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_COMMON_H
25 #define BLOCK_INT_COMMON_H
26 
27 #include "block/aio.h"
28 #include "block/block-common.h"
29 #include "block/block-global-state.h"
30 #include "block/snapshot.h"
31 #include "qemu/iov.h"
32 #include "qemu/rcu.h"
33 #include "qemu/stats64.h"
34 
35 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
36 
37 #define BLOCK_OPT_SIZE              "size"
38 #define BLOCK_OPT_ENCRYPT           "encryption"
39 #define BLOCK_OPT_ENCRYPT_FORMAT    "encrypt.format"
40 #define BLOCK_OPT_COMPAT6           "compat6"
41 #define BLOCK_OPT_HWVERSION         "hwversion"
42 #define BLOCK_OPT_BACKING_FILE      "backing_file"
43 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
44 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
45 #define BLOCK_OPT_TABLE_SIZE        "table_size"
46 #define BLOCK_OPT_PREALLOC          "preallocation"
47 #define BLOCK_OPT_SUBFMT            "subformat"
48 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
49 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
50 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
51 #define BLOCK_OPT_REDUNDANCY        "redundancy"
52 #define BLOCK_OPT_NOCOW             "nocow"
53 #define BLOCK_OPT_EXTENT_SIZE_HINT  "extent_size_hint"
54 #define BLOCK_OPT_OBJECT_SIZE       "object_size"
55 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
56 #define BLOCK_OPT_DATA_FILE         "data_file"
57 #define BLOCK_OPT_DATA_FILE_RAW     "data_file_raw"
58 #define BLOCK_OPT_COMPRESSION_TYPE  "compression_type"
59 #define BLOCK_OPT_EXTL2             "extended_l2"
60 
61 #define BLOCK_PROBE_BUF_SIZE        512
62 
63 enum BdrvTrackedRequestType {
64     BDRV_TRACKED_READ,
65     BDRV_TRACKED_WRITE,
66     BDRV_TRACKED_DISCARD,
67     BDRV_TRACKED_TRUNCATE,
68 };
69 
70 /*
71  * That is not quite good that BdrvTrackedRequest structure is public,
72  * as block/io.c is very careful about incoming offset/bytes being
73  * correct. Be sure to assert bdrv_check_request() succeeded after any
74  * modification of BdrvTrackedRequest object out of block/io.c
75  */
76 typedef struct BdrvTrackedRequest {
77     BlockDriverState *bs;
78     int64_t offset;
79     int64_t bytes;
80     enum BdrvTrackedRequestType type;
81 
82     bool serialising;
83     int64_t overlap_offset;
84     int64_t overlap_bytes;
85 
86     QLIST_ENTRY(BdrvTrackedRequest) list;
87     Coroutine *co; /* owner, used for deadlock detection */
88     CoQueue wait_queue; /* coroutines blocked on this request */
89 
90     struct BdrvTrackedRequest *waiting_for;
91 } BdrvTrackedRequest;
92 
93 
94 struct BlockDriver {
95     /*
96      * These fields are initialized when this object is created,
97      * and are never changed afterwards.
98      */
99 
100     const char *format_name;
101     int instance_size;
102 
103     /*
104      * Set to true if the BlockDriver is a block filter. Block filters pass
105      * certain callbacks that refer to data (see block.c) to their bs->file
106      * or bs->backing (whichever one exists) if the driver doesn't implement
107      * them. Drivers that do not wish to forward must implement them and return
108      * -ENOTSUP.
109      * Note that filters are not allowed to modify data.
110      *
111      * Filters generally cannot have more than a single filtered child,
112      * because the data they present must at all times be the same as
113      * that on their filtered child.  That would be impossible to
114      * achieve for multiple filtered children.
115      * (And this filtered child must then be bs->file or bs->backing.)
116      */
117     bool is_filter;
118     /*
119      * Only make sense for filter drivers, for others must be false.
120      * If true, filtered child is bs->backing. Otherwise it's bs->file.
121      * Two internal filters use bs->backing as filtered child and has this
122      * field set to true: mirror_top and commit_top. There also two such test
123      * filters in tests/unit/test-bdrv-graph-mod.c.
124      *
125      * Never create any more such filters!
126      *
127      * TODO: imagine how to deprecate this behavior and make all filters work
128      * similarly using bs->file as filtered child.
129      */
130     bool filtered_child_is_backing;
131 
132     /*
133      * Set to true if the BlockDriver is a format driver.  Format nodes
134      * generally do not expect their children to be other format nodes
135      * (except for backing files), and so format probing is disabled
136      * on those children.
137      */
138     bool is_format;
139 
140     /*
141      * Set to true if the BlockDriver supports zoned children.
142      */
143     bool supports_zoned_children;
144 
145     /*
146      * Drivers not implementing bdrv_parse_filename nor bdrv_open should have
147      * this field set to true, except ones that are defined only by their
148      * child's bs.
149      * An example of the last type will be the quorum block driver.
150      */
151     bool bdrv_needs_filename;
152 
153     /*
154      * Set if a driver can support backing files. This also implies the
155      * following semantics:
156      *
157      *  - Return status 0 of .bdrv_co_block_status means that corresponding
158      *    blocks are not allocated in this layer of backing-chain
159      *  - For such (unallocated) blocks, read will:
160      *    - fill buffer with zeros if there is no backing file
161      *    - read from the backing file otherwise, where the block layer
162      *      takes care of reading zeros beyond EOF if backing file is short
163      */
164     bool supports_backing;
165 
166     /*
167      * Drivers setting this field must be able to work with just a plain
168      * filename with '<protocol_name>:' as a prefix, and no other options.
169      * Options may be extracted from the filename by implementing
170      * bdrv_parse_filename.
171      */
172     const char *protocol_name;
173 
174     /* List of options for creating images, terminated by name == NULL */
175     QemuOptsList *create_opts;
176 
177     /* List of options for image amend */
178     QemuOptsList *amend_opts;
179 
180     /*
181      * If this driver supports reopening images this contains a
182      * NULL-terminated list of the runtime options that can be
183      * modified. If an option in this list is unspecified during
184      * reopen then it _must_ be reset to its default value or return
185      * an error.
186      */
187     const char *const *mutable_opts;
188 
189     /*
190      * Pointer to a NULL-terminated array of names of strong options
191      * that can be specified for bdrv_open(). A strong option is one
192      * that changes the data of a BDS.
193      * If this pointer is NULL, the array is considered empty.
194      * "filename" and "driver" are always considered strong.
195      */
196     const char *const *strong_runtime_opts;
197 
198 
199     /*
200      * Global state (GS) API. These functions run under the BQL.
201      *
202      * See include/block/block-global-state.h for more information about
203      * the GS API.
204      */
205 
206     /*
207      * This function is invoked under BQL before .bdrv_co_amend()
208      * (which in contrast does not necessarily run under the BQL)
209      * to allow driver-specific initialization code that requires
210      * the BQL, like setting up specific permission flags.
211      */
212     int GRAPH_RDLOCK_PTR (*bdrv_amend_pre_run)(
213         BlockDriverState *bs, Error **errp);
214     /*
215      * This function is invoked under BQL after .bdrv_co_amend()
216      * to allow cleaning up what was done in .bdrv_amend_pre_run().
217      */
218     void GRAPH_RDLOCK_PTR (*bdrv_amend_clean)(BlockDriverState *bs);
219 
220     /*
221      * Return true if @to_replace can be replaced by a BDS with the
222      * same data as @bs without it affecting @bs's behavior (that is,
223      * without it being visible to @bs's parents).
224      */
225     bool GRAPH_RDLOCK_PTR (*bdrv_recurse_can_replace)(
226         BlockDriverState *bs, BlockDriverState *to_replace);
227 
228     int (*bdrv_probe_device)(const char *filename);
229 
230     /*
231      * Any driver implementing this callback is expected to be able to handle
232      * NULL file names in its .bdrv_open() implementation.
233      */
234     void (*bdrv_parse_filename)(const char *filename, QDict *options,
235                                 Error **errp);
236 
237     /* For handling image reopen for split or non-split files. */
238     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
239                                BlockReopenQueue *queue, Error **errp);
240     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
241     void (*bdrv_reopen_commit_post)(BDRVReopenState *reopen_state);
242     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
243     void (*bdrv_join_options)(QDict *options, QDict *old_options);
244 
245     int GRAPH_UNLOCKED_PTR (*bdrv_open)(
246         BlockDriverState *bs, QDict *options, int flags, Error **errp);
247 
248     /* Protocol drivers should implement this instead of bdrv_open */
249     int GRAPH_UNLOCKED_PTR (*bdrv_file_open)(
250         BlockDriverState *bs, QDict *options, int flags, Error **errp);
251     void (*bdrv_close)(BlockDriverState *bs);
252 
253     int coroutine_fn GRAPH_UNLOCKED_PTR (*bdrv_co_create)(
254         BlockdevCreateOptions *opts, Error **errp);
255 
256     int coroutine_fn GRAPH_UNLOCKED_PTR (*bdrv_co_create_opts)(
257         BlockDriver *drv, const char *filename, QemuOpts *opts, Error **errp);
258 
259     int (*bdrv_amend_options)(BlockDriverState *bs,
260                               QemuOpts *opts,
261                               BlockDriverAmendStatusCB *status_cb,
262                               void *cb_opaque,
263                               bool force,
264                               Error **errp);
265 
266     int (*bdrv_make_empty)(BlockDriverState *bs);
267 
268     /*
269      * Refreshes the bs->exact_filename field. If that is impossible,
270      * bs->exact_filename has to be left empty.
271      */
272     void (*bdrv_refresh_filename)(BlockDriverState *bs);
273 
274     /*
275      * Gathers the open options for all children into @target.
276      * A simple format driver (without backing file support) might
277      * implement this function like this:
278      *
279      *     QINCREF(bs->file->bs->full_open_options);
280      *     qdict_put(target, "file", bs->file->bs->full_open_options);
281      *
282      * If not specified, the generic implementation will simply put
283      * all children's options under their respective name.
284      *
285      * @backing_overridden is true when bs->backing seems not to be
286      * the child that would result from opening bs->backing_file.
287      * Therefore, if it is true, the backing child's options should be
288      * gathered; otherwise, there is no need since the backing child
289      * is the one implied by the image header.
290      *
291      * Note that ideally this function would not be needed.  Every
292      * block driver which implements it is probably doing something
293      * shady regarding its runtime option structure.
294      */
295     void (*bdrv_gather_child_options)(BlockDriverState *bs, QDict *target,
296                                       bool backing_overridden);
297 
298     /*
299      * Returns an allocated string which is the directory name of this BDS: It
300      * will be used to make relative filenames absolute by prepending this
301      * function's return value to them.
302      */
303     char *(*bdrv_dirname)(BlockDriverState *bs, Error **errp);
304 
305     /*
306      * This informs the driver that we are no longer interested in the result
307      * of in-flight requests, so don't waste the time if possible.
308      *
309      * One example usage is to avoid waiting for an nbd target node reconnect
310      * timeout during job-cancel with force=true.
311      */
312     void (*bdrv_cancel_in_flight)(BlockDriverState *bs);
313 
314     int GRAPH_RDLOCK_PTR (*bdrv_inactivate)(BlockDriverState *bs);
315 
316     int (*bdrv_snapshot_create)(BlockDriverState *bs,
317                                 QEMUSnapshotInfo *sn_info);
318     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
319                               const char *snapshot_id);
320     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
321                                 const char *snapshot_id,
322                                 const char *name,
323                                 Error **errp);
324     int (*bdrv_snapshot_list)(BlockDriverState *bs,
325                               QEMUSnapshotInfo **psn_info);
326     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
327                                   const char *snapshot_id,
328                                   const char *name,
329                                   Error **errp);
330 
331     int (*bdrv_change_backing_file)(BlockDriverState *bs,
332         const char *backing_file, const char *backing_fmt);
333 
334     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
335     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
336         const char *tag);
337     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
338         const char *tag);
339     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
340     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
341 
342     void GRAPH_RDLOCK_PTR (*bdrv_refresh_limits)(
343         BlockDriverState *bs, Error **errp);
344 
345     /*
346      * Returns 1 if newly created images are guaranteed to contain only
347      * zeros, 0 otherwise.
348      */
349     int (*bdrv_has_zero_init)(BlockDriverState *bs);
350 
351     /*
352      * Remove fd handlers, timers, and other event loop callbacks so the event
353      * loop is no longer in use.  Called with no in-flight requests and in
354      * depth-first traversal order with parents before child nodes.
355      */
356     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
357 
358     /*
359      * Add fd handlers, timers, and other event loop callbacks so I/O requests
360      * can be processed again.  Called with no in-flight requests and in
361      * depth-first traversal order with child nodes before parent nodes.
362      */
363     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
364                                     AioContext *new_context);
365 
366     /**
367      * bdrv_drain_begin is called if implemented in the beginning of a
368      * drain operation to drain and stop any internal sources of requests in
369      * the driver.
370      * bdrv_drain_end is called if implemented at the end of the drain.
371      *
372      * They should be used by the driver to e.g. manage scheduled I/O
373      * requests, or toggle an internal state. After the end of the drain new
374      * requests will continue normally.
375      *
376      * Implementations of both functions must not call aio_poll().
377      */
378     void (*bdrv_drain_begin)(BlockDriverState *bs);
379     void (*bdrv_drain_end)(BlockDriverState *bs);
380 
381     /**
382      * Try to get @bs's logical and physical block size.
383      * On success, store them in @bsz and return zero.
384      * On failure, return negative errno.
385      */
386     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
387     /**
388      * Try to get @bs's geometry (cyls, heads, sectors)
389      * On success, store them in @geo and return 0.
390      * On failure return -errno.
391      * Only drivers that want to override guest geometry implement this
392      * callback; see hd_geometry_guess().
393      */
394     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
395 
396     void GRAPH_WRLOCK_PTR (*bdrv_add_child)(
397         BlockDriverState *parent, BlockDriverState *child, Error **errp);
398 
399     void GRAPH_WRLOCK_PTR (*bdrv_del_child)(
400         BlockDriverState *parent, BdrvChild *child, Error **errp);
401 
402     /**
403      * Informs the block driver that a permission change is intended. The
404      * driver checks whether the change is permissible and may take other
405      * preparations for the change (e.g. get file system locks). This operation
406      * is always followed either by a call to either .bdrv_set_perm or
407      * .bdrv_abort_perm_update.
408      *
409      * Checks whether the requested set of cumulative permissions in @perm
410      * can be granted for accessing @bs and whether no other users are using
411      * permissions other than those given in @shared (both arguments take
412      * BLK_PERM_* bitmasks).
413      *
414      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
415      * and errp is set to an error describing the conflict.
416      */
417     int GRAPH_RDLOCK_PTR (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
418                                             uint64_t shared, Error **errp);
419 
420     /**
421      * Called to inform the driver that the set of cumulative set of used
422      * permissions for @bs has changed to @perm, and the set of shareable
423      * permission to @shared. The driver can use this to propagate changes to
424      * its children (i.e. request permissions only if a parent actually needs
425      * them).
426      *
427      * This function is only invoked after bdrv_check_perm(), so block drivers
428      * may rely on preparations made in their .bdrv_check_perm implementation.
429      */
430     void GRAPH_RDLOCK_PTR (*bdrv_set_perm)(
431         BlockDriverState *bs, uint64_t perm, uint64_t shared);
432 
433     /*
434      * Called to inform the driver that after a previous bdrv_check_perm()
435      * call, the permission update is not performed and any preparations made
436      * for it (e.g. taken file locks) need to be undone.
437      *
438      * This function can be called even for nodes that never saw a
439      * bdrv_check_perm() call. It is a no-op then.
440      */
441     void GRAPH_RDLOCK_PTR (*bdrv_abort_perm_update)(BlockDriverState *bs);
442 
443     /**
444      * Returns in @nperm and @nshared the permissions that the driver for @bs
445      * needs on its child @c, based on the cumulative permissions requested by
446      * the parents in @parent_perm and @parent_shared.
447      *
448      * If @c is NULL, return the permissions for attaching a new child for the
449      * given @child_class and @role.
450      *
451      * If @reopen_queue is non-NULL, don't return the currently needed
452      * permissions, but those that will be needed after applying the
453      * @reopen_queue.
454      */
455      void GRAPH_RDLOCK_PTR (*bdrv_child_perm)(
456         BlockDriverState *bs, BdrvChild *c, BdrvChildRole role,
457         BlockReopenQueue *reopen_queue,
458         uint64_t parent_perm, uint64_t parent_shared,
459         uint64_t *nperm, uint64_t *nshared);
460 
461     /**
462      * Register/unregister a buffer for I/O. For example, when the driver is
463      * interested to know the memory areas that will later be used in iovs, so
464      * that it can do IOMMU mapping with VFIO etc., in order to get better
465      * performance. In the case of VFIO drivers, this callback is used to do
466      * DMA mapping for hot buffers.
467      *
468      * Returns: true on success, false on failure
469      */
470     bool GRAPH_RDLOCK_PTR (*bdrv_register_buf)(
471         BlockDriverState *bs, void *host, size_t size, Error **errp);
472     void GRAPH_RDLOCK_PTR (*bdrv_unregister_buf)(
473         BlockDriverState *bs, void *host, size_t size);
474 
475     /*
476      * This field is modified only under the BQL, and is part of
477      * the global state.
478      */
479     QLIST_ENTRY(BlockDriver) list;
480 
481     /*
482      * I/O API functions. These functions are thread-safe.
483      *
484      * See include/block/block-io.h for more information about
485      * the I/O API.
486      */
487 
488     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
489 
490     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_amend)(
491         BlockDriverState *bs, BlockdevAmendOptions *opts, bool force,
492         Error **errp);
493 
494     /* aio */
495     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_preadv)(BlockDriverState *bs,
496         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
497         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
498 
499     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_pwritev)(BlockDriverState *bs,
500         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
501         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
502 
503     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_flush)(
504         BlockDriverState *bs, BlockCompletionFunc *cb, void *opaque);
505 
506     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_pdiscard)(
507         BlockDriverState *bs, int64_t offset, int bytes,
508         BlockCompletionFunc *cb, void *opaque);
509 
510     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_readv)(BlockDriverState *bs,
511         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
512 
513     /**
514      * @offset: position in bytes to read at
515      * @bytes: number of bytes to read
516      * @qiov: the buffers to fill with read data
517      * @flags: currently unused, always 0
518      *
519      * @offset and @bytes will be a multiple of 'request_alignment',
520      * but the length of individual @qiov elements does not have to
521      * be a multiple.
522      *
523      * @bytes will always equal the total size of @qiov, and will be
524      * no larger than 'max_transfer'.
525      *
526      * The buffer in @qiov may point directly to guest memory.
527      */
528     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv)(BlockDriverState *bs,
529         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
530         BdrvRequestFlags flags);
531 
532     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_part)(
533         BlockDriverState *bs, int64_t offset, int64_t bytes,
534         QEMUIOVector *qiov, size_t qiov_offset,
535         BdrvRequestFlags flags);
536 
537     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_writev)(BlockDriverState *bs,
538         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
539         int flags);
540     /**
541      * @offset: position in bytes to write at
542      * @bytes: number of bytes to write
543      * @qiov: the buffers containing data to write
544      * @flags: zero or more bits allowed by 'supported_write_flags'
545      *
546      * @offset and @bytes will be a multiple of 'request_alignment',
547      * but the length of individual @qiov elements does not have to
548      * be a multiple.
549      *
550      * @bytes will always equal the total size of @qiov, and will be
551      * no larger than 'max_transfer'.
552      *
553      * The buffer in @qiov may point directly to guest memory.
554      */
555     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev)(
556         BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov,
557         BdrvRequestFlags flags);
558     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_part)(
559         BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov,
560         size_t qiov_offset, BdrvRequestFlags flags);
561 
562     /*
563      * Efficiently zero a region of the disk image.  Typically an image format
564      * would use a compact metadata representation to implement this.  This
565      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
566      * will be called instead.
567      */
568     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwrite_zeroes)(
569         BlockDriverState *bs, int64_t offset, int64_t bytes,
570         BdrvRequestFlags flags);
571 
572     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard)(
573         BlockDriverState *bs, int64_t offset, int64_t bytes);
574 
575     /*
576      * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
577      * and invoke bdrv_co_copy_range_from(child, ...), or invoke
578      * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
579      *
580      * See the comment of bdrv_co_copy_range for the parameter and return value
581      * semantics.
582      */
583     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_from)(
584         BlockDriverState *bs, BdrvChild *src, int64_t offset,
585         BdrvChild *dst, int64_t dst_offset, int64_t bytes,
586         BdrvRequestFlags read_flags, BdrvRequestFlags write_flags);
587 
588     /*
589      * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
590      * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
591      * operation if @bs is the leaf and @src has the same BlockDriver.  Return
592      * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
593      *
594      * See the comment of bdrv_co_copy_range for the parameter and return value
595      * semantics.
596      */
597     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_to)(
598         BlockDriverState *bs, BdrvChild *src, int64_t src_offset,
599         BdrvChild *dst, int64_t dst_offset, int64_t bytes,
600         BdrvRequestFlags read_flags, BdrvRequestFlags write_flags);
601 
602     /*
603      * Building block for bdrv_block_status[_above] and
604      * bdrv_is_allocated[_above].  The driver should answer only
605      * according to the current layer, and should only need to set
606      * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
607      * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
608      * layer, the result should be 0 (and not BDRV_BLOCK_ZERO).  See
609      * block.h for the overall meaning of the bits.  As a hint, the
610      * flag want_zero is true if the caller cares more about precise
611      * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
612      * overall allocation (favor larger *pnum, perhaps by reporting
613      * _DATA instead of _ZERO).  The block layer guarantees input
614      * clamped to bdrv_getlength() and aligned to request_alignment,
615      * as well as non-NULL pnum, map, and file; in turn, the driver
616      * must return an error or set pnum to an aligned non-zero value.
617      *
618      * Note that @bytes is just a hint on how big of a region the
619      * caller wants to inspect.  It is not a limit on *pnum.
620      * Implementations are free to return larger values of *pnum if
621      * doing so does not incur a performance penalty.
622      *
623      * block/io.c's bdrv_co_block_status() will utilize an unclamped
624      * *pnum value for the block-status cache on protocol nodes, prior
625      * to clamping *pnum for return to its caller.
626      */
627     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_block_status)(
628         BlockDriverState *bs,
629         bool want_zero, int64_t offset, int64_t bytes, int64_t *pnum,
630         int64_t *map, BlockDriverState **file);
631 
632     /*
633      * Snapshot-access API.
634      *
635      * Block-driver may provide snapshot-access API: special functions to access
636      * some internal "snapshot". The functions are similar with normal
637      * read/block_status/discard handler, but don't have any specific handling
638      * in generic block-layer: no serializing, no alignment, no tracked
639      * requests. So, block-driver that realizes these APIs is fully responsible
640      * for synchronization between snapshot-access API and normal IO requests.
641      *
642      * TODO: To be able to support qcow2's internal snapshots, this API will
643      * need to be extended to:
644      * - be able to select a specific snapshot
645      * - receive the snapshot's actual length (which may differ from bs's
646      *   length)
647      */
648     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_snapshot)(
649         BlockDriverState *bs, int64_t offset, int64_t bytes,
650         QEMUIOVector *qiov, size_t qiov_offset);
651 
652     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_snapshot_block_status)(
653         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
654         int64_t *pnum, int64_t *map, BlockDriverState **file);
655 
656     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard_snapshot)(
657         BlockDriverState *bs, int64_t offset, int64_t bytes);
658 
659     /*
660      * Invalidate any cached meta-data.
661      */
662     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_invalidate_cache)(
663         BlockDriverState *bs, Error **errp);
664 
665     /*
666      * Flushes all data for all layers by calling bdrv_co_flush for underlying
667      * layers, if needed. This function is needed for deterministic
668      * synchronization of the flush finishing callback.
669      */
670     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush)(BlockDriverState *bs);
671 
672     /* Delete a created file. */
673     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_delete_file)(
674         BlockDriverState *bs, Error **errp);
675 
676     /*
677      * Flushes all data that was already written to the OS all the way down to
678      * the disk (for example file-posix.c calls fsync()).
679      */
680     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_disk)(
681         BlockDriverState *bs);
682 
683     /*
684      * Flushes all internal caches to the OS. The data may still sit in a
685      * writeback cache of the host OS, but it will survive a crash of the qemu
686      * process.
687      */
688     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_os)(
689         BlockDriverState *bs);
690 
691     /*
692      * Truncate @bs to @offset bytes using the given @prealloc mode
693      * when growing.  Modes other than PREALLOC_MODE_OFF should be
694      * rejected when shrinking @bs.
695      *
696      * If @exact is true, @bs must be resized to exactly @offset.
697      * Otherwise, it is sufficient for @bs (if it is a host block
698      * device and thus there is no way to resize it) to be at least
699      * @offset bytes in length.
700      *
701      * If @exact is true and this function fails but would succeed
702      * with @exact = false, it should return -ENOTSUP.
703      */
704     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_truncate)(
705         BlockDriverState *bs, int64_t offset, bool exact,
706         PreallocMode prealloc, BdrvRequestFlags flags, Error **errp);
707 
708     int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_getlength)(
709         BlockDriverState *bs);
710 
711     int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_allocated_file_size)(
712         BlockDriverState *bs);
713 
714     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
715                                       Error **errp);
716 
717     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed)(
718         BlockDriverState *bs, int64_t offset, int64_t bytes,
719         QEMUIOVector *qiov);
720 
721     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed_part)(
722         BlockDriverState *bs, int64_t offset, int64_t bytes,
723         QEMUIOVector *qiov, size_t qiov_offset);
724 
725     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_info)(
726         BlockDriverState *bs, BlockDriverInfo *bdi);
727 
728     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs,
729                                                  Error **errp);
730     BlockStatsSpecific *(*bdrv_get_specific_stats)(BlockDriverState *bs);
731 
732     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_save_vmstate)(
733         BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
734 
735     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_load_vmstate)(
736         BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
737 
738     int coroutine_fn (*bdrv_co_zone_report)(BlockDriverState *bs,
739             int64_t offset, unsigned int *nr_zones,
740             BlockZoneDescriptor *zones);
741     int coroutine_fn (*bdrv_co_zone_mgmt)(BlockDriverState *bs, BlockZoneOp op,
742             int64_t offset, int64_t len);
743     int coroutine_fn (*bdrv_co_zone_append)(BlockDriverState *bs,
744             int64_t *offset, QEMUIOVector *qiov,
745             BdrvRequestFlags flags);
746 
747     /* removable device specific */
748     bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_is_inserted)(
749         BlockDriverState *bs);
750     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_eject)(
751         BlockDriverState *bs, bool eject_flag);
752     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_lock_medium)(
753         BlockDriverState *bs, bool locked);
754 
755     /* to control generic scsi devices */
756     BlockAIOCB *coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_aio_ioctl)(
757         BlockDriverState *bs, unsigned long int req, void *buf,
758         BlockCompletionFunc *cb, void *opaque);
759 
760     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_ioctl)(
761         BlockDriverState *bs, unsigned long int req, void *buf);
762 
763     /*
764      * Returns 0 for completed check, -errno for internal errors.
765      * The check results are stored in result.
766      */
767     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_check)(
768         BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix);
769 
770     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_debug_event)(
771         BlockDriverState *bs, BlkdebugEvent event);
772 
773     bool (*bdrv_supports_persistent_dirty_bitmap)(BlockDriverState *bs);
774 
775     bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_can_store_new_dirty_bitmap)(
776         BlockDriverState *bs, const char *name, uint32_t granularity,
777         Error **errp);
778 
779     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_remove_persistent_dirty_bitmap)(
780         BlockDriverState *bs, const char *name, Error **errp);
781 };
782 
783 static inline bool TSA_NO_TSA block_driver_can_compress(BlockDriver *drv)
784 {
785     return drv->bdrv_co_pwritev_compressed ||
786            drv->bdrv_co_pwritev_compressed_part;
787 }
788 
789 typedef struct BlockLimits {
790     /*
791      * Alignment requirement, in bytes, for offset/length of I/O
792      * requests. Must be a power of 2 less than INT_MAX; defaults to
793      * 1 for drivers with modern byte interfaces, and to 512
794      * otherwise.
795      */
796     uint32_t request_alignment;
797 
798     /*
799      * Maximum number of bytes that can be discarded at once. Must be multiple
800      * of pdiscard_alignment, but need not be power of 2. May be 0 if no
801      * inherent 64-bit limit.
802      */
803     int64_t max_pdiscard;
804 
805     /*
806      * Optimal alignment for discard requests in bytes. A power of 2
807      * is best but not mandatory.  Must be a multiple of
808      * bl.request_alignment, and must be less than max_pdiscard if
809      * that is set. May be 0 if bl.request_alignment is good enough
810      */
811     uint32_t pdiscard_alignment;
812 
813     /*
814      * Maximum number of bytes that can zeroized at once. Must be multiple of
815      * pwrite_zeroes_alignment. 0 means no limit.
816      */
817     int64_t max_pwrite_zeroes;
818 
819     /*
820      * Optimal alignment for write zeroes requests in bytes. A power
821      * of 2 is best but not mandatory.  Must be a multiple of
822      * bl.request_alignment, and must be less than max_pwrite_zeroes
823      * if that is set. May be 0 if bl.request_alignment is good
824      * enough
825      */
826     uint32_t pwrite_zeroes_alignment;
827 
828     /*
829      * Optimal transfer length in bytes.  A power of 2 is best but not
830      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
831      * no preferred size
832      */
833     uint32_t opt_transfer;
834 
835     /*
836      * Maximal transfer length in bytes.  Need not be power of 2, but
837      * must be multiple of opt_transfer and bl.request_alignment, or 0
838      * for no 32-bit limit.  For now, anything larger than INT_MAX is
839      * clamped down.
840      */
841     uint32_t max_transfer;
842 
843     /*
844      * Maximal hardware transfer length in bytes.  Applies whenever
845      * transfers to the device bypass the kernel I/O scheduler, for
846      * example with SG_IO.  If larger than max_transfer or if zero,
847      * blk_get_max_hw_transfer will fall back to max_transfer.
848      */
849     uint64_t max_hw_transfer;
850 
851     /*
852      * Maximal number of scatter/gather elements allowed by the hardware.
853      * Applies whenever transfers to the device bypass the kernel I/O
854      * scheduler, for example with SG_IO.  If larger than max_iov
855      * or if zero, blk_get_max_hw_iov will fall back to max_iov.
856      */
857     int max_hw_iov;
858 
859 
860     /* memory alignment, in bytes so that no bounce buffer is needed */
861     size_t min_mem_alignment;
862 
863     /* memory alignment, in bytes, for bounce buffer */
864     size_t opt_mem_alignment;
865 
866     /* maximum number of iovec elements */
867     int max_iov;
868 
869     /*
870      * true if the length of the underlying file can change, and QEMU
871      * is expected to adjust automatically.  Mostly for CD-ROM drives,
872      * whose length is zero when the tray is empty (they don't need
873      * an explicit monitor command to load the disk inside the guest).
874      */
875     bool has_variable_length;
876 
877     /* device zone model */
878     BlockZoneModel zoned;
879 
880     /* zone size expressed in bytes */
881     uint32_t zone_size;
882 
883     /* total number of zones */
884     uint32_t nr_zones;
885 
886     /* maximum sectors of a zone append write operation */
887     uint32_t max_append_sectors;
888 
889     /* maximum number of open zones */
890     uint32_t max_open_zones;
891 
892     /* maximum number of active zones */
893     uint32_t max_active_zones;
894 
895     uint32_t write_granularity;
896 } BlockLimits;
897 
898 typedef struct BdrvOpBlocker BdrvOpBlocker;
899 
900 typedef struct BdrvAioNotifier {
901     void (*attached_aio_context)(AioContext *new_context, void *opaque);
902     void (*detach_aio_context)(void *opaque);
903 
904     void *opaque;
905     bool deleted;
906 
907     QLIST_ENTRY(BdrvAioNotifier) list;
908 } BdrvAioNotifier;
909 
910 struct BdrvChildClass {
911     /*
912      * If true, bdrv_replace_node() doesn't change the node this BdrvChild
913      * points to.
914      */
915     bool stay_at_node;
916 
917     /*
918      * If true, the parent is a BlockDriverState and bdrv_next_all_states()
919      * will return it. This information is used for drain_all, where every node
920      * will be drained separately, so the drain only needs to be propagated to
921      * non-BDS parents.
922      */
923     bool parent_is_bds;
924 
925     /*
926      * Global state (GS) API. These functions run under the BQL.
927      *
928      * See include/block/block-global-state.h for more information about
929      * the GS API.
930      */
931     void (*inherit_options)(BdrvChildRole role, bool parent_is_format,
932                             int *child_flags, QDict *child_options,
933                             int parent_flags, QDict *parent_options);
934     void (*change_media)(BdrvChild *child, bool load);
935 
936     /*
937      * Returns a malloced string that describes the parent of the child for a
938      * human reader. This could be a node-name, BlockBackend name, qdev ID or
939      * QOM path of the device owning the BlockBackend, job type and ID etc. The
940      * caller is responsible for freeing the memory.
941      */
942     char *(*get_parent_desc)(BdrvChild *child);
943 
944     /*
945      * Notifies the parent that the child has been activated/inactivated (e.g.
946      * when migration is completing) and it can start/stop requesting
947      * permissions and doing I/O on it.
948      */
949     void GRAPH_RDLOCK_PTR (*activate)(BdrvChild *child, Error **errp);
950     int GRAPH_RDLOCK_PTR (*inactivate)(BdrvChild *child);
951 
952     void GRAPH_WRLOCK_PTR (*attach)(BdrvChild *child);
953     void GRAPH_WRLOCK_PTR (*detach)(BdrvChild *child);
954 
955     /*
956      * If this pair of functions is implemented, the parent doesn't issue new
957      * requests after returning from .drained_begin() until .drained_end() is
958      * called.
959      *
960      * These functions must not change the graph (and therefore also must not
961      * call aio_poll(), which could change the graph indirectly).
962      *
963      * Note that this can be nested. If drained_begin() was called twice, new
964      * I/O is allowed only after drained_end() was called twice, too.
965      */
966     void (*drained_begin)(BdrvChild *child);
967     void (*drained_end)(BdrvChild *child);
968 
969     /*
970      * Returns whether the parent has pending requests for the child. This
971      * callback is polled after .drained_begin() has been called until all
972      * activity on the child has stopped.
973      */
974     bool (*drained_poll)(BdrvChild *child);
975 
976     /*
977      * Notifies the parent that the filename of its child has changed (e.g.
978      * because the direct child was removed from the backing chain), so that it
979      * can update its reference.
980      */
981     int (*update_filename)(BdrvChild *child, BlockDriverState *new_base,
982                            const char *filename, Error **errp);
983 
984     bool (*change_aio_ctx)(BdrvChild *child, AioContext *ctx,
985                            GHashTable *visited, Transaction *tran,
986                            Error **errp);
987 
988     /*
989      * I/O API functions. These functions are thread-safe.
990      *
991      * See include/block/block-io.h for more information about
992      * the I/O API.
993      */
994 
995     void (*resize)(BdrvChild *child);
996 
997     /*
998      * Returns a name that is supposedly more useful for human users than the
999      * node name for identifying the node in question (in particular, a BB
1000      * name), or NULL if the parent can't provide a better name.
1001      */
1002     const char *(*get_name)(BdrvChild *child);
1003 
1004     AioContext *(*get_parent_aio_context)(BdrvChild *child);
1005 };
1006 
1007 extern const BdrvChildClass child_of_bds;
1008 
1009 struct BdrvChild {
1010     BlockDriverState *bs;
1011     char *name;
1012     const BdrvChildClass *klass;
1013     BdrvChildRole role;
1014     void *opaque;
1015 
1016     /**
1017      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
1018      */
1019     uint64_t perm;
1020 
1021     /**
1022      * Permissions that can still be granted to other users of @bs while this
1023      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
1024      */
1025     uint64_t shared_perm;
1026 
1027     /*
1028      * This link is frozen: the child can neither be replaced nor
1029      * detached from the parent.
1030      */
1031     bool frozen;
1032 
1033     /*
1034      * True if the parent of this child has been drained by this BdrvChild
1035      * (through klass->drained_*).
1036      *
1037      * It is generally true if bs->quiesce_counter > 0. It may differ while the
1038      * child is entering or leaving a drained section.
1039      */
1040     bool quiesced_parent;
1041 
1042     QLIST_ENTRY(BdrvChild) next;
1043     QLIST_ENTRY(BdrvChild) next_parent;
1044 };
1045 
1046 /*
1047  * Allows bdrv_co_block_status() to cache one data region for a
1048  * protocol node.
1049  *
1050  * @valid: Whether the cache is valid (should be accessed with atomic
1051  *         functions so this can be reset by RCU readers)
1052  * @data_start: Offset where we know (or strongly assume) is data
1053  * @data_end: Offset where the data region ends (which is not necessarily
1054  *            the start of a zeroed region)
1055  */
1056 typedef struct BdrvBlockStatusCache {
1057     struct rcu_head rcu;
1058 
1059     bool valid;
1060     int64_t data_start;
1061     int64_t data_end;
1062 } BdrvBlockStatusCache;
1063 
1064 struct BlockDriverState {
1065     /*
1066      * Protected by big QEMU lock or read-only after opening.  No special
1067      * locking needed during I/O...
1068      */
1069     int open_flags; /* flags used to open the file, re-used for re-open */
1070     bool encrypted; /* if true, the media is encrypted */
1071     bool sg;        /* if true, the device is a /dev/sg* */
1072     bool probed;    /* if true, format was probed rather than specified */
1073     bool force_share; /* if true, always allow all shared permissions */
1074     bool implicit;  /* if true, this filter node was automatically inserted */
1075 
1076     BlockDriver *drv; /* NULL means no media */
1077     void *opaque;
1078 
1079     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
1080     /*
1081      * long-running tasks intended to always use the same AioContext as this
1082      * BDS may register themselves in this list to be notified of changes
1083      * regarding this BDS's context
1084      */
1085     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
1086     bool walking_aio_notifiers; /* to make removal during iteration safe */
1087 
1088     char filename[PATH_MAX];
1089     /*
1090      * If not empty, this image is a diff in relation to backing_file.
1091      * Note that this is the name given in the image header and
1092      * therefore may or may not be equal to .backing->bs->filename.
1093      * If this field contains a relative path, it is to be resolved
1094      * relatively to the overlay's location.
1095      */
1096     char backing_file[PATH_MAX];
1097     /*
1098      * The backing filename indicated by the image header.  Contrary
1099      * to backing_file, if we ever open this file, auto_backing_file
1100      * is replaced by the resulting BDS's filename (i.e. after a
1101      * bdrv_refresh_filename() run).
1102      */
1103     char auto_backing_file[PATH_MAX];
1104     char backing_format[16]; /* if non-zero and backing_file exists */
1105 
1106     QDict *full_open_options;
1107     char exact_filename[PATH_MAX];
1108 
1109     /* I/O Limits */
1110     BlockLimits bl;
1111 
1112     /*
1113      * Flags honored during pread
1114      */
1115     BdrvRequestFlags supported_read_flags;
1116     /*
1117      * Flags honored during pwrite (so far: BDRV_REQ_FUA,
1118      * BDRV_REQ_WRITE_UNCHANGED).
1119      * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
1120      * writes will be issued as normal writes without the flag set.
1121      * This is important to note for drivers that do not explicitly
1122      * request a WRITE permission for their children and instead take
1123      * the same permissions as their parent did (this is commonly what
1124      * block filters do).  Such drivers have to be aware that the
1125      * parent may have taken a WRITE_UNCHANGED permission only and is
1126      * issuing such requests.  Drivers either must make sure that
1127      * these requests do not result in plain WRITE accesses (usually
1128      * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1129      * every incoming write request as-is, including potentially that
1130      * flag), or they have to explicitly take the WRITE permission for
1131      * their children.
1132      */
1133     BdrvRequestFlags supported_write_flags;
1134     /*
1135      * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1136      * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1137      */
1138     BdrvRequestFlags supported_zero_flags;
1139     /*
1140      * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1141      *
1142      * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1143      * that any added space reads as all zeros. If this can't be guaranteed,
1144      * the operation must fail.
1145      */
1146     BdrvRequestFlags supported_truncate_flags;
1147 
1148     /* the following member gives a name to every node on the bs graph. */
1149     char node_name[32];
1150     /* element of the list of named nodes building the graph */
1151     QTAILQ_ENTRY(BlockDriverState) node_list;
1152     /* element of the list of all BlockDriverStates (all_bdrv_states) */
1153     QTAILQ_ENTRY(BlockDriverState) bs_list;
1154     /* element of the list of monitor-owned BDS */
1155     QTAILQ_ENTRY(BlockDriverState) monitor_list;
1156     int refcnt;
1157 
1158     /* operation blockers. Protected by BQL. */
1159     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
1160 
1161     /*
1162      * The node that this node inherited default options from (and a reopen on
1163      * which can affect this node by changing these defaults). This is always a
1164      * parent node of this node.
1165      */
1166     BlockDriverState *inherits_from;
1167 
1168     /*
1169      * @backing and @file are some of @children or NULL. All these three fields
1170      * (@file, @backing and @children) are modified only in
1171      * bdrv_child_cb_attach() and bdrv_child_cb_detach().
1172      *
1173      * See also comment in include/block/block.h, to learn how backing and file
1174      * are connected with BdrvChildRole.
1175      */
1176     QLIST_HEAD(, BdrvChild) children;
1177     BdrvChild *backing;
1178     BdrvChild *file;
1179 
1180     QLIST_HEAD(, BdrvChild) parents;
1181 
1182     QDict *options;
1183     QDict *explicit_options;
1184     BlockdevDetectZeroesOptions detect_zeroes;
1185 
1186     /* The error object in use for blocking operations on backing_hd */
1187     Error *backing_blocker;
1188 
1189     /* Protected by AioContext lock */
1190 
1191     /*
1192      * If we are reading a disk image, give its size in sectors.
1193      * Generally read-only; it is written to by load_snapshot and
1194      * save_snaphost, but the block layer is quiescent during those.
1195      */
1196     int64_t total_sectors;
1197 
1198     /* threshold limit for writes, in bytes. "High water mark". */
1199     uint64_t write_threshold_offset;
1200 
1201     /*
1202      * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1203      * Reading from the list can be done with either the BQL or the
1204      * dirty_bitmap_mutex.  Modifying a bitmap only requires
1205      * dirty_bitmap_mutex.
1206      */
1207     QemuMutex dirty_bitmap_mutex;
1208     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
1209 
1210     /* Offset after the highest byte written to */
1211     Stat64 wr_highest_offset;
1212 
1213     /*
1214      * If true, copy read backing sectors into image.  Can be >1 if more
1215      * than one client has requested copy-on-read.  Accessed with atomic
1216      * ops.
1217      */
1218     int copy_on_read;
1219 
1220     /*
1221      * number of in-flight requests; overall and serialising.
1222      * Accessed with atomic ops.
1223      */
1224     unsigned int in_flight;
1225     unsigned int serialising_in_flight;
1226 
1227     /* do we need to tell the quest if we have a volatile write cache? */
1228     int enable_write_cache;
1229 
1230     /* Accessed with atomic ops.  */
1231     int quiesce_counter;
1232 
1233     unsigned int write_gen;               /* Current data generation */
1234 
1235     /* Protected by reqs_lock.  */
1236     QemuMutex reqs_lock;
1237     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
1238     CoQueue flush_queue;                  /* Serializing flush queue */
1239     bool active_flush_req;                /* Flush request in flight? */
1240 
1241     /* Only read/written by whoever has set active_flush_req to true.  */
1242     unsigned int flushed_gen;             /* Flushed write generation */
1243 
1244     /* BdrvChild links to this node may never be frozen */
1245     bool never_freeze;
1246 
1247     /* Lock for block-status cache RCU writers */
1248     CoMutex bsc_modify_lock;
1249     /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1250     BdrvBlockStatusCache *block_status_cache;
1251 
1252     /* array of write pointers' location of each zone in the zoned device. */
1253     BlockZoneWps *wps;
1254 };
1255 
1256 struct BlockBackendRootState {
1257     int open_flags;
1258     BlockdevDetectZeroesOptions detect_zeroes;
1259 };
1260 
1261 typedef enum BlockMirrorBackingMode {
1262     /*
1263      * Reuse the existing backing chain from the source for the target.
1264      * - sync=full: Set backing BDS to NULL.
1265      * - sync=top:  Use source's backing BDS.
1266      * - sync=none: Use source as the backing BDS.
1267      */
1268     MIRROR_SOURCE_BACKING_CHAIN,
1269 
1270     /* Open the target's backing chain completely anew */
1271     MIRROR_OPEN_BACKING_CHAIN,
1272 
1273     /* Do not change the target's backing BDS after job completion */
1274     MIRROR_LEAVE_BACKING_CHAIN,
1275 } BlockMirrorBackingMode;
1276 
1277 
1278 /*
1279  * Essential block drivers which must always be statically linked into qemu, and
1280  * which therefore can be accessed without using bdrv_find_format()
1281  */
1282 extern BlockDriver bdrv_file;
1283 extern BlockDriver bdrv_raw;
1284 extern BlockDriver bdrv_qcow2;
1285 
1286 extern unsigned int bdrv_drain_all_count;
1287 extern QemuOptsList bdrv_create_opts_simple;
1288 
1289 /*
1290  * Common functions that are neither I/O nor Global State.
1291  *
1292  * See include/block/block-common.h for more information about
1293  * the Common API.
1294  */
1295 
1296 static inline BlockDriverState *child_bs(BdrvChild *child)
1297 {
1298     return child ? child->bs : NULL;
1299 }
1300 
1301 int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp);
1302 char *create_tmp_file(Error **errp);
1303 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
1304                                       QDict *options);
1305 
1306 
1307 int bdrv_check_qiov_request(int64_t offset, int64_t bytes,
1308                             QEMUIOVector *qiov, size_t qiov_offset,
1309                             Error **errp);
1310 
1311 #ifdef _WIN32
1312 int is_windows_drive(const char *filename);
1313 #endif
1314 
1315 #endif /* BLOCK_INT_COMMON_H */
1316