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