include/block/block.h

   1 #ifndef BLOCK_H
   2 #define BLOCK_H
   3
   4 #include "block/aio.h"
   5 #include "qapi/qapi-types-block-core.h"
   6 #include "block/aio-wait.h"
   7 #include "qemu/iov.h"
   8 #include "qemu/coroutine.h"
   9 #include "block/accounting.h"
  10 #include "block/dirty-bitmap.h"
  11 #include "block/blockjob.h"
  12 #include "qemu/hbitmap.h"
  13
  14 /* block.c */
  15 typedef struct BlockDriver BlockDriver;
  16 typedef struct BdrvChild BdrvChild;
  17 typedef struct BdrvChildRole BdrvChildRole;
  18
  19 typedef struct BlockDriverInfo {
  20     /* in bytes, 0 if irrelevant */
  21     int cluster_size;
  22     /* offset at which the VM state can be saved (0 if not possible) */
  23     int64_t vm_state_offset;
  24     bool is_dirty;
  25     /*
  26      * True if unallocated blocks read back as zeroes. This is equivalent
  27      * to the LBPRZ flag in the SCSI logical block provisioning page.
  28      */
  29     bool unallocated_blocks_are_zero;
  30     /*
  31      * True if this block driver only supports compressed writes
  32      */
  33     bool needs_compressed_writes;
  34 } BlockDriverInfo;
  35
  36 typedef struct BlockFragInfo {
  37     uint64_t allocated_clusters;
  38     uint64_t total_clusters;
  39     uint64_t fragmented_clusters;
  40     uint64_t compressed_clusters;
  41 } BlockFragInfo;
  42
  43 typedef enum {
  44     BDRV_REQ_COPY_ON_READ       = 0x1,
  45     BDRV_REQ_ZERO_WRITE         = 0x2,
  46     /* The BDRV_REQ_MAY_UNMAP flag is used to indicate that the block driver
  47      * is allowed to optimize a write zeroes request by unmapping (discarding)
  48      * blocks if it is guaranteed that the result will read back as
  49      * zeroes. The flag is only passed to the driver if the block device is
  50      * opened with BDRV_O_UNMAP.
  51      */
  52     BDRV_REQ_MAY_UNMAP          = 0x4,
  53     BDRV_REQ_NO_SERIALISING     = 0x8,
  54     BDRV_REQ_FUA                = 0x10,
  55     BDRV_REQ_WRITE_COMPRESSED   = 0x20,
  56
  57     /* Signifies that this write request will not change the visible disk
  58      * content. */
  59     BDRV_REQ_WRITE_UNCHANGED    = 0x40,
  60
  61     /* Mask of valid flags */
  62     BDRV_REQ_MASK               = 0x7f,
  63 } BdrvRequestFlags;
  64
  65 typedef struct BlockSizes {
  66     uint32_t phys;
  67     uint32_t log;
  68 } BlockSizes;
  69
  70 typedef struct HDGeometry {
  71     uint32_t heads;
  72     uint32_t sectors;
  73     uint32_t cylinders;
  74 } HDGeometry;
  75
  76 #define BDRV_O_RDWR        0x0002
  77 #define BDRV_O_RESIZE      0x0004 /* request permission for resizing the node */
  78 #define BDRV_O_SNAPSHOT    0x0008 /* open the file read only and save writes in a snapshot */
  79 #define BDRV_O_TEMPORARY   0x0010 /* delete the file after use */
  80 #define BDRV_O_NOCACHE     0x0020 /* do not use the host page cache */
  81 #define BDRV_O_NATIVE_AIO  0x0080 /* use native AIO instead of the thread pool */
  82 #define BDRV_O_NO_BACKING  0x0100 /* don't open the backing file */
  83 #define BDRV_O_NO_FLUSH    0x0200 /* disable flushing on this disk */
  84 #define BDRV_O_COPY_ON_READ 0x0400 /* copy read backing sectors into image */
  85 #define BDRV_O_INACTIVE    0x0800  /* consistency hint for migration handoff */
  86 #define BDRV_O_CHECK       0x1000  /* open solely for consistency check */
  87 #define BDRV_O_ALLOW_RDWR  0x2000  /* allow reopen to change from r/o to r/w */
  88 #define BDRV_O_UNMAP       0x4000  /* execute guest UNMAP/TRIM operations */
  89 #define BDRV_O_PROTOCOL    0x8000  /* if no block driver is explicitly given:
  90                                       select an appropriate protocol driver,
  91                                       ignoring the format layer */
  92 #define BDRV_O_NO_IO       0x10000 /* don't initialize for I/O */
  93
  94 #define BDRV_O_CACHE_MASK  (BDRV_O_NOCACHE | BDRV_O_NO_FLUSH)
  95
  96
  97 /* Option names of options parsed by the block layer */
  98
  99 #define BDRV_OPT_CACHE_WB       "cache.writeback"
 100 #define BDRV_OPT_CACHE_DIRECT   "cache.direct"
 101 #define BDRV_OPT_CACHE_NO_FLUSH "cache.no-flush"
 102 #define BDRV_OPT_READ_ONLY      "read-only"
 103 #define BDRV_OPT_DISCARD        "discard"
 104 #define BDRV_OPT_FORCE_SHARE    "force-share"
 105
 106
 107 #define BDRV_SECTOR_BITS   9
 108 #define BDRV_SECTOR_SIZE   (1ULL << BDRV_SECTOR_BITS)
 109 #define BDRV_SECTOR_MASK   ~(BDRV_SECTOR_SIZE - 1)
 110
 111 #define BDRV_REQUEST_MAX_SECTORS MIN(SIZE_MAX >> BDRV_SECTOR_BITS, \
 112                                      INT_MAX >> BDRV_SECTOR_BITS)
 113 #define BDRV_REQUEST_MAX_BYTES (BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS)
 114
 115 /*
 116  * Allocation status flags for bdrv_block_status() and friends.
 117  *
 118  * Public flags:
 119  * BDRV_BLOCK_DATA: allocation for data at offset is tied to this layer
 120  * BDRV_BLOCK_ZERO: offset reads as zero
 121  * BDRV_BLOCK_OFFSET_VALID: an associated offset exists for accessing raw data
 122  * BDRV_BLOCK_ALLOCATED: the content of the block is determined by this
 123  *                       layer rather than any backing, set by block layer
 124  * BDRV_BLOCK_EOF: the returned pnum covers through end of file for this
 125  *                 layer, set by block layer
 126  *
 127  * Internal flag:
 128  * BDRV_BLOCK_RAW: for use by passthrough drivers, such as raw, to request
 129  *                 that the block layer recompute the answer from the returned
 130  *                 BDS; must be accompanied by just BDRV_BLOCK_OFFSET_VALID.
 131  *
 132  * If BDRV_BLOCK_OFFSET_VALID is set, the map parameter represents the
 133  * host offset within the returned BDS that is allocated for the
 134  * corresponding raw guest data.  However, whether that offset
 135  * actually contains data also depends on BDRV_BLOCK_DATA, as follows:
 136  *
 137  * DATA ZERO OFFSET_VALID
 138  *  t    t        t       sectors read as zero, returned file is zero at offset
 139  *  t    f        t       sectors read as valid from file at offset
 140  *  f    t        t       sectors preallocated, read as zero, returned file not
 141  *                        necessarily zero at offset
 142  *  f    f        t       sectors preallocated but read from backing_hd,
 143  *                        returned file contains garbage at offset
 144  *  t    t        f       sectors preallocated, read as zero, unknown offset
 145  *  t    f        f       sectors read from unknown file or offset
 146  *  f    t        f       not allocated or unknown offset, read as zero
 147  *  f    f        f       not allocated or unknown offset, read from backing_hd
 148  */
 149 #define BDRV_BLOCK_DATA         0x01
 150 #define BDRV_BLOCK_ZERO         0x02
 151 #define BDRV_BLOCK_OFFSET_VALID 0x04
 152 #define BDRV_BLOCK_RAW          0x08
 153 #define BDRV_BLOCK_ALLOCATED    0x10
 154 #define BDRV_BLOCK_EOF          0x20
 155 #define BDRV_BLOCK_OFFSET_MASK  BDRV_SECTOR_MASK
 156
 157 typedef QSIMPLEQ_HEAD(BlockReopenQueue, BlockReopenQueueEntry) BlockReopenQueue;
 158
 159 typedef struct BDRVReopenState {
 160     BlockDriverState *bs;
 161     int flags;
 162     uint64_t perm, shared_perm;
 163     QDict *options;
 164     QDict *explicit_options;
 165     void *opaque;
 166 } BDRVReopenState;
 167
 168 /*
 169  * Block operation types
 170  */
 171 typedef enum BlockOpType {
 172     BLOCK_OP_TYPE_BACKUP_SOURCE,
 173     BLOCK_OP_TYPE_BACKUP_TARGET,
 174     BLOCK_OP_TYPE_CHANGE,
 175     BLOCK_OP_TYPE_COMMIT_SOURCE,
 176     BLOCK_OP_TYPE_COMMIT_TARGET,
 177     BLOCK_OP_TYPE_DATAPLANE,
 178     BLOCK_OP_TYPE_DRIVE_DEL,
 179     BLOCK_OP_TYPE_EJECT,
 180     BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
 181     BLOCK_OP_TYPE_INTERNAL_SNAPSHOT,
 182     BLOCK_OP_TYPE_INTERNAL_SNAPSHOT_DELETE,
 183     BLOCK_OP_TYPE_MIRROR_SOURCE,
 184     BLOCK_OP_TYPE_MIRROR_TARGET,
 185     BLOCK_OP_TYPE_RESIZE,
 186     BLOCK_OP_TYPE_STREAM,
 187     BLOCK_OP_TYPE_REPLACE,
 188     BLOCK_OP_TYPE_MAX,
 189 } BlockOpType;
 190
 191 /* Block node permission constants */
 192 enum {
 193     /**
 194      * A user that has the "permission" of consistent reads is guaranteed that
 195      * their view of the contents of the block device is complete and
 196      * self-consistent, representing the contents of a disk at a specific
 197      * point.
 198      *
 199      * For most block devices (including their backing files) this is true, but
 200      * the property cannot be maintained in a few situations like for
 201      * intermediate nodes of a commit block job.
 202      */
 203     BLK_PERM_CONSISTENT_READ    = 0x01,
 204
 205     /** This permission is required to change the visible disk contents. */
 206     BLK_PERM_WRITE              = 0x02,
 207
 208     /**
 209      * This permission (which is weaker than BLK_PERM_WRITE) is both enough and
 210      * required for writes to the block node when the caller promises that
 211      * the visible disk content doesn't change.
 212      *
 213      * As the BLK_PERM_WRITE permission is strictly stronger, either is
 214      * sufficient to perform an unchanging write.
 215      */
 216     BLK_PERM_WRITE_UNCHANGED    = 0x04,
 217
 218     /** This permission is required to change the size of a block node. */
 219     BLK_PERM_RESIZE             = 0x08,
 220
 221     /**
 222      * This permission is required to change the node that this BdrvChild
 223      * points to.
 224      */
 225     BLK_PERM_GRAPH_MOD          = 0x10,
 226
 227     BLK_PERM_ALL                = 0x1f,
 228 };
 229
 230 char *bdrv_perm_names(uint64_t perm);
 231
 232 /* disk I/O throttling */
 233 void bdrv_init(void);
 234 void bdrv_init_with_whitelist(void);
 235 bool bdrv_uses_whitelist(void);
 236 int bdrv_is_whitelisted(BlockDriver *drv, bool read_only);
 237 BlockDriver *bdrv_find_protocol(const char *filename,
 238                                 bool allow_protocol_prefix,
 239                                 Error **errp);
 240 BlockDriver *bdrv_find_format(const char *format_name);
 241 int bdrv_create(BlockDriver *drv, const char* filename,
 242                 QemuOpts *opts, Error **errp);
 243 int bdrv_create_file(const char *filename, QemuOpts *opts, Error **errp);
 244 BlockDriverState *bdrv_new(void);
 245 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
 246                  Error **errp);
 247 void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,
 248                        Error **errp);
 249
 250 int bdrv_parse_cache_mode(const char *mode, int *flags, bool *writethrough);
 251 int bdrv_parse_discard_flags(const char *mode, int *flags);
 252 BdrvChild *bdrv_open_child(const char *filename,
 253                            QDict *options, const char *bdref_key,
 254                            BlockDriverState* parent,
 255                            const BdrvChildRole *child_role,
 256                            bool allow_none, Error **errp);
 257 BlockDriverState *bdrv_open_blockdev_ref(BlockdevRef *ref, Error **errp);
 258 void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
 259                          Error **errp);
 260 int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
 261                            const char *bdref_key, Error **errp);
 262 BlockDriverState *bdrv_open(const char *filename, const char *reference,
 263                             QDict *options, int flags, Error **errp);
 264 BlockDriverState *bdrv_new_open_driver(BlockDriver *drv, const char *node_name,
 265                                        int flags, Error **errp);
 266 BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
 267                                     BlockDriverState *bs,
 268                                     QDict *options, int flags);
 269 int bdrv_reopen_multiple(AioContext *ctx, BlockReopenQueue *bs_queue, Error **errp);
 270 int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp);
 271 int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
 272                         BlockReopenQueue *queue, Error **errp);
 273 void bdrv_reopen_commit(BDRVReopenState *reopen_state);
 274 void bdrv_reopen_abort(BDRVReopenState *reopen_state);
 275 int bdrv_read(BdrvChild *child, int64_t sector_num,
 276               uint8_t *buf, int nb_sectors);
 277 int bdrv_write(BdrvChild *child, int64_t sector_num,
 278                const uint8_t *buf, int nb_sectors);
 279 int bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset,
 280                        int bytes, BdrvRequestFlags flags);
 281 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags);
 282 int bdrv_pread(BdrvChild *child, int64_t offset, void *buf, int bytes);
 283 int bdrv_preadv(BdrvChild *child, int64_t offset, QEMUIOVector *qiov);
 284 int bdrv_pwrite(BdrvChild *child, int64_t offset, const void *buf, int bytes);
 285 int bdrv_pwritev(BdrvChild *child, int64_t offset, QEMUIOVector *qiov);
 286 int bdrv_pwrite_sync(BdrvChild *child, int64_t offset,
 287                      const void *buf, int count);
 288 int coroutine_fn bdrv_co_readv(BdrvChild *child, int64_t sector_num,
 289                                int nb_sectors, QEMUIOVector *qiov);
 290 int coroutine_fn bdrv_co_writev(BdrvChild *child, int64_t sector_num,
 291                                int nb_sectors, QEMUIOVector *qiov);
 292 /*
 293  * Efficiently zero a region of the disk image.  Note that this is a regular
 294  * I/O request like read or write and should have a reasonable size.  This
 295  * function is not suitable for zeroing the entire image in a single request
 296  * because it may allocate memory for the entire region.
 297  */
 298 int coroutine_fn bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset,
 299                                        int bytes, BdrvRequestFlags flags);
 300 BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs,
 301     const char *backing_file);
 302 void bdrv_refresh_filename(BlockDriverState *bs);
 303 int bdrv_truncate(BdrvChild *child, int64_t offset, PreallocMode prealloc,
 304                   Error **errp);
 305 int64_t bdrv_nb_sectors(BlockDriverState *bs);
 306 int64_t bdrv_getlength(BlockDriverState *bs);
 307 int64_t bdrv_get_allocated_file_size(BlockDriverState *bs);
 308 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
 309                                BlockDriverState *in_bs, Error **errp);
 310 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr);
 311 void bdrv_refresh_limits(BlockDriverState *bs, Error **errp);
 312 int bdrv_commit(BlockDriverState *bs);
 313 int bdrv_change_backing_file(BlockDriverState *bs,
 314     const char *backing_file, const char *backing_fmt);
 315 void bdrv_register(BlockDriver *bdrv);
 316 int bdrv_drop_intermediate(BlockDriverState *top, BlockDriverState *base,
 317                            const char *backing_file_str);
 318 BlockDriverState *bdrv_find_overlay(BlockDriverState *active,
 319                                     BlockDriverState *bs);
 320 BlockDriverState *bdrv_find_base(BlockDriverState *bs);
 321
 322
 323 typedef struct BdrvCheckResult {
 324     int corruptions;
 325     int leaks;
 326     int check_errors;
 327     int corruptions_fixed;
 328     int leaks_fixed;
 329     int64_t image_end_offset;
 330     BlockFragInfo bfi;
 331 } BdrvCheckResult;
 332
 333 typedef enum {
 334     BDRV_FIX_LEAKS    = 1,
 335     BDRV_FIX_ERRORS   = 2,
 336 } BdrvCheckMode;
 337
 338 int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix);
 339
 340 /* The units of offset and total_work_size may be chosen arbitrarily by the
 341  * block driver; total_work_size may change during the course of the amendment
 342  * operation */
 343 typedef void BlockDriverAmendStatusCB(BlockDriverState *bs, int64_t offset,
 344                                       int64_t total_work_size, void *opaque);
 345 int bdrv_amend_options(BlockDriverState *bs_new, QemuOpts *opts,
 346                        BlockDriverAmendStatusCB *status_cb, void *cb_opaque,
 347                        Error **errp);
 348
 349 /* external snapshots */
 350 bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs,
 351                                       BlockDriverState *candidate);
 352 bool bdrv_is_first_non_filter(BlockDriverState *candidate);
 353
 354 /* check if a named node can be replaced when doing drive-mirror */
 355 BlockDriverState *check_to_replace_node(BlockDriverState *parent_bs,
 356                                         const char *node_name, Error **errp);
 357
 358 /* async block I/O */
 359 void bdrv_aio_cancel(BlockAIOCB *acb);
 360 void bdrv_aio_cancel_async(BlockAIOCB *acb);
 361
 362 /* sg packet commands */
 363 int bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf);
 364
 365 /* Invalidate any cached metadata used by image formats */
 366 void bdrv_invalidate_cache(BlockDriverState *bs, Error **errp);
 367 void bdrv_invalidate_cache_all(Error **errp);
 368 int bdrv_inactivate_all(void);
 369
 370 /* Ensure contents are flushed to disk.  */
 371 int bdrv_flush(BlockDriverState *bs);
 372 int coroutine_fn bdrv_co_flush(BlockDriverState *bs);
 373 int bdrv_flush_all(void);
 374 void bdrv_close_all(void);
 375 void bdrv_drain(BlockDriverState *bs);
 376 void coroutine_fn bdrv_co_drain(BlockDriverState *bs);
 377 void bdrv_drain_all_begin(void);
 378 void bdrv_drain_all_end(void);
 379 void bdrv_drain_all(void);
 380
 381 /* Returns NULL when bs == NULL */
 382 AioWait *bdrv_get_aio_wait(BlockDriverState *bs);
 383
 384 #define BDRV_POLL_WHILE(bs, cond) ({                       \
 385     BlockDriverState *bs_ = (bs);                          \
 386     AIO_WAIT_WHILE(bdrv_get_aio_wait(bs_),                 \
 387                    bdrv_get_aio_context(bs_),              \
 388                    cond); })
 389
 390 int bdrv_pdiscard(BlockDriverState *bs, int64_t offset, int bytes);
 391 int bdrv_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes);
 392 int bdrv_has_zero_init_1(BlockDriverState *bs);
 393 int bdrv_has_zero_init(BlockDriverState *bs);
 394 bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs);
 395 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs);
 396 int bdrv_block_status(BlockDriverState *bs, int64_t offset,
 397                       int64_t bytes, int64_t *pnum, int64_t *map,
 398                       BlockDriverState **file);
 399 int bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base,
 400                             int64_t offset, int64_t bytes, int64_t *pnum,
 401                             int64_t *map, BlockDriverState **file);
 402 int bdrv_is_allocated(BlockDriverState *bs, int64_t offset, int64_t bytes,
 403                       int64_t *pnum);
 404 int bdrv_is_allocated_above(BlockDriverState *top, BlockDriverState *base,
 405                             int64_t offset, int64_t bytes, int64_t *pnum);
 406
 407 bool bdrv_is_read_only(BlockDriverState *bs);
 408 int bdrv_can_set_read_only(BlockDriverState *bs, bool read_only,
 409                            bool ignore_allow_rdw, Error **errp);
 410 int bdrv_set_read_only(BlockDriverState *bs, bool read_only, Error **errp);
 411 bool bdrv_is_writable(BlockDriverState *bs);
 412 bool bdrv_is_sg(BlockDriverState *bs);
 413 bool bdrv_is_inserted(BlockDriverState *bs);
 414 void bdrv_lock_medium(BlockDriverState *bs, bool locked);
 415 void bdrv_eject(BlockDriverState *bs, bool eject_flag);
 416 const char *bdrv_get_format_name(BlockDriverState *bs);
 417 BlockDriverState *bdrv_find_node(const char *node_name);
 418 BlockDeviceInfoList *bdrv_named_nodes_list(Error **errp);
 419 BlockDriverState *bdrv_lookup_bs(const char *device,
 420                                  const char *node_name,
 421                                  Error **errp);
 422 bool bdrv_chain_contains(BlockDriverState *top, BlockDriverState *base);
 423 BlockDriverState *bdrv_next_node(BlockDriverState *bs);
 424
 425 typedef struct BdrvNextIterator {
 426     enum {
 427         BDRV_NEXT_BACKEND_ROOTS,
 428         BDRV_NEXT_MONITOR_OWNED,
 429     } phase;
 430     BlockBackend *blk;
 431     BlockDriverState *bs;
 432 } BdrvNextIterator;
 433
 434 BlockDriverState *bdrv_first(BdrvNextIterator *it);
 435 BlockDriverState *bdrv_next(BdrvNextIterator *it);
 436 void bdrv_next_cleanup(BdrvNextIterator *it);
 437
 438 BlockDriverState *bdrv_next_monitor_owned(BlockDriverState *bs);
 439 bool bdrv_is_encrypted(BlockDriverState *bs);
 440 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 441                          void *opaque);
 442 const char *bdrv_get_node_name(const BlockDriverState *bs);
 443 const char *bdrv_get_device_name(const BlockDriverState *bs);
 444 const char *bdrv_get_device_or_node_name(const BlockDriverState *bs);
 445 int bdrv_get_flags(BlockDriverState *bs);
 446 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi);
 447 ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs);
 448 void bdrv_round_to_clusters(BlockDriverState *bs,
 449                             int64_t offset, int64_t bytes,
 450                             int64_t *cluster_offset,
 451                             int64_t *cluster_bytes);
 452
 453 const char *bdrv_get_encrypted_filename(BlockDriverState *bs);
 454 void bdrv_get_backing_filename(BlockDriverState *bs,
 455                                char *filename, int filename_size);
 456 void bdrv_get_full_backing_filename(BlockDriverState *bs,
 457                                     char *dest, size_t sz, Error **errp);
 458 void bdrv_get_full_backing_filename_from_filename(const char *backed,
 459                                                   const char *backing,
 460                                                   char *dest, size_t sz,
 461                                                   Error **errp);
 462
 463 int path_has_protocol(const char *path);
 464 int path_is_absolute(const char *path);
 465 void path_combine(char *dest, int dest_size,
 466                   const char *base_path,
 467                   const char *filename);
 468
 469 int bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
 470 int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
 471 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
 472                       int64_t pos, int size);
 473
 474 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
 475                       int64_t pos, int size);
 476
 477 void bdrv_img_create(const char *filename, const char *fmt,
 478                      const char *base_filename, const char *base_fmt,
 479                      char *options, uint64_t img_size, int flags,
 480                      bool quiet, Error **errp);
 481
 482 /* Returns the alignment in bytes that is required so that no bounce buffer
 483  * is required throughout the stack */
 484 size_t bdrv_min_mem_align(BlockDriverState *bs);
 485 /* Returns optimal alignment in bytes for bounce buffer */
 486 size_t bdrv_opt_mem_align(BlockDriverState *bs);
 487 void *qemu_blockalign(BlockDriverState *bs, size_t size);
 488 void *qemu_blockalign0(BlockDriverState *bs, size_t size);
 489 void *qemu_try_blockalign(BlockDriverState *bs, size_t size);
 490 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size);
 491 bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov);
 492
 493 void bdrv_enable_copy_on_read(BlockDriverState *bs);
 494 void bdrv_disable_copy_on_read(BlockDriverState *bs);
 495
 496 void bdrv_ref(BlockDriverState *bs);
 497 void bdrv_unref(BlockDriverState *bs);
 498 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child);
 499 BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
 500                              BlockDriverState *child_bs,
 501                              const char *child_name,
 502                              const BdrvChildRole *child_role,
 503                              Error **errp);
 504
 505 bool bdrv_op_is_blocked(BlockDriverState *bs, BlockOpType op, Error **errp);
 506 void bdrv_op_block(BlockDriverState *bs, BlockOpType op, Error *reason);
 507 void bdrv_op_unblock(BlockDriverState *bs, BlockOpType op, Error *reason);
 508 void bdrv_op_block_all(BlockDriverState *bs, Error *reason);
 509 void bdrv_op_unblock_all(BlockDriverState *bs, Error *reason);
 510 bool bdrv_op_blocker_is_empty(BlockDriverState *bs);
 511
 512 #define BLKDBG_EVENT(child, evt) \
 513     do { \
 514         if (child) { \
 515             bdrv_debug_event(child->bs, evt); \
 516         } \
 517     } while (0)
 518
 519 void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event);
 520
 521 int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event,
 522                            const char *tag);
 523 int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag);
 524 int bdrv_debug_resume(BlockDriverState *bs, const char *tag);
 525 bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag);
 526
 527 /**
 528  * bdrv_get_aio_context:
 529  *
 530  * Returns: the currently bound #AioContext
 531  */
 532 AioContext *bdrv_get_aio_context(BlockDriverState *bs);
 533
 534 /**
 535  * Transfer control to @co in the aio context of @bs
 536  */
 537 void bdrv_coroutine_enter(BlockDriverState *bs, Coroutine *co);
 538
 539 /**
 540  * bdrv_set_aio_context:
 541  *
 542  * Changes the #AioContext used for fd handlers, timers, and BHs by this
 543  * BlockDriverState and all its children.
 544  *
 545  * This function must be called with iothread lock held.
 546  */
 547 void bdrv_set_aio_context(BlockDriverState *bs, AioContext *new_context);
 548 int bdrv_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz);
 549 int bdrv_probe_geometry(BlockDriverState *bs, HDGeometry *geo);
 550
 551 void bdrv_io_plug(BlockDriverState *bs);
 552 void bdrv_io_unplug(BlockDriverState *bs);
 553
 554 /**
 555  * bdrv_parent_drained_begin:
 556  *
 557  * Begin a quiesced section of all users of @bs. This is part of
 558  * bdrv_drained_begin.
 559  */
 560 void bdrv_parent_drained_begin(BlockDriverState *bs, BdrvChild *ignore,
 561                                bool ignore_bds_parents);
 562
 563 /**
 564  * bdrv_parent_drained_end:
 565  *
 566  * End a quiesced section of all users of @bs. This is part of
 567  * bdrv_drained_end.
 568  */
 569 void bdrv_parent_drained_end(BlockDriverState *bs, BdrvChild *ignore,
 570                              bool ignore_bds_parents);
 571
 572 /**
 573  * bdrv_drain_poll:
 574  *
 575  * Poll for pending requests in @bs, its parents (except for @ignore_parent),
 576  * and if @recursive is true its children as well (used for subtree drain).
 577  *
 578  * If @ignore_bds_parents is true, parents that are BlockDriverStates must
 579  * ignore the drain request because they will be drained separately (used for
 580  * drain_all).
 581  *
 582  * This is part of bdrv_drained_begin.
 583  */
 584 bool bdrv_drain_poll(BlockDriverState *bs, bool recursive,
 585                      BdrvChild *ignore_parent, bool ignore_bds_parents);
 586
 587 /**
 588  * bdrv_drained_begin:
 589  *
 590  * Begin a quiesced section for exclusive access to the BDS, by disabling
 591  * external request sources including NBD server and device model. Note that
 592  * this doesn't block timers or coroutines from submitting more requests, which
 593  * means block_job_pause is still necessary.
 594  *
 595  * This function can be recursive.
 596  */
 597 void bdrv_drained_begin(BlockDriverState *bs);
 598
 599 /**
 600  * bdrv_do_drained_begin_quiesce:
 601  *
 602  * Quiesces a BDS like bdrv_drained_begin(), but does not wait for already
 603  * running requests to complete.
 604  */
 605 void bdrv_do_drained_begin_quiesce(BlockDriverState *bs,
 606                                    BdrvChild *parent, bool ignore_bds_parents);
 607
 608 /**
 609  * Like bdrv_drained_begin, but recursively begins a quiesced section for
 610  * exclusive access to all child nodes as well.
 611  */
 612 void bdrv_subtree_drained_begin(BlockDriverState *bs);
 613
 614 /**
 615  * bdrv_drained_end:
 616  *
 617  * End a quiescent section started by bdrv_drained_begin().
 618  */
 619 void bdrv_drained_end(BlockDriverState *bs);
 620
 621 /**
 622  * End a quiescent section started by bdrv_subtree_drained_begin().
 623  */
 624 void bdrv_subtree_drained_end(BlockDriverState *bs);
 625
 626 void bdrv_add_child(BlockDriverState *parent, BlockDriverState *child,
 627                     Error **errp);
 628 void bdrv_del_child(BlockDriverState *parent, BdrvChild *child, Error **errp);
 629
 630 bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
 631                                      uint32_t granularity, Error **errp);
 632 /**
 633  *
 634  * bdrv_register_buf/bdrv_unregister_buf:
 635  *
 636  * Register/unregister a buffer for I/O. For example, VFIO drivers are
 637  * interested to know the memory areas that would later be used for I/O, so
 638  * that they can prepare IOMMU mapping etc., to get better performance.
 639  */
 640 void bdrv_register_buf(BlockDriverState *bs, void *host, size_t size);
 641 void bdrv_unregister_buf(BlockDriverState *bs, void *host);
 642
 643 /**
 644  *
 645  * bdrv_co_copy_range:
 646  *
 647  * Do offloaded copy between two children. If the operation is not implemented
 648  * by the driver, or if the backend storage doesn't support it, a negative
 649  * error code will be returned.
 650  *
 651  * Note: block layer doesn't emulate or fallback to a bounce buffer approach
 652  * because usually the caller shouldn't attempt offloaded copy any more (e.g.
 653  * calling copy_file_range(2)) after the first error, thus it should fall back
 654  * to a read+write path in the caller level.
 655  *
 656  * @src: Source child to copy data from
 657  * @src_offset: offset in @src image to read data
 658  * @dst: Destination child to copy data to
 659  * @dst_offset: offset in @dst image to write data
 660  * @bytes: number of bytes to copy
 661  * @flags: request flags. Must be one of:
 662  *         0 - actually read data from src;
 663  *         BDRV_REQ_ZERO_WRITE - treat the @src range as zero data and do zero
 664  *                               write on @dst as if bdrv_co_pwrite_zeroes is
 665  *                               called. Used to simplify caller code, or
 666  *                               during BlockDriver.bdrv_co_copy_range_from()
 667  *                               recursion.
 668  *
 669  * Returns: 0 if succeeded; negative error code if failed.
 670  **/
 671 int coroutine_fn bdrv_co_copy_range(BdrvChild *src, uint64_t src_offset,
 672                                     BdrvChild *dst, uint64_t dst_offset,
 673                                     uint64_t bytes, BdrvRequestFlags flags);
 674 #endif