xref: /openbmc/qemu/include/block/block_int.h (revision 795c40b8)
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/timer.h"
33 #include "qapi-types.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_ENCRYPT          1
40 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
41 
42 #define BLOCK_OPT_SIZE              "size"
43 #define BLOCK_OPT_ENCRYPT           "encryption"
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 */
90     bool is_filter;
91     /* for snapshots block filter like Quorum can implement the
92      * following recursive callback.
93      * It's purpose is to recurse on the filter children while calling
94      * bdrv_recurse_is_first_non_filter on them.
95      * For a sample implementation look in the future Quorum block filter.
96      */
97     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
98                                              BlockDriverState *candidate);
99 
100     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
101     int (*bdrv_probe_device)(const char *filename);
102 
103     /* Any driver implementing this callback is expected to be able to handle
104      * NULL file names in its .bdrv_open() implementation */
105     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
106     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
107      * this field set to true, except ones that are defined only by their
108      * child's bs.
109      * An example of the last type will be the quorum block driver.
110      */
111     bool bdrv_needs_filename;
112 
113     /* Set if a driver can support backing files */
114     bool supports_backing;
115 
116     /* For handling image reopen for split or non-split files */
117     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
118                                BlockReopenQueue *queue, Error **errp);
119     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
120     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
121     void (*bdrv_join_options)(QDict *options, QDict *old_options);
122 
123     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
124                      Error **errp);
125     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
126                           Error **errp);
127     void (*bdrv_close)(BlockDriverState *bs);
128     int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp);
129     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
130     int (*bdrv_make_empty)(BlockDriverState *bs);
131 
132     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
133 
134     /* aio */
135     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
136         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
137         BlockCompletionFunc *cb, void *opaque);
138     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
139         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
140         BlockCompletionFunc *cb, void *opaque);
141     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
142         BlockCompletionFunc *cb, void *opaque);
143     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
144         int64_t offset, int count,
145         BlockCompletionFunc *cb, void *opaque);
146 
147     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
148         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
149     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
150         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
151     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
152         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
153     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
154         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
155     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
156         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
157 
158     /*
159      * Efficiently zero a region of the disk image.  Typically an image format
160      * would use a compact metadata representation to implement this.  This
161      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
162      * will be called instead.
163      */
164     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
165         int64_t offset, int count, BdrvRequestFlags flags);
166     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
167         int64_t offset, int count);
168 
169     /*
170      * Building block for bdrv_block_status[_above]. The driver should
171      * answer only according to the current layer, and should not
172      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
173      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
174      */
175     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
176         int64_t sector_num, int nb_sectors, int *pnum,
177         BlockDriverState **file);
178 
179     /*
180      * Invalidate any cached meta-data.
181      */
182     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
183     int (*bdrv_inactivate)(BlockDriverState *bs);
184 
185     /*
186      * Flushes all data for all layers by calling bdrv_co_flush for underlying
187      * layers, if needed. This function is needed for deterministic
188      * synchronization of the flush finishing callback.
189      */
190     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
191 
192     /*
193      * Flushes all data that was already written to the OS all the way down to
194      * the disk (for example file-posix.c calls fsync()).
195      */
196     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
197 
198     /*
199      * Flushes all internal caches to the OS. The data may still sit in a
200      * writeback cache of the host OS, but it will survive a crash of the qemu
201      * process.
202      */
203     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
204 
205     const char *protocol_name;
206     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset, Error **errp);
207 
208     int64_t (*bdrv_getlength)(BlockDriverState *bs);
209     bool has_variable_length;
210     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
211 
212     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
213         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
214 
215     int (*bdrv_snapshot_create)(BlockDriverState *bs,
216                                 QEMUSnapshotInfo *sn_info);
217     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
218                               const char *snapshot_id);
219     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
220                                 const char *snapshot_id,
221                                 const char *name,
222                                 Error **errp);
223     int (*bdrv_snapshot_list)(BlockDriverState *bs,
224                               QEMUSnapshotInfo **psn_info);
225     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
226                                   const char *snapshot_id,
227                                   const char *name,
228                                   Error **errp);
229     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
230     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
231 
232     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
233                                           QEMUIOVector *qiov,
234                                           int64_t pos);
235     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
236                                           QEMUIOVector *qiov,
237                                           int64_t pos);
238 
239     int (*bdrv_change_backing_file)(BlockDriverState *bs,
240         const char *backing_file, const char *backing_fmt);
241 
242     /* removable device specific */
243     bool (*bdrv_is_inserted)(BlockDriverState *bs);
244     int (*bdrv_media_changed)(BlockDriverState *bs);
245     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
246     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
247 
248     /* to control generic scsi devices */
249     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
250         unsigned long int req, void *buf,
251         BlockCompletionFunc *cb, void *opaque);
252     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
253                                       unsigned long int req, void *buf);
254 
255     /* List of options for creating images, terminated by name == NULL */
256     QemuOptsList *create_opts;
257 
258     /*
259      * Returns 0 for completed check, -errno for internal errors.
260      * The check results are stored in result.
261      */
262     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
263         BdrvCheckMode fix);
264 
265     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
266                               BlockDriverAmendStatusCB *status_cb,
267                               void *cb_opaque);
268 
269     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
270 
271     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
272     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
273         const char *tag);
274     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
275         const char *tag);
276     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
277     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
278 
279     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
280 
281     /*
282      * Returns 1 if newly created images are guaranteed to contain only
283      * zeros, 0 otherwise.
284      */
285     int (*bdrv_has_zero_init)(BlockDriverState *bs);
286 
287     /* Remove fd handlers, timers, and other event loop callbacks so the event
288      * loop is no longer in use.  Called with no in-flight requests and in
289      * depth-first traversal order with parents before child nodes.
290      */
291     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
292 
293     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
294      * can be processed again.  Called with no in-flight requests and in
295      * depth-first traversal order with child nodes before parent nodes.
296      */
297     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
298                                     AioContext *new_context);
299 
300     /* io queue for linux-aio */
301     void (*bdrv_io_plug)(BlockDriverState *bs);
302     void (*bdrv_io_unplug)(BlockDriverState *bs);
303 
304     /**
305      * Try to get @bs's logical and physical block size.
306      * On success, store them in @bsz and return zero.
307      * On failure, return negative errno.
308      */
309     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
310     /**
311      * Try to get @bs's geometry (cyls, heads, sectors)
312      * On success, store them in @geo and return 0.
313      * On failure return -errno.
314      * Only drivers that want to override guest geometry implement this
315      * callback; see hd_geometry_guess().
316      */
317     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
318 
319     /**
320      * Drain and stop any internal sources of requests in the driver, and
321      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
322      */
323     void (*bdrv_drain)(BlockDriverState *bs);
324 
325     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
326                            Error **errp);
327     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
328                            Error **errp);
329 
330     /**
331      * Informs the block driver that a permission change is intended. The
332      * driver checks whether the change is permissible and may take other
333      * preparations for the change (e.g. get file system locks). This operation
334      * is always followed either by a call to either .bdrv_set_perm or
335      * .bdrv_abort_perm_update.
336      *
337      * Checks whether the requested set of cumulative permissions in @perm
338      * can be granted for accessing @bs and whether no other users are using
339      * permissions other than those given in @shared (both arguments take
340      * BLK_PERM_* bitmasks).
341      *
342      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
343      * and errp is set to an error describing the conflict.
344      */
345     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
346                            uint64_t shared, Error **errp);
347 
348     /**
349      * Called to inform the driver that the set of cumulative set of used
350      * permissions for @bs has changed to @perm, and the set of sharable
351      * permission to @shared. The driver can use this to propagate changes to
352      * its children (i.e. request permissions only if a parent actually needs
353      * them).
354      *
355      * This function is only invoked after bdrv_check_perm(), so block drivers
356      * may rely on preparations made in their .bdrv_check_perm implementation.
357      */
358     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
359 
360     /*
361      * Called to inform the driver that after a previous bdrv_check_perm()
362      * call, the permission update is not performed and any preparations made
363      * for it (e.g. taken file locks) need to be undone.
364      *
365      * This function can be called even for nodes that never saw a
366      * bdrv_check_perm() call. It is a no-op then.
367      */
368     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
369 
370     /**
371      * Returns in @nperm and @nshared the permissions that the driver for @bs
372      * needs on its child @c, based on the cumulative permissions requested by
373      * the parents in @parent_perm and @parent_shared.
374      *
375      * If @c is NULL, return the permissions for attaching a new child for the
376      * given @role.
377      */
378      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
379                              const BdrvChildRole *role,
380                              uint64_t parent_perm, uint64_t parent_shared,
381                              uint64_t *nperm, uint64_t *nshared);
382 
383     QLIST_ENTRY(BlockDriver) list;
384 };
385 
386 typedef struct BlockLimits {
387     /* Alignment requirement, in bytes, for offset/length of I/O
388      * requests. Must be a power of 2 less than INT_MAX; defaults to
389      * 1 for drivers with modern byte interfaces, and to 512
390      * otherwise. */
391     uint32_t request_alignment;
392 
393     /* Maximum number of bytes that can be discarded at once (since it
394      * is signed, it must be < 2G, if set). Must be multiple of
395      * pdiscard_alignment, but need not be power of 2. May be 0 if no
396      * inherent 32-bit limit */
397     int32_t max_pdiscard;
398 
399     /* Optimal alignment for discard requests in bytes. A power of 2
400      * is best but not mandatory.  Must be a multiple of
401      * bl.request_alignment, and must be less than max_pdiscard if
402      * that is set. May be 0 if bl.request_alignment is good enough */
403     uint32_t pdiscard_alignment;
404 
405     /* Maximum number of bytes that can zeroized at once (since it is
406      * signed, it must be < 2G, if set). Must be multiple of
407      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
408     int32_t max_pwrite_zeroes;
409 
410     /* Optimal alignment for write zeroes requests in bytes. A power
411      * of 2 is best but not mandatory.  Must be a multiple of
412      * bl.request_alignment, and must be less than max_pwrite_zeroes
413      * if that is set. May be 0 if bl.request_alignment is good
414      * enough */
415     uint32_t pwrite_zeroes_alignment;
416 
417     /* Optimal transfer length in bytes.  A power of 2 is best but not
418      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
419      * no preferred size */
420     uint32_t opt_transfer;
421 
422     /* Maximal transfer length in bytes.  Need not be power of 2, but
423      * must be multiple of opt_transfer and bl.request_alignment, or 0
424      * for no 32-bit limit.  For now, anything larger than INT_MAX is
425      * clamped down. */
426     uint32_t max_transfer;
427 
428     /* memory alignment, in bytes so that no bounce buffer is needed */
429     size_t min_mem_alignment;
430 
431     /* memory alignment, in bytes, for bounce buffer */
432     size_t opt_mem_alignment;
433 
434     /* maximum number of iovec elements */
435     int max_iov;
436 } BlockLimits;
437 
438 typedef struct BdrvOpBlocker BdrvOpBlocker;
439 
440 typedef struct BdrvAioNotifier {
441     void (*attached_aio_context)(AioContext *new_context, void *opaque);
442     void (*detach_aio_context)(void *opaque);
443 
444     void *opaque;
445     bool deleted;
446 
447     QLIST_ENTRY(BdrvAioNotifier) list;
448 } BdrvAioNotifier;
449 
450 struct BdrvChildRole {
451     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
452      * points to. */
453     bool stay_at_node;
454 
455     void (*inherit_options)(int *child_flags, QDict *child_options,
456                             int parent_flags, QDict *parent_options);
457 
458     void (*change_media)(BdrvChild *child, bool load);
459     void (*resize)(BdrvChild *child);
460 
461     /* Returns a name that is supposedly more useful for human users than the
462      * node name for identifying the node in question (in particular, a BB
463      * name), or NULL if the parent can't provide a better name. */
464     const char *(*get_name)(BdrvChild *child);
465 
466     /* Returns a malloced string that describes the parent of the child for a
467      * human reader. This could be a node-name, BlockBackend name, qdev ID or
468      * QOM path of the device owning the BlockBackend, job type and ID etc. The
469      * caller is responsible for freeing the memory. */
470     char *(*get_parent_desc)(BdrvChild *child);
471 
472     /*
473      * If this pair of functions is implemented, the parent doesn't issue new
474      * requests after returning from .drained_begin() until .drained_end() is
475      * called.
476      *
477      * Note that this can be nested. If drained_begin() was called twice, new
478      * I/O is allowed only after drained_end() was called twice, too.
479      */
480     void (*drained_begin)(BdrvChild *child);
481     void (*drained_end)(BdrvChild *child);
482 
483     /* Notifies the parent that the child has been activated/inactivated (e.g.
484      * when migration is completing) and it can start/stop requesting
485      * permissions and doing I/O on it. */
486     void (*activate)(BdrvChild *child, Error **errp);
487     int (*inactivate)(BdrvChild *child);
488 
489     void (*attach)(BdrvChild *child);
490     void (*detach)(BdrvChild *child);
491 };
492 
493 extern const BdrvChildRole child_file;
494 extern const BdrvChildRole child_format;
495 extern const BdrvChildRole child_backing;
496 
497 struct BdrvChild {
498     BlockDriverState *bs;
499     char *name;
500     const BdrvChildRole *role;
501     void *opaque;
502 
503     /**
504      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
505      */
506     uint64_t perm;
507 
508     /**
509      * Permissions that can still be granted to other users of @bs while this
510      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
511      */
512     uint64_t shared_perm;
513 
514     QLIST_ENTRY(BdrvChild) next;
515     QLIST_ENTRY(BdrvChild) next_parent;
516 };
517 
518 /*
519  * Note: the function bdrv_append() copies and swaps contents of
520  * BlockDriverStates, so if you add new fields to this struct, please
521  * inspect bdrv_append() to determine if the new fields need to be
522  * copied as well.
523  */
524 struct BlockDriverState {
525     /* Protected by big QEMU lock or read-only after opening.  No special
526      * locking needed during I/O...
527      */
528     int open_flags; /* flags used to open the file, re-used for re-open */
529     bool read_only; /* if true, the media is read only */
530     bool encrypted; /* if true, the media is encrypted */
531     bool valid_key; /* if true, a valid encryption key has been set */
532     bool sg;        /* if true, the device is a /dev/sg* */
533     bool probed;    /* if true, format was probed rather than specified */
534     bool force_share; /* if true, always allow all shared permissions */
535 
536     BlockDriver *drv; /* NULL means no media */
537     void *opaque;
538 
539     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
540     /* long-running tasks intended to always use the same AioContext as this
541      * BDS may register themselves in this list to be notified of changes
542      * regarding this BDS's context */
543     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
544     bool walking_aio_notifiers; /* to make removal during iteration safe */
545 
546     char filename[PATH_MAX];
547     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
548                                     this file image */
549     char backing_format[16]; /* if non-zero and backing_file exists */
550 
551     QDict *full_open_options;
552     char exact_filename[PATH_MAX];
553 
554     BdrvChild *backing;
555     BdrvChild *file;
556 
557     /* I/O Limits */
558     BlockLimits bl;
559 
560     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
561     unsigned int supported_write_flags;
562     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
563      * BDRV_REQ_MAY_UNMAP) */
564     unsigned int supported_zero_flags;
565 
566     /* the following member gives a name to every node on the bs graph. */
567     char node_name[32];
568     /* element of the list of named nodes building the graph */
569     QTAILQ_ENTRY(BlockDriverState) node_list;
570     /* element of the list of all BlockDriverStates (all_bdrv_states) */
571     QTAILQ_ENTRY(BlockDriverState) bs_list;
572     /* element of the list of monitor-owned BDS */
573     QTAILQ_ENTRY(BlockDriverState) monitor_list;
574     int refcnt;
575 
576     /* operation blockers */
577     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
578 
579     /* long-running background operation */
580     BlockJob *job;
581 
582     /* The node that this node inherited default options from (and a reopen on
583      * which can affect this node by changing these defaults). This is always a
584      * parent node of this node. */
585     BlockDriverState *inherits_from;
586     QLIST_HEAD(, BdrvChild) children;
587     QLIST_HEAD(, BdrvChild) parents;
588 
589     QDict *options;
590     QDict *explicit_options;
591     BlockdevDetectZeroesOptions detect_zeroes;
592 
593     /* The error object in use for blocking operations on backing_hd */
594     Error *backing_blocker;
595 
596     /* Protected by AioContext lock */
597 
598     /* If true, copy read backing sectors into image.  Can be >1 if more
599      * than one client has requested copy-on-read.
600      */
601     int copy_on_read;
602 
603     /* If we are reading a disk image, give its size in sectors.
604      * Generally read-only; it is written to by load_vmstate and save_vmstate,
605      * but the block layer is quiescent during those.
606      */
607     int64_t total_sectors;
608 
609     /* Callback before write request is processed */
610     NotifierWithReturnList before_write_notifiers;
611 
612     /* number of in-flight requests; overall and serialising */
613     unsigned int in_flight;
614     unsigned int serialising_in_flight;
615 
616     bool wakeup;
617 
618     /* Offset after the highest byte written to */
619     uint64_t wr_highest_offset;
620 
621     /* threshold limit for writes, in bytes. "High water mark". */
622     uint64_t write_threshold_offset;
623     NotifierWithReturn write_threshold_notifier;
624 
625     /* counter for nested bdrv_io_plug */
626     unsigned io_plugged;
627 
628     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
629     CoQueue flush_queue;                  /* Serializing flush queue */
630     bool active_flush_req;                /* Flush request in flight? */
631     unsigned int write_gen;               /* Current data generation */
632     unsigned int flushed_gen;             /* Flushed write generation */
633 
634     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
635 
636     /* do we need to tell the quest if we have a volatile write cache? */
637     int enable_write_cache;
638 
639     int quiesce_counter;
640 };
641 
642 struct BlockBackendRootState {
643     int open_flags;
644     bool read_only;
645     BlockdevDetectZeroesOptions detect_zeroes;
646 };
647 
648 typedef enum BlockMirrorBackingMode {
649     /* Reuse the existing backing chain from the source for the target.
650      * - sync=full: Set backing BDS to NULL.
651      * - sync=top:  Use source's backing BDS.
652      * - sync=none: Use source as the backing BDS. */
653     MIRROR_SOURCE_BACKING_CHAIN,
654 
655     /* Open the target's backing chain completely anew */
656     MIRROR_OPEN_BACKING_CHAIN,
657 
658     /* Do not change the target's backing BDS after job completion */
659     MIRROR_LEAVE_BACKING_CHAIN,
660 } BlockMirrorBackingMode;
661 
662 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
663 {
664     return bs->backing ? bs->backing->bs : NULL;
665 }
666 
667 
668 /* Essential block drivers which must always be statically linked into qemu, and
669  * which therefore can be accessed without using bdrv_find_format() */
670 extern BlockDriver bdrv_file;
671 extern BlockDriver bdrv_raw;
672 extern BlockDriver bdrv_qcow2;
673 
674 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
675     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
676     BdrvRequestFlags flags);
677 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
678     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
679     BdrvRequestFlags flags);
680 
681 int get_tmp_filename(char *filename, int size);
682 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
683                             const char *filename);
684 
685 
686 /**
687  * bdrv_add_before_write_notifier:
688  *
689  * Register a callback that is invoked before write requests are processed but
690  * after any throttling or waiting for overlapping requests.
691  */
692 void bdrv_add_before_write_notifier(BlockDriverState *bs,
693                                     NotifierWithReturn *notifier);
694 
695 /**
696  * bdrv_detach_aio_context:
697  *
698  * May be called from .bdrv_detach_aio_context() to detach children from the
699  * current #AioContext.  This is only needed by block drivers that manage their
700  * own children.  Both ->file and ->backing are automatically handled and
701  * block drivers should not call this function on them explicitly.
702  */
703 void bdrv_detach_aio_context(BlockDriverState *bs);
704 
705 /**
706  * bdrv_attach_aio_context:
707  *
708  * May be called from .bdrv_attach_aio_context() to attach children to the new
709  * #AioContext.  This is only needed by block drivers that manage their own
710  * children.  Both ->file and ->backing are automatically handled and block
711  * drivers should not call this function on them explicitly.
712  */
713 void bdrv_attach_aio_context(BlockDriverState *bs,
714                              AioContext *new_context);
715 
716 /**
717  * bdrv_add_aio_context_notifier:
718  *
719  * If a long-running job intends to be always run in the same AioContext as a
720  * certain BDS, it may use this function to be notified of changes regarding the
721  * association of the BDS to an AioContext.
722  *
723  * attached_aio_context() is called after the target BDS has been attached to a
724  * new AioContext; detach_aio_context() is called before the target BDS is being
725  * detached from its old AioContext.
726  */
727 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
728         void (*attached_aio_context)(AioContext *new_context, void *opaque),
729         void (*detach_aio_context)(void *opaque), void *opaque);
730 
731 /**
732  * bdrv_remove_aio_context_notifier:
733  *
734  * Unsubscribe of change notifications regarding the BDS's AioContext. The
735  * parameters given here have to be the same as those given to
736  * bdrv_add_aio_context_notifier().
737  */
738 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
739                                       void (*aio_context_attached)(AioContext *,
740                                                                    void *),
741                                       void (*aio_context_detached)(void *),
742                                       void *opaque);
743 
744 /**
745  * bdrv_wakeup:
746  * @bs: The BlockDriverState for which an I/O operation has been completed.
747  *
748  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
749  * synchronous I/O on a BlockDriverState that is attached to another
750  * I/O thread, the main thread lets the I/O thread's event loop run,
751  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
752  * up the main thread if necessary.
753  *
754  * Manual calls to bdrv_wakeup are rarely necessary, because
755  * bdrv_dec_in_flight already calls it.
756  */
757 void bdrv_wakeup(BlockDriverState *bs);
758 
759 #ifdef _WIN32
760 int is_windows_drive(const char *filename);
761 #endif
762 
763 /**
764  * stream_start:
765  * @job_id: The id of the newly-created job, or %NULL to use the
766  * device name of @bs.
767  * @bs: Block device to operate on.
768  * @base: Block device that will become the new base, or %NULL to
769  * flatten the whole backing file chain onto @bs.
770  * @backing_file_str: The file name that will be written to @bs as the
771  * the new backing file if the job completes. Ignored if @base is %NULL.
772  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
773  * @on_error: The action to take upon error.
774  * @errp: Error object.
775  *
776  * Start a streaming operation on @bs.  Clusters that are unallocated
777  * in @bs, but allocated in any image between @base and @bs (both
778  * exclusive) will be written to @bs.  At the end of a successful
779  * streaming job, the backing file of @bs will be changed to
780  * @backing_file_str in the written image and to @base in the live
781  * BlockDriverState.
782  */
783 void stream_start(const char *job_id, BlockDriverState *bs,
784                   BlockDriverState *base, const char *backing_file_str,
785                   int64_t speed, BlockdevOnError on_error, Error **errp);
786 
787 /**
788  * commit_start:
789  * @job_id: The id of the newly-created job, or %NULL to use the
790  * device name of @bs.
791  * @bs: Active block device.
792  * @top: Top block device to be committed.
793  * @base: Block device that will be written into, and become the new top.
794  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
795  * @on_error: The action to take upon error.
796  * @backing_file_str: String to use as the backing file in @top's overlay
797  * @filter_node_name: The node name that should be assigned to the filter
798  * driver that the commit job inserts into the graph above @top. NULL means
799  * that a node name should be autogenerated.
800  * @errp: Error object.
801  *
802  */
803 void commit_start(const char *job_id, BlockDriverState *bs,
804                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
805                   BlockdevOnError on_error, const char *backing_file_str,
806                   const char *filter_node_name, Error **errp);
807 /**
808  * commit_active_start:
809  * @job_id: The id of the newly-created job, or %NULL to use the
810  * device name of @bs.
811  * @bs: Active block device to be committed.
812  * @base: Block device that will be written into, and become the new top.
813  * @creation_flags: Flags that control the behavior of the Job lifetime.
814  *                  See @BlockJobCreateFlags
815  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
816  * @on_error: The action to take upon error.
817  * @filter_node_name: The node name that should be assigned to the filter
818  * driver that the commit job inserts into the graph above @bs. NULL means that
819  * a node name should be autogenerated.
820  * @cb: Completion function for the job.
821  * @opaque: Opaque pointer value passed to @cb.
822  * @auto_complete: Auto complete the job.
823  * @errp: Error object.
824  *
825  */
826 void commit_active_start(const char *job_id, BlockDriverState *bs,
827                          BlockDriverState *base, int creation_flags,
828                          int64_t speed, BlockdevOnError on_error,
829                          const char *filter_node_name,
830                          BlockCompletionFunc *cb, void *opaque,
831                          bool auto_complete, Error **errp);
832 /*
833  * mirror_start:
834  * @job_id: The id of the newly-created job, or %NULL to use the
835  * device name of @bs.
836  * @bs: Block device to operate on.
837  * @target: Block device to write to.
838  * @replaces: Block graph node name to replace once the mirror is done. Can
839  *            only be used when full mirroring is selected.
840  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
841  * @granularity: The chosen granularity for the dirty bitmap.
842  * @buf_size: The amount of data that can be in flight at one time.
843  * @mode: Whether to collapse all images in the chain to the target.
844  * @backing_mode: How to establish the target's backing chain after completion.
845  * @on_source_error: The action to take upon error reading from the source.
846  * @on_target_error: The action to take upon error writing to the target.
847  * @unmap: Whether to unmap target where source sectors only contain zeroes.
848  * @filter_node_name: The node name that should be assigned to the filter
849  * driver that the mirror job inserts into the graph above @bs. NULL means that
850  * a node name should be autogenerated.
851  * @errp: Error object.
852  *
853  * Start a mirroring operation on @bs.  Clusters that are allocated
854  * in @bs will be written to @target until the job is cancelled or
855  * manually completed.  At the end of a successful mirroring job,
856  * @bs will be switched to read from @target.
857  */
858 void mirror_start(const char *job_id, BlockDriverState *bs,
859                   BlockDriverState *target, const char *replaces,
860                   int64_t speed, uint32_t granularity, int64_t buf_size,
861                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
862                   BlockdevOnError on_source_error,
863                   BlockdevOnError on_target_error,
864                   bool unmap, const char *filter_node_name, Error **errp);
865 
866 /*
867  * backup_job_create:
868  * @job_id: The id of the newly-created job, or %NULL to use the
869  * device name of @bs.
870  * @bs: Block device to operate on.
871  * @target: Block device to write to.
872  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
873  * @sync_mode: What parts of the disk image should be copied to the destination.
874  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
875  * @on_source_error: The action to take upon error reading from the source.
876  * @on_target_error: The action to take upon error writing to the target.
877  * @creation_flags: Flags that control the behavior of the Job lifetime.
878  *                  See @BlockJobCreateFlags
879  * @cb: Completion function for the job.
880  * @opaque: Opaque pointer value passed to @cb.
881  * @txn: Transaction that this job is part of (may be NULL).
882  *
883  * Create a backup operation on @bs.  Clusters in @bs are written to @target
884  * until the job is cancelled or manually completed.
885  */
886 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
887                             BlockDriverState *target, int64_t speed,
888                             MirrorSyncMode sync_mode,
889                             BdrvDirtyBitmap *sync_bitmap,
890                             bool compress,
891                             BlockdevOnError on_source_error,
892                             BlockdevOnError on_target_error,
893                             int creation_flags,
894                             BlockCompletionFunc *cb, void *opaque,
895                             BlockJobTxn *txn, Error **errp);
896 
897 void hmp_drive_add_node(Monitor *mon, const char *optstr);
898 
899 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
900                                   const char *child_name,
901                                   const BdrvChildRole *child_role,
902                                   uint64_t perm, uint64_t shared_perm,
903                                   void *opaque, Error **errp);
904 void bdrv_root_unref_child(BdrvChild *child);
905 
906 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
907                             Error **errp);
908 
909 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
910  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
911  * all children */
912 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
913                                const BdrvChildRole *role,
914                                uint64_t perm, uint64_t shared,
915                                uint64_t *nperm, uint64_t *nshared);
916 
917 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
918  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
919  * requires WRITE | RESIZE for read-write images, always requires
920  * CONSISTENT_READ and doesn't share WRITE. */
921 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
922                                const BdrvChildRole *role,
923                                uint64_t perm, uint64_t shared,
924                                uint64_t *nperm, uint64_t *nshared);
925 
926 const char *bdrv_get_parent_name(const BlockDriverState *bs);
927 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
928 bool blk_dev_has_removable_media(BlockBackend *blk);
929 bool blk_dev_has_tray(BlockBackend *blk);
930 void blk_dev_eject_request(BlockBackend *blk, bool force);
931 bool blk_dev_is_tray_open(BlockBackend *blk);
932 bool blk_dev_is_medium_locked(BlockBackend *blk);
933 
934 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
935 bool bdrv_requests_pending(BlockDriverState *bs);
936 
937 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
938 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
939 
940 void bdrv_inc_in_flight(BlockDriverState *bs);
941 void bdrv_dec_in_flight(BlockDriverState *bs);
942 
943 void blockdev_close_all_bdrv_states(void);
944 
945 #endif /* BLOCK_INT_H */
946