xref: /openbmc/qemu/include/block/block_int.h (revision 60d7cacac8f08d7832b9b4081f32b8e953867b55)
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_ENCRYPT          1
41 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
42 
43 #define BLOCK_OPT_SIZE              "size"
44 #define BLOCK_OPT_ENCRYPT           "encryption"
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 */
91     bool is_filter;
92     /* for snapshots block filter like Quorum can implement the
93      * following recursive callback.
94      * It's purpose is to recurse on the filter children while calling
95      * bdrv_recurse_is_first_non_filter on them.
96      * For a sample implementation look in the future Quorum block filter.
97      */
98     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
99                                              BlockDriverState *candidate);
100 
101     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
102     int (*bdrv_probe_device)(const char *filename);
103 
104     /* Any driver implementing this callback is expected to be able to handle
105      * NULL file names in its .bdrv_open() implementation */
106     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
107     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
108      * this field set to true, except ones that are defined only by their
109      * child's bs.
110      * An example of the last type will be the quorum block driver.
111      */
112     bool bdrv_needs_filename;
113 
114     /* Set if a driver can support backing files */
115     bool supports_backing;
116 
117     /* For handling image reopen for split or non-split files */
118     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
119                                BlockReopenQueue *queue, Error **errp);
120     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
121     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
122     void (*bdrv_join_options)(QDict *options, QDict *old_options);
123 
124     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
125                      Error **errp);
126     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
127                           Error **errp);
128     void (*bdrv_close)(BlockDriverState *bs);
129     int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp);
130     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
131     int (*bdrv_make_empty)(BlockDriverState *bs);
132 
133     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
134 
135     /* aio */
136     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
137         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
138         BlockCompletionFunc *cb, void *opaque);
139     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
140         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
141         BlockCompletionFunc *cb, void *opaque);
142     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
143         BlockCompletionFunc *cb, void *opaque);
144     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
145         int64_t offset, int bytes,
146         BlockCompletionFunc *cb, void *opaque);
147 
148     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
149         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
150     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
151         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
152     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
153         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
154     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
155         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
156     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
157         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
158 
159     /*
160      * Efficiently zero a region of the disk image.  Typically an image format
161      * would use a compact metadata representation to implement this.  This
162      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
163      * will be called instead.
164      */
165     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
166         int64_t offset, int bytes, BdrvRequestFlags flags);
167     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
168         int64_t offset, int bytes);
169 
170     /*
171      * Building block for bdrv_block_status[_above]. The driver should
172      * answer only according to the current layer, and should not
173      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
174      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
175      */
176     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
177         int64_t sector_num, int nb_sectors, int *pnum,
178         BlockDriverState **file);
179 
180     /*
181      * Invalidate any cached meta-data.
182      */
183     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
184     int (*bdrv_inactivate)(BlockDriverState *bs);
185 
186     /*
187      * Flushes all data for all layers by calling bdrv_co_flush for underlying
188      * layers, if needed. This function is needed for deterministic
189      * synchronization of the flush finishing callback.
190      */
191     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
192 
193     /*
194      * Flushes all data that was already written to the OS all the way down to
195      * the disk (for example file-posix.c calls fsync()).
196      */
197     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
198 
199     /*
200      * Flushes all internal caches to the OS. The data may still sit in a
201      * writeback cache of the host OS, but it will survive a crash of the qemu
202      * process.
203      */
204     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
205 
206     const char *protocol_name;
207     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset, Error **errp);
208 
209     int64_t (*bdrv_getlength)(BlockDriverState *bs);
210     bool has_variable_length;
211     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
212 
213     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
214         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
215 
216     int (*bdrv_snapshot_create)(BlockDriverState *bs,
217                                 QEMUSnapshotInfo *sn_info);
218     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
219                               const char *snapshot_id);
220     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
221                                 const char *snapshot_id,
222                                 const char *name,
223                                 Error **errp);
224     int (*bdrv_snapshot_list)(BlockDriverState *bs,
225                               QEMUSnapshotInfo **psn_info);
226     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
227                                   const char *snapshot_id,
228                                   const char *name,
229                                   Error **errp);
230     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
231     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
232 
233     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
234                                           QEMUIOVector *qiov,
235                                           int64_t pos);
236     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
237                                           QEMUIOVector *qiov,
238                                           int64_t pos);
239 
240     int (*bdrv_change_backing_file)(BlockDriverState *bs,
241         const char *backing_file, const char *backing_fmt);
242 
243     /* removable device specific */
244     bool (*bdrv_is_inserted)(BlockDriverState *bs);
245     int (*bdrv_media_changed)(BlockDriverState *bs);
246     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
247     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
248 
249     /* to control generic scsi devices */
250     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
251         unsigned long int req, void *buf,
252         BlockCompletionFunc *cb, void *opaque);
253     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
254                                       unsigned long int req, void *buf);
255 
256     /* List of options for creating images, terminated by name == NULL */
257     QemuOptsList *create_opts;
258 
259     /*
260      * Returns 0 for completed check, -errno for internal errors.
261      * The check results are stored in result.
262      */
263     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
264         BdrvCheckMode fix);
265 
266     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
267                               BlockDriverAmendStatusCB *status_cb,
268                               void *cb_opaque);
269 
270     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
271 
272     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
273     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
274         const char *tag);
275     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
276         const char *tag);
277     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
278     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
279 
280     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
281 
282     /*
283      * Returns 1 if newly created images are guaranteed to contain only
284      * zeros, 0 otherwise.
285      */
286     int (*bdrv_has_zero_init)(BlockDriverState *bs);
287 
288     /* Remove fd handlers, timers, and other event loop callbacks so the event
289      * loop is no longer in use.  Called with no in-flight requests and in
290      * depth-first traversal order with parents before child nodes.
291      */
292     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
293 
294     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
295      * can be processed again.  Called with no in-flight requests and in
296      * depth-first traversal order with child nodes before parent nodes.
297      */
298     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
299                                     AioContext *new_context);
300 
301     /* io queue for linux-aio */
302     void (*bdrv_io_plug)(BlockDriverState *bs);
303     void (*bdrv_io_unplug)(BlockDriverState *bs);
304 
305     /**
306      * Try to get @bs's logical and physical block size.
307      * On success, store them in @bsz and return zero.
308      * On failure, return negative errno.
309      */
310     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
311     /**
312      * Try to get @bs's geometry (cyls, heads, sectors)
313      * On success, store them in @geo and return 0.
314      * On failure return -errno.
315      * Only drivers that want to override guest geometry implement this
316      * callback; see hd_geometry_guess().
317      */
318     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
319 
320     /**
321      * Drain and stop any internal sources of requests in the driver, and
322      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
323      */
324     void (*bdrv_drain)(BlockDriverState *bs);
325 
326     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
327                            Error **errp);
328     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
329                            Error **errp);
330 
331     /**
332      * Informs the block driver that a permission change is intended. The
333      * driver checks whether the change is permissible and may take other
334      * preparations for the change (e.g. get file system locks). This operation
335      * is always followed either by a call to either .bdrv_set_perm or
336      * .bdrv_abort_perm_update.
337      *
338      * Checks whether the requested set of cumulative permissions in @perm
339      * can be granted for accessing @bs and whether no other users are using
340      * permissions other than those given in @shared (both arguments take
341      * BLK_PERM_* bitmasks).
342      *
343      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
344      * and errp is set to an error describing the conflict.
345      */
346     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
347                            uint64_t shared, Error **errp);
348 
349     /**
350      * Called to inform the driver that the set of cumulative set of used
351      * permissions for @bs has changed to @perm, and the set of sharable
352      * permission to @shared. The driver can use this to propagate changes to
353      * its children (i.e. request permissions only if a parent actually needs
354      * them).
355      *
356      * This function is only invoked after bdrv_check_perm(), so block drivers
357      * may rely on preparations made in their .bdrv_check_perm implementation.
358      */
359     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
360 
361     /*
362      * Called to inform the driver that after a previous bdrv_check_perm()
363      * call, the permission update is not performed and any preparations made
364      * for it (e.g. taken file locks) need to be undone.
365      *
366      * This function can be called even for nodes that never saw a
367      * bdrv_check_perm() call. It is a no-op then.
368      */
369     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
370 
371     /**
372      * Returns in @nperm and @nshared the permissions that the driver for @bs
373      * needs on its child @c, based on the cumulative permissions requested by
374      * the parents in @parent_perm and @parent_shared.
375      *
376      * If @c is NULL, return the permissions for attaching a new child for the
377      * given @role.
378      */
379      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
380                              const BdrvChildRole *role,
381                              uint64_t parent_perm, uint64_t parent_shared,
382                              uint64_t *nperm, uint64_t *nshared);
383 
384     QLIST_ENTRY(BlockDriver) list;
385 };
386 
387 typedef struct BlockLimits {
388     /* Alignment requirement, in bytes, for offset/length of I/O
389      * requests. Must be a power of 2 less than INT_MAX; defaults to
390      * 1 for drivers with modern byte interfaces, and to 512
391      * otherwise. */
392     uint32_t request_alignment;
393 
394     /* Maximum number of bytes that can be discarded at once (since it
395      * is signed, it must be < 2G, if set). Must be multiple of
396      * pdiscard_alignment, but need not be power of 2. May be 0 if no
397      * inherent 32-bit limit */
398     int32_t max_pdiscard;
399 
400     /* Optimal alignment for discard requests in bytes. A power of 2
401      * is best but not mandatory.  Must be a multiple of
402      * bl.request_alignment, and must be less than max_pdiscard if
403      * that is set. May be 0 if bl.request_alignment is good enough */
404     uint32_t pdiscard_alignment;
405 
406     /* Maximum number of bytes that can zeroized at once (since it is
407      * signed, it must be < 2G, if set). Must be multiple of
408      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
409     int32_t max_pwrite_zeroes;
410 
411     /* Optimal alignment for write zeroes requests in bytes. A power
412      * of 2 is best but not mandatory.  Must be a multiple of
413      * bl.request_alignment, and must be less than max_pwrite_zeroes
414      * if that is set. May be 0 if bl.request_alignment is good
415      * enough */
416     uint32_t pwrite_zeroes_alignment;
417 
418     /* Optimal transfer length in bytes.  A power of 2 is best but not
419      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
420      * no preferred size */
421     uint32_t opt_transfer;
422 
423     /* Maximal transfer length in bytes.  Need not be power of 2, but
424      * must be multiple of opt_transfer and bl.request_alignment, or 0
425      * for no 32-bit limit.  For now, anything larger than INT_MAX is
426      * clamped down. */
427     uint32_t max_transfer;
428 
429     /* memory alignment, in bytes so that no bounce buffer is needed */
430     size_t min_mem_alignment;
431 
432     /* memory alignment, in bytes, for bounce buffer */
433     size_t opt_mem_alignment;
434 
435     /* maximum number of iovec elements */
436     int max_iov;
437 } BlockLimits;
438 
439 typedef struct BdrvOpBlocker BdrvOpBlocker;
440 
441 typedef struct BdrvAioNotifier {
442     void (*attached_aio_context)(AioContext *new_context, void *opaque);
443     void (*detach_aio_context)(void *opaque);
444 
445     void *opaque;
446     bool deleted;
447 
448     QLIST_ENTRY(BdrvAioNotifier) list;
449 } BdrvAioNotifier;
450 
451 struct BdrvChildRole {
452     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
453      * points to. */
454     bool stay_at_node;
455 
456     void (*inherit_options)(int *child_flags, QDict *child_options,
457                             int parent_flags, QDict *parent_options);
458 
459     void (*change_media)(BdrvChild *child, bool load);
460     void (*resize)(BdrvChild *child);
461 
462     /* Returns a name that is supposedly more useful for human users than the
463      * node name for identifying the node in question (in particular, a BB
464      * name), or NULL if the parent can't provide a better name. */
465     const char *(*get_name)(BdrvChild *child);
466 
467     /* Returns a malloced string that describes the parent of the child for a
468      * human reader. This could be a node-name, BlockBackend name, qdev ID or
469      * QOM path of the device owning the BlockBackend, job type and ID etc. The
470      * caller is responsible for freeing the memory. */
471     char *(*get_parent_desc)(BdrvChild *child);
472 
473     /*
474      * If this pair of functions is implemented, the parent doesn't issue new
475      * requests after returning from .drained_begin() until .drained_end() is
476      * called.
477      *
478      * Note that this can be nested. If drained_begin() was called twice, new
479      * I/O is allowed only after drained_end() was called twice, too.
480      */
481     void (*drained_begin)(BdrvChild *child);
482     void (*drained_end)(BdrvChild *child);
483 
484     /* Notifies the parent that the child has been activated/inactivated (e.g.
485      * when migration is completing) and it can start/stop requesting
486      * permissions and doing I/O on it. */
487     void (*activate)(BdrvChild *child, Error **errp);
488     int (*inactivate)(BdrvChild *child);
489 
490     void (*attach)(BdrvChild *child);
491     void (*detach)(BdrvChild *child);
492 };
493 
494 extern const BdrvChildRole child_file;
495 extern const BdrvChildRole child_format;
496 extern const BdrvChildRole child_backing;
497 
498 struct BdrvChild {
499     BlockDriverState *bs;
500     char *name;
501     const BdrvChildRole *role;
502     void *opaque;
503 
504     /**
505      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
506      */
507     uint64_t perm;
508 
509     /**
510      * Permissions that can still be granted to other users of @bs while this
511      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
512      */
513     uint64_t shared_perm;
514 
515     QLIST_ENTRY(BdrvChild) next;
516     QLIST_ENTRY(BdrvChild) next_parent;
517 };
518 
519 /*
520  * Note: the function bdrv_append() copies and swaps contents of
521  * BlockDriverStates, so if you add new fields to this struct, please
522  * inspect bdrv_append() to determine if the new fields need to be
523  * copied as well.
524  */
525 struct BlockDriverState {
526     /* Protected by big QEMU lock or read-only after opening.  No special
527      * locking needed during I/O...
528      */
529     int open_flags; /* flags used to open the file, re-used for re-open */
530     bool read_only; /* if true, the media is read only */
531     bool encrypted; /* if true, the media is encrypted */
532     bool valid_key; /* if true, a valid encryption key has been set */
533     bool sg;        /* if true, the device is a /dev/sg* */
534     bool probed;    /* if true, format was probed rather than specified */
535     bool force_share; /* if true, always allow all shared permissions */
536 
537     BlockDriver *drv; /* NULL means no media */
538     void *opaque;
539 
540     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
541     /* long-running tasks intended to always use the same AioContext as this
542      * BDS may register themselves in this list to be notified of changes
543      * regarding this BDS's context */
544     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
545     bool walking_aio_notifiers; /* to make removal during iteration safe */
546 
547     char filename[PATH_MAX];
548     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
549                                     this file image */
550     char backing_format[16]; /* if non-zero and backing_file exists */
551 
552     QDict *full_open_options;
553     char exact_filename[PATH_MAX];
554 
555     BdrvChild *backing;
556     BdrvChild *file;
557 
558     /* I/O Limits */
559     BlockLimits bl;
560 
561     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
562     unsigned int supported_write_flags;
563     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
564      * BDRV_REQ_MAY_UNMAP) */
565     unsigned int supported_zero_flags;
566 
567     /* the following member gives a name to every node on the bs graph. */
568     char node_name[32];
569     /* element of the list of named nodes building the graph */
570     QTAILQ_ENTRY(BlockDriverState) node_list;
571     /* element of the list of all BlockDriverStates (all_bdrv_states) */
572     QTAILQ_ENTRY(BlockDriverState) bs_list;
573     /* element of the list of monitor-owned BDS */
574     QTAILQ_ENTRY(BlockDriverState) monitor_list;
575     int refcnt;
576 
577     /* operation blockers */
578     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
579 
580     /* long-running background operation */
581     BlockJob *job;
582 
583     /* The node that this node inherited default options from (and a reopen on
584      * which can affect this node by changing these defaults). This is always a
585      * parent node of this node. */
586     BlockDriverState *inherits_from;
587     QLIST_HEAD(, BdrvChild) children;
588     QLIST_HEAD(, BdrvChild) parents;
589 
590     QDict *options;
591     QDict *explicit_options;
592     BlockdevDetectZeroesOptions detect_zeroes;
593 
594     /* The error object in use for blocking operations on backing_hd */
595     Error *backing_blocker;
596 
597     /* Protected by AioContext lock */
598 
599     /* If we are reading a disk image, give its size in sectors.
600      * Generally read-only; it is written to by load_snapshot and
601      * save_snaphost, but the block layer is quiescent during those.
602      */
603     int64_t total_sectors;
604 
605     /* Callback before write request is processed */
606     NotifierWithReturnList before_write_notifiers;
607 
608     /* threshold limit for writes, in bytes. "High water mark". */
609     uint64_t write_threshold_offset;
610     NotifierWithReturn write_threshold_notifier;
611 
612     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
613      * Reading from the list can be done with either the BQL or the
614      * dirty_bitmap_mutex.  Modifying a bitmap only requires
615      * dirty_bitmap_mutex.  */
616     QemuMutex dirty_bitmap_mutex;
617     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
618 
619     /* Offset after the highest byte written to */
620     Stat64 wr_highest_offset;
621 
622     /* If true, copy read backing sectors into image.  Can be >1 if more
623      * than one client has requested copy-on-read.  Accessed with atomic
624      * ops.
625      */
626     int copy_on_read;
627 
628     /* number of in-flight requests; overall and serialising.
629      * Accessed with atomic ops.
630      */
631     unsigned int in_flight;
632     unsigned int serialising_in_flight;
633 
634     /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
635      * ops.
636      */
637     bool wakeup;
638 
639     /* counter for nested bdrv_io_plug.
640      * Accessed with atomic ops.
641     */
642     unsigned io_plugged;
643 
644     /* do we need to tell the quest if we have a volatile write cache? */
645     int enable_write_cache;
646 
647     /* Accessed with atomic ops.  */
648     int quiesce_counter;
649     unsigned int write_gen;               /* Current data generation */
650 
651     /* Protected by reqs_lock.  */
652     CoMutex reqs_lock;
653     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
654     CoQueue flush_queue;                  /* Serializing flush queue */
655     bool active_flush_req;                /* Flush request in flight? */
656 
657     /* Only read/written by whoever has set active_flush_req to true.  */
658     unsigned int flushed_gen;             /* Flushed write generation */
659 };
660 
661 struct BlockBackendRootState {
662     int open_flags;
663     bool read_only;
664     BlockdevDetectZeroesOptions detect_zeroes;
665 };
666 
667 typedef enum BlockMirrorBackingMode {
668     /* Reuse the existing backing chain from the source for the target.
669      * - sync=full: Set backing BDS to NULL.
670      * - sync=top:  Use source's backing BDS.
671      * - sync=none: Use source as the backing BDS. */
672     MIRROR_SOURCE_BACKING_CHAIN,
673 
674     /* Open the target's backing chain completely anew */
675     MIRROR_OPEN_BACKING_CHAIN,
676 
677     /* Do not change the target's backing BDS after job completion */
678     MIRROR_LEAVE_BACKING_CHAIN,
679 } BlockMirrorBackingMode;
680 
681 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
682 {
683     return bs->backing ? bs->backing->bs : NULL;
684 }
685 
686 
687 /* Essential block drivers which must always be statically linked into qemu, and
688  * which therefore can be accessed without using bdrv_find_format() */
689 extern BlockDriver bdrv_file;
690 extern BlockDriver bdrv_raw;
691 extern BlockDriver bdrv_qcow2;
692 
693 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
694     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
695     BdrvRequestFlags flags);
696 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
697     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
698     BdrvRequestFlags flags);
699 
700 int get_tmp_filename(char *filename, int size);
701 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
702                             const char *filename);
703 
704 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
705                                       QDict *options);
706 
707 
708 /**
709  * bdrv_add_before_write_notifier:
710  *
711  * Register a callback that is invoked before write requests are processed but
712  * after any throttling or waiting for overlapping requests.
713  */
714 void bdrv_add_before_write_notifier(BlockDriverState *bs,
715                                     NotifierWithReturn *notifier);
716 
717 /**
718  * bdrv_detach_aio_context:
719  *
720  * May be called from .bdrv_detach_aio_context() to detach children from the
721  * current #AioContext.  This is only needed by block drivers that manage their
722  * own children.  Both ->file and ->backing are automatically handled and
723  * block drivers should not call this function on them explicitly.
724  */
725 void bdrv_detach_aio_context(BlockDriverState *bs);
726 
727 /**
728  * bdrv_attach_aio_context:
729  *
730  * May be called from .bdrv_attach_aio_context() to attach children to the new
731  * #AioContext.  This is only needed by block drivers that manage their own
732  * children.  Both ->file and ->backing are automatically handled and block
733  * drivers should not call this function on them explicitly.
734  */
735 void bdrv_attach_aio_context(BlockDriverState *bs,
736                              AioContext *new_context);
737 
738 /**
739  * bdrv_add_aio_context_notifier:
740  *
741  * If a long-running job intends to be always run in the same AioContext as a
742  * certain BDS, it may use this function to be notified of changes regarding the
743  * association of the BDS to an AioContext.
744  *
745  * attached_aio_context() is called after the target BDS has been attached to a
746  * new AioContext; detach_aio_context() is called before the target BDS is being
747  * detached from its old AioContext.
748  */
749 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
750         void (*attached_aio_context)(AioContext *new_context, void *opaque),
751         void (*detach_aio_context)(void *opaque), void *opaque);
752 
753 /**
754  * bdrv_remove_aio_context_notifier:
755  *
756  * Unsubscribe of change notifications regarding the BDS's AioContext. The
757  * parameters given here have to be the same as those given to
758  * bdrv_add_aio_context_notifier().
759  */
760 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
761                                       void (*aio_context_attached)(AioContext *,
762                                                                    void *),
763                                       void (*aio_context_detached)(void *),
764                                       void *opaque);
765 
766 /**
767  * bdrv_wakeup:
768  * @bs: The BlockDriverState for which an I/O operation has been completed.
769  *
770  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
771  * synchronous I/O on a BlockDriverState that is attached to another
772  * I/O thread, the main thread lets the I/O thread's event loop run,
773  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
774  * up the main thread if necessary.
775  *
776  * Manual calls to bdrv_wakeup are rarely necessary, because
777  * bdrv_dec_in_flight already calls it.
778  */
779 void bdrv_wakeup(BlockDriverState *bs);
780 
781 #ifdef _WIN32
782 int is_windows_drive(const char *filename);
783 #endif
784 
785 /**
786  * stream_start:
787  * @job_id: The id of the newly-created job, or %NULL to use the
788  * device name of @bs.
789  * @bs: Block device to operate on.
790  * @base: Block device that will become the new base, or %NULL to
791  * flatten the whole backing file chain onto @bs.
792  * @backing_file_str: The file name that will be written to @bs as the
793  * the new backing file if the job completes. Ignored if @base is %NULL.
794  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
795  * @on_error: The action to take upon error.
796  * @errp: Error object.
797  *
798  * Start a streaming operation on @bs.  Clusters that are unallocated
799  * in @bs, but allocated in any image between @base and @bs (both
800  * exclusive) will be written to @bs.  At the end of a successful
801  * streaming job, the backing file of @bs will be changed to
802  * @backing_file_str in the written image and to @base in the live
803  * BlockDriverState.
804  */
805 void stream_start(const char *job_id, BlockDriverState *bs,
806                   BlockDriverState *base, const char *backing_file_str,
807                   int64_t speed, BlockdevOnError on_error, Error **errp);
808 
809 /**
810  * commit_start:
811  * @job_id: The id of the newly-created job, or %NULL to use the
812  * device name of @bs.
813  * @bs: Active block device.
814  * @top: Top block device to be committed.
815  * @base: Block device that will be written into, and become the new top.
816  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
817  * @on_error: The action to take upon error.
818  * @backing_file_str: String to use as the backing file in @top's overlay
819  * @filter_node_name: The node name that should be assigned to the filter
820  * driver that the commit job inserts into the graph above @top. NULL means
821  * that a node name should be autogenerated.
822  * @errp: Error object.
823  *
824  */
825 void commit_start(const char *job_id, BlockDriverState *bs,
826                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
827                   BlockdevOnError on_error, const char *backing_file_str,
828                   const char *filter_node_name, Error **errp);
829 /**
830  * commit_active_start:
831  * @job_id: The id of the newly-created job, or %NULL to use the
832  * device name of @bs.
833  * @bs: Active block device to be committed.
834  * @base: Block device that will be written into, and become the new top.
835  * @creation_flags: Flags that control the behavior of the Job lifetime.
836  *                  See @BlockJobCreateFlags
837  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
838  * @on_error: The action to take upon error.
839  * @filter_node_name: The node name that should be assigned to the filter
840  * driver that the commit job inserts into the graph above @bs. NULL means that
841  * a node name should be autogenerated.
842  * @cb: Completion function for the job.
843  * @opaque: Opaque pointer value passed to @cb.
844  * @auto_complete: Auto complete the job.
845  * @errp: Error object.
846  *
847  */
848 void commit_active_start(const char *job_id, BlockDriverState *bs,
849                          BlockDriverState *base, int creation_flags,
850                          int64_t speed, BlockdevOnError on_error,
851                          const char *filter_node_name,
852                          BlockCompletionFunc *cb, void *opaque,
853                          bool auto_complete, Error **errp);
854 /*
855  * mirror_start:
856  * @job_id: The id of the newly-created job, or %NULL to use the
857  * device name of @bs.
858  * @bs: Block device to operate on.
859  * @target: Block device to write to.
860  * @replaces: Block graph node name to replace once the mirror is done. Can
861  *            only be used when full mirroring is selected.
862  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
863  * @granularity: The chosen granularity for the dirty bitmap.
864  * @buf_size: The amount of data that can be in flight at one time.
865  * @mode: Whether to collapse all images in the chain to the target.
866  * @backing_mode: How to establish the target's backing chain after completion.
867  * @on_source_error: The action to take upon error reading from the source.
868  * @on_target_error: The action to take upon error writing to the target.
869  * @unmap: Whether to unmap target where source sectors only contain zeroes.
870  * @filter_node_name: The node name that should be assigned to the filter
871  * driver that the mirror job inserts into the graph above @bs. NULL means that
872  * a node name should be autogenerated.
873  * @errp: Error object.
874  *
875  * Start a mirroring operation on @bs.  Clusters that are allocated
876  * in @bs will be written to @target until the job is cancelled or
877  * manually completed.  At the end of a successful mirroring job,
878  * @bs will be switched to read from @target.
879  */
880 void mirror_start(const char *job_id, BlockDriverState *bs,
881                   BlockDriverState *target, const char *replaces,
882                   int64_t speed, uint32_t granularity, int64_t buf_size,
883                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
884                   BlockdevOnError on_source_error,
885                   BlockdevOnError on_target_error,
886                   bool unmap, const char *filter_node_name, Error **errp);
887 
888 /*
889  * backup_job_create:
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  * @target: Block device to write to.
894  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
895  * @sync_mode: What parts of the disk image should be copied to the destination.
896  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
897  * @on_source_error: The action to take upon error reading from the source.
898  * @on_target_error: The action to take upon error writing to the target.
899  * @creation_flags: Flags that control the behavior of the Job lifetime.
900  *                  See @BlockJobCreateFlags
901  * @cb: Completion function for the job.
902  * @opaque: Opaque pointer value passed to @cb.
903  * @txn: Transaction that this job is part of (may be NULL).
904  *
905  * Create a backup operation on @bs.  Clusters in @bs are written to @target
906  * until the job is cancelled or manually completed.
907  */
908 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
909                             BlockDriverState *target, int64_t speed,
910                             MirrorSyncMode sync_mode,
911                             BdrvDirtyBitmap *sync_bitmap,
912                             bool compress,
913                             BlockdevOnError on_source_error,
914                             BlockdevOnError on_target_error,
915                             int creation_flags,
916                             BlockCompletionFunc *cb, void *opaque,
917                             BlockJobTxn *txn, Error **errp);
918 
919 void hmp_drive_add_node(Monitor *mon, const char *optstr);
920 
921 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
922                                   const char *child_name,
923                                   const BdrvChildRole *child_role,
924                                   uint64_t perm, uint64_t shared_perm,
925                                   void *opaque, Error **errp);
926 void bdrv_root_unref_child(BdrvChild *child);
927 
928 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
929                             Error **errp);
930 
931 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
932  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
933  * all children */
934 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
935                                const BdrvChildRole *role,
936                                uint64_t perm, uint64_t shared,
937                                uint64_t *nperm, uint64_t *nshared);
938 
939 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
940  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
941  * requires WRITE | RESIZE for read-write images, always requires
942  * CONSISTENT_READ and doesn't share WRITE. */
943 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
944                                const BdrvChildRole *role,
945                                uint64_t perm, uint64_t shared,
946                                uint64_t *nperm, uint64_t *nshared);
947 
948 const char *bdrv_get_parent_name(const BlockDriverState *bs);
949 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
950 bool blk_dev_has_removable_media(BlockBackend *blk);
951 bool blk_dev_has_tray(BlockBackend *blk);
952 void blk_dev_eject_request(BlockBackend *blk, bool force);
953 bool blk_dev_is_tray_open(BlockBackend *blk);
954 bool blk_dev_is_medium_locked(BlockBackend *blk);
955 
956 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
957 bool bdrv_requests_pending(BlockDriverState *bs);
958 
959 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
960 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
961 
962 void bdrv_inc_in_flight(BlockDriverState *bs);
963 void bdrv_dec_in_flight(BlockDriverState *bs);
964 
965 void blockdev_close_all_bdrv_states(void);
966 
967 #endif /* BLOCK_INT_H */
968