xref: /openbmc/qemu/include/block/block_int.h (revision eab15862)
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 "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/main-loop.h"
37 #include "qemu/throttle.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_OBJECT_SIZE       "object_size"
58 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
59 
60 #define BLOCK_PROBE_BUF_SIZE        512
61 
62 enum BdrvTrackedRequestType {
63     BDRV_TRACKED_READ,
64     BDRV_TRACKED_WRITE,
65     BDRV_TRACKED_DISCARD,
66 };
67 
68 typedef struct BdrvTrackedRequest {
69     BlockDriverState *bs;
70     int64_t offset;
71     unsigned int bytes;
72     enum BdrvTrackedRequestType type;
73 
74     bool serialising;
75     int64_t overlap_offset;
76     unsigned int overlap_bytes;
77 
78     QLIST_ENTRY(BdrvTrackedRequest) list;
79     Coroutine *co; /* owner, used for deadlock detection */
80     CoQueue wait_queue; /* coroutines blocked on this request */
81 
82     struct BdrvTrackedRequest *waiting_for;
83 } BdrvTrackedRequest;
84 
85 struct BlockDriver {
86     const char *format_name;
87     int instance_size;
88 
89     /* set to true if the BlockDriver is a block filter. Block filters pass
90      * certain callbacks that refer to data (see block.c) to their bs->file if
91      * the driver doesn't implement them. Drivers that do not wish to forward
92      * must implement them and return -ENOTSUP.
93      */
94     bool is_filter;
95     /* for snapshots block filter like Quorum can implement the
96      * following recursive callback.
97      * It's purpose is to recurse on the filter children while calling
98      * bdrv_recurse_is_first_non_filter on them.
99      * For a sample implementation look in the future Quorum block filter.
100      */
101     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
102                                              BlockDriverState *candidate);
103 
104     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
105     int (*bdrv_probe_device)(const char *filename);
106 
107     /* Any driver implementing this callback is expected to be able to handle
108      * NULL file names in its .bdrv_open() implementation */
109     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
110     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
111      * this field set to true, except ones that are defined only by their
112      * child's bs.
113      * An example of the last type will be the quorum block driver.
114      */
115     bool bdrv_needs_filename;
116 
117     /* Set if a driver can support backing files */
118     bool supports_backing;
119 
120     /* For handling image reopen for split or non-split files */
121     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
122                                BlockReopenQueue *queue, Error **errp);
123     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
124     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
125     void (*bdrv_join_options)(QDict *options, QDict *old_options);
126 
127     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
128                      Error **errp);
129 
130     /* Protocol drivers should implement this instead of bdrv_open */
131     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
132                           Error **errp);
133     void (*bdrv_close)(BlockDriverState *bs);
134     int coroutine_fn (*bdrv_co_create)(BlockdevCreateOptions *opts,
135                                        Error **errp);
136     int coroutine_fn (*bdrv_co_create_opts)(const char *filename,
137                                             QemuOpts *opts,
138                                             Error **errp);
139     int (*bdrv_make_empty)(BlockDriverState *bs);
140 
141     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
142 
143     /* aio */
144     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
145         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
146         BlockCompletionFunc *cb, void *opaque);
147     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
148         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
149         BlockCompletionFunc *cb, void *opaque);
150     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
151         BlockCompletionFunc *cb, void *opaque);
152     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
153         int64_t offset, int bytes,
154         BlockCompletionFunc *cb, void *opaque);
155 
156     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
157         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
158 
159     /**
160      * @offset: position in bytes to read at
161      * @bytes: number of bytes to read
162      * @qiov: the buffers to fill with read data
163      * @flags: currently unused, always 0
164      *
165      * @offset and @bytes will be a multiple of 'request_alignment',
166      * but the length of individual @qiov elements does not have to
167      * be a multiple.
168      *
169      * @bytes will always equal the total size of @qiov, and will be
170      * no larger than 'max_transfer'.
171      *
172      * The buffer in @qiov may point directly to guest memory.
173      */
174     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
175         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
176     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
177         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
178     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
179         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
180     /**
181      * @offset: position in bytes to write at
182      * @bytes: number of bytes to write
183      * @qiov: the buffers containing data to write
184      * @flags: zero or more bits allowed by 'supported_write_flags'
185      *
186      * @offset and @bytes will be a multiple of 'request_alignment',
187      * but the length of individual @qiov elements does not have to
188      * be a multiple.
189      *
190      * @bytes will always equal the total size of @qiov, and will be
191      * no larger than 'max_transfer'.
192      *
193      * The buffer in @qiov may point directly to guest memory.
194      */
195     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
196         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
197 
198     /*
199      * Efficiently zero a region of the disk image.  Typically an image format
200      * would use a compact metadata representation to implement this.  This
201      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
202      * will be called instead.
203      */
204     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
205         int64_t offset, int bytes, BdrvRequestFlags flags);
206     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
207         int64_t offset, int bytes);
208 
209     /*
210      * Building block for bdrv_block_status[_above] and
211      * bdrv_is_allocated[_above].  The driver should answer only
212      * according to the current layer, and should only need to set
213      * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID,
214      * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing
215      * layer, the result should be 0 (and not BDRV_BLOCK_ZERO).  See
216      * block.h for the overall meaning of the bits.  As a hint, the
217      * flag want_zero is true if the caller cares more about precise
218      * mappings (favor accurate _OFFSET_VALID/_ZERO) or false for
219      * overall allocation (favor larger *pnum, perhaps by reporting
220      * _DATA instead of _ZERO).  The block layer guarantees input
221      * clamped to bdrv_getlength() and aligned to request_alignment,
222      * as well as non-NULL pnum, map, and file; in turn, the driver
223      * must return an error or set pnum to an aligned non-zero value.
224      */
225     int coroutine_fn (*bdrv_co_block_status)(BlockDriverState *bs,
226         bool want_zero, int64_t offset, int64_t bytes, int64_t *pnum,
227         int64_t *map, BlockDriverState **file);
228 
229     /*
230      * Invalidate any cached meta-data.
231      */
232     void coroutine_fn (*bdrv_co_invalidate_cache)(BlockDriverState *bs,
233                                                   Error **errp);
234     int (*bdrv_inactivate)(BlockDriverState *bs);
235 
236     /*
237      * Flushes all data for all layers by calling bdrv_co_flush for underlying
238      * layers, if needed. This function is needed for deterministic
239      * synchronization of the flush finishing callback.
240      */
241     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
242 
243     /*
244      * Flushes all data that was already written to the OS all the way down to
245      * the disk (for example file-posix.c calls fsync()).
246      */
247     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
248 
249     /*
250      * Flushes all internal caches to the OS. The data may still sit in a
251      * writeback cache of the host OS, but it will survive a crash of the qemu
252      * process.
253      */
254     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
255 
256     /*
257      * Drivers setting this field must be able to work with just a plain
258      * filename with '<protocol_name>:' as a prefix, and no other options.
259      * Options may be extracted from the filename by implementing
260      * bdrv_parse_filename.
261      */
262     const char *protocol_name;
263     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset,
264                          PreallocMode prealloc, Error **errp);
265 
266     int64_t (*bdrv_getlength)(BlockDriverState *bs);
267     bool has_variable_length;
268     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
269     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
270                                       Error **errp);
271 
272     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
273         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
274 
275     int (*bdrv_snapshot_create)(BlockDriverState *bs,
276                                 QEMUSnapshotInfo *sn_info);
277     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
278                               const char *snapshot_id);
279     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
280                                 const char *snapshot_id,
281                                 const char *name,
282                                 Error **errp);
283     int (*bdrv_snapshot_list)(BlockDriverState *bs,
284                               QEMUSnapshotInfo **psn_info);
285     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
286                                   const char *snapshot_id,
287                                   const char *name,
288                                   Error **errp);
289     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
290     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
291 
292     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
293                                           QEMUIOVector *qiov,
294                                           int64_t pos);
295     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
296                                           QEMUIOVector *qiov,
297                                           int64_t pos);
298 
299     int (*bdrv_change_backing_file)(BlockDriverState *bs,
300         const char *backing_file, const char *backing_fmt);
301 
302     /* removable device specific */
303     bool (*bdrv_is_inserted)(BlockDriverState *bs);
304     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
305     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
306 
307     /* to control generic scsi devices */
308     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
309         unsigned long int req, void *buf,
310         BlockCompletionFunc *cb, void *opaque);
311     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
312                                       unsigned long int req, void *buf);
313 
314     /* List of options for creating images, terminated by name == NULL */
315     QemuOptsList *create_opts;
316 
317     /*
318      * Returns 0 for completed check, -errno for internal errors.
319      * The check results are stored in result.
320      */
321     int coroutine_fn (*bdrv_co_check)(BlockDriverState *bs,
322                                       BdrvCheckResult *result,
323                                       BdrvCheckMode fix);
324 
325     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
326                               BlockDriverAmendStatusCB *status_cb,
327                               void *cb_opaque);
328 
329     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
330 
331     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
332     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
333         const char *tag);
334     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
335         const char *tag);
336     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
337     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
338 
339     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
340 
341     /*
342      * Returns 1 if newly created images are guaranteed to contain only
343      * zeros, 0 otherwise.
344      */
345     int (*bdrv_has_zero_init)(BlockDriverState *bs);
346 
347     /* Remove fd handlers, timers, and other event loop callbacks so the event
348      * loop is no longer in use.  Called with no in-flight requests and in
349      * depth-first traversal order with parents before child nodes.
350      */
351     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
352 
353     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
354      * can be processed again.  Called with no in-flight requests and in
355      * depth-first traversal order with child nodes before parent nodes.
356      */
357     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
358                                     AioContext *new_context);
359 
360     /* io queue for linux-aio */
361     void (*bdrv_io_plug)(BlockDriverState *bs);
362     void (*bdrv_io_unplug)(BlockDriverState *bs);
363 
364     /**
365      * Try to get @bs's logical and physical block size.
366      * On success, store them in @bsz and return zero.
367      * On failure, return negative errno.
368      */
369     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
370     /**
371      * Try to get @bs's geometry (cyls, heads, sectors)
372      * On success, store them in @geo and return 0.
373      * On failure return -errno.
374      * Only drivers that want to override guest geometry implement this
375      * callback; see hd_geometry_guess().
376      */
377     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
378 
379     /**
380      * bdrv_co_drain_begin is called if implemented in the beginning of a
381      * drain operation to drain and stop any internal sources of requests in
382      * the driver.
383      * bdrv_co_drain_end is called if implemented at the end of the drain.
384      *
385      * They should be used by the driver to e.g. manage scheduled I/O
386      * requests, or toggle an internal state. After the end of the drain new
387      * requests will continue normally.
388      */
389     void coroutine_fn (*bdrv_co_drain_begin)(BlockDriverState *bs);
390     void coroutine_fn (*bdrv_co_drain_end)(BlockDriverState *bs);
391 
392     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
393                            Error **errp);
394     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
395                            Error **errp);
396 
397     /**
398      * Informs the block driver that a permission change is intended. The
399      * driver checks whether the change is permissible and may take other
400      * preparations for the change (e.g. get file system locks). This operation
401      * is always followed either by a call to either .bdrv_set_perm or
402      * .bdrv_abort_perm_update.
403      *
404      * Checks whether the requested set of cumulative permissions in @perm
405      * can be granted for accessing @bs and whether no other users are using
406      * permissions other than those given in @shared (both arguments take
407      * BLK_PERM_* bitmasks).
408      *
409      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
410      * and errp is set to an error describing the conflict.
411      */
412     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
413                            uint64_t shared, Error **errp);
414 
415     /**
416      * Called to inform the driver that the set of cumulative set of used
417      * permissions for @bs has changed to @perm, and the set of sharable
418      * permission to @shared. The driver can use this to propagate changes to
419      * its children (i.e. request permissions only if a parent actually needs
420      * them).
421      *
422      * This function is only invoked after bdrv_check_perm(), so block drivers
423      * may rely on preparations made in their .bdrv_check_perm implementation.
424      */
425     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
426 
427     /*
428      * Called to inform the driver that after a previous bdrv_check_perm()
429      * call, the permission update is not performed and any preparations made
430      * for it (e.g. taken file locks) need to be undone.
431      *
432      * This function can be called even for nodes that never saw a
433      * bdrv_check_perm() call. It is a no-op then.
434      */
435     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
436 
437     /**
438      * Returns in @nperm and @nshared the permissions that the driver for @bs
439      * needs on its child @c, based on the cumulative permissions requested by
440      * the parents in @parent_perm and @parent_shared.
441      *
442      * If @c is NULL, return the permissions for attaching a new child for the
443      * given @role.
444      *
445      * If @reopen_queue is non-NULL, don't return the currently needed
446      * permissions, but those that will be needed after applying the
447      * @reopen_queue.
448      */
449      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
450                              const BdrvChildRole *role,
451                              BlockReopenQueue *reopen_queue,
452                              uint64_t parent_perm, uint64_t parent_shared,
453                              uint64_t *nperm, uint64_t *nshared);
454 
455     /**
456      * Bitmaps should be marked as 'IN_USE' in the image on reopening image
457      * as rw. This handler should realize it. It also should unset readonly
458      * field of BlockDirtyBitmap's in case of success.
459      */
460     int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
461     bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
462                                             const char *name,
463                                             uint32_t granularity,
464                                             Error **errp);
465     void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
466                                                 const char *name,
467                                                 Error **errp);
468 
469     /**
470      * Register/unregister a buffer for I/O. For example, when the driver is
471      * interested to know the memory areas that will later be used in iovs, so
472      * that it can do IOMMU mapping with VFIO etc., in order to get better
473      * performance. In the case of VFIO drivers, this callback is used to do
474      * DMA mapping for hot buffers.
475      */
476     void (*bdrv_register_buf)(BlockDriverState *bs, void *host, size_t size);
477     void (*bdrv_unregister_buf)(BlockDriverState *bs, void *host);
478     QLIST_ENTRY(BlockDriver) list;
479 };
480 
481 typedef struct BlockLimits {
482     /* Alignment requirement, in bytes, for offset/length of I/O
483      * requests. Must be a power of 2 less than INT_MAX; defaults to
484      * 1 for drivers with modern byte interfaces, and to 512
485      * otherwise. */
486     uint32_t request_alignment;
487 
488     /* Maximum number of bytes that can be discarded at once (since it
489      * is signed, it must be < 2G, if set). Must be multiple of
490      * pdiscard_alignment, but need not be power of 2. May be 0 if no
491      * inherent 32-bit limit */
492     int32_t max_pdiscard;
493 
494     /* Optimal alignment for discard requests in bytes. A power of 2
495      * is best but not mandatory.  Must be a multiple of
496      * bl.request_alignment, and must be less than max_pdiscard if
497      * that is set. May be 0 if bl.request_alignment is good enough */
498     uint32_t pdiscard_alignment;
499 
500     /* Maximum number of bytes that can zeroized at once (since it is
501      * signed, it must be < 2G, if set). Must be multiple of
502      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
503     int32_t max_pwrite_zeroes;
504 
505     /* Optimal alignment for write zeroes requests in bytes. A power
506      * of 2 is best but not mandatory.  Must be a multiple of
507      * bl.request_alignment, and must be less than max_pwrite_zeroes
508      * if that is set. May be 0 if bl.request_alignment is good
509      * enough */
510     uint32_t pwrite_zeroes_alignment;
511 
512     /* Optimal transfer length in bytes.  A power of 2 is best but not
513      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
514      * no preferred size */
515     uint32_t opt_transfer;
516 
517     /* Maximal transfer length in bytes.  Need not be power of 2, but
518      * must be multiple of opt_transfer and bl.request_alignment, or 0
519      * for no 32-bit limit.  For now, anything larger than INT_MAX is
520      * clamped down. */
521     uint32_t max_transfer;
522 
523     /* memory alignment, in bytes so that no bounce buffer is needed */
524     size_t min_mem_alignment;
525 
526     /* memory alignment, in bytes, for bounce buffer */
527     size_t opt_mem_alignment;
528 
529     /* maximum number of iovec elements */
530     int max_iov;
531 } BlockLimits;
532 
533 typedef struct BdrvOpBlocker BdrvOpBlocker;
534 
535 typedef struct BdrvAioNotifier {
536     void (*attached_aio_context)(AioContext *new_context, void *opaque);
537     void (*detach_aio_context)(void *opaque);
538 
539     void *opaque;
540     bool deleted;
541 
542     QLIST_ENTRY(BdrvAioNotifier) list;
543 } BdrvAioNotifier;
544 
545 struct BdrvChildRole {
546     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
547      * points to. */
548     bool stay_at_node;
549 
550     void (*inherit_options)(int *child_flags, QDict *child_options,
551                             int parent_flags, QDict *parent_options);
552 
553     void (*change_media)(BdrvChild *child, bool load);
554     void (*resize)(BdrvChild *child);
555 
556     /* Returns a name that is supposedly more useful for human users than the
557      * node name for identifying the node in question (in particular, a BB
558      * name), or NULL if the parent can't provide a better name. */
559     const char *(*get_name)(BdrvChild *child);
560 
561     /* Returns a malloced string that describes the parent of the child for a
562      * human reader. This could be a node-name, BlockBackend name, qdev ID or
563      * QOM path of the device owning the BlockBackend, job type and ID etc. The
564      * caller is responsible for freeing the memory. */
565     char *(*get_parent_desc)(BdrvChild *child);
566 
567     /*
568      * If this pair of functions is implemented, the parent doesn't issue new
569      * requests after returning from .drained_begin() until .drained_end() is
570      * called.
571      *
572      * Note that this can be nested. If drained_begin() was called twice, new
573      * I/O is allowed only after drained_end() was called twice, too.
574      */
575     void (*drained_begin)(BdrvChild *child);
576     void (*drained_end)(BdrvChild *child);
577 
578     /* Notifies the parent that the child has been activated/inactivated (e.g.
579      * when migration is completing) and it can start/stop requesting
580      * permissions and doing I/O on it. */
581     void (*activate)(BdrvChild *child, Error **errp);
582     int (*inactivate)(BdrvChild *child);
583 
584     void (*attach)(BdrvChild *child);
585     void (*detach)(BdrvChild *child);
586 
587     /* Notifies the parent that the filename of its child has changed (e.g.
588      * because the direct child was removed from the backing chain), so that it
589      * can update its reference. */
590     int (*update_filename)(BdrvChild *child, BlockDriverState *new_base,
591                            const char *filename, Error **errp);
592 };
593 
594 extern const BdrvChildRole child_file;
595 extern const BdrvChildRole child_format;
596 extern const BdrvChildRole child_backing;
597 
598 struct BdrvChild {
599     BlockDriverState *bs;
600     char *name;
601     const BdrvChildRole *role;
602     void *opaque;
603 
604     /**
605      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
606      */
607     uint64_t perm;
608 
609     /**
610      * Permissions that can still be granted to other users of @bs while this
611      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
612      */
613     uint64_t shared_perm;
614 
615     QLIST_ENTRY(BdrvChild) next;
616     QLIST_ENTRY(BdrvChild) next_parent;
617 };
618 
619 /*
620  * Note: the function bdrv_append() copies and swaps contents of
621  * BlockDriverStates, so if you add new fields to this struct, please
622  * inspect bdrv_append() to determine if the new fields need to be
623  * copied as well.
624  */
625 struct BlockDriverState {
626     /* Protected by big QEMU lock or read-only after opening.  No special
627      * locking needed during I/O...
628      */
629     int open_flags; /* flags used to open the file, re-used for re-open */
630     bool read_only; /* if true, the media is read only */
631     bool encrypted; /* if true, the media is encrypted */
632     bool sg;        /* if true, the device is a /dev/sg* */
633     bool probed;    /* if true, format was probed rather than specified */
634     bool force_share; /* if true, always allow all shared permissions */
635     bool implicit;  /* if true, this filter node was automatically inserted */
636 
637     BlockDriver *drv; /* NULL means no media */
638     void *opaque;
639 
640     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
641     /* long-running tasks intended to always use the same AioContext as this
642      * BDS may register themselves in this list to be notified of changes
643      * regarding this BDS's context */
644     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
645     bool walking_aio_notifiers; /* to make removal during iteration safe */
646 
647     char filename[PATH_MAX];
648     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
649                                     this file image */
650     char backing_format[16]; /* if non-zero and backing_file exists */
651 
652     QDict *full_open_options;
653     char exact_filename[PATH_MAX];
654 
655     BdrvChild *backing;
656     BdrvChild *file;
657 
658     /* I/O Limits */
659     BlockLimits bl;
660 
661     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
662     unsigned int supported_write_flags;
663     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
664      * BDRV_REQ_MAY_UNMAP) */
665     unsigned int supported_zero_flags;
666 
667     /* the following member gives a name to every node on the bs graph. */
668     char node_name[32];
669     /* element of the list of named nodes building the graph */
670     QTAILQ_ENTRY(BlockDriverState) node_list;
671     /* element of the list of all BlockDriverStates (all_bdrv_states) */
672     QTAILQ_ENTRY(BlockDriverState) bs_list;
673     /* element of the list of monitor-owned BDS */
674     QTAILQ_ENTRY(BlockDriverState) monitor_list;
675     int refcnt;
676 
677     /* operation blockers */
678     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
679 
680     /* long-running background operation */
681     BlockJob *job;
682 
683     /* The node that this node inherited default options from (and a reopen on
684      * which can affect this node by changing these defaults). This is always a
685      * parent node of this node. */
686     BlockDriverState *inherits_from;
687     QLIST_HEAD(, BdrvChild) children;
688     QLIST_HEAD(, BdrvChild) parents;
689 
690     QDict *options;
691     QDict *explicit_options;
692     BlockdevDetectZeroesOptions detect_zeroes;
693 
694     /* The error object in use for blocking operations on backing_hd */
695     Error *backing_blocker;
696 
697     /* Protected by AioContext lock */
698 
699     /* If we are reading a disk image, give its size in sectors.
700      * Generally read-only; it is written to by load_snapshot and
701      * save_snaphost, but the block layer is quiescent during those.
702      */
703     int64_t total_sectors;
704 
705     /* Callback before write request is processed */
706     NotifierWithReturnList before_write_notifiers;
707 
708     /* threshold limit for writes, in bytes. "High water mark". */
709     uint64_t write_threshold_offset;
710     NotifierWithReturn write_threshold_notifier;
711 
712     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
713      * Reading from the list can be done with either the BQL or the
714      * dirty_bitmap_mutex.  Modifying a bitmap only requires
715      * dirty_bitmap_mutex.  */
716     QemuMutex dirty_bitmap_mutex;
717     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
718 
719     /* Offset after the highest byte written to */
720     Stat64 wr_highest_offset;
721 
722     /* If true, copy read backing sectors into image.  Can be >1 if more
723      * than one client has requested copy-on-read.  Accessed with atomic
724      * ops.
725      */
726     int copy_on_read;
727 
728     /* number of in-flight requests; overall and serialising.
729      * Accessed with atomic ops.
730      */
731     unsigned int in_flight;
732     unsigned int serialising_in_flight;
733 
734     /* Kicked to signal main loop when a request completes. */
735     AioWait wait;
736 
737     /* counter for nested bdrv_io_plug.
738      * Accessed with atomic ops.
739     */
740     unsigned io_plugged;
741 
742     /* do we need to tell the quest if we have a volatile write cache? */
743     int enable_write_cache;
744 
745     /* Accessed with atomic ops.  */
746     int quiesce_counter;
747     int recursive_quiesce_counter;
748 
749     unsigned int write_gen;               /* Current data generation */
750 
751     /* Protected by reqs_lock.  */
752     CoMutex reqs_lock;
753     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
754     CoQueue flush_queue;                  /* Serializing flush queue */
755     bool active_flush_req;                /* Flush request in flight? */
756 
757     /* Only read/written by whoever has set active_flush_req to true.  */
758     unsigned int flushed_gen;             /* Flushed write generation */
759 };
760 
761 struct BlockBackendRootState {
762     int open_flags;
763     bool read_only;
764     BlockdevDetectZeroesOptions detect_zeroes;
765 };
766 
767 typedef enum BlockMirrorBackingMode {
768     /* Reuse the existing backing chain from the source for the target.
769      * - sync=full: Set backing BDS to NULL.
770      * - sync=top:  Use source's backing BDS.
771      * - sync=none: Use source as the backing BDS. */
772     MIRROR_SOURCE_BACKING_CHAIN,
773 
774     /* Open the target's backing chain completely anew */
775     MIRROR_OPEN_BACKING_CHAIN,
776 
777     /* Do not change the target's backing BDS after job completion */
778     MIRROR_LEAVE_BACKING_CHAIN,
779 } BlockMirrorBackingMode;
780 
781 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
782 {
783     return bs->backing ? bs->backing->bs : NULL;
784 }
785 
786 
787 /* Essential block drivers which must always be statically linked into qemu, and
788  * which therefore can be accessed without using bdrv_find_format() */
789 extern BlockDriver bdrv_file;
790 extern BlockDriver bdrv_raw;
791 extern BlockDriver bdrv_qcow2;
792 
793 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
794     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
795     BdrvRequestFlags flags);
796 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
797     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
798     BdrvRequestFlags flags);
799 
800 void bdrv_apply_subtree_drain(BdrvChild *child, BlockDriverState *new_parent);
801 void bdrv_unapply_subtree_drain(BdrvChild *child, BlockDriverState *old_parent);
802 
803 int get_tmp_filename(char *filename, int size);
804 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
805                             const char *filename);
806 
807 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
808                                       QDict *options);
809 
810 
811 /**
812  * bdrv_add_before_write_notifier:
813  *
814  * Register a callback that is invoked before write requests are processed but
815  * after any throttling or waiting for overlapping requests.
816  */
817 void bdrv_add_before_write_notifier(BlockDriverState *bs,
818                                     NotifierWithReturn *notifier);
819 
820 /**
821  * bdrv_detach_aio_context:
822  *
823  * May be called from .bdrv_detach_aio_context() to detach children from the
824  * current #AioContext.  This is only needed by block drivers that manage their
825  * own children.  Both ->file and ->backing are automatically handled and
826  * block drivers should not call this function on them explicitly.
827  */
828 void bdrv_detach_aio_context(BlockDriverState *bs);
829 
830 /**
831  * bdrv_attach_aio_context:
832  *
833  * May be called from .bdrv_attach_aio_context() to attach children to the new
834  * #AioContext.  This is only needed by block drivers that manage their own
835  * children.  Both ->file and ->backing are automatically handled and block
836  * drivers should not call this function on them explicitly.
837  */
838 void bdrv_attach_aio_context(BlockDriverState *bs,
839                              AioContext *new_context);
840 
841 /**
842  * bdrv_add_aio_context_notifier:
843  *
844  * If a long-running job intends to be always run in the same AioContext as a
845  * certain BDS, it may use this function to be notified of changes regarding the
846  * association of the BDS to an AioContext.
847  *
848  * attached_aio_context() is called after the target BDS has been attached to a
849  * new AioContext; detach_aio_context() is called before the target BDS is being
850  * detached from its old AioContext.
851  */
852 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
853         void (*attached_aio_context)(AioContext *new_context, void *opaque),
854         void (*detach_aio_context)(void *opaque), void *opaque);
855 
856 /**
857  * bdrv_remove_aio_context_notifier:
858  *
859  * Unsubscribe of change notifications regarding the BDS's AioContext. The
860  * parameters given here have to be the same as those given to
861  * bdrv_add_aio_context_notifier().
862  */
863 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
864                                       void (*aio_context_attached)(AioContext *,
865                                                                    void *),
866                                       void (*aio_context_detached)(void *),
867                                       void *opaque);
868 
869 /**
870  * bdrv_wakeup:
871  * @bs: The BlockDriverState for which an I/O operation has been completed.
872  *
873  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
874  * synchronous I/O on a BlockDriverState that is attached to another
875  * I/O thread, the main thread lets the I/O thread's event loop run,
876  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
877  * up the main thread if necessary.
878  *
879  * Manual calls to bdrv_wakeup are rarely necessary, because
880  * bdrv_dec_in_flight already calls it.
881  */
882 void bdrv_wakeup(BlockDriverState *bs);
883 
884 #ifdef _WIN32
885 int is_windows_drive(const char *filename);
886 #endif
887 
888 /**
889  * stream_start:
890  * @job_id: The id of the newly-created job, or %NULL to use the
891  * device name of @bs.
892  * @bs: Block device to operate on.
893  * @base: Block device that will become the new base, or %NULL to
894  * flatten the whole backing file chain onto @bs.
895  * @backing_file_str: The file name that will be written to @bs as the
896  * the new backing file if the job completes. Ignored if @base is %NULL.
897  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
898  * @on_error: The action to take upon error.
899  * @errp: Error object.
900  *
901  * Start a streaming operation on @bs.  Clusters that are unallocated
902  * in @bs, but allocated in any image between @base and @bs (both
903  * exclusive) will be written to @bs.  At the end of a successful
904  * streaming job, the backing file of @bs will be changed to
905  * @backing_file_str in the written image and to @base in the live
906  * BlockDriverState.
907  */
908 void stream_start(const char *job_id, BlockDriverState *bs,
909                   BlockDriverState *base, const char *backing_file_str,
910                   int64_t speed, BlockdevOnError on_error, Error **errp);
911 
912 /**
913  * commit_start:
914  * @job_id: The id of the newly-created job, or %NULL to use the
915  * device name of @bs.
916  * @bs: Active block device.
917  * @top: Top block device to be committed.
918  * @base: Block device that will be written into, and become the new top.
919  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
920  * @on_error: The action to take upon error.
921  * @backing_file_str: String to use as the backing file in @top's overlay
922  * @filter_node_name: The node name that should be assigned to the filter
923  * driver that the commit job inserts into the graph above @top. NULL means
924  * that a node name should be autogenerated.
925  * @errp: Error object.
926  *
927  */
928 void commit_start(const char *job_id, BlockDriverState *bs,
929                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
930                   BlockdevOnError on_error, const char *backing_file_str,
931                   const char *filter_node_name, Error **errp);
932 /**
933  * commit_active_start:
934  * @job_id: The id of the newly-created job, or %NULL to use the
935  * device name of @bs.
936  * @bs: Active block device to be committed.
937  * @base: Block device that will be written into, and become the new top.
938  * @creation_flags: Flags that control the behavior of the Job lifetime.
939  *                  See @BlockJobCreateFlags
940  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
941  * @on_error: The action to take upon error.
942  * @filter_node_name: The node name that should be assigned to the filter
943  * driver that the commit job inserts into the graph above @bs. NULL means that
944  * a node name should be autogenerated.
945  * @cb: Completion function for the job.
946  * @opaque: Opaque pointer value passed to @cb.
947  * @auto_complete: Auto complete the job.
948  * @errp: Error object.
949  *
950  */
951 void commit_active_start(const char *job_id, BlockDriverState *bs,
952                          BlockDriverState *base, int creation_flags,
953                          int64_t speed, BlockdevOnError on_error,
954                          const char *filter_node_name,
955                          BlockCompletionFunc *cb, void *opaque,
956                          bool auto_complete, Error **errp);
957 /*
958  * mirror_start:
959  * @job_id: The id of the newly-created job, or %NULL to use the
960  * device name of @bs.
961  * @bs: Block device to operate on.
962  * @target: Block device to write to.
963  * @replaces: Block graph node name to replace once the mirror is done. Can
964  *            only be used when full mirroring is selected.
965  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
966  * @granularity: The chosen granularity for the dirty bitmap.
967  * @buf_size: The amount of data that can be in flight at one time.
968  * @mode: Whether to collapse all images in the chain to the target.
969  * @backing_mode: How to establish the target's backing chain after completion.
970  * @on_source_error: The action to take upon error reading from the source.
971  * @on_target_error: The action to take upon error writing to the target.
972  * @unmap: Whether to unmap target where source sectors only contain zeroes.
973  * @filter_node_name: The node name that should be assigned to the filter
974  * driver that the mirror job inserts into the graph above @bs. NULL means that
975  * a node name should be autogenerated.
976  * @errp: Error object.
977  *
978  * Start a mirroring operation on @bs.  Clusters that are allocated
979  * in @bs will be written to @target until the job is cancelled or
980  * manually completed.  At the end of a successful mirroring job,
981  * @bs will be switched to read from @target.
982  */
983 void mirror_start(const char *job_id, BlockDriverState *bs,
984                   BlockDriverState *target, const char *replaces,
985                   int64_t speed, uint32_t granularity, int64_t buf_size,
986                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
987                   BlockdevOnError on_source_error,
988                   BlockdevOnError on_target_error,
989                   bool unmap, const char *filter_node_name, Error **errp);
990 
991 /*
992  * backup_job_create:
993  * @job_id: The id of the newly-created job, or %NULL to use the
994  * device name of @bs.
995  * @bs: Block device to operate on.
996  * @target: Block device to write to.
997  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
998  * @sync_mode: What parts of the disk image should be copied to the destination.
999  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
1000  * @on_source_error: The action to take upon error reading from the source.
1001  * @on_target_error: The action to take upon error writing to the target.
1002  * @creation_flags: Flags that control the behavior of the Job lifetime.
1003  *                  See @BlockJobCreateFlags
1004  * @cb: Completion function for the job.
1005  * @opaque: Opaque pointer value passed to @cb.
1006  * @txn: Transaction that this job is part of (may be NULL).
1007  *
1008  * Create a backup operation on @bs.  Clusters in @bs are written to @target
1009  * until the job is cancelled or manually completed.
1010  */
1011 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
1012                             BlockDriverState *target, int64_t speed,
1013                             MirrorSyncMode sync_mode,
1014                             BdrvDirtyBitmap *sync_bitmap,
1015                             bool compress,
1016                             BlockdevOnError on_source_error,
1017                             BlockdevOnError on_target_error,
1018                             int creation_flags,
1019                             BlockCompletionFunc *cb, void *opaque,
1020                             BlockJobTxn *txn, Error **errp);
1021 
1022 void hmp_drive_add_node(Monitor *mon, const char *optstr);
1023 
1024 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
1025                                   const char *child_name,
1026                                   const BdrvChildRole *child_role,
1027                                   uint64_t perm, uint64_t shared_perm,
1028                                   void *opaque, Error **errp);
1029 void bdrv_root_unref_child(BdrvChild *child);
1030 
1031 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
1032                             Error **errp);
1033 
1034 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
1035  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
1036  * all children */
1037 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
1038                                const BdrvChildRole *role,
1039                                BlockReopenQueue *reopen_queue,
1040                                uint64_t perm, uint64_t shared,
1041                                uint64_t *nperm, uint64_t *nshared);
1042 
1043 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
1044  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
1045  * requires WRITE | RESIZE for read-write images, always requires
1046  * CONSISTENT_READ and doesn't share WRITE. */
1047 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1048                                const BdrvChildRole *role,
1049                                BlockReopenQueue *reopen_queue,
1050                                uint64_t perm, uint64_t shared,
1051                                uint64_t *nperm, uint64_t *nshared);
1052 
1053 /*
1054  * Default implementation for drivers to pass bdrv_co_block_status() to
1055  * their file.
1056  */
1057 int coroutine_fn bdrv_co_block_status_from_file(BlockDriverState *bs,
1058                                                 bool want_zero,
1059                                                 int64_t offset,
1060                                                 int64_t bytes,
1061                                                 int64_t *pnum,
1062                                                 int64_t *map,
1063                                                 BlockDriverState **file);
1064 /*
1065  * Default implementation for drivers to pass bdrv_co_block_status() to
1066  * their backing file.
1067  */
1068 int coroutine_fn bdrv_co_block_status_from_backing(BlockDriverState *bs,
1069                                                    bool want_zero,
1070                                                    int64_t offset,
1071                                                    int64_t bytes,
1072                                                    int64_t *pnum,
1073                                                    int64_t *map,
1074                                                    BlockDriverState **file);
1075 const char *bdrv_get_parent_name(const BlockDriverState *bs);
1076 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
1077 bool blk_dev_has_removable_media(BlockBackend *blk);
1078 bool blk_dev_has_tray(BlockBackend *blk);
1079 void blk_dev_eject_request(BlockBackend *blk, bool force);
1080 bool blk_dev_is_tray_open(BlockBackend *blk);
1081 bool blk_dev_is_medium_locked(BlockBackend *blk);
1082 
1083 void bdrv_set_dirty(BlockDriverState *bs, int64_t offset, int64_t bytes);
1084 
1085 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
1086 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
1087 
1088 void bdrv_inc_in_flight(BlockDriverState *bs);
1089 void bdrv_dec_in_flight(BlockDriverState *bs);
1090 
1091 void blockdev_close_all_bdrv_states(void);
1092 
1093 #endif /* BLOCK_INT_H */
1094