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 (*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      * Try to get @bs's logical and physical block size.
368      * On success, store them in @bsz and return zero.
369      * On failure, return negative errno.
370      */
371     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
372     /**
373      * Try to get @bs's geometry (cyls, heads, sectors)
374      * On success, store them in @geo and return 0.
375      * On failure return -errno.
376      * Only drivers that want to override guest geometry implement this
377      * callback; see hd_geometry_guess().
378      */
379     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
380 
381     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
382                            Error **errp);
383     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
384                            Error **errp);
385 
386     /**
387      * Informs the block driver that a permission change is intended. The
388      * driver checks whether the change is permissible and may take other
389      * preparations for the change (e.g. get file system locks). This operation
390      * is always followed either by a call to either .bdrv_set_perm or
391      * .bdrv_abort_perm_update.
392      *
393      * Checks whether the requested set of cumulative permissions in @perm
394      * can be granted for accessing @bs and whether no other users are using
395      * permissions other than those given in @shared (both arguments take
396      * BLK_PERM_* bitmasks).
397      *
398      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
399      * and errp is set to an error describing the conflict.
400      */
401     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
402                            uint64_t shared, Error **errp);
403 
404     /**
405      * Called to inform the driver that the set of cumulative set of used
406      * permissions for @bs has changed to @perm, and the set of sharable
407      * permission to @shared. The driver can use this to propagate changes to
408      * its children (i.e. request permissions only if a parent actually needs
409      * them).
410      *
411      * This function is only invoked after bdrv_check_perm(), so block drivers
412      * may rely on preparations made in their .bdrv_check_perm implementation.
413      */
414     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
415 
416     /*
417      * Called to inform the driver that after a previous bdrv_check_perm()
418      * call, the permission update is not performed and any preparations made
419      * for it (e.g. taken file locks) need to be undone.
420      *
421      * This function can be called even for nodes that never saw a
422      * bdrv_check_perm() call. It is a no-op then.
423      */
424     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
425 
426     /**
427      * Returns in @nperm and @nshared the permissions that the driver for @bs
428      * needs on its child @c, based on the cumulative permissions requested by
429      * the parents in @parent_perm and @parent_shared.
430      *
431      * If @c is NULL, return the permissions for attaching a new child for the
432      * given @child_class and @role.
433      *
434      * If @reopen_queue is non-NULL, don't return the currently needed
435      * permissions, but those that will be needed after applying the
436      * @reopen_queue.
437      */
438      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
439                              BdrvChildRole role,
440                              BlockReopenQueue *reopen_queue,
441                              uint64_t parent_perm, uint64_t parent_shared,
442                              uint64_t *nperm, uint64_t *nshared);
443 
444     /**
445      * Register/unregister a buffer for I/O. For example, when the driver is
446      * interested to know the memory areas that will later be used in iovs, so
447      * that it can do IOMMU mapping with VFIO etc., in order to get better
448      * performance. In the case of VFIO drivers, this callback is used to do
449      * DMA mapping for hot buffers.
450      *
451      * Returns: true on success, false on failure
452      */
453     bool GRAPH_RDLOCK_PTR (*bdrv_register_buf)(
454         BlockDriverState *bs, void *host, size_t size, Error **errp);
455     void GRAPH_RDLOCK_PTR (*bdrv_unregister_buf)(
456         BlockDriverState *bs, void *host, size_t size);
457 
458     /*
459      * This field is modified only under the BQL, and is part of
460      * the global state.
461      */
462     QLIST_ENTRY(BlockDriver) list;
463 
464     /*
465      * I/O API functions. These functions are thread-safe.
466      *
467      * See include/block/block-io.h for more information about
468      * the I/O API.
469      */
470 
471     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
472 
473     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_amend)(
474         BlockDriverState *bs, BlockdevAmendOptions *opts, bool force,
475         Error **errp);
476 
477     /* aio */
478     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_preadv)(BlockDriverState *bs,
479         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
480         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
481 
482     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_pwritev)(BlockDriverState *bs,
483         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
484         BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque);
485 
486     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_flush)(
487         BlockDriverState *bs, BlockCompletionFunc *cb, void *opaque);
488 
489     BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_pdiscard)(
490         BlockDriverState *bs, int64_t offset, int bytes,
491         BlockCompletionFunc *cb, void *opaque);
492 
493     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_readv)(BlockDriverState *bs,
494         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
495 
496     /**
497      * @offset: position in bytes to read at
498      * @bytes: number of bytes to read
499      * @qiov: the buffers to fill with read data
500      * @flags: currently unused, always 0
501      *
502      * @offset and @bytes will be a multiple of 'request_alignment',
503      * but the length of individual @qiov elements does not have to
504      * be a multiple.
505      *
506      * @bytes will always equal the total size of @qiov, and will be
507      * no larger than 'max_transfer'.
508      *
509      * The buffer in @qiov may point directly to guest memory.
510      */
511     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv)(BlockDriverState *bs,
512         int64_t offset, int64_t bytes, QEMUIOVector *qiov,
513         BdrvRequestFlags flags);
514 
515     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_part)(
516         BlockDriverState *bs, int64_t offset, int64_t bytes,
517         QEMUIOVector *qiov, size_t qiov_offset,
518         BdrvRequestFlags flags);
519 
520     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_writev)(BlockDriverState *bs,
521         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov,
522         int flags);
523     /**
524      * @offset: position in bytes to write at
525      * @bytes: number of bytes to write
526      * @qiov: the buffers containing data to write
527      * @flags: zero or more bits allowed by 'supported_write_flags'
528      *
529      * @offset and @bytes will be a multiple of 'request_alignment',
530      * but the length of individual @qiov elements does not have to
531      * be a multiple.
532      *
533      * @bytes will always equal the total size of @qiov, and will be
534      * no larger than 'max_transfer'.
535      *
536      * The buffer in @qiov may point directly to guest memory.
537      */
538     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev)(
539         BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov,
540         BdrvRequestFlags flags);
541     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_part)(
542         BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov,
543         size_t qiov_offset, BdrvRequestFlags flags);
544 
545     /*
546      * Efficiently zero a region of the disk image.  Typically an image format
547      * would use a compact metadata representation to implement this.  This
548      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
549      * will be called instead.
550      */
551     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwrite_zeroes)(
552         BlockDriverState *bs, int64_t offset, int64_t bytes,
553         BdrvRequestFlags flags);
554 
555     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard)(
556         BlockDriverState *bs, int64_t offset, int64_t bytes);
557 
558     /*
559      * Map [offset, offset + nbytes) range onto a child of @bs to copy from,
560      * and invoke bdrv_co_copy_range_from(child, ...), or invoke
561      * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from.
562      *
563      * See the comment of bdrv_co_copy_range for the parameter and return value
564      * semantics.
565      */
566     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_from)(
567         BlockDriverState *bs, BdrvChild *src, int64_t offset,
568         BdrvChild *dst, int64_t dst_offset, int64_t bytes,
569         BdrvRequestFlags read_flags, BdrvRequestFlags write_flags);
570 
571     /*
572      * Map [offset, offset + nbytes) range onto a child of bs to copy data to,
573      * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy
574      * operation if @bs is the leaf and @src has the same BlockDriver.  Return
575      * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver.
576      *
577      * See the comment of bdrv_co_copy_range for the parameter and return value
578      * semantics.
579      */
580     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_to)(
581         BlockDriverState *bs, BdrvChild *src, int64_t src_offset,
582         BdrvChild *dst, int64_t dst_offset, int64_t bytes,
583         BdrvRequestFlags read_flags, BdrvRequestFlags write_flags);
584 
585     /*
586      * Building block for bdrv_block_status[_above] and
587      * bdrv_is_allocated[_above].  The driver should answer only
588      * according to the current layer, and should only need to set
589      * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
590      * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
591      * layer, the result should be 0 (and not BDRV_BLOCK_ZERO).  See
592      * block.h for the overall meaning of the bits.  As a hint, the
593      * flag want_zero is true if the caller cares more about precise
594      * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
595      * overall allocation (favor larger *pnum, perhaps by reporting
596      * _DATA instead of _ZERO).  The block layer guarantees input
597      * clamped to bdrv_getlength() and aligned to request_alignment,
598      * as well as non-NULL pnum, map, and file; in turn, the driver
599      * must return an error or set pnum to an aligned non-zero value.
600      *
601      * Note that @bytes is just a hint on how big of a region the
602      * caller wants to inspect.  It is not a limit on *pnum.
603      * Implementations are free to return larger values of *pnum if
604      * doing so does not incur a performance penalty.
605      *
606      * block/io.c's bdrv_co_block_status() will utilize an unclamped
607      * *pnum value for the block-status cache on protocol nodes, prior
608      * to clamping *pnum for return to its caller.
609      */
610     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_block_status)(
611         BlockDriverState *bs,
612         bool want_zero, int64_t offset, int64_t bytes, int64_t *pnum,
613         int64_t *map, BlockDriverState **file);
614 
615     /*
616      * Snapshot-access API.
617      *
618      * Block-driver may provide snapshot-access API: special functions to access
619      * some internal "snapshot". The functions are similar with normal
620      * read/block_status/discard handler, but don't have any specific handling
621      * in generic block-layer: no serializing, no alignment, no tracked
622      * requests. So, block-driver that realizes these APIs is fully responsible
623      * for synchronization between snapshot-access API and normal IO requests.
624      *
625      * TODO: To be able to support qcow2's internal snapshots, this API will
626      * need to be extended to:
627      * - be able to select a specific snapshot
628      * - receive the snapshot's actual length (which may differ from bs's
629      *   length)
630      */
631     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_snapshot)(
632         BlockDriverState *bs, int64_t offset, int64_t bytes,
633         QEMUIOVector *qiov, size_t qiov_offset);
634 
635     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_snapshot_block_status)(
636         BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
637         int64_t *pnum, int64_t *map, BlockDriverState **file);
638 
639     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard_snapshot)(
640         BlockDriverState *bs, int64_t offset, int64_t bytes);
641 
642     /*
643      * Invalidate any cached meta-data.
644      */
645     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_invalidate_cache)(
646         BlockDriverState *bs, Error **errp);
647 
648     /*
649      * Flushes all data for all layers by calling bdrv_co_flush for underlying
650      * layers, if needed. This function is needed for deterministic
651      * synchronization of the flush finishing callback.
652      */
653     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush)(BlockDriverState *bs);
654 
655     /* Delete a created file. */
656     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_delete_file)(
657         BlockDriverState *bs, Error **errp);
658 
659     /*
660      * Flushes all data that was already written to the OS all the way down to
661      * the disk (for example file-posix.c calls fsync()).
662      */
663     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_disk)(
664         BlockDriverState *bs);
665 
666     /*
667      * Flushes all internal caches to the OS. The data may still sit in a
668      * writeback cache of the host OS, but it will survive a crash of the qemu
669      * process.
670      */
671     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_os)(
672         BlockDriverState *bs);
673 
674     /*
675      * Truncate @bs to @offset bytes using the given @prealloc mode
676      * when growing.  Modes other than PREALLOC_MODE_OFF should be
677      * rejected when shrinking @bs.
678      *
679      * If @exact is true, @bs must be resized to exactly @offset.
680      * Otherwise, it is sufficient for @bs (if it is a host block
681      * device and thus there is no way to resize it) to be at least
682      * @offset bytes in length.
683      *
684      * If @exact is true and this function fails but would succeed
685      * with @exact = false, it should return -ENOTSUP.
686      */
687     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_truncate)(
688         BlockDriverState *bs, int64_t offset, bool exact,
689         PreallocMode prealloc, BdrvRequestFlags flags, Error **errp);
690 
691     int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_getlength)(
692         BlockDriverState *bs);
693 
694     int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_allocated_file_size)(
695         BlockDriverState *bs);
696 
697     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
698                                       Error **errp);
699 
700     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed)(
701         BlockDriverState *bs, int64_t offset, int64_t bytes,
702         QEMUIOVector *qiov);
703 
704     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed_part)(
705         BlockDriverState *bs, int64_t offset, int64_t bytes,
706         QEMUIOVector *qiov, size_t qiov_offset);
707 
708     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_info)(
709         BlockDriverState *bs, BlockDriverInfo *bdi);
710 
711     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs,
712                                                  Error **errp);
713     BlockStatsSpecific *(*bdrv_get_specific_stats)(BlockDriverState *bs);
714 
715     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_save_vmstate)(
716         BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
717 
718     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_load_vmstate)(
719         BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
720 
721     int coroutine_fn (*bdrv_co_zone_report)(BlockDriverState *bs,
722             int64_t offset, unsigned int *nr_zones,
723             BlockZoneDescriptor *zones);
724     int coroutine_fn (*bdrv_co_zone_mgmt)(BlockDriverState *bs, BlockZoneOp op,
725             int64_t offset, int64_t len);
726     int coroutine_fn (*bdrv_co_zone_append)(BlockDriverState *bs,
727             int64_t *offset, QEMUIOVector *qiov,
728             BdrvRequestFlags flags);
729 
730     /* removable device specific */
731     bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_is_inserted)(
732         BlockDriverState *bs);
733     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_eject)(
734         BlockDriverState *bs, bool eject_flag);
735     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_lock_medium)(
736         BlockDriverState *bs, bool locked);
737 
738     /* to control generic scsi devices */
739     BlockAIOCB *coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_aio_ioctl)(
740         BlockDriverState *bs, unsigned long int req, void *buf,
741         BlockCompletionFunc *cb, void *opaque);
742 
743     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_ioctl)(
744         BlockDriverState *bs, unsigned long int req, void *buf);
745 
746     /*
747      * Returns 0 for completed check, -errno for internal errors.
748      * The check results are stored in result.
749      */
750     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_check)(
751         BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix);
752 
753     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_debug_event)(
754         BlockDriverState *bs, BlkdebugEvent event);
755 
756     /* io queue for linux-aio */
757     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_io_plug)(BlockDriverState *bs);
758     void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_io_unplug)(
759         BlockDriverState *bs);
760 
761     /**
762      * bdrv_drain_begin is called if implemented in the beginning of a
763      * drain operation to drain and stop any internal sources of requests in
764      * the driver.
765      * bdrv_drain_end is called if implemented at the end of the drain.
766      *
767      * They should be used by the driver to e.g. manage scheduled I/O
768      * requests, or toggle an internal state. After the end of the drain new
769      * requests will continue normally.
770      *
771      * Implementations of both functions must not call aio_poll().
772      */
773     void (*bdrv_drain_begin)(BlockDriverState *bs);
774     void (*bdrv_drain_end)(BlockDriverState *bs);
775 
776     bool (*bdrv_supports_persistent_dirty_bitmap)(BlockDriverState *bs);
777 
778     bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_can_store_new_dirty_bitmap)(
779         BlockDriverState *bs, const char *name, uint32_t granularity,
780         Error **errp);
781 
782     int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_remove_persistent_dirty_bitmap)(
783         BlockDriverState *bs, const char *name, Error **errp);
784 };
785 
786 static inline bool TSA_NO_TSA block_driver_can_compress(BlockDriver *drv)
787 {
788     return drv->bdrv_co_pwritev_compressed ||
789            drv->bdrv_co_pwritev_compressed_part;
790 }
791 
792 typedef struct BlockLimits {
793     /*
794      * Alignment requirement, in bytes, for offset/length of I/O
795      * requests. Must be a power of 2 less than INT_MAX; defaults to
796      * 1 for drivers with modern byte interfaces, and to 512
797      * otherwise.
798      */
799     uint32_t request_alignment;
800 
801     /*
802      * Maximum number of bytes that can be discarded at once. Must be multiple
803      * of pdiscard_alignment, but need not be power of 2. May be 0 if no
804      * inherent 64-bit limit.
805      */
806     int64_t max_pdiscard;
807 
808     /*
809      * Optimal alignment for discard requests in bytes. A power of 2
810      * is best but not mandatory.  Must be a multiple of
811      * bl.request_alignment, and must be less than max_pdiscard if
812      * that is set. May be 0 if bl.request_alignment is good enough
813      */
814     uint32_t pdiscard_alignment;
815 
816     /*
817      * Maximum number of bytes that can zeroized at once. Must be multiple of
818      * pwrite_zeroes_alignment. 0 means no limit.
819      */
820     int64_t max_pwrite_zeroes;
821 
822     /*
823      * Optimal alignment for write zeroes requests in bytes. A power
824      * of 2 is best but not mandatory.  Must be a multiple of
825      * bl.request_alignment, and must be less than max_pwrite_zeroes
826      * if that is set. May be 0 if bl.request_alignment is good
827      * enough
828      */
829     uint32_t pwrite_zeroes_alignment;
830 
831     /*
832      * Optimal transfer length in bytes.  A power of 2 is best but not
833      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
834      * no preferred size
835      */
836     uint32_t opt_transfer;
837 
838     /*
839      * Maximal transfer length in bytes.  Need not be power of 2, but
840      * must be multiple of opt_transfer and bl.request_alignment, or 0
841      * for no 32-bit limit.  For now, anything larger than INT_MAX is
842      * clamped down.
843      */
844     uint32_t max_transfer;
845 
846     /*
847      * Maximal hardware transfer length in bytes.  Applies whenever
848      * transfers to the device bypass the kernel I/O scheduler, for
849      * example with SG_IO.  If larger than max_transfer or if zero,
850      * blk_get_max_hw_transfer will fall back to max_transfer.
851      */
852     uint64_t max_hw_transfer;
853 
854     /*
855      * Maximal number of scatter/gather elements allowed by the hardware.
856      * Applies whenever transfers to the device bypass the kernel I/O
857      * scheduler, for example with SG_IO.  If larger than max_iov
858      * or if zero, blk_get_max_hw_iov will fall back to max_iov.
859      */
860     int max_hw_iov;
861 
862 
863     /* memory alignment, in bytes so that no bounce buffer is needed */
864     size_t min_mem_alignment;
865 
866     /* memory alignment, in bytes, for bounce buffer */
867     size_t opt_mem_alignment;
868 
869     /* maximum number of iovec elements */
870     int max_iov;
871 
872     /*
873      * true if the length of the underlying file can change, and QEMU
874      * is expected to adjust automatically.  Mostly for CD-ROM drives,
875      * whose length is zero when the tray is empty (they don't need
876      * an explicit monitor command to load the disk inside the guest).
877      */
878     bool has_variable_length;
879 
880     /* device zone model */
881     BlockZoneModel zoned;
882 
883     /* zone size expressed in bytes */
884     uint32_t zone_size;
885 
886     /* total number of zones */
887     uint32_t nr_zones;
888 
889     /* maximum sectors of a zone append write operation */
890     uint32_t max_append_sectors;
891 
892     /* maximum number of open zones */
893     uint32_t max_open_zones;
894 
895     /* maximum number of active zones */
896     uint32_t max_active_zones;
897 
898     uint32_t write_granularity;
899 } BlockLimits;
900 
901 typedef struct BdrvOpBlocker BdrvOpBlocker;
902 
903 typedef struct BdrvAioNotifier {
904     void (*attached_aio_context)(AioContext *new_context, void *opaque);
905     void (*detach_aio_context)(void *opaque);
906 
907     void *opaque;
908     bool deleted;
909 
910     QLIST_ENTRY(BdrvAioNotifier) list;
911 } BdrvAioNotifier;
912 
913 struct BdrvChildClass {
914     /*
915      * If true, bdrv_replace_node() doesn't change the node this BdrvChild
916      * points to.
917      */
918     bool stay_at_node;
919 
920     /*
921      * If true, the parent is a BlockDriverState and bdrv_next_all_states()
922      * will return it. This information is used for drain_all, where every node
923      * will be drained separately, so the drain only needs to be propagated to
924      * non-BDS parents.
925      */
926     bool parent_is_bds;
927 
928     /*
929      * Global state (GS) API. These functions run under the BQL.
930      *
931      * See include/block/block-global-state.h for more information about
932      * the GS API.
933      */
934     void (*inherit_options)(BdrvChildRole role, bool parent_is_format,
935                             int *child_flags, QDict *child_options,
936                             int parent_flags, QDict *parent_options);
937     void (*change_media)(BdrvChild *child, bool load);
938 
939     /*
940      * Returns a malloced string that describes the parent of the child for a
941      * human reader. This could be a node-name, BlockBackend name, qdev ID or
942      * QOM path of the device owning the BlockBackend, job type and ID etc. The
943      * caller is responsible for freeing the memory.
944      */
945     char *(*get_parent_desc)(BdrvChild *child);
946 
947     /*
948      * Notifies the parent that the child has been activated/inactivated (e.g.
949      * when migration is completing) and it can start/stop requesting
950      * permissions and doing I/O on it.
951      */
952     void (*activate)(BdrvChild *child, Error **errp);
953     int (*inactivate)(BdrvChild *child);
954 
955     void GRAPH_WRLOCK_PTR (*attach)(BdrvChild *child);
956     void GRAPH_WRLOCK_PTR (*detach)(BdrvChild *child);
957 
958     /*
959      * Notifies the parent that the filename of its child has changed (e.g.
960      * because the direct child was removed from the backing chain), so that it
961      * can update its reference.
962      */
963     int (*update_filename)(BdrvChild *child, BlockDriverState *new_base,
964                            const char *filename, Error **errp);
965 
966     bool (*change_aio_ctx)(BdrvChild *child, AioContext *ctx,
967                            GHashTable *visited, Transaction *tran,
968                            Error **errp);
969 
970     /*
971      * I/O API functions. These functions are thread-safe.
972      *
973      * See include/block/block-io.h for more information about
974      * the I/O API.
975      */
976 
977     void (*resize)(BdrvChild *child);
978 
979     /*
980      * Returns a name that is supposedly more useful for human users than the
981      * node name for identifying the node in question (in particular, a BB
982      * name), or NULL if the parent can't provide a better name.
983      */
984     const char *(*get_name)(BdrvChild *child);
985 
986     AioContext *(*get_parent_aio_context)(BdrvChild *child);
987 
988     /*
989      * If this pair of functions is implemented, the parent doesn't issue new
990      * requests after returning from .drained_begin() until .drained_end() is
991      * called.
992      *
993      * These functions must not change the graph (and therefore also must not
994      * call aio_poll(), which could change the graph indirectly).
995      *
996      * Note that this can be nested. If drained_begin() was called twice, new
997      * I/O is allowed only after drained_end() was called twice, too.
998      */
999     void (*drained_begin)(BdrvChild *child);
1000     void (*drained_end)(BdrvChild *child);
1001 
1002     /*
1003      * Returns whether the parent has pending requests for the child. This
1004      * callback is polled after .drained_begin() has been called until all
1005      * activity on the child has stopped.
1006      */
1007     bool (*drained_poll)(BdrvChild *child);
1008 };
1009 
1010 extern const BdrvChildClass child_of_bds;
1011 
1012 struct BdrvChild {
1013     BlockDriverState *bs;
1014     char *name;
1015     const BdrvChildClass *klass;
1016     BdrvChildRole role;
1017     void *opaque;
1018 
1019     /**
1020      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
1021      */
1022     uint64_t perm;
1023 
1024     /**
1025      * Permissions that can still be granted to other users of @bs while this
1026      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
1027      */
1028     uint64_t shared_perm;
1029 
1030     /*
1031      * This link is frozen: the child can neither be replaced nor
1032      * detached from the parent.
1033      */
1034     bool frozen;
1035 
1036     /*
1037      * True if the parent of this child has been drained by this BdrvChild
1038      * (through klass->drained_*).
1039      *
1040      * It is generally true if bs->quiesce_counter > 0. It may differ while the
1041      * child is entering or leaving a drained section.
1042      */
1043     bool quiesced_parent;
1044 
1045     QLIST_ENTRY(BdrvChild) next;
1046     QLIST_ENTRY(BdrvChild) next_parent;
1047 };
1048 
1049 /*
1050  * Allows bdrv_co_block_status() to cache one data region for a
1051  * protocol node.
1052  *
1053  * @valid: Whether the cache is valid (should be accessed with atomic
1054  *         functions so this can be reset by RCU readers)
1055  * @data_start: Offset where we know (or strongly assume) is data
1056  * @data_end: Offset where the data region ends (which is not necessarily
1057  *            the start of a zeroed region)
1058  */
1059 typedef struct BdrvBlockStatusCache {
1060     struct rcu_head rcu;
1061 
1062     bool valid;
1063     int64_t data_start;
1064     int64_t data_end;
1065 } BdrvBlockStatusCache;
1066 
1067 struct BlockDriverState {
1068     /*
1069      * Protected by big QEMU lock or read-only after opening.  No special
1070      * locking needed during I/O...
1071      */
1072     int open_flags; /* flags used to open the file, re-used for re-open */
1073     bool encrypted; /* if true, the media is encrypted */
1074     bool sg;        /* if true, the device is a /dev/sg* */
1075     bool probed;    /* if true, format was probed rather than specified */
1076     bool force_share; /* if true, always allow all shared permissions */
1077     bool implicit;  /* if true, this filter node was automatically inserted */
1078 
1079     BlockDriver *drv; /* NULL means no media */
1080     void *opaque;
1081 
1082     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
1083     /*
1084      * long-running tasks intended to always use the same AioContext as this
1085      * BDS may register themselves in this list to be notified of changes
1086      * regarding this BDS's context
1087      */
1088     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
1089     bool walking_aio_notifiers; /* to make removal during iteration safe */
1090 
1091     char filename[PATH_MAX];
1092     /*
1093      * If not empty, this image is a diff in relation to backing_file.
1094      * Note that this is the name given in the image header and
1095      * therefore may or may not be equal to .backing->bs->filename.
1096      * If this field contains a relative path, it is to be resolved
1097      * relatively to the overlay's location.
1098      */
1099     char backing_file[PATH_MAX];
1100     /*
1101      * The backing filename indicated by the image header.  Contrary
1102      * to backing_file, if we ever open this file, auto_backing_file
1103      * is replaced by the resulting BDS's filename (i.e. after a
1104      * bdrv_refresh_filename() run).
1105      */
1106     char auto_backing_file[PATH_MAX];
1107     char backing_format[16]; /* if non-zero and backing_file exists */
1108 
1109     QDict *full_open_options;
1110     char exact_filename[PATH_MAX];
1111 
1112     /* I/O Limits */
1113     BlockLimits bl;
1114 
1115     /*
1116      * Flags honored during pread
1117      */
1118     BdrvRequestFlags supported_read_flags;
1119     /*
1120      * Flags honored during pwrite (so far: BDRV_REQ_FUA,
1121      * BDRV_REQ_WRITE_UNCHANGED).
1122      * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those
1123      * writes will be issued as normal writes without the flag set.
1124      * This is important to note for drivers that do not explicitly
1125      * request a WRITE permission for their children and instead take
1126      * the same permissions as their parent did (this is commonly what
1127      * block filters do).  Such drivers have to be aware that the
1128      * parent may have taken a WRITE_UNCHANGED permission only and is
1129      * issuing such requests.  Drivers either must make sure that
1130      * these requests do not result in plain WRITE accesses (usually
1131      * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding
1132      * every incoming write request as-is, including potentially that
1133      * flag), or they have to explicitly take the WRITE permission for
1134      * their children.
1135      */
1136     BdrvRequestFlags supported_write_flags;
1137     /*
1138      * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
1139      * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED)
1140      */
1141     BdrvRequestFlags supported_zero_flags;
1142     /*
1143      * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE).
1144      *
1145      * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure
1146      * that any added space reads as all zeros. If this can't be guaranteed,
1147      * the operation must fail.
1148      */
1149     BdrvRequestFlags supported_truncate_flags;
1150 
1151     /* the following member gives a name to every node on the bs graph. */
1152     char node_name[32];
1153     /* element of the list of named nodes building the graph */
1154     QTAILQ_ENTRY(BlockDriverState) node_list;
1155     /* element of the list of all BlockDriverStates (all_bdrv_states) */
1156     QTAILQ_ENTRY(BlockDriverState) bs_list;
1157     /* element of the list of monitor-owned BDS */
1158     QTAILQ_ENTRY(BlockDriverState) monitor_list;
1159     int refcnt;
1160 
1161     /* operation blockers. Protected by BQL. */
1162     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
1163 
1164     /*
1165      * The node that this node inherited default options from (and a reopen on
1166      * which can affect this node by changing these defaults). This is always a
1167      * parent node of this node.
1168      */
1169     BlockDriverState *inherits_from;
1170 
1171     /*
1172      * @backing and @file are some of @children or NULL. All these three fields
1173      * (@file, @backing and @children) are modified only in
1174      * bdrv_child_cb_attach() and bdrv_child_cb_detach().
1175      *
1176      * See also comment in include/block/block.h, to learn how backing and file
1177      * are connected with BdrvChildRole.
1178      */
1179     QLIST_HEAD(, BdrvChild) children;
1180     BdrvChild *backing;
1181     BdrvChild *file;
1182 
1183     QLIST_HEAD(, BdrvChild) parents;
1184 
1185     QDict *options;
1186     QDict *explicit_options;
1187     BlockdevDetectZeroesOptions detect_zeroes;
1188 
1189     /* The error object in use for blocking operations on backing_hd */
1190     Error *backing_blocker;
1191 
1192     /* Protected by AioContext lock */
1193 
1194     /*
1195      * If we are reading a disk image, give its size in sectors.
1196      * Generally read-only; it is written to by load_snapshot and
1197      * save_snaphost, but the block layer is quiescent during those.
1198      */
1199     int64_t total_sectors;
1200 
1201     /* threshold limit for writes, in bytes. "High water mark". */
1202     uint64_t write_threshold_offset;
1203 
1204     /*
1205      * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
1206      * Reading from the list can be done with either the BQL or the
1207      * dirty_bitmap_mutex.  Modifying a bitmap only requires
1208      * dirty_bitmap_mutex.
1209      */
1210     QemuMutex dirty_bitmap_mutex;
1211     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
1212 
1213     /* Offset after the highest byte written to */
1214     Stat64 wr_highest_offset;
1215 
1216     /*
1217      * If true, copy read backing sectors into image.  Can be >1 if more
1218      * than one client has requested copy-on-read.  Accessed with atomic
1219      * ops.
1220      */
1221     int copy_on_read;
1222 
1223     /*
1224      * number of in-flight requests; overall and serialising.
1225      * Accessed with atomic ops.
1226      */
1227     unsigned int in_flight;
1228     unsigned int serialising_in_flight;
1229 
1230     /*
1231      * counter for nested bdrv_io_plug.
1232      * Accessed with atomic ops.
1233      */
1234     unsigned io_plugged;
1235 
1236     /* do we need to tell the quest if we have a volatile write cache? */
1237     int enable_write_cache;
1238 
1239     /* Accessed with atomic ops.  */
1240     int quiesce_counter;
1241 
1242     unsigned int write_gen;               /* Current data generation */
1243 
1244     /* Protected by reqs_lock.  */
1245     CoMutex reqs_lock;
1246     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
1247     CoQueue flush_queue;                  /* Serializing flush queue */
1248     bool active_flush_req;                /* Flush request in flight? */
1249 
1250     /* Only read/written by whoever has set active_flush_req to true.  */
1251     unsigned int flushed_gen;             /* Flushed write generation */
1252 
1253     /* BdrvChild links to this node may never be frozen */
1254     bool never_freeze;
1255 
1256     /* Lock for block-status cache RCU writers */
1257     CoMutex bsc_modify_lock;
1258     /* Always non-NULL, but must only be dereferenced under an RCU read guard */
1259     BdrvBlockStatusCache *block_status_cache;
1260 
1261     /* array of write pointers' location of each zone in the zoned device. */
1262     BlockZoneWps *wps;
1263 };
1264 
1265 struct BlockBackendRootState {
1266     int open_flags;
1267     BlockdevDetectZeroesOptions detect_zeroes;
1268 };
1269 
1270 typedef enum BlockMirrorBackingMode {
1271     /*
1272      * Reuse the existing backing chain from the source for the target.
1273      * - sync=full: Set backing BDS to NULL.
1274      * - sync=top:  Use source's backing BDS.
1275      * - sync=none: Use source as the backing BDS.
1276      */
1277     MIRROR_SOURCE_BACKING_CHAIN,
1278 
1279     /* Open the target's backing chain completely anew */
1280     MIRROR_OPEN_BACKING_CHAIN,
1281 
1282     /* Do not change the target's backing BDS after job completion */
1283     MIRROR_LEAVE_BACKING_CHAIN,
1284 } BlockMirrorBackingMode;
1285 
1286 
1287 /*
1288  * Essential block drivers which must always be statically linked into qemu, and
1289  * which therefore can be accessed without using bdrv_find_format()
1290  */
1291 extern BlockDriver bdrv_file;
1292 extern BlockDriver bdrv_raw;
1293 extern BlockDriver bdrv_qcow2;
1294 
1295 extern unsigned int bdrv_drain_all_count;
1296 extern QemuOptsList bdrv_create_opts_simple;
1297 
1298 /*
1299  * Common functions that are neither I/O nor Global State.
1300  *
1301  * See include/block/block-common.h for more information about
1302  * the Common API.
1303  */
1304 
1305 static inline BlockDriverState *child_bs(BdrvChild *child)
1306 {
1307     return child ? child->bs : NULL;
1308 }
1309 
1310 int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp);
1311 char *create_tmp_file(Error **errp);
1312 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
1313                                       QDict *options);
1314 
1315 
1316 int bdrv_check_qiov_request(int64_t offset, int64_t bytes,
1317                             QEMUIOVector *qiov, size_t qiov_offset,
1318                             Error **errp);
1319 
1320 #ifdef _WIN32
1321 int is_windows_drive(const char *filename);
1322 #endif
1323 
1324 #endif /* BLOCK_INT_COMMON_H */
1325