xref: /openbmc/qemu/include/block/block_int.h (revision bb2e0039)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #ifndef BLOCK_INT_H
25 #define BLOCK_INT_H
26 
27 #include "block/accounting.h"
28 #include "block/block.h"
29 #include "qemu/option.h"
30 #include "qemu/queue.h"
31 #include "qemu/coroutine.h"
32 #include "qemu/stats64.h"
33 #include "qemu/timer.h"
34 #include "qapi-types.h"
35 #include "qemu/hbitmap.h"
36 #include "block/snapshot.h"
37 #include "qemu/main-loop.h"
38 #include "qemu/throttle.h"
39 
40 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
41 
42 #define BLOCK_OPT_SIZE              "size"
43 #define BLOCK_OPT_ENCRYPT           "encryption"
44 #define BLOCK_OPT_ENCRYPT_FORMAT    "encrypt.format"
45 #define BLOCK_OPT_COMPAT6           "compat6"
46 #define BLOCK_OPT_HWVERSION         "hwversion"
47 #define BLOCK_OPT_BACKING_FILE      "backing_file"
48 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
49 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
50 #define BLOCK_OPT_TABLE_SIZE        "table_size"
51 #define BLOCK_OPT_PREALLOC          "preallocation"
52 #define BLOCK_OPT_SUBFMT            "subformat"
53 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
54 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
55 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
56 #define BLOCK_OPT_REDUNDANCY        "redundancy"
57 #define BLOCK_OPT_NOCOW             "nocow"
58 #define BLOCK_OPT_OBJECT_SIZE       "object_size"
59 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
60 
61 #define BLOCK_PROBE_BUF_SIZE        512
62 
63 enum BdrvTrackedRequestType {
64     BDRV_TRACKED_READ,
65     BDRV_TRACKED_WRITE,
66     BDRV_TRACKED_DISCARD,
67 };
68 
69 typedef struct BdrvTrackedRequest {
70     BlockDriverState *bs;
71     int64_t offset;
72     unsigned int bytes;
73     enum BdrvTrackedRequestType type;
74 
75     bool serialising;
76     int64_t overlap_offset;
77     unsigned int overlap_bytes;
78 
79     QLIST_ENTRY(BdrvTrackedRequest) list;
80     Coroutine *co; /* owner, used for deadlock detection */
81     CoQueue wait_queue; /* coroutines blocked on this request */
82 
83     struct BdrvTrackedRequest *waiting_for;
84 } BdrvTrackedRequest;
85 
86 struct BlockDriver {
87     const char *format_name;
88     int instance_size;
89 
90     /* set to true if the BlockDriver is a block filter */
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_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 bytes,
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 
150     /**
151      * @offset: position in bytes to read at
152      * @bytes: number of bytes to read
153      * @qiov: the buffers to fill with read data
154      * @flags: currently unused, always 0
155      *
156      * @offset and @bytes will be a multiple of 'request_alignment',
157      * but the length of individual @qiov elements does not have to
158      * be a multiple.
159      *
160      * @bytes will always equal the total size of @qiov, and will be
161      * no larger than 'max_transfer'.
162      *
163      * The buffer in @qiov may point directly to guest memory.
164      */
165     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
166         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
167     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
168         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
169     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
170         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
171     /**
172      * @offset: position in bytes to write at
173      * @bytes: number of bytes to write
174      * @qiov: the buffers containing data to write
175      * @flags: zero or more bits allowed by 'supported_write_flags'
176      *
177      * @offset and @bytes will be a multiple of 'request_alignment',
178      * but the length of individual @qiov elements does not have to
179      * be a multiple.
180      *
181      * @bytes will always equal the total size of @qiov, and will be
182      * no larger than 'max_transfer'.
183      *
184      * The buffer in @qiov may point directly to guest memory.
185      */
186     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
187         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
188 
189     /*
190      * Efficiently zero a region of the disk image.  Typically an image format
191      * would use a compact metadata representation to implement this.  This
192      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
193      * will be called instead.
194      */
195     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
196         int64_t offset, int bytes, BdrvRequestFlags flags);
197     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
198         int64_t offset, int bytes);
199 
200     /*
201      * Building block for bdrv_block_status[_above]. The driver should
202      * answer only according to the current layer, and should not
203      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
204      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
205      */
206     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
207         int64_t sector_num, int nb_sectors, int *pnum,
208         BlockDriverState **file);
209 
210     /*
211      * Invalidate any cached meta-data.
212      */
213     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
214     int (*bdrv_inactivate)(BlockDriverState *bs);
215 
216     /*
217      * Flushes all data for all layers by calling bdrv_co_flush for underlying
218      * layers, if needed. This function is needed for deterministic
219      * synchronization of the flush finishing callback.
220      */
221     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
222 
223     /*
224      * Flushes all data that was already written to the OS all the way down to
225      * the disk (for example file-posix.c calls fsync()).
226      */
227     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
228 
229     /*
230      * Flushes all internal caches to the OS. The data may still sit in a
231      * writeback cache of the host OS, but it will survive a crash of the qemu
232      * process.
233      */
234     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
235 
236     const char *protocol_name;
237     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset,
238                          PreallocMode prealloc, Error **errp);
239 
240     int64_t (*bdrv_getlength)(BlockDriverState *bs);
241     bool has_variable_length;
242     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
243     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
244                                       Error **errp);
245 
246     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
247         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
248 
249     int (*bdrv_snapshot_create)(BlockDriverState *bs,
250                                 QEMUSnapshotInfo *sn_info);
251     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
252                               const char *snapshot_id);
253     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
254                                 const char *snapshot_id,
255                                 const char *name,
256                                 Error **errp);
257     int (*bdrv_snapshot_list)(BlockDriverState *bs,
258                               QEMUSnapshotInfo **psn_info);
259     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
260                                   const char *snapshot_id,
261                                   const char *name,
262                                   Error **errp);
263     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
264     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
265 
266     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
267                                           QEMUIOVector *qiov,
268                                           int64_t pos);
269     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
270                                           QEMUIOVector *qiov,
271                                           int64_t pos);
272 
273     int (*bdrv_change_backing_file)(BlockDriverState *bs,
274         const char *backing_file, const char *backing_fmt);
275 
276     /* removable device specific */
277     bool (*bdrv_is_inserted)(BlockDriverState *bs);
278     int (*bdrv_media_changed)(BlockDriverState *bs);
279     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
280     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
281 
282     /* to control generic scsi devices */
283     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
284         unsigned long int req, void *buf,
285         BlockCompletionFunc *cb, void *opaque);
286     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
287                                       unsigned long int req, void *buf);
288 
289     /* List of options for creating images, terminated by name == NULL */
290     QemuOptsList *create_opts;
291 
292     /*
293      * Returns 0 for completed check, -errno for internal errors.
294      * The check results are stored in result.
295      */
296     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
297         BdrvCheckMode fix);
298 
299     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
300                               BlockDriverAmendStatusCB *status_cb,
301                               void *cb_opaque);
302 
303     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
304 
305     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
306     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
307         const char *tag);
308     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
309         const char *tag);
310     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
311     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
312 
313     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
314 
315     /*
316      * Returns 1 if newly created images are guaranteed to contain only
317      * zeros, 0 otherwise.
318      */
319     int (*bdrv_has_zero_init)(BlockDriverState *bs);
320 
321     /* Remove fd handlers, timers, and other event loop callbacks so the event
322      * loop is no longer in use.  Called with no in-flight requests and in
323      * depth-first traversal order with parents before child nodes.
324      */
325     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
326 
327     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
328      * can be processed again.  Called with no in-flight requests and in
329      * depth-first traversal order with child nodes before parent nodes.
330      */
331     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
332                                     AioContext *new_context);
333 
334     /* io queue for linux-aio */
335     void (*bdrv_io_plug)(BlockDriverState *bs);
336     void (*bdrv_io_unplug)(BlockDriverState *bs);
337 
338     /**
339      * Try to get @bs's logical and physical block size.
340      * On success, store them in @bsz and return zero.
341      * On failure, return negative errno.
342      */
343     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
344     /**
345      * Try to get @bs's geometry (cyls, heads, sectors)
346      * On success, store them in @geo and return 0.
347      * On failure return -errno.
348      * Only drivers that want to override guest geometry implement this
349      * callback; see hd_geometry_guess().
350      */
351     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
352 
353     /**
354      * Drain and stop any internal sources of requests in the driver, and
355      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
356      */
357     void coroutine_fn (*bdrv_co_drain)(BlockDriverState *bs);
358 
359     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
360                            Error **errp);
361     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
362                            Error **errp);
363 
364     /**
365      * Informs the block driver that a permission change is intended. The
366      * driver checks whether the change is permissible and may take other
367      * preparations for the change (e.g. get file system locks). This operation
368      * is always followed either by a call to either .bdrv_set_perm or
369      * .bdrv_abort_perm_update.
370      *
371      * Checks whether the requested set of cumulative permissions in @perm
372      * can be granted for accessing @bs and whether no other users are using
373      * permissions other than those given in @shared (both arguments take
374      * BLK_PERM_* bitmasks).
375      *
376      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
377      * and errp is set to an error describing the conflict.
378      */
379     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
380                            uint64_t shared, Error **errp);
381 
382     /**
383      * Called to inform the driver that the set of cumulative set of used
384      * permissions for @bs has changed to @perm, and the set of sharable
385      * permission to @shared. The driver can use this to propagate changes to
386      * its children (i.e. request permissions only if a parent actually needs
387      * them).
388      *
389      * This function is only invoked after bdrv_check_perm(), so block drivers
390      * may rely on preparations made in their .bdrv_check_perm implementation.
391      */
392     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
393 
394     /*
395      * Called to inform the driver that after a previous bdrv_check_perm()
396      * call, the permission update is not performed and any preparations made
397      * for it (e.g. taken file locks) need to be undone.
398      *
399      * This function can be called even for nodes that never saw a
400      * bdrv_check_perm() call. It is a no-op then.
401      */
402     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
403 
404     /**
405      * Returns in @nperm and @nshared the permissions that the driver for @bs
406      * needs on its child @c, based on the cumulative permissions requested by
407      * the parents in @parent_perm and @parent_shared.
408      *
409      * If @c is NULL, return the permissions for attaching a new child for the
410      * given @role.
411      */
412      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
413                              const BdrvChildRole *role,
414                              uint64_t parent_perm, uint64_t parent_shared,
415                              uint64_t *nperm, uint64_t *nshared);
416 
417     /**
418      * Bitmaps should be marked as 'IN_USE' in the image on reopening image
419      * as rw. This handler should realize it. It also should unset readonly
420      * field of BlockDirtyBitmap's in case of success.
421      */
422     int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
423     bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
424                                             const char *name,
425                                             uint32_t granularity,
426                                             Error **errp);
427     void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
428                                                 const char *name,
429                                                 Error **errp);
430 
431     QLIST_ENTRY(BlockDriver) list;
432 };
433 
434 typedef struct BlockLimits {
435     /* Alignment requirement, in bytes, for offset/length of I/O
436      * requests. Must be a power of 2 less than INT_MAX; defaults to
437      * 1 for drivers with modern byte interfaces, and to 512
438      * otherwise. */
439     uint32_t request_alignment;
440 
441     /* Maximum number of bytes that can be discarded at once (since it
442      * is signed, it must be < 2G, if set). Must be multiple of
443      * pdiscard_alignment, but need not be power of 2. May be 0 if no
444      * inherent 32-bit limit */
445     int32_t max_pdiscard;
446 
447     /* Optimal alignment for discard requests in bytes. A power of 2
448      * is best but not mandatory.  Must be a multiple of
449      * bl.request_alignment, and must be less than max_pdiscard if
450      * that is set. May be 0 if bl.request_alignment is good enough */
451     uint32_t pdiscard_alignment;
452 
453     /* Maximum number of bytes that can zeroized at once (since it is
454      * signed, it must be < 2G, if set). Must be multiple of
455      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
456     int32_t max_pwrite_zeroes;
457 
458     /* Optimal alignment for write zeroes requests in bytes. A power
459      * of 2 is best but not mandatory.  Must be a multiple of
460      * bl.request_alignment, and must be less than max_pwrite_zeroes
461      * if that is set. May be 0 if bl.request_alignment is good
462      * enough */
463     uint32_t pwrite_zeroes_alignment;
464 
465     /* Optimal transfer length in bytes.  A power of 2 is best but not
466      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
467      * no preferred size */
468     uint32_t opt_transfer;
469 
470     /* Maximal transfer length in bytes.  Need not be power of 2, but
471      * must be multiple of opt_transfer and bl.request_alignment, or 0
472      * for no 32-bit limit.  For now, anything larger than INT_MAX is
473      * clamped down. */
474     uint32_t max_transfer;
475 
476     /* memory alignment, in bytes so that no bounce buffer is needed */
477     size_t min_mem_alignment;
478 
479     /* memory alignment, in bytes, for bounce buffer */
480     size_t opt_mem_alignment;
481 
482     /* maximum number of iovec elements */
483     int max_iov;
484 } BlockLimits;
485 
486 typedef struct BdrvOpBlocker BdrvOpBlocker;
487 
488 typedef struct BdrvAioNotifier {
489     void (*attached_aio_context)(AioContext *new_context, void *opaque);
490     void (*detach_aio_context)(void *opaque);
491 
492     void *opaque;
493     bool deleted;
494 
495     QLIST_ENTRY(BdrvAioNotifier) list;
496 } BdrvAioNotifier;
497 
498 struct BdrvChildRole {
499     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
500      * points to. */
501     bool stay_at_node;
502 
503     void (*inherit_options)(int *child_flags, QDict *child_options,
504                             int parent_flags, QDict *parent_options);
505 
506     void (*change_media)(BdrvChild *child, bool load);
507     void (*resize)(BdrvChild *child);
508 
509     /* Returns a name that is supposedly more useful for human users than the
510      * node name for identifying the node in question (in particular, a BB
511      * name), or NULL if the parent can't provide a better name. */
512     const char *(*get_name)(BdrvChild *child);
513 
514     /* Returns a malloced string that describes the parent of the child for a
515      * human reader. This could be a node-name, BlockBackend name, qdev ID or
516      * QOM path of the device owning the BlockBackend, job type and ID etc. The
517      * caller is responsible for freeing the memory. */
518     char *(*get_parent_desc)(BdrvChild *child);
519 
520     /*
521      * If this pair of functions is implemented, the parent doesn't issue new
522      * requests after returning from .drained_begin() until .drained_end() is
523      * called.
524      *
525      * Note that this can be nested. If drained_begin() was called twice, new
526      * I/O is allowed only after drained_end() was called twice, too.
527      */
528     void (*drained_begin)(BdrvChild *child);
529     void (*drained_end)(BdrvChild *child);
530 
531     /* Notifies the parent that the child has been activated/inactivated (e.g.
532      * when migration is completing) and it can start/stop requesting
533      * permissions and doing I/O on it. */
534     void (*activate)(BdrvChild *child, Error **errp);
535     int (*inactivate)(BdrvChild *child);
536 
537     void (*attach)(BdrvChild *child);
538     void (*detach)(BdrvChild *child);
539 };
540 
541 extern const BdrvChildRole child_file;
542 extern const BdrvChildRole child_format;
543 extern const BdrvChildRole child_backing;
544 
545 struct BdrvChild {
546     BlockDriverState *bs;
547     char *name;
548     const BdrvChildRole *role;
549     void *opaque;
550 
551     /**
552      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
553      */
554     uint64_t perm;
555 
556     /**
557      * Permissions that can still be granted to other users of @bs while this
558      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
559      */
560     uint64_t shared_perm;
561 
562     QLIST_ENTRY(BdrvChild) next;
563     QLIST_ENTRY(BdrvChild) next_parent;
564 };
565 
566 /*
567  * Note: the function bdrv_append() copies and swaps contents of
568  * BlockDriverStates, so if you add new fields to this struct, please
569  * inspect bdrv_append() to determine if the new fields need to be
570  * copied as well.
571  */
572 struct BlockDriverState {
573     /* Protected by big QEMU lock or read-only after opening.  No special
574      * locking needed during I/O...
575      */
576     int open_flags; /* flags used to open the file, re-used for re-open */
577     bool read_only; /* if true, the media is read only */
578     bool encrypted; /* if true, the media is encrypted */
579     bool sg;        /* if true, the device is a /dev/sg* */
580     bool probed;    /* if true, format was probed rather than specified */
581     bool force_share; /* if true, always allow all shared permissions */
582     bool implicit;  /* if true, this filter node was automatically inserted */
583 
584     BlockDriver *drv; /* NULL means no media */
585     void *opaque;
586 
587     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
588     /* long-running tasks intended to always use the same AioContext as this
589      * BDS may register themselves in this list to be notified of changes
590      * regarding this BDS's context */
591     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
592     bool walking_aio_notifiers; /* to make removal during iteration safe */
593 
594     char filename[PATH_MAX];
595     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
596                                     this file image */
597     char backing_format[16]; /* if non-zero and backing_file exists */
598 
599     QDict *full_open_options;
600     char exact_filename[PATH_MAX];
601 
602     BdrvChild *backing;
603     BdrvChild *file;
604 
605     /* I/O Limits */
606     BlockLimits bl;
607 
608     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
609     unsigned int supported_write_flags;
610     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
611      * BDRV_REQ_MAY_UNMAP) */
612     unsigned int supported_zero_flags;
613 
614     /* the following member gives a name to every node on the bs graph. */
615     char node_name[32];
616     /* element of the list of named nodes building the graph */
617     QTAILQ_ENTRY(BlockDriverState) node_list;
618     /* element of the list of all BlockDriverStates (all_bdrv_states) */
619     QTAILQ_ENTRY(BlockDriverState) bs_list;
620     /* element of the list of monitor-owned BDS */
621     QTAILQ_ENTRY(BlockDriverState) monitor_list;
622     int refcnt;
623 
624     /* operation blockers */
625     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
626 
627     /* long-running background operation */
628     BlockJob *job;
629 
630     /* The node that this node inherited default options from (and a reopen on
631      * which can affect this node by changing these defaults). This is always a
632      * parent node of this node. */
633     BlockDriverState *inherits_from;
634     QLIST_HEAD(, BdrvChild) children;
635     QLIST_HEAD(, BdrvChild) parents;
636 
637     QDict *options;
638     QDict *explicit_options;
639     BlockdevDetectZeroesOptions detect_zeroes;
640 
641     /* The error object in use for blocking operations on backing_hd */
642     Error *backing_blocker;
643 
644     /* Protected by AioContext lock */
645 
646     /* If we are reading a disk image, give its size in sectors.
647      * Generally read-only; it is written to by load_snapshot and
648      * save_snaphost, but the block layer is quiescent during those.
649      */
650     int64_t total_sectors;
651 
652     /* Callback before write request is processed */
653     NotifierWithReturnList before_write_notifiers;
654 
655     /* threshold limit for writes, in bytes. "High water mark". */
656     uint64_t write_threshold_offset;
657     NotifierWithReturn write_threshold_notifier;
658 
659     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
660      * Reading from the list can be done with either the BQL or the
661      * dirty_bitmap_mutex.  Modifying a bitmap only requires
662      * dirty_bitmap_mutex.  */
663     QemuMutex dirty_bitmap_mutex;
664     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
665 
666     /* Offset after the highest byte written to */
667     Stat64 wr_highest_offset;
668 
669     /* If true, copy read backing sectors into image.  Can be >1 if more
670      * than one client has requested copy-on-read.  Accessed with atomic
671      * ops.
672      */
673     int copy_on_read;
674 
675     /* number of in-flight requests; overall and serialising.
676      * Accessed with atomic ops.
677      */
678     unsigned int in_flight;
679     unsigned int serialising_in_flight;
680 
681     /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
682      * ops.
683      */
684     bool wakeup;
685 
686     /* counter for nested bdrv_io_plug.
687      * Accessed with atomic ops.
688     */
689     unsigned io_plugged;
690 
691     /* do we need to tell the quest if we have a volatile write cache? */
692     int enable_write_cache;
693 
694     /* Accessed with atomic ops.  */
695     int quiesce_counter;
696     unsigned int write_gen;               /* Current data generation */
697 
698     /* Protected by reqs_lock.  */
699     CoMutex reqs_lock;
700     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
701     CoQueue flush_queue;                  /* Serializing flush queue */
702     bool active_flush_req;                /* Flush request in flight? */
703 
704     /* Only read/written by whoever has set active_flush_req to true.  */
705     unsigned int flushed_gen;             /* Flushed write generation */
706 };
707 
708 struct BlockBackendRootState {
709     int open_flags;
710     bool read_only;
711     BlockdevDetectZeroesOptions detect_zeroes;
712 };
713 
714 typedef enum BlockMirrorBackingMode {
715     /* Reuse the existing backing chain from the source for the target.
716      * - sync=full: Set backing BDS to NULL.
717      * - sync=top:  Use source's backing BDS.
718      * - sync=none: Use source as the backing BDS. */
719     MIRROR_SOURCE_BACKING_CHAIN,
720 
721     /* Open the target's backing chain completely anew */
722     MIRROR_OPEN_BACKING_CHAIN,
723 
724     /* Do not change the target's backing BDS after job completion */
725     MIRROR_LEAVE_BACKING_CHAIN,
726 } BlockMirrorBackingMode;
727 
728 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
729 {
730     return bs->backing ? bs->backing->bs : NULL;
731 }
732 
733 
734 /* Essential block drivers which must always be statically linked into qemu, and
735  * which therefore can be accessed without using bdrv_find_format() */
736 extern BlockDriver bdrv_file;
737 extern BlockDriver bdrv_raw;
738 extern BlockDriver bdrv_qcow2;
739 
740 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
741     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
742     BdrvRequestFlags flags);
743 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
744     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
745     BdrvRequestFlags flags);
746 
747 int get_tmp_filename(char *filename, int size);
748 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
749                             const char *filename);
750 
751 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
752                                       QDict *options);
753 
754 
755 /**
756  * bdrv_add_before_write_notifier:
757  *
758  * Register a callback that is invoked before write requests are processed but
759  * after any throttling or waiting for overlapping requests.
760  */
761 void bdrv_add_before_write_notifier(BlockDriverState *bs,
762                                     NotifierWithReturn *notifier);
763 
764 /**
765  * bdrv_detach_aio_context:
766  *
767  * May be called from .bdrv_detach_aio_context() to detach children from the
768  * current #AioContext.  This is only needed by block drivers that manage their
769  * own children.  Both ->file and ->backing are automatically handled and
770  * block drivers should not call this function on them explicitly.
771  */
772 void bdrv_detach_aio_context(BlockDriverState *bs);
773 
774 /**
775  * bdrv_attach_aio_context:
776  *
777  * May be called from .bdrv_attach_aio_context() to attach children to the new
778  * #AioContext.  This is only needed by block drivers that manage their own
779  * children.  Both ->file and ->backing are automatically handled and block
780  * drivers should not call this function on them explicitly.
781  */
782 void bdrv_attach_aio_context(BlockDriverState *bs,
783                              AioContext *new_context);
784 
785 /**
786  * bdrv_add_aio_context_notifier:
787  *
788  * If a long-running job intends to be always run in the same AioContext as a
789  * certain BDS, it may use this function to be notified of changes regarding the
790  * association of the BDS to an AioContext.
791  *
792  * attached_aio_context() is called after the target BDS has been attached to a
793  * new AioContext; detach_aio_context() is called before the target BDS is being
794  * detached from its old AioContext.
795  */
796 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
797         void (*attached_aio_context)(AioContext *new_context, void *opaque),
798         void (*detach_aio_context)(void *opaque), void *opaque);
799 
800 /**
801  * bdrv_remove_aio_context_notifier:
802  *
803  * Unsubscribe of change notifications regarding the BDS's AioContext. The
804  * parameters given here have to be the same as those given to
805  * bdrv_add_aio_context_notifier().
806  */
807 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
808                                       void (*aio_context_attached)(AioContext *,
809                                                                    void *),
810                                       void (*aio_context_detached)(void *),
811                                       void *opaque);
812 
813 /**
814  * bdrv_wakeup:
815  * @bs: The BlockDriverState for which an I/O operation has been completed.
816  *
817  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
818  * synchronous I/O on a BlockDriverState that is attached to another
819  * I/O thread, the main thread lets the I/O thread's event loop run,
820  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
821  * up the main thread if necessary.
822  *
823  * Manual calls to bdrv_wakeup are rarely necessary, because
824  * bdrv_dec_in_flight already calls it.
825  */
826 void bdrv_wakeup(BlockDriverState *bs);
827 
828 #ifdef _WIN32
829 int is_windows_drive(const char *filename);
830 #endif
831 
832 /**
833  * stream_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  * @base: Block device that will become the new base, or %NULL to
838  * flatten the whole backing file chain onto @bs.
839  * @backing_file_str: The file name that will be written to @bs as the
840  * the new backing file if the job completes. Ignored if @base is %NULL.
841  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
842  * @on_error: The action to take upon error.
843  * @errp: Error object.
844  *
845  * Start a streaming operation on @bs.  Clusters that are unallocated
846  * in @bs, but allocated in any image between @base and @bs (both
847  * exclusive) will be written to @bs.  At the end of a successful
848  * streaming job, the backing file of @bs will be changed to
849  * @backing_file_str in the written image and to @base in the live
850  * BlockDriverState.
851  */
852 void stream_start(const char *job_id, BlockDriverState *bs,
853                   BlockDriverState *base, const char *backing_file_str,
854                   int64_t speed, BlockdevOnError on_error, Error **errp);
855 
856 /**
857  * commit_start:
858  * @job_id: The id of the newly-created job, or %NULL to use the
859  * device name of @bs.
860  * @bs: Active block device.
861  * @top: Top block device to be committed.
862  * @base: Block device that will be written into, and become the new top.
863  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
864  * @on_error: The action to take upon error.
865  * @backing_file_str: String to use as the backing file in @top's overlay
866  * @filter_node_name: The node name that should be assigned to the filter
867  * driver that the commit job inserts into the graph above @top. NULL means
868  * that a node name should be autogenerated.
869  * @errp: Error object.
870  *
871  */
872 void commit_start(const char *job_id, BlockDriverState *bs,
873                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
874                   BlockdevOnError on_error, const char *backing_file_str,
875                   const char *filter_node_name, Error **errp);
876 /**
877  * commit_active_start:
878  * @job_id: The id of the newly-created job, or %NULL to use the
879  * device name of @bs.
880  * @bs: Active block device to be committed.
881  * @base: Block device that will be written into, and become the new top.
882  * @creation_flags: Flags that control the behavior of the Job lifetime.
883  *                  See @BlockJobCreateFlags
884  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
885  * @on_error: The action to take upon error.
886  * @filter_node_name: The node name that should be assigned to the filter
887  * driver that the commit job inserts into the graph above @bs. NULL means that
888  * a node name should be autogenerated.
889  * @cb: Completion function for the job.
890  * @opaque: Opaque pointer value passed to @cb.
891  * @auto_complete: Auto complete the job.
892  * @errp: Error object.
893  *
894  */
895 void commit_active_start(const char *job_id, BlockDriverState *bs,
896                          BlockDriverState *base, int creation_flags,
897                          int64_t speed, BlockdevOnError on_error,
898                          const char *filter_node_name,
899                          BlockCompletionFunc *cb, void *opaque,
900                          bool auto_complete, Error **errp);
901 /*
902  * mirror_start:
903  * @job_id: The id of the newly-created job, or %NULL to use the
904  * device name of @bs.
905  * @bs: Block device to operate on.
906  * @target: Block device to write to.
907  * @replaces: Block graph node name to replace once the mirror is done. Can
908  *            only be used when full mirroring is selected.
909  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
910  * @granularity: The chosen granularity for the dirty bitmap.
911  * @buf_size: The amount of data that can be in flight at one time.
912  * @mode: Whether to collapse all images in the chain to the target.
913  * @backing_mode: How to establish the target's backing chain after completion.
914  * @on_source_error: The action to take upon error reading from the source.
915  * @on_target_error: The action to take upon error writing to the target.
916  * @unmap: Whether to unmap target where source sectors only contain zeroes.
917  * @filter_node_name: The node name that should be assigned to the filter
918  * driver that the mirror job inserts into the graph above @bs. NULL means that
919  * a node name should be autogenerated.
920  * @errp: Error object.
921  *
922  * Start a mirroring operation on @bs.  Clusters that are allocated
923  * in @bs will be written to @target until the job is cancelled or
924  * manually completed.  At the end of a successful mirroring job,
925  * @bs will be switched to read from @target.
926  */
927 void mirror_start(const char *job_id, BlockDriverState *bs,
928                   BlockDriverState *target, const char *replaces,
929                   int64_t speed, uint32_t granularity, int64_t buf_size,
930                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
931                   BlockdevOnError on_source_error,
932                   BlockdevOnError on_target_error,
933                   bool unmap, const char *filter_node_name, Error **errp);
934 
935 /*
936  * backup_job_create:
937  * @job_id: The id of the newly-created job, or %NULL to use the
938  * device name of @bs.
939  * @bs: Block device to operate on.
940  * @target: Block device to write to.
941  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
942  * @sync_mode: What parts of the disk image should be copied to the destination.
943  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
944  * @on_source_error: The action to take upon error reading from the source.
945  * @on_target_error: The action to take upon error writing to the target.
946  * @creation_flags: Flags that control the behavior of the Job lifetime.
947  *                  See @BlockJobCreateFlags
948  * @cb: Completion function for the job.
949  * @opaque: Opaque pointer value passed to @cb.
950  * @txn: Transaction that this job is part of (may be NULL).
951  *
952  * Create a backup operation on @bs.  Clusters in @bs are written to @target
953  * until the job is cancelled or manually completed.
954  */
955 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
956                             BlockDriverState *target, int64_t speed,
957                             MirrorSyncMode sync_mode,
958                             BdrvDirtyBitmap *sync_bitmap,
959                             bool compress,
960                             BlockdevOnError on_source_error,
961                             BlockdevOnError on_target_error,
962                             int creation_flags,
963                             BlockCompletionFunc *cb, void *opaque,
964                             BlockJobTxn *txn, Error **errp);
965 
966 void hmp_drive_add_node(Monitor *mon, const char *optstr);
967 
968 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
969                                   const char *child_name,
970                                   const BdrvChildRole *child_role,
971                                   uint64_t perm, uint64_t shared_perm,
972                                   void *opaque, Error **errp);
973 void bdrv_root_unref_child(BdrvChild *child);
974 
975 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
976                             Error **errp);
977 
978 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
979  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
980  * all children */
981 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
982                                const BdrvChildRole *role,
983                                uint64_t perm, uint64_t shared,
984                                uint64_t *nperm, uint64_t *nshared);
985 
986 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
987  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
988  * requires WRITE | RESIZE for read-write images, always requires
989  * CONSISTENT_READ and doesn't share WRITE. */
990 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
991                                const BdrvChildRole *role,
992                                uint64_t perm, uint64_t shared,
993                                uint64_t *nperm, uint64_t *nshared);
994 
995 const char *bdrv_get_parent_name(const BlockDriverState *bs);
996 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
997 bool blk_dev_has_removable_media(BlockBackend *blk);
998 bool blk_dev_has_tray(BlockBackend *blk);
999 void blk_dev_eject_request(BlockBackend *blk, bool force);
1000 bool blk_dev_is_tray_open(BlockBackend *blk);
1001 bool blk_dev_is_medium_locked(BlockBackend *blk);
1002 
1003 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
1004 bool bdrv_requests_pending(BlockDriverState *bs);
1005 
1006 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
1007 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
1008 
1009 void bdrv_inc_in_flight(BlockDriverState *bs);
1010 void bdrv_dec_in_flight(BlockDriverState *bs);
1011 
1012 void blockdev_close_all_bdrv_states(void);
1013 
1014 #endif /* BLOCK_INT_H */
1015