include/block/block-io.h

   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_IO_H
  25 #define BLOCK_IO_H
  26
  27 #include "block/aio-wait.h"
  28 #include "block/block-common.h"
  29 #include "qemu/coroutine.h"
  30 #include "qemu/iov.h"
  31
  32 /*
  33  * I/O API functions. These functions are thread-safe, and therefore
  34  * can run in any thread.
  35  *
  36  * These functions can only call functions from I/O and Common categories,
  37  * but can be invoked by GS, "I/O or GS" and I/O APIs.
  38  *
  39  * All functions in this category must use the macro
  40  * IO_CODE();
  41  * to catch when they are accidentally called by the wrong API.
  42  */
  43
  44 int co_wrapper_mixed_bdrv_rdlock
  45 bdrv_pwrite_zeroes(BdrvChild *child, int64_t offset, int64_t bytes,
  46                    BdrvRequestFlags flags);
  47
  48 int bdrv_make_zero(BdrvChild *child, BdrvRequestFlags flags);
  49
  50 int co_wrapper_mixed_bdrv_rdlock
  51 bdrv_pread(BdrvChild *child, int64_t offset, int64_t bytes, void *buf,
  52            BdrvRequestFlags flags);
  53
  54 int co_wrapper_mixed_bdrv_rdlock
  55 bdrv_pwrite(BdrvChild *child, int64_t offset,int64_t bytes,
  56             const void *buf, BdrvRequestFlags flags);
  57
  58 int co_wrapper_mixed_bdrv_rdlock
  59 bdrv_pwrite_sync(BdrvChild *child, int64_t offset, int64_t bytes,
  60                  const void *buf, BdrvRequestFlags flags);
  61
  62 int coroutine_fn GRAPH_RDLOCK
  63 bdrv_co_pwrite_sync(BdrvChild *child, int64_t offset, int64_t bytes,
  64                     const void *buf, BdrvRequestFlags flags);
  65
  66 /*
  67  * Efficiently zero a region of the disk image.  Note that this is a regular
  68  * I/O request like read or write and should have a reasonable size.  This
  69  * function is not suitable for zeroing the entire image in a single request
  70  * because it may allocate memory for the entire region.
  71  */
  72 int coroutine_fn GRAPH_RDLOCK
  73 bdrv_co_pwrite_zeroes(BdrvChild *child, int64_t offset, int64_t bytes,
  74                       BdrvRequestFlags flags);
  75
  76 int coroutine_fn GRAPH_RDLOCK
  77 bdrv_co_truncate(BdrvChild *child, int64_t offset, bool exact,
  78                  PreallocMode prealloc, BdrvRequestFlags flags, Error **errp);
  79
  80 int64_t coroutine_fn GRAPH_RDLOCK bdrv_co_nb_sectors(BlockDriverState *bs);
  81 int64_t coroutine_mixed_fn bdrv_nb_sectors(BlockDriverState *bs);
  82
  83 int64_t coroutine_fn GRAPH_RDLOCK bdrv_co_getlength(BlockDriverState *bs);
  84 int64_t co_wrapper_mixed_bdrv_rdlock bdrv_getlength(BlockDriverState *bs);
  85
  86 int64_t coroutine_fn GRAPH_RDLOCK
  87 bdrv_co_get_allocated_file_size(BlockDriverState *bs);
  88
  89 int64_t co_wrapper_bdrv_rdlock
  90 bdrv_get_allocated_file_size(BlockDriverState *bs);
  91
  92 BlockMeasureInfo *bdrv_measure(BlockDriver *drv, QemuOpts *opts,
  93                                BlockDriverState *in_bs, Error **errp);
  94
  95 int coroutine_fn GRAPH_RDLOCK
  96 bdrv_co_delete_file(BlockDriverState *bs, Error **errp);
  97
  98 void coroutine_fn GRAPH_RDLOCK
  99 bdrv_co_delete_file_noerr(BlockDriverState *bs);
 100
 101
 102 /* async block I/O */
 103 void bdrv_aio_cancel_async(BlockAIOCB *acb);
 104
 105 /* sg packet commands */
 106 int coroutine_fn GRAPH_RDLOCK
 107 bdrv_co_ioctl(BlockDriverState *bs, int req, void *buf);
 108
 109 /* Ensure contents are flushed to disk.  */
 110 int coroutine_fn GRAPH_RDLOCK bdrv_co_flush(BlockDriverState *bs);
 111
 112 int coroutine_fn GRAPH_RDLOCK bdrv_co_pdiscard(BdrvChild *child, int64_t offset,
 113                                                int64_t bytes);
 114
 115 /* Report zone information of zone block device. */
 116 int coroutine_fn GRAPH_RDLOCK bdrv_co_zone_report(BlockDriverState *bs,
 117                                                   int64_t offset,
 118                                                   unsigned int *nr_zones,
 119                                                   BlockZoneDescriptor *zones);
 120 int coroutine_fn GRAPH_RDLOCK bdrv_co_zone_mgmt(BlockDriverState *bs,
 121                                                 BlockZoneOp op,
 122                                                 int64_t offset, int64_t len);
 123 int coroutine_fn GRAPH_RDLOCK bdrv_co_zone_append(BlockDriverState *bs,
 124                                                   int64_t *offset,
 125                                                   QEMUIOVector *qiov,
 126                                                   BdrvRequestFlags flags);
 127
 128 bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs);
 129
 130 int coroutine_fn GRAPH_RDLOCK
 131 bdrv_co_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
 132                      int64_t *pnum, int64_t *map, BlockDriverState **file);
 133 int co_wrapper_mixed_bdrv_rdlock
 134 bdrv_block_status(BlockDriverState *bs, int64_t offset, int64_t bytes,
 135                   int64_t *pnum, int64_t *map, BlockDriverState **file);
 136
 137 int coroutine_fn GRAPH_RDLOCK
 138 bdrv_co_block_status_above(BlockDriverState *bs, BlockDriverState *base,
 139                            int64_t offset, int64_t bytes, int64_t *pnum,
 140                            int64_t *map, BlockDriverState **file);
 141 int co_wrapper_mixed_bdrv_rdlock
 142 bdrv_block_status_above(BlockDriverState *bs, BlockDriverState *base,
 143                         int64_t offset, int64_t bytes, int64_t *pnum,
 144                         int64_t *map, BlockDriverState **file);
 145
 146 int coroutine_fn GRAPH_RDLOCK
 147 bdrv_co_is_allocated(BlockDriverState *bs, int64_t offset, int64_t bytes,
 148                      int64_t *pnum);
 149 int co_wrapper_mixed_bdrv_rdlock
 150 bdrv_is_allocated(BlockDriverState *bs, int64_t offset,
 151                   int64_t bytes, int64_t *pnum);
 152
 153 int coroutine_fn GRAPH_RDLOCK
 154 bdrv_co_is_allocated_above(BlockDriverState *top, BlockDriverState *base,
 155                            bool include_base, int64_t offset, int64_t bytes,
 156                            int64_t *pnum);
 157 int co_wrapper_mixed_bdrv_rdlock
 158 bdrv_is_allocated_above(BlockDriverState *bs, BlockDriverState *base,
 159                         bool include_base, int64_t offset,
 160                         int64_t bytes, int64_t *pnum);
 161
 162 int coroutine_fn GRAPH_RDLOCK
 163 bdrv_co_is_zero_fast(BlockDriverState *bs, int64_t offset, int64_t bytes);
 164
 165 int GRAPH_RDLOCK
 166 bdrv_apply_auto_read_only(BlockDriverState *bs, const char *errmsg,
 167                           Error **errp);
 168
 169 bool bdrv_is_read_only(BlockDriverState *bs);
 170 bool bdrv_is_writable(BlockDriverState *bs);
 171 bool bdrv_is_sg(BlockDriverState *bs);
 172 int bdrv_get_flags(BlockDriverState *bs);
 173
 174 bool coroutine_fn GRAPH_RDLOCK bdrv_co_is_inserted(BlockDriverState *bs);
 175 bool co_wrapper_bdrv_rdlock bdrv_is_inserted(BlockDriverState *bs);
 176
 177 void coroutine_fn GRAPH_RDLOCK
 178 bdrv_co_lock_medium(BlockDriverState *bs, bool locked);
 179
 180 void coroutine_fn GRAPH_RDLOCK
 181 bdrv_co_eject(BlockDriverState *bs, bool eject_flag);
 182
 183 const char *bdrv_get_format_name(BlockDriverState *bs);
 184
 185 bool GRAPH_RDLOCK bdrv_supports_compressed_writes(BlockDriverState *bs);
 186 const char *bdrv_get_node_name(const BlockDriverState *bs);
 187
 188 const char * GRAPH_RDLOCK
 189 bdrv_get_device_name(const BlockDriverState *bs);
 190
 191 const char * GRAPH_RDLOCK
 192 bdrv_get_device_or_node_name(const BlockDriverState *bs);
 193
 194 int coroutine_fn GRAPH_RDLOCK
 195 bdrv_co_get_info(BlockDriverState *bs, BlockDriverInfo *bdi);
 196
 197 int co_wrapper_mixed_bdrv_rdlock
 198 bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi);
 199
 200 ImageInfoSpecific * GRAPH_RDLOCK
 201 bdrv_get_specific_info(BlockDriverState *bs, Error **errp);
 202
 203 BlockStatsSpecific *bdrv_get_specific_stats(BlockDriverState *bs);
 204 void bdrv_round_to_subclusters(BlockDriverState *bs,
 205                                int64_t offset, int64_t bytes,
 206                                int64_t *cluster_offset,
 207                                int64_t *cluster_bytes);
 208
 209 void bdrv_get_backing_filename(BlockDriverState *bs,
 210                                char *filename, int filename_size);
 211
 212 int coroutine_fn GRAPH_RDLOCK
 213 bdrv_co_change_backing_file(BlockDriverState *bs, const char *backing_file,
 214                             const char *backing_fmt, bool warn);
 215
 216 int co_wrapper_bdrv_rdlock
 217 bdrv_change_backing_file(BlockDriverState *bs, const char *backing_file,
 218                          const char *backing_fmt, bool warn);
 219
 220 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
 221                       int64_t pos, int size);
 222
 223 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
 224                       int64_t pos, int size);
 225
 226 /*
 227  * Returns the alignment in bytes that is required so that no bounce buffer
 228  * is required throughout the stack
 229  */
 230 size_t bdrv_min_mem_align(BlockDriverState *bs);
 231 /* Returns optimal alignment in bytes for bounce buffer */
 232 size_t bdrv_opt_mem_align(BlockDriverState *bs);
 233 void *qemu_blockalign(BlockDriverState *bs, size_t size);
 234 void *qemu_blockalign0(BlockDriverState *bs, size_t size);
 235 void *qemu_try_blockalign(BlockDriverState *bs, size_t size);
 236 void *qemu_try_blockalign0(BlockDriverState *bs, size_t size);
 237
 238 void bdrv_enable_copy_on_read(BlockDriverState *bs);
 239 void bdrv_disable_copy_on_read(BlockDriverState *bs);
 240
 241 void coroutine_fn GRAPH_RDLOCK
 242 bdrv_co_debug_event(BlockDriverState *bs, BlkdebugEvent event);
 243
 244 void co_wrapper_mixed_bdrv_rdlock
 245 bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event);
 246
 247 #define BLKDBG_CO_EVENT(child, evt) \
 248     do { \
 249         if (child) { \
 250             bdrv_co_debug_event(child->bs, evt); \
 251         } \
 252     } while (0)
 253
 254 #define BLKDBG_EVENT(child, evt) \
 255     do { \
 256         if (child) { \
 257             bdrv_debug_event(child->bs, evt); \
 258         } \
 259     } while (0)
 260
 261 /**
 262  * bdrv_get_aio_context:
 263  *
 264  * Returns: the currently bound #AioContext
 265  */
 266 AioContext *bdrv_get_aio_context(BlockDriverState *bs);
 267
 268 AioContext *bdrv_child_get_parent_aio_context(BdrvChild *c);
 269
 270 /**
 271  * Move the current coroutine to the AioContext of @bs and return the old
 272  * AioContext of the coroutine. Increase bs->in_flight so that draining @bs
 273  * will wait for the operation to proceed until the corresponding
 274  * bdrv_co_leave().
 275  *
 276  * Consequently, you can't call drain inside a bdrv_co_enter/leave() section as
 277  * this will deadlock.
 278  */
 279 AioContext *coroutine_fn bdrv_co_enter(BlockDriverState *bs);
 280
 281 /**
 282  * Ends a section started by bdrv_co_enter(). Move the current coroutine back
 283  * to old_ctx and decrease bs->in_flight again.
 284  */
 285 void coroutine_fn bdrv_co_leave(BlockDriverState *bs, AioContext *old_ctx);
 286
 287 AioContext *child_of_bds_get_parent_aio_context(BdrvChild *c);
 288
 289 bool coroutine_fn GRAPH_RDLOCK
 290 bdrv_co_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
 291                                    uint32_t granularity, Error **errp);
 292 bool co_wrapper_bdrv_rdlock
 293 bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char *name,
 294                                 uint32_t granularity, Error **errp);
 295
 296 /**
 297  *
 298  * bdrv_co_copy_range:
 299  *
 300  * Do offloaded copy between two children. If the operation is not implemented
 301  * by the driver, or if the backend storage doesn't support it, a negative
 302  * error code will be returned.
 303  *
 304  * Note: block layer doesn't emulate or fallback to a bounce buffer approach
 305  * because usually the caller shouldn't attempt offloaded copy any more (e.g.
 306  * calling copy_file_range(2)) after the first error, thus it should fall back
 307  * to a read+write path in the caller level.
 308  *
 309  * @src: Source child to copy data from
 310  * @src_offset: offset in @src image to read data
 311  * @dst: Destination child to copy data to
 312  * @dst_offset: offset in @dst image to write data
 313  * @bytes: number of bytes to copy
 314  * @flags: request flags. Supported flags:
 315  *         BDRV_REQ_ZERO_WRITE - treat the @src range as zero data and do zero
 316  *                               write on @dst as if bdrv_co_pwrite_zeroes is
 317  *                               called. Used to simplify caller code, or
 318  *                               during BlockDriver.bdrv_co_copy_range_from()
 319  *                               recursion.
 320  *         BDRV_REQ_NO_SERIALISING - do not serialize with other overlapping
 321  *                                   requests currently in flight.
 322  *
 323  * Returns: 0 if succeeded; negative error code if failed.
 324  **/
 325 int coroutine_fn GRAPH_RDLOCK
 326 bdrv_co_copy_range(BdrvChild *src, int64_t src_offset,
 327                    BdrvChild *dst, int64_t dst_offset,
 328                    int64_t bytes, BdrvRequestFlags read_flags,
 329                    BdrvRequestFlags write_flags);
 330
 331 /*
 332  * "I/O or GS" API functions. These functions can run without
 333  * the BQL, but only in one specific iothread/main loop.
 334  *
 335  * More specifically, these functions use BDRV_POLL_WHILE(bs), which requires
 336  * the caller to be either in the main thread or directly in the home thread
 337  * that runs the bs AioContext. Calling them from another thread in another
 338  * AioContext would cause deadlocks.
 339  *
 340  * Therefore, these functions are not proper I/O, because they
 341  * can't run in *any* iothreads, but only in a specific one.
 342  *
 343  * These functions can call any function from I/O, Common and this
 344  * categories, but must be invoked only by other "I/O or GS" and GS APIs.
 345  *
 346  * All functions in this category must use the macro
 347  * IO_OR_GS_CODE();
 348  * to catch when they are accidentally called by the wrong API.
 349  */
 350
 351 #define BDRV_POLL_WHILE(bs, cond) ({                       \
 352     BlockDriverState *bs_ = (bs);                          \
 353     IO_OR_GS_CODE();                                       \
 354     AIO_WAIT_WHILE(bdrv_get_aio_context(bs_),              \
 355                    cond); })
 356
 357 void bdrv_drain(BlockDriverState *bs);
 358
 359 int co_wrapper_mixed_bdrv_rdlock
 360 bdrv_truncate(BdrvChild *child, int64_t offset, bool exact,
 361               PreallocMode prealloc, BdrvRequestFlags flags, Error **errp);
 362
 363 int co_wrapper_mixed_bdrv_rdlock
 364 bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix);
 365
 366 /* Invalidate any cached metadata used by image formats */
 367 int co_wrapper_mixed_bdrv_rdlock
 368 bdrv_invalidate_cache(BlockDriverState *bs, Error **errp);
 369
 370 int co_wrapper_mixed_bdrv_rdlock bdrv_flush(BlockDriverState *bs);
 371
 372 int co_wrapper_mixed_bdrv_rdlock
 373 bdrv_pdiscard(BdrvChild *child, int64_t offset, int64_t bytes);
 374
 375 int co_wrapper_mixed_bdrv_rdlock
 376 bdrv_readv_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
 377
 378 int co_wrapper_mixed_bdrv_rdlock
 379 bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos);
 380
 381 /**
 382  * bdrv_parent_drained_begin_single:
 383  *
 384  * Begin a quiesced section for the parent of @c.
 385  */
 386 void GRAPH_RDLOCK bdrv_parent_drained_begin_single(BdrvChild *c);
 387
 388 /**
 389  * bdrv_parent_drained_poll_single:
 390  *
 391  * Returns true if there is any pending activity to cease before @c can be
 392  * called quiesced, false otherwise.
 393  */
 394 bool GRAPH_RDLOCK bdrv_parent_drained_poll_single(BdrvChild *c);
 395
 396 /**
 397  * bdrv_parent_drained_end_single:
 398  *
 399  * End a quiesced section for the parent of @c.
 400  */
 401 void GRAPH_RDLOCK bdrv_parent_drained_end_single(BdrvChild *c);
 402
 403 /**
 404  * bdrv_drain_poll:
 405  *
 406  * Poll for pending requests in @bs and its parents (except for @ignore_parent).
 407  *
 408  * If @ignore_bds_parents is true, parents that are BlockDriverStates must
 409  * ignore the drain request because they will be drained separately (used for
 410  * drain_all).
 411  *
 412  * This is part of bdrv_drained_begin.
 413  */
 414 bool GRAPH_RDLOCK
 415 bdrv_drain_poll(BlockDriverState *bs, BdrvChild *ignore_parent,
 416                 bool ignore_bds_parents);
 417
 418 /**
 419  * bdrv_drained_begin:
 420  *
 421  * Begin a quiesced section for exclusive access to the BDS, by disabling
 422  * external request sources including NBD server, block jobs, and device model.
 423  *
 424  * This function can only be invoked by the main loop or a coroutine
 425  * (regardless of the AioContext where it is running).
 426  * If the coroutine is running in an Iothread AioContext, this function will
 427  * just schedule a BH to run in the main loop.
 428  * However, it cannot be directly called by an Iothread.
 429  *
 430  * This function can be recursive.
 431  */
 432 void bdrv_drained_begin(BlockDriverState *bs);
 433
 434 /**
 435  * bdrv_do_drained_begin_quiesce:
 436  *
 437  * Quiesces a BDS like bdrv_drained_begin(), but does not wait for already
 438  * running requests to complete.
 439  */
 440 void bdrv_do_drained_begin_quiesce(BlockDriverState *bs, BdrvChild *parent);
 441
 442 /**
 443  * bdrv_drained_end:
 444  *
 445  * End a quiescent section started by bdrv_drained_begin().
 446  *
 447  * This function can only be invoked by the main loop or a coroutine
 448  * (regardless of the AioContext where it is running).
 449  * If the coroutine is running in an Iothread AioContext, this function will
 450  * just schedule a BH to run in the main loop.
 451  * However, it cannot be directly called by an Iothread.
 452  */
 453 void bdrv_drained_end(BlockDriverState *bs);
 454
 455 #endif /* BLOCK_IO_H */