;                  /* Serializing flush queue */
1041     bool active_flush_req;                /* Flush request in flight? */
1042 
1043     /* Only read/written by whoever has set active_flush_req to true.  */
1044     unsigned int flushed_gen;             /* Flushed write generation */
1045 
1046     /* BdrvChild links to this node may never be frozen */
1047     bool never_freeze;
1048 
1049     /* Lock for block-status cache RCU writers */
1050     CoMutex bsc_modify_lock;
1051     /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1052     BdrvBlockStatusCache *block_status_cache;
1053 };
1054 
1055 struct BlockBackendRootState {
1056     int open_flags;
1057     BlockdevDetectZeroesOptions detect_zeroes;
1058 };
1059 
1060 typedef enum BlockMirrorBackingMode {
1061     /* Reuse the existing backing chain from the source for the target.
1062      * - sync=full: Set backing BDS to NULL.
1063      * - sync=top:  Use source's backing BDS.
1064      * - sync=none: Use source as the backing BDS. */
1065     MIRROR_SOURCE_BACKING_CHAIN,
1066 
1067     /* Open the target's backing chain completely anew */
1068     MIRROR_OPEN_BACKING_CHAIN,
1069 
1070     /* Do not change the target's backing BDS after job completion */
1071     MIRROR_LEAVE_BACKING_CHAIN,
1072 } BlockMirrorBackingMode;
1073 
1074 
1075 /* Essential block drivers which must always be statically linked into qemu, and
1076  * which therefore can be accessed without using bdrv_find_format() */
1077 extern BlockDriver bdrv_file;
1078 extern BlockDriver bdrv_raw;
1079 extern BlockDriver bdrv_qcow2;
1080 
1081 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
1082     int64_t offset, int64_t bytes, QEMUIOVector *qiov,
1083     BdrvRequestFlags flags);
1084 int coroutine_fn bdrv_co_preadv_part(BdrvChild *child,
1085     int64_t offset, int64_t bytes,
1086     QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags);
1087 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
1088     int64_t offset, int64_t bytes, QEMUIOVector *qiov,
1089     BdrvRequestFlags flags);
1090 int coroutine_fn bdrv_co_pwritev_part(BdrvChild *child,
1091     int64_t offset, int64_t bytes,
1092     QEMUIOVector *qiov, size_t qiov_offset, BdrvRequestFlags flags);
1093 
1094 static inline int coroutine_fn bdrv_co_pread(BdrvChild *child,
1095     int64_t offset, unsigned int bytes, void *buf, BdrvRequestFlags flags)
1096 {
1097     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1098 
1099     return bdrv_co_preadv(child, offset, bytes, &qiov, flags);
1100 }
1101 
1102 static inline int coroutine_fn bdrv_co_pwrite(BdrvChild *child,
1103     int64_t offset, unsigned int bytes, void *buf, BdrvRequestFlags flags)
1104 {
1105     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
1106 
1107     return bdrv_co_pwritev(child, offset, bytes, &qiov, flags);
1108 }
1109 
1110 extern unsigned int bdrv_drain_all_count;
1111 void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent);
1112 void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent);
1113 
1114 bool coroutine_fn bdrv_make_request_serialising(BdrvTrackedRequest *req,
1115                                                 uint64_t align);
1116 BdrvTrackedRequest *coroutine_fn bdrv_co_get_self_request(BlockDriverState *bs);
1117 
1118 int get_tmp_filename(char *filename, int size);
1119 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
1120                             const char *filename);
1121 
1122 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
1123                                       QDict *options);
1124 
1125 bool bdrv_backing_overridden(BlockDriverState *bs);
1126 
1127 
1128 /**
1129  * bdrv_add_aio_context_notifier:
1130  *
1131  * If a long-running job intends to be always run in the same AioContext as a
1132  * certain BDS, it may use this function to be notified of changes regarding the
1133  * association of the BDS to an AioContext.
1134  *
1135  * attached_aio_context() is called after the target BDS has been attached to a
1136  * new AioContext; detach_aio_context() is called before the target BDS is being
1137  * detached from its old AioContext.
1138  */
1139 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
1140         void (*attached_aio_context)(AioContext *new_context, void *opaque),
1141         void (*detach_aio_context)(void *opaque), void *opaque);
1142 
1143 /**
1144  * bdrv_remove_aio_context_notifier:
1145  *
1146  * Unsubscribe of change notifications regarding the BDS's AioContext. The
1147  * parameters given here have to be the same as those given to
1148  * bdrv_add_aio_context_notifier().
1149  */
1150 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
1151                                       void (*aio_context_attached)(AioContext *,
1152                                                                    void *),
1153                                       void (*aio_context_detached)(void *),
1154                                       void *opaque);
1155 
1156 /**
1157  * bdrv_wakeup:
1158  * @bs: The BlockDriverState for which an I/O operation has been completed.
1159  *
1160  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
1161  * synchronous I/O on a BlockDriverState that is attached to another
1162  * I/O thread, the main thread lets the I/O thread's event loop run,
1163  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
1164  * up the main thread if necessary.
1165  *
1166  * Manual calls to bdrv_wakeup are rarely necessary, because
1167  * bdrv_dec_in_flight already calls it.
1168  */
1169 void bdrv_wakeup(BlockDriverState *bs);
1170 
1171 #ifdef _WIN32
1172 int is_windows_drive(const char *filename);
1173 #endif
1174 
1175 /**
1176  * stream_start:
1177  * @job_id: The id of the newly-created job, or %NULL to use the
1178  * device name of @bs.
1179  * @bs: Block device to operate on.
1180  * @base: Block device that will become the new base, or %NULL to
1181  * flatten the whole backing file chain onto @bs.
1182  * @backing_file_str: The file name that will be written to @bs as the
1183  * the new backing file if the job completes. Ignored if @base is %NULL.
1184  * @creation_flags: Flags that control the behavior of the Job lifetime.
1185  *                  See @BlockJobCreateFlags
1186  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
1187  * @on_error: The action to take upon error.
1188  * @filter_node_name: The node name that should be assigned to the filter
1189  *                    driver that the stream job inserts into the graph above
1190  *                    @bs. NULL means that a node name should be autogenerated.
1191  * @errp: Error object.
1192  *
1193  * Start a streaming operation on @bs.  Clusters that are unallocated
1194  * in @bs, but allocated in any image between @base and @bs (both
1195  * exclusive) will be written to @bs.  At the end of a successful
1196  * streaming job, the backing file of @bs will be changed to
1197  * @backing_file_str in the written image and to @base in the live
1198  * BlockDriverState.
1199  */
1200 void stream_start(const char *job_id, BlockDriverState *bs,
1201                   BlockDriverState *base, const char *backing_file_str,
1202                   BlockDriverState *bottom,
1203                   int creation_flags, int64_t speed,
1204                   BlockdevOnError on_error,
1205                   const char *filter_node_name,
1206                   Error **errp);
1207 
1208 /**
1209  * commit_start:
1210  * @job_id: The id of the newly-created job, or %NULL to use the
1211  * device name of @bs.
1212  * @bs: Active block device.
1213  * @top: Top block device to be committed.
1214  * @base: Block device that will be written into, and become the new top.
1215  * @creation_flags: Flags that control the behavior of the Job lifetime.
1216  *                  See @BlockJobCreateFlags
1217  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
1218  * @on_error: The action to take upon error.
1219  * @backing_file_str: String to use as the backing file in @top's overlay
1220  * @filter_node_name: The node name that should be assigned to the filter
1221  * driver that the commit job inserts into the graph above @top. NULL means
1222  * that a node name should be autogenerated.
1223  * @errp: Error object.
1224  *
1225  */
1226 void commit_start(const char *job_id, BlockDriverState *bs,
1227                   BlockDriverState *base, BlockDriverState *top,
1228                   int creation_flags, int64_t speed,
1229                   BlockdevOnError on_error, const char *backing_file_str,
1230                   const char *filter_node_name, Error **errp);
1231 /**
1232  * commit_active_start:
1233  * @job_id: The id of the newly-created job, or %NULL to use the
1234  * device name of @bs.
1235  * @bs: Active block device to be committed.
1236  * @base: Block device that will be written into, and become the new top.
1237  * @creation_flags: Flags that control the behavior of the Job lifetime.
1238  *                  See @BlockJobCreateFlags
1239  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
1240  * @on_error: The action to take upon error.
1241  * @filter_node_name: The node name that should be assigned to the filter
1242  * driver that the commit job inserts into the graph above @bs. NULL means that
1243  * a node name should be autogenerated.
1244  * @cb: Completion function for the job.
1245  * @opaque: Opaque pointer value passed to @cb.
1246  * @auto_complete: Auto complete the job.
1247  * @errp: Error object.
1248  *
1249  */
1250 BlockJob *commit_active_start(const char *job_id, BlockDriverState *bs,
1251                               BlockDriverState *base, int creation_flags,
1252                               int64_t speed, BlockdevOnError on_error,
1253                               const char *filter_node_name,
1254                               BlockCompletionFunc *cb, void *opaque,
1255                               bool auto_complete, Error **errp);
1256 /*
1257  * mirror_start:
1258  * @job_id: The id of the newly-created job, or %NULL to use the
1259  * device name of @bs.
1260  * @bs: Block device to operate on.
1261  * @target: Block device to write to.
1262  * @replaces: Block graph node name to replace once the mirror is done. Can
1263  *            only be used when full mirroring is selected.
1264  * @creation_flags: Flags that control the behavior of the Job lifetime.
1265  *                  See @BlockJobCreateFlags
1266  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
1267  * @granularity: The chosen granularity for the dirty bitmap.
1268  * @buf_size: The amount of data that can be in flight at one time.
1269  * @mode: Whether to collapse all images in the chain to the target.
1270  * @backing_mode: How to establish the target's backing chain after completion.
1271  * @zero_target: Whether the target should be explicitly zero-initialized
1272  * @on_source_error: The action to take upon error reading from the source.
1273  * @on_target_error: The action to take upon error writing to the target.
1274  * @unmap: Whether to unmap target where source sectors only contain zeroes.
1275  * @filter_node_name: The node name that should be assigned to the filter
1276  * driver that the mirror job inserts into the graph above @bs. NULL means that
1277  * a node name should be autogenerated.
1278  * @copy_mode: When to trigger writes to the target.
1279  * @errp: Error object.
1280  *
1281  * Start a mirroring operation on @bs.  Clusters that are allocated
1282  * in @bs will be written to @target until the job is cancelled or
1283  * manually completed.  At the end of a successful mirroring job,
1284  * @bs will be switched to read from @target.
1285  */
1286 void mirror_start(const char *job_id, BlockDriverState *bs,
1287                   BlockDriverState *target, const char *replaces,
1288                   int creation_flags, int64_t speed,
1289                   uint32_t granularity, int64_t buf_size,
1290                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
1291                   bool zero_target,
1292                   BlockdevOnError on_source_error,
1293                   BlockdevOnError on_target_error,
1294                   bool unmap, const char *filter_node_name,
1295                   MirrorCopyMode copy_mode, Error **errp);
1296 
1297 /*
1298  * backup_job_create:
1299  * @job_id: The id of the newly-created job, or %NULL to use the
1300  * device name of @bs.
1301  * @bs: Block device to operate on.
1302  * @target: Block device to write to.
1303  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
1304  * @sync_mode: What parts of the disk image should be copied to the destination.
1305  * @sync_bitmap: The dirty bitmap if sync_mode is 'bitmap' or 'incremental'
1306  * @bitmap_mode: The bitmap synchronization policy to use.
1307  * @perf: Performance options. All actual fields assumed to be present,
1308  *        all ".has_*" fields are ignored.
1309  * @on_source_error: The action to take upon error reading from the source.
1310  * @on_target_error: The action to take upon error writing to the target.
1311  * @creation_flags: Flags that control the behavior of the Job lifetime.
1312  *                  See @BlockJobCreateFlags
1313  * @cb: Completion function for the job.
1314  * @opaque: Opaque pointer value passed to @cb.
1315  * @txn: Transaction that this job is part of (may be NULL).
1316  *
1317  * Create a backup operation on @bs.  Clusters in @bs are written to @target
1318  * until the job is cancelled or manually completed.
1319  */
1320 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
1321                             BlockDriverState *target, int64_t speed,
1322                             MirrorSyncMode sync_mode,
1323                             BdrvDirtyBitmap *sync_bitmap,
1324                             BitmapSyncMode bitmap_mode,
1325                             bool compress,
1326                             const char *filter_node_name,
1327                             BackupPerf *perf,
1328                             BlockdevOnError on_source_error,
1329                             BlockdevOnError on_target_error,
1330                             int creation_flags,
1331                             BlockCompletionFunc *cb, void *opaque,
1332                             JobTxn *txn, Error **errp);
1333 
1334 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1335                                   const char *child_name,
1336                                   const BdrvChildClass *child_class,
1337                                   BdrvChildRole child_role,
1338                                   uint64_t perm, uint64_t shared_perm,
1339                                   void *opaque, Error **errp);
1340 void bdrv_root_unref_child(BdrvChild *child);
1341 
1342 void bdrv_get_cumulative_perm(BlockDriverState *bs, uint64_t *perm,
1343                               uint64_t *shared_perm);
1344 
1345 /**
1346  * Sets a BdrvChild's permissions.  Avoid if the parent is a BDS; use
1347  * bdrv_child_refresh_perms() instead and make the parent's
1348  * .bdrv_child_perm() implementation return the correct values.
1349  */
1350 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1351                             Error **errp);
1352 
1353 /**
1354  * Calls bs->drv->bdrv_child_perm() and updates the child's permission
1355  * masks with the result.
1356  * Drivers should invoke this function whenever an event occurs that
1357  * makes their .bdrv_child_perm() implementation return different
1358  * values than before, but which will not result in the block layer
1359  * automatically refreshing the permissions.
1360  */
1361 int bdrv_child_refresh_perms(BlockDriverState *bs, BdrvChild *c, Error **errp);
1362 
1363 bool bdrv_recurse_can_replace(BlockDriverState *bs,
1364                               BlockDriverState *to_replace);
1365 
1366 /*
1367  * Default implementation for BlockDriver.bdrv_child_perm() that can
1368  * be used by block filters and image formats, as long as they use the
1369  * child_of_bds child class and set an appropriate BdrvChildRole.
1370  */
1371 void bdrv_default_perms(BlockDriverState *bs, BdrvChild *c,
1372                         BdrvChildRole role, BlockReopenQueue *reopen_queue,
1373                         uint64_t perm, uint64_t shared,
1374                         uint64_t *nperm, uint64_t *nshared);
1375 
1376 const char *bdrv_get_parent_name(const BlockDriverState *bs);
1377 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
1378 bool blk_dev_has_removable_media(BlockBackend *blk);
1379 bool blk_dev_has_tray(BlockBackend *blk);
1380 void blk_dev_eject_request(BlockBackend *blk, bool force);
1381 bool blk_dev_is_tray_open(BlockBackend *blk);
1382 bool blk_dev_is_medium_locked(BlockBackend *blk);
1383 
1384 void bdrv_set_dirty(BlockDriverState *bs, int64_t offset, int64_t bytes);
1385 
1386 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
1387 void bdrv_restore_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *backup);
1388 bool bdrv_dirty_bitmap_merge_internal(BdrvDirtyBitmap *dest,
1389                                       const BdrvDirtyBitmap *src,
1390                                       HBitmap **backup, bool lock);
1391 
1392 void bdrv_inc_in_flight(BlockDriverState *bs);
1393 void bdrv_dec_in_flight(BlockDriverState *bs);
1394 
1395 void blockdev_close_all_bdrv_states(void);
1396 
1397 int coroutine_fn bdrv_co_copy_range_from(BdrvChild *src, int64_t src_offset,
1398                                          BdrvChild *dst, int64_t dst_offset,
1399                                          int64_t bytes,
1400                                          BdrvRequestFlags read_flags,
1401                                          BdrvRequestFlags write_flags);
1402 int coroutine_fn bdrv_co_copy_range_to(BdrvChild *src, int64_t src_offset,
1403                                        BdrvChild *dst, int64_t dst_offset,
1404                                        int64_t bytes,
1405                                        BdrvRequestFlags read_flags,
1406                                        BdrvRequestFlags write_flags);
1407 
1408 int refresh_total_sectors(BlockDriverState *bs, int64_t hint);
1409 
1410 void bdrv_set_monitor_owned(BlockDriverState *bs);
1411 BlockDriverState *bds_tree_init(QDict *bs_opts, Error **errp);
1412 
1413 /**
1414  * Simple implementation of bdrv_co_create_opts for protocol drivers
1415  * which only support creation via opening a file
1416  * (usually existing raw storage device)
1417  */
1418 int coroutine_fn bdrv_co_create_opts_simple(BlockDriver *drv,
1419                                             const char *filename,
1420                                             QemuOpts *opts,
1421                                             Error **errp);
1422 extern QemuOptsList bdrv_create_opts_simple;
1423 
1424 BdrvDirtyBitmap *block_dirty_bitmap_lookup(const char *node,
1425                                            const char *name,
1426                                            BlockDriverState **pbs,
1427                                            Error **errp);
1428 BdrvDirtyBitmap *block_dirty_bitmap_merge(const char *node, const char *target,
1429                                           BlockDirtyBitmapMergeSourceList *bms,
1430                                           HBitmap **backup, Error **errp);
1431 BdrvDirtyBitmap *block_dirty_bitmap_remove(const char *node, const char *name,
1432                                            bool release,
1433                                            BlockDriverState **bitmap_bs,
1434                                            Error **errp);
1435 
1436 BdrvChild *bdrv_cow_child(BlockDriverState *bs);
1437 BdrvChild *bdrv_filter_child(BlockDriverState *bs);
1438 BdrvChild *bdrv_filter_or_cow_child(BlockDriverState *bs);
1439 BdrvChild *bdrv_primary_child(BlockDriverState *bs);
1440 BlockDriverState *bdrv_skip_implicit_filters(BlockDriverState *bs);
1441 BlockDriverState *bdrv_skip_filters(BlockDriverState *bs);
1442 BlockDriverState *bdrv_backing_chain_next(BlockDriverState *bs);
1443 
1444 static inline BlockDriverState *child_bs(BdrvChild *child)
1445 {
1446     return child ? child->bs : NULL;
1447 }
1448 
1449 static inline BlockDriverState *bdrv_cow_bs(BlockDriverState *bs)
1450 {
1451     return child_bs(bdrv_cow_child(bs));
1452 }
1453 
1454 static inline BlockDriverState *bdrv_filter_bs(BlockDriverState *bs)
1455 {
1456     return child_bs(bdrv_filter_child(bs));
1457 }
1458 
1459 static inline BlockDriverState *bdrv_filter_or_cow_bs(BlockDriverState *bs)
1460 {
1461     return child_bs(bdrv_filter_or_cow_child(bs));
1462 }
1463 
1464 static inline BlockDriverState *bdrv_primary_bs(BlockDriverState *bs)
1465 {
1466     return child_bs(bdrv_primary_child(bs));
1467 }
1468 
1469 /**
1470  * End all quiescent sections started by bdrv_drain_all_begin(). This is
1471  * needed when deleting a BDS before bdrv_drain_all_end() is called.
1472  *
1473  * NOTE: this is an internal helper for bdrv_close() *only*. No one else
1474  * should call it.
1475  */
1476 void bdrv_drain_all_end_quiesce(BlockDriverState *bs);
1477 
1478 /**
1479  * Check whether the given offset is in the cached block-status data
1480  * region.
1481  *
1482  * If it is, and @pnum is not NULL, *pnum is set to
1483  * `bsc.data_end - offset`, i.e. how many bytes, starting from
1484  * @offset, are data (according to the cache).
1485  * Otherwise, *pnum is not touched.
1486  */
1487 bool bdrv_bsc_is_data(BlockDriverState *bs, int64_t offset, int64_t *pnum);
1488 
1489 /**
1490  * If [offset, offset + bytes) overlaps with the currently cached
1491  * block-status region, invalidate the cache.
1492  *
1493  * (To be used by I/O paths that cause data regions to be zero or
1494  * holes.)
1495  */
1496 void bdrv_bsc_invalidate_range(BlockDriverState *bs,
1497                                int64_t offset, int64_t bytes);
1498 
1499 /**
1500  * Mark the range [offset, offset + bytes) as a data region.
1501  */
1502 void bdrv_bsc_fill(BlockDriverState *bs, int64_t offset, int64_t bytes);
1503 
1504 #endif /* BLOCK_INT_H */
1505