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/dm-ioctl.h>
14 #include <linux/math64.h>
15 #include <linux/ratelimit.h>
16
17 struct dm_dev;
18 struct dm_target;
19 struct dm_table;
20 struct dm_report_zones_args;
21 struct mapped_device;
22 struct bio_vec;
23
24 /*
25 * Type of table, mapped_device's mempool and request_queue
26 */
27 enum dm_queue_mode {
28 DM_TYPE_NONE = 0,
29 DM_TYPE_BIO_BASED = 1,
30 DM_TYPE_REQUEST_BASED = 2,
31 DM_TYPE_DAX_BIO_BASED = 3,
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 union map_info *map_context);
67
68 /*
69 * Returns:
70 * < 0 : error (currently ignored)
71 * 0 : ended successfully
72 * 1 : for some reason the io has still not completed (eg,
73 * multipath target might want to requeue a failed io).
74 * 2 : The target wants to push back the io
75 */
76 typedef int (*dm_endio_fn) (struct dm_target *ti,
77 struct bio *bio, blk_status_t *error);
78 typedef int (*dm_request_endio_fn) (struct dm_target *ti,
79 struct request *clone, blk_status_t error,
80 union map_info *map_context);
81
82 typedef void (*dm_presuspend_fn) (struct dm_target *ti);
83 typedef void (*dm_presuspend_undo_fn) (struct dm_target *ti);
84 typedef void (*dm_postsuspend_fn) (struct dm_target *ti);
85 typedef int (*dm_preresume_fn) (struct dm_target *ti);
86 typedef void (*dm_resume_fn) (struct dm_target *ti);
87
88 typedef void (*dm_status_fn) (struct dm_target *ti, status_type_t status_type,
89 unsigned status_flags, char *result, unsigned maxlen);
90
91 typedef int (*dm_message_fn) (struct dm_target *ti, unsigned argc, char **argv,
92 char *result, unsigned maxlen);
93
94 typedef int (*dm_prepare_ioctl_fn) (struct dm_target *ti, struct block_device **bdev);
95
96 #ifdef CONFIG_BLK_DEV_ZONED
97 typedef int (*dm_report_zones_fn) (struct dm_target *ti,
98 struct dm_report_zones_args *args,
99 unsigned int nr_zones);
100 #else
101 /*
102 * Define dm_report_zones_fn so that targets can assign to NULL if
103 * CONFIG_BLK_DEV_ZONED disabled. Otherwise each target needs to do
104 * awkward #ifdefs in their target_type, etc.
105 */
106 typedef int (*dm_report_zones_fn) (struct dm_target *dummy);
107 #endif
108
109 /*
110 * These iteration functions are typically used to check (and combine)
111 * properties of underlying devices.
112 * E.g. Does at least one underlying device support flush?
113 * Does any underlying device not support WRITE_SAME?
114 *
115 * The callout function is called once for each contiguous section of
116 * an underlying device. State can be maintained in *data.
117 * Return non-zero to stop iterating through any further devices.
118 */
119 typedef int (*iterate_devices_callout_fn) (struct dm_target *ti,
120 struct dm_dev *dev,
121 sector_t start, sector_t len,
122 void *data);
123
124 /*
125 * This function must iterate through each section of device used by the
126 * target until it encounters a non-zero return code, which it then returns.
127 * Returns zero if no callout returned non-zero.
128 */
129 typedef int (*dm_iterate_devices_fn) (struct dm_target *ti,
130 iterate_devices_callout_fn fn,
131 void *data);
132
133 typedef void (*dm_io_hints_fn) (struct dm_target *ti,
134 struct queue_limits *limits);
135
136 /*
137 * Returns:
138 * 0: The target can handle the next I/O immediately.
139 * 1: The target can't handle the next I/O immediately.
140 */
141 typedef int (*dm_busy_fn) (struct dm_target *ti);
142
143 /*
144 * Returns:
145 * < 0 : error
146 * >= 0 : the number of bytes accessible at the address
147 */
148 typedef long (*dm_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff,
149 long nr_pages, void **kaddr, pfn_t *pfn);
150 typedef size_t (*dm_dax_copy_iter_fn)(struct dm_target *ti, pgoff_t pgoff,
151 void *addr, size_t bytes, struct iov_iter *i);
152 typedef int (*dm_dax_zero_page_range_fn)(struct dm_target *ti, pgoff_t pgoff,
153 size_t nr_pages);
154 #define PAGE_SECTORS (PAGE_SIZE / 512)
155
156 void dm_error(const char *message);
157
158 struct dm_dev {
159 struct block_device *bdev;
160 struct dax_device *dax_dev;
161 fmode_t mode;
162 char name[16];
163 };
164
165 dev_t dm_get_dev_t(const char *path);
166
167 /*
168 * Constructors should call these functions to ensure destination devices
169 * are opened/closed correctly.
170 */
171 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
172 struct dm_dev **result);
173 void dm_put_device(struct dm_target *ti, struct dm_dev *d);
174
175 /*
176 * Information about a target type
177 */
178
179 struct target_type {
180 uint64_t features;
181 const char *name;
182 struct module *module;
183 unsigned version[3];
184 dm_ctr_fn ctr;
185 dm_dtr_fn dtr;
186 dm_map_fn map;
187 dm_clone_and_map_request_fn clone_and_map_rq;
188 dm_release_clone_request_fn release_clone_rq;
189 dm_endio_fn end_io;
190 dm_request_endio_fn rq_end_io;
191 dm_presuspend_fn presuspend;
192 dm_presuspend_undo_fn presuspend_undo;
193 dm_postsuspend_fn postsuspend;
194 dm_preresume_fn preresume;
195 dm_resume_fn resume;
196 dm_status_fn status;
197 dm_message_fn message;
198 dm_prepare_ioctl_fn prepare_ioctl;
199 dm_report_zones_fn report_zones;
200 dm_busy_fn busy;
201 dm_iterate_devices_fn iterate_devices;
202 dm_io_hints_fn io_hints;
203 dm_dax_direct_access_fn direct_access;
204 dm_dax_copy_iter_fn dax_copy_from_iter;
205 dm_dax_copy_iter_fn dax_copy_to_iter;
206 dm_dax_zero_page_range_fn dax_zero_page_range;
207
208 /* For internal device-mapper use. */
209 struct list_head list;
210 };
211
212 /*
213 * Target features
214 */
215
216 /*
217 * Any table that contains an instance of this target must have only one.
218 */
219 #define DM_TARGET_SINGLETON 0x00000001
220 #define dm_target_needs_singleton(type) ((type)->features & DM_TARGET_SINGLETON)
221
222 /*
223 * Indicates that a target does not support read-only devices.
224 */
225 #define DM_TARGET_ALWAYS_WRITEABLE 0x00000002
226 #define dm_target_always_writeable(type) \
227 ((type)->features & DM_TARGET_ALWAYS_WRITEABLE)
228
229 /*
230 * Any device that contains a table with an instance of this target may never
231 * have tables containing any different target type.
232 */
233 #define DM_TARGET_IMMUTABLE 0x00000004
234 #define dm_target_is_immutable(type) ((type)->features & DM_TARGET_IMMUTABLE)
235
236 /*
237 * Indicates that a target may replace any target; even immutable targets.
238 * .map, .map_rq, .clone_and_map_rq and .release_clone_rq are all defined.
239 */
240 #define DM_TARGET_WILDCARD 0x00000008
241 #define dm_target_is_wildcard(type) ((type)->features & DM_TARGET_WILDCARD)
242
243 /*
244 * A target implements own bio data integrity.
245 */
246 #define DM_TARGET_INTEGRITY 0x00000010
247 #define dm_target_has_integrity(type) ((type)->features & DM_TARGET_INTEGRITY)
248
249 /*
250 * A target passes integrity data to the lower device.
251 */
252 #define DM_TARGET_PASSES_INTEGRITY 0x00000020
253 #define dm_target_passes_integrity(type) ((type)->features & DM_TARGET_PASSES_INTEGRITY)
254
255 /*
256 * Indicates support for zoned block devices:
257 * - DM_TARGET_ZONED_HM: the target also supports host-managed zoned
258 * block devices but does not support combining different zoned models.
259 * - DM_TARGET_MIXED_ZONED_MODEL: the target supports combining multiple
260 * devices with different zoned models.
261 */
262 #ifdef CONFIG_BLK_DEV_ZONED
263 #define DM_TARGET_ZONED_HM 0x00000040
264 #define dm_target_supports_zoned_hm(type) ((type)->features & DM_TARGET_ZONED_HM)
265 #else
266 #define DM_TARGET_ZONED_HM 0x00000000
267 #define dm_target_supports_zoned_hm(type) (false)
268 #endif
269
270 /*
271 * A target handles REQ_NOWAIT
272 */
273 #define DM_TARGET_NOWAIT 0x00000080
274 #define dm_target_supports_nowait(type) ((type)->features & DM_TARGET_NOWAIT)
275
276 /*
277 * A target supports passing through inline crypto support.
278 */
279 #define DM_TARGET_PASSES_CRYPTO 0x00000100
280 #define dm_target_passes_crypto(type) ((type)->features & DM_TARGET_PASSES_CRYPTO)
281
282 #ifdef CONFIG_BLK_DEV_ZONED
283 #define DM_TARGET_MIXED_ZONED_MODEL 0x00000200
284 #define dm_target_supports_mixed_zoned_model(type) \
285 ((type)->features & DM_TARGET_MIXED_ZONED_MODEL)
286 #else
287 #define DM_TARGET_MIXED_ZONED_MODEL 0x00000000
288 #define dm_target_supports_mixed_zoned_model(type) (false)
289 #endif
290
291 struct dm_target {
292 struct dm_table *table;
293 struct target_type *type;
294
295 /* target limits */
296 sector_t begin;
297 sector_t len;
298
299 /* If non-zero, maximum size of I/O submitted to a target. */
300 uint32_t max_io_len;
301
302 /*
303 * A number of zero-length barrier bios that will be submitted
304 * to the target for the purpose of flushing cache.
305 *
306 * The bio number can be accessed with dm_bio_get_target_bio_nr.
307 * It is a responsibility of the target driver to remap these bios
308 * to the real underlying devices.
309 */
310 unsigned num_flush_bios;
311
312 /*
313 * The number of discard bios that will be submitted to the target.
314 * The bio number can be accessed with dm_bio_get_target_bio_nr.
315 */
316 unsigned num_discard_bios;
317
318 /*
319 * The number of secure erase bios that will be submitted to the target.
320 * The bio number can be accessed with dm_bio_get_target_bio_nr.
321 */
322 unsigned num_secure_erase_bios;
323
324 /*
325 * The number of WRITE SAME bios that will be submitted to the target.
326 * The bio number can be accessed with dm_bio_get_target_bio_nr.
327 */
328 unsigned num_write_same_bios;
329
330 /*
331 * The number of WRITE ZEROES bios that will be submitted to the target.
332 * The bio number can be accessed with dm_bio_get_target_bio_nr.
333 */
334 unsigned num_write_zeroes_bios;
335
336 /*
337 * The minimum number of extra bytes allocated in each io for the
338 * target to use.
339 */
340 unsigned per_io_data_size;
341
342 /* target specific data */
343 void *private;
344
345 /* Used to provide an error string from the ctr */
346 char *error;
347
348 /*
349 * Set if this target needs to receive flushes regardless of
350 * whether or not its underlying devices have support.
351 */
352 bool flush_supported:1;
353
354 /*
355 * Set if this target needs to receive discards regardless of
356 * whether or not its underlying devices have support.
357 */
358 bool discards_supported:1;
359
360 /*
361 * Set if we need to limit the number of in-flight bios when swapping.
362 */
363 bool limit_swap_bios:1;
364 };
365
366 void *dm_per_bio_data(struct bio *bio, size_t data_size);
367 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size);
368 unsigned dm_bio_get_target_bio_nr(const struct bio *bio);
369
370 u64 dm_start_time_ns_from_clone(struct bio *bio);
371
372 int dm_register_target(struct target_type *t);
373 void dm_unregister_target(struct target_type *t);
374
375 /*
376 * Target argument parsing.
377 */
378 struct dm_arg_set {
379 unsigned argc;
380 char **argv;
381 };
382
383 /*
384 * The minimum and maximum value of a numeric argument, together with
385 * the error message to use if the number is found to be outside that range.
386 */
387 struct dm_arg {
388 unsigned min;
389 unsigned max;
390 char *error;
391 };
392
393 /*
394 * Validate the next argument, either returning it as *value or, if invalid,
395 * returning -EINVAL and setting *error.
396 */
397 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
398 unsigned *value, char **error);
399
400 /*
401 * Process the next argument as the start of a group containing between
402 * arg->min and arg->max further arguments. Either return the size as
403 * *num_args or, if invalid, return -EINVAL and set *error.
404 */
405 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
406 unsigned *num_args, char **error);
407
408 /*
409 * Return the current argument and shift to the next.
410 */
411 const char *dm_shift_arg(struct dm_arg_set *as);
412
413 /*
414 * Move through num_args arguments.
415 */
416 void dm_consume_args(struct dm_arg_set *as, unsigned num_args);
417
418 /*-----------------------------------------------------------------
419 * Functions for creating and manipulating mapped devices.
420 * Drop the reference with dm_put when you finish with the object.
421 *---------------------------------------------------------------*/
422
423 /*
424 * DM_ANY_MINOR chooses the next available minor number.
425 */
426 #define DM_ANY_MINOR (-1)
427 int dm_create(int minor, struct mapped_device **md);
428
429 /*
430 * Reference counting for md.
431 */
432 struct mapped_device *dm_get_md(dev_t dev);
433 void dm_get(struct mapped_device *md);
434 int dm_hold(struct mapped_device *md);
435 void dm_put(struct mapped_device *md);
436
437 /*
438 * An arbitrary pointer may be stored alongside a mapped device.
439 */
440 void dm_set_mdptr(struct mapped_device *md, void *ptr);
441 void *dm_get_mdptr(struct mapped_device *md);
442
443 /*
444 * A device can still be used while suspended, but I/O is deferred.
445 */
446 int dm_suspend(struct mapped_device *md, unsigned suspend_flags);
447 int dm_resume(struct mapped_device *md);
448
449 /*
450 * Event functions.
451 */
452 uint32_t dm_get_event_nr(struct mapped_device *md);
453 int dm_wait_event(struct mapped_device *md, int event_nr);
454 uint32_t dm_next_uevent_seq(struct mapped_device *md);
455 void dm_uevent_add(struct mapped_device *md, struct list_head *elist);
456
457 /*
458 * Info functions.
459 */
460 const char *dm_device_name(struct mapped_device *md);
461 int dm_copy_name_and_uuid(struct mapped_device *md, char *name, char *uuid);
462 struct gendisk *dm_disk(struct mapped_device *md);
463 int dm_suspended(struct dm_target *ti);
464 int dm_post_suspending(struct dm_target *ti);
465 int dm_noflush_suspending(struct dm_target *ti);
466 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors);
467 union map_info *dm_get_rq_mapinfo(struct request *rq);
468
469 #ifdef CONFIG_BLK_DEV_ZONED
470 struct dm_report_zones_args {
471 struct dm_target *tgt;
472 sector_t next_sector;
473
474 void *orig_data;
475 report_zones_cb orig_cb;
476 unsigned int zone_idx;
477
478 /* must be filled by ->report_zones before calling dm_report_zones_cb */
479 sector_t start;
480 };
481 int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx, void *data);
482 #endif /* CONFIG_BLK_DEV_ZONED */
483
484 /*
485 * Device mapper functions to parse and create devices specified by the
486 * parameter "dm-mod.create="
487 */
488 int __init dm_early_create(struct dm_ioctl *dmi,
489 struct dm_target_spec **spec_array,
490 char **target_params_array);
491
492 struct queue_limits *dm_get_queue_limits(struct mapped_device *md);
493
494 /*
495 * Geometry functions.
496 */
497 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo);
498 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo);
499
500 /*-----------------------------------------------------------------
501 * Functions for manipulating device-mapper tables.
502 *---------------------------------------------------------------*/
503
504 /*
505 * First create an empty table.
506 */
507 int dm_table_create(struct dm_table **result, fmode_t mode,
508 unsigned num_targets, struct mapped_device *md);
509
510 /*
511 * Then call this once for each target.
512 */
513 int dm_table_add_target(struct dm_table *t, const char *type,
514 sector_t start, sector_t len, char *params);
515
516 /*
517 * Target can use this to set the table's type.
518 * Can only ever be called from a target's ctr.
519 * Useful for "hybrid" target (supports both bio-based
520 * and request-based).
521 */
522 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type);
523
524 /*
525 * Finally call this to make the table ready for use.
526 */
527 int dm_table_complete(struct dm_table *t);
528
529 /*
530 * Destroy the table when finished.
531 */
532 void dm_table_destroy(struct dm_table *t);
533
534 /*
535 * Target may require that it is never sent I/O larger than len.
536 */
537 int __must_check dm_set_target_max_io_len(struct dm_target *ti, sector_t len);
538
539 /*
540 * Table reference counting.
541 */
542 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx);
543 void dm_put_live_table(struct mapped_device *md, int srcu_idx);
544 void dm_sync_table(struct mapped_device *md);
545
546 /*
547 * Queries
548 */
549 sector_t dm_table_get_size(struct dm_table *t);
550 unsigned int dm_table_get_num_targets(struct dm_table *t);
551 fmode_t dm_table_get_mode(struct dm_table *t);
552 struct mapped_device *dm_table_get_md(struct dm_table *t);
553 const char *dm_table_device_name(struct dm_table *t);
554
555 /*
556 * Trigger an event.
557 */
558 void dm_table_event(struct dm_table *t);
559
560 /*
561 * Run the queue for request-based targets.
562 */
563 void dm_table_run_md_queue_async(struct dm_table *t);
564
565 /*
566 * The device must be suspended before calling this method.
567 * Returns the previous table, which the caller must destroy.
568 */
569 struct dm_table *dm_swap_table(struct mapped_device *md,
570 struct dm_table *t);
571
572 /*
573 * Table keyslot manager functions
574 */
575 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm);
576
577 /*-----------------------------------------------------------------
578 * Macros.
579 *---------------------------------------------------------------*/
580 #define DM_NAME "device-mapper"
581
582 #define DM_FMT(fmt) DM_NAME ": " DM_MSG_PREFIX ": " fmt "\n"
583
584 #define DMCRIT(fmt, ...) pr_crit(DM_FMT(fmt), ##__VA_ARGS__)
585
586 #define DMERR(fmt, ...) pr_err(DM_FMT(fmt), ##__VA_ARGS__)
587 #define DMERR_LIMIT(fmt, ...) pr_err_ratelimited(DM_FMT(fmt), ##__VA_ARGS__)
588 #define DMWARN(fmt, ...) pr_warn(DM_FMT(fmt), ##__VA_ARGS__)
589 #define DMWARN_LIMIT(fmt, ...) pr_warn_ratelimited(DM_FMT(fmt), ##__VA_ARGS__)
590 #define DMINFO(fmt, ...) pr_info(DM_FMT(fmt), ##__VA_ARGS__)
591 #define DMINFO_LIMIT(fmt, ...) pr_info_ratelimited(DM_FMT(fmt), ##__VA_ARGS__)
592
593 #define DMDEBUG(fmt, ...) pr_debug(DM_FMT(fmt), ##__VA_ARGS__)
594 #define DMDEBUG_LIMIT(fmt, ...) pr_debug_ratelimited(DM_FMT(fmt), ##__VA_ARGS__)
595
596 #define DMEMIT(x...) sz += ((sz >= maxlen) ? \
597 0 : scnprintf(result + sz, maxlen - sz, x))
598
599 /*
600 * Definitions of return values from target end_io function.
601 */
602 #define DM_ENDIO_DONE 0
603 #define DM_ENDIO_INCOMPLETE 1
604 #define DM_ENDIO_REQUEUE 2
605 #define DM_ENDIO_DELAY_REQUEUE 3
606
607 /*
608 * Definitions of return values from target map function.
609 */
610 #define DM_MAPIO_SUBMITTED 0
611 #define DM_MAPIO_REMAPPED 1
612 #define DM_MAPIO_REQUEUE DM_ENDIO_REQUEUE
613 #define DM_MAPIO_DELAY_REQUEUE DM_ENDIO_DELAY_REQUEUE
614 #define DM_MAPIO_KILL 4
615
616 #define dm_sector_div64(x, y)( \
617 { \
618 u64 _res; \
619 (x) = div64_u64_rem(x, y, &_res); \
620 _res; \
621 } \
622 )
623
624 /*
625 * Ceiling(n / sz)
626 */
627 #define dm_div_up(n, sz) (((n) + (sz) - 1) / (sz))
628
629 #define dm_sector_div_up(n, sz) ( \
630 { \
631 sector_t _r = ((n) + (sz) - 1); \
632 sector_div(_r, (sz)); \
633 _r; \
634 } \
635 )
636
637 /*
638 * ceiling(n / size) * size
639 */
640 #define dm_round_up(n, sz) (dm_div_up((n), (sz)) * (sz))
641
642 /*
643 * Sector offset taken relative to the start of the target instead of
644 * relative to the start of the device.
645 */
646 #define dm_target_offset(ti, sector) ((sector) - (ti)->begin)
647
to_sector(unsigned long long n)648 static inline sector_t to_sector(unsigned long long n)
649 {
650 return (n >> SECTOR_SHIFT);
651 }
652
to_bytes(sector_t n)653 static inline unsigned long to_bytes(sector_t n)
654 {
655 return (n << SECTOR_SHIFT);
656 }
657
658 #endif /* _LINUX_DEVICE_MAPPER_H */
659