xref: /openbmc/qemu/include/block/block_int.h (revision 87e0331c)
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_set_key)(BlockDriverState *bs, const char *key);
131     int (*bdrv_make_empty)(BlockDriverState *bs);
132 
133     void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options);
134 
135     /* aio */
136     BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
137         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
138         BlockCompletionFunc *cb, void *opaque);
139     BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
140         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
141         BlockCompletionFunc *cb, void *opaque);
142     BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
143         BlockCompletionFunc *cb, void *opaque);
144     BlockAIOCB *(*bdrv_aio_pdiscard)(BlockDriverState *bs,
145         int64_t offset, int bytes,
146         BlockCompletionFunc *cb, void *opaque);
147 
148     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
149         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
150     int coroutine_fn (*bdrv_co_preadv)(BlockDriverState *bs,
151         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
152     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
153         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
154     int coroutine_fn (*bdrv_co_writev_flags)(BlockDriverState *bs,
155         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, int flags);
156     int coroutine_fn (*bdrv_co_pwritev)(BlockDriverState *bs,
157         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov, int flags);
158 
159     /*
160      * Efficiently zero a region of the disk image.  Typically an image format
161      * would use a compact metadata representation to implement this.  This
162      * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev()
163      * will be called instead.
164      */
165     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
166         int64_t offset, int bytes, BdrvRequestFlags flags);
167     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
168         int64_t offset, int bytes);
169 
170     /*
171      * Building block for bdrv_block_status[_above]. The driver should
172      * answer only according to the current layer, and should not
173      * set BDRV_BLOCK_ALLOCATED, but may set BDRV_BLOCK_RAW.  See block.h
174      * for the meaning of _DATA, _ZERO, and _OFFSET_VALID.
175      */
176     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
177         int64_t sector_num, int nb_sectors, int *pnum,
178         BlockDriverState **file);
179 
180     /*
181      * Invalidate any cached meta-data.
182      */
183     void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp);
184     int (*bdrv_inactivate)(BlockDriverState *bs);
185 
186     /*
187      * Flushes all data for all layers by calling bdrv_co_flush for underlying
188      * layers, if needed. This function is needed for deterministic
189      * synchronization of the flush finishing callback.
190      */
191     int coroutine_fn (*bdrv_co_flush)(BlockDriverState *bs);
192 
193     /*
194      * Flushes all data that was already written to the OS all the way down to
195      * the disk (for example file-posix.c calls fsync()).
196      */
197     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
198 
199     /*
200      * Flushes all internal caches to the OS. The data may still sit in a
201      * writeback cache of the host OS, but it will survive a crash of the qemu
202      * process.
203      */
204     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
205 
206     const char *protocol_name;
207     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset,
208                          PreallocMode prealloc, Error **errp);
209 
210     int64_t (*bdrv_getlength)(BlockDriverState *bs);
211     bool has_variable_length;
212     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
213     BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs,
214                                       Error **errp);
215 
216     int coroutine_fn (*bdrv_co_pwritev_compressed)(BlockDriverState *bs,
217         uint64_t offset, uint64_t bytes, QEMUIOVector *qiov);
218 
219     int (*bdrv_snapshot_create)(BlockDriverState *bs,
220                                 QEMUSnapshotInfo *sn_info);
221     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
222                               const char *snapshot_id);
223     int (*bdrv_snapshot_delete)(BlockDriverState *bs,
224                                 const char *snapshot_id,
225                                 const char *name,
226                                 Error **errp);
227     int (*bdrv_snapshot_list)(BlockDriverState *bs,
228                               QEMUSnapshotInfo **psn_info);
229     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
230                                   const char *snapshot_id,
231                                   const char *name,
232                                   Error **errp);
233     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
234     ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs);
235 
236     int coroutine_fn (*bdrv_save_vmstate)(BlockDriverState *bs,
237                                           QEMUIOVector *qiov,
238                                           int64_t pos);
239     int coroutine_fn (*bdrv_load_vmstate)(BlockDriverState *bs,
240                                           QEMUIOVector *qiov,
241                                           int64_t pos);
242 
243     int (*bdrv_change_backing_file)(BlockDriverState *bs,
244         const char *backing_file, const char *backing_fmt);
245 
246     /* removable device specific */
247     bool (*bdrv_is_inserted)(BlockDriverState *bs);
248     int (*bdrv_media_changed)(BlockDriverState *bs);
249     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
250     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
251 
252     /* to control generic scsi devices */
253     BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
254         unsigned long int req, void *buf,
255         BlockCompletionFunc *cb, void *opaque);
256     int coroutine_fn (*bdrv_co_ioctl)(BlockDriverState *bs,
257                                       unsigned long int req, void *buf);
258 
259     /* List of options for creating images, terminated by name == NULL */
260     QemuOptsList *create_opts;
261 
262     /*
263      * Returns 0 for completed check, -errno for internal errors.
264      * The check results are stored in result.
265      */
266     int (*bdrv_check)(BlockDriverState *bs, BdrvCheckResult *result,
267         BdrvCheckMode fix);
268 
269     int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts,
270                               BlockDriverAmendStatusCB *status_cb,
271                               void *cb_opaque);
272 
273     void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event);
274 
275     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
276     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
277         const char *tag);
278     int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs,
279         const char *tag);
280     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
281     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
282 
283     void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp);
284 
285     /*
286      * Returns 1 if newly created images are guaranteed to contain only
287      * zeros, 0 otherwise.
288      */
289     int (*bdrv_has_zero_init)(BlockDriverState *bs);
290 
291     /* Remove fd handlers, timers, and other event loop callbacks so the event
292      * loop is no longer in use.  Called with no in-flight requests and in
293      * depth-first traversal order with parents before child nodes.
294      */
295     void (*bdrv_detach_aio_context)(BlockDriverState *bs);
296 
297     /* Add fd handlers, timers, and other event loop callbacks so I/O requests
298      * can be processed again.  Called with no in-flight requests and in
299      * depth-first traversal order with child nodes before parent nodes.
300      */
301     void (*bdrv_attach_aio_context)(BlockDriverState *bs,
302                                     AioContext *new_context);
303 
304     /* io queue for linux-aio */
305     void (*bdrv_io_plug)(BlockDriverState *bs);
306     void (*bdrv_io_unplug)(BlockDriverState *bs);
307 
308     /**
309      * Try to get @bs's logical and physical block size.
310      * On success, store them in @bsz and return zero.
311      * On failure, return negative errno.
312      */
313     int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz);
314     /**
315      * Try to get @bs's geometry (cyls, heads, sectors)
316      * On success, store them in @geo and return 0.
317      * On failure return -errno.
318      * Only drivers that want to override guest geometry implement this
319      * callback; see hd_geometry_guess().
320      */
321     int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo);
322 
323     /**
324      * Drain and stop any internal sources of requests in the driver, and
325      * remain so until next I/O callback (e.g. bdrv_co_writev) is called.
326      */
327     void coroutine_fn (*bdrv_co_drain)(BlockDriverState *bs);
328 
329     void (*bdrv_add_child)(BlockDriverState *parent, BlockDriverState *child,
330                            Error **errp);
331     void (*bdrv_del_child)(BlockDriverState *parent, BdrvChild *child,
332                            Error **errp);
333 
334     /**
335      * Informs the block driver that a permission change is intended. The
336      * driver checks whether the change is permissible and may take other
337      * preparations for the change (e.g. get file system locks). This operation
338      * is always followed either by a call to either .bdrv_set_perm or
339      * .bdrv_abort_perm_update.
340      *
341      * Checks whether the requested set of cumulative permissions in @perm
342      * can be granted for accessing @bs and whether no other users are using
343      * permissions other than those given in @shared (both arguments take
344      * BLK_PERM_* bitmasks).
345      *
346      * If both conditions are met, 0 is returned. Otherwise, -errno is returned
347      * and errp is set to an error describing the conflict.
348      */
349     int (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm,
350                            uint64_t shared, Error **errp);
351 
352     /**
353      * Called to inform the driver that the set of cumulative set of used
354      * permissions for @bs has changed to @perm, and the set of sharable
355      * permission to @shared. The driver can use this to propagate changes to
356      * its children (i.e. request permissions only if a parent actually needs
357      * them).
358      *
359      * This function is only invoked after bdrv_check_perm(), so block drivers
360      * may rely on preparations made in their .bdrv_check_perm implementation.
361      */
362     void (*bdrv_set_perm)(BlockDriverState *bs, uint64_t perm, uint64_t shared);
363 
364     /*
365      * Called to inform the driver that after a previous bdrv_check_perm()
366      * call, the permission update is not performed and any preparations made
367      * for it (e.g. taken file locks) need to be undone.
368      *
369      * This function can be called even for nodes that never saw a
370      * bdrv_check_perm() call. It is a no-op then.
371      */
372     void (*bdrv_abort_perm_update)(BlockDriverState *bs);
373 
374     /**
375      * Returns in @nperm and @nshared the permissions that the driver for @bs
376      * needs on its child @c, based on the cumulative permissions requested by
377      * the parents in @parent_perm and @parent_shared.
378      *
379      * If @c is NULL, return the permissions for attaching a new child for the
380      * given @role.
381      */
382      void (*bdrv_child_perm)(BlockDriverState *bs, BdrvChild *c,
383                              const BdrvChildRole *role,
384                              uint64_t parent_perm, uint64_t parent_shared,
385                              uint64_t *nperm, uint64_t *nshared);
386 
387     /**
388      * Bitmaps should be marked as 'IN_USE' in the image on reopening image
389      * as rw. This handler should realize it. It also should unset readonly
390      * field of BlockDirtyBitmap's in case of success.
391      */
392     int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
393     bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
394                                             const char *name,
395                                             uint32_t granularity,
396                                             Error **errp);
397     void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
398                                                 const char *name,
399                                                 Error **errp);
400 
401     QLIST_ENTRY(BlockDriver) list;
402 };
403 
404 typedef struct BlockLimits {
405     /* Alignment requirement, in bytes, for offset/length of I/O
406      * requests. Must be a power of 2 less than INT_MAX; defaults to
407      * 1 for drivers with modern byte interfaces, and to 512
408      * otherwise. */
409     uint32_t request_alignment;
410 
411     /* Maximum number of bytes that can be discarded at once (since it
412      * is signed, it must be < 2G, if set). Must be multiple of
413      * pdiscard_alignment, but need not be power of 2. May be 0 if no
414      * inherent 32-bit limit */
415     int32_t max_pdiscard;
416 
417     /* Optimal alignment for discard requests in bytes. A power of 2
418      * is best but not mandatory.  Must be a multiple of
419      * bl.request_alignment, and must be less than max_pdiscard if
420      * that is set. May be 0 if bl.request_alignment is good enough */
421     uint32_t pdiscard_alignment;
422 
423     /* Maximum number of bytes that can zeroized at once (since it is
424      * signed, it must be < 2G, if set). Must be multiple of
425      * pwrite_zeroes_alignment. May be 0 if no inherent 32-bit limit */
426     int32_t max_pwrite_zeroes;
427 
428     /* Optimal alignment for write zeroes requests in bytes. A power
429      * of 2 is best but not mandatory.  Must be a multiple of
430      * bl.request_alignment, and must be less than max_pwrite_zeroes
431      * if that is set. May be 0 if bl.request_alignment is good
432      * enough */
433     uint32_t pwrite_zeroes_alignment;
434 
435     /* Optimal transfer length in bytes.  A power of 2 is best but not
436      * mandatory.  Must be a multiple of bl.request_alignment, or 0 if
437      * no preferred size */
438     uint32_t opt_transfer;
439 
440     /* Maximal transfer length in bytes.  Need not be power of 2, but
441      * must be multiple of opt_transfer and bl.request_alignment, or 0
442      * for no 32-bit limit.  For now, anything larger than INT_MAX is
443      * clamped down. */
444     uint32_t max_transfer;
445 
446     /* memory alignment, in bytes so that no bounce buffer is needed */
447     size_t min_mem_alignment;
448 
449     /* memory alignment, in bytes, for bounce buffer */
450     size_t opt_mem_alignment;
451 
452     /* maximum number of iovec elements */
453     int max_iov;
454 } BlockLimits;
455 
456 typedef struct BdrvOpBlocker BdrvOpBlocker;
457 
458 typedef struct BdrvAioNotifier {
459     void (*attached_aio_context)(AioContext *new_context, void *opaque);
460     void (*detach_aio_context)(void *opaque);
461 
462     void *opaque;
463     bool deleted;
464 
465     QLIST_ENTRY(BdrvAioNotifier) list;
466 } BdrvAioNotifier;
467 
468 struct BdrvChildRole {
469     /* If true, bdrv_replace_node() doesn't change the node this BdrvChild
470      * points to. */
471     bool stay_at_node;
472 
473     void (*inherit_options)(int *child_flags, QDict *child_options,
474                             int parent_flags, QDict *parent_options);
475 
476     void (*change_media)(BdrvChild *child, bool load);
477     void (*resize)(BdrvChild *child);
478 
479     /* Returns a name that is supposedly more useful for human users than the
480      * node name for identifying the node in question (in particular, a BB
481      * name), or NULL if the parent can't provide a better name. */
482     const char *(*get_name)(BdrvChild *child);
483 
484     /* Returns a malloced string that describes the parent of the child for a
485      * human reader. This could be a node-name, BlockBackend name, qdev ID or
486      * QOM path of the device owning the BlockBackend, job type and ID etc. The
487      * caller is responsible for freeing the memory. */
488     char *(*get_parent_desc)(BdrvChild *child);
489 
490     /*
491      * If this pair of functions is implemented, the parent doesn't issue new
492      * requests after returning from .drained_begin() until .drained_end() is
493      * called.
494      *
495      * Note that this can be nested. If drained_begin() was called twice, new
496      * I/O is allowed only after drained_end() was called twice, too.
497      */
498     void (*drained_begin)(BdrvChild *child);
499     void (*drained_end)(BdrvChild *child);
500 
501     /* Notifies the parent that the child has been activated/inactivated (e.g.
502      * when migration is completing) and it can start/stop requesting
503      * permissions and doing I/O on it. */
504     void (*activate)(BdrvChild *child, Error **errp);
505     int (*inactivate)(BdrvChild *child);
506 
507     void (*attach)(BdrvChild *child);
508     void (*detach)(BdrvChild *child);
509 };
510 
511 extern const BdrvChildRole child_file;
512 extern const BdrvChildRole child_format;
513 extern const BdrvChildRole child_backing;
514 
515 struct BdrvChild {
516     BlockDriverState *bs;
517     char *name;
518     const BdrvChildRole *role;
519     void *opaque;
520 
521     /**
522      * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask)
523      */
524     uint64_t perm;
525 
526     /**
527      * Permissions that can still be granted to other users of @bs while this
528      * BdrvChild is still attached to it. (BLK_PERM_* bitmask)
529      */
530     uint64_t shared_perm;
531 
532     QLIST_ENTRY(BdrvChild) next;
533     QLIST_ENTRY(BdrvChild) next_parent;
534 };
535 
536 /*
537  * Note: the function bdrv_append() copies and swaps contents of
538  * BlockDriverStates, so if you add new fields to this struct, please
539  * inspect bdrv_append() to determine if the new fields need to be
540  * copied as well.
541  */
542 struct BlockDriverState {
543     /* Protected by big QEMU lock or read-only after opening.  No special
544      * locking needed during I/O...
545      */
546     int open_flags; /* flags used to open the file, re-used for re-open */
547     bool read_only; /* if true, the media is read only */
548     bool encrypted; /* if true, the media is encrypted */
549     bool sg;        /* if true, the device is a /dev/sg* */
550     bool probed;    /* if true, format was probed rather than specified */
551     bool force_share; /* if true, always allow all shared permissions */
552     bool implicit;  /* if true, this filter node was automatically inserted */
553 
554     BlockDriver *drv; /* NULL means no media */
555     void *opaque;
556 
557     AioContext *aio_context; /* event loop used for fd handlers, timers, etc */
558     /* long-running tasks intended to always use the same AioContext as this
559      * BDS may register themselves in this list to be notified of changes
560      * regarding this BDS's context */
561     QLIST_HEAD(, BdrvAioNotifier) aio_notifiers;
562     bool walking_aio_notifiers; /* to make removal during iteration safe */
563 
564     char filename[PATH_MAX];
565     char backing_file[PATH_MAX]; /* if non zero, the image is a diff of
566                                     this file image */
567     char backing_format[16]; /* if non-zero and backing_file exists */
568 
569     QDict *full_open_options;
570     char exact_filename[PATH_MAX];
571 
572     BdrvChild *backing;
573     BdrvChild *file;
574 
575     /* I/O Limits */
576     BlockLimits bl;
577 
578     /* Flags honored during pwrite (so far: BDRV_REQ_FUA) */
579     unsigned int supported_write_flags;
580     /* Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA,
581      * BDRV_REQ_MAY_UNMAP) */
582     unsigned int supported_zero_flags;
583 
584     /* the following member gives a name to every node on the bs graph. */
585     char node_name[32];
586     /* element of the list of named nodes building the graph */
587     QTAILQ_ENTRY(BlockDriverState) node_list;
588     /* element of the list of all BlockDriverStates (all_bdrv_states) */
589     QTAILQ_ENTRY(BlockDriverState) bs_list;
590     /* element of the list of monitor-owned BDS */
591     QTAILQ_ENTRY(BlockDriverState) monitor_list;
592     int refcnt;
593 
594     /* operation blockers */
595     QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX];
596 
597     /* long-running background operation */
598     BlockJob *job;
599 
600     /* The node that this node inherited default options from (and a reopen on
601      * which can affect this node by changing these defaults). This is always a
602      * parent node of this node. */
603     BlockDriverState *inherits_from;
604     QLIST_HEAD(, BdrvChild) children;
605     QLIST_HEAD(, BdrvChild) parents;
606 
607     QDict *options;
608     QDict *explicit_options;
609     BlockdevDetectZeroesOptions detect_zeroes;
610 
611     /* The error object in use for blocking operations on backing_hd */
612     Error *backing_blocker;
613 
614     /* Protected by AioContext lock */
615 
616     /* If we are reading a disk image, give its size in sectors.
617      * Generally read-only; it is written to by load_snapshot and
618      * save_snaphost, but the block layer is quiescent during those.
619      */
620     int64_t total_sectors;
621 
622     /* Callback before write request is processed */
623     NotifierWithReturnList before_write_notifiers;
624 
625     /* threshold limit for writes, in bytes. "High water mark". */
626     uint64_t write_threshold_offset;
627     NotifierWithReturn write_threshold_notifier;
628 
629     /* Writing to the list requires the BQL _and_ the dirty_bitmap_mutex.
630      * Reading from the list can be done with either the BQL or the
631      * dirty_bitmap_mutex.  Modifying a bitmap only requires
632      * dirty_bitmap_mutex.  */
633     QemuMutex dirty_bitmap_mutex;
634     QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps;
635 
636     /* Offset after the highest byte written to */
637     Stat64 wr_highest_offset;
638 
639     /* If true, copy read backing sectors into image.  Can be >1 if more
640      * than one client has requested copy-on-read.  Accessed with atomic
641      * ops.
642      */
643     int copy_on_read;
644 
645     /* number of in-flight requests; overall and serialising.
646      * Accessed with atomic ops.
647      */
648     unsigned int in_flight;
649     unsigned int serialising_in_flight;
650 
651     /* Internal to BDRV_POLL_WHILE and bdrv_wakeup.  Accessed with atomic
652      * ops.
653      */
654     bool wakeup;
655 
656     /* counter for nested bdrv_io_plug.
657      * Accessed with atomic ops.
658     */
659     unsigned io_plugged;
660 
661     /* do we need to tell the quest if we have a volatile write cache? */
662     int enable_write_cache;
663 
664     /* Accessed with atomic ops.  */
665     int quiesce_counter;
666     unsigned int write_gen;               /* Current data generation */
667 
668     /* Protected by reqs_lock.  */
669     CoMutex reqs_lock;
670     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
671     CoQueue flush_queue;                  /* Serializing flush queue */
672     bool active_flush_req;                /* Flush request in flight? */
673 
674     /* Only read/written by whoever has set active_flush_req to true.  */
675     unsigned int flushed_gen;             /* Flushed write generation */
676 };
677 
678 struct BlockBackendRootState {
679     int open_flags;
680     bool read_only;
681     BlockdevDetectZeroesOptions detect_zeroes;
682 };
683 
684 typedef enum BlockMirrorBackingMode {
685     /* Reuse the existing backing chain from the source for the target.
686      * - sync=full: Set backing BDS to NULL.
687      * - sync=top:  Use source's backing BDS.
688      * - sync=none: Use source as the backing BDS. */
689     MIRROR_SOURCE_BACKING_CHAIN,
690 
691     /* Open the target's backing chain completely anew */
692     MIRROR_OPEN_BACKING_CHAIN,
693 
694     /* Do not change the target's backing BDS after job completion */
695     MIRROR_LEAVE_BACKING_CHAIN,
696 } BlockMirrorBackingMode;
697 
698 static inline BlockDriverState *backing_bs(BlockDriverState *bs)
699 {
700     return bs->backing ? bs->backing->bs : NULL;
701 }
702 
703 
704 /* Essential block drivers which must always be statically linked into qemu, and
705  * which therefore can be accessed without using bdrv_find_format() */
706 extern BlockDriver bdrv_file;
707 extern BlockDriver bdrv_raw;
708 extern BlockDriver bdrv_qcow2;
709 
710 int coroutine_fn bdrv_co_preadv(BdrvChild *child,
711     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
712     BdrvRequestFlags flags);
713 int coroutine_fn bdrv_co_pwritev(BdrvChild *child,
714     int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
715     BdrvRequestFlags flags);
716 
717 int get_tmp_filename(char *filename, int size);
718 BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size,
719                             const char *filename);
720 
721 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix,
722                                       QDict *options);
723 
724 
725 /**
726  * bdrv_add_before_write_notifier:
727  *
728  * Register a callback that is invoked before write requests are processed but
729  * after any throttling or waiting for overlapping requests.
730  */
731 void bdrv_add_before_write_notifier(BlockDriverState *bs,
732                                     NotifierWithReturn *notifier);
733 
734 /**
735  * bdrv_detach_aio_context:
736  *
737  * May be called from .bdrv_detach_aio_context() to detach children from the
738  * current #AioContext.  This is only needed by block drivers that manage their
739  * own children.  Both ->file and ->backing are automatically handled and
740  * block drivers should not call this function on them explicitly.
741  */
742 void bdrv_detach_aio_context(BlockDriverState *bs);
743 
744 /**
745  * bdrv_attach_aio_context:
746  *
747  * May be called from .bdrv_attach_aio_context() to attach children to the new
748  * #AioContext.  This is only needed by block drivers that manage their own
749  * children.  Both ->file and ->backing are automatically handled and block
750  * drivers should not call this function on them explicitly.
751  */
752 void bdrv_attach_aio_context(BlockDriverState *bs,
753                              AioContext *new_context);
754 
755 /**
756  * bdrv_add_aio_context_notifier:
757  *
758  * If a long-running job intends to be always run in the same AioContext as a
759  * certain BDS, it may use this function to be notified of changes regarding the
760  * association of the BDS to an AioContext.
761  *
762  * attached_aio_context() is called after the target BDS has been attached to a
763  * new AioContext; detach_aio_context() is called before the target BDS is being
764  * detached from its old AioContext.
765  */
766 void bdrv_add_aio_context_notifier(BlockDriverState *bs,
767         void (*attached_aio_context)(AioContext *new_context, void *opaque),
768         void (*detach_aio_context)(void *opaque), void *opaque);
769 
770 /**
771  * bdrv_remove_aio_context_notifier:
772  *
773  * Unsubscribe of change notifications regarding the BDS's AioContext. The
774  * parameters given here have to be the same as those given to
775  * bdrv_add_aio_context_notifier().
776  */
777 void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
778                                       void (*aio_context_attached)(AioContext *,
779                                                                    void *),
780                                       void (*aio_context_detached)(void *),
781                                       void *opaque);
782 
783 /**
784  * bdrv_wakeup:
785  * @bs: The BlockDriverState for which an I/O operation has been completed.
786  *
787  * Wake up the main thread if it is waiting on BDRV_POLL_WHILE.  During
788  * synchronous I/O on a BlockDriverState that is attached to another
789  * I/O thread, the main thread lets the I/O thread's event loop run,
790  * waiting for the I/O operation to complete.  A bdrv_wakeup will wake
791  * up the main thread if necessary.
792  *
793  * Manual calls to bdrv_wakeup are rarely necessary, because
794  * bdrv_dec_in_flight already calls it.
795  */
796 void bdrv_wakeup(BlockDriverState *bs);
797 
798 #ifdef _WIN32
799 int is_windows_drive(const char *filename);
800 #endif
801 
802 /**
803  * stream_start:
804  * @job_id: The id of the newly-created job, or %NULL to use the
805  * device name of @bs.
806  * @bs: Block device to operate on.
807  * @base: Block device that will become the new base, or %NULL to
808  * flatten the whole backing file chain onto @bs.
809  * @backing_file_str: The file name that will be written to @bs as the
810  * the new backing file if the job completes. Ignored if @base is %NULL.
811  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
812  * @on_error: The action to take upon error.
813  * @errp: Error object.
814  *
815  * Start a streaming operation on @bs.  Clusters that are unallocated
816  * in @bs, but allocated in any image between @base and @bs (both
817  * exclusive) will be written to @bs.  At the end of a successful
818  * streaming job, the backing file of @bs will be changed to
819  * @backing_file_str in the written image and to @base in the live
820  * BlockDriverState.
821  */
822 void stream_start(const char *job_id, BlockDriverState *bs,
823                   BlockDriverState *base, const char *backing_file_str,
824                   int64_t speed, BlockdevOnError on_error, Error **errp);
825 
826 /**
827  * commit_start:
828  * @job_id: The id of the newly-created job, or %NULL to use the
829  * device name of @bs.
830  * @bs: Active block device.
831  * @top: Top block device to be committed.
832  * @base: Block device that will be written into, and become the new top.
833  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
834  * @on_error: The action to take upon error.
835  * @backing_file_str: String to use as the backing file in @top's overlay
836  * @filter_node_name: The node name that should be assigned to the filter
837  * driver that the commit job inserts into the graph above @top. NULL means
838  * that a node name should be autogenerated.
839  * @errp: Error object.
840  *
841  */
842 void commit_start(const char *job_id, BlockDriverState *bs,
843                   BlockDriverState *base, BlockDriverState *top, int64_t speed,
844                   BlockdevOnError on_error, const char *backing_file_str,
845                   const char *filter_node_name, Error **errp);
846 /**
847  * commit_active_start:
848  * @job_id: The id of the newly-created job, or %NULL to use the
849  * device name of @bs.
850  * @bs: Active block device to be committed.
851  * @base: Block device that will be written into, and become the new top.
852  * @creation_flags: Flags that control the behavior of the Job lifetime.
853  *                  See @BlockJobCreateFlags
854  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
855  * @on_error: The action to take upon error.
856  * @filter_node_name: The node name that should be assigned to the filter
857  * driver that the commit job inserts into the graph above @bs. NULL means that
858  * a node name should be autogenerated.
859  * @cb: Completion function for the job.
860  * @opaque: Opaque pointer value passed to @cb.
861  * @auto_complete: Auto complete the job.
862  * @errp: Error object.
863  *
864  */
865 void commit_active_start(const char *job_id, BlockDriverState *bs,
866                          BlockDriverState *base, int creation_flags,
867                          int64_t speed, BlockdevOnError on_error,
868                          const char *filter_node_name,
869                          BlockCompletionFunc *cb, void *opaque,
870                          bool auto_complete, Error **errp);
871 /*
872  * mirror_start:
873  * @job_id: The id of the newly-created job, or %NULL to use the
874  * device name of @bs.
875  * @bs: Block device to operate on.
876  * @target: Block device to write to.
877  * @replaces: Block graph node name to replace once the mirror is done. Can
878  *            only be used when full mirroring is selected.
879  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
880  * @granularity: The chosen granularity for the dirty bitmap.
881  * @buf_size: The amount of data that can be in flight at one time.
882  * @mode: Whether to collapse all images in the chain to the target.
883  * @backing_mode: How to establish the target's backing chain after completion.
884  * @on_source_error: The action to take upon error reading from the source.
885  * @on_target_error: The action to take upon error writing to the target.
886  * @unmap: Whether to unmap target where source sectors only contain zeroes.
887  * @filter_node_name: The node name that should be assigned to the filter
888  * driver that the mirror job inserts into the graph above @bs. NULL means that
889  * a node name should be autogenerated.
890  * @errp: Error object.
891  *
892  * Start a mirroring operation on @bs.  Clusters that are allocated
893  * in @bs will be written to @target until the job is cancelled or
894  * manually completed.  At the end of a successful mirroring job,
895  * @bs will be switched to read from @target.
896  */
897 void mirror_start(const char *job_id, BlockDriverState *bs,
898                   BlockDriverState *target, const char *replaces,
899                   int64_t speed, uint32_t granularity, int64_t buf_size,
900                   MirrorSyncMode mode, BlockMirrorBackingMode backing_mode,
901                   BlockdevOnError on_source_error,
902                   BlockdevOnError on_target_error,
903                   bool unmap, const char *filter_node_name, Error **errp);
904 
905 /*
906  * backup_job_create:
907  * @job_id: The id of the newly-created job, or %NULL to use the
908  * device name of @bs.
909  * @bs: Block device to operate on.
910  * @target: Block device to write to.
911  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
912  * @sync_mode: What parts of the disk image should be copied to the destination.
913  * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL.
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  * @creation_flags: Flags that control the behavior of the Job lifetime.
917  *                  See @BlockJobCreateFlags
918  * @cb: Completion function for the job.
919  * @opaque: Opaque pointer value passed to @cb.
920  * @txn: Transaction that this job is part of (may be NULL).
921  *
922  * Create a backup operation on @bs.  Clusters in @bs are written to @target
923  * until the job is cancelled or manually completed.
924  */
925 BlockJob *backup_job_create(const char *job_id, BlockDriverState *bs,
926                             BlockDriverState *target, int64_t speed,
927                             MirrorSyncMode sync_mode,
928                             BdrvDirtyBitmap *sync_bitmap,
929                             bool compress,
930                             BlockdevOnError on_source_error,
931                             BlockdevOnError on_target_error,
932                             int creation_flags,
933                             BlockCompletionFunc *cb, void *opaque,
934                             BlockJobTxn *txn, Error **errp);
935 
936 void hmp_drive_add_node(Monitor *mon, const char *optstr);
937 
938 BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs,
939                                   const char *child_name,
940                                   const BdrvChildRole *child_role,
941                                   uint64_t perm, uint64_t shared_perm,
942                                   void *opaque, Error **errp);
943 void bdrv_root_unref_child(BdrvChild *child);
944 
945 int bdrv_child_try_set_perm(BdrvChild *c, uint64_t perm, uint64_t shared,
946                             Error **errp);
947 
948 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
949  * block filters: Forward CONSISTENT_READ, WRITE, WRITE_UNCHANGED and RESIZE to
950  * all children */
951 void bdrv_filter_default_perms(BlockDriverState *bs, BdrvChild *c,
952                                const BdrvChildRole *role,
953                                uint64_t perm, uint64_t shared,
954                                uint64_t *nperm, uint64_t *nshared);
955 
956 /* Default implementation for BlockDriver.bdrv_child_perm() that can be used by
957  * (non-raw) image formats: Like above for bs->backing, but for bs->file it
958  * requires WRITE | RESIZE for read-write images, always requires
959  * CONSISTENT_READ and doesn't share WRITE. */
960 void bdrv_format_default_perms(BlockDriverState *bs, BdrvChild *c,
961                                const BdrvChildRole *role,
962                                uint64_t perm, uint64_t shared,
963                                uint64_t *nperm, uint64_t *nshared);
964 
965 const char *bdrv_get_parent_name(const BlockDriverState *bs);
966 void blk_dev_change_media_cb(BlockBackend *blk, bool load, Error **errp);
967 bool blk_dev_has_removable_media(BlockBackend *blk);
968 bool blk_dev_has_tray(BlockBackend *blk);
969 void blk_dev_eject_request(BlockBackend *blk, bool force);
970 bool blk_dev_is_tray_open(BlockBackend *blk);
971 bool blk_dev_is_medium_locked(BlockBackend *blk);
972 
973 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int64_t nr_sect);
974 bool bdrv_requests_pending(BlockDriverState *bs);
975 
976 void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out);
977 void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in);
978 
979 void bdrv_inc_in_flight(BlockDriverState *bs);
980 void bdrv_dec_in_flight(BlockDriverState *bs);
981 
982 void blockdev_close_all_bdrv_states(void);
983 
984 #endif /* BLOCK_INT_H */
985