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