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_DAX_BIO_BASED    = 3,
  30         DM_TYPE_NVME_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, blk_status_t *error);
  76 typedef int (*dm_request_endio_fn) (struct dm_target *ti,
  77                                     struct request *clone, blk_status_t 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                               char *result, unsigned maxlen);
  91
  92 typedef int (*dm_prepare_ioctl_fn) (struct dm_target *ti, struct block_device **bdev);
  93
  94 typedef int (*dm_report_zones_fn) (struct dm_target *ti, sector_t sector,
  95                                    struct blk_zone *zones,
  96                                    unsigned int *nr_zones,
  97                                    gfp_t gfp_mask);
  98
  99 /*
 100  * These iteration functions are typically used to check (and combine)
 101  * properties of underlying devices.
 102  * E.g. Does at least one underlying device support flush?
 103  *      Does any underlying device not support WRITE_SAME?
 104  *
 105  * The callout function is called once for each contiguous section of
 106  * an underlying device.  State can be maintained in *data.
 107  * Return non-zero to stop iterating through any further devices.
 108  */
 109 typedef int (*iterate_devices_callout_fn) (struct dm_target *ti,
 110                                            struct dm_dev *dev,
 111                                            sector_t start, sector_t len,
 112                                            void *data);
 113
 114 /*
 115  * This function must iterate through each section of device used by the
 116  * target until it encounters a non-zero return code, which it then returns.
 117  * Returns zero if no callout returned non-zero.
 118  */
 119 typedef int (*dm_iterate_devices_fn) (struct dm_target *ti,
 120                                       iterate_devices_callout_fn fn,
 121                                       void *data);
 122
 123 typedef void (*dm_io_hints_fn) (struct dm_target *ti,
 124                                 struct queue_limits *limits);
 125
 126 /*
 127  * Returns:
 128  *    0: The target can handle the next I/O immediately.
 129  *    1: The target can't handle the next I/O immediately.
 130  */
 131 typedef int (*dm_busy_fn) (struct dm_target *ti);
 132
 133 /*
 134  * Returns:
 135  *  < 0 : error
 136  * >= 0 : the number of bytes accessible at the address
 137  */
 138 typedef long (*dm_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff,
 139                 long nr_pages, void **kaddr, pfn_t *pfn);
 140 typedef size_t (*dm_dax_copy_iter_fn)(struct dm_target *ti, pgoff_t pgoff,
 141                 void *addr, size_t bytes, struct iov_iter *i);
 142 #define PAGE_SECTORS (PAGE_SIZE / 512)
 143
 144 void dm_error(const char *message);
 145
 146 struct dm_dev {
 147         struct block_device *bdev;
 148         struct dax_device *dax_dev;
 149         fmode_t mode;
 150         char name[16];
 151 };
 152
 153 dev_t dm_get_dev_t(const char *path);
 154
 155 /*
 156  * Constructors should call these functions to ensure destination devices
 157  * are opened/closed correctly.
 158  */
 159 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 160                   struct dm_dev **result);
 161 void dm_put_device(struct dm_target *ti, struct dm_dev *d);
 162
 163 /*
 164  * Information about a target type
 165  */
 166
 167 struct target_type {
 168         uint64_t features;
 169         const char *name;
 170         struct module *module;
 171         unsigned version[3];
 172         dm_ctr_fn ctr;
 173         dm_dtr_fn dtr;
 174         dm_map_fn map;
 175         dm_clone_and_map_request_fn clone_and_map_rq;
 176         dm_release_clone_request_fn release_clone_rq;
 177         dm_endio_fn end_io;
 178         dm_request_endio_fn rq_end_io;
 179         dm_presuspend_fn presuspend;
 180         dm_presuspend_undo_fn presuspend_undo;
 181         dm_postsuspend_fn postsuspend;
 182         dm_preresume_fn preresume;
 183         dm_resume_fn resume;
 184         dm_status_fn status;
 185         dm_message_fn message;
 186         dm_prepare_ioctl_fn prepare_ioctl;
 187 #ifdef CONFIG_BLK_DEV_ZONED
 188         dm_report_zones_fn report_zones;
 189 #endif
 190         dm_busy_fn busy;
 191         dm_iterate_devices_fn iterate_devices;
 192         dm_io_hints_fn io_hints;
 193         dm_dax_direct_access_fn direct_access;
 194         dm_dax_copy_iter_fn dax_copy_from_iter;
 195         dm_dax_copy_iter_fn dax_copy_to_iter;
 196
 197         /* For internal device-mapper use. */
 198         struct list_head list;
 199 };
 200
 201 /*
 202  * Target features
 203  */
 204
 205 /*
 206  * Any table that contains an instance of this target must have only one.
 207  */
 208 #define DM_TARGET_SINGLETON             0x00000001
 209 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
 210
 211 /*
 212  * Indicates that a target does not support read-only devices.
 213  */
 214 #define DM_TARGET_ALWAYS_WRITEABLE      0x00000002
 215 #define dm_target_always_writeable(type) \
 216                 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
 217
 218 /*
 219  * Any device that contains a table with an instance of this target may never
 220  * have tables containing any different target type.
 221  */
 222 #define DM_TARGET_IMMUTABLE             0x00000004
 223 #define dm_target_is_immutable(type)    ((type)->features & DM_TARGET_IMMUTABLE)
 224
 225 /*
 226  * Indicates that a target may replace any target; even immutable targets.
 227  * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
 228  */
 229 #define DM_TARGET_WILDCARD              0x00000008
 230 #define dm_target_is_wildcard(type)     ((type)->features & DM_TARGET_WILDCARD)
 231
 232 /*
 233  * A target implements own bio data integrity.
 234  */
 235 #define DM_TARGET_INTEGRITY             0x00000010
 236 #define dm_target_has_integrity(type)   ((type)->features & DM_TARGET_INTEGRITY)
 237
 238 /*
 239  * A target passes integrity data to the lower device.
 240  */
 241 #define DM_TARGET_PASSES_INTEGRITY      0x00000020
 242 #define dm_target_passes_integrity(type) ((type)->features & DM_TARGET_PASSES_INTEGRITY)
 243
 244 /*
 245  * Indicates that a target supports host-managed zoned block devices.
 246  */
 247 #define DM_TARGET_ZONED_HM              0x00000040
 248 #define dm_target_supports_zoned_hm(type) ((type)->features & DM_TARGET_ZONED_HM)
 249
 250 struct dm_target {
 251         struct dm_table *table;
 252         struct target_type *type;
 253
 254         /* target limits */
 255         sector_t begin;
 256         sector_t len;
 257
 258         /* If non-zero, maximum size of I/O submitted to a target. */
 259         uint32_t max_io_len;
 260
 261         /*
 262          * A number of zero-length barrier bios that will be submitted
 263          * to the target for the purpose of flushing cache.
 264          *
 265          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 266          * It is a responsibility of the target driver to remap these bios
 267          * to the real underlying devices.
 268          */
 269         unsigned num_flush_bios;
 270
 271         /*
 272          * The number of discard 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_discard_bios;
 276
 277         /*
 278          * The number of secure erase bios that will be submitted to the target.
 279          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 280          */
 281         unsigned num_secure_erase_bios;
 282
 283         /*
 284          * The number of WRITE SAME bios that will be submitted to the target.
 285          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 286          */
 287         unsigned num_write_same_bios;
 288
 289         /*
 290          * The number of WRITE ZEROES bios that will be submitted to the target.
 291          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 292          */
 293         unsigned num_write_zeroes_bios;
 294
 295         /*
 296          * The minimum number of extra bytes allocated in each io for the
 297          * target to use.
 298          */
 299         unsigned per_io_data_size;
 300
 301         /* target specific data */
 302         void *private;
 303
 304         /* Used to provide an error string from the ctr */
 305         char *error;
 306
 307         /*
 308          * Set if this target needs to receive flushes regardless of
 309          * whether or not its underlying devices have support.
 310          */
 311         bool flush_supported:1;
 312
 313         /*
 314          * Set if this target needs to receive discards regardless of
 315          * whether or not its underlying devices have support.
 316          */
 317         bool discards_supported:1;
 318
 319         /*
 320          * Set if the target required discard bios to be split
 321          * on max_io_len boundary.
 322          */
 323         bool split_discard_bios:1;
 324 };
 325
 326 /* Each target can link one of these into the table */
 327 struct dm_target_callbacks {
 328         struct list_head list;
 329         int (*congested_fn) (struct dm_target_callbacks *, int);
 330 };
 331
 332 void *dm_per_bio_data(struct bio *bio, size_t data_size);
 333 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size);
 334 unsigned dm_bio_get_target_bio_nr(const struct bio *bio);
 335
 336 int dm_register_target(struct target_type *t);
 337 void dm_unregister_target(struct target_type *t);
 338
 339 /*
 340  * Target argument parsing.
 341  */
 342 struct dm_arg_set {
 343         unsigned argc;
 344         char **argv;
 345 };
 346
 347 /*
 348  * The minimum and maximum value of a numeric argument, together with
 349  * the error message to use if the number is found to be outside that range.
 350  */
 351 struct dm_arg {
 352         unsigned min;
 353         unsigned max;
 354         char *error;
 355 };
 356
 357 /*
 358  * Validate the next argument, either returning it as *value or, if invalid,
 359  * returning -EINVAL and setting *error.
 360  */
 361 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 362                 unsigned *value, char **error);
 363
 364 /*
 365  * Process the next argument as the start of a group containing between
 366  * arg->min and arg->max further arguments. Either return the size as
 367  * *num_args or, if invalid, return -EINVAL and set *error.
 368  */
 369 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 370                       unsigned *num_args, char **error);
 371
 372 /*
 373  * Return the current argument and shift to the next.
 374  */
 375 const char *dm_shift_arg(struct dm_arg_set *as);
 376
 377 /*
 378  * Move through num_args arguments.
 379  */
 380 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
 381
 382 /*-----------------------------------------------------------------
 383  * Functions for creating and manipulating mapped devices.
 384  * Drop the reference with dm_put when you finish with the object.
 385  *---------------------------------------------------------------*/
 386
 387 /*
 388  * DM_ANY_MINOR chooses the next available minor number.
 389  */
 390 #define DM_ANY_MINOR (-1)
 391 int dm_create(int minor, struct mapped_device **md);
 392
 393 /*
 394  * Reference counting for md.
 395  */
 396 struct mapped_device *dm_get_md(dev_t dev);
 397 void dm_get(struct mapped_device *md);
 398 int dm_hold(struct mapped_device *md);
 399 void dm_put(struct mapped_device *md);
 400
 401 /*
 402  * An arbitrary pointer may be stored alongside a mapped device.
 403  */
 404 void dm_set_mdptr(struct mapped_device *md, void *ptr);
 405 void *dm_get_mdptr(struct mapped_device *md);
 406
 407 /*
 408  * A device can still be used while suspended, but I/O is deferred.
 409  */
 410 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
 411 int dm_resume(struct mapped_device *md);
 412
 413 /*
 414  * Event functions.
 415  */
 416 uint32_t dm_get_event_nr(struct mapped_device *md);
 417 int dm_wait_event(struct mapped_device *md, int event_nr);
 418 uint32_t dm_next_uevent_seq(struct mapped_device *md);
 419 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
 420
 421 /*
 422  * Info functions.
 423  */
 424 const char *dm_device_name(struct mapped_device *md);
 425 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
 426 struct gendisk *dm_disk(struct mapped_device *md);
 427 int dm_suspended(struct dm_target *ti);
 428 int dm_noflush_suspending(struct dm_target *ti);
 429 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
 430 void dm_remap_zone_report(struct dm_target *ti, sector_t start,
 431                           struct blk_zone *zones, unsigned int *nr_zones);
 432 union map_info *dm_get_rq_mapinfo(struct request *rq);
 433
 434 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
 435
 436 /*
 437  * Geometry functions.
 438  */
 439 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
 440 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
 441
 442 /*-----------------------------------------------------------------
 443  * Functions for manipulating device-mapper tables.
 444  *---------------------------------------------------------------*/
 445
 446 /*
 447  * First create an empty table.
 448  */
 449 int dm_table_create(struct dm_table **result, fmode_t mode,
 450                     unsigned num_targets, struct mapped_device *md);
 451
 452 /*
 453  * Then call this once for each target.
 454  */
 455 int dm_table_add_target(struct dm_table *t, const char *type,
 456                         sector_t start, sector_t len, char *params);
 457
 458 /*
 459  * Target_ctr should call this if it needs to add any callbacks.
 460  */
 461 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 462
 463 /*
 464  * Target can use this to set the table's type.
 465  * Can only ever be called from a target's ctr.
 466  * Useful for "hybrid" target (supports both bio-based
 467  * and request-based).
 468  */
 469 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type);
 470
 471 /*
 472  * Finally call this to make the table ready for use.
 473  */
 474 int dm_table_complete(struct dm_table *t);
 475
 476 /*
 477  * Destroy the table when finished.
 478  */
 479 void dm_table_destroy(struct dm_table *t);
 480
 481 /*
 482  * Target may require that it is never sent I/O larger than len.
 483  */
 484 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
 485
 486 /*
 487  * Table reference counting.
 488  */
 489 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
 490 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
 491 void dm_sync_table(struct mapped_device *md);
 492
 493 /*
 494  * Queries
 495  */
 496 sector_t dm_table_get_size(struct dm_table *t);
 497 unsigned int dm_table_get_num_targets(struct dm_table *t);
 498 fmode_t dm_table_get_mode(struct dm_table *t);
 499 struct mapped_device *dm_table_get_md(struct dm_table *t);
 500 const char *dm_table_device_name(struct dm_table *t);
 501
 502 /*
 503  * Trigger an event.
 504  */
 505 void dm_table_event(struct dm_table *t);
 506
 507 /*
 508  * Run the queue for request-based targets.
 509  */
 510 void dm_table_run_md_queue_async(struct dm_table *t);
 511
 512 /*
 513  * The device must be suspended before calling this method.
 514  * Returns the previous table, which the caller must destroy.
 515  */
 516 struct dm_table *dm_swap_table(struct mapped_device *md,
 517                                struct dm_table *t);
 518
 519 /*
 520  * A wrapper around vmalloc.
 521  */
 522 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size);
 523
 524 /*-----------------------------------------------------------------
 525  * Macros.
 526  *---------------------------------------------------------------*/
 527 #define DM_NAME "device-mapper"
 528
 529 #define DM_RATELIMIT(pr_func, fmt, ...)                                 \
 530 do {                                                                    \
 531         static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,   \
 532                                       DEFAULT_RATELIMIT_BURST);         \
 533                                                                         \
 534         if (__ratelimit(&rs))                                           \
 535                 pr_func(DM_FMT(fmt), ##__VA_ARGS__);                    \
 536 } while (0)
 537
 538 #define DM_FMT(fmt) DM_NAME ": " DM_MSG_PREFIX ": " fmt "\n"
 539
 540 #define DMCRIT(fmt, ...) pr_crit(DM_FMT(fmt), ##__VA_ARGS__)
 541
 542 #define DMERR(fmt, ...) pr_err(DM_FMT(fmt), ##__VA_ARGS__)
 543 #define DMERR_LIMIT(fmt, ...) DM_RATELIMIT(pr_err, fmt, ##__VA_ARGS__)
 544 #define DMWARN(fmt, ...) pr_warn(DM_FMT(fmt), ##__VA_ARGS__)
 545 #define DMWARN_LIMIT(fmt, ...) DM_RATELIMIT(pr_warn, fmt, ##__VA_ARGS__)
 546 #define DMINFO(fmt, ...) pr_info(DM_FMT(fmt), ##__VA_ARGS__)
 547 #define DMINFO_LIMIT(fmt, ...) DM_RATELIMIT(pr_info, fmt, ##__VA_ARGS__)
 548
 549 #ifdef CONFIG_DM_DEBUG
 550 #define DMDEBUG(fmt, ...) printk(KERN_DEBUG DM_FMT(fmt), ##__VA_ARGS__)
 551 #define DMDEBUG_LIMIT(fmt, ...) DM_RATELIMIT(pr_debug, fmt, ##__VA_ARGS__)
 552 #else
 553 #define DMDEBUG(fmt, ...) no_printk(fmt, ##__VA_ARGS__)
 554 #define DMDEBUG_LIMIT(fmt, ...) no_printk(fmt, ##__VA_ARGS__)
 555 #endif
 556
 557 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
 558                           0 : scnprintf(result + sz, maxlen - sz, x))
 559
 560 /*
 561  * Definitions of return values from target end_io function.
 562  */
 563 #define DM_ENDIO_DONE           0
 564 #define DM_ENDIO_INCOMPLETE     1
 565 #define DM_ENDIO_REQUEUE        2
 566 #define DM_ENDIO_DELAY_REQUEUE  3
 567
 568 /*
 569  * Definitions of return values from target map function.
 570  */
 571 #define DM_MAPIO_SUBMITTED      0
 572 #define DM_MAPIO_REMAPPED       1
 573 #define DM_MAPIO_REQUEUE        DM_ENDIO_REQUEUE
 574 #define DM_MAPIO_DELAY_REQUEUE  DM_ENDIO_DELAY_REQUEUE
 575 #define DM_MAPIO_KILL           4
 576
 577 #define dm_sector_div64(x, y)( \
 578 { \
 579         u64 _res; \
 580         (x) = div64_u64_rem(x, y, &_res); \
 581         _res; \
 582 } \
 583 )
 584
 585 /*
 586  * Ceiling(n / sz)
 587  */
 588 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
 589
 590 #define dm_sector_div_up(n, sz) ( \
 591 { \
 592         sector_t _r = ((n) + (sz) - 1); \
 593         sector_div(_r, (sz)); \
 594         _r; \
 595 } \
 596 )
 597
 598 /*
 599  * ceiling(n / size) * size
 600  */
 601 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
 602
 603 #define dm_array_too_big(fixed, obj, num) \
 604         ((num) > (UINT_MAX - (fixed)) / (obj))
 605
 606 /*
 607  * Sector offset taken relative to the start of the target instead of
 608  * relative to the start of the device.
 609  */
 610 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
 611
 612 static inline sector_t to_sector(unsigned long n)
 613 {
 614         return (n >> SECTOR_SHIFT);
 615 }
 616
 617 static inline unsigned long to_bytes(sector_t n)
 618 {
 619         return (n << SECTOR_SHIFT);
 620 }
 621
 622 #endif  /* _LINUX_DEVICE_MAPPER_H */