]>
Commit | Line | Data |
---|---|---|
ea2384d3 FB |
1 | /* |
2 | * QEMU System Emulator block driver | |
5fafdf24 | 3 | * |
ea2384d3 | 4 | * Copyright (c) 2003 Fabrice Bellard |
5fafdf24 | 5 | * |
ea2384d3 FB |
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 | ||
5e5a94b6 | 27 | #include "block/accounting.h" |
737e150e | 28 | #include "block/block.h" |
281d22d8 | 29 | #include "block/throttle-groups.h" |
1de7afc9 PB |
30 | #include "qemu/option.h" |
31 | #include "qemu/queue.h" | |
10817bf0 | 32 | #include "qemu/coroutine.h" |
1de7afc9 | 33 | #include "qemu/timer.h" |
b2023818 | 34 | #include "qapi-types.h" |
8f0720ec | 35 | #include "qemu/hbitmap.h" |
f364ec65 | 36 | #include "block/snapshot.h" |
6a1751b7 | 37 | #include "qemu/main-loop.h" |
cc0681c4 | 38 | #include "qemu/throttle.h" |
faf07963 | 39 | |
bfe8043e SH |
40 | #define BLOCK_FLAG_ENCRYPT 1 |
41 | #define BLOCK_FLAG_COMPAT6 4 | |
42 | #define BLOCK_FLAG_LAZY_REFCOUNTS 8 | |
ec36ba14 | 43 | |
bfe8043e SH |
44 | #define BLOCK_OPT_SIZE "size" |
45 | #define BLOCK_OPT_ENCRYPT "encryption" | |
46 | #define BLOCK_OPT_COMPAT6 "compat6" | |
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" | |
7f2039f6 | 55 | #define BLOCK_OPT_ADAPTER_TYPE "adapter_type" |
b3af018f | 56 | #define BLOCK_OPT_REDUNDANCY "redundancy" |
4ab15590 | 57 | #define BLOCK_OPT_NOCOW "nocow" |
876eb1b0 | 58 | #define BLOCK_OPT_OBJECT_SIZE "object_size" |
06d05fa7 | 59 | #define BLOCK_OPT_REFCOUNT_BITS "refcount_bits" |
0e7e1989 | 60 | |
7cddd372 KW |
61 | #define BLOCK_PROBE_BUF_SIZE 512 |
62 | ||
ebde595c FZ |
63 | enum BdrvTrackedRequestType { |
64 | BDRV_TRACKED_READ, | |
65 | BDRV_TRACKED_WRITE, | |
66 | BDRV_TRACKED_FLUSH, | |
67 | BDRV_TRACKED_IOCTL, | |
68 | BDRV_TRACKED_DISCARD, | |
69 | }; | |
70 | ||
d616b224 SH |
71 | typedef struct BdrvTrackedRequest { |
72 | BlockDriverState *bs; | |
793ed47a KW |
73 | int64_t offset; |
74 | unsigned int bytes; | |
ebde595c | 75 | enum BdrvTrackedRequestType type; |
7327145f | 76 | |
2dbafdc0 | 77 | bool serialising; |
7327145f KW |
78 | int64_t overlap_offset; |
79 | unsigned int overlap_bytes; | |
80 | ||
d616b224 SH |
81 | QLIST_ENTRY(BdrvTrackedRequest) list; |
82 | Coroutine *co; /* owner, used for deadlock detection */ | |
83 | CoQueue wait_queue; /* coroutines blocked on this request */ | |
6460440f KW |
84 | |
85 | struct BdrvTrackedRequest *waiting_for; | |
d616b224 SH |
86 | } BdrvTrackedRequest; |
87 | ||
ea2384d3 FB |
88 | struct BlockDriver { |
89 | const char *format_name; | |
90 | int instance_size; | |
f6186f49 | 91 | |
b5042a36 BC |
92 | /* set to true if the BlockDriver is a block filter */ |
93 | bool is_filter; | |
94 | /* for snapshots block filter like Quorum can implement the | |
95 | * following recursive callback. | |
212a5a8f BC |
96 | * It's purpose is to recurse on the filter children while calling |
97 | * bdrv_recurse_is_first_non_filter on them. | |
98 | * For a sample implementation look in the future Quorum block filter. | |
f6186f49 | 99 | */ |
212a5a8f BC |
100 | bool (*bdrv_recurse_is_first_non_filter)(BlockDriverState *bs, |
101 | BlockDriverState *candidate); | |
f6186f49 | 102 | |
ea2384d3 | 103 | int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename); |
508c7cb3 | 104 | int (*bdrv_probe_device)(const char *filename); |
c2ad1b0c KW |
105 | |
106 | /* Any driver implementing this callback is expected to be able to handle | |
107 | * NULL file names in its .bdrv_open() implementation */ | |
6963a30d | 108 | void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp); |
030be321 BC |
109 | /* Drivers not implementing bdrv_parse_filename nor bdrv_open should have |
110 | * this field set to true, except ones that are defined only by their | |
111 | * child's bs. | |
112 | * An example of the last type will be the quorum block driver. | |
113 | */ | |
114 | bool bdrv_needs_filename; | |
e971aa12 | 115 | |
8ee79e70 KW |
116 | /* Set if a driver can support backing files */ |
117 | bool supports_backing; | |
118 | ||
e971aa12 JC |
119 | /* For handling image reopen for split or non-split files */ |
120 | int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state, | |
121 | BlockReopenQueue *queue, Error **errp); | |
122 | void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state); | |
123 | void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state); | |
5365f44d | 124 | void (*bdrv_join_options)(QDict *options, QDict *old_options); |
e971aa12 | 125 | |
015a1036 HR |
126 | int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags, |
127 | Error **errp); | |
128 | int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags, | |
129 | Error **errp); | |
5fafdf24 | 130 | int (*bdrv_read)(BlockDriverState *bs, int64_t sector_num, |
ea2384d3 | 131 | uint8_t *buf, int nb_sectors); |
5fafdf24 | 132 | int (*bdrv_write)(BlockDriverState *bs, int64_t sector_num, |
ea2384d3 | 133 | const uint8_t *buf, int nb_sectors); |
e2731add | 134 | void (*bdrv_close)(BlockDriverState *bs); |
c282e1fd | 135 | int (*bdrv_create)(const char *filename, QemuOpts *opts, Error **errp); |
ea2384d3 | 136 | int (*bdrv_set_key)(BlockDriverState *bs, const char *key); |
95389c86 | 137 | int (*bdrv_make_empty)(BlockDriverState *bs); |
91af7014 | 138 | |
4cdd01d3 | 139 | void (*bdrv_refresh_filename)(BlockDriverState *bs, QDict *options); |
91af7014 | 140 | |
83f64091 | 141 | /* aio */ |
7c84b1b8 | 142 | BlockAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs, |
f141eafe | 143 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
097310b5 | 144 | BlockCompletionFunc *cb, void *opaque); |
7c84b1b8 | 145 | BlockAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs, |
f141eafe | 146 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
097310b5 | 147 | BlockCompletionFunc *cb, void *opaque); |
7c84b1b8 | 148 | BlockAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs, |
097310b5 | 149 | BlockCompletionFunc *cb, void *opaque); |
7c84b1b8 | 150 | BlockAIOCB *(*bdrv_aio_discard)(BlockDriverState *bs, |
4265d620 | 151 | int64_t sector_num, int nb_sectors, |
097310b5 | 152 | BlockCompletionFunc *cb, void *opaque); |
83f64091 | 153 | |
da1fa91d KW |
154 | int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs, |
155 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov); | |
156 | int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs, | |
157 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov); | |
f08f2dda SH |
158 | /* |
159 | * Efficiently zero a region of the disk image. Typically an image format | |
160 | * would use a compact metadata representation to implement this. This | |
161 | * function pointer may be NULL and .bdrv_co_writev() will be called | |
162 | * instead. | |
163 | */ | |
164 | int coroutine_fn (*bdrv_co_write_zeroes)(BlockDriverState *bs, | |
aa7bfbff | 165 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags); |
4265d620 PB |
166 | int coroutine_fn (*bdrv_co_discard)(BlockDriverState *bs, |
167 | int64_t sector_num, int nb_sectors); | |
b6b8a333 | 168 | int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs, |
376ae3f1 | 169 | int64_t sector_num, int nb_sectors, int *pnum); |
da1fa91d | 170 | |
0f15423c AL |
171 | /* |
172 | * Invalidate any cached meta-data. | |
173 | */ | |
5a8a30db | 174 | void (*bdrv_invalidate_cache)(BlockDriverState *bs, Error **errp); |
0f15423c | 175 | |
c68b89ac KW |
176 | /* |
177 | * Flushes all data that was already written to the OS all the way down to | |
178 | * the disk (for example raw-posix calls fsync()). | |
179 | */ | |
180 | int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs); | |
181 | ||
eb489bb1 KW |
182 | /* |
183 | * Flushes all internal caches to the OS. The data may still sit in a | |
184 | * writeback cache of the host OS, but it will survive a crash of the qemu | |
185 | * process. | |
186 | */ | |
187 | int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs); | |
188 | ||
83f64091 | 189 | const char *protocol_name; |
83f64091 | 190 | int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset); |
b94a2610 | 191 | |
83f64091 | 192 | int64_t (*bdrv_getlength)(BlockDriverState *bs); |
b94a2610 | 193 | bool has_variable_length; |
4a1d5e1f | 194 | int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs); |
b94a2610 | 195 | |
5fafdf24 | 196 | int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num, |
faea38e7 FB |
197 | const uint8_t *buf, int nb_sectors); |
198 | ||
5fafdf24 | 199 | int (*bdrv_snapshot_create)(BlockDriverState *bs, |
faea38e7 | 200 | QEMUSnapshotInfo *sn_info); |
5fafdf24 | 201 | int (*bdrv_snapshot_goto)(BlockDriverState *bs, |
faea38e7 | 202 | const char *snapshot_id); |
a89d89d3 WX |
203 | int (*bdrv_snapshot_delete)(BlockDriverState *bs, |
204 | const char *snapshot_id, | |
205 | const char *name, | |
206 | Error **errp); | |
5fafdf24 | 207 | int (*bdrv_snapshot_list)(BlockDriverState *bs, |
faea38e7 | 208 | QEMUSnapshotInfo **psn_info); |
51ef6727 | 209 | int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs, |
7b4c4781 WX |
210 | const char *snapshot_id, |
211 | const char *name, | |
212 | Error **errp); | |
faea38e7 | 213 | int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi); |
eae041fe | 214 | ImageInfoSpecific *(*bdrv_get_specific_info)(BlockDriverState *bs); |
83f64091 | 215 | |
cf8074b3 KW |
216 | int (*bdrv_save_vmstate)(BlockDriverState *bs, QEMUIOVector *qiov, |
217 | int64_t pos); | |
45566e9c CH |
218 | int (*bdrv_load_vmstate)(BlockDriverState *bs, uint8_t *buf, |
219 | int64_t pos, int size); | |
178e08a5 | 220 | |
756e6736 KW |
221 | int (*bdrv_change_backing_file)(BlockDriverState *bs, |
222 | const char *backing_file, const char *backing_fmt); | |
223 | ||
19cb3738 | 224 | /* removable device specific */ |
e031f750 | 225 | bool (*bdrv_is_inserted)(BlockDriverState *bs); |
19cb3738 | 226 | int (*bdrv_media_changed)(BlockDriverState *bs); |
f36f3949 | 227 | void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag); |
025e849a | 228 | void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked); |
3b46e624 | 229 | |
985a03b0 | 230 | /* to control generic scsi devices */ |
7c84b1b8 | 231 | BlockAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs, |
221f715d | 232 | unsigned long int req, void *buf, |
097310b5 | 233 | BlockCompletionFunc *cb, void *opaque); |
985a03b0 | 234 | |
0e7e1989 | 235 | /* List of options for creating images, terminated by name == NULL */ |
83d0521a | 236 | QemuOptsList *create_opts; |
5eb45639 | 237 | |
9ac228e0 KW |
238 | /* |
239 | * Returns 0 for completed check, -errno for internal errors. | |
240 | * The check results are stored in result. | |
241 | */ | |
4534ff54 KW |
242 | int (*bdrv_check)(BlockDriverState* bs, BdrvCheckResult *result, |
243 | BdrvCheckMode fix); | |
e97fc193 | 244 | |
77485434 HR |
245 | int (*bdrv_amend_options)(BlockDriverState *bs, QemuOpts *opts, |
246 | BlockDriverAmendStatusCB *status_cb); | |
6f176b48 | 247 | |
a31939e6 | 248 | void (*bdrv_debug_event)(BlockDriverState *bs, BlkdebugEvent event); |
8b9b0cc2 | 249 | |
41c695c7 KW |
250 | /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */ |
251 | int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event, | |
252 | const char *tag); | |
4cc70e93 FZ |
253 | int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs, |
254 | const char *tag); | |
41c695c7 KW |
255 | int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag); |
256 | bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag); | |
257 | ||
3baca891 | 258 | void (*bdrv_refresh_limits)(BlockDriverState *bs, Error **errp); |
d34682cd | 259 | |
336c1c12 KW |
260 | /* |
261 | * Returns 1 if newly created images are guaranteed to contain only | |
262 | * zeros, 0 otherwise. | |
263 | */ | |
264 | int (*bdrv_has_zero_init)(BlockDriverState *bs); | |
12c09b8c | 265 | |
dcd04228 SH |
266 | /* Remove fd handlers, timers, and other event loop callbacks so the event |
267 | * loop is no longer in use. Called with no in-flight requests and in | |
268 | * depth-first traversal order with parents before child nodes. | |
269 | */ | |
270 | void (*bdrv_detach_aio_context)(BlockDriverState *bs); | |
271 | ||
272 | /* Add fd handlers, timers, and other event loop callbacks so I/O requests | |
273 | * can be processed again. Called with no in-flight requests and in | |
274 | * depth-first traversal order with child nodes before parent nodes. | |
275 | */ | |
276 | void (*bdrv_attach_aio_context)(BlockDriverState *bs, | |
277 | AioContext *new_context); | |
278 | ||
448ad91d ML |
279 | /* io queue for linux-aio */ |
280 | void (*bdrv_io_plug)(BlockDriverState *bs); | |
281 | void (*bdrv_io_unplug)(BlockDriverState *bs); | |
282 | void (*bdrv_flush_io_queue)(BlockDriverState *bs); | |
283 | ||
892b7de8 ET |
284 | /** |
285 | * Try to get @bs's logical and physical block size. | |
286 | * On success, store them in @bsz and return zero. | |
287 | * On failure, return negative errno. | |
288 | */ | |
289 | int (*bdrv_probe_blocksizes)(BlockDriverState *bs, BlockSizes *bsz); | |
290 | /** | |
291 | * Try to get @bs's geometry (cyls, heads, sectors) | |
292 | * On success, store them in @geo and return 0. | |
293 | * On failure return -errno. | |
294 | * Only drivers that want to override guest geometry implement this | |
295 | * callback; see hd_geometry_guess(). | |
296 | */ | |
297 | int (*bdrv_probe_geometry)(BlockDriverState *bs, HDGeometry *geo); | |
298 | ||
67da1dc5 FZ |
299 | /** |
300 | * Drain and stop any internal sources of requests in the driver, and | |
301 | * remain so until next I/O callback (e.g. bdrv_co_writev) is called. | |
302 | */ | |
303 | void (*bdrv_drain)(BlockDriverState *bs); | |
304 | ||
8a22f02a | 305 | QLIST_ENTRY(BlockDriver) list; |
ea2384d3 FB |
306 | }; |
307 | ||
fe81c2cc PL |
308 | typedef struct BlockLimits { |
309 | /* maximum number of sectors that can be discarded at once */ | |
310 | int max_discard; | |
311 | ||
312 | /* optimal alignment for discard requests in sectors */ | |
313 | int64_t discard_alignment; | |
314 | ||
315 | /* maximum number of sectors that can zeroized at once */ | |
316 | int max_write_zeroes; | |
317 | ||
318 | /* optimal alignment for write zeroes requests in sectors */ | |
319 | int64_t write_zeroes_alignment; | |
7337acaf PL |
320 | |
321 | /* optimal transfer length in sectors */ | |
322 | int opt_transfer_length; | |
339064d5 | 323 | |
2647fab5 PL |
324 | /* maximal transfer length in sectors */ |
325 | int max_transfer_length; | |
326 | ||
339064d5 | 327 | /* memory alignment so that no bounce buffer is needed */ |
4196d2f0 DL |
328 | size_t min_mem_alignment; |
329 | ||
330 | /* memory alignment for bounce buffer */ | |
339064d5 | 331 | size_t opt_mem_alignment; |
fe81c2cc PL |
332 | } BlockLimits; |
333 | ||
fbe40ff7 FZ |
334 | typedef struct BdrvOpBlocker BdrvOpBlocker; |
335 | ||
33384421 HR |
336 | typedef struct BdrvAioNotifier { |
337 | void (*attached_aio_context)(AioContext *new_context, void *opaque); | |
338 | void (*detach_aio_context)(void *opaque); | |
339 | ||
340 | void *opaque; | |
341 | ||
342 | QLIST_ENTRY(BdrvAioNotifier) list; | |
343 | } BdrvAioNotifier; | |
344 | ||
f3930ed0 | 345 | struct BdrvChildRole { |
8e2160e2 KW |
346 | void (*inherit_options)(int *child_flags, QDict *child_options, |
347 | int parent_flags, QDict *parent_options); | |
f3930ed0 KW |
348 | }; |
349 | ||
350 | extern const BdrvChildRole child_file; | |
351 | extern const BdrvChildRole child_format; | |
352 | ||
b4b059f6 | 353 | struct BdrvChild { |
6e93e7c4 | 354 | BlockDriverState *bs; |
260fecf1 | 355 | char *name; |
6e93e7c4 KW |
356 | const BdrvChildRole *role; |
357 | QLIST_ENTRY(BdrvChild) next; | |
d42a8a93 | 358 | QLIST_ENTRY(BdrvChild) next_parent; |
b4b059f6 | 359 | }; |
6e93e7c4 | 360 | |
8802d1fd JC |
361 | /* |
362 | * Note: the function bdrv_append() copies and swaps contents of | |
363 | * BlockDriverStates, so if you add new fields to this struct, please | |
364 | * inspect bdrv_append() to determine if the new fields need to be | |
365 | * copied as well. | |
366 | */ | |
ea2384d3 | 367 | struct BlockDriverState { |
d15a771d FB |
368 | int64_t total_sectors; /* if we are reading a disk image, give its |
369 | size in sectors */ | |
ea2384d3 | 370 | int read_only; /* if true, the media is read only */ |
4dca4b63 | 371 | int open_flags; /* flags used to open the file, re-used for re-open */ |
ea2384d3 | 372 | int encrypted; /* if true, the media is encrypted */ |
c0f4ce77 | 373 | int valid_key; /* if true, a valid encryption key has been set */ |
985a03b0 | 374 | int sg; /* if true, the device is a /dev/sg* */ |
53fec9d3 SH |
375 | int copy_on_read; /* if true, copy read backing sectors into image |
376 | note this is a reference count */ | |
38f3ef57 | 377 | bool probed; |
ea2384d3 | 378 | |
19cb3738 | 379 | BlockDriver *drv; /* NULL means no media */ |
ea2384d3 FB |
380 | void *opaque; |
381 | ||
7e7d56d9 MA |
382 | BlockBackend *blk; /* owning backend, if any */ |
383 | ||
dcd04228 | 384 | AioContext *aio_context; /* event loop used for fd handlers, timers, etc */ |
33384421 HR |
385 | /* long-running tasks intended to always use the same AioContext as this |
386 | * BDS may register themselves in this list to be notified of changes | |
387 | * regarding this BDS's context */ | |
388 | QLIST_HEAD(, BdrvAioNotifier) aio_notifiers; | |
dcd04228 | 389 | |
9a29e18f JC |
390 | char filename[PATH_MAX]; |
391 | char backing_file[PATH_MAX]; /* if non zero, the image is a diff of | |
392 | this file image */ | |
5eb45639 | 393 | char backing_format[16]; /* if non-zero and backing_file exists */ |
19cb3738 | 394 | |
91af7014 | 395 | QDict *full_open_options; |
9a29e18f | 396 | char exact_filename[PATH_MAX]; |
91af7014 | 397 | |
760e0063 | 398 | BdrvChild *backing; |
9a4f4c31 | 399 | BdrvChild *file; |
66f82cee | 400 | |
d7d512f6 PB |
401 | NotifierList close_notifiers; |
402 | ||
d616b224 SH |
403 | /* Callback before write request is processed */ |
404 | NotifierWithReturnList before_write_notifiers; | |
405 | ||
2dbafdc0 KW |
406 | /* number of in-flight serialising requests */ |
407 | unsigned int serialising_in_flight; | |
470c0504 | 408 | |
a0d64a61 AG |
409 | /* I/O throttling. |
410 | * throttle_state tells us if this BDS has I/O limits configured. | |
411 | * io_limits_enabled tells us if they are currently being | |
412 | * enforced, but it can be temporarily set to false */ | |
cc0681c4 | 413 | CoQueue throttled_reqs[2]; |
0563e191 | 414 | bool io_limits_enabled; |
76f4afb4 AG |
415 | /* The following fields are protected by the ThrottleGroup lock. |
416 | * See the ThrottleGroup documentation for details. */ | |
417 | ThrottleState *throttle_state; | |
418 | ThrottleTimers throttle_timers; | |
419 | unsigned pending_reqs[2]; | |
2ff1f2e3 | 420 | QLIST_ENTRY(BlockDriverState) round_robin; |
0563e191 | 421 | |
53d8f9d8 HR |
422 | /* Offset after the highest byte written to */ |
423 | uint64_t wr_highest_offset; | |
424 | ||
fe81c2cc PL |
425 | /* I/O Limits */ |
426 | BlockLimits bl; | |
427 | ||
0d51b4de AH |
428 | /* Whether produces zeros when read beyond eof */ |
429 | bool zero_beyond_eof; | |
430 | ||
c25f53b0 PB |
431 | /* Alignment requirement for offset/length of I/O requests */ |
432 | unsigned int request_alignment; | |
433 | ||
e900a7b7 CH |
434 | /* do we need to tell the quest if we have a volatile write cache? */ |
435 | int enable_write_cache; | |
436 | ||
dc364f4c BC |
437 | /* the following member gives a name to every node on the bs graph. */ |
438 | char node_name[32]; | |
439 | /* element of the list of named nodes building the graph */ | |
440 | QTAILQ_ENTRY(BlockDriverState) node_list; | |
dc364f4c BC |
441 | /* element of the list of "drives" the guest sees */ |
442 | QTAILQ_ENTRY(BlockDriverState) device_list; | |
e4654d2d | 443 | QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps; |
9fcb0251 | 444 | int refcnt; |
dbffbdcf SH |
445 | |
446 | QLIST_HEAD(, BdrvTrackedRequest) tracked_requests; | |
eeec61f2 | 447 | |
fbe40ff7 FZ |
448 | /* operation blockers */ |
449 | QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX]; | |
450 | ||
eeec61f2 SH |
451 | /* long-running background operation */ |
452 | BlockJob *job; | |
e971aa12 | 453 | |
bddcec37 KW |
454 | /* The node that this node inherited default options from (and a reopen on |
455 | * which can affect this node by changing these defaults). This is always a | |
456 | * parent node of this node. */ | |
457 | BlockDriverState *inherits_from; | |
6e93e7c4 | 458 | QLIST_HEAD(, BdrvChild) children; |
d42a8a93 | 459 | QLIST_HEAD(, BdrvChild) parents; |
6e93e7c4 | 460 | |
de9c0cec | 461 | QDict *options; |
145f598e | 462 | QDict *explicit_options; |
465bee1d | 463 | BlockdevDetectZeroesOptions detect_zeroes; |
826b6ca0 FZ |
464 | |
465 | /* The error object in use for blocking operations on backing_hd */ | |
466 | Error *backing_blocker; | |
e2462113 FR |
467 | |
468 | /* threshold limit for writes, in bytes. "High water mark". */ | |
469 | uint64_t write_threshold_offset; | |
470 | NotifierWithReturn write_threshold_notifier; | |
51288d79 FZ |
471 | |
472 | int quiesce_counter; | |
ea2384d3 FB |
473 | }; |
474 | ||
281d22d8 HR |
475 | struct BlockBackendRootState { |
476 | int open_flags; | |
477 | bool read_only; | |
478 | BlockdevDetectZeroesOptions detect_zeroes; | |
479 | ||
480 | char *throttle_group; | |
481 | ThrottleState *throttle_state; | |
482 | }; | |
483 | ||
760e0063 KW |
484 | static inline BlockDriverState *backing_bs(BlockDriverState *bs) |
485 | { | |
486 | return bs->backing ? bs->backing->bs : NULL; | |
487 | } | |
488 | ||
5f535a94 HR |
489 | |
490 | /* Essential block drivers which must always be statically linked into qemu, and | |
491 | * which therefore can be accessed without using bdrv_find_format() */ | |
492 | extern BlockDriver bdrv_file; | |
493 | extern BlockDriver bdrv_raw; | |
494 | extern BlockDriver bdrv_qcow2; | |
495 | ||
c69a4dd8 HR |
496 | extern QTAILQ_HEAD(BdrvStates, BlockDriverState) bdrv_states; |
497 | ||
0eb7217e SH |
498 | /** |
499 | * bdrv_setup_io_funcs: | |
500 | * | |
501 | * Prepare a #BlockDriver for I/O request processing by populating | |
502 | * unimplemented coroutine and AIO interfaces with generic wrapper functions | |
503 | * that fall back to implemented interfaces. | |
504 | */ | |
505 | void bdrv_setup_io_funcs(BlockDriver *bdrv); | |
5f535a94 | 506 | |
eba25057 | 507 | int get_tmp_filename(char *filename, int size); |
38f3ef57 KW |
508 | BlockDriver *bdrv_probe_all(const uint8_t *buf, int buf_size, |
509 | const char *filename); | |
95389c86 | 510 | |
0563e191 | 511 | void bdrv_set_io_limits(BlockDriverState *bs, |
cc0681c4 BC |
512 | ThrottleConfig *cfg); |
513 | ||
0563e191 | 514 | |
d616b224 SH |
515 | /** |
516 | * bdrv_add_before_write_notifier: | |
517 | * | |
518 | * Register a callback that is invoked before write requests are processed but | |
519 | * after any throttling or waiting for overlapping requests. | |
520 | */ | |
521 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
522 | NotifierWithReturn *notifier); | |
523 | ||
dcd04228 SH |
524 | /** |
525 | * bdrv_detach_aio_context: | |
526 | * | |
527 | * May be called from .bdrv_detach_aio_context() to detach children from the | |
528 | * current #AioContext. This is only needed by block drivers that manage their | |
760e0063 | 529 | * own children. Both ->file and ->backing are automatically handled and |
dcd04228 SH |
530 | * block drivers should not call this function on them explicitly. |
531 | */ | |
532 | void bdrv_detach_aio_context(BlockDriverState *bs); | |
533 | ||
534 | /** | |
535 | * bdrv_attach_aio_context: | |
536 | * | |
537 | * May be called from .bdrv_attach_aio_context() to attach children to the new | |
538 | * #AioContext. This is only needed by block drivers that manage their own | |
760e0063 | 539 | * children. Both ->file and ->backing are automatically handled and block |
dcd04228 SH |
540 | * drivers should not call this function on them explicitly. |
541 | */ | |
542 | void bdrv_attach_aio_context(BlockDriverState *bs, | |
543 | AioContext *new_context); | |
544 | ||
33384421 HR |
545 | /** |
546 | * bdrv_add_aio_context_notifier: | |
547 | * | |
548 | * If a long-running job intends to be always run in the same AioContext as a | |
549 | * certain BDS, it may use this function to be notified of changes regarding the | |
550 | * association of the BDS to an AioContext. | |
551 | * | |
552 | * attached_aio_context() is called after the target BDS has been attached to a | |
553 | * new AioContext; detach_aio_context() is called before the target BDS is being | |
554 | * detached from its old AioContext. | |
555 | */ | |
556 | void bdrv_add_aio_context_notifier(BlockDriverState *bs, | |
557 | void (*attached_aio_context)(AioContext *new_context, void *opaque), | |
558 | void (*detach_aio_context)(void *opaque), void *opaque); | |
559 | ||
560 | /** | |
561 | * bdrv_remove_aio_context_notifier: | |
562 | * | |
563 | * Unsubscribe of change notifications regarding the BDS's AioContext. The | |
564 | * parameters given here have to be the same as those given to | |
565 | * bdrv_add_aio_context_notifier(). | |
566 | */ | |
567 | void bdrv_remove_aio_context_notifier(BlockDriverState *bs, | |
568 | void (*aio_context_attached)(AioContext *, | |
569 | void *), | |
570 | void (*aio_context_detached)(void *), | |
571 | void *opaque); | |
572 | ||
508c7cb3 CH |
573 | #ifdef _WIN32 |
574 | int is_windows_drive(const char *filename); | |
575 | #endif | |
576 | ||
dc534f8f PB |
577 | /** |
578 | * stream_start: | |
579 | * @bs: Block device to operate on. | |
580 | * @base: Block device that will become the new base, or %NULL to | |
581 | * flatten the whole backing file chain onto @bs. | |
582 | * @base_id: The file name that will be written to @bs as the new | |
583 | * backing file if the job completes. Ignored if @base is %NULL. | |
c83c66c3 | 584 | * @speed: The maximum speed, in bytes per second, or 0 for unlimited. |
1d809098 | 585 | * @on_error: The action to take upon error. |
dc534f8f PB |
586 | * @cb: Completion function for the job. |
587 | * @opaque: Opaque pointer value passed to @cb. | |
fd7f8c65 | 588 | * @errp: Error object. |
dc534f8f PB |
589 | * |
590 | * Start a streaming operation on @bs. Clusters that are unallocated | |
591 | * in @bs, but allocated in any image between @base and @bs (both | |
592 | * exclusive) will be written to @bs. At the end of a successful | |
593 | * streaming job, the backing file of @bs will be changed to | |
594 | * @base_id in the written image and to @base in the live BlockDriverState. | |
595 | */ | |
fd7f8c65 | 596 | void stream_start(BlockDriverState *bs, BlockDriverState *base, |
1d809098 | 597 | const char *base_id, int64_t speed, BlockdevOnError on_error, |
097310b5 | 598 | BlockCompletionFunc *cb, |
fd7f8c65 | 599 | void *opaque, Error **errp); |
4f1043b4 | 600 | |
747ff602 JC |
601 | /** |
602 | * commit_start: | |
03544a6e FZ |
603 | * @bs: Active block device. |
604 | * @top: Top block device to be committed. | |
605 | * @base: Block device that will be written into, and become the new top. | |
747ff602 JC |
606 | * @speed: The maximum speed, in bytes per second, or 0 for unlimited. |
607 | * @on_error: The action to take upon error. | |
608 | * @cb: Completion function for the job. | |
609 | * @opaque: Opaque pointer value passed to @cb. | |
54e26900 | 610 | * @backing_file_str: String to use as the backing file in @top's overlay |
747ff602 JC |
611 | * @errp: Error object. |
612 | * | |
613 | */ | |
614 | void commit_start(BlockDriverState *bs, BlockDriverState *base, | |
615 | BlockDriverState *top, int64_t speed, | |
097310b5 | 616 | BlockdevOnError on_error, BlockCompletionFunc *cb, |
54e26900 | 617 | void *opaque, const char *backing_file_str, Error **errp); |
03544a6e FZ |
618 | /** |
619 | * commit_active_start: | |
620 | * @bs: Active block device to be committed. | |
621 | * @base: Block device that will be written into, and become the new top. | |
622 | * @speed: The maximum speed, in bytes per second, or 0 for unlimited. | |
623 | * @on_error: The action to take upon error. | |
624 | * @cb: Completion function for the job. | |
625 | * @opaque: Opaque pointer value passed to @cb. | |
626 | * @errp: Error object. | |
627 | * | |
628 | */ | |
629 | void commit_active_start(BlockDriverState *bs, BlockDriverState *base, | |
630 | int64_t speed, | |
631 | BlockdevOnError on_error, | |
097310b5 | 632 | BlockCompletionFunc *cb, |
03544a6e | 633 | void *opaque, Error **errp); |
893f7eba PB |
634 | /* |
635 | * mirror_start: | |
636 | * @bs: Block device to operate on. | |
637 | * @target: Block device to write to. | |
09158f00 BC |
638 | * @replaces: Block graph node name to replace once the mirror is done. Can |
639 | * only be used when full mirroring is selected. | |
893f7eba | 640 | * @speed: The maximum speed, in bytes per second, or 0 for unlimited. |
eee13dfe | 641 | * @granularity: The chosen granularity for the dirty bitmap. |
08e4ed6c | 642 | * @buf_size: The amount of data that can be in flight at one time. |
893f7eba | 643 | * @mode: Whether to collapse all images in the chain to the target. |
b952b558 PB |
644 | * @on_source_error: The action to take upon error reading from the source. |
645 | * @on_target_error: The action to take upon error writing to the target. | |
0fc9f8ea | 646 | * @unmap: Whether to unmap target where source sectors only contain zeroes. |
893f7eba PB |
647 | * @cb: Completion function for the job. |
648 | * @opaque: Opaque pointer value passed to @cb. | |
649 | * @errp: Error object. | |
650 | * | |
651 | * Start a mirroring operation on @bs. Clusters that are allocated | |
652 | * in @bs will be written to @bs until the job is cancelled or | |
653 | * manually completed. At the end of a successful mirroring job, | |
654 | * @bs will be switched to read from @target. | |
655 | */ | |
656 | void mirror_start(BlockDriverState *bs, BlockDriverState *target, | |
09158f00 | 657 | const char *replaces, |
5fba6c0e | 658 | int64_t speed, uint32_t granularity, int64_t buf_size, |
08e4ed6c | 659 | MirrorSyncMode mode, BlockdevOnError on_source_error, |
b952b558 | 660 | BlockdevOnError on_target_error, |
0fc9f8ea | 661 | bool unmap, |
097310b5 | 662 | BlockCompletionFunc *cb, |
893f7eba PB |
663 | void *opaque, Error **errp); |
664 | ||
98d2c6f2 DM |
665 | /* |
666 | * backup_start: | |
667 | * @bs: Block device to operate on. | |
668 | * @target: Block device to write to. | |
669 | * @speed: The maximum speed, in bytes per second, or 0 for unlimited. | |
fc5d3f84 | 670 | * @sync_mode: What parts of the disk image should be copied to the destination. |
4b80ab2b | 671 | * @sync_bitmap: The dirty bitmap if sync_mode is MIRROR_SYNC_MODE_INCREMENTAL. |
98d2c6f2 DM |
672 | * @on_source_error: The action to take upon error reading from the source. |
673 | * @on_target_error: The action to take upon error writing to the target. | |
674 | * @cb: Completion function for the job. | |
675 | * @opaque: Opaque pointer value passed to @cb. | |
78f51fde | 676 | * @txn: Transaction that this job is part of (may be NULL). |
98d2c6f2 DM |
677 | * |
678 | * Start a backup operation on @bs. Clusters in @bs are written to @target | |
679 | * until the job is cancelled or manually completed. | |
680 | */ | |
681 | void backup_start(BlockDriverState *bs, BlockDriverState *target, | |
fc5d3f84 | 682 | int64_t speed, MirrorSyncMode sync_mode, |
d58d8453 | 683 | BdrvDirtyBitmap *sync_bitmap, |
fc5d3f84 | 684 | BlockdevOnError on_source_error, |
98d2c6f2 | 685 | BlockdevOnError on_target_error, |
097310b5 | 686 | BlockCompletionFunc *cb, void *opaque, |
78f51fde | 687 | BlockJobTxn *txn, Error **errp); |
98d2c6f2 | 688 | |
a2d61900 KW |
689 | void blk_set_bs(BlockBackend *blk, BlockDriverState *bs); |
690 | ||
a7f53e26 MA |
691 | void blk_dev_change_media_cb(BlockBackend *blk, bool load); |
692 | bool blk_dev_has_removable_media(BlockBackend *blk); | |
693 | void blk_dev_eject_request(BlockBackend *blk, bool force); | |
694 | bool blk_dev_is_tray_open(BlockBackend *blk); | |
695 | bool blk_dev_is_medium_locked(BlockBackend *blk); | |
696 | void blk_dev_resize_cb(BlockBackend *blk); | |
697 | ||
e0c47b6c | 698 | void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors); |
439db28c | 699 | bool bdrv_requests_pending(BlockDriverState *bs); |
e0c47b6c | 700 | |
df9a681d FZ |
701 | void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out); |
702 | void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in); | |
703 | ||
ea2384d3 | 704 | #endif /* BLOCK_INT_H */ |