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         DM_TYPE_NVME_BIO_BASED   = 5,
  32 };
  33
  34 typedef enum { STATUSTYPE_INFO, STATUSTYPE_TABLE } status_type_t;
  35
  36 union map_info {
  37         void *ptr;
  38 };
  39
  40 /*
  41  * In the constructor the target parameter will already have the
  42  * table, type, begin and len fields filled in.
  43  */
  44 typedef int (*dm_ctr_fn) (struct dm_target *target,
  45                           unsigned int argc, char **argv);
  46
  47 /*
  48  * The destructor doesn't need to free the dm_target, just
  49  * anything hidden ti->private.
  50  */
  51 typedef void (*dm_dtr_fn) (struct dm_target *ti);
  52
  53 /*
  54  * The map function must return:
  55  * < 0: error
  56  * = 0: The target will handle the io by resubmitting it later
  57  * = 1: simple remap complete
  58  * = 2: The target wants to push back the io
  59  */
  60 typedef int (*dm_map_fn) (struct dm_target *ti, struct bio *bio);
  61 typedef int (*dm_clone_and_map_request_fn) (struct dm_target *ti,
  62                                             struct request *rq,
  63                                             union map_info *map_context,
  64                                             struct request **clone);
  65 typedef void (*dm_release_clone_request_fn) (struct request *clone);
  66
  67 /*
  68  * Returns:
  69  * < 0 : error (currently ignored)
  70  * 0   : ended successfully
  71  * 1   : for some reason the io has still not completed (eg,
  72  *       multipath target might want to requeue a failed io).
  73  * 2   : The target wants to push back the io
  74  */
  75 typedef int (*dm_endio_fn) (struct dm_target *ti,
  76                             struct bio *bio, blk_status_t *error);
  77 typedef int (*dm_request_endio_fn) (struct dm_target *ti,
  78                                     struct request *clone, blk_status_t error,
  79                                     union map_info *map_context);
  80
  81 typedef void (*dm_presuspend_fn) (struct dm_target *ti);
  82 typedef void (*dm_presuspend_undo_fn) (struct dm_target *ti);
  83 typedef void (*dm_postsuspend_fn) (struct dm_target *ti);
  84 typedef int (*dm_preresume_fn) (struct dm_target *ti);
  85 typedef void (*dm_resume_fn) (struct dm_target *ti);
  86
  87 typedef void (*dm_status_fn) (struct dm_target *ti, status_type_t status_type,
  88                               unsigned status_flags, char *result, unsigned maxlen);
  89
  90 typedef int (*dm_message_fn) (struct dm_target *ti, unsigned argc, char **argv);
  91
  92 typedef int (*dm_prepare_ioctl_fn) (struct dm_target *ti,
  93                             struct block_device **bdev, fmode_t *mode);
  94
  95 /*
  96  * These iteration functions are typically used to check (and combine)
  97  * properties of underlying devices.
  98  * E.g. Does at least one underlying device support flush?
  99  *      Does any underlying device not support WRITE_SAME?
 100  *
 101  * The callout function is called once for each contiguous section of
 102  * an underlying device.  State can be maintained in *data.
 103  * Return non-zero to stop iterating through any further devices.
 104  */
 105 typedef int (*iterate_devices_callout_fn) (struct dm_target *ti,
 106                                            struct dm_dev *dev,
 107                                            sector_t start, sector_t len,
 108                                            void *data);
 109
 110 /*
 111  * This function must iterate through each section of device used by the
 112  * target until it encounters a non-zero return code, which it then returns.
 113  * Returns zero if no callout returned non-zero.
 114  */
 115 typedef int (*dm_iterate_devices_fn) (struct dm_target *ti,
 116                                       iterate_devices_callout_fn fn,
 117                                       void *data);
 118
 119 typedef void (*dm_io_hints_fn) (struct dm_target *ti,
 120                                 struct queue_limits *limits);
 121
 122 /*
 123  * Returns:
 124  *    0: The target can handle the next I/O immediately.
 125  *    1: The target can't handle the next I/O immediately.
 126  */
 127 typedef int (*dm_busy_fn) (struct dm_target *ti);
 128
 129 /*
 130  * Returns:
 131  *  < 0 : error
 132  * >= 0 : the number of bytes accessible at the address
 133  */
 134 typedef long (*dm_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff,
 135                 long nr_pages, void **kaddr, pfn_t *pfn);
 136 typedef size_t (*dm_dax_copy_from_iter_fn)(struct dm_target *ti, pgoff_t pgoff,
 137                 void *addr, size_t bytes, struct iov_iter *i);
 138 #define PAGE_SECTORS (PAGE_SIZE / 512)
 139
 140 void dm_error(const char *message);
 141
 142 struct dm_dev {
 143         struct block_device *bdev;
 144         struct dax_device *dax_dev;
 145         fmode_t mode;
 146         char name[16];
 147 };
 148
 149 dev_t dm_get_dev_t(const char *path);
 150
 151 /*
 152  * Constructors should call these functions to ensure destination devices
 153  * are opened/closed correctly.
 154  */
 155 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 156                   struct dm_dev **result);
 157 void dm_put_device(struct dm_target *ti, struct dm_dev *d);
 158
 159 /*
 160  * Information about a target type
 161  */
 162
 163 struct target_type {
 164         uint64_t features;
 165         const char *name;
 166         struct module *module;
 167         unsigned version[3];
 168         dm_ctr_fn ctr;
 169         dm_dtr_fn dtr;
 170         dm_map_fn map;
 171         dm_clone_and_map_request_fn clone_and_map_rq;
 172         dm_release_clone_request_fn release_clone_rq;
 173         dm_endio_fn end_io;
 174         dm_request_endio_fn rq_end_io;
 175         dm_presuspend_fn presuspend;
 176         dm_presuspend_undo_fn presuspend_undo;
 177         dm_postsuspend_fn postsuspend;
 178         dm_preresume_fn preresume;
 179         dm_resume_fn resume;
 180         dm_status_fn status;
 181         dm_message_fn message;
 182         dm_prepare_ioctl_fn prepare_ioctl;
 183         dm_busy_fn busy;
 184         dm_iterate_devices_fn iterate_devices;
 185         dm_io_hints_fn io_hints;
 186         dm_dax_direct_access_fn direct_access;
 187         dm_dax_copy_from_iter_fn dax_copy_from_iter;
 188
 189         /* For internal device-mapper use. */
 190         struct list_head list;
 191 };
 192
 193 /*
 194  * Target features
 195  */
 196
 197 /*
 198  * Any table that contains an instance of this target must have only one.
 199  */
 200 #define DM_TARGET_SINGLETON             0x00000001
 201 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
 202
 203 /*
 204  * Indicates that a target does not support read-only devices.
 205  */
 206 #define DM_TARGET_ALWAYS_WRITEABLE      0x00000002
 207 #define dm_target_always_writeable(type) \
 208                 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
 209
 210 /*
 211  * Any device that contains a table with an instance of this target may never
 212  * have tables containing any different target type.
 213  */
 214 #define DM_TARGET_IMMUTABLE             0x00000004
 215 #define dm_target_is_immutable(type)    ((type)->features & DM_TARGET_IMMUTABLE)
 216
 217 /*
 218  * Indicates that a target may replace any target; even immutable targets.
 219  * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
 220  */
 221 #define DM_TARGET_WILDCARD              0x00000008
 222 #define dm_target_is_wildcard(type)     ((type)->features & DM_TARGET_WILDCARD)
 223
 224 /*
 225  * A target implements own bio data integrity.
 226  */
 227 #define DM_TARGET_INTEGRITY             0x00000010
 228 #define dm_target_has_integrity(type)   ((type)->features & DM_TARGET_INTEGRITY)
 229
 230 /*
 231  * A target passes integrity data to the lower device.
 232  */
 233 #define DM_TARGET_PASSES_INTEGRITY      0x00000020
 234 #define dm_target_passes_integrity(type) ((type)->features & DM_TARGET_PASSES_INTEGRITY)
 235
 236 /*
 237  * Indicates that a target supports host-managed zoned block devices.
 238  */
 239 #define DM_TARGET_ZONED_HM              0x00000040
 240 #define dm_target_supports_zoned_hm(type) ((type)->features & DM_TARGET_ZONED_HM)
 241
 242 struct dm_target {
 243         struct dm_table *table;
 244         struct target_type *type;
 245
 246         /* target limits */
 247         sector_t begin;
 248         sector_t len;
 249
 250         /* If non-zero, maximum size of I/O submitted to a target. */
 251         uint32_t max_io_len;
 252
 253         /*
 254          * A number of zero-length barrier bios that will be submitted
 255          * to the target for the purpose of flushing cache.
 256          *
 257          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 258          * It is a responsibility of the target driver to remap these bios
 259          * to the real underlying devices.
 260          */
 261         unsigned num_flush_bios;
 262
 263         /*
 264          * The number of discard bios that will be submitted to the target.
 265          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 266          */
 267         unsigned num_discard_bios;
 268
 269         /*
 270          * The number of WRITE SAME bios that will be submitted to the target.
 271          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 272          */
 273         unsigned num_write_same_bios;
 274
 275         /*
 276          * The number of WRITE ZEROES bios that will be submitted to the target.
 277          * The bio number can be accessed with dm_bio_get_target_bio_nr.
 278          */
 279         unsigned num_write_zeroes_bios;
 280
 281         /*
 282          * The minimum number of extra bytes allocated in each io for the
 283          * target to use.
 284          */
 285         unsigned per_io_data_size;
 286
 287         /* target specific data */
 288         void *private;
 289
 290         /* Used to provide an error string from the ctr */
 291         char *error;
 292
 293         /*
 294          * Set if this target needs to receive flushes regardless of
 295          * whether or not its underlying devices have support.
 296          */
 297         bool flush_supported:1;
 298
 299         /*
 300          * Set if this target needs to receive discards regardless of
 301          * whether or not its underlying devices have support.
 302          */
 303         bool discards_supported:1;
 304
 305         /*
 306          * Set if the target required discard bios to be split
 307          * on max_io_len boundary.
 308          */
 309         bool split_discard_bios:1;
 310 };
 311
 312 /* Each target can link one of these into the table */
 313 struct dm_target_callbacks {
 314         struct list_head list;
 315         int (*congested_fn) (struct dm_target_callbacks *, int);
 316 };
 317
 318 void *dm_per_bio_data(struct bio *bio, size_t data_size);
 319 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size);
 320 unsigned dm_bio_get_target_bio_nr(const struct bio *bio);
 321
 322 int dm_register_target(struct target_type *t);
 323 void dm_unregister_target(struct target_type *t);
 324
 325 /*
 326  * Target argument parsing.
 327  */
 328 struct dm_arg_set {
 329         unsigned argc;
 330         char **argv;
 331 };
 332
 333 /*
 334  * The minimum and maximum value of a numeric argument, together with
 335  * the error message to use if the number is found to be outside that range.
 336  */
 337 struct dm_arg {
 338         unsigned min;
 339         unsigned max;
 340         char *error;
 341 };
 342
 343 /*
 344  * Validate the next argument, either returning it as *value or, if invalid,
 345  * returning -EINVAL and setting *error.
 346  */
 347 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 348                 unsigned *value, char **error);
 349
 350 /*
 351  * Process the next argument as the start of a group containing between
 352  * arg->min and arg->max further arguments. Either return the size as
 353  * *num_args or, if invalid, return -EINVAL and set *error.
 354  */
 355 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
 356                       unsigned *num_args, char **error);
 357
 358 /*
 359  * Return the current argument and shift to the next.
 360  */
 361 const char *dm_shift_arg(struct dm_arg_set *as);
 362
 363 /*
 364  * Move through num_args arguments.
 365  */
 366 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
 367
 368 /*-----------------------------------------------------------------
 369  * Functions for creating and manipulating mapped devices.
 370  * Drop the reference with dm_put when you finish with the object.
 371  *---------------------------------------------------------------*/
 372
 373 /*
 374  * DM_ANY_MINOR chooses the next available minor number.
 375  */
 376 #define DM_ANY_MINOR (-1)
 377 int dm_create(int minor, struct mapped_device **md);
 378
 379 /*
 380  * Reference counting for md.
 381  */
 382 struct mapped_device *dm_get_md(dev_t dev);
 383 void dm_get(struct mapped_device *md);
 384 int dm_hold(struct mapped_device *md);
 385 void dm_put(struct mapped_device *md);
 386
 387 /*
 388  * An arbitrary pointer may be stored alongside a mapped device.
 389  */
 390 void dm_set_mdptr(struct mapped_device *md, void *ptr);
 391 void *dm_get_mdptr(struct mapped_device *md);
 392
 393 /*
 394  * A device can still be used while suspended, but I/O is deferred.
 395  */
 396 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
 397 int dm_resume(struct mapped_device *md);
 398
 399 /*
 400  * Event functions.
 401  */
 402 uint32_t dm_get_event_nr(struct mapped_device *md);
 403 int dm_wait_event(struct mapped_device *md, int event_nr);
 404 uint32_t dm_next_uevent_seq(struct mapped_device *md);
 405 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
 406
 407 /*
 408  * Info functions.
 409  */
 410 const char *dm_device_name(struct mapped_device *md);
 411 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
 412 struct gendisk *dm_disk(struct mapped_device *md);
 413 int dm_suspended(struct dm_target *ti);
 414 int dm_noflush_suspending(struct dm_target *ti);
 415 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
 416 void dm_remap_zone_report(struct dm_target *ti, struct bio *bio,
 417                           sector_t start);
 418 union map_info *dm_get_rq_mapinfo(struct request *rq);
 419
 420 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
 421
 422 /*
 423  * Geometry functions.
 424  */
 425 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
 426 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
 427
 428 /*-----------------------------------------------------------------
 429  * Functions for manipulating device-mapper tables.
 430  *---------------------------------------------------------------*/
 431
 432 /*
 433  * First create an empty table.
 434  */
 435 int dm_table_create(struct dm_table **result, fmode_t mode,
 436                     unsigned num_targets, struct mapped_device *md);
 437
 438 /*
 439  * Then call this once for each target.
 440  */
 441 int dm_table_add_target(struct dm_table *t, const char *type,
 442                         sector_t start, sector_t len, char *params);
 443
 444 /*
 445  * Target_ctr should call this if it needs to add any callbacks.
 446  */
 447 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb);
 448
 449 /*
 450  * Target can use this to set the table's type.
 451  * Can only ever be called from a target's ctr.
 452  * Useful for "hybrid" target (supports both bio-based
 453  * and request-based).
 454  */
 455 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type);
 456
 457 /*
 458  * Finally call this to make the table ready for use.
 459  */
 460 int dm_table_complete(struct dm_table *t);
 461
 462 /*
 463  * Destroy the table when finished.
 464  */
 465 void dm_table_destroy(struct dm_table *t);
 466
 467 /*
 468  * Target may require that it is never sent I/O larger than len.
 469  */
 470 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
 471
 472 /*
 473  * Table reference counting.
 474  */
 475 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
 476 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
 477 void dm_sync_table(struct mapped_device *md);
 478
 479 /*
 480  * Queries
 481  */
 482 sector_t dm_table_get_size(struct dm_table *t);
 483 unsigned int dm_table_get_num_targets(struct dm_table *t);
 484 fmode_t dm_table_get_mode(struct dm_table *t);
 485 struct mapped_device *dm_table_get_md(struct dm_table *t);
 486
 487 /*
 488  * Trigger an event.
 489  */
 490 void dm_table_event(struct dm_table *t);
 491
 492 /*
 493  * Run the queue for request-based targets.
 494  */
 495 void dm_table_run_md_queue_async(struct dm_table *t);
 496
 497 /*
 498  * The device must be suspended before calling this method.
 499  * Returns the previous table, which the caller must destroy.
 500  */
 501 struct dm_table *dm_swap_table(struct mapped_device *md,
 502                                struct dm_table *t);
 503
 504 /*
 505  * A wrapper around vmalloc.
 506  */
 507 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size);
 508
 509 /*-----------------------------------------------------------------
 510  * Macros.
 511  *---------------------------------------------------------------*/
 512 #define DM_NAME "device-mapper"
 513
 514 #define DM_RATELIMIT(pr_func, fmt, ...)                                 \
 515 do {                                                                    \
 516         static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,   \
 517                                       DEFAULT_RATELIMIT_BURST);         \
 518                                                                         \
 519         if (__ratelimit(&rs))                                           \
 520                 pr_func(DM_FMT(fmt), ##__VA_ARGS__);                    \
 521 } while (0)
 522
 523 #define DM_FMT(fmt) DM_NAME ": " DM_MSG_PREFIX ": " fmt "\n"
 524
 525 #define DMCRIT(fmt, ...) pr_crit(DM_FMT(fmt), ##__VA_ARGS__)
 526
 527 #define DMERR(fmt, ...) pr_err(DM_FMT(fmt), ##__VA_ARGS__)
 528 #define DMERR_LIMIT(fmt, ...) DM_RATELIMIT(pr_err, fmt, ##__VA_ARGS__)
 529 #define DMWARN(fmt, ...) pr_warn(DM_FMT(fmt), ##__VA_ARGS__)
 530 #define DMWARN_LIMIT(fmt, ...) DM_RATELIMIT(pr_warn, fmt, ##__VA_ARGS__)
 531 #define DMINFO(fmt, ...) pr_info(DM_FMT(fmt), ##__VA_ARGS__)
 532 #define DMINFO_LIMIT(fmt, ...) DM_RATELIMIT(pr_info, fmt, ##__VA_ARGS__)
 533
 534 #ifdef CONFIG_DM_DEBUG
 535 #define DMDEBUG(fmt, ...) printk(KERN_DEBUG DM_FMT(fmt), ##__VA_ARGS__)
 536 #define DMDEBUG_LIMIT(fmt, ...) DM_RATELIMIT(pr_debug, fmt, ##__VA_ARGS__)
 537 #else
 538 #define DMDEBUG(fmt, ...) no_printk(fmt, ##__VA_ARGS__)
 539 #define DMDEBUG_LIMIT(fmt, ...) no_printk(fmt, ##__VA_ARGS__)
 540 #endif
 541
 542 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
 543                           0 : scnprintf(result + sz, maxlen - sz, x))
 544
 545 #define SECTOR_SHIFT 9
 546
 547 /*
 548  * Definitions of return values from target end_io function.
 549  */
 550 #define DM_ENDIO_DONE           0
 551 #define DM_ENDIO_INCOMPLETE     1
 552 #define DM_ENDIO_REQUEUE        2
 553 #define DM_ENDIO_DELAY_REQUEUE  3
 554
 555 /*
 556  * Definitions of return values from target map function.
 557  */
 558 #define DM_MAPIO_SUBMITTED      0
 559 #define DM_MAPIO_REMAPPED       1
 560 #define DM_MAPIO_REQUEUE        DM_ENDIO_REQUEUE
 561 #define DM_MAPIO_DELAY_REQUEUE  DM_ENDIO_DELAY_REQUEUE
 562 #define DM_MAPIO_KILL           4
 563
 564 #define dm_sector_div64(x, y)( \
 565 { \
 566         u64 _res; \
 567         (x) = div64_u64_rem(x, y, &_res); \
 568         _res; \
 569 } \
 570 )
 571
 572 /*
 573  * Ceiling(n / sz)
 574  */
 575 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
 576
 577 #define dm_sector_div_up(n, sz) ( \
 578 { \
 579         sector_t _r = ((n) + (sz) - 1); \
 580         sector_div(_r, (sz)); \
 581         _r; \
 582 } \
 583 )
 584
 585 /*
 586  * ceiling(n / size) * size
 587  */
 588 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
 589
 590 #define dm_array_too_big(fixed, obj, num) \
 591         ((num) > (UINT_MAX - (fixed)) / (obj))
 592
 593 /*
 594  * Sector offset taken relative to the start of the target instead of
 595  * relative to the start of the device.
 596  */
 597 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
 598
 599 static inline sector_t to_sector(unsigned long n)
 600 {
 601         return (n >> SECTOR_SHIFT);
 602 }
 603
 604 static inline unsigned long to_bytes(sector_t n)
 605 {
 606         return (n << SECTOR_SHIFT);
 607 }
 608
 609 #endif  /* _LINUX_DEVICE_MAPPER_H */