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_COMMON_H 25 #define BLOCK_INT_COMMON_H 26 27 #include "block/aio.h" 28 #include "block/block-common.h" 29 #include "block/block-global-state.h" 30 #include "block/snapshot.h" 31 #include "qemu/iov.h" 32 #include "qemu/rcu.h" 33 #include "qemu/stats64.h" 34 35 #define BLOCK_FLAG_LAZY_REFCOUNTS 8 36 37 #define BLOCK_OPT_SIZE "size" 38 #define BLOCK_OPT_ENCRYPT "encryption" 39 #define BLOCK_OPT_ENCRYPT_FORMAT "encrypt.format" 40 #define BLOCK_OPT_COMPAT6 "compat6" 41 #define BLOCK_OPT_HWVERSION "hwversion" 42 #define BLOCK_OPT_BACKING_FILE "backing_file" 43 #define BLOCK_OPT_BACKING_FMT "backing_fmt" 44 #define BLOCK_OPT_CLUSTER_SIZE "cluster_size" 45 #define BLOCK_OPT_TABLE_SIZE "table_size" 46 #define BLOCK_OPT_PREALLOC "preallocation" 47 #define BLOCK_OPT_SUBFMT "subformat" 48 #define BLOCK_OPT_COMPAT_LEVEL "compat" 49 #define BLOCK_OPT_LAZY_REFCOUNTS "lazy_refcounts" 50 #define BLOCK_OPT_ADAPTER_TYPE "adapter_type" 51 #define BLOCK_OPT_REDUNDANCY "redundancy" 52 #define BLOCK_OPT_NOCOW "nocow" 53 #define BLOCK_OPT_EXTENT_SIZE_HINT "extent_size_hint" 54 #define BLOCK_OPT_OBJECT_SIZE "object_size" 55 #define BLOCK_OPT_REFCOUNT_BITS "refcount_bits" 56 #define BLOCK_OPT_DATA_FILE "data_file" 57 #define BLOCK_OPT_DATA_FILE_RAW "data_file_raw" 58 #define BLOCK_OPT_COMPRESSION_TYPE "compression_type" 59 #define BLOCK_OPT_EXTL2 "extended_l2" 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 BDRV_TRACKED_TRUNCATE, 68 }; 69 70 /* 71 * That is not quite good that BdrvTrackedRequest structure is public, 72 * as block/io.c is very careful about incoming offset/bytes being 73 * correct. Be sure to assert bdrv_check_request() succeeded after any 74 * modification of BdrvTrackedRequest object out of block/io.c 75 */ 76 typedef struct BdrvTrackedRequest { 77 BlockDriverState *bs; 78 int64_t offset; 79 int64_t bytes; 80 enum BdrvTrackedRequestType type; 81 82 bool serialising; 83 int64_t overlap_offset; 84 int64_t overlap_bytes; 85 86 QLIST_ENTRY(BdrvTrackedRequest) list; 87 Coroutine *co; /* owner, used for deadlock detection */ 88 CoQueue wait_queue; /* coroutines blocked on this request */ 89 90 struct BdrvTrackedRequest *waiting_for; 91 } BdrvTrackedRequest; 92 93 94 struct BlockDriver { 95 /* 96 * These fields are initialized when this object is created, 97 * and are never changed afterwards. 98 */ 99 100 const char *format_name; 101 int instance_size; 102 103 /* 104 * Set to true if the BlockDriver is a block filter. Block filters pass 105 * certain callbacks that refer to data (see block.c) to their bs->file 106 * or bs->backing (whichever one exists) if the driver doesn't implement 107 * them. Drivers that do not wish to forward must implement them and return 108 * -ENOTSUP. 109 * Note that filters are not allowed to modify data. 110 * 111 * Filters generally cannot have more than a single filtered child, 112 * because the data they present must at all times be the same as 113 * that on their filtered child. That would be impossible to 114 * achieve for multiple filtered children. 115 * (And this filtered child must then be bs->file or bs->backing.) 116 */ 117 bool is_filter; 118 /* 119 * Only make sense for filter drivers, for others must be false. 120 * If true, filtered child is bs->backing. Otherwise it's bs->file. 121 * Two internal filters use bs->backing as filtered child and has this 122 * field set to true: mirror_top and commit_top. There also two such test 123 * filters in tests/unit/test-bdrv-graph-mod.c. 124 * 125 * Never create any more such filters! 126 * 127 * TODO: imagine how to deprecate this behavior and make all filters work 128 * similarly using bs->file as filtered child. 129 */ 130 bool filtered_child_is_backing; 131 132 /* 133 * Set to true if the BlockDriver is a format driver. Format nodes 134 * generally do not expect their children to be other format nodes 135 * (except for backing files), and so format probing is disabled 136 * on those children. 137 */ 138 bool is_format; 139 140 /* 141 * Set to true if the BlockDriver supports zoned children. 142 */ 143 bool supports_zoned_children; 144 145 /* 146 * Drivers not implementing bdrv_parse_filename nor bdrv_open should have 147 * this field set to true, except ones that are defined only by their 148 * child's bs. 149 * An example of the last type will be the quorum block driver. 150 */ 151 bool bdrv_needs_filename; 152 153 /* 154 * Set if a driver can support backing files. This also implies the 155 * following semantics: 156 * 157 * - Return status 0 of .bdrv_co_block_status means that corresponding 158 * blocks are not allocated in this layer of backing-chain 159 * - For such (unallocated) blocks, read will: 160 * - fill buffer with zeros if there is no backing file 161 * - read from the backing file otherwise, where the block layer 162 * takes care of reading zeros beyond EOF if backing file is short 163 */ 164 bool supports_backing; 165 166 /* 167 * Drivers setting this field must be able to work with just a plain 168 * filename with '<protocol_name>:' as a prefix, and no other options. 169 * Options may be extracted from the filename by implementing 170 * bdrv_parse_filename. 171 */ 172 const char *protocol_name; 173 174 /* List of options for creating images, terminated by name == NULL */ 175 QemuOptsList *create_opts; 176 177 /* List of options for image amend */ 178 QemuOptsList *amend_opts; 179 180 /* 181 * If this driver supports reopening images this contains a 182 * NULL-terminated list of the runtime options that can be 183 * modified. If an option in this list is unspecified during 184 * reopen then it _must_ be reset to its default value or return 185 * an error. 186 */ 187 const char *const *mutable_opts; 188 189 /* 190 * Pointer to a NULL-terminated array of names of strong options 191 * that can be specified for bdrv_open(). A strong option is one 192 * that changes the data of a BDS. 193 * If this pointer is NULL, the array is considered empty. 194 * "filename" and "driver" are always considered strong. 195 */ 196 const char *const *strong_runtime_opts; 197 198 199 /* 200 * Global state (GS) API. These functions run under the BQL. 201 * 202 * See include/block/block-global-state.h for more information about 203 * the GS API. 204 */ 205 206 /* 207 * This function is invoked under BQL before .bdrv_co_amend() 208 * (which in contrast does not necessarily run under the BQL) 209 * to allow driver-specific initialization code that requires 210 * the BQL, like setting up specific permission flags. 211 */ 212 int GRAPH_RDLOCK_PTR (*bdrv_amend_pre_run)( 213 BlockDriverState *bs, Error **errp); 214 /* 215 * This function is invoked under BQL after .bdrv_co_amend() 216 * to allow cleaning up what was done in .bdrv_amend_pre_run(). 217 */ 218 void GRAPH_RDLOCK_PTR (*bdrv_amend_clean)(BlockDriverState *bs); 219 220 /* 221 * Return true if @to_replace can be replaced by a BDS with the 222 * same data as @bs without it affecting @bs's behavior (that is, 223 * without it being visible to @bs's parents). 224 */ 225 bool GRAPH_RDLOCK_PTR (*bdrv_recurse_can_replace)( 226 BlockDriverState *bs, BlockDriverState *to_replace); 227 228 int (*bdrv_probe_device)(const char *filename); 229 230 /* 231 * Any driver implementing this callback is expected to be able to handle 232 * NULL file names in its .bdrv_open() implementation. 233 */ 234 void (*bdrv_parse_filename)(const char *filename, QDict *options, 235 Error **errp); 236 237 /* For handling image reopen for split or non-split files. */ 238 int GRAPH_UNLOCKED_PTR (*bdrv_reopen_prepare)( 239 BDRVReopenState *reopen_state, BlockReopenQueue *queue, Error **errp); 240 void GRAPH_UNLOCKED_PTR (*bdrv_reopen_commit)( 241 BDRVReopenState *reopen_state); 242 void GRAPH_UNLOCKED_PTR (*bdrv_reopen_commit_post)( 243 BDRVReopenState *reopen_state); 244 void GRAPH_UNLOCKED_PTR (*bdrv_reopen_abort)( 245 BDRVReopenState *reopen_state); 246 void (*bdrv_join_options)(QDict *options, QDict *old_options); 247 248 int GRAPH_UNLOCKED_PTR (*bdrv_open)( 249 BlockDriverState *bs, QDict *options, int flags, Error **errp); 250 251 void (*bdrv_close)(BlockDriverState *bs); 252 253 int coroutine_fn GRAPH_UNLOCKED_PTR (*bdrv_co_create)( 254 BlockdevCreateOptions *opts, Error **errp); 255 256 int coroutine_fn GRAPH_UNLOCKED_PTR (*bdrv_co_create_opts)( 257 BlockDriver *drv, const char *filename, QemuOpts *opts, Error **errp); 258 259 int GRAPH_RDLOCK_PTR (*bdrv_amend_options)( 260 BlockDriverState *bs, QemuOpts *opts, 261 BlockDriverAmendStatusCB *status_cb, void *cb_opaque, 262 bool force, Error **errp); 263 264 int GRAPH_RDLOCK_PTR (*bdrv_make_empty)(BlockDriverState *bs); 265 266 /* 267 * Refreshes the bs->exact_filename field. If that is impossible, 268 * bs->exact_filename has to be left empty. 269 */ 270 void GRAPH_RDLOCK_PTR (*bdrv_refresh_filename)(BlockDriverState *bs); 271 272 /* 273 * Gathers the open options for all children into @target. 274 * A simple format driver (without backing file support) might 275 * implement this function like this: 276 * 277 * QINCREF(bs->file->bs->full_open_options); 278 * qdict_put(target, "file", bs->file->bs->full_open_options); 279 * 280 * If not specified, the generic implementation will simply put 281 * all children's options under their respective name. 282 * 283 * @backing_overridden is true when bs->backing seems not to be 284 * the child that would result from opening bs->backing_file. 285 * Therefore, if it is true, the backing child's options should be 286 * gathered; otherwise, there is no need since the backing child 287 * is the one implied by the image header. 288 * 289 * Note that ideally this function would not be needed. Every 290 * block driver which implements it is probably doing something 291 * shady regarding its runtime option structure. 292 */ 293 void GRAPH_RDLOCK_PTR (*bdrv_gather_child_options)( 294 BlockDriverState *bs, QDict *target, bool backing_overridden); 295 296 /* 297 * Returns an allocated string which is the directory name of this BDS: It 298 * will be used to make relative filenames absolute by prepending this 299 * function's return value to them. 300 */ 301 char * GRAPH_RDLOCK_PTR (*bdrv_dirname)(BlockDriverState *bs, Error **errp); 302 303 /* 304 * This informs the driver that we are no longer interested in the result 305 * of in-flight requests, so don't waste the time if possible. 306 * 307 * One example usage is to avoid waiting for an nbd target node reconnect 308 * timeout during job-cancel with force=true. 309 */ 310 void GRAPH_RDLOCK_PTR (*bdrv_cancel_in_flight)(BlockDriverState *bs); 311 312 int GRAPH_RDLOCK_PTR (*bdrv_inactivate)(BlockDriverState *bs); 313 314 int GRAPH_RDLOCK_PTR (*bdrv_snapshot_create)( 315 BlockDriverState *bs, QEMUSnapshotInfo *sn_info); 316 317 int GRAPH_UNLOCKED_PTR (*bdrv_snapshot_goto)( 318 BlockDriverState *bs, const char *snapshot_id); 319 320 int GRAPH_RDLOCK_PTR (*bdrv_snapshot_delete)( 321 BlockDriverState *bs, const char *snapshot_id, const char *name, 322 Error **errp); 323 324 int GRAPH_RDLOCK_PTR (*bdrv_snapshot_list)( 325 BlockDriverState *bs, QEMUSnapshotInfo **psn_info); 326 327 int GRAPH_RDLOCK_PTR (*bdrv_snapshot_load_tmp)( 328 BlockDriverState *bs, const char *snapshot_id, const char *name, 329 Error **errp); 330 331 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_change_backing_file)( 332 BlockDriverState *bs, const char *backing_file, 333 const char *backing_fmt); 334 335 /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */ 336 int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event, 337 const char *tag); 338 int (*bdrv_debug_remove_breakpoint)(BlockDriverState *bs, 339 const char *tag); 340 int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag); 341 bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag); 342 343 void GRAPH_RDLOCK_PTR (*bdrv_refresh_limits)( 344 BlockDriverState *bs, Error **errp); 345 346 /* 347 * Returns 1 if newly created images are guaranteed to contain only 348 * zeros, 0 otherwise. 349 */ 350 int GRAPH_RDLOCK_PTR (*bdrv_has_zero_init)(BlockDriverState *bs); 351 352 /* 353 * Remove fd handlers, timers, and other event loop callbacks so the event 354 * loop is no longer in use. Called with no in-flight requests and in 355 * depth-first traversal order with parents before child nodes. 356 */ 357 void (*bdrv_detach_aio_context)(BlockDriverState *bs); 358 359 /* 360 * Add fd handlers, timers, and other event loop callbacks so I/O requests 361 * can be processed again. Called with no in-flight requests and in 362 * depth-first traversal order with child nodes before parent nodes. 363 */ 364 void (*bdrv_attach_aio_context)(BlockDriverState *bs, 365 AioContext *new_context); 366 367 /** 368 * bdrv_drain_begin is called if implemented in the beginning of a 369 * drain operation to drain and stop any internal sources of requests in 370 * the driver. 371 * bdrv_drain_end is called if implemented at the end of the drain. 372 * 373 * They should be used by the driver to e.g. manage scheduled I/O 374 * requests, or toggle an internal state. After the end of the drain new 375 * requests will continue normally. 376 * 377 * Implementations of both functions must not call aio_poll(). 378 */ 379 void (*bdrv_drain_begin)(BlockDriverState *bs); 380 void (*bdrv_drain_end)(BlockDriverState *bs); 381 382 /** 383 * Try to get @bs's logical and physical block size. 384 * On success, store them in @bsz and return zero. 385 * On failure, return negative errno. 386 */ 387 int GRAPH_RDLOCK_PTR (*bdrv_probe_blocksizes)( 388 BlockDriverState *bs, BlockSizes *bsz); 389 /** 390 * Try to get @bs's geometry (cyls, heads, sectors) 391 * On success, store them in @geo and return 0. 392 * On failure return -errno. 393 * Only drivers that want to override guest geometry implement this 394 * callback; see hd_geometry_guess(). 395 */ 396 int GRAPH_RDLOCK_PTR (*bdrv_probe_geometry)( 397 BlockDriverState *bs, HDGeometry *geo); 398 399 /** 400 * Hot add a BDS's child. Used in combination with bdrv_del_child, so the 401 * user can take a child offline when it is broken and take a new child 402 * online. 403 * 404 * All block nodes must be drained. 405 */ 406 void GRAPH_WRLOCK_PTR (*bdrv_add_child)( 407 BlockDriverState *parent, BlockDriverState *child, Error **errp); 408 409 /** 410 * Hot remove a BDS's child. Used in combination with bdrv_add_child, so the 411 * user can take a child offline when it is broken and take a new child 412 * online. 413 * 414 * All block nodes must be drained. 415 */ 416 void GRAPH_WRLOCK_PTR (*bdrv_del_child)( 417 BlockDriverState *parent, BdrvChild *child, Error **errp); 418 419 /** 420 * Informs the block driver that a permission change is intended. The 421 * driver checks whether the change is permissible and may take other 422 * preparations for the change (e.g. get file system locks). This operation 423 * is always followed either by a call to either .bdrv_set_perm or 424 * .bdrv_abort_perm_update. 425 * 426 * Checks whether the requested set of cumulative permissions in @perm 427 * can be granted for accessing @bs and whether no other users are using 428 * permissions other than those given in @shared (both arguments take 429 * BLK_PERM_* bitmasks). 430 * 431 * If both conditions are met, 0 is returned. Otherwise, -errno is returned 432 * and errp is set to an error describing the conflict. 433 */ 434 int GRAPH_RDLOCK_PTR (*bdrv_check_perm)(BlockDriverState *bs, uint64_t perm, 435 uint64_t shared, Error **errp); 436 437 /** 438 * Called to inform the driver that the set of cumulative set of used 439 * permissions for @bs has changed to @perm, and the set of shareable 440 * permission to @shared. The driver can use this to propagate changes to 441 * its children (i.e. request permissions only if a parent actually needs 442 * them). 443 * 444 * This function is only invoked after bdrv_check_perm(), so block drivers 445 * may rely on preparations made in their .bdrv_check_perm implementation. 446 */ 447 void GRAPH_RDLOCK_PTR (*bdrv_set_perm)( 448 BlockDriverState *bs, uint64_t perm, uint64_t shared); 449 450 /* 451 * Called to inform the driver that after a previous bdrv_check_perm() 452 * call, the permission update is not performed and any preparations made 453 * for it (e.g. taken file locks) need to be undone. 454 * 455 * This function can be called even for nodes that never saw a 456 * bdrv_check_perm() call. It is a no-op then. 457 */ 458 void GRAPH_RDLOCK_PTR (*bdrv_abort_perm_update)(BlockDriverState *bs); 459 460 /** 461 * Returns in @nperm and @nshared the permissions that the driver for @bs 462 * needs on its child @c, based on the cumulative permissions requested by 463 * the parents in @parent_perm and @parent_shared. 464 * 465 * If @c is NULL, return the permissions for attaching a new child for the 466 * given @child_class and @role. 467 * 468 * If @reopen_queue is non-NULL, don't return the currently needed 469 * permissions, but those that will be needed after applying the 470 * @reopen_queue. 471 */ 472 void GRAPH_RDLOCK_PTR (*bdrv_child_perm)( 473 BlockDriverState *bs, BdrvChild *c, BdrvChildRole role, 474 BlockReopenQueue *reopen_queue, 475 uint64_t parent_perm, uint64_t parent_shared, 476 uint64_t *nperm, uint64_t *nshared); 477 478 /** 479 * Register/unregister a buffer for I/O. For example, when the driver is 480 * interested to know the memory areas that will later be used in iovs, so 481 * that it can do IOMMU mapping with VFIO etc., in order to get better 482 * performance. In the case of VFIO drivers, this callback is used to do 483 * DMA mapping for hot buffers. 484 * 485 * Returns: true on success, false on failure 486 */ 487 bool GRAPH_RDLOCK_PTR (*bdrv_register_buf)( 488 BlockDriverState *bs, void *host, size_t size, Error **errp); 489 void GRAPH_RDLOCK_PTR (*bdrv_unregister_buf)( 490 BlockDriverState *bs, void *host, size_t size); 491 492 /* 493 * This field is modified only under the BQL, and is part of 494 * the global state. 495 */ 496 QLIST_ENTRY(BlockDriver) list; 497 498 /* 499 * I/O API functions. These functions are thread-safe. 500 * 501 * See include/block/block-io.h for more information about 502 * the I/O API. 503 */ 504 505 int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename); 506 507 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_amend)( 508 BlockDriverState *bs, BlockdevAmendOptions *opts, bool force, 509 Error **errp); 510 511 /* aio */ 512 BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_preadv)(BlockDriverState *bs, 513 int64_t offset, int64_t bytes, QEMUIOVector *qiov, 514 BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque); 515 516 BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_pwritev)(BlockDriverState *bs, 517 int64_t offset, int64_t bytes, QEMUIOVector *qiov, 518 BdrvRequestFlags flags, BlockCompletionFunc *cb, void *opaque); 519 520 BlockAIOCB * GRAPH_RDLOCK_PTR (*bdrv_aio_flush)( 521 BlockDriverState *bs, BlockCompletionFunc *cb, void *opaque); 522 523 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_readv)(BlockDriverState *bs, 524 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov); 525 526 /** 527 * @offset: position in bytes to read at 528 * @bytes: number of bytes to read 529 * @qiov: the buffers to fill with read data 530 * @flags: currently unused, always 0 531 * 532 * @offset and @bytes will be a multiple of 'request_alignment', 533 * but the length of individual @qiov elements does not have to 534 * be a multiple. 535 * 536 * @bytes will always equal the total size of @qiov, and will be 537 * no larger than 'max_transfer'. 538 * 539 * The buffer in @qiov may point directly to guest memory. 540 */ 541 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv)(BlockDriverState *bs, 542 int64_t offset, int64_t bytes, QEMUIOVector *qiov, 543 BdrvRequestFlags flags); 544 545 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_part)( 546 BlockDriverState *bs, int64_t offset, int64_t bytes, 547 QEMUIOVector *qiov, size_t qiov_offset, 548 BdrvRequestFlags flags); 549 550 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_writev)(BlockDriverState *bs, 551 int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, 552 int flags); 553 /** 554 * @offset: position in bytes to write at 555 * @bytes: number of bytes to write 556 * @qiov: the buffers containing data to write 557 * @flags: zero or more bits allowed by 'supported_write_flags' 558 * 559 * @offset and @bytes will be a multiple of 'request_alignment', 560 * but the length of individual @qiov elements does not have to 561 * be a multiple. 562 * 563 * @bytes will always equal the total size of @qiov, and will be 564 * no larger than 'max_transfer'. 565 * 566 * The buffer in @qiov may point directly to guest memory. 567 */ 568 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev)( 569 BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov, 570 BdrvRequestFlags flags); 571 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_part)( 572 BlockDriverState *bs, int64_t offset, int64_t bytes, QEMUIOVector *qiov, 573 size_t qiov_offset, BdrvRequestFlags flags); 574 575 /* 576 * Efficiently zero a region of the disk image. Typically an image format 577 * would use a compact metadata representation to implement this. This 578 * function pointer may be NULL or return -ENOSUP and .bdrv_co_writev() 579 * will be called instead. 580 */ 581 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwrite_zeroes)( 582 BlockDriverState *bs, int64_t offset, int64_t bytes, 583 BdrvRequestFlags flags); 584 585 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard)( 586 BlockDriverState *bs, int64_t offset, int64_t bytes); 587 588 /* 589 * Map [offset, offset + nbytes) range onto a child of @bs to copy from, 590 * and invoke bdrv_co_copy_range_from(child, ...), or invoke 591 * bdrv_co_copy_range_to() if @bs is the leaf child to copy data from. 592 * 593 * See the comment of bdrv_co_copy_range for the parameter and return value 594 * semantics. 595 */ 596 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_from)( 597 BlockDriverState *bs, BdrvChild *src, int64_t offset, 598 BdrvChild *dst, int64_t dst_offset, int64_t bytes, 599 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags); 600 601 /* 602 * Map [offset, offset + nbytes) range onto a child of bs to copy data to, 603 * and invoke bdrv_co_copy_range_to(child, src, ...), or perform the copy 604 * operation if @bs is the leaf and @src has the same BlockDriver. Return 605 * -ENOTSUP if @bs is the leaf but @src has a different BlockDriver. 606 * 607 * See the comment of bdrv_co_copy_range for the parameter and return value 608 * semantics. 609 */ 610 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_copy_range_to)( 611 BlockDriverState *bs, BdrvChild *src, int64_t src_offset, 612 BdrvChild *dst, int64_t dst_offset, int64_t bytes, 613 BdrvRequestFlags read_flags, BdrvRequestFlags write_flags); 614 615 /* 616 * Building block for bdrv_block_status[_above] and 617 * bdrv_is_allocated[_above]. The driver should answer only 618 * according to the current layer, and should only need to set 619 * BDRV_BLOCK_DATA, BDRV_BLOCK_ZERO, BDRV_BLOCK_OFFSET_VALID, 620 * and/or BDRV_BLOCK_RAW; if the current layer defers to a backing 621 * layer, the result should be 0 (and not BDRV_BLOCK_ZERO). The 622 * caller will synthesize BDRV_BLOCK_ALLOCATED based on the 623 * non-zero results. See block.h for the overall meaning of the 624 * bits. As a hint, the flags in @mode may include a bitwise-or 625 * of BDRV_WANT_ALLOCATED, BDRV_WANT_OFFSET_VALID, or 626 * BDRV_WANT_ZERO based on what the caller is looking for in the 627 * results. The block layer guarantees input clamped to 628 * bdrv_getlength() and aligned to request_alignment, as well as 629 * non-NULL pnum, map, and file; in turn, the driver must return 630 * an error or set pnum to an aligned non-zero value. 631 * 632 * Note that @bytes is just a hint on how big of a region the 633 * caller wants to inspect. It is not a limit on *pnum. 634 * Implementations are free to return larger values of *pnum if 635 * doing so does not incur a performance penalty. 636 * 637 * block/io.c's bdrv_co_block_status() will utilize an unclamped 638 * *pnum value for the block-status cache on protocol nodes, prior 639 * to clamping *pnum for return to its caller. 640 */ 641 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_block_status)( 642 BlockDriverState *bs, unsigned int mode, 643 int64_t offset, int64_t bytes, int64_t *pnum, 644 int64_t *map, BlockDriverState **file); 645 646 /* 647 * Snapshot-access API. 648 * 649 * Block-driver may provide snapshot-access API: special functions to access 650 * some internal "snapshot". The functions are similar with normal 651 * read/block_status/discard handler, but don't have any specific handling 652 * in generic block-layer: no serializing, no alignment, no tracked 653 * requests. So, block-driver that realizes these APIs is fully responsible 654 * for synchronization between snapshot-access API and normal IO requests. 655 * 656 * TODO: To be able to support qcow2's internal snapshots, this API will 657 * need to be extended to: 658 * - be able to select a specific snapshot 659 * - receive the snapshot's actual length (which may differ from bs's 660 * length) 661 */ 662 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_preadv_snapshot)( 663 BlockDriverState *bs, int64_t offset, int64_t bytes, 664 QEMUIOVector *qiov, size_t qiov_offset); 665 666 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_snapshot_block_status)( 667 BlockDriverState *bs, unsigned int mode, int64_t offset, 668 int64_t bytes, int64_t *pnum, int64_t *map, BlockDriverState **file); 669 670 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pdiscard_snapshot)( 671 BlockDriverState *bs, int64_t offset, int64_t bytes); 672 673 /* 674 * Invalidate any cached meta-data. 675 */ 676 void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_invalidate_cache)( 677 BlockDriverState *bs, Error **errp); 678 679 /* 680 * Flushes all data for all layers by calling bdrv_co_flush for underlying 681 * layers, if needed. This function is needed for deterministic 682 * synchronization of the flush finishing callback. 683 */ 684 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush)(BlockDriverState *bs); 685 686 /* Delete a created file. */ 687 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_delete_file)( 688 BlockDriverState *bs, Error **errp); 689 690 /* 691 * Flushes all data that was already written to the OS all the way down to 692 * the disk (for example file-posix.c calls fsync()). 693 */ 694 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_disk)( 695 BlockDriverState *bs); 696 697 /* 698 * Flushes all internal caches to the OS. The data may still sit in a 699 * writeback cache of the host OS, but it will survive a crash of the qemu 700 * process. 701 */ 702 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_flush_to_os)( 703 BlockDriverState *bs); 704 705 /* 706 * Truncate @bs to @offset bytes using the given @prealloc mode 707 * when growing. Modes other than PREALLOC_MODE_OFF should be 708 * rejected when shrinking @bs. 709 * 710 * If @exact is true, @bs must be resized to exactly @offset. 711 * Otherwise, it is sufficient for @bs (if it is a host block 712 * device and thus there is no way to resize it) to be at least 713 * @offset bytes in length. 714 * 715 * If @exact is true and this function fails but would succeed 716 * with @exact = false, it should return -ENOTSUP. 717 */ 718 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_truncate)( 719 BlockDriverState *bs, int64_t offset, bool exact, 720 PreallocMode prealloc, BdrvRequestFlags flags, Error **errp); 721 722 int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_getlength)( 723 BlockDriverState *bs); 724 725 int64_t coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_allocated_file_size)( 726 BlockDriverState *bs); 727 728 BlockMeasureInfo *(*bdrv_measure)(QemuOpts *opts, BlockDriverState *in_bs, 729 Error **errp); 730 731 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed)( 732 BlockDriverState *bs, int64_t offset, int64_t bytes, 733 QEMUIOVector *qiov); 734 735 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_pwritev_compressed_part)( 736 BlockDriverState *bs, int64_t offset, int64_t bytes, 737 QEMUIOVector *qiov, size_t qiov_offset); 738 739 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_get_info)( 740 BlockDriverState *bs, BlockDriverInfo *bdi); 741 742 ImageInfoSpecific * GRAPH_RDLOCK_PTR (*bdrv_get_specific_info)( 743 BlockDriverState *bs, Error **errp); 744 BlockStatsSpecific *(*bdrv_get_specific_stats)(BlockDriverState *bs); 745 746 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_save_vmstate)( 747 BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos); 748 749 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_load_vmstate)( 750 BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos); 751 752 int coroutine_fn (*bdrv_co_zone_report)(BlockDriverState *bs, 753 int64_t offset, unsigned int *nr_zones, 754 BlockZoneDescriptor *zones); 755 int coroutine_fn (*bdrv_co_zone_mgmt)(BlockDriverState *bs, BlockZoneOp op, 756 int64_t offset, int64_t len); 757 int coroutine_fn (*bdrv_co_zone_append)(BlockDriverState *bs, 758 int64_t *offset, QEMUIOVector *qiov, 759 BdrvRequestFlags flags); 760 761 /* removable device specific */ 762 bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_is_inserted)( 763 BlockDriverState *bs); 764 void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_eject)( 765 BlockDriverState *bs, bool eject_flag); 766 void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_lock_medium)( 767 BlockDriverState *bs, bool locked); 768 769 /* to control generic scsi devices */ 770 BlockAIOCB *coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_aio_ioctl)( 771 BlockDriverState *bs, unsigned long int req, void *buf, 772 BlockCompletionFunc *cb, void *opaque); 773 774 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_ioctl)( 775 BlockDriverState *bs, unsigned long int req, void *buf); 776 777 /* 778 * Returns 0 for completed check, -errno for internal errors. 779 * The check results are stored in result. 780 */ 781 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_check)( 782 BlockDriverState *bs, BdrvCheckResult *result, BdrvCheckMode fix); 783 784 void coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_debug_event)( 785 BlockDriverState *bs, BlkdebugEvent event); 786 787 bool (*bdrv_supports_persistent_dirty_bitmap)(BlockDriverState *bs); 788 789 bool coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_can_store_new_dirty_bitmap)( 790 BlockDriverState *bs, const char *name, uint32_t granularity, 791 Error **errp); 792 793 int coroutine_fn GRAPH_RDLOCK_PTR (*bdrv_co_remove_persistent_dirty_bitmap)( 794 BlockDriverState *bs, const char *name, Error **errp); 795 }; 796 797 static inline bool TSA_NO_TSA block_driver_can_compress(BlockDriver *drv) 798 { 799 return drv->bdrv_co_pwritev_compressed || 800 drv->bdrv_co_pwritev_compressed_part; 801 } 802 803 typedef struct BlockLimits { 804 /* 805 * Alignment requirement, in bytes, for offset/length of I/O 806 * requests. Must be a power of 2 less than INT_MAX; defaults to 807 * 1 for drivers with modern byte interfaces, and to 512 808 * otherwise. 809 */ 810 uint32_t request_alignment; 811 812 /* 813 * Maximum number of bytes that can be discarded at once. Must be multiple 814 * of pdiscard_alignment, but need not be power of 2. May be 0 if no 815 * inherent 64-bit limit. 816 */ 817 int64_t max_pdiscard; 818 819 /* 820 * Optimal alignment for discard requests in bytes. A power of 2 821 * is best but not mandatory. Must be a multiple of 822 * bl.request_alignment, and must be less than max_pdiscard if 823 * that is set. May be 0 if bl.request_alignment is good enough 824 */ 825 uint32_t pdiscard_alignment; 826 827 /* 828 * Maximum number of bytes that can zeroized at once. Must be multiple of 829 * pwrite_zeroes_alignment. 0 means no limit. 830 */ 831 int64_t max_pwrite_zeroes; 832 833 /* 834 * Optimal alignment for write zeroes requests in bytes. A power 835 * of 2 is best but not mandatory. Must be a multiple of 836 * bl.request_alignment, and must be less than max_pwrite_zeroes 837 * if that is set. May be 0 if bl.request_alignment is good 838 * enough 839 */ 840 uint32_t pwrite_zeroes_alignment; 841 842 /* 843 * Optimal transfer length in bytes. A power of 2 is best but not 844 * mandatory. Must be a multiple of bl.request_alignment, or 0 if 845 * no preferred size 846 */ 847 uint32_t opt_transfer; 848 849 /* 850 * Maximal transfer length in bytes. Need not be power of 2, but 851 * must be multiple of opt_transfer and bl.request_alignment, or 0 852 * for no 32-bit limit. For now, anything larger than INT_MAX is 853 * clamped down. 854 */ 855 uint32_t max_transfer; 856 857 /* 858 * Maximal hardware transfer length in bytes. Applies whenever 859 * transfers to the device bypass the kernel I/O scheduler, for 860 * example with SG_IO. If larger than max_transfer or if zero, 861 * blk_get_max_hw_transfer will fall back to max_transfer. 862 */ 863 uint64_t max_hw_transfer; 864 865 /* 866 * Maximal number of scatter/gather elements allowed by the hardware. 867 * Applies whenever transfers to the device bypass the kernel I/O 868 * scheduler, for example with SG_IO. If larger than max_iov 869 * or if zero, blk_get_max_hw_iov will fall back to max_iov. 870 */ 871 int max_hw_iov; 872 873 874 /* memory alignment, in bytes so that no bounce buffer is needed */ 875 size_t min_mem_alignment; 876 877 /* memory alignment, in bytes, for bounce buffer */ 878 size_t opt_mem_alignment; 879 880 /* maximum number of iovec elements */ 881 int max_iov; 882 883 /* 884 * true if the length of the underlying file can change, and QEMU 885 * is expected to adjust automatically. Mostly for CD-ROM drives, 886 * whose length is zero when the tray is empty (they don't need 887 * an explicit monitor command to load the disk inside the guest). 888 */ 889 bool has_variable_length; 890 891 /* device zone model */ 892 BlockZoneModel zoned; 893 894 /* zone size expressed in bytes */ 895 uint32_t zone_size; 896 897 /* total number of zones */ 898 uint32_t nr_zones; 899 900 /* maximum sectors of a zone append write operation */ 901 uint32_t max_append_sectors; 902 903 /* maximum number of open zones */ 904 uint32_t max_open_zones; 905 906 /* maximum number of active zones */ 907 uint32_t max_active_zones; 908 909 uint32_t write_granularity; 910 } BlockLimits; 911 912 typedef struct BdrvOpBlocker BdrvOpBlocker; 913 914 typedef struct BdrvAioNotifier { 915 void (*attached_aio_context)(AioContext *new_context, void *opaque); 916 void (*detach_aio_context)(void *opaque); 917 918 void *opaque; 919 bool deleted; 920 921 QLIST_ENTRY(BdrvAioNotifier) list; 922 } BdrvAioNotifier; 923 924 struct BdrvChildClass { 925 /* 926 * If true, bdrv_replace_node() doesn't change the node this BdrvChild 927 * points to. 928 */ 929 bool stay_at_node; 930 931 /* 932 * If true, the parent is a BlockDriverState and bdrv_next_all_states() 933 * will return it. This information is used for drain_all, where every node 934 * will be drained separately, so the drain only needs to be propagated to 935 * non-BDS parents. 936 */ 937 bool parent_is_bds; 938 939 /* 940 * Global state (GS) API. These functions run under the BQL. 941 * 942 * See include/block/block-global-state.h for more information about 943 * the GS API. 944 */ 945 void (*inherit_options)(BdrvChildRole role, bool parent_is_format, 946 int *child_flags, QDict *child_options, 947 int parent_flags, QDict *parent_options); 948 void (*change_media)(BdrvChild *child, bool load); 949 950 /* 951 * Returns a malloced string that describes the parent of the child for a 952 * human reader. This could be a node-name, BlockBackend name, qdev ID or 953 * QOM path of the device owning the BlockBackend, job type and ID etc. The 954 * caller is responsible for freeing the memory. 955 */ 956 char *(*get_parent_desc)(BdrvChild *child); 957 958 /* 959 * Notifies the parent that the child has been activated/inactivated (e.g. 960 * when migration is completing) and it can start/stop requesting 961 * permissions and doing I/O on it. 962 */ 963 void GRAPH_RDLOCK_PTR (*activate)(BdrvChild *child, Error **errp); 964 int GRAPH_RDLOCK_PTR (*inactivate)(BdrvChild *child); 965 966 void GRAPH_WRLOCK_PTR (*attach)(BdrvChild *child); 967 void GRAPH_WRLOCK_PTR (*detach)(BdrvChild *child); 968 969 /* 970 * If this pair of functions is implemented, the parent doesn't issue new 971 * requests after returning from .drained_begin() until .drained_end() is 972 * called. 973 * 974 * These functions must not change the graph (and therefore also must not 975 * call aio_poll(), which could change the graph indirectly). 976 * 977 * Note that this can be nested. If drained_begin() was called twice, new 978 * I/O is allowed only after drained_end() was called twice, too. 979 */ 980 void GRAPH_RDLOCK_PTR (*drained_begin)(BdrvChild *child); 981 void GRAPH_RDLOCK_PTR (*drained_end)(BdrvChild *child); 982 983 /* 984 * Returns whether the parent has pending requests for the child. This 985 * callback is polled after .drained_begin() has been called until all 986 * activity on the child has stopped. 987 */ 988 bool GRAPH_RDLOCK_PTR (*drained_poll)(BdrvChild *child); 989 990 /* 991 * Notifies the parent that the filename of its child has changed (e.g. 992 * because the direct child was removed from the backing chain), so that it 993 * can update its reference. 994 */ 995 int (*update_filename)(BdrvChild *child, BlockDriverState *new_base, 996 const char *filename, 997 bool backing_mask_protocol, 998 Error **errp); 999 1000 /* 1001 * Notifies the parent that the child is trying to change its AioContext. 1002 * The parent may in turn change the AioContext of other nodes in the same 1003 * transaction. Returns true if the change is possible and the transaction 1004 * can be continued. Returns false and sets @errp if not and the transaction 1005 * must be aborted. 1006 * 1007 * @visited will accumulate all visited BdrvChild objects. The caller is 1008 * responsible for freeing the list afterwards. 1009 * 1010 * Must be called with the affected block nodes drained. 1011 */ 1012 bool GRAPH_RDLOCK_PTR (*change_aio_ctx)(BdrvChild *child, AioContext *ctx, 1013 GHashTable *visited, 1014 Transaction *tran, Error **errp); 1015 1016 /* 1017 * I/O API functions. These functions are thread-safe. 1018 * 1019 * See include/block/block-io.h for more information about 1020 * the I/O API. 1021 */ 1022 1023 void (*resize)(BdrvChild *child); 1024 1025 /* 1026 * Returns a name that is supposedly more useful for human users than the 1027 * node name for identifying the node in question (in particular, a BB 1028 * name), or NULL if the parent can't provide a better name. 1029 */ 1030 const char *(*get_name)(BdrvChild *child); 1031 1032 AioContext *(*get_parent_aio_context)(BdrvChild *child); 1033 }; 1034 1035 extern const BdrvChildClass child_of_bds; 1036 1037 struct BdrvChild { 1038 BlockDriverState *bs; 1039 char *name; 1040 const BdrvChildClass *klass; 1041 BdrvChildRole role; 1042 void *opaque; 1043 1044 /** 1045 * Granted permissions for operating on this BdrvChild (BLK_PERM_* bitmask) 1046 */ 1047 uint64_t perm; 1048 1049 /** 1050 * Permissions that can still be granted to other users of @bs while this 1051 * BdrvChild is still attached to it. (BLK_PERM_* bitmask) 1052 */ 1053 uint64_t shared_perm; 1054 1055 /* 1056 * This link is frozen: the child can neither be replaced nor 1057 * detached from the parent. 1058 */ 1059 bool frozen; 1060 1061 /* 1062 * True if the parent of this child has been drained by this BdrvChild 1063 * (through klass->drained_*). 1064 * 1065 * It is generally true if bs->quiesce_counter > 0. It may differ while the 1066 * child is entering or leaving a drained section. 1067 */ 1068 bool quiesced_parent; 1069 1070 QLIST_ENTRY(BdrvChild GRAPH_RDLOCK_PTR) next; 1071 QLIST_ENTRY(BdrvChild GRAPH_RDLOCK_PTR) next_parent; 1072 }; 1073 1074 /* 1075 * Allows bdrv_co_block_status() to cache one data region for a 1076 * protocol node. 1077 * 1078 * @valid: Whether the cache is valid (should be accessed with atomic 1079 * functions so this can be reset by RCU readers) 1080 * @data_start: Offset where we know (or strongly assume) is data 1081 * @data_end: Offset where the data region ends (which is not necessarily 1082 * the start of a zeroed region) 1083 */ 1084 typedef struct BdrvBlockStatusCache { 1085 struct rcu_head rcu; 1086 1087 bool valid; 1088 int64_t data_start; 1089 int64_t data_end; 1090 } BdrvBlockStatusCache; 1091 1092 struct BlockDriverState { 1093 /* 1094 * Protected by big QEMU lock or read-only after opening. No special 1095 * locking needed during I/O... 1096 */ 1097 int open_flags; /* flags used to open the file, re-used for re-open */ 1098 bool encrypted; /* if true, the media is encrypted */ 1099 bool sg; /* if true, the device is a /dev/sg* */ 1100 bool probed; /* if true, format was probed rather than specified */ 1101 bool force_share; /* if true, always allow all shared permissions */ 1102 bool implicit; /* if true, this filter node was automatically inserted */ 1103 1104 BlockDriver *drv; /* NULL means no media */ 1105 void *opaque; 1106 1107 AioContext *aio_context; /* event loop used for fd handlers, timers, etc */ 1108 /* 1109 * long-running tasks intended to always use the same AioContext as this 1110 * BDS may register themselves in this list to be notified of changes 1111 * regarding this BDS's context 1112 */ 1113 QLIST_HEAD(, BdrvAioNotifier) aio_notifiers; 1114 bool walking_aio_notifiers; /* to make removal during iteration safe */ 1115 1116 char filename[PATH_MAX]; 1117 /* 1118 * If not empty, this image is a diff in relation to backing_file. 1119 * Note that this is the name given in the image header and 1120 * therefore may or may not be equal to .backing->bs->filename. 1121 * If this field contains a relative path, it is to be resolved 1122 * relatively to the overlay's location. 1123 */ 1124 char backing_file[PATH_MAX]; 1125 /* 1126 * The backing filename indicated by the image header. Contrary 1127 * to backing_file, if we ever open this file, auto_backing_file 1128 * is replaced by the resulting BDS's filename (i.e. after a 1129 * bdrv_refresh_filename() run). 1130 */ 1131 char auto_backing_file[PATH_MAX]; 1132 char backing_format[16]; /* if non-zero and backing_file exists */ 1133 1134 QDict *full_open_options; 1135 char exact_filename[PATH_MAX]; 1136 1137 /* I/O Limits */ 1138 BlockLimits bl; 1139 1140 /* 1141 * Flags honored during pread 1142 */ 1143 BdrvRequestFlags supported_read_flags; 1144 /* 1145 * Flags honored during pwrite (so far: BDRV_REQ_FUA, 1146 * BDRV_REQ_WRITE_UNCHANGED). 1147 * If a driver does not support BDRV_REQ_WRITE_UNCHANGED, those 1148 * writes will be issued as normal writes without the flag set. 1149 * This is important to note for drivers that do not explicitly 1150 * request a WRITE permission for their children and instead take 1151 * the same permissions as their parent did (this is commonly what 1152 * block filters do). Such drivers have to be aware that the 1153 * parent may have taken a WRITE_UNCHANGED permission only and is 1154 * issuing such requests. Drivers either must make sure that 1155 * these requests do not result in plain WRITE accesses (usually 1156 * by supporting BDRV_REQ_WRITE_UNCHANGED, and then forwarding 1157 * every incoming write request as-is, including potentially that 1158 * flag), or they have to explicitly take the WRITE permission for 1159 * their children. 1160 */ 1161 BdrvRequestFlags supported_write_flags; 1162 /* 1163 * Flags honored during pwrite_zeroes (so far: BDRV_REQ_FUA, 1164 * BDRV_REQ_MAY_UNMAP, BDRV_REQ_WRITE_UNCHANGED) 1165 */ 1166 BdrvRequestFlags supported_zero_flags; 1167 /* 1168 * Flags honoured during truncate (so far: BDRV_REQ_ZERO_WRITE). 1169 * 1170 * If BDRV_REQ_ZERO_WRITE is given, the truncate operation must make sure 1171 * that any added space reads as all zeros. If this can't be guaranteed, 1172 * the operation must fail. 1173 */ 1174 BdrvRequestFlags supported_truncate_flags; 1175 1176 /* the following member gives a name to every node on the bs graph. */ 1177 char node_name[32]; 1178 /* element of the list of named nodes building the graph */ 1179 QTAILQ_ENTRY(BlockDriverState) node_list; 1180 /* element of the list of all BlockDriverStates (all_bdrv_states) */ 1181 QTAILQ_ENTRY(BlockDriverState) bs_list; 1182 /* element of the list of monitor-owned BDS */ 1183 QTAILQ_ENTRY(BlockDriverState) monitor_list; 1184 int refcnt; 1185 1186 /* operation blockers. Protected by BQL. */ 1187 QLIST_HEAD(, BdrvOpBlocker) op_blockers[BLOCK_OP_TYPE_MAX]; 1188 1189 /* 1190 * The node that this node inherited default options from (and a reopen on 1191 * which can affect this node by changing these defaults). This is always a 1192 * parent node of this node. 1193 */ 1194 BlockDriverState *inherits_from; 1195 1196 /* 1197 * @backing and @file are some of @children or NULL. All these three fields 1198 * (@file, @backing and @children) are modified only in 1199 * bdrv_child_cb_attach() and bdrv_child_cb_detach(). 1200 * 1201 * See also comment in include/block/block.h, to learn how backing and file 1202 * are connected with BdrvChildRole. 1203 */ 1204 QLIST_HEAD(, BdrvChild GRAPH_RDLOCK_PTR) children; 1205 BdrvChild * GRAPH_RDLOCK_PTR backing; 1206 BdrvChild * GRAPH_RDLOCK_PTR file; 1207 1208 QLIST_HEAD(, BdrvChild GRAPH_RDLOCK_PTR) parents; 1209 1210 QDict *options; 1211 QDict *explicit_options; 1212 BlockdevDetectZeroesOptions detect_zeroes; 1213 1214 /* The error object in use for blocking operations on backing_hd */ 1215 Error *backing_blocker; 1216 1217 /* 1218 * If we are reading a disk image, give its size in sectors. 1219 * Generally read-only; it is written to by load_snapshot and 1220 * save_snaphost, but the block layer is quiescent during those. 1221 */ 1222 int64_t total_sectors; 1223 1224 /* threshold limit for writes, in bytes. "High water mark". */ 1225 uint64_t write_threshold_offset; 1226 1227 /* 1228 * Writing to the list requires the BQL _and_ the dirty_bitmap_mutex. 1229 * Reading from the list can be done with either the BQL or the 1230 * dirty_bitmap_mutex. Modifying a bitmap only requires 1231 * dirty_bitmap_mutex. 1232 */ 1233 QemuMutex dirty_bitmap_mutex; 1234 QLIST_HEAD(, BdrvDirtyBitmap) dirty_bitmaps; 1235 1236 /* Offset after the highest byte written to */ 1237 Stat64 wr_highest_offset; 1238 1239 /* 1240 * If true, copy read backing sectors into image. Can be >1 if more 1241 * than one client has requested copy-on-read. Accessed with atomic 1242 * ops. 1243 */ 1244 int copy_on_read; 1245 1246 /* 1247 * number of in-flight requests; overall and serialising. 1248 * Accessed with atomic ops. 1249 */ 1250 unsigned int in_flight; 1251 unsigned int serialising_in_flight; 1252 1253 /* do we need to tell the quest if we have a volatile write cache? */ 1254 int enable_write_cache; 1255 1256 /* Accessed with atomic ops. */ 1257 int quiesce_counter; 1258 1259 unsigned int write_gen; /* Current data generation */ 1260 1261 /* Protected by reqs_lock. */ 1262 QemuMutex reqs_lock; 1263 QLIST_HEAD(, BdrvTrackedRequest) tracked_requests; 1264 CoQueue flush_queue; /* Serializing flush queue */ 1265 bool active_flush_req; /* Flush request in flight? */ 1266 1267 /* Only read/written by whoever has set active_flush_req to true. */ 1268 unsigned int flushed_gen; /* Flushed write generation */ 1269 1270 /* BdrvChild links to this node may never be frozen */ 1271 bool never_freeze; 1272 1273 /* Lock for block-status cache RCU writers */ 1274 CoMutex bsc_modify_lock; 1275 /* Always non-NULL, but must only be dereferenced under an RCU read guard */ 1276 BdrvBlockStatusCache *block_status_cache; 1277 1278 /* array of write pointers' location of each zone in the zoned device. */ 1279 BlockZoneWps *wps; 1280 }; 1281 1282 struct BlockBackendRootState { 1283 int open_flags; 1284 BlockdevDetectZeroesOptions detect_zeroes; 1285 }; 1286 1287 typedef enum BlockMirrorBackingMode { 1288 /* 1289 * Reuse the existing backing chain from the source for the target. 1290 * - sync=full: Set backing BDS to NULL. 1291 * - sync=top: Use source's backing BDS. 1292 * - sync=none: Use source as the backing BDS. 1293 */ 1294 MIRROR_SOURCE_BACKING_CHAIN, 1295 1296 /* Open the target's backing chain completely anew */ 1297 MIRROR_OPEN_BACKING_CHAIN, 1298 1299 /* Do not change the target's backing BDS after job completion */ 1300 MIRROR_LEAVE_BACKING_CHAIN, 1301 } BlockMirrorBackingMode; 1302 1303 1304 /* 1305 * Essential block drivers which must always be statically linked into qemu, and 1306 * which therefore can be accessed without using bdrv_find_format() 1307 */ 1308 extern BlockDriver bdrv_file; 1309 extern BlockDriver bdrv_raw; 1310 extern BlockDriver bdrv_qcow2; 1311 1312 extern unsigned int bdrv_drain_all_count; 1313 extern QemuOptsList bdrv_create_opts_simple; 1314 1315 /* 1316 * Common functions that are neither I/O nor Global State. 1317 * 1318 * See include/block/block-common.h for more information about 1319 * the Common API. 1320 */ 1321 1322 static inline BlockDriverState *child_bs(BdrvChild *child) 1323 { 1324 return child ? child->bs : NULL; 1325 } 1326 1327 int bdrv_check_request(int64_t offset, int64_t bytes, Error **errp); 1328 char *create_tmp_file(Error **errp); 1329 void bdrv_parse_filename_strip_prefix(const char *filename, const char *prefix, 1330 QDict *options); 1331 1332 1333 int bdrv_check_qiov_request(int64_t offset, int64_t bytes, 1334 QEMUIOVector *qiov, size_t qiov_offset, 1335 Error **errp); 1336 1337 #ifdef _WIN32 1338 int is_windows_drive(const char *filename); 1339 #endif 1340 1341 #endif /* BLOCK_INT_COMMON_H */ 1342