xref: /openbmc/qemu/include/block/block_int.h (revision 83974cf4)
1 /*
2  * QEMU System Emulator block driver
3  *
4  * Copyright (c) 2003 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 #ifndef BLOCK_INT_H
25 #define BLOCK_INT_H
26 
27 #include "block/accounting.h"
28 #include "block/block.h"
29 #include "qemu/option.h"
30 #include "qemu/queue.h"
31 #include "qemu/coroutine.h"
32 #include "qemu/stats64.h"
33 #include "qemu/timer.h"
34 #include "qapi-types.h"
35 #include "qemu/hbitmap.h"
36 #include "block/snapshot.h"
37 #include "qemu/main-loop.h"
38 #include "qemu/throttle.h"
39 
40 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
41 
42 #define BLOCK_OPT_SIZE              "size"
43 #define BLOCK_OPT_ENCRYPT           "encryption"
44 #define BLOCK_OPT_ENCRYPT_FORMAT    "encrypt.format"
45 #define BLOCK_OPT_COMPAT6           "compat6"
46 #define BLOCK_OPT_HWVERSION         "hwversion"
47 #define BLOCK_OPT_BACKING_FILE      "backing_file"
48 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
49 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
50 #define BLOCK_OPT_TABLE_SIZE        "table_size"
51 #define BLOCK_OPT_PREALLOC          "preallocation"
52 #define BLOCK_OPT_SUBFMT            "subformat"
53 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
54 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
55 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
56 #define BLOCK_OPT_REDUNDANCY        "redundancy"
57 #define BLOCK_OPT_NOCOW             "nocow"
58 #define BLOCK_OPT_OBJECT_SIZE       "object_size"
59 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
60 
61 #define BLOCK_PROBE_BUF_SIZE        512
62 
63 enum BdrvTrackedRequestType {
64     BDRV_TRACKED_READ,
65     BDRV_TRACKED_WRITE,
66     BDRV_TRACKED_DISCARD,
67 };
68 
69 typedef struct BdrvTrackedRequest {
70     BlockDriverState *bs;
71     int64_t offset;
72     unsigned int bytes;
73     enum BdrvTrackedRequestType type;
74 
75     bool serialising;
76     int64_t overlap_offset;
77     unsigned int overlap_bytes;
78 
79     QLIST_ENTRY(BdrvTrackedRequest) list;
80     Coroutine *co; /* owner, used for deadlock detection */
81     CoQueue wait_queue; /* coroutines blocked on this request */
82 
83     struct BdrvTrackedRequest *waiting_for;
84 } BdrvTrackedRequest;
85 
86 struct BlockDriver {
87     const char *format_name;
88     int instance_size;
89 
90     /* set to true if the BlockDriver is a block filter. Block filters pass
91      * certain callbacks that refer to data (see block.c) to their bs->file if
92      * the driver doesn't implement them. Drivers that do not wish to forward
93      * must implement them and return -ENOTSUP.
94      */
95     bool is_filter;
96     /* for snapshots block filter like Quorum can implement the
97      * following recursive callback.
98      * It's purpose is to recurse on the filter children while calling
99      * bdrv_recurse_is_first_non_filter on them.
100      * For a sample implementation look in the future Quorum block filter.
101      */
102     bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs,
103                                              BlockDriverState *candidate);
104 
105     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
106     int (*bdrv_probe_device)(const char *filename);
107 
108     /* Any driver implementing this callback is expected to be able to handle
109      * NULL file names in its .bdrv_open() implementation */
110     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
111     /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have
112      * this field set to true, except ones that are defined only by their
113      * child's bs.
114      * An example of the last type will be the quorum block driver.
115      */
116     bool bdrv_needs_filename;
117 
118     /* Set if a driver can support backing files */
119     bool supports_backing;
120 
121     /* For handling image reopen for split or non-split files */
122     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
123                                BlockReopenQueue *queue, Error **errp);
124     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
125     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
126     void (*bdrv_join_options)(QDict *options, QDict *old_options);
127 
128     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags,
129                      Error **errp);
130     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags,
131                           Error **errp);
132     void (*bdrv_close)(BlockDriverState *bs);
133     int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp);
134     int (*bdrv_make_empty)(BlockDriverState *bs);
135 
136     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
137 
138     /* aio */
139     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
140         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
141         BlockCompletionFunc *cb, void *opaque);
142     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
143         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
144         BlockCompletionFunc *cb, void *opaque);
145     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
146         BlockCompletionFunc *cb, void *opaque);
147     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
148         int64_t offset, int bytes,
149         BlockCompletionFunc *cb, void *opaque);
150 
151     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
152         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
153 
154     /**
155      * @offset: position in bytes to read at
156      * @bytes: number of bytes to read
157      * @qiov: the buffers to fill with read data
158      * @flags: currently unused, always 0
159      *
160      * @offset and @bytes will be a multiple of 'request_alignment',
161      * but the length of individual @qiov elements does not have to
162      * be a multiple.
163      *
164      * @bytes will always equal the total size of @qiov, and will be
165      * no larger than 'max_transfer'.
166      *
167      * The buffer in @qiov may point directly to guest memory.
168      */
169     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
170         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
171     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
172         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
173     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
174         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
175     /**
176      * @offset: position in bytes to write at
177      * @bytes: number of bytes to write
178      * @qiov: the buffers containing data to write
179      * @flags: zero or more bits allowed by 'supported_write_flags'
180      *
181      * @offset and @bytes will be a multiple of 'request_alignment',
182      * but the length of individual @qiov elements does not have to
183      * be a multiple.
184      *
185      * @bytes will always equal the total size of @qiov, and will be
186      * no larger than 'max_transfer'.
187      *
188      * The buffer in @qiov may point directly to guest memory.
189      */
190     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
191         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
192 
193     /*
194      * Efficiently zero a region of the disk image.  Typically an image format
195      * would use a compact metadata representation to implement this.  This
196      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
197      * will be called instead.
198      */
199     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
200         int64_t offset, int bytes, BdrvRequestFlags flags);
201     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
202         int64_t offset, int bytes);
203 
204     /*
205      * Building block for bdrv_block_status[_above]. The driver should
206      * answer only according to the current layer, and should not
207      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
208      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
209      */
210     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
211         int64_t sector_num, int nb_sectors, int *pnum,
212         BlockDriverState **file);
213 
214     /*
215      * Invalidate any cached meta-data.
216      */
217     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
218     int (*bdrv_inactivate)(BlockDriverState *bs);
219 
220     /*
221      * Flushes all data for all layers by calling bdrv_co_flush for underlying
222      * layers, if needed. This function is needed for deterministic
223      * synchronization of the flush finishing callback.
224      */
225     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
226 
227     /*
228      * Flushes all data that was already written to the OS all the way down to
229      * the disk (for example file-posix.c calls fsync()).
230      */
231     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
232 
233     /*
234      * Flushes all internal caches to the OS. The data may still sit in a
235      * writeback cache of the host OS, but it will survive a crash of the qemu
236      * process.
237      */
238     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
239 
240     const char *protocol_name;
241     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset,
242                          PreallocMode prealloc, Error **errp);
243 
244     int64_t (*bdrv_getlength)(BlockDriverState *bs);
245     bool has_variable_length;
246     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
247     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
248                                       Error **errp);
249 
250     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
251         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
252 
253     int (*bdrv_snapshot_create)(BlockDriverState *bs,
254                                 QEMUSnapshotInfo *sn_info);
255     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
256                               const char *snapshot_id);
257     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
258                                 const char *snapshot_id,
259                                 const char *name,
260                                 Error **errp);
261     int (*bdrv_snapshot_list)(BlockDriverState *bs,
262                               QEMUSnapshotInfo **psn_info);
263     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
264                                   const char *snapshot_id,
265                                   const char *name,
266                                   Error **errp);
267     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
268     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
269 
270     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
271                                           QEMUIOVector *qiov,
272                                           int64_t pos);
273     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
274                                           QEMUIOVector *qiov,
275                                           int64_t pos);
276 
277     int (*bdrv_change_backing_file)(BlockDriverState *bs,
278         const char *backing_file, const char *backing_fmt);
279 
280     /* removable device specific */
281     bool (*bdrv_is_inserted)(BlockDriverState *bs);
282     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
283     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
284 
285     /* to control generic scsi devices */
286     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
287         unsigned long int req, void *buf,
288         BlockCompletionFunc *cb, void *opaque);
289     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
290                                       unsigned long int req, void *buf);
291 
292     /* List of options for creating images, terminated by name == NULL */
293     QemuOptsList *create_opts;
294 
295     /*
296      * Returns 0 for completed check, -errno for internal errors.
297      * The check results are stored in result.
298      */
299     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
300         BdrvCheckMode fix);
301 
302     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
303                               BlockDriverAmendStatusCB *status_cb,
304                               void *cb_opaque);
305 
306     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
307 
308     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
309     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
310         const char *tag);
311     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
312         const char *tag);
313     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
314     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
315 
316     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
317 
318     /*
319      * Returns 1 if newly created images are guaranteed to contain only
320      * zeros, 0 otherwise.
321      */
322     int (*bdrv_has_zero_init)(BlockDriverState *bs);
323 
324     /* Remove fd handlers, timers, and other event loop callbacks so the event
325      * loop is no longer in use.  Called with no in-flight requests and in
326      * depth-first traversal order with parents before child nodes.
327      */
328     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
329 
330     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
331      * can be processed again.  Called with no in-flight requests and in
332      * depth-first traversal order with child nodes before parent nodes.
333      */
334     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
335                                     AioContext *new_context);
336 
337     /* io queue for linux-aio */
338     void (*bdrv_io_plug)(BlockDriverState *bs);
339     void (*bdrv_io_unplug)(BlockDriverState *bs);
340 
341     /**
342      * Try to get @bs's logical and physical block size.
343      * On success, store them in @bsz and return zero.
344      * On failure, return negative errno.
345      */
346     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
347     /**
348      * Try to get @bs's geometry (cyls, heads, sectors)
349      * On success, store them in @geo and return 0.
350      * On failure return -errno.
351      * Only drivers that want to override guest geometry implement this
352      * callback; see hd_geometry_guess().
353      */
354     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
355 
356     /**
357      * Drain and stop any internal sources of requests in the driver, and
358      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
359      */
360     void coroutine_fn (*bdrv_co_drain)(BlockDriverState *bs);
361 
362     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
363                            Error **errp);
364     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
365                            Error **errp);
366 
367     /**
368      * Informs the block driver that a permission change is intended. The
369      * driver checks whether the change is permissible and may take other
370      * preparations for the change (e.g. get file system locks). This operation
371      * is always followed either by a call to either .bdrv_set_perm or
372      * .bdrv_abort_perm_update.
373      *
374      * Checks whether the requested set of cumulative permissions in @perm
375      * can be granted for accessing @bs and whether no other users are using
376      * permissions other than those given in @shared (both arguments take
377      * BLK_PERM_* bitmasks).
378      *
379      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
380      * and errp is set to an error describing the conflict.
381      */
382     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
383                            uint64_t shared, Error **errp);
384 
385     /**
386      * Called to inform the driver that the set of cumulative set of used
387      * permissions for @bs has changed to @perm, and the set of sharable
388      * permission to @shared. The driver can use this to propagate changes to
389      * its children (i.e. request permissions only if a parent actually needs
390      * them).
391      *
392      * This function is only invoked after bdrv_check_perm(), so block drivers
393      * may rely on preparations made in their .bdrv_check_perm implementation.
394      */
395     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
396 
397     /*
398      * Called to inform the driver that after a previous bdrv_check_perm()
399      * call, the permission update is not performed and any preparations made
400      * for it (e.g. taken file locks) need to be undone.
401      *
402      * This function can be called even for nodes that never saw a
403      * bdrv_check_perm() call. It is a no-op then.
404      */
405     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
406 
407     /**
408      * Returns in @nperm and @nshared the permissions that the driver for @bs
409      * needs on its child @c, based on the cumulative permissions requested by
410      * the parents in @parent_perm and @parent_shared.
411      *
412      * If @c is NULL, return the permissions for attaching a new child for the
413      * given @role.
414      *
415      * If @reopen_queue is non-NULL, don't return the currently needed
416      * permissions, but those that will be needed after applying the
417      * @reopen_queue.
418      */
419      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
420                              const BdrvChildRole *role,
421                              BlockReopenQueue *reopen_queue,
422                              uint64_t parent_perm, uint64_t parent_shared,
423                              uint64_t *nperm, uint64_t *nshared);
424 
425     /**
426      * Bitmaps should be marked as 'IN_USE' in the image on reopening image
427      * as rw. This handler should realize it. It also should unset readonly
428      * field of BlockDirtyBitmap's in case of success.
429      */
430     int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
431     bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
432                                             const char *name,
433                                             uint32_t granularity,
434                                             Error **errp);
435     void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
436                                                 const char *name,
437                                                 Error **errp);
438 
439     QLIST_ENTRY(BlockDriver) list;
440 };
441 
442 typedef struct BlockLimits {
443     /* Alignment requirement, in bytes, for offset/length of I/O
444      * requests. Must be a power of 2 less than INT_MAX; defaults to
445      * 1 for drivers with modern byte interfaces, and to 512
446      * otherwise. */
447     uint32_t request_alignment;
448 
449     /* Maximum number of bytes that can be discarded at once (since it
450      * is signed, it must be < 2G, if set). Must be multiple of
451      * pdiscard_alignment, but need not be power of 2. May be 0 if no
452      * inherent 32-bit limit */
453     int32_t max_pdiscard;
454 
455     /* Optimal alignment for discard requests in bytes. A power of 2
456      * is best but not mandatory.  Must be a multiple of
457      * bl.request_alignment, and must be less than max_pdiscard if
458      * that is set. May be 0 if bl.request_alignment is good enough */
459     uint32_t pdiscard_alignment;
460 
461     /* Maximum number of bytes that can zeroized at once (since it is
462      * signed, it must be < 2G, if set). Must be multiple of
463      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
464     int32_t max_pwrite_zeroes;
465 
466     /* Optimal alignment for write zeroes requests in bytes. A power
467      * of 2 is best but not mandatory.  Must be a multiple of
468      * bl.request_alignment, and must be less than max_pwrite_zeroes
469      * if that is set. May be 0 if bl.request_alignment is good
470      * enough */
471     uint32_t pwrite_zeroes_alignment;
472 
473     /* Optimal transfer length in bytes.  A power of 2 is best but not
474      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
475      * no preferred size */
476     uint32_t opt_transfer;
477 
478     /* Maximal transfer length in bytes.  Need not be power of 2, but
479      * must be multiple of opt_transfer and bl.request_alignment, or 0
480      * for no 32-bit limit.  For now, anything larger than INT_MAX is
481      * clamped down. */
482     uint32_t max_transfer;
483 
484     /* memory alignment, in bytes so that no bounce buffer is needed */
485     size_t min_mem_alignment;
486 
487     /* memory alignment, in bytes, for bounce buffer */
488     size_t opt_mem_alignment;
489 
490     /* maximum number of iovec elements */
491     int max_iov;
492 } BlockLimits;
493 
494 typedef struct BdrvOpBlocker BdrvOpBlocker;
495 
496 typedef struct BdrvAioNotifier {
497     void (*attached_aio_context)(AioContext *new_context, void *opaque);
498     void (*detach_aio_context)(void *opaque);
499 
500     void *opaque;
501     bool deleted;
502 
503     QLIST_ENTRY(BdrvAioNotifier) list;
504 } BdrvAioNotifier;
505 
506 struct BdrvChildRole {
507     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
508      * points to. */
509     bool stay_at_node;
510 
511     void (*inherit_options)(int *child_flags, QDict *child_options,
512                             int parent_flags, QDict *parent_options);
513 
514     void (*change_media)(BdrvChild *child, bool load);
515     void (*resize)(BdrvChild *child);
516 
517     /* Returns a name that is supposedly more useful for human users than the
518      * node name for identifying the node in question (in particular, a BB
519      * name), or NULL if the parent can't provide a better name. */
520     const char *(*get_name)(BdrvChild *child);
521 
522     /* Returns a malloced string that describes the parent of the child for a
523      * human reader. This could be a node-name, BlockBackend name, qdev ID or
524      * QOM path of the device owning the BlockBackend, job type and ID etc. The
525      * caller is responsible for freeing the memory. */
526     char *(*get_parent_desc)(BdrvChild *child);
527 
528     /*
529      * If this pair of functions is implemented, the parent doesn't issue new
530      * requests after returning from .drained_begin() until .drained_end() is
531      * called.
532      *
533      * Note that this can be nested. If drained_begin() was called twice, new
534      * I/O is allowed only after drained_end() was called twice, too.
535      */
536     void (*drained_begin)(BdrvChild *child);
537     void (*drained_end)(BdrvChild *child);
538 
539     /* Notifies the parent that the child has been activated/inactivated (e.g.
540      * when migration is completing) and it can start/stop requesting
541      * permissions and doing I/O on it. */
542     void (*activate)(BdrvChild *child, Error **errp);
543     int (*inactivate)(BdrvChild *child);
544 
545     void (*attach)(BdrvChild *child);
546     void (*detach)(BdrvChild *child);
547 
548     /* Notifies the parent that the filename of its child has changed (e.g.
549      * because the direct child was removed from the backing chain), so that it
550      * can update its reference. */
551     int (*update_filename)(BdrvChild *child, BlockDriverState *new_base,
552                            const char *filename, Error **errp);
553 };
554 
555 extern const BdrvChildRole child_file;
556 extern const BdrvChildRole child_format;
557 extern const BdrvChildRole child_backing;
558 
559 struct BdrvChild {
560     BlockDriverState *bs;
561     char *name;
562     const BdrvChildRole *role;
563     void *opaque;
564 
565     /**
566      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
567      */
568     uint64_t perm;
569 
570     /**
571      * Permissions that can still be granted to other users of @bs while this
572      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
573      */
574     uint64_t shared_perm;
575 
576     QLIST_ENTRY(BdrvChild) next;
577     QLIST_ENTRY(BdrvChild) next_parent;
578 };
579 
580 /*
581  * Note: the function bdrv_append() copies and swaps contents of
582  * BlockDriverStates, so if you add new fields to this struct, please
583  * inspect bdrv_append() to determine if the new fields need to be
584  * copied as well.
585  */
586 struct BlockDriverState {
587     /* Protected by big QEMU lock or read-only after opening.  No special
588      * locking needed during I/O...
589      */
590     int open_flags; /* flags used to open the file, re-used for re-open */
591     bool read_only; /* if true, the media is read only */
592     bool encrypted; /* if true, the media is encrypted */
593     bool sg;        /* if true, the device is a /dev/sg* */
594     bool probed;    /* if true, format was probed rather than specified */
595     bool force_share; /* if true, always allow all shared permissions */
596     bool implicit;  /* if true, this filter node was automatically inserted */
597 
598     BlockDriver *drv; /* NULL means no media */
599     void *opaque;
600 
601     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
602     /* long-running tasks intended to always use the same AioContext as this
603      * BDS may register themselves in this list to be notified of changes
604      * regarding this BDS's context */
605     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
606     bool walking_aio_notifiers; /* to make removal during iteration safe */
607 
608     char filename[PATH_MAX];
609     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
610                                     this file image */
611     char backing_format[16]; /* if non-zero and backing_file exists */
612 
613     QDict *full_open_options;
614     char exact_filename[PATH_MAX];
615 
616     BdrvChild *backing;
617     BdrvChild *file;
618 
619     /* I/O Limits */
620     BlockLimits bl;
621 
622     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
623     unsigned int supported_write_flags;
624     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
625      * BDRV_REQ_MAY_UNMAP) */
626     unsigned int supported_zero_flags;
627 
628     /* the following member gives a name to every node on the bs graph. */
629     char node_name[32];
630     /* element of the list of named nodes building the graph */
631     QTAILQ_ENTRY(BlockDriverState) node_list;
632     /* element of the list of all BlockDriverStates (all_bdrv_states) */
633     QTAILQ_ENTRY(BlockDriverState) bs_list;
634     /* element of the list of monitor-owned BDS */
635     QTAILQ_ENTRY(BlockDriverState) monitor_list;
636     int refcnt;
637 
638     /* operation blockers */
639     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
640 
641     /* long-running background operation */
642     BlockJob *job;
643 
644     /* The node that this node inherited default options from (and a reopen on
645      * which can affect this node by changing these defaults). This is always a
646      * parent node of this node. */
647     BlockDriverState *inherits_from;
648     QLIST_HEAD(, BdrvChild) children;
649     QLIST_HEAD(, BdrvChild) parents;
650 
651     QDict *options;
652     QDict *explicit_options;
653     BlockdevDetectZeroesOptions detect_zeroes;
654 
655     /* The error object in use for blocking operations on backing_hd */
656     Error *backing_blocker;
657 
658     /* Protected by AioContext lock */
659 
660     /* If we are reading a disk image, give its size in sectors.
661      * Generally read-only; it is written to by load_snapshot and
662      * save_snaphost, but the block layer is quiescent during those.
663      */
664     int64_t total_sectors;
665 
666     /* Callback before write request is processed */
667     NotifierWithReturnList before_write_notifiers;
668 
669     /* threshold limit for writes, in bytes. "High water mark". */
670     uint64_t write_threshold_offset;
671     NotifierWithReturn write_threshold_notifier;
672 
673     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
674      * Reading from the list can be done with either the BQL or the
675      * dirty_bitmap_mutex.  Modifying a bitmap only requires
676      * dirty_bitmap_mutex.  */
677     QemuMutex dirty_bitmap_mutex;
678     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
679 
680     /* Offset after the highest byte written to */
681     Stat64 wr_highest_offset;
682 
683     /* If true, copy read backing sectors into image.  Can be >1 if more
684      * than one client has requested copy-on-read.  Accessed with atomic
685      * ops.
686      */
687     int copy_on_read;
688 
689     /* number of in-flight requests; overall and serialising.
690      * Accessed with atomic ops.
691      */
692     unsigned int in_flight;
693     unsigned int serialising_in_flight;
694 
695     /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
696      * ops.
697      */
698     bool wakeup;
699 
700     /* counter for nested bdrv_io_plug.
701      * Accessed with atomic ops.
702     */
703     unsigned io_plugged;
704 
705     /* do we need to tell the quest if we have a volatile write cache? */
706     int enable_write_cache;
707 
708     /* Accessed with atomic ops.  */
709     int quiesce_counter;
710     unsigned int write_gen;               /* Current data generation */
711 
712     /* Protected by reqs_lock.  */
713     CoMutex reqs_lock;
714     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
715     CoQueue flush_queue;                  /* Serializing flush queue */
716     bool active_flush_req;                /* Flush request in flight? */
717 
718     /* Only read/written by whoever has set active_flush_req to true.  */
719     unsigned int flushed_gen;             /* Flushed write generation */
720 };
721 
722 struct BlockBackendRootState {
723     int open_flags;
724     bool read_only;
725     BlockdevDetectZeroesOptions detect_zeroes;
726 };
727 
728 typedef enum BlockMirrorBackingMode {
729     /* Reuse the existing backing chain from the source for the target.
730      * - sync=full: Set backing BDS to NULL.
731      * - sync=top:  Use source's backing BDS.
732      * - sync=none: Use source as the backing BDS. */
733     MIRROR_SOURCE_BACKING_CHAIN,
734 
735     /* Open the target's backing chain completely anew */
736     MIRROR_OPEN_BACKING_CHAIN,
737 
738     /* Do not change the target's backing BDS after job completion */
739     MIRROR_LEAVE_BACKING_CHAIN,
740 } BlockMirrorBackingMode;
741 
742 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
743 {
744     return bs->backing ? bs->backing->bs : NULL;
745 }
746 
747 
748 /* Essential block drivers which must always be statically linked into qemu, and
749  * which therefore can be accessed without using bdrv_find_format() */
750 extern BlockDriver bdrv_file;
751 extern BlockDriver bdrv_raw;
752 extern BlockDriver bdrv_qcow2;
753 
754 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
755     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
756     BdrvRequestFlags flags);
757 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
758     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
759     BdrvRequestFlags flags);
760 
761 int get_tmp_filename(char *filename, int size);
762 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
763                             const char *filename);
764 
765 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
766                                       QDict *options);
767 
768 
769 /**
770  * bdrv_add_before_write_notifier:
771  *
772  * Register a callback that is invoked before write requests are processed but
773  * after any throttling or waiting for overlapping requests.
774  */
775 void bdrv_add_before_write_notifier(BlockDriverState *bs,
776                                     NotifierWithReturn *notifier);
777 
778 /**
779  * bdrv_detach_aio_context:
780  *
781  * May be called from .bdrv_detach_aio_context() to detach children from the
782  * current #AioContext.  This is only needed by block drivers that manage their
783  * own children.  Both ->file and ->backing are automatically handled and
784  * block drivers should not call this function on them explicitly.
785  */
786 void bdrv_detach_aio_context(BlockDriverState *bs);
787 
788 /**
789  * bdrv_attach_aio_context:
790  *
791  * May be called from .bdrv_attach_aio_context() to attach children to the new
792  * #AioContext.  This is only needed by block drivers that manage their own
793  * children.  Both ->file and ->backing are automatically handled and block
794  * drivers should not call this function on them explicitly.
795  */
796 void bdrv_attach_aio_context(BlockDriverState *bs,
797                              AioContext *new_context);
798 
799 /**
800  * bdrv_add_aio_context_notifier:
801  *
802  * If a long-running job intends to be always run in the same AioContext as a
803  * certain BDS, it may use this function to be notified of changes regarding the
804  * association of the BDS to an AioContext.
805  *
806  * attached_aio_context() is called after the target BDS has been attached to a
807  * new AioContext; detach_aio_context() is called before the target BDS is being
808  * detached from its old AioContext.
809  */
810 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
811         void (*attached_aio_context)(AioContext *new_context, void *opaque),
812         void (*detach_aio_context)(void *opaque), void *opaque);
813 
814 /**
815  * bdrv_remove_aio_context_notifier:
816  *
817  * Unsubscribe of change notifications regarding the BDS's AioContext. The
818  * parameters given here have to be the same as those given to
819  * bdrv_add_aio_context_notifier().
820  */
821 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
822                                       void (*aio_context_attached)(AioContext *,
823                                                                    void *),
824                                       void (*aio_context_detached)(void *),
825                                       void *opaque);
826 
827 /**
828  * bdrv_wakeup:
829  * @bs: The BlockDriverState for which an I/O operation has been completed.
830  *
831  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
832  * synchronous I/O on a BlockDriverState that is attached to another
833  * I/O thread, the main thread lets the I/O thread's event loop run,
834  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
835  * up the main thread if necessary.
836  *
837  * Manual calls to bdrv_wakeup are rarely necessary, because
838  * bdrv_dec_in_flight already calls it.
839  */
840 void bdrv_wakeup(BlockDriverState *bs);
841 
842 #ifdef _WIN32
843 int is_windows_drive(const char *filename);
844 #endif
845 
846 /**
847  * stream_start:
848  * @job_id: The id of the newly-created job, or %NULL to use the
849  * device name of @bs.
850  * @bs: Block device to operate on.
851  * @base: Block device that will become the new base, or %NULL to
852  * flatten the whole backing file chain onto @bs.
853  * @backing_file_str: The file name that will be written to @bs as the
854  * the new backing file if the job completes. Ignored if @base is %NULL.
855  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
856  * @on_error: The action to take upon error.
857  * @errp: Error object.
858  *
859  * Start a streaming operation on @bs.  Clusters that are unallocated
860  * in @bs, but allocated in any image between @base and @bs (both
861  * exclusive) will be written to @bs.  At the end of a successful
862  * streaming job, the backing file of @bs will be changed to
863  * @backing_file_str in the written image and to @base in the live
864  * BlockDriverState.
865  */
866 void stream_start(const char *job_id, BlockDriverState *bs,
867                   BlockDriverState *base, const char *backing_file_str,
868                   int64_t speed, BlockdevOnError on_error, Error **errp);
869 
870 /**
871  * commit_start:
872  * @job_id: The id of the newly-created job, or %NULL to use the
873  * device name of @bs.
874  * @bs: Active block device.
875  * @top: Top block device to be committed.
876  * @base: Block device that will be written into, and become the new top.
877  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
878  * @on_error: The action to take upon error.
879  * @backing_file_str: String to use as the backing file in @top's overlay
880  * @filter_node_name: The node name that should be assigned to the filter
881  * driver that the commit job inserts into the graph above @top. NULL means
882  * that a node name should be autogenerated.
883  * @errp: Error object.
884  *
885  */
886 void commit_start(const char *job_id, BlockDriverState *bs,
887                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
888                   BlockdevOnError on_error, const char *backing_file_str,
889                   const char *filter_node_name, Error **errp);
890 /**
891  * commit_active_start:
892  * @job_id: The id of the newly-created job, or %NULL to use the
893  * device name of @bs.
894  * @bs: Active block device to be committed.
895  * @base: Block device that will be written into, and become the new top.
896  * @creation_flags: Flags that control the behavior of the Job lifetime.
897  *                  See @BlockJobCreateFlags
898  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
899  * @on_error: The action to take upon error.
900  * @filter_node_name: The node name that should be assigned to the filter
901  * driver that the commit job inserts into the graph above @bs. NULL means that
902  * a node name should be autogenerated.
903  * @cb: Completion function for the job.
904  * @opaque: Opaque pointer value passed to @cb.
905  * @auto_complete: Auto complete the job.
906  * @errp: Error object.
907  *
908  */
909 void commit_active_start(const char *job_id, BlockDriverState *bs,
910                          BlockDriverState *base, int creation_flags,
911                          int64_t speed, BlockdevOnError on_error,
912                          const char *filter_node_name,
913                          BlockCompletionFunc *cb, void *opaque,
914                          bool auto_complete, Error **errp);
915 /*
916  * mirror_start:
917  * @job_id: The id of the newly-created job, or %NULL to use the
918  * device name of @bs.
919  * @bs: Block device to operate on.
920  * @target: Block device to write to.
921  * @replaces: Block graph node name to replace once the mirror is done. Can
922  *            only be used when full mirroring is selected.
923  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
924  * @granularity: The chosen granularity for the dirty bitmap.
925  * @buf_size: The amount of data that can be in flight at one time.
926  * @mode: Whether to collapse all images in the chain to the target.
927  * @backing_mode: How to establish the target's backing chain after completion.
928  * @on_source_error: The action to take upon error reading from the source.
929  * @on_target_error: The action to take upon error writing to the target.
930  * @unmap: Whether to unmap target where source sectors only contain zeroes.
931  * @filter_node_name: The node name that should be assigned to the filter
932  * driver that the mirror job inserts into the graph above @bs. NULL means that
933  * a node name should be autogenerated.
934  * @errp: Error object.
935  *
936  * Start a mirroring operation on @bs.  Clusters that are allocated
937  * in @bs will be written to @target until the job is cancelled or
938  * manually completed.  At the end of a successful mirroring job,
939  * @bs will be switched to read from @target.
940  */
941 void mirror_start(const char *job_id, BlockDriverState *bs,
942                   BlockDriverState *target, const char *replaces,
943                   int64_t speed, uint32_t granularity, int64_t buf_size,
944                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
945                   BlockdevOnError on_source_error,
946                   BlockdevOnError on_target_error,
947                   bool unmap, const char *filter_node_name, Error **errp);
948 
949 /*
950  * backup_job_create:
951  * @job_id: The id of the newly-created job, or %NULL to use the
952  * device name of @bs.
953  * @bs: Block device to operate on.
954  * @target: Block device to write to.
955  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
956  * @sync_mode: What parts of the disk image should be copied to the destination.
957  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
958  * @on_source_error: The action to take upon error reading from the source.
959  * @on_target_error: The action to take upon error writing to the target.
960  * @creation_flags: Flags that control the behavior of the Job lifetime.
961  *                  See @BlockJobCreateFlags
962  * @cb: Completion function for the job.
963  * @opaque: Opaque pointer value passed to @cb.
964  * @txn: Transaction that this job is part of (may be NULL).
965  *
966  * Create a backup operation on @bs.  Clusters in @bs are written to @target
967  * until the job is cancelled or manually completed.
968  */
969 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
970                             BlockDriverState *target, int64_t speed,
971                             MirrorSyncMode sync_mode,
972                             BdrvDirtyBitmap *sync_bitmap,
973                             bool compress,
974                             BlockdevOnError on_source_error,
975                             BlockdevOnError on_target_error,
976                             int creation_flags,
977                             BlockCompletionFunc *cb, void *opaque,
978                             BlockJobTxn *txn, Error **errp);
979 
980 void hmp_drive_add_node(Monitor *mon, const char *optstr);
981 
982 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
983                                   const char *child_name,
984                                   const BdrvChildRole *child_role,
985                                   uint64_t perm, uint64_t shared_perm,
986                                   void *opaque, Error **errp);
987 void bdrv_root_unref_child(BdrvChild *child);
988 
989 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
990                             Error **errp);
991 
992 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
993  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
994  * all children */
995 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
996                                const BdrvChildRole *role,
997                                BlockReopenQueue *reopen_queue,
998                                uint64_t perm, uint64_t shared,
999                                uint64_t *nperm, uint64_t *nshared);
1000 
1001 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
1002  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
1003  * requires WRITE | RESIZE for read-write images, always requires
1004  * CONSISTENT_READ and doesn't share WRITE. */
1005 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
1006                                const BdrvChildRole *role,
1007                                BlockReopenQueue *reopen_queue,
1008                                uint64_t perm, uint64_t shared,
1009                                uint64_t *nperm, uint64_t *nshared);
1010 
1011 /*
1012  * Default implementation for drivers to pass bdrv_co_get_block_status() to
1013  * their file.
1014  */
1015 int64_t coroutine_fn bdrv_co_get_block_status_from_file(BlockDriverState *bs,
1016                                                         int64_t sector_num,
1017                                                         int nb_sectors,
1018                                                         int *pnum,
1019                                                         BlockDriverState **file);
1020 /*
1021  * Default implementation for drivers to pass bdrv_co_get_block_status() to
1022  * their backing file.
1023  */
1024 int64_t coroutine_fn bdrv_co_get_block_status_from_backing(BlockDriverState *bs,
1025                                                            int64_t sector_num,
1026                                                            int nb_sectors,
1027                                                            int *pnum,
1028                                                            BlockDriverState **file);
1029 const char *bdrv_get_parent_name(const BlockDriverState *bs);
1030 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
1031 bool blk_dev_has_removable_media(BlockBackend *blk);
1032 bool blk_dev_has_tray(BlockBackend *blk);
1033 void blk_dev_eject_request(BlockBackend *blk, bool force);
1034 bool blk_dev_is_tray_open(BlockBackend *blk);
1035 bool blk_dev_is_medium_locked(BlockBackend *blk);
1036 
1037 void bdrv_set_dirty(BlockDriverState *bs, int64_t offset, int64_t bytes);
1038 bool bdrv_requests_pending(BlockDriverState *bs);
1039 
1040 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
1041 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
1042 
1043 void bdrv_inc_in_flight(BlockDriverState *bs);
1044 void bdrv_dec_in_flight(BlockDriverState *bs);
1045 
1046 void blockdev_close_all_bdrv_states(void);
1047 
1048 #endif /* BLOCK_INT_H */
1049