include/block/block_int.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_INT_H
  25 #define BLOCK_INT_H
  26
  27 #include "block/block.h"
  28 #include "qemu/option.h"
  29 #include "qemu/queue.h"
  30 #include "block/coroutine.h"
  31 #include "qemu/timer.h"
  32 #include "qapi-types.h"
  33 #include "qapi/qmp/qerror.h"
  34 #include "monitor/monitor.h"
  35 #include "qemu/hbitmap.h"
  36 #include "block/snapshot.h"
  37 #include "qemu/main-loop.h"
  38 #include "qemu/throttle.h"
  39
  40 #define BLOCK_FLAG_ENCRYPT          1
  41 #define BLOCK_FLAG_COMPAT6          4
  42 #define BLOCK_FLAG_LAZY_REFCOUNTS   8
  43
  44 #define BLOCK_OPT_SIZE              "size"
  45 #define BLOCK_OPT_ENCRYPT           "encryption"
  46 #define BLOCK_OPT_COMPAT6           "compat6"
  47 #define BLOCK_OPT_BACKING_FILE      "backing_file"
  48 #define BLOCK_OPT_BACKING_FMT       "backing_fmt"
  49 #define BLOCK_OPT_CLUSTER_SIZE      "cluster_size"
  50 #define BLOCK_OPT_TABLE_SIZE        "table_size"
  51 #define BLOCK_OPT_PREALLOC          "preallocation"
  52 #define BLOCK_OPT_SUBFMT            "subformat"
  53 #define BLOCK_OPT_COMPAT_LEVEL      "compat"
  54 #define BLOCK_OPT_LAZY_REFCOUNTS    "lazy_refcounts"
  55 #define BLOCK_OPT_ADAPTER_TYPE      "adapter_type"
  56
  57 typedef struct BdrvTrackedRequest {
  58     BlockDriverState *bs;
  59     int64_t sector_num;
  60     int nb_sectors;
  61     bool is_write;
  62     QLIST_ENTRY(BdrvTrackedRequest) list;
  63     Coroutine *co; /* owner, used for deadlock detection */
  64     CoQueue wait_queue; /* coroutines blocked on this request */
  65 } BdrvTrackedRequest;
  66
  67 struct BlockDriver {
  68     const char *format_name;
  69     int instance_size;
  70     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
  71     int (*bdrv_probe_device)(const char *filename);
  72
  73     /* Any driver implementing this callback is expected to be able to handle
  74      * NULL file names in its .bdrv_open() implementation */
  75     void (*bdrv_parse_filename)(const char *filename, QDict *options, Error **errp);
  76
  77     /* For handling image reopen for split or non-split files */
  78     int (*bdrv_reopen_prepare)(BDRVReopenState *reopen_state,
  79                                BlockReopenQueue *queue, Error **errp);
  80     void (*bdrv_reopen_commit)(BDRVReopenState *reopen_state);
  81     void (*bdrv_reopen_abort)(BDRVReopenState *reopen_state);
  82
  83     int (*bdrv_open)(BlockDriverState *bs, QDict *options, int flags);
  84     int (*bdrv_file_open)(BlockDriverState *bs, QDict *options, int flags);
  85     int (*bdrv_read)(BlockDriverState *bs, int64_t sector_num,
  86                      uint8_t *buf, int nb_sectors);
  87     int (*bdrv_write)(BlockDriverState *bs, int64_t sector_num,
  88                       const uint8_t *buf, int nb_sectors);
  89     void (*bdrv_close)(BlockDriverState *bs);
  90     void (*bdrv_rebind)(BlockDriverState *bs);
  91     int (*bdrv_create)(const char *filename, QEMUOptionParameter *options);
  92     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
  93     int (*bdrv_make_empty)(BlockDriverState *bs);
  94     /* aio */
  95     BlockDriverAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
  96         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  97         BlockDriverCompletionFunc *cb, void *opaque);
  98     BlockDriverAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
  99         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 100         BlockDriverCompletionFunc *cb, void *opaque);
 101     BlockDriverAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
 102         BlockDriverCompletionFunc *cb, void *opaque);
 103     BlockDriverAIOCB *(*bdrv_aio_discard)(BlockDriverState *bs,
 104         int64_t sector_num, int nb_sectors,
 105         BlockDriverCompletionFunc *cb, void *opaque);
 106
 107     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
 108         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
 109     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
 110         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
 111     /*
 112      * Efficiently zero a region of the disk image.  Typically an image format
 113      * would use a compact metadata representation to implement this.  This
 114      * function pointer may be NULL and .bdrv_co_writev() will be called
 115      * instead.
 116      */
 117     int coroutine_fn (*bdrv_co_write_zeroes)(BlockDriverState *bs,
 118         int64_t sector_num, int nb_sectors);
 119     int coroutine_fn (*bdrv_co_discard)(BlockDriverState *bs,
 120         int64_t sector_num, int nb_sectors);
 121     int64_t coroutine_fn (*bdrv_co_get_block_status)(BlockDriverState *bs,
 122         int64_t sector_num, int nb_sectors, int *pnum);
 123
 124     /*
 125      * Invalidate any cached meta-data.
 126      */
 127     void (*bdrv_invalidate_cache)(BlockDriverState *bs);
 128
 129     /*
 130      * Flushes all data that was already written to the OS all the way down to
 131      * the disk (for example raw-posix calls fsync()).
 132      */
 133     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
 134
 135     /*
 136      * Flushes all internal caches to the OS. The data may still sit in a
 137      * writeback cache of the host OS, but it will survive a crash of the qemu
 138      * process.
 139      */
 140     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
 141
 142     const char *protocol_name;
 143     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset);
 144     int64_t (*bdrv_getlength)(BlockDriverState *bs);
 145     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
 146     int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
 147                                  const uint8_t *buf, int nb_sectors);
 148
 149     int (*bdrv_snapshot_create)(BlockDriverState *bs,
 150                                 QEMUSnapshotInfo *sn_info);
 151     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
 152                               const char *snapshot_id);
 153     int (*bdrv_snapshot_delete)(BlockDriverState *bs, const char *snapshot_id);
 154     int (*bdrv_snapshot_list)(BlockDriverState *bs,
 155                               QEMUSnapshotInfo **psn_info);
 156     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
 157                                   const char *snapshot_name);
 158     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
 159
 160     int (*bdrv_save_vmstate)(BlockDriverState *bs, QEMUIOVector *qiov,
 161                              int64_t pos);
 162     int (*bdrv_load_vmstate)(BlockDriverState *bs, uint8_t *buf,
 163                              int64_t pos, int size);
 164
 165     int (*bdrv_change_backing_file)(BlockDriverState *bs,
 166         const char *backing_file, const char *backing_fmt);
 167
 168     /* removable device specific */
 169     int (*bdrv_is_inserted)(BlockDriverState *bs);
 170     int (*bdrv_media_changed)(BlockDriverState *bs);
 171     void (*bdrv_eject)(BlockDriverState *bs, bool eject_flag);
 172     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
 173
 174     /* to control generic scsi devices */
 175     int (*bdrv_ioctl)(BlockDriverState *bs, unsigned long int req, void *buf);
 176     BlockDriverAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
 177         unsigned long int req, void *buf,
 178         BlockDriverCompletionFunc *cb, void *opaque);
 179
 180     /* List of options for creating images, terminated by name == NULL */
 181     QEMUOptionParameter *create_options;
 182
 183
 184     /*
 185      * Returns 0 for completed check, -errno for internal errors.
 186      * The check results are stored in result.
 187      */
 188     int (*bdrv_check)(BlockDriverState* bs, BdrvCheckResult *result,
 189         BdrvCheckMode fix);
 190
 191     void (*bdrv_debug_event)(BlockDriverState *bs, BlkDebugEvent event);
 192
 193     /* TODO Better pass a option string/QDict/QemuOpts to add any rule? */
 194     int (*bdrv_debug_breakpoint)(BlockDriverState *bs, const char *event,
 195         const char *tag);
 196     int (*bdrv_debug_resume)(BlockDriverState *bs, const char *tag);
 197     bool (*bdrv_debug_is_suspended)(BlockDriverState *bs, const char *tag);
 198
 199     /*
 200      * Returns 1 if newly created images are guaranteed to contain only
 201      * zeros, 0 otherwise.
 202      */
 203     int (*bdrv_has_zero_init)(BlockDriverState *bs);
 204
 205     QLIST_ENTRY(BlockDriver) list;
 206 };
 207
 208 /*
 209  * Note: the function bdrv_append() copies and swaps contents of
 210  * BlockDriverStates, so if you add new fields to this struct, please
 211  * inspect bdrv_append() to determine if the new fields need to be
 212  * copied as well.
 213  */
 214 struct BlockDriverState {
 215     int64_t total_sectors; /* if we are reading a disk image, give its
 216                               size in sectors */
 217     int read_only; /* if true, the media is read only */
 218     int open_flags; /* flags used to open the file, re-used for re-open */
 219     int encrypted; /* if true, the media is encrypted */
 220     int valid_key; /* if true, a valid encryption key has been set */
 221     int sg;        /* if true, the device is a /dev/sg* */
 222     int copy_on_read; /* if true, copy read backing sectors into image
 223                          note this is a reference count */
 224
 225     BlockDriver *drv; /* NULL means no media */
 226     void *opaque;
 227
 228     void *dev;                  /* attached device model, if any */
 229     /* TODO change to DeviceState when all users are qdevified */
 230     const BlockDevOps *dev_ops;
 231     void *dev_opaque;
 232
 233     char filename[1024];
 234     char backing_file[1024]; /* if non zero, the image is a diff of
 235                                 this file image */
 236     char backing_format[16]; /* if non-zero and backing_file exists */
 237     int is_temporary;
 238
 239     BlockDriverState *backing_hd;
 240     BlockDriverState *file;
 241
 242     NotifierList close_notifiers;
 243
 244     /* Callback before write request is processed */
 245     NotifierWithReturnList before_write_notifiers;
 246
 247     /* number of in-flight copy-on-read requests */
 248     unsigned int copy_on_read_in_flight;
 249
 250     /* I/O throttling */
 251     ThrottleState throttle_state;
 252     CoQueue      throttled_reqs[2];
 253     bool         io_limits_enabled;
 254
 255     /* I/O stats (display with "info blockstats"). */
 256     uint64_t nr_bytes[BDRV_MAX_IOTYPE];
 257     uint64_t nr_ops[BDRV_MAX_IOTYPE];
 258     uint64_t total_time_ns[BDRV_MAX_IOTYPE];
 259     uint64_t wr_highest_sector;
 260
 261     /* Whether the disk can expand beyond total_sectors */
 262     int growable;
 263
 264     /* Whether produces zeros when read beyond eof */
 265     bool zero_beyond_eof;
 266
 267     /* the memory alignment required for the buffers handled by this driver */
 268     int buffer_alignment;
 269
 270     /* do we need to tell the quest if we have a volatile write cache? */
 271     int enable_write_cache;
 272
 273     /* NOTE: the following infos are only hints for real hardware
 274        drivers. They are not used by the block driver */
 275     BlockdevOnError on_read_error, on_write_error;
 276     bool iostatus_enabled;
 277     BlockDeviceIoStatus iostatus;
 278     char device_name[32];
 279     HBitmap *dirty_bitmap;
 280     int refcnt;
 281     int in_use; /* users other than guest access, eg. block migration */
 282     QTAILQ_ENTRY(BlockDriverState) list;
 283
 284     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
 285
 286     /* long-running background operation */
 287     BlockJob *job;
 288
 289     QDict *options;
 290 };
 291
 292 int get_tmp_filename(char *filename, int size);
 293
 294 void bdrv_set_io_limits(BlockDriverState *bs,
 295                         ThrottleConfig *cfg);
 296
 297
 298 /**
 299  * bdrv_add_before_write_notifier:
 300  *
 301  * Register a callback that is invoked before write requests are processed but
 302  * after any throttling or waiting for overlapping requests.
 303  */
 304 void bdrv_add_before_write_notifier(BlockDriverState *bs,
 305                                     NotifierWithReturn *notifier);
 306
 307 /**
 308  * bdrv_get_aio_context:
 309  *
 310  * Returns: the currently bound #AioContext
 311  */
 312 AioContext *bdrv_get_aio_context(BlockDriverState *bs);
 313
 314 #ifdef _WIN32
 315 int is_windows_drive(const char *filename);
 316 #endif
 317 void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv,
 318                                enum MonitorEvent ev,
 319                                BlockErrorAction action, bool is_read);
 320
 321 /**
 322  * stream_start:
 323  * @bs: Block device to operate on.
 324  * @base: Block device that will become the new base, or %NULL to
 325  * flatten the whole backing file chain onto @bs.
 326  * @base_id: The file name that will be written to @bs as the new
 327  * backing file if the job completes.  Ignored if @base is %NULL.
 328  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
 329  * @on_error: The action to take upon error.
 330  * @cb: Completion function for the job.
 331  * @opaque: Opaque pointer value passed to @cb.
 332  * @errp: Error object.
 333  *
 334  * Start a streaming operation on @bs.  Clusters that are unallocated
 335  * in @bs, but allocated in any image between @base and @bs (both
 336  * exclusive) will be written to @bs.  At the end of a successful
 337  * streaming job, the backing file of @bs will be changed to
 338  * @base_id in the written image and to @base in the live BlockDriverState.
 339  */
 340 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 341                   const char *base_id, int64_t speed, BlockdevOnError on_error,
 342                   BlockDriverCompletionFunc *cb,
 343                   void *opaque, Error **errp);
 344
 345 /**
 346  * commit_start:
 347  * @bs: Top Block device
 348  * @base: Block device that will be written into, and become the new top
 349  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
 350  * @on_error: The action to take upon error.
 351  * @cb: Completion function for the job.
 352  * @opaque: Opaque pointer value passed to @cb.
 353  * @errp: Error object.
 354  *
 355  */
 356 void commit_start(BlockDriverState *bs, BlockDriverState *base,
 357                  BlockDriverState *top, int64_t speed,
 358                  BlockdevOnError on_error, BlockDriverCompletionFunc *cb,
 359                  void *opaque, Error **errp);
 360
 361 /*
 362  * mirror_start:
 363  * @bs: Block device to operate on.
 364  * @target: Block device to write to.
 365  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
 366  * @granularity: The chosen granularity for the dirty bitmap.
 367  * @buf_size: The amount of data that can be in flight at one time.
 368  * @mode: Whether to collapse all images in the chain to the target.
 369  * @on_source_error: The action to take upon error reading from the source.
 370  * @on_target_error: The action to take upon error writing to the target.
 371  * @cb: Completion function for the job.
 372  * @opaque: Opaque pointer value passed to @cb.
 373  * @errp: Error object.
 374  *
 375  * Start a mirroring operation on @bs.  Clusters that are allocated
 376  * in @bs will be written to @bs until the job is cancelled or
 377  * manually completed.  At the end of a successful mirroring job,
 378  * @bs will be switched to read from @target.
 379  */
 380 void mirror_start(BlockDriverState *bs, BlockDriverState *target,
 381                   int64_t speed, int64_t granularity, int64_t buf_size,
 382                   MirrorSyncMode mode, BlockdevOnError on_source_error,
 383                   BlockdevOnError on_target_error,
 384                   BlockDriverCompletionFunc *cb,
 385                   void *opaque, Error **errp);
 386
 387 /*
 388  * backup_start:
 389  * @bs: Block device to operate on.
 390  * @target: Block device to write to.
 391  * @speed: The maximum speed, in bytes per second, or 0 for unlimited.
 392  * @sync_mode: What parts of the disk image should be copied to the destination.
 393  * @on_source_error: The action to take upon error reading from the source.
 394  * @on_target_error: The action to take upon error writing to the target.
 395  * @cb: Completion function for the job.
 396  * @opaque: Opaque pointer value passed to @cb.
 397  *
 398  * Start a backup operation on @bs.  Clusters in @bs are written to @target
 399  * until the job is cancelled or manually completed.
 400  */
 401 void backup_start(BlockDriverState *bs, BlockDriverState *target,
 402                   int64_t speed, MirrorSyncMode sync_mode,
 403                   BlockdevOnError on_source_error,
 404                   BlockdevOnError on_target_error,
 405                   BlockDriverCompletionFunc *cb, void *opaque,
 406                   Error **errp);
 407
 408 #endif /* BLOCK_INT_H */