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 #define DM_TYPE_NONE                    0
  26 #define DM_TYPE_BIO_BASED               1
  27 #define DM_TYPE_REQUEST_BASED           2
  28 #define DM_TYPE_MQ_REQUEST_BASED        3
  29 #define DM_TYPE_DAX_BIO_BASED           4
  30
  31 typedef enum { STATUSTYPE_INFO, STATUSTYPE_TABLE } status_type_t;
  32
  33 union map_info {
  34         void *ptr;
  35 };
  36
  37 /*
  38  * In the constructor the target parameter will already have the
  39  * table, type, begin and len fields filled in.
  40  */
  41 typedef int (*dm_ctr_fn) (struct dm_target *target,
  42                           unsigned int argc, char **argv);
  43
  44 /*
  45  * The destructor doesn't need to free the dm_target, just
  46  * anything hidden ti->private.
  47  */
  48 typedef void (*dm_dtr_fn) (struct dm_target *ti);
  49
  50 /*
  51  * The map function must return:
  52  * < 0: error
  53  * = 0: The target will handle the io by resubmitting it later
  54  * = 1: simple remap complete
  55  * = 2: The target wants to push back the io
  56  */
  57 typedef int (*dm_map_fn) (struct dm_target *ti, struct bio *bio);
  58 typedef int (*dm_map_request_fn) (struct dm_target *ti, struct request *clone,
  59                                   union map_info *map_context);
  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_map_request_fn map_rq;
 167         dm_clone_and_map_request_fn clone_and_map_rq;
 168         dm_release_clone_request_fn release_clone_rq;
 169         dm_endio_fn end_io;
 170         dm_request_endio_fn rq_end_io;
 171         dm_presuspend_fn presuspend;
 172         dm_presuspend_undo_fn presuspend_undo;
 173         dm_postsuspend_fn postsuspend;
 174         dm_preresume_fn preresume;
 175         dm_resume_fn resume;
 176         dm_status_fn status;
 177         dm_message_fn message;
 178         dm_prepare_ioctl_fn prepare_ioctl;
 179         dm_busy_fn busy;
 180         dm_iterate_devices_fn iterate_devices;
 181         dm_io_hints_fn io_hints;
 182         dm_direct_access_fn direct_access;
 183
 184         /* For internal device-mapper use. */
 185         struct list_head list;
 186 };
 187
 188 /*
 189  * Target features
 190  */
 191
 192 /*
 193  * Any table that contains an instance of this target must have only one.
 194  */
 195 #define DM_TARGET_SINGLETON             0x00000001
 196 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
 197
 198 /*
 199  * Indicates that a target does not support read-only devices.
 200  */
 201 #define DM_TARGET_ALWAYS_WRITEABLE      0x00000002
 202 #define dm_target_always_writeable(type) \
 203                 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
 204
 205 /*
 206  * Any device that contains a table with an instance of this target may never
 207  * have tables containing any different target type.
 208  */
 209 #define DM_TARGET_IMMUTABLE             0x00000004
 210 #define dm_target_is_immutable(type)    ((type)->features & DM_TARGET_IMMUTABLE)
 211
 212 /*
 213  * Indicates that a target may replace any target; even immutable targets.
 214  * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
 215  */
 216 #define DM_TARGET_WILDCARD              0x00000008
 217 #define dm_target_is_wildcard(type)     ((type)->features & DM_TARGET_WILDCARD)
 218
 219 /*
 220  * Some targets need to be sent the same WRITE bio severals times so
 221  * that they can send copies of it to different devices.  This function
 222  * examines any supplied bio and returns the number of copies of it the
 223  * target requires.
 224  */
 225 typedef unsigned (*dm_num_write_bios_fn) (struct dm_target *ti, struct bio *bio);
 226
 227 struct dm_target {
 228         struct dm_table *table;
 229         struct target_type *type;
 230
 231         /* target limits */
 232         sector_t begin;
 233         sector_t len;
 234
 235         /* If non-zero, maximum size of I/O submitted to a target. */
 236         uint32_t max_io_len;
 237
 238         /*
 239          * A number of zero-length barrier bios that will be submitted
 240          * to the target for the purpose of flushing cache.
 241          *
 242          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 243          * It is a responsibility of the target driver to remap these bios
 244          * to the real underlying devices.
 245          */
 246         unsigned num_flush_bios;
 247
 248         /*
 249          * The number of discard bios that will be submitted to the target.
 250          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 251          */
 252         unsigned num_discard_bios;
 253
 254         /*
 255          * The number of WRITE SAME bios that will be submitted to the target.
 256          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 257          */
 258         unsigned num_write_same_bios;
 259
 260         /*
 261          * The minimum number of extra bytes allocated in each io for the
 262          * target to use.
 263          */
 264         unsigned per_io_data_size;
 265
 266         /*
 267          * If defined, this function is called to find out how many
 268          * duplicate bios should be sent to the target when writing
 269          * data.
 270          */
 271         dm_num_write_bios_fn num_write_bios;
 272
 273         /* target specific data */
 274         void *private;
 275
 276         /* Used to provide an error string from the ctr */
 277         char *error;
 278
 279         /*
 280          * Set if this target needs to receive flushes regardless of
 281          * whether or not its underlying devices have support.
 282          */
 283         bool flush_supported:1;
 284
 285         /*
 286          * Set if this target needs to receive discards regardless of
 287          * whether or not its underlying devices have support.
 288          */
 289         bool discards_supported:1;
 290
 291         /*
 292          * Set if the target required discard bios to be split
 293          * on max_io_len boundary.
 294          */
 295         bool split_discard_bios:1;
 296
 297         /*
 298          * Set if this target does not return zeroes on discarded blocks.
 299          */
 300         bool discard_zeroes_data_unsupported:1;
 301 };
 302
 303 /* Each target can link one of these into the table */
 304 struct dm_target_callbacks {
 305         struct list_head list;
 306         int (*congested_fn) (struct dm_target_callbacks *, int);
 307 };
 308
 309 /*
 310  * For bio-based dm.
 311  * One of these is allocated for each bio.
 312  * This structure shouldn't be touched directly by target drivers.
 313  * It is here so that we can inline dm_per_bio_data and
 314  * dm_bio_from_per_bio_data
 315  */
 316 struct dm_target_io {
 317         struct dm_io *io;
 318         struct dm_target *ti;
 319         unsigned target_bio_nr;
 320         unsigned *len_ptr;
 321         struct bio clone;
 322 };
 323
 324 static inline void *dm_per_bio_data(struct bio *bio, size_t data_size)
 325 {
 326         return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
 327 }
 328
 329 static inline struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 330 {
 331         return (struct bio *)((char *)data + data_size + offsetof(struct dm_target_io, clone));
 332 }
 333
 334 static inline unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
 335 {
 336         return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 337 }
 338
 339 int dm_register_target(struct target_type *t);
 340 void dm_unregister_target(struct target_type *t);
 341
 342 /*
 343  * Target argument parsing.
 344  */
 345 struct dm_arg_set {
 346         unsigned argc;
 347         char **argv;
 348 };
 349
 350 /*
 351  * The minimum and maximum value of a numeric argument, together with
 352  * the error message to use if the number is found to be outside that range.
 353  */
 354 struct dm_arg {
 355         unsigned min;
 356         unsigned max;
 357         char *error;
 358 };
 359
 360 /*
 361  * Validate the next argument, either returning it as *value or, if invalid,
 362  * returning -EINVAL and setting *error.
 363  */
 364 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 365                 unsigned *value, char **error);
 366
 367 /*
 368  * Process the next argument as the start of a group containing between
 369  * arg->min and arg->max further arguments. Either return the size as
 370  * *num_args or, if invalid, return -EINVAL and set *error.
 371  */
 372 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 373                       unsigned *num_args, char **error);
 374
 375 /*
 376  * Return the current argument and shift to the next.
 377  */
 378 const char *dm_shift_arg(struct dm_arg_set *as);
 379
 380 /*
 381  * Move through num_args arguments.
 382  */
 383 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
 384
 385 /*-----------------------------------------------------------------
 386  * Functions for creating and manipulating mapped devices.
 387  * Drop the reference with dm_put when you finish with the object.
 388  *---------------------------------------------------------------*/
 389
 390 /*
 391  * DM_ANY_MINOR chooses the next available minor number.
 392  */
 393 #define DM_ANY_MINOR (-1)
 394 int dm_create(int minor, struct mapped_device **md);
 395
 396 /*
 397  * Reference counting for md.
 398  */
 399 struct mapped_device *dm_get_md(dev_t dev);
 400 void dm_get(struct mapped_device *md);
 401 int dm_hold(struct mapped_device *md);
 402 void dm_put(struct mapped_device *md);
 403
 404 /*
 405  * An arbitrary pointer may be stored alongside a mapped device.
 406  */
 407 void dm_set_mdptr(struct mapped_device *md, void *ptr);
 408 void *dm_get_mdptr(struct mapped_device *md);
 409
 410 /*
 411  * A device can still be used while suspended, but I/O is deferred.
 412  */
 413 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
 414 int dm_resume(struct mapped_device *md);
 415
 416 /*
 417  * Event functions.
 418  */
 419 uint32_t dm_get_event_nr(struct mapped_device *md);
 420 int dm_wait_event(struct mapped_device *md, int event_nr);
 421 uint32_t dm_next_uevent_seq(struct mapped_device *md);
 422 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
 423
 424 /*
 425  * Info functions.
 426  */
 427 const char *dm_device_name(struct mapped_device *md);
 428 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
 429 struct gendisk *dm_disk(struct mapped_device *md);
 430 int dm_suspended(struct dm_target *ti);
 431 int dm_noflush_suspending(struct dm_target *ti);
 432 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
 433 union map_info *dm_get_rq_mapinfo(struct request *rq);
 434
 435 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
 436
 437 /*
 438  * Geometry functions.
 439  */
 440 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
 441 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
 442
 443 /*-----------------------------------------------------------------
 444  * Functions for manipulating device-mapper tables.
 445  *---------------------------------------------------------------*/
 446
 447 /*
 448  * First create an empty table.
 449  */
 450 int dm_table_create(struct dm_table **result, fmode_t mode,
 451                     unsigned num_targets, struct mapped_device *md);
 452
 453 /*
 454  * Then call this once for each target.
 455  */
 456 int dm_table_add_target(struct dm_table *t, const char *type,
 457                         sector_t start, sector_t len, char *params);
 458
 459 /*
 460  * Target_ctr should call this if it needs to add any callbacks.
 461  */
 462 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 463
 464 /*
 465  * Target can use this to set the table's type.
 466  * Can only ever be called from a target's ctr.
 467  * Useful for "hybrid" target (supports both bio-based
 468  * and request-based).
 469  */
 470 void dm_table_set_type(struct dm_table *t, unsigned type);
 471
 472 /*
 473  * Finally call this to make the table ready for use.
 474  */
 475 int dm_table_complete(struct dm_table *t);
 476
 477 /*
 478  * Target may require that it is never sent I/O larger than len.
 479  */
 480 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
 481
 482 /*
 483  * Table reference counting.
 484  */
 485 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
 486 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
 487 void dm_sync_table(struct mapped_device *md);
 488
 489 /*
 490  * Queries
 491  */
 492 sector_t dm_table_get_size(struct dm_table *t);
 493 unsigned int dm_table_get_num_targets(struct dm_table *t);
 494 fmode_t dm_table_get_mode(struct dm_table *t);
 495 struct mapped_device *dm_table_get_md(struct dm_table *t);
 496
 497 /*
 498  * Trigger an event.
 499  */
 500 void dm_table_event(struct dm_table *t);
 501
 502 /*
 503  * Run the queue for request-based targets.
 504  */
 505 void dm_table_run_md_queue_async(struct dm_table *t);
 506
 507 /*
 508  * The device must be suspended before calling this method.
 509  * Returns the previous table, which the caller must destroy.
 510  */
 511 struct dm_table *dm_swap_table(struct mapped_device *md,
 512                                struct dm_table *t);
 513
 514 /*
 515  * A wrapper around vmalloc.
 516  */
 517 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size);
 518
 519 /*-----------------------------------------------------------------
 520  * Macros.
 521  *---------------------------------------------------------------*/
 522 #define DM_NAME "device-mapper"
 523
 524 #ifdef CONFIG_PRINTK
 525 extern struct ratelimit_state dm_ratelimit_state;
 526
 527 #define dm_ratelimit()  __ratelimit(&dm_ratelimit_state)
 528 #else
 529 #define dm_ratelimit()  0
 530 #endif
 531
 532 #define DMCRIT(f, arg...) \
 533         printk(KERN_CRIT DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 534
 535 #define DMERR(f, arg...) \
 536         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 537 #define DMERR_LIMIT(f, arg...) \
 538         do { \
 539                 if (dm_ratelimit())     \
 540                         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " \
 541                                f "\n", ## arg); \
 542         } while (0)
 543
 544 #define DMWARN(f, arg...) \
 545         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 546 #define DMWARN_LIMIT(f, arg...) \
 547         do { \
 548                 if (dm_ratelimit())     \
 549                         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " \
 550                                f "\n", ## arg); \
 551         } while (0)
 552
 553 #define DMINFO(f, arg...) \
 554         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 555 #define DMINFO_LIMIT(f, arg...) \
 556         do { \
 557                 if (dm_ratelimit())     \
 558                         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f \
 559                                "\n", ## arg); \
 560         } while (0)
 561
 562 #ifdef CONFIG_DM_DEBUG
 563 #  define DMDEBUG(f, arg...) \
 564         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX " DEBUG: " f "\n", ## arg)
 565 #  define DMDEBUG_LIMIT(f, arg...) \
 566         do { \
 567                 if (dm_ratelimit())     \
 568                         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX ": " f \
 569                                "\n", ## arg); \
 570         } while (0)
 571 #else
 572 #  define DMDEBUG(f, arg...) do {} while (0)
 573 #  define DMDEBUG_LIMIT(f, arg...) do {} while (0)
 574 #endif
 575
 576 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
 577                           0 : scnprintf(result + sz, maxlen - sz, x))
 578
 579 #define SECTOR_SHIFT 9
 580
 581 /*
 582  * Definitions of return values from target end_io function.
 583  */
 584 #define DM_ENDIO_INCOMPLETE     1
 585 #define DM_ENDIO_REQUEUE        2
 586
 587 /*
 588  * Definitions of return values from target map function.
 589  */
 590 #define DM_MAPIO_SUBMITTED      0
 591 #define DM_MAPIO_REMAPPED       1
 592 #define DM_MAPIO_REQUEUE        DM_ENDIO_REQUEUE
 593
 594 #define dm_sector_div64(x, y)( \
 595 { \
 596         u64 _res; \
 597         (x) = div64_u64_rem(x, y, &_res); \
 598         _res; \
 599 } \
 600 )
 601
 602 /*
 603  * Ceiling(n / sz)
 604  */
 605 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
 606
 607 #define dm_sector_div_up(n, sz) ( \
 608 { \
 609         sector_t _r = ((n) + (sz) - 1); \
 610         sector_div(_r, (sz)); \
 611         _r; \
 612 } \
 613 )
 614
 615 /*
 616  * ceiling(n / size) * size
 617  */
 618 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
 619
 620 #define dm_array_too_big(fixed, obj, num) \
 621         ((num) > (UINT_MAX - (fixed)) / (obj))
 622
 623 /*
 624  * Sector offset taken relative to the start of the target instead of
 625  * relative to the start of the device.
 626  */
 627 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
 628
 629 static inline sector_t to_sector(unsigned long n)
 630 {
 631         return (n >> SECTOR_SHIFT);
 632 }
 633
 634 static inline unsigned long to_bytes(sector_t n)
 635 {
 636         return (n << SECTOR_SHIFT);
 637 }
 638
 639 #endif  /* _LINUX_DEVICE_MAPPER_H */