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