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_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff,
 134                 long nr_pages, void **kaddr, pfn_t *pfn);
 135 #define PAGE_SECTORS (PAGE_SIZE / 512)
 136
 137 void dm_error(const char *message);
 138
 139 struct dm_dev {
 140         struct block_device *bdev;
 141         struct dax_device *dax_dev;
 142         fmode_t mode;
 143         char name[16];
 144 };
 145
 146 dev_t dm_get_dev_t(const char *path);
 147
 148 /*
 149  * Constructors should call these functions to ensure destination devices
 150  * are opened/closed correctly.
 151  */
 152 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 153                   struct dm_dev **result);
 154 void dm_put_device(struct dm_target *ti, struct dm_dev *d);
 155
 156 /*
 157  * Information about a target type
 158  */
 159
 160 struct target_type {
 161         uint64_t features;
 162         const char *name;
 163         struct module *module;
 164         unsigned version[3];
 165         dm_ctr_fn ctr;
 166         dm_dtr_fn dtr;
 167         dm_map_fn map;
 168         dm_clone_and_map_request_fn clone_and_map_rq;
 169         dm_release_clone_request_fn release_clone_rq;
 170         dm_endio_fn end_io;
 171         dm_request_endio_fn rq_end_io;
 172         dm_presuspend_fn presuspend;
 173         dm_presuspend_undo_fn presuspend_undo;
 174         dm_postsuspend_fn postsuspend;
 175         dm_preresume_fn preresume;
 176         dm_resume_fn resume;
 177         dm_status_fn status;
 178         dm_message_fn message;
 179         dm_prepare_ioctl_fn prepare_ioctl;
 180         dm_busy_fn busy;
 181         dm_iterate_devices_fn iterate_devices;
 182         dm_io_hints_fn io_hints;
 183         dm_dax_direct_access_fn direct_access;
 184
 185         /* For internal device-mapper use. */
 186         struct list_head list;
 187 };
 188
 189 /*
 190  * Target features
 191  */
 192
 193 /*
 194  * Any table that contains an instance of this target must have only one.
 195  */
 196 #define DM_TARGET_SINGLETON             0x00000001
 197 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
 198
 199 /*
 200  * Indicates that a target does not support read-only devices.
 201  */
 202 #define DM_TARGET_ALWAYS_WRITEABLE      0x00000002
 203 #define dm_target_always_writeable(type) \
 204                 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
 205
 206 /*
 207  * Any device that contains a table with an instance of this target may never
 208  * have tables containing any different target type.
 209  */
 210 #define DM_TARGET_IMMUTABLE             0x00000004
 211 #define dm_target_is_immutable(type)    ((type)->features & DM_TARGET_IMMUTABLE)
 212
 213 /*
 214  * Indicates that a target may replace any target; even immutable targets.
 215  * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
 216  */
 217 #define DM_TARGET_WILDCARD              0x00000008
 218 #define dm_target_is_wildcard(type)     ((type)->features & DM_TARGET_WILDCARD)
 219
 220 /*
 221  * Some targets need to be sent the same WRITE bio severals times so
 222  * that they can send copies of it to different devices.  This function
 223  * examines any supplied bio and returns the number of copies of it the
 224  * target requires.
 225  */
 226 typedef unsigned (*dm_num_write_bios_fn) (struct dm_target *ti, struct bio *bio);
 227
 228 /*
 229  * A target implements own bio data integrity.
 230  */
 231 #define DM_TARGET_INTEGRITY             0x00000010
 232 #define dm_target_has_integrity(type)   ((type)->features & DM_TARGET_INTEGRITY)
 233
 234 /*
 235  * A target passes integrity data to the lower device.
 236  */
 237 #define DM_TARGET_PASSES_INTEGRITY      0x00000020
 238 #define dm_target_passes_integrity(type) ((type)->features & DM_TARGET_PASSES_INTEGRITY)
 239
 240 struct dm_target {
 241         struct dm_table *table;
 242         struct target_type *type;
 243
 244         /* target limits */
 245         sector_t begin;
 246         sector_t len;
 247
 248         /* If non-zero, maximum size of I/O submitted to a target. */
 249         uint32_t max_io_len;
 250
 251         /*
 252          * A number of zero-length barrier bios that will be submitted
 253          * to the target for the purpose of flushing cache.
 254          *
 255          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 256          * It is a responsibility of the target driver to remap these bios
 257          * to the real underlying devices.
 258          */
 259         unsigned num_flush_bios;
 260
 261         /*
 262          * The number of discard bios that will be submitted to the target.
 263          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 264          */
 265         unsigned num_discard_bios;
 266
 267         /*
 268          * The number of WRITE SAME bios that will be submitted to the target.
 269          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 270          */
 271         unsigned num_write_same_bios;
 272
 273         /*
 274          * The number of WRITE ZEROES bios that will be submitted to the target.
 275          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 276          */
 277         unsigned num_write_zeroes_bios;
 278
 279         /*
 280          * The minimum number of extra bytes allocated in each io for the
 281          * target to use.
 282          */
 283         unsigned per_io_data_size;
 284
 285         /*
 286          * If defined, this function is called to find out how many
 287          * duplicate bios should be sent to the target when writing
 288          * data.
 289          */
 290         dm_num_write_bios_fn num_write_bios;
 291
 292         /* target specific data */
 293         void *private;
 294
 295         /* Used to provide an error string from the ctr */
 296         char *error;
 297
 298         /*
 299          * Set if this target needs to receive flushes regardless of
 300          * whether or not its underlying devices have support.
 301          */
 302         bool flush_supported:1;
 303
 304         /*
 305          * Set if this target needs to receive discards regardless of
 306          * whether or not its underlying devices have support.
 307          */
 308         bool discards_supported:1;
 309
 310         /*
 311          * Set if the target required discard bios to be split
 312          * on max_io_len boundary.
 313          */
 314         bool split_discard_bios:1;
 315 };
 316
 317 /* Each target can link one of these into the table */
 318 struct dm_target_callbacks {
 319         struct list_head list;
 320         int (*congested_fn) (struct dm_target_callbacks *, int);
 321 };
 322
 323 /*
 324  * For bio-based dm.
 325  * One of these is allocated for each bio.
 326  * This structure shouldn't be touched directly by target drivers.
 327  * It is here so that we can inline dm_per_bio_data and
 328  * dm_bio_from_per_bio_data
 329  */
 330 struct dm_target_io {
 331         struct dm_io *io;
 332         struct dm_target *ti;
 333         unsigned target_bio_nr;
 334         unsigned *len_ptr;
 335         struct bio clone;
 336 };
 337
 338 static inline void *dm_per_bio_data(struct bio *bio, size_t data_size)
 339 {
 340         return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
 341 }
 342
 343 static inline struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 344 {
 345         return (struct bio *)((char *)data + data_size + offsetof(struct dm_target_io, clone));
 346 }
 347
 348 static inline unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
 349 {
 350         return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 351 }
 352
 353 int dm_register_target(struct target_type *t);
 354 void dm_unregister_target(struct target_type *t);
 355
 356 /*
 357  * Target argument parsing.
 358  */
 359 struct dm_arg_set {
 360         unsigned argc;
 361         char **argv;
 362 };
 363
 364 /*
 365  * The minimum and maximum value of a numeric argument, together with
 366  * the error message to use if the number is found to be outside that range.
 367  */
 368 struct dm_arg {
 369         unsigned min;
 370         unsigned max;
 371         char *error;
 372 };
 373
 374 /*
 375  * Validate the next argument, either returning it as *value or, if invalid,
 376  * returning -EINVAL and setting *error.
 377  */
 378 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 379                 unsigned *value, char **error);
 380
 381 /*
 382  * Process the next argument as the start of a group containing between
 383  * arg->min and arg->max further arguments. Either return the size as
 384  * *num_args or, if invalid, return -EINVAL and set *error.
 385  */
 386 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 387                       unsigned *num_args, char **error);
 388
 389 /*
 390  * Return the current argument and shift to the next.
 391  */
 392 const char *dm_shift_arg(struct dm_arg_set *as);
 393
 394 /*
 395  * Move through num_args arguments.
 396  */
 397 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
 398
 399 /*-----------------------------------------------------------------
 400  * Functions for creating and manipulating mapped devices.
 401  * Drop the reference with dm_put when you finish with the object.
 402  *---------------------------------------------------------------*/
 403
 404 /*
 405  * DM_ANY_MINOR chooses the next available minor number.
 406  */
 407 #define DM_ANY_MINOR (-1)
 408 int dm_create(int minor, struct mapped_device **md);
 409
 410 /*
 411  * Reference counting for md.
 412  */
 413 struct mapped_device *dm_get_md(dev_t dev);
 414 void dm_get(struct mapped_device *md);
 415 int dm_hold(struct mapped_device *md);
 416 void dm_put(struct mapped_device *md);
 417
 418 /*
 419  * An arbitrary pointer may be stored alongside a mapped device.
 420  */
 421 void dm_set_mdptr(struct mapped_device *md, void *ptr);
 422 void *dm_get_mdptr(struct mapped_device *md);
 423
 424 /*
 425  * A device can still be used while suspended, but I/O is deferred.
 426  */
 427 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
 428 int dm_resume(struct mapped_device *md);
 429
 430 /*
 431  * Event functions.
 432  */
 433 uint32_t dm_get_event_nr(struct mapped_device *md);
 434 int dm_wait_event(struct mapped_device *md, int event_nr);
 435 uint32_t dm_next_uevent_seq(struct mapped_device *md);
 436 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
 437
 438 /*
 439  * Info functions.
 440  */
 441 const char *dm_device_name(struct mapped_device *md);
 442 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
 443 struct gendisk *dm_disk(struct mapped_device *md);
 444 int dm_suspended(struct dm_target *ti);
 445 int dm_noflush_suspending(struct dm_target *ti);
 446 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
 447 union map_info *dm_get_rq_mapinfo(struct request *rq);
 448
 449 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
 450
 451 /*
 452  * Geometry functions.
 453  */
 454 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
 455 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
 456
 457 /*-----------------------------------------------------------------
 458  * Functions for manipulating device-mapper tables.
 459  *---------------------------------------------------------------*/
 460
 461 /*
 462  * First create an empty table.
 463  */
 464 int dm_table_create(struct dm_table **result, fmode_t mode,
 465                     unsigned num_targets, struct mapped_device *md);
 466
 467 /*
 468  * Then call this once for each target.
 469  */
 470 int dm_table_add_target(struct dm_table *t, const char *type,
 471                         sector_t start, sector_t len, char *params);
 472
 473 /*
 474  * Target_ctr should call this if it needs to add any callbacks.
 475  */
 476 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 477
 478 /*
 479  * Target can use this to set the table's type.
 480  * Can only ever be called from a target's ctr.
 481  * Useful for "hybrid" target (supports both bio-based
 482  * and request-based).
 483  */
 484 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type);
 485
 486 /*
 487  * Finally call this to make the table ready for use.
 488  */
 489 int dm_table_complete(struct dm_table *t);
 490
 491 /*
 492  * Target may require that it is never sent I/O larger than len.
 493  */
 494 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
 495
 496 /*
 497  * Table reference counting.
 498  */
 499 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
 500 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
 501 void dm_sync_table(struct mapped_device *md);
 502
 503 /*
 504  * Queries
 505  */
 506 sector_t dm_table_get_size(struct dm_table *t);
 507 unsigned int dm_table_get_num_targets(struct dm_table *t);
 508 fmode_t dm_table_get_mode(struct dm_table *t);
 509 struct mapped_device *dm_table_get_md(struct dm_table *t);
 510
 511 /*
 512  * Trigger an event.
 513  */
 514 void dm_table_event(struct dm_table *t);
 515
 516 /*
 517  * Run the queue for request-based targets.
 518  */
 519 void dm_table_run_md_queue_async(struct dm_table *t);
 520
 521 /*
 522  * The device must be suspended before calling this method.
 523  * Returns the previous table, which the caller must destroy.
 524  */
 525 struct dm_table *dm_swap_table(struct mapped_device *md,
 526                                struct dm_table *t);
 527
 528 /*
 529  * A wrapper around vmalloc.
 530  */
 531 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size);
 532
 533 /*-----------------------------------------------------------------
 534  * Macros.
 535  *---------------------------------------------------------------*/
 536 #define DM_NAME "device-mapper"
 537
 538 #ifdef CONFIG_PRINTK
 539 extern struct ratelimit_state dm_ratelimit_state;
 540
 541 #define dm_ratelimit()  __ratelimit(&dm_ratelimit_state)
 542 #else
 543 #define dm_ratelimit()  0
 544 #endif
 545
 546 #define DMCRIT(f, arg...) \
 547         printk(KERN_CRIT DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 548
 549 #define DMERR(f, arg...) \
 550         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 551 #define DMERR_LIMIT(f, arg...) \
 552         do { \
 553                 if (dm_ratelimit())     \
 554                         printk(KERN_ERR DM_NAME ": " DM_MSG_PREFIX ": " \
 555                                f "\n", ## arg); \
 556         } while (0)
 557
 558 #define DMWARN(f, arg...) \
 559         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 560 #define DMWARN_LIMIT(f, arg...) \
 561         do { \
 562                 if (dm_ratelimit())     \
 563                         printk(KERN_WARNING DM_NAME ": " DM_MSG_PREFIX ": " \
 564                                f "\n", ## arg); \
 565         } while (0)
 566
 567 #define DMINFO(f, arg...) \
 568         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f "\n", ## arg)
 569 #define DMINFO_LIMIT(f, arg...) \
 570         do { \
 571                 if (dm_ratelimit())     \
 572                         printk(KERN_INFO DM_NAME ": " DM_MSG_PREFIX ": " f \
 573                                "\n", ## arg); \
 574         } while (0)
 575
 576 #ifdef CONFIG_DM_DEBUG
 577 #  define DMDEBUG(f, arg...) \
 578         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX " DEBUG: " f "\n", ## arg)
 579 #  define DMDEBUG_LIMIT(f, arg...) \
 580         do { \
 581                 if (dm_ratelimit())     \
 582                         printk(KERN_DEBUG DM_NAME ": " DM_MSG_PREFIX ": " f \
 583                                "\n", ## arg); \
 584         } while (0)
 585 #else
 586 #  define DMDEBUG(f, arg...) do {} while (0)
 587 #  define DMDEBUG_LIMIT(f, arg...) do {} while (0)
 588 #endif
 589
 590 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
 591                           0 : scnprintf(result + sz, maxlen - sz, x))
 592
 593 #define SECTOR_SHIFT 9
 594
 595 /*
 596  * Definitions of return values from target end_io function.
 597  */
 598 #define DM_ENDIO_DONE           0
 599 #define DM_ENDIO_INCOMPLETE     1
 600 #define DM_ENDIO_REQUEUE        2
 601
 602 /*
 603  * Definitions of return values from target map function.
 604  */
 605 #define DM_MAPIO_SUBMITTED      0
 606 #define DM_MAPIO_REMAPPED       1
 607 #define DM_MAPIO_REQUEUE        DM_ENDIO_REQUEUE
 608 #define DM_MAPIO_DELAY_REQUEUE  3
 609 #define DM_MAPIO_KILL           4
 610
 611 #define dm_sector_div64(x, y)( \
 612 { \
 613         u64 _res; \
 614         (x) = div64_u64_rem(x, y, &_res); \
 615         _res; \
 616 } \
 617 )
 618
 619 /*
 620  * Ceiling(n / sz)
 621  */
 622 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
 623
 624 #define dm_sector_div_up(n, sz) ( \
 625 { \
 626         sector_t _r = ((n) + (sz) - 1); \
 627         sector_div(_r, (sz)); \
 628         _r; \
 629 } \
 630 )
 631
 632 /*
 633  * ceiling(n / size) * size
 634  */
 635 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
 636
 637 #define dm_array_too_big(fixed, obj, num) \
 638         ((num) > (UINT_MAX - (fixed)) / (obj))
 639
 640 /*
 641  * Sector offset taken relative to the start of the target instead of
 642  * relative to the start of the device.
 643  */
 644 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
 645
 646 static inline sector_t to_sector(unsigned long n)
 647 {
 648         return (n >> SECTOR_SHIFT);
 649 }
 650
 651 static inline unsigned long to_bytes(sector_t n)
 652 {
 653         return (n << SECTOR_SHIFT);
 654 }
 655
 656 #endif  /* _LINUX_DEVICE_MAPPER_H */