xref: /openbmc/qemu/include/block/block_int.h (revision a8583393)
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_H
25 #define BLOCK_INT_H
26 
27 #include "block/accounting.h"
28 #include "block/block.h"
29 #include "qemu/option.h"
30 #include "qemu/queue.h"
31 #include "qemu/coroutine.h"
32 #include "qemu/stats64.h"
33 #include "qemu/timer.h"
34 #include "qapi-types.h"
35 #include "qemu/hbitmap.h"
36 #include "block/snapshot.h"
37 #include "qemu/main-loop.h"
38 #include "qemu/throttle.h"
39 
40 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
41 
42 #define BLOCK_OPT_SIZE              "size"
43 #define BLOCK_OPT_ENCRYPT           "encryption"
44 #define BLOCK_OPT_ENCRYPT_FORMAT    "encrypt.format"
45 #define BLOCK_OPT_COMPAT6           "compat6"
46 #define BLOCK_OPT_HWVERSION         "hwversion"
47 #define BLOCK_OPT_BACKING_FILE      "backing_file"
48 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
49 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
50 #define BLOCK_OPT_TABLE_SIZE        "table_size"
51 #define BLOCK_OPT_PREALLOC          "preallocation"
52 #define BLOCK_OPT_SUBFMT            "subformat"
53 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
54 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
55 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
56 #define BLOCK_OPT_REDUNDANCY        "redundancy"
57 #define BLOCK_OPT_NOCOW             "nocow"
58 #define BLOCK_OPT_OBJECT_SIZE       "object_size"
59 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
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 };
68 
69 typedef struct BdrvTrackedRequest {
70     BlockDriverState *bs;
71     int64_t offset;
72     unsigned int bytes;
73     enum BdrvTrackedRequestType type;
74 
75     bool serialising;
76     int64_t overlap_offset;
77     unsigned int overlap_bytes;
78 
79     QLIST_ENTRY(BdrvTrackedRequest) list;
80     Coroutine *co; /* owner, used for deadlock detection */
81     CoQueue wait_queue; /* coroutines blocked on this request */
82 
83     struct BdrvTrackedRequest *waiting_for;
84 } BdrvTrackedRequest;
85 
86 struct BlockDriver {
87     const char *format_name;
88     int instance_size;
89 
90     /* set to true if the BlockDriver is a block filter. Block filters pass
91      * certain callbacks that refer to data (see block.c) to their bs->file if
92      * the driver doesn't implement them. Drivers that do not wish to forward
93      * must implement them and return -ENOTSUP.
94      */
95     bool is_filter;
96     /* for snapshots block filter like Quorum can implement the
97      * following recursive callback.
98      * It's purpose is to recurse on the filter children while calling
99      * bdrv_recurse_is_first_non_filter on them.
100      * For a sample implementation look in the future Quorum block filter.
101      */
102     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
103                                              BlockDriverState *candidate);
104 
105     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
106     int (*bdrv_probe_device)(const char *filename);
107 
108     /* Any driver implementing this callback is expected to be able to handle
109      * NULL file names in its .bdrv_open() implementation */
110     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
111     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
112      * this field set to true, except ones that are defined only by their
113      * child's bs.
114      * An example of the last type will be the quorum block driver.
115      */
116     bool bdrv_needs_filename;
117 
118     /* Set if a driver can support backing files */
119     bool supports_backing;
120 
121     /* For handling image reopen for split or non-split files */
122     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
123                                BlockReopenQueue *queue, Error **errp);
124     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
125     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
126     void (*bdrv_join_options)(QDict *options, QDict *old_options);
127 
128     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
129                      Error **errp);
130     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
131                           Error **errp);
132     void (*bdrv_close)(BlockDriverState *bs);
133     int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp);
134     int (*bdrv_make_empty)(BlockDriverState *bs);
135 
136     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
137 
138     /* aio */
139     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
140         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
141         BlockCompletionFunc *cb, void *opaque);
142     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
143         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
144         BlockCompletionFunc *cb, void *opaque);
145     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
146         BlockCompletionFunc *cb, void *opaque);
147     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
148         int64_t offset, int bytes,
149         BlockCompletionFunc *cb, void *opaque);
150 
151     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
152         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
153 
154     /**
155      * @offset: position in bytes to read at
156      * @bytes: number of bytes to read
157      * @qiov: the buffers to fill with read data
158      * @flags: currently unused, always 0
159      *
160      * @offset and @bytes will be a multiple of 'request_alignment',
161      * but the length of individual @qiov elements does not have to
162      * be a multiple.
163      *
164      * @bytes will always equal the total size of @qiov, and will be
165      * no larger than 'max_transfer'.
166      *
167      * The buffer in @qiov may point directly to guest memory.
168      */
169     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
170         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
171     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
172         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
173     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
174         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
175     /**
176      * @offset: position in bytes to write at
177      * @bytes: number of bytes to write
178      * @qiov: the buffers containing data to write
179      * @flags: zero or more bits allowed by 'supported_write_flags'
180      *
181      * @offset and @bytes will be a multiple of 'request_alignment',
182      * but the length of individual @qiov elements does not have to
183      * be a multiple.
184      *
185      * @bytes will always equal the total size of @qiov, and will be
186      * no larger than 'max_transfer'.
187      *
188      * The buffer in @qiov may point directly to guest memory.
189      */
190     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
191         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
192 
193     /*
194      * Efficiently zero a region of the disk image.  Typically an image format
195      * would use a compact metadata representation to implement this.  This
196      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
197      * will be called instead.
198      */
199     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
200         int64_t offset, int bytes, BdrvRequestFlags flags);
201     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
202         int64_t offset, int bytes);
203 
204     /*
205      * Building block for bdrv_block_status[_above]. The driver should
206      * answer only according to the current layer, and should not
207      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
208      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
209      */
210     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
211         int64_t sector_num, int nb_sectors, int *pnum,
212         BlockDriverState **file);
213 
214     /*
215      * Invalidate any cached meta-data.
216      */
217     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
218     int (*bdrv_inactivate)(BlockDriverState *bs);
219 
220     /*
221      * Flushes all data for all layers by calling bdrv_co_flush for underlying
222      * layers, if needed. This function is needed for deterministic
223      * synchronization of the flush finishing callback.
224      */
225     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
226 
227     /*
228      * Flushes all data that was already written to the OS all the way down to
229      * the disk (for example file-posix.c calls fsync()).
230      */
231     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
232 
233     /*
234      * Flushes all internal caches to the OS. The data may still sit in a
235      * writeback cache of the host OS, but it will survive a crash of the qemu
236      * process.
237      */
238     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
239 
240     const char *protocol_name;
241     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset,
242                          PreallocMode prealloc, Error **errp);
243 
244     int64_t (*bdrv_getlength)(BlockDriverState *bs);
245     bool has_variable_length;
246     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
247     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
248                                       Error **errp);
249 
250     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
251         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
252 
253     int (*bdrv_snapshot_create)(BlockDriverState *bs,
254                                 QEMUSnapshotInfo *sn_info);
255     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
256                               const char *snapshot_id);
257     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
258                                 const char *snapshot_id,
259                                 const char *name,
260                                 Error **errp);
261     int (*bdrv_snapshot_list)(BlockDriverState *bs,
262                               QEMUSnapshotInfo **psn_info);
263     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
264                                   const char *snapshot_id,
265                                   const char *name,
266                                   Error **errp);
267     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
268     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
269 
270     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
271                                           QEMUIOVector *qiov,
272                                           int64_t pos);
273     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
274                                           QEMUIOVector *qiov,
275                                           int64_t pos);
276 
277     int (*bdrv_change_backing_file)(BlockDriverState *bs,
278         const char *backing_file, const char *backing_fmt);
279 
280     /* removable device specific */
281     bool (*bdrv_is_inserted)(BlockDriverState *bs);
282     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
283     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
284 
285     /* to control generic scsi devices */
286     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
287         unsigned long int req, void *buf,
288         BlockCompletionFunc *cb, void *opaque);
289     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
290                                       unsigned long int req, void *buf);
291 
292     /* List of options for creating images, terminated by name == NULL */
293     QemuOptsList *create_opts;
294 
295     /*
296      * Returns 0 for completed check, -errno for internal errors.
297      * The check results are stored in result.
298      */
299     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
300         BdrvCheckMode fix);
301 
302     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
303                               BlockDriverAmendStatusCB *status_cb,
304                               void *cb_opaque);
305 
306     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
307 
308     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
309     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
310         const char *tag);
311     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
312         const char *tag);
313     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
314     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
315 
316     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
317 
318     /*
319      * Returns 1 if newly created images are guaranteed to contain only
320      * zeros, 0 otherwise.
321      */
322     int (*bdrv_has_zero_init)(BlockDriverState *bs);
323 
324     /* Remove fd handlers, timers, and other event loop callbacks so the event
325      * loop is no longer in use.  Called with no in-flight requests and in
326      * depth-first traversal order with parents before child nodes.
327      */
328     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
329 
330     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
331      * can be processed again.  Called with no in-flight requests and in
332      * depth-first traversal order with child nodes before parent nodes.
333      */
334     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
335                                     AioContext *new_context);
336 
337     /* io queue for linux-aio */
338     void (*bdrv_io_plug)(BlockDriverState *bs);
339     void (*bdrv_io_unplug)(BlockDriverState *bs);
340 
341     /**
342      * Try to get @bs's logical and physical block size.
343      * On success, store them in @bsz and return zero.
344      * On failure, return negative errno.
345      */
346     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
347     /**
348      * Try to get @bs's geometry (cyls, heads, sectors)
349      * On success, store them in @geo and return 0.
350      * On failure return -errno.
351      * Only drivers that want to override guest geometry implement this
352      * callback; see hd_geometry_guess().
353      */
354     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
355 
356     /**
357      * Drain and stop any internal sources of requests in the driver, and
358      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
359      */
360     void coroutine_fn (*bdrv_co_drain)(BlockDriverState *bs);
361 
362     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
363                            Error **errp);
364     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
365                            Error **errp);
366 
367     /**
368      * Informs the block driver that a permission change is intended. The
369      * driver checks whether the change is permissible and may take other
370      * preparations for the change (e.g. get file system locks). This operation
371      * is always followed either by a call to either .bdrv_set_perm or
372      * .bdrv_abort_perm_update.
373      *
374      * Checks whether the requested set of cumulative permissions in @perm
375      * can be granted for accessing @bs and whether no other users are using
376      * permissions other than those given in @shared (both arguments take
377      * BLK_PERM_* bitmasks).
378      *
379      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
380      * and errp is set to an error describing the conflict.
381      */
382     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
383                            uint64_t shared, Error **errp);
384 
385     /**
386      * Called to inform the driver that the set of cumulative set of used
387      * permissions for @bs has changed to @perm, and the set of sharable
388      * permission to @shared. The driver can use this to propagate changes to
389      * its children (i.e. request permissions only if a parent actually needs
390      * them).
391      *
392      * This function is only invoked after bdrv_check_perm(), so block drivers
393      * may rely on preparations made in their .bdrv_check_perm implementation.
394      */
395     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
396 
397     /*
398      * Called to inform the driver that after a previous bdrv_check_perm()
399      * call, the permission update is not performed and any preparations made
400      * for it (e.g. taken file locks) need to be undone.
401      *
402      * This function can be called even for nodes that never saw a
403      * bdrv_check_perm() call. It is a no-op then.
404      */
405     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
406 
407     /**
408      * Returns in @nperm and @nshared the permissions that the driver for @bs
409      * needs on its child @c, based on the cumulative permissions requested by
410      * the parents in @parent_perm and @parent_shared.
411      *
412      * If @c is NULL, return the permissions for attaching a new child for the
413      * given @role.
414      */
415      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
416                              const BdrvChildRole *role,
417                              uint64_t parent_perm, uint64_t parent_shared,
418                              uint64_t *nperm, uint64_t *nshared);
419 
420     /**
421      * Bitmaps should be marked as 'IN_USE' in the image on reopening image
422      * as rw. This handler should realize it. It also should unset readonly
423      * field of BlockDirtyBitmap's in case of success.
424      */
425     int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
426     bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
427                                             const char *name,
428                                             uint32_t granularity,
429                                             Error **errp);
430     void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
431                                                 const char *name,
432                                                 Error **errp);
433 
434     QLIST_ENTRY(BlockDriver) list;
435 };
436 
437 typedef struct BlockLimits {
438     /* Alignment requirement, in bytes, for offset/length of I/O
439      * requests. Must be a power of 2 less than INT_MAX; defaults to
440      * 1 for drivers with modern byte interfaces, and to 512
441      * otherwise. */
442     uint32_t request_alignment;
443 
444     /* Maximum number of bytes that can be discarded at once (since it
445      * is signed, it must be < 2G, if set). Must be multiple of
446      * pdiscard_alignment, but need not be power of 2. May be 0 if no
447      * inherent 32-bit limit */
448     int32_t max_pdiscard;
449 
450     /* Optimal alignment for discard requests in bytes. A power of 2
451      * is best but not mandatory.  Must be a multiple of
452      * bl.request_alignment, and must be less than max_pdiscard if
453      * that is set. May be 0 if bl.request_alignment is good enough */
454     uint32_t pdiscard_alignment;
455 
456     /* Maximum number of bytes that can zeroized at once (since it is
457      * signed, it must be < 2G, if set). Must be multiple of
458      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
459     int32_t max_pwrite_zeroes;
460 
461     /* Optimal alignment for write zeroes requests in bytes. A power
462      * of 2 is best but not mandatory.  Must be a multiple of
463      * bl.request_alignment, and must be less than max_pwrite_zeroes
464      * if that is set. May be 0 if bl.request_alignment is good
465      * enough */
466     uint32_t pwrite_zeroes_alignment;
467 
468     /* Optimal transfer length in bytes.  A power of 2 is best but not
469      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
470      * no preferred size */
471     uint32_t opt_transfer;
472 
473     /* Maximal transfer length in bytes.  Need not be power of 2, but
474      * must be multiple of opt_transfer and bl.request_alignment, or 0
475      * for no 32-bit limit.  For now, anything larger than INT_MAX is
476      * clamped down. */
477     uint32_t max_transfer;
478 
479     /* memory alignment, in bytes so that no bounce buffer is needed */
480     size_t min_mem_alignment;
481 
482     /* memory alignment, in bytes, for bounce buffer */
483     size_t opt_mem_alignment;
484 
485     /* maximum number of iovec elements */
486     int max_iov;
487 } BlockLimits;
488 
489 typedef struct BdrvOpBlocker BdrvOpBlocker;
490 
491 typedef struct BdrvAioNotifier {
492     void (*attached_aio_context)(AioContext *new_context, void *opaque);
493     void (*detach_aio_context)(void *opaque);
494 
495     void *opaque;
496     bool deleted;
497 
498     QLIST_ENTRY(BdrvAioNotifier) list;
499 } BdrvAioNotifier;
500 
501 struct BdrvChildRole {
502     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
503      * points to. */
504     bool stay_at_node;
505 
506     void (*inherit_options)(int *child_flags, QDict *child_options,
507                             int parent_flags, QDict *parent_options);
508 
509     void (*change_media)(BdrvChild *child, bool load);
510     void (*resize)(BdrvChild *child);
511 
512     /* Returns a name that is supposedly more useful for human users than the
513      * node name for identifying the node in question (in particular, a BB
514      * name), or NULL if the parent can't provide a better name. */
515     const char *(*get_name)(BdrvChild *child);
516 
517     /* Returns a malloced string that describes the parent of the child for a
518      * human reader. This could be a node-name, BlockBackend name, qdev ID or
519      * QOM path of the device owning the BlockBackend, job type and ID etc. The
520      * caller is responsible for freeing the memory. */
521     char *(*get_parent_desc)(BdrvChild *child);
522 
523     /*
524      * If this pair of functions is implemented, the parent doesn't issue new
525      * requests after returning from .drained_begin() until .drained_end() is
526      * called.
527      *
528      * Note that this can be nested. If drained_begin() was called twice, new
529      * I/O is allowed only after drained_end() was called twice, too.
530      */
531     void (*drained_begin)(BdrvChild *child);
532     void (*drained_end)(BdrvChild *child);
533 
534     /* Notifies the parent that the child has been activated/inactivated (e.g.
535      * when migration is completing) and it can start/stop requesting
536      * permissions and doing I/O on it. */
537     void (*activate)(BdrvChild *child, Error **errp);
538     int (*inactivate)(BdrvChild *child);
539 
540     void (*attach)(BdrvChild *child);
541     void (*detach)(BdrvChild *child);
542 };
543 
544 extern const BdrvChildRole child_file;
545 extern const BdrvChildRole child_format;
546 extern const BdrvChildRole child_backing;
547 
548 struct BdrvChild {
549     BlockDriverState *bs;
550     char *name;
551     const BdrvChildRole *role;
552     void *opaque;
553 
554     /**
555      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
556      */
557     uint64_t perm;
558 
559     /**
560      * Permissions that can still be granted to other users of @bs while this
561      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
562      */
563     uint64_t shared_perm;
564 
565     QLIST_ENTRY(BdrvChild) next;
566     QLIST_ENTRY(BdrvChild) next_parent;
567 };
568 
569 /*
570  * Note: the function bdrv_append() copies and swaps contents of
571  * BlockDriverStates, so if you add new fields to this struct, please
572  * inspect bdrv_append() to determine if the new fields need to be
573  * copied as well.
574  */
575 struct BlockDriverState {
576     /* Protected by big QEMU lock or read-only after opening.  No special
577      * locking needed during I/O...
578      */
579     int open_flags; /* flags used to open the file, re-used for re-open */
580     bool read_only; /* if true, the media is read only */
581     bool encrypted; /* if true, the media is encrypted */
582     bool sg;        /* if true, the device is a /dev/sg* */
583     bool probed;    /* if true, format was probed rather than specified */
584     bool force_share; /* if true, always allow all shared permissions */
585     bool implicit;  /* if true, this filter node was automatically inserted */
586 
587     BlockDriver *drv; /* NULL means no media */
588     void *opaque;
589 
590     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
591     /* long-running tasks intended to always use the same AioContext as this
592      * BDS may register themselves in this list to be notified of changes
593      * regarding this BDS's context */
594     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
595     bool walking_aio_notifiers; /* to make removal during iteration safe */
596 
597     char filename[PATH_MAX];
598     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
599                                     this file image */
600     char backing_format[16]; /* if non-zero and backing_file exists */
601 
602     QDict *full_open_options;
603     char exact_filename[PATH_MAX];
604 
605     BdrvChild *backing;
606     BdrvChild *file;
607 
608     /* I/O Limits */
609     BlockLimits bl;
610 
611     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
612     unsigned int supported_write_flags;
613     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
614      * BDRV_REQ_MAY_UNMAP) */
615     unsigned int supported_zero_flags;
616 
617     /* the following member gives a name to every node on the bs graph. */
618     char node_name[32];
619     /* element of the list of named nodes building the graph */
620     QTAILQ_ENTRY(BlockDriverState) node_list;
621     /* element of the list of all BlockDriverStates (all_bdrv_states) */
622     QTAILQ_ENTRY(BlockDriverState) bs_list;
623     /* element of the list of monitor-owned BDS */
624     QTAILQ_ENTRY(BlockDriverState) monitor_list;
625     int refcnt;
626 
627     /* operation blockers */
628     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
629 
630     /* long-running background operation */
631     BlockJob *job;
632 
633     /* The node that this node inherited default options from (and a reopen on
634      * which can affect this node by changing these defaults). This is always a
635      * parent node of this node. */
636     BlockDriverState *inherits_from;
637     QLIST_HEAD(, BdrvChild) children;
638     QLIST_HEAD(, BdrvChild) parents;
639 
640     QDict *options;
641     QDict *explicit_options;
642     BlockdevDetectZeroesOptions detect_zeroes;
643 
644     /* The error object in use for blocking operations on backing_hd */
645     Error *backing_blocker;
646 
647     /* Protected by AioContext lock */
648 
649     /* If we are reading a disk image, give its size in sectors.
650      * Generally read-only; it is written to by load_snapshot and
651      * save_snaphost, but the block layer is quiescent during those.
652      */
653     int64_t total_sectors;
654 
655     /* Callback before write request is processed */
656     NotifierWithReturnList before_write_notifiers;
657 
658     /* threshold limit for writes, in bytes. "High water mark". */
659     uint64_t write_threshold_offset;
660     NotifierWithReturn write_threshold_notifier;
661 
662     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
663      * Reading from the list can be done with either the BQL or the
664      * dirty_bitmap_mutex.  Modifying a bitmap only requires
665      * dirty_bitmap_mutex.  */
666     QemuMutex dirty_bitmap_mutex;
667     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
668 
669     /* Offset after the highest byte written to */
670     Stat64 wr_highest_offset;
671 
672     /* If true, copy read backing sectors into image.  Can be >1 if more
673      * than one client has requested copy-on-read.  Accessed with atomic
674      * ops.
675      */
676     int copy_on_read;
677 
678     /* number of in-flight requests; overall and serialising.
679      * Accessed with atomic ops.
680      */
681     unsigned int in_flight;
682     unsigned int serialising_in_flight;
683 
684     /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
685      * ops.
686      */
687     bool wakeup;
688 
689     /* counter for nested bdrv_io_plug.
690      * Accessed with atomic ops.
691     */
692     unsigned io_plugged;
693 
694     /* do we need to tell the quest if we have a volatile write cache? */
695     int enable_write_cache;
696 
697     /* Accessed with atomic ops.  */
698     int quiesce_counter;
699     unsigned int write_gen;               /* Current data generation */
700 
701     /* Protected by reqs_lock.  */
702     CoMutex reqs_lock;
703     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
704     CoQueue flush_queue;                  /* Serializing flush queue */
705     bool active_flush_req;                /* Flush request in flight? */
706 
707     /* Only read/written by whoever has set active_flush_req to true.  */
708     unsigned int flushed_gen;             /* Flushed write generation */
709 };
710 
711 struct BlockBackendRootState {
712     int open_flags;
713     bool read_only;
714     BlockdevDetectZeroesOptions detect_zeroes;
715 };
716 
717 typedef enum BlockMirrorBackingMode {
718     /* Reuse the existing backing chain from the source for the target.
719      * - sync=full: Set backing BDS to NULL.
720      * - sync=top:  Use source's backing BDS.
721      * - sync=none: Use source as the backing BDS. */
722     MIRROR_SOURCE_BACKING_CHAIN,
723 
724     /* Open the target's backing chain completely anew */
725     MIRROR_OPEN_BACKING_CHAIN,
726 
727     /* Do not change the target's backing BDS after job completion */
728     MIRROR_LEAVE_BACKING_CHAIN,
729 } BlockMirrorBackingMode;
730 
731 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
732 {
733     return bs->backing ? bs->backing->bs : NULL;
734 }
735 
736 
737 /* Essential block drivers which must always be statically linked into qemu, and
738  * which therefore can be accessed without using bdrv_find_format() */
739 extern BlockDriver bdrv_file;
740 extern BlockDriver bdrv_raw;
741 extern BlockDriver bdrv_qcow2;
742 
743 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
744     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
745     BdrvRequestFlags flags);
746 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
747     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
748     BdrvRequestFlags flags);
749 
750 int get_tmp_filename(char *filename, int size);
751 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
752                             const char *filename);
753 
754 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
755                                       QDict *options);
756 
757 
758 /**
759  * bdrv_add_before_write_notifier:
760  *
761  * Register a callback that is invoked before write requests are processed but
762  * after any throttling or waiting for overlapping requests.
763  */
764 void bdrv_add_before_write_notifier(BlockDriverState *bs,
765                                     NotifierWithReturn *notifier);
766 
767 /**
768  * bdrv_detach_aio_context:
769  *
770  * May be called from .bdrv_detach_aio_context() to detach children from the
771  * current #AioContext.  This is only needed by block drivers that manage their
772  * own children.  Both ->file and ->backing are automatically handled and
773  * block drivers should not call this function on them explicitly.
774  */
775 void bdrv_detach_aio_context(BlockDriverState *bs);
776 
777 /**
778  * bdrv_attach_aio_context:
779  *
780  * May be called from .bdrv_attach_aio_context() to attach children to the new
781  * #AioContext.  This is only needed by block drivers that manage their own
782  * children.  Both ->file and ->backing are automatically handled and block
783  * drivers should not call this function on them explicitly.
784  */
785 void bdrv_attach_aio_context(BlockDriverState *bs,
786                              AioContext *new_context);
787 
788 /**
789  * bdrv_add_aio_context_notifier:
790  *
791  * If a long-running job intends to be always run in the same AioContext as a
792  * certain BDS, it may use this function to be notified of changes regarding the
793  * association of the BDS to an AioContext.
794  *
795  * attached_aio_context() is called after the target BDS has been attached to a
796  * new AioContext; detach_aio_context() is called before the target BDS is being
797  * detached from its old AioContext.
798  */
799 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
800         void (*attached_aio_context)(AioContext *new_context, void *opaque),
801         void (*detach_aio_context)(void *opaque), void *opaque);
802 
803 /**
804  * bdrv_remove_aio_context_notifier:
805  *
806  * Unsubscribe of change notifications regarding the BDS's AioContext. The
807  * parameters given here have to be the same as those given to
808  * bdrv_add_aio_context_notifier().
809  */
810 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
811                                       void (*aio_context_attached)(AioContext *,
812                                                                    void *),
813                                       void (*aio_context_detached)(void *),
814                                       void *opaque);
815 
816 /**
817  * bdrv_wakeup:
818  * @bs: The BlockDriverState for which an I/O operation has been completed.
819  *
820  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
821  * synchronous I/O on a BlockDriverState that is attached to another
822  * I/O thread, the main thread lets the I/O thread's event loop run,
823  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
824  * up the main thread if necessary.
825  *
826  * Manual calls to bdrv_wakeup are rarely necessary, because
827  * bdrv_dec_in_flight already calls it.
828  */
829 void bdrv_wakeup(BlockDriverState *bs);
830 
831 #ifdef _WIN32
832 int is_windows_drive(const char *filename);
833 #endif
834 
835 /**
836  * stream_start:
837  * @job_id: The id of the newly-created job, or %NULL to use the
838  * device name of @bs.
839  * @bs: Block device to operate on.
840  * @base: Block device that will become the new base, or %NULL to
841  * flatten the whole backing file chain onto @bs.
842  * @backing_file_str: The file name that will be written to @bs as the
843  * the new backing file if the job completes. Ignored if @base is %NULL.
844  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
845  * @on_error: The action to take upon error.
846  * @errp: Error object.
847  *
848  * Start a streaming operation on @bs.  Clusters that are unallocated
849  * in @bs, but allocated in any image between @base and @bs (both
850  * exclusive) will be written to @bs.  At the end of a successful
851  * streaming job, the backing file of @bs will be changed to
852  * @backing_file_str in the written image and to @base in the live
853  * BlockDriverState.
854  */
855 void stream_start(const char *job_id, BlockDriverState *bs,
856                   BlockDriverState *base, const char *backing_file_str,
857                   int64_t speed, BlockdevOnError on_error, Error **errp);
858 
859 /**
860  * commit_start:
861  * @job_id: The id of the newly-created job, or %NULL to use the
862  * device name of @bs.
863  * @bs: Active block device.
864  * @top: Top block device to be committed.
865  * @base: Block device that will be written into, and become the new top.
866  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
867  * @on_error: The action to take upon error.
868  * @backing_file_str: String to use as the backing file in @top's overlay
869  * @filter_node_name: The node name that should be assigned to the filter
870  * driver that the commit job inserts into the graph above @top. NULL means
871  * that a node name should be autogenerated.
872  * @errp: Error object.
873  *
874  */
875 void commit_start(const char *job_id, BlockDriverState *bs,
876                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
877                   BlockdevOnError on_error, const char *backing_file_str,
878                   const char *filter_node_name, Error **errp);
879 /**
880  * commit_active_start:
881  * @job_id: The id of the newly-created job, or %NULL to use the
882  * device name of @bs.
883  * @bs: Active block device to be committed.
884  * @base: Block device that will be written into, and become the new top.
885  * @creation_flags: Flags that control the behavior of the Job lifetime.
886  *                  See @BlockJobCreateFlags
887  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
888  * @on_error: The action to take upon error.
889  * @filter_node_name: The node name that should be assigned to the filter
890  * driver that the commit job inserts into the graph above @bs. NULL means that
891  * a node name should be autogenerated.
892  * @cb: Completion function for the job.
893  * @opaque: Opaque pointer value passed to @cb.
894  * @auto_complete: Auto complete the job.
895  * @errp: Error object.
896  *
897  */
898 void commit_active_start(const char *job_id, BlockDriverState *bs,
899                          BlockDriverState *base, int creation_flags,
900                          int64_t speed, BlockdevOnError on_error,
901                          const char *filter_node_name,
902                          BlockCompletionFunc *cb, void *opaque,
903                          bool auto_complete, Error **errp);
904 /*
905  * mirror_start:
906  * @job_id: The id of the newly-created job, or %NULL to use the
907  * device name of @bs.
908  * @bs: Block device to operate on.
909  * @target: Block device to write to.
910  * @replaces: Block graph node name to replace once the mirror is done. Can
911  *            only be used when full mirroring is selected.
912  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
913  * @granularity: The chosen granularity for the dirty bitmap.
914  * @buf_size: The amount of data that can be in flight at one time.
915  * @mode: Whether to collapse all images in the chain to the target.
916  * @backing_mode: How to establish the target's backing chain after completion.
917  * @on_source_error: The action to take upon error reading from the source.
918  * @on_target_error: The action to take upon error writing to the target.
919  * @unmap: Whether to unmap target where source sectors only contain zeroes.
920  * @filter_node_name: The node name that should be assigned to the filter
921  * driver that the mirror job inserts into the graph above @bs. NULL means that
922  * a node name should be autogenerated.
923  * @errp: Error object.
924  *
925  * Start a mirroring operation on @bs.  Clusters that are allocated
926  * in @bs will be written to @target until the job is cancelled or
927  * manually completed.  At the end of a successful mirroring job,
928  * @bs will be switched to read from @target.
929  */
930 void mirror_start(const char *job_id, BlockDriverState *bs,
931                   BlockDriverState *target, const char *replaces,
932                   int64_t speed, uint32_t granularity, int64_t buf_size,
933                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
934                   BlockdevOnError on_source_error,
935                   BlockdevOnError on_target_error,
936                   bool unmap, const char *filter_node_name, Error **errp);
937 
938 /*
939  * backup_job_create:
940  * @job_id: The id of the newly-created job, or %NULL to use the
941  * device name of @bs.
942  * @bs: Block device to operate on.
943  * @target: Block device to write to.
944  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
945  * @sync_mode: What parts of the disk image should be copied to the destination.
946  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
947  * @on_source_error: The action to take upon error reading from the source.
948  * @on_target_error: The action to take upon error writing to the target.
949  * @creation_flags: Flags that control the behavior of the Job lifetime.
950  *                  See @BlockJobCreateFlags
951  * @cb: Completion function for the job.
952  * @opaque: Opaque pointer value passed to @cb.
953  * @txn: Transaction that this job is part of (may be NULL).
954  *
955  * Create a backup operation on @bs.  Clusters in @bs are written to @target
956  * until the job is cancelled or manually completed.
957  */
958 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
959                             BlockDriverState *target, int64_t speed,
960                             MirrorSyncMode sync_mode,
961                             BdrvDirtyBitmap *sync_bitmap,
962                             bool compress,
963                             BlockdevOnError on_source_error,
964                             BlockdevOnError on_target_error,
965                             int creation_flags,
966                             BlockCompletionFunc *cb, void *opaque,
967                             BlockJobTxn *txn, Error **errp);
968 
969 void hmp_drive_add_node(Monitor *mon, const char *optstr);
970 
971 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
972                                   const char *child_name,
973                                   const BdrvChildRole *child_role,
974                                   uint64_t perm, uint64_t shared_perm,
975                                   void *opaque, Error **errp);
976 void bdrv_root_unref_child(BdrvChild *child);
977 
978 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
979                             Error **errp);
980 
981 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
982  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
983  * all children */
984 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
985                                const BdrvChildRole *role,
986                                uint64_t perm, uint64_t shared,
987                                uint64_t *nperm, uint64_t *nshared);
988 
989 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
990  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
991  * requires WRITE | RESIZE for read-write images, always requires
992  * CONSISTENT_READ and doesn't share WRITE. */
993 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
994                                const BdrvChildRole *role,
995                                uint64_t perm, uint64_t shared,
996                                uint64_t *nperm, uint64_t *nshared);
997 
998 /*
999  * Default implementation for drivers to pass bdrv_co_get_block_status() to
1000  * their file.
1001  */
1002 int64_t coroutine_fn bdrv_co_get_block_status_from_file(BlockDriverState *bs,
1003                                                         int64_t sector_num,
1004                                                         int nb_sectors,
1005                                                         int *pnum,
1006                                                         BlockDriverState **file);
1007 /*
1008  * Default implementation for drivers to pass bdrv_co_get_block_status() to
1009  * their backing file.
1010  */
1011 int64_t coroutine_fn bdrv_co_get_block_status_from_backing(BlockDriverState *bs,
1012                                                            int64_t sector_num,
1013                                                            int nb_sectors,
1014                                                            int *pnum,
1015                                                            BlockDriverState **file);
1016 const char *bdrv_get_parent_name(const BlockDriverState *bs);
1017 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
1018 bool blk_dev_has_removable_media(BlockBackend *blk);
1019 bool blk_dev_has_tray(BlockBackend *blk);
1020 void blk_dev_eject_request(BlockBackend *blk, bool force);
1021 bool blk_dev_is_tray_open(BlockBackend *blk);
1022 bool blk_dev_is_medium_locked(BlockBackend *blk);
1023 
1024 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
1025 bool bdrv_requests_pending(BlockDriverState *bs);
1026 
1027 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
1028 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
1029 
1030 void bdrv_inc_in_flight(BlockDriverState *bs);
1031 void bdrv_dec_in_flight(BlockDriverState *bs);
1032 
1033 void blockdev_close_all_bdrv_states(void);
1034 
1035 #endif /* BLOCK_INT_H */
1036