include/linux/device-mapper.h

   1 /*
   2  * Copyright (C) 2001 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the LGPL.
   6  */
   7
   8 #ifndef _LINUX_DEVICE_MAPPER_H
   9 #define _LINUX_DEVICE_MAPPER_H
  10
  11 #include <linux/bio.h>
  12 #include <linux/blkdev.h>
  13 #include <linux/math64.h>
  14 #include <linux/ratelimit.h>
  15
  16 struct dm_dev;
  17 struct dm_target;
  18 struct dm_table;
  19 struct mapped_device;
  20 struct bio_vec;
  21
  22 /*
  23  * Type of table, mapped_device's mempool and request_queue
  24  */
  25 enum dm_queue_mode {
  26         DM_TYPE_NONE             = 0,
  27         DM_TYPE_BIO_BASED        = 1,
  28         DM_TYPE_REQUEST_BASED    = 2,
  29         DM_TYPE_MQ_REQUEST_BASED = 3,
  30         DM_TYPE_DAX_BIO_BASED    = 4,
  31 };
  32
  33 typedef enum { STATUSTYPE_INFO, STATUSTYPE_TABLE } status_type_t;
  34
  35 union map_info {
  36         void *ptr;
  37 };
  38
  39 /*
  40  * In the constructor the target parameter will already have the
  41  * table, type, begin and len fields filled in.
  42  */
  43 typedef int (*dm_ctr_fn) (struct dm_target *target,
  44                           unsigned int argc, char **argv);
  45
  46 /*
  47  * The destructor doesn't need to free the dm_target, just
  48  * anything hidden ti->private.
  49  */
  50 typedef void (*dm_dtr_fn) (struct dm_target *ti);
  51
  52 /*
  53  * The map function must return:
  54  * < 0: error
  55  * = 0: The target will handle the io by resubmitting it later
  56  * = 1: simple remap complete
  57  * = 2: The target wants to push back the io
  58  */
  59 typedef int (*dm_map_fn) (struct dm_target *ti, struct bio *bio);
  60 typedef int (*dm_clone_and_map_request_fn) (struct dm_target *ti,
  61                                             struct request *rq,
  62                                             union map_info *map_context,
  63                                             struct request **clone);
  64 typedef void (*dm_release_clone_request_fn) (struct request *clone);
  65
  66 /*
  67  * Returns:
  68  * < 0 : error (currently ignored)
  69  * 0   : ended successfully
  70  * 1   : for some reason the io has still not completed (eg,
  71  *       multipath target might want to requeue a failed io).
  72  * 2   : The target wants to push back the io
  73  */
  74 typedef int (*dm_endio_fn) (struct dm_target *ti,
  75                             struct bio *bio, int error);
  76 typedef int (*dm_request_endio_fn) (struct dm_target *ti,
  77                                     struct request *clone, int error,
  78                                     union map_info *map_context);
  79
  80 typedef void (*dm_presuspend_fn) (struct dm_target *ti);
  81 typedef void (*dm_presuspend_undo_fn) (struct dm_target *ti);
  82 typedef void (*dm_postsuspend_fn) (struct dm_target *ti);
  83 typedef int (*dm_preresume_fn) (struct dm_target *ti);
  84 typedef void (*dm_resume_fn) (struct dm_target *ti);
  85
  86 typedef void (*dm_status_fn) (struct dm_target *ti, status_type_t status_type,
  87                               unsigned status_flags, char *result, unsigned maxlen);
  88
  89 typedef int (*dm_message_fn) (struct dm_target *ti, unsigned argc, char **argv);
  90
  91 typedef int (*dm_prepare_ioctl_fn) (struct dm_target *ti,
  92                             struct block_device **bdev, fmode_t *mode);
  93
  94 /*
  95  * These iteration functions are typically used to check (and combine)
  96  * properties of underlying devices.
  97  * E.g. Does at least one underlying device support flush?
  98  *      Does any underlying device not support WRITE_SAME?
  99  *
 100  * The callout function is called once for each contiguous section of
 101  * an underlying device.  State can be maintained in *data.
 102  * Return non-zero to stop iterating through any further devices.
 103  */
 104 typedef int (*iterate_devices_callout_fn) (struct dm_target *ti,
 105                                            struct dm_dev *dev,
 106                                            sector_t start, sector_t len,
 107                                            void *data);
 108
 109 /*
 110  * This function must iterate through each section of device used by the
 111  * target until it encounters a non-zero return code, which it then returns.
 112  * Returns zero if no callout returned non-zero.
 113  */
 114 typedef int (*dm_iterate_devices_fn) (struct dm_target *ti,
 115                                       iterate_devices_callout_fn fn,
 116                                       void *data);
 117
 118 typedef void (*dm_io_hints_fn) (struct dm_target *ti,
 119                                 struct queue_limits *limits);
 120
 121 /*
 122  * Returns:
 123  *    0: The target can handle the next I/O immediately.
 124  *    1: The target can't handle the next I/O immediately.
 125  */
 126 typedef int (*dm_busy_fn) (struct dm_target *ti);
 127
 128 /*
 129  * Returns:
 130  *  < 0 : error
 131  * >= 0 : the number of bytes accessible at the address
 132  */
 133 typedef long (*dm_direct_access_fn) (struct dm_target *ti, sector_t sector,
 134                                      void **kaddr, pfn_t *pfn, long size);
 135
 136 void dm_error(const char *message);
 137
 138 struct dm_dev {
 139         struct block_device *bdev;
 140         fmode_t mode;
 141         char name[16];
 142 };
 143
 144 dev_t dm_get_dev_t(const char *path);
 145
 146 /*
 147  * Constructors should call these functions to ensure destination devices
 148  * are opened/closed correctly.
 149  */
 150 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 151                   struct dm_dev **result);
 152 void dm_put_device(struct dm_target *ti, struct dm_dev *d);
 153
 154 /*
 155  * Information about a target type
 156  */
 157
 158 struct target_type {
 159         uint64_t features;
 160         const char *name;
 161         struct module *module;
 162         unsigned version[3];
 163         dm_ctr_fn ctr;
 164         dm_dtr_fn dtr;
 165         dm_map_fn map;
 166         dm_clone_and_map_request_fn clone_and_map_rq;
 167         dm_release_clone_request_fn release_clone_rq;
 168         dm_endio_fn end_io;
 169         dm_request_endio_fn rq_end_io;
 170         dm_presuspend_fn presuspend;
 171         dm_presuspend_undo_fn presuspend_undo;
 172         dm_postsuspend_fn postsuspend;
 173         dm_preresume_fn preresume;
 174         dm_resume_fn resume;
 175         dm_status_fn status;
 176         dm_message_fn message;
 177         dm_prepare_ioctl_fn prepare_ioctl;
 178         dm_busy_fn busy;
 179         dm_iterate_devices_fn iterate_devices;
 180         dm_io_hints_fn io_hints;
 181         dm_direct_access_fn direct_access;
 182
 183         /* For internal device-mapper use. */
 184         struct list_head list;
 185 };
 186
 187 /*
 188  * Target features
 189  */
 190
 191 /*
 192  * Any table that contains an instance of this target must have only one.
 193  */
 194 #define DM_TARGET_SINGLETON             0x00000001
 195 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
 196
 197 /*
 198  * Indicates that a target does not support read-only devices.
 199  */
 200 #define DM_TARGET_ALWAYS_WRITEABLE      0x00000002
 201 #define dm_target_always_writeable(type) \
 202                 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
 203
 204 /*
 205  * Any device that contains a table with an instance of this target may never
 206  * have tables containing any different target type.
 207  */
 208 #define DM_TARGET_IMMUTABLE             0x00000004
 209 #define dm_target_is_immutable(type)    ((type)->features & DM_TARGET_IMMUTABLE)
 210
 211 /*
 212  * Indicates that a target may replace any target; even immutable targets.
 213  * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
 214  */
 215 #define DM_TARGET_WILDCARD              0x00000008
 216 #define dm_target_is_wildcard(type)     ((type)->features & DM_TARGET_WILDCARD)
 217
 218 /*
 219  * Some targets need to be sent the same WRITE bio severals times so
 220  * that they can send copies of it to different devices.  This function
 221  * examines any supplied bio and returns the number of copies of it the
 222  * target requires.
 223  */
 224 typedef unsigned (*dm_num_write_bios_fn) (struct dm_target *ti, struct bio *bio);
 225
 226 /*
 227  * A target implements own bio data integrity.
 228  */
 229 #define DM_TARGET_INTEGRITY             0x00000010
 230 #define dm_target_has_integrity(type)   ((type)->features & DM_TARGET_INTEGRITY)
 231
 232 /*
 233  * A target passes integrity data to the lower device.
 234  */
 235 #define DM_TARGET_PASSES_INTEGRITY      0x00000020
 236 #define dm_target_passes_integrity(type) ((type)->features & DM_TARGET_PASSES_INTEGRITY)
 237
 238 struct dm_target {
 239         struct dm_table *table;
 240         struct target_type *type;
 241
 242         /* target limits */
 243         sector_t begin;
 244         sector_t len;
 245
 246         /* If non-zero, maximum size of I/O submitted to a target. */
 247         uint32_t max_io_len;
 248
 249         /*
 250          * A number of zero-length barrier bios that will be submitted
 251          * to the target for the purpose of flushing cache.
 252          *
 253          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 254          * It is a responsibility of the target driver to remap these bios
 255          * to the real underlying devices.
 256          */
 257         unsigned num_flush_bios;
 258
 259         /*
 260          * The number of discard bios that will be submitted to the target.
 261          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 262          */
 263         unsigned num_discard_bios;
 264
 265         /*
 266          * The number of WRITE SAME bios that will be submitted to the target.
 267          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 268          */
 269         unsigned num_write_same_bios;
 270
 271         /*
 272          * The number of WRITE ZEROES bios that will be submitted to the target.
 273          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 274          */
 275         unsigned num_write_zeroes_bios;
 276
 277         /*
 278          * The minimum number of extra bytes allocated in each io for the
 279          * target to use.
 280          */
 281         unsigned per_io_data_size;
 282
 283         /*
 284          * If defined, this function is called to find out how many
 285          * duplicate bios should be sent to the target when writing
 286          * data.
 287          */
 288         dm_num_write_bios_fn num_write_bios;
 289
 290         /* target specific data */
 291         void *private;
 292
 293         /* Used to provide an error string from the ctr */
 294         char *error;
 295
 296         /*
 297          * Set if this target needs to receive flushes regardless of
 298          * whether or not its underlying devices have support.
 299          */
 300         bool flush_supported:1;
 301
 302         /*
 303          * Set if this target needs to receive discards regardless of
 304          * whether or not its underlying devices have support.
 305          */
 306         bool discards_supported:1;
 307
 308         /*
 309          * Set if the target required discard bios to be split
 310          * on max_io_len boundary.
 311          */
 312         bool split_discard_bios:1;
 313 };
 314
 315 /* Each target can link one of these into the table */
 316 struct dm_target_callbacks {
 317         struct list_head list;
 318         int (*congested_fn) (struct dm_target_callbacks *, int);
 319 };
 320
 321 /*
 322  * For bio-based dm.
 323  * One of these is allocated for each bio.
 324  * This structure shouldn't be touched directly by target drivers.
 325  * It is here so that we can inline dm_per_bio_data and
 326  * dm_bio_from_per_bio_data
 327  */
 328 struct dm_target_io {
 329         struct dm_io *io;
 330         struct dm_target *ti;
 331         unsigned target_bio_nr;
 332         unsigned *len_ptr;
 333         struct bio clone;
 334 };
 335
 336 static inline void *dm_per_bio_data(struct bio *bio, size_t data_size)
 337 {
 338         return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
 339 }
 340
 341 static inline struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 342 {
 343         return (struct bio *)((char *)data + data_size + offsetof(struct dm_target_io, clone));
 344 }
 345
 346 static inline unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
 347 {
 348         return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 349 }
 350
 351 int dm_register_target(struct target_type *t);
 352 void dm_unregister_target(struct target_type *t);
 353
 354 /*
 355  * Target argument parsing.
 356  */
 357 struct dm_arg_set {
 358         unsigned argc;
 359         char **argv;
 360 };
 361
 362 /*
 363  * The minimum and maximum value of a numeric argument, together with
 364  * the error message to use if the number is found to be outside that range.
 365  */
 366 struct dm_arg {
 367         unsigned min;
 368         unsigned max;
 369         char *error;
 370 };
 371
 372 /*
 373  * Validate the next argument, either returning it as *value or, if invalid,
 374  * returning -EINVAL and setting *error.
 375  */
 376 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 377                 unsigned *value, char **error);
 378
 379 /*
 380  * Process the next argument as the start of a group containing between
 381  * arg->min and arg->max further arguments. Either return the size as
 382  * *num_args or, if invalid, return -EINVAL and set *error.
 383  */
 384 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 385                       unsigned *num_args, char **error);
 386
 387 /*
 388  * Return the current argument and shift to the next.
 389  */
 390 const char *dm_shift_arg(struct dm_arg_set *as);
 391
 392 /*
 393  * Move through num_args arguments.
 394  */
 395 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
 396
 397 /*-----------------------------------------------------------------
 398  * Functions for creating and manipulating mapped devices.
 399  * Drop the reference with dm_put when you finish with the object.
 400  *---------------------------------------------------------------*/
 401
 402 /*
 403  * DM_ANY_MINOR chooses the next available minor number.
 404  */
 405 #define DM_ANY_MINOR (-1)
 406 int dm_create(int minor, struct mapped_device **md);
 407
 408 /*
 409  * Reference counting for md.
 410  */
 411 struct mapped_device *dm_get_md(dev_t dev);
 412 void dm_get(struct mapped_device *md);
 413 int dm_hold(struct mapped_device *md);
 414 void dm_put(struct mapped_device *md);
 415
 416 /*
 417  * An arbitrary pointer may be stored alongside a mapped device.
 418  */
 419 void dm_set_mdptr(struct mapped_device *md, void *ptr);
 420 void *dm_get_mdptr(struct mapped_device *md);
 421
 422 /*
 423  * A device can still be used while suspended, but I/O is deferred.
 424  */
 425 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
 426 int dm_resume(struct mapped_device *md);
 427
 428 /*
 429  * Event functions.
 430  */
 431 uint32_t dm_get_event_nr(struct mapped_device *md);
 432 int dm_wait_event(struct mapped_device *md, int event_nr);
 433 uint32_t dm_next_uevent_seq(struct mapped_device *md);
 434 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
 435
 436 /*
 437  * Info functions.
 438  */
 439 const char *dm_device_name(struct mapped_device *md);
 440 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
 441 struct gendisk *dm_disk(struct mapped_device *md);
 442 int dm_suspended(struct dm_target *ti);
 443 int dm_noflush_suspending(struct dm_target *ti);
 444 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
 445 union map_info *dm_get_rq_mapinfo(struct request *rq);
 446
 447 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
 448
 449 /*
 450  * Geometry functions.
 451  */
 452 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
 453 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
 454
 455 /*-----------------------------------------------------------------
 456  * Functions for manipulating device-mapper tables.
 457  *---------------------------------------------------------------*/
 458
 459 /*
 460  * First create an empty table.
 461  */
 462 int dm_table_create(struct dm_table **result, fmode_t mode,
 463                     unsigned num_targets, struct mapped_device *md);
 464
 465 /*
 466  * Then call this once for each target.
 467  */
 468 int dm_table_add_target(struct dm_table *t, const char *type,
 469                         sector_t start, sector_t len, char *params);
 470
 471 /*
 472  * Target_ctr should call this if it needs to add any callbacks.
 473  */
 474 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 475
 476 /*
 477  * Target can use this to set the table's type.
 478  * Can only ever be called from a target's ctr.
 479  * Useful for "hybrid" target (supports both bio-based
 480  * and request-based).
 481  */
 482 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type);
 483
 484 /*
 485  * Finally call this to make the table ready for use.
 486  */
 487 int dm_table_complete(struct dm_table *t);
 488
 489 /*
 490  * Target may require that it is never sent I/O larger than len.
 491  */
 492 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
 493
 494 /*
 495  * Table reference counting.
 496  */
 497 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
 498 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
 499 void dm_sync_table(struct mapped_device *md);
 500
 501 /*
 502  * Queries
 503  */
 504 sector_t dm_table_get_size(struct dm_table *t);
 505 unsigned int dm_table_get_num_targets(struct dm_table *t);
 506 fmode_t dm_table_get_mode(struct dm_table *t);
 507 struct mapped_device *dm_table_get_md(struct dm_table *t);
 508
 509 /*
 510  * Trigger an event.
 511  */
 512 void dm_table_event(struct dm_table *t);
 513
 514 /*
 515  * Run the queue for request-based targets.
 516  */
 517 void dm_table_run_md_queue_async(struct dm_table *t);
 518
 519 /*
 520  * The device must be suspended before calling this method.
 521  * Returns the previous table, which the caller must destroy.
 522  */
 523 struct dm_table *dm_swap_table(struct mapped_device *md,
 524                                struct dm_table *t);
 525
 526 /*
 527  * A wrapper around vmalloc.
 528  */
 529 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size);
 530
 531 /*-----------------------------------------------------------------
 532  * Macros.
 533  *---------------------------------------------------------------*/
 534 #define DM_NAME "device-mapper"
 535
 536 #ifdef CONFIG_PRINTK
 537 extern struct ratelimit_state dm_ratelimit_state;
 538
 539 #define dm_ratelimit()  __ratelimit(&dm_ratelimit_state)
 540 #else
 541 #define dm_ratelimit()  0
 542 #endif
 543
 544 #define DMCRIT(f, arg...) \
 545         printk(KERN_CRIT DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 546
 547 #define DMERR(f, arg...) \
 548         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 549 #define DMERR_LIMIT(f, arg...) \
 550         do { \
 551                 if (dm_ratelimit())     \
 552                         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " \
 553                                f "\n", ## arg); \
 554         } while (0)
 555
 556 #define DMWARN(f, arg...) \
 557         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 558 #define DMWARN_LIMIT(f, arg...) \
 559         do { \
 560                 if (dm_ratelimit())     \
 561                         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " \
 562                                f "\n", ## arg); \
 563         } while (0)
 564
 565 #define DMINFO(f, arg...) \
 566         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 567 #define DMINFO_LIMIT(f, arg...) \
 568         do { \
 569                 if (dm_ratelimit())     \
 570                         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f \
 571                                "\n", ## arg); \
 572         } while (0)
 573
 574 #ifdef CONFIG_DM_DEBUG
 575 #  define DMDEBUG(f, arg...) \
 576         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX " DEBUG: " f "\n", ## arg)
 577 #  define DMDEBUG_LIMIT(f, arg...) \
 578         do { \
 579                 if (dm_ratelimit())     \
 580                         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX ": " f \
 581                                "\n", ## arg); \
 582         } while (0)
 583 #else
 584 #  define DMDEBUG(f, arg...) do {} while (0)
 585 #  define DMDEBUG_LIMIT(f, arg...) do {} while (0)
 586 #endif
 587
 588 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
 589                           0 : scnprintf(result + sz, maxlen - sz, x))
 590
 591 #define SECTOR_SHIFT 9
 592
 593 /*
 594  * Definitions of return values from target end_io function.
 595  */
 596 #define DM_ENDIO_DONE           0
 597 #define DM_ENDIO_INCOMPLETE     1
 598 #define DM_ENDIO_REQUEUE        2
 599
 600 /*
 601  * Definitions of return values from target map function.
 602  */
 603 #define DM_MAPIO_SUBMITTED      0
 604 #define DM_MAPIO_REMAPPED       1
 605 #define DM_MAPIO_REQUEUE        DM_ENDIO_REQUEUE
 606 #define DM_MAPIO_DELAY_REQUEUE  3
 607 #define DM_MAPIO_KILL           4
 608
 609 #define dm_sector_div64(x, y)( \
 610 { \
 611         u64 _res; \
 612         (x) = div64_u64_rem(x, y, &_res); \
 613         _res; \
 614 } \
 615 )
 616
 617 /*
 618  * Ceiling(n / sz)
 619  */
 620 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
 621
 622 #define dm_sector_div_up(n, sz) ( \
 623 { \
 624         sector_t _r = ((n) + (sz) - 1); \
 625         sector_div(_r, (sz)); \
 626         _r; \
 627 } \
 628 )
 629
 630 /*
 631  * ceiling(n / size) * size
 632  */
 633 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
 634
 635 #define dm_array_too_big(fixed, obj, num) \
 636         ((num) > (UINT_MAX - (fixed)) / (obj))
 637
 638 /*
 639  * Sector offset taken relative to the start of the target instead of
 640  * relative to the start of the device.
 641  */
 642 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
 643
 644 static inline sector_t to_sector(unsigned long n)
 645 {
 646         return (n >> SECTOR_SHIFT);
 647 }
 648
 649 static inline unsigned long to_bytes(sector_t n)
 650 {
 651         return (n << SECTOR_SHIFT);
 652 }
 653
 654 #endif  /* _LINUX_DEVICE_MAPPER_H */