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.h"
  28 #include "qemu-option.h"
  29 #include "qemu-queue.h"
  30 #include "qemu-coroutine.h"
  31 #include "qemu-timer.h"
  32 #include "qapi-types.h"
  33
  34 #define BLOCK_FLAG_ENCRYPT      1
  35 #define BLOCK_FLAG_COMPAT6      4
  36
  37 #define BLOCK_IO_LIMIT_READ     0
  38 #define BLOCK_IO_LIMIT_WRITE    1
  39 #define BLOCK_IO_LIMIT_TOTAL    2
  40
  41 #define BLOCK_IO_SLICE_TIME     100000000
  42 #define NANOSECONDS_PER_SECOND  1000000000.0
  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
  54 typedef struct BdrvTrackedRequest BdrvTrackedRequest;
  55
  56 typedef struct AIOPool {
  57     void (*cancel)(BlockDriverAIOCB *acb);
  58     int aiocb_size;
  59     BlockDriverAIOCB *free_aiocb;
  60 } AIOPool;
  61
  62 typedef struct BlockIOLimit {
  63     int64_t bps[3];
  64     int64_t iops[3];
  65 } BlockIOLimit;
  66
  67 typedef struct BlockIOBaseValue {
  68     uint64_t bytes[2];
  69     uint64_t ios[2];
  70 } BlockIOBaseValue;
  71
  72 typedef void BlockJobCancelFunc(void *opaque);
  73 typedef struct BlockJob BlockJob;
  74 typedef struct BlockJobType {
  75     /** Derived BlockJob struct size */
  76     size_t instance_size;
  77
  78     /** String describing the operation, part of query-block-jobs QMP API */
  79     const char *job_type;
  80
  81     /** Optional callback for job types that support setting a speed limit */
  82     int (*set_speed)(BlockJob *job, int64_t value);
  83 } BlockJobType;
  84
  85 /**
  86  * Long-running operation on a BlockDriverState
  87  */
  88 struct BlockJob {
  89     const BlockJobType *job_type;
  90     BlockDriverState *bs;
  91     bool cancelled;
  92
  93     /* These fields are published by the query-block-jobs QMP API */
  94     int64_t offset;
  95     int64_t len;
  96     int64_t speed;
  97
  98     BlockDriverCompletionFunc *cb;
  99     void *opaque;
 100 };
 101
 102 struct BlockDriver {
 103     const char *format_name;
 104     int instance_size;
 105     int (*bdrv_probe)(const uint8_t *buf, int buf_size, const char *filename);
 106     int (*bdrv_probe_device)(const char *filename);
 107     int (*bdrv_open)(BlockDriverState *bs, int flags);
 108     int (*bdrv_file_open)(BlockDriverState *bs, const char *filename, int flags);
 109     int (*bdrv_read)(BlockDriverState *bs, int64_t sector_num,
 110                      uint8_t *buf, int nb_sectors);
 111     int (*bdrv_write)(BlockDriverState *bs, int64_t sector_num,
 112                       const uint8_t *buf, int nb_sectors);
 113     void (*bdrv_close)(BlockDriverState *bs);
 114     int (*bdrv_create)(const char *filename, QEMUOptionParameter *options);
 115     int (*bdrv_set_key)(BlockDriverState *bs, const char *key);
 116     int (*bdrv_make_empty)(BlockDriverState *bs);
 117     /* aio */
 118     BlockDriverAIOCB *(*bdrv_aio_readv)(BlockDriverState *bs,
 119         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 120         BlockDriverCompletionFunc *cb, void *opaque);
 121     BlockDriverAIOCB *(*bdrv_aio_writev)(BlockDriverState *bs,
 122         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 123         BlockDriverCompletionFunc *cb, void *opaque);
 124     BlockDriverAIOCB *(*bdrv_aio_flush)(BlockDriverState *bs,
 125         BlockDriverCompletionFunc *cb, void *opaque);
 126     BlockDriverAIOCB *(*bdrv_aio_discard)(BlockDriverState *bs,
 127         int64_t sector_num, int nb_sectors,
 128         BlockDriverCompletionFunc *cb, void *opaque);
 129
 130     int coroutine_fn (*bdrv_co_readv)(BlockDriverState *bs,
 131         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
 132     int coroutine_fn (*bdrv_co_writev)(BlockDriverState *bs,
 133         int64_t sector_num, int nb_sectors, QEMUIOVector *qiov);
 134     /*
 135      * Efficiently zero a region of the disk image.  Typically an image format
 136      * would use a compact metadata representation to implement this.  This
 137      * function pointer may be NULL and .bdrv_co_writev() will be called
 138      * instead.
 139      */
 140     int coroutine_fn (*bdrv_co_write_zeroes)(BlockDriverState *bs,
 141         int64_t sector_num, int nb_sectors);
 142     int coroutine_fn (*bdrv_co_discard)(BlockDriverState *bs,
 143         int64_t sector_num, int nb_sectors);
 144     int coroutine_fn (*bdrv_co_is_allocated)(BlockDriverState *bs,
 145         int64_t sector_num, int nb_sectors, int *pnum);
 146
 147     /*
 148      * Invalidate any cached meta-data.
 149      */
 150     void (*bdrv_invalidate_cache)(BlockDriverState *bs);
 151
 152     /*
 153      * Flushes all data that was already written to the OS all the way down to
 154      * the disk (for example raw-posix calls fsync()).
 155      */
 156     int coroutine_fn (*bdrv_co_flush_to_disk)(BlockDriverState *bs);
 157
 158     /*
 159      * Flushes all internal caches to the OS. The data may still sit in a
 160      * writeback cache of the host OS, but it will survive a crash of the qemu
 161      * process.
 162      */
 163     int coroutine_fn (*bdrv_co_flush_to_os)(BlockDriverState *bs);
 164
 165     int (*bdrv_aio_multiwrite)(BlockDriverState *bs, BlockRequest *reqs,
 166         int num_reqs);
 167     int (*bdrv_merge_requests)(BlockDriverState *bs, BlockRequest* a,
 168         BlockRequest *b);
 169
 170
 171     const char *protocol_name;
 172     int (*bdrv_truncate)(BlockDriverState *bs, int64_t offset);
 173     int64_t (*bdrv_getlength)(BlockDriverState *bs);
 174     int64_t (*bdrv_get_allocated_file_size)(BlockDriverState *bs);
 175     int (*bdrv_write_compressed)(BlockDriverState *bs, int64_t sector_num,
 176                                  const uint8_t *buf, int nb_sectors);
 177
 178     int (*bdrv_snapshot_create)(BlockDriverState *bs,
 179                                 QEMUSnapshotInfo *sn_info);
 180     int (*bdrv_snapshot_goto)(BlockDriverState *bs,
 181                               const char *snapshot_id);
 182     int (*bdrv_snapshot_delete)(BlockDriverState *bs, const char *snapshot_id);
 183     int (*bdrv_snapshot_list)(BlockDriverState *bs,
 184                               QEMUSnapshotInfo **psn_info);
 185     int (*bdrv_snapshot_load_tmp)(BlockDriverState *bs,
 186                                   const char *snapshot_name);
 187     int (*bdrv_get_info)(BlockDriverState *bs, BlockDriverInfo *bdi);
 188
 189     int (*bdrv_save_vmstate)(BlockDriverState *bs, const uint8_t *buf,
 190                              int64_t pos, int size);
 191     int (*bdrv_load_vmstate)(BlockDriverState *bs, uint8_t *buf,
 192                              int64_t pos, int size);
 193
 194     int (*bdrv_change_backing_file)(BlockDriverState *bs,
 195         const char *backing_file, const char *backing_fmt);
 196
 197     /* removable device specific */
 198     int (*bdrv_is_inserted)(BlockDriverState *bs);
 199     int (*bdrv_media_changed)(BlockDriverState *bs);
 200     void (*bdrv_eject)(BlockDriverState *bs, int eject_flag);
 201     void (*bdrv_lock_medium)(BlockDriverState *bs, bool locked);
 202
 203     /* to control generic scsi devices */
 204     int (*bdrv_ioctl)(BlockDriverState *bs, unsigned long int req, void *buf);
 205     BlockDriverAIOCB *(*bdrv_aio_ioctl)(BlockDriverState *bs,
 206         unsigned long int req, void *buf,
 207         BlockDriverCompletionFunc *cb, void *opaque);
 208
 209     /* List of options for creating images, terminated by name == NULL */
 210     QEMUOptionParameter *create_options;
 211
 212
 213     /*
 214      * Returns 0 for completed check, -errno for internal errors.
 215      * The check results are stored in result.
 216      */
 217     int (*bdrv_check)(BlockDriverState* bs, BdrvCheckResult *result);
 218
 219     void (*bdrv_debug_event)(BlockDriverState *bs, BlkDebugEvent event);
 220
 221     /*
 222      * Returns 1 if newly created images are guaranteed to contain only
 223      * zeros, 0 otherwise.
 224      */
 225     int (*bdrv_has_zero_init)(BlockDriverState *bs);
 226
 227     QLIST_ENTRY(BlockDriver) list;
 228 };
 229
 230 struct BlockDriverState {
 231     int64_t total_sectors; /* if we are reading a disk image, give its
 232                               size in sectors */
 233     int read_only; /* if true, the media is read only */
 234     int keep_read_only; /* if true, the media was requested to stay read only */
 235     int open_flags; /* flags used to open the file, re-used for re-open */
 236     int encrypted; /* if true, the media is encrypted */
 237     int valid_key; /* if true, a valid encryption key has been set */
 238     int sg;        /* if true, the device is a /dev/sg* */
 239     int copy_on_read; /* if true, copy read backing sectors into image
 240                          note this is a reference count */
 241
 242     BlockDriver *drv; /* NULL means no media */
 243     void *opaque;
 244
 245     void *dev;                  /* attached device model, if any */
 246     /* TODO change to DeviceState when all users are qdevified */
 247     const BlockDevOps *dev_ops;
 248     void *dev_opaque;
 249
 250     char filename[1024];
 251     char backing_file[1024]; /* if non zero, the image is a diff of
 252                                 this file image */
 253     char backing_format[16]; /* if non-zero and backing_file exists */
 254     int is_temporary;
 255
 256     BlockDriverState *backing_hd;
 257     BlockDriverState *file;
 258
 259     /* number of in-flight copy-on-read requests */
 260     unsigned int copy_on_read_in_flight;
 261
 262     /* async read/write emulation */
 263
 264     void *sync_aiocb;
 265
 266     /* the time for latest disk I/O */
 267     int64_t slice_time;
 268     int64_t slice_start;
 269     int64_t slice_end;
 270     BlockIOLimit io_limits;
 271     BlockIOBaseValue  io_base;
 272     CoQueue      throttled_reqs;
 273     QEMUTimer    *block_timer;
 274     bool         io_limits_enabled;
 275
 276     /* I/O stats (display with "info blockstats"). */
 277     uint64_t nr_bytes[BDRV_MAX_IOTYPE];
 278     uint64_t nr_ops[BDRV_MAX_IOTYPE];
 279     uint64_t total_time_ns[BDRV_MAX_IOTYPE];
 280     uint64_t wr_highest_sector;
 281
 282     /* Whether the disk can expand beyond total_sectors */
 283     int growable;
 284
 285     /* the memory alignment required for the buffers handled by this driver */
 286     int buffer_alignment;
 287
 288     /* do we need to tell the quest if we have a volatile write cache? */
 289     int enable_write_cache;
 290
 291     /* NOTE: the following infos are only hints for real hardware
 292        drivers. They are not used by the block driver */
 293     int cyls, heads, secs, translation;
 294     BlockErrorAction on_read_error, on_write_error;
 295     bool iostatus_enabled;
 296     BlockDeviceIoStatus iostatus;
 297     char device_name[32];
 298     unsigned long *dirty_bitmap;
 299     int64_t dirty_count;
 300     int in_use; /* users other than guest access, eg. block migration */
 301     QTAILQ_ENTRY(BlockDriverState) list;
 302     void *private;
 303
 304     QLIST_HEAD(, BdrvTrackedRequest) tracked_requests;
 305
 306     /* long-running background operation */
 307     BlockJob *job;
 308 };
 309
 310 struct BlockDriverAIOCB {
 311     AIOPool *pool;
 312     BlockDriverState *bs;
 313     BlockDriverCompletionFunc *cb;
 314     void *opaque;
 315     BlockDriverAIOCB *next;
 316 };
 317
 318 void get_tmp_filename(char *filename, int size);
 319
 320 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
 321                    BlockDriverCompletionFunc *cb, void *opaque);
 322 void qemu_aio_release(void *p);
 323
 324 void bdrv_set_io_limits(BlockDriverState *bs,
 325                         BlockIOLimit *io_limits);
 326
 327 #ifdef _WIN32
 328 int is_windows_drive(const char *filename);
 329 #endif
 330
 331 void *block_job_create(const BlockJobType *job_type, BlockDriverState *bs,
 332                        BlockDriverCompletionFunc *cb, void *opaque);
 333 void block_job_complete(BlockJob *job, int ret);
 334 int block_job_set_speed(BlockJob *job, int64_t value);
 335 void block_job_cancel(BlockJob *job);
 336 bool block_job_is_cancelled(BlockJob *job);
 337
 338 int stream_start(BlockDriverState *bs, BlockDriverState *base,
 339                  const char *base_id, BlockDriverCompletionFunc *cb,
 340                  void *opaque);
 341
 342 #endif /* BLOCK_INT_H */