root/drivers/md/dm.c

DEFINITIONS

1. dm_issue_global_event
2. dm_per_bio_data
3. dm_bio_from_per_bio_data
4. dm_bio_get_target_bio_nr
5. __dm_get_module_param_int
6. __dm_get_module_param
7. dm_get_reserved_bio_based_ios
8. dm_get_numa_node
9. local_init
10. local_exit
11. dm_init
12. dm_exit
13. dm_deleting_md
14. dm_blk_open
15. dm_blk_close
16. dm_open_count
17. dm_lock_for_deletion
18. dm_cancel_deferred_remove
19. do_deferred_remove
20. dm_get_size
21. dm_get_md_queue
22. dm_get_stats
23. dm_blk_getgeo
24. dm_blk_report_zones
25. dm_prepare_ioctl
26. dm_unprepare_ioctl
27. dm_blk_ioctl
28. alloc_io
29. free_io
30. alloc_tio
31. free_tio
32. md_in_flight_bios
33. md_in_flight
34. start_io_acct
35. end_io_acct
36. queue_io
37. dm_get_live_table
38. dm_put_live_table
39. dm_sync_table
40. dm_get_live_table_fast
41. dm_put_live_table_fast
42. open_table_device
43. close_table_device
44. find_table_device
45. dm_get_table_device
46. dm_put_table_device
47. free_table_devices
48. dm_get_geometry
49. dm_set_geometry
50. __noflush_suspending
51. dec_pending
52. disable_discard
53. disable_write_same
54. disable_write_zeroes
55. clone_endio
56. max_io_len_target_boundary
57. max_io_len
58. dm_set_target_max_io_len
59. dm_dax_get_live_target
60. dm_dax_direct_access
61. dm_dax_supported
62. dm_dax_copy_from_iter
63. dm_dax_copy_to_iter
64. dm_accept_partial_bio
65. dm_remap_zone_report
66. __map_bio
67. bio_setup_sector
68. clone_bio
69. alloc_multiple_bios
70. __clone_and_map_simple_bio
71. __send_duplicate_bios
72. __send_empty_flush
73. __clone_and_map_data_bio
74. get_num_discard_bios
75. get_num_secure_erase_bios
76. get_num_write_same_bios
77. get_num_write_zeroes_bios
78. __send_changing_extent_only
79. __send_discard
80. __send_secure_erase
81. __send_write_same
82. __send_write_zeroes
83. is_abnormal_io
84. __process_abnormal_io
85. __split_and_process_non_flush
86. init_clone_info
87. __split_and_process_bio
88. __process_bio
89. dm_queue_split
90. dm_process_bio
91. dm_make_request
92. dm_any_congested
93. free_minor
94. specific_minor
95. next_free_minor
96. cleanup_mapped_device
97. alloc_dev
98. free_dev
99. __bind_mempools
100. event_callback
101. __set_size
102. __bind
103. __unbind
104. dm_create
105. dm_lock_md_type
106. dm_unlock_md_type
107. dm_set_md_type
108. dm_get_md_type
109. dm_get_immutable_target_type
110. dm_get_queue_limits
111. dm_init_congested_fn
112. dm_setup_md_queue
113. dm_get_md
114. dm_get_mdptr
115. dm_set_mdptr
116. dm_get
117. dm_hold
118. dm_device_name
119. __dm_destroy
120. dm_destroy
121. dm_destroy_immediate
122. dm_put
123. dm_wait_for_completion
124. dm_wq_work
125. dm_queue_flush
126. dm_swap_table
127. lock_fs
128. unlock_fs
129. __dm_suspend
130. dm_suspend
131. __dm_resume
132. dm_resume
133. __dm_internal_suspend
134. __dm_internal_resume
135. dm_internal_suspend_noflush
136. dm_internal_resume
137. dm_internal_suspend_fast
138. dm_internal_resume_fast
139. dm_kobject_uevent
140. dm_next_uevent_seq
141. dm_get_event_nr
142. dm_wait_event
143. dm_uevent_add
144. dm_disk
145. dm_kobject
146. dm_get_from_kobject
147. dm_suspended_md
148. dm_suspended_internally_md
149. dm_test_deferred_remove_flag
150. dm_suspended
151. dm_noflush_suspending
152. dm_alloc_md_mempools
153. dm_free_md_mempools
154. dm_call_pr
155. __dm_pr_register
156. dm_pr_register
157. dm_pr_reserve
158. dm_pr_release
159. dm_pr_preempt
160. dm_pr_clear

   1 /*
   2  * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
   3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4  *
   5  * This file is released under the GPL.
   6  */
   7 
   8 #include "dm-core.h"
   9 #include "dm-rq.h"
  10 #include "dm-uevent.h"
  11 
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/mutex.h>
  15 #include <linux/sched/signal.h>
  16 #include <linux/blkpg.h>
  17 #include <linux/bio.h>
  18 #include <linux/mempool.h>
  19 #include <linux/dax.h>
  20 #include <linux/slab.h>
  21 #include <linux/idr.h>
  22 #include <linux/uio.h>
  23 #include <linux/hdreg.h>
  24 #include <linux/delay.h>
  25 #include <linux/wait.h>
  26 #include <linux/pr.h>
  27 #include <linux/refcount.h>
  28 
  29 #define DM_MSG_PREFIX "core"
  30 
  31 /*
  32  * Cookies are numeric values sent with CHANGE and REMOVE
  33  * uevents while resuming, removing or renaming the device.
  34  */
  35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
  36 #define DM_COOKIE_LENGTH 24
  37 
  38 static const char *_name = DM_NAME;
  39 
  40 static unsigned int major = 0;
  41 static unsigned int _major = 0;
  42 
  43 static DEFINE_IDR(_minor_idr);
  44 
  45 static DEFINE_SPINLOCK(_minor_lock);
  46 
  47 static void do_deferred_remove(struct work_struct *w);
  48 
  49 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
  50 
  51 static struct workqueue_struct *deferred_remove_workqueue;
  52 
  53 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
  54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
  55 
  56 void dm_issue_global_event(void)
  57 {
  58         atomic_inc(&dm_global_event_nr);
  59         wake_up(&dm_global_eventq);
  60 }
  61 
  62 /*
  63  * One of these is allocated (on-stack) per original bio.
  64  */
  65 struct clone_info {
  66         struct dm_table *map;
  67         struct bio *bio;
  68         struct dm_io *io;
  69         sector_t sector;
  70         unsigned sector_count;
  71 };
  72 
  73 /*
  74  * One of these is allocated per clone bio.
  75  */
  76 #define DM_TIO_MAGIC 7282014
  77 struct dm_target_io {
  78         unsigned magic;
  79         struct dm_io *io;
  80         struct dm_target *ti;
  81         unsigned target_bio_nr;
  82         unsigned *len_ptr;
  83         bool inside_dm_io;
  84         struct bio clone;
  85 };
  86 
  87 /*
  88  * One of these is allocated per original bio.
  89  * It contains the first clone used for that original.
  90  */
  91 #define DM_IO_MAGIC 5191977
  92 struct dm_io {
  93         unsigned magic;
  94         struct mapped_device *md;
  95         blk_status_t status;
  96         atomic_t io_count;
  97         struct bio *orig_bio;
  98         unsigned long start_time;
  99         spinlock_t endio_lock;
 100         struct dm_stats_aux stats_aux;
 101         /* last member of dm_target_io is 'struct bio' */
 102         struct dm_target_io tio;
 103 };
 104 
 105 void *dm_per_bio_data(struct bio *bio, size_t data_size)
 106 {
 107         struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 108         if (!tio->inside_dm_io)
 109                 return (char *)bio - offsetof(struct dm_target_io, clone) - data_size;
 110         return (char *)bio - offsetof(struct dm_target_io, clone) - offsetof(struct dm_io, tio) - data_size;
 111 }
 112 EXPORT_SYMBOL_GPL(dm_per_bio_data);
 113 
 114 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
 115 {
 116         struct dm_io *io = (struct dm_io *)((char *)data + data_size);
 117         if (io->magic == DM_IO_MAGIC)
 118                 return (struct bio *)((char *)io + offsetof(struct dm_io, tio) + offsetof(struct dm_target_io, clone));
 119         BUG_ON(io->magic != DM_TIO_MAGIC);
 120         return (struct bio *)((char *)io + offsetof(struct dm_target_io, clone));
 121 }
 122 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
 123 
 124 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
 125 {
 126         return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
 127 }
 128 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
 129 
 130 #define MINOR_ALLOCED ((void *)-1)
 131 
 132 /*
 133  * Bits for the md->flags field.
 134  */
 135 #define DMF_BLOCK_IO_FOR_SUSPEND 0
 136 #define DMF_SUSPENDED 1
 137 #define DMF_FROZEN 2
 138 #define DMF_FREEING 3
 139 #define DMF_DELETING 4
 140 #define DMF_NOFLUSH_SUSPENDING 5
 141 #define DMF_DEFERRED_REMOVE 6
 142 #define DMF_SUSPENDED_INTERNALLY 7
 143 
 144 #define DM_NUMA_NODE NUMA_NO_NODE
 145 static int dm_numa_node = DM_NUMA_NODE;
 146 
 147 /*
 148  * For mempools pre-allocation at the table loading time.
 149  */
 150 struct dm_md_mempools {
 151         struct bio_set bs;
 152         struct bio_set io_bs;
 153 };
 154 
 155 struct table_device {
 156         struct list_head list;
 157         refcount_t count;
 158         struct dm_dev dm_dev;
 159 };
 160 
 161 /*
 162  * Bio-based DM's mempools' reserved IOs set by the user.
 163  */
 164 #define RESERVED_BIO_BASED_IOS          16
 165 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
 166 
 167 static int __dm_get_module_param_int(int *module_param, int min, int max)
 168 {
 169         int param = READ_ONCE(*module_param);
 170         int modified_param = 0;
 171         bool modified = true;
 172 
 173         if (param < min)
 174                 modified_param = min;
 175         else if (param > max)
 176                 modified_param = max;
 177         else
 178                 modified = false;
 179 
 180         if (modified) {
 181                 (void)cmpxchg(module_param, param, modified_param);
 182                 param = modified_param;
 183         }
 184 
 185         return param;
 186 }
 187 
 188 unsigned __dm_get_module_param(unsigned *module_param,
 189                                unsigned def, unsigned max)
 190 {
 191         unsigned param = READ_ONCE(*module_param);
 192         unsigned modified_param = 0;
 193 
 194         if (!param)
 195                 modified_param = def;
 196         else if (param > max)
 197                 modified_param = max;
 198 
 199         if (modified_param) {
 200                 (void)cmpxchg(module_param, param, modified_param);
 201                 param = modified_param;
 202         }
 203 
 204         return param;
 205 }
 206 
 207 unsigned dm_get_reserved_bio_based_ios(void)
 208 {
 209         return __dm_get_module_param(&reserved_bio_based_ios,
 210                                      RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
 211 }
 212 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
 213 
 214 static unsigned dm_get_numa_node(void)
 215 {
 216         return __dm_get_module_param_int(&dm_numa_node,
 217                                          DM_NUMA_NODE, num_online_nodes() - 1);
 218 }
 219 
 220 static int __init local_init(void)
 221 {
 222         int r;
 223 
 224         r = dm_uevent_init();
 225         if (r)
 226                 return r;
 227 
 228         deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
 229         if (!deferred_remove_workqueue) {
 230                 r = -ENOMEM;
 231                 goto out_uevent_exit;
 232         }
 233 
 234         _major = major;
 235         r = register_blkdev(_major, _name);
 236         if (r < 0)
 237                 goto out_free_workqueue;
 238 
 239         if (!_major)
 240                 _major = r;
 241 
 242         return 0;
 243 
 244 out_free_workqueue:
 245         destroy_workqueue(deferred_remove_workqueue);
 246 out_uevent_exit:
 247         dm_uevent_exit();
 248 
 249         return r;
 250 }
 251 
 252 static void local_exit(void)
 253 {
 254         flush_scheduled_work();
 255         destroy_workqueue(deferred_remove_workqueue);
 256 
 257         unregister_blkdev(_major, _name);
 258         dm_uevent_exit();
 259 
 260         _major = 0;
 261 
 262         DMINFO("cleaned up");
 263 }
 264 
 265 static int (*_inits[])(void) __initdata = {
 266         local_init,
 267         dm_target_init,
 268         dm_linear_init,
 269         dm_stripe_init,
 270         dm_io_init,
 271         dm_kcopyd_init,
 272         dm_interface_init,
 273         dm_statistics_init,
 274 };
 275 
 276 static void (*_exits[])(void) = {
 277         local_exit,
 278         dm_target_exit,
 279         dm_linear_exit,
 280         dm_stripe_exit,
 281         dm_io_exit,
 282         dm_kcopyd_exit,
 283         dm_interface_exit,
 284         dm_statistics_exit,
 285 };
 286 
 287 static int __init dm_init(void)
 288 {
 289         const int count = ARRAY_SIZE(_inits);
 290 
 291         int r, i;
 292 
 293         for (i = 0; i < count; i++) {
 294                 r = _inits[i]();
 295                 if (r)
 296                         goto bad;
 297         }
 298 
 299         return 0;
 300 
 301       bad:
 302         while (i--)
 303                 _exits[i]();
 304 
 305         return r;
 306 }
 307 
 308 static void __exit dm_exit(void)
 309 {
 310         int i = ARRAY_SIZE(_exits);
 311 
 312         while (i--)
 313                 _exits[i]();
 314 
 315         /*
 316          * Should be empty by this point.
 317          */
 318         idr_destroy(&_minor_idr);
 319 }
 320 
 321 /*
 322  * Block device functions
 323  */
 324 int dm_deleting_md(struct mapped_device *md)
 325 {
 326         return test_bit(DMF_DELETING, &md->flags);
 327 }
 328 
 329 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
 330 {
 331         struct mapped_device *md;
 332 
 333         spin_lock(&_minor_lock);
 334 
 335         md = bdev->bd_disk->private_data;
 336         if (!md)
 337                 goto out;
 338 
 339         if (test_bit(DMF_FREEING, &md->flags) ||
 340             dm_deleting_md(md)) {
 341                 md = NULL;
 342                 goto out;
 343         }
 344 
 345         dm_get(md);
 346         atomic_inc(&md->open_count);
 347 out:
 348         spin_unlock(&_minor_lock);
 349 
 350         return md ? 0 : -ENXIO;
 351 }
 352 
 353 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
 354 {
 355         struct mapped_device *md;
 356 
 357         spin_lock(&_minor_lock);
 358 
 359         md = disk->private_data;
 360         if (WARN_ON(!md))
 361                 goto out;
 362 
 363         if (atomic_dec_and_test(&md->open_count) &&
 364             (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
 365                 queue_work(deferred_remove_workqueue, &deferred_remove_work);
 366 
 367         dm_put(md);
 368 out:
 369         spin_unlock(&_minor_lock);
 370 }
 371 
 372 int dm_open_count(struct mapped_device *md)
 373 {
 374         return atomic_read(&md->open_count);
 375 }
 376 
 377 /*
 378  * Guarantees nothing is using the device before it's deleted.
 379  */
 380 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
 381 {
 382         int r = 0;
 383 
 384         spin_lock(&_minor_lock);
 385 
 386         if (dm_open_count(md)) {
 387                 r = -EBUSY;
 388                 if (mark_deferred)
 389                         set_bit(DMF_DEFERRED_REMOVE, &md->flags);
 390         } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
 391                 r = -EEXIST;
 392         else
 393                 set_bit(DMF_DELETING, &md->flags);
 394 
 395         spin_unlock(&_minor_lock);
 396 
 397         return r;
 398 }
 399 
 400 int dm_cancel_deferred_remove(struct mapped_device *md)
 401 {
 402         int r = 0;
 403 
 404         spin_lock(&_minor_lock);
 405 
 406         if (test_bit(DMF_DELETING, &md->flags))
 407                 r = -EBUSY;
 408         else
 409                 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
 410 
 411         spin_unlock(&_minor_lock);
 412 
 413         return r;
 414 }
 415 
 416 static void do_deferred_remove(struct work_struct *w)
 417 {
 418         dm_deferred_remove();
 419 }
 420 
 421 sector_t dm_get_size(struct mapped_device *md)
 422 {
 423         return get_capacity(md->disk);
 424 }
 425 
 426 struct request_queue *dm_get_md_queue(struct mapped_device *md)
 427 {
 428         return md->queue;
 429 }
 430 
 431 struct dm_stats *dm_get_stats(struct mapped_device *md)
 432 {
 433         return &md->stats;
 434 }
 435 
 436 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 437 {
 438         struct mapped_device *md = bdev->bd_disk->private_data;
 439 
 440         return dm_get_geometry(md, geo);
 441 }
 442 
 443 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
 444                                struct blk_zone *zones, unsigned int *nr_zones)
 445 {
 446 #ifdef CONFIG_BLK_DEV_ZONED
 447         struct mapped_device *md = disk->private_data;
 448         struct dm_target *tgt;
 449         struct dm_table *map;
 450         int srcu_idx, ret;
 451 
 452         if (dm_suspended_md(md))
 453                 return -EAGAIN;
 454 
 455         map = dm_get_live_table(md, &srcu_idx);
 456         if (!map)
 457                 return -EIO;
 458 
 459         tgt = dm_table_find_target(map, sector);
 460         if (!tgt) {
 461                 ret = -EIO;
 462                 goto out;
 463         }
 464 
 465         /*
 466          * If we are executing this, we already know that the block device
 467          * is a zoned device and so each target should have support for that
 468          * type of drive. A missing report_zones method means that the target
 469          * driver has a problem.
 470          */
 471         if (WARN_ON(!tgt->type->report_zones)) {
 472                 ret = -EIO;
 473                 goto out;
 474         }
 475 
 476         /*
 477          * blkdev_report_zones() will loop and call this again to cover all the
 478          * zones of the target, eventually moving on to the next target.
 479          * So there is no need to loop here trying to fill the entire array
 480          * of zones.
 481          */
 482         ret = tgt->type->report_zones(tgt, sector, zones, nr_zones);
 483 
 484 out:
 485         dm_put_live_table(md, srcu_idx);
 486         return ret;
 487 #else
 488         return -ENOTSUPP;
 489 #endif
 490 }
 491 
 492 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
 493                             struct block_device **bdev)
 494         __acquires(md->io_barrier)
 495 {
 496         struct dm_target *tgt;
 497         struct dm_table *map;
 498         int r;
 499 
 500 retry:
 501         r = -ENOTTY;
 502         map = dm_get_live_table(md, srcu_idx);
 503         if (!map || !dm_table_get_size(map))
 504                 return r;
 505 
 506         /* We only support devices that have a single target */
 507         if (dm_table_get_num_targets(map) != 1)
 508                 return r;
 509 
 510         tgt = dm_table_get_target(map, 0);
 511         if (!tgt->type->prepare_ioctl)
 512                 return r;
 513 
 514         if (dm_suspended_md(md))
 515                 return -EAGAIN;
 516 
 517         r = tgt->type->prepare_ioctl(tgt, bdev);
 518         if (r == -ENOTCONN && !fatal_signal_pending(current)) {
 519                 dm_put_live_table(md, *srcu_idx);
 520                 msleep(10);
 521                 goto retry;
 522         }
 523 
 524         return r;
 525 }
 526 
 527 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
 528         __releases(md->io_barrier)
 529 {
 530         dm_put_live_table(md, srcu_idx);
 531 }
 532 
 533 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
 534                         unsigned int cmd, unsigned long arg)
 535 {
 536         struct mapped_device *md = bdev->bd_disk->private_data;
 537         int r, srcu_idx;
 538 
 539         r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
 540         if (r < 0)
 541                 goto out;
 542 
 543         if (r > 0) {
 544                 /*
 545                  * Target determined this ioctl is being issued against a
 546                  * subset of the parent bdev; require extra privileges.
 547                  */
 548                 if (!capable(CAP_SYS_RAWIO)) {
 549                         DMWARN_LIMIT(
 550         "%s: sending ioctl %x to DM device without required privilege.",
 551                                 current->comm, cmd);
 552                         r = -ENOIOCTLCMD;
 553                         goto out;
 554                 }
 555         }
 556 
 557         r =  __blkdev_driver_ioctl(bdev, mode, cmd, arg);
 558 out:
 559         dm_unprepare_ioctl(md, srcu_idx);
 560         return r;
 561 }
 562 
 563 static void start_io_acct(struct dm_io *io);
 564 
 565 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
 566 {
 567         struct dm_io *io;
 568         struct dm_target_io *tio;
 569         struct bio *clone;
 570 
 571         clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
 572         if (!clone)
 573                 return NULL;
 574 
 575         tio = container_of(clone, struct dm_target_io, clone);
 576         tio->inside_dm_io = true;
 577         tio->io = NULL;
 578 
 579         io = container_of(tio, struct dm_io, tio);
 580         io->magic = DM_IO_MAGIC;
 581         io->status = 0;
 582         atomic_set(&io->io_count, 1);
 583         io->orig_bio = bio;
 584         io->md = md;
 585         spin_lock_init(&io->endio_lock);
 586 
 587         start_io_acct(io);
 588 
 589         return io;
 590 }
 591 
 592 static void free_io(struct mapped_device *md, struct dm_io *io)
 593 {
 594         bio_put(&io->tio.clone);
 595 }
 596 
 597 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
 598                                       unsigned target_bio_nr, gfp_t gfp_mask)
 599 {
 600         struct dm_target_io *tio;
 601 
 602         if (!ci->io->tio.io) {
 603                 /* the dm_target_io embedded in ci->io is available */
 604                 tio = &ci->io->tio;
 605         } else {
 606                 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
 607                 if (!clone)
 608                         return NULL;
 609 
 610                 tio = container_of(clone, struct dm_target_io, clone);
 611                 tio->inside_dm_io = false;
 612         }
 613 
 614         tio->magic = DM_TIO_MAGIC;
 615         tio->io = ci->io;
 616         tio->ti = ti;
 617         tio->target_bio_nr = target_bio_nr;
 618 
 619         return tio;
 620 }
 621 
 622 static void free_tio(struct dm_target_io *tio)
 623 {
 624         if (tio->inside_dm_io)
 625                 return;
 626         bio_put(&tio->clone);
 627 }
 628 
 629 static bool md_in_flight_bios(struct mapped_device *md)
 630 {
 631         int cpu;
 632         struct hd_struct *part = &dm_disk(md)->part0;
 633         long sum = 0;
 634 
 635         for_each_possible_cpu(cpu) {
 636                 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
 637                 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
 638         }
 639 
 640         return sum != 0;
 641 }
 642 
 643 static bool md_in_flight(struct mapped_device *md)
 644 {
 645         if (queue_is_mq(md->queue))
 646                 return blk_mq_queue_inflight(md->queue);
 647         else
 648                 return md_in_flight_bios(md);
 649 }
 650 
 651 static void start_io_acct(struct dm_io *io)
 652 {
 653         struct mapped_device *md = io->md;
 654         struct bio *bio = io->orig_bio;
 655 
 656         io->start_time = jiffies;
 657 
 658         generic_start_io_acct(md->queue, bio_op(bio), bio_sectors(bio),
 659                               &dm_disk(md)->part0);
 660 
 661         if (unlikely(dm_stats_used(&md->stats)))
 662                 dm_stats_account_io(&md->stats, bio_data_dir(bio),
 663                                     bio->bi_iter.bi_sector, bio_sectors(bio),
 664                                     false, 0, &io->stats_aux);
 665 }
 666 
 667 static void end_io_acct(struct dm_io *io)
 668 {
 669         struct mapped_device *md = io->md;
 670         struct bio *bio = io->orig_bio;
 671         unsigned long duration = jiffies - io->start_time;
 672 
 673         generic_end_io_acct(md->queue, bio_op(bio), &dm_disk(md)->part0,
 674                             io->start_time);
 675 
 676         if (unlikely(dm_stats_used(&md->stats)))
 677                 dm_stats_account_io(&md->stats, bio_data_dir(bio),
 678                                     bio->bi_iter.bi_sector, bio_sectors(bio),
 679                                     true, duration, &io->stats_aux);
 680 
 681         /* nudge anyone waiting on suspend queue */
 682         if (unlikely(wq_has_sleeper(&md->wait)))
 683                 wake_up(&md->wait);
 684 }
 685 
 686 /*
 687  * Add the bio to the list of deferred io.
 688  */
 689 static void queue_io(struct mapped_device *md, struct bio *bio)
 690 {
 691         unsigned long flags;
 692 
 693         spin_lock_irqsave(&md->deferred_lock, flags);
 694         bio_list_add(&md->deferred, bio);
 695         spin_unlock_irqrestore(&md->deferred_lock, flags);
 696         queue_work(md->wq, &md->work);
 697 }
 698 
 699 /*
 700  * Everyone (including functions in this file), should use this
 701  * function to access the md->map field, and make sure they call
 702  * dm_put_live_table() when finished.
 703  */
 704 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
 705 {
 706         *srcu_idx = srcu_read_lock(&md->io_barrier);
 707 
 708         return srcu_dereference(md->map, &md->io_barrier);
 709 }
 710 
 711 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
 712 {
 713         srcu_read_unlock(&md->io_barrier, srcu_idx);
 714 }
 715 
 716 void dm_sync_table(struct mapped_device *md)
 717 {
 718         synchronize_srcu(&md->io_barrier);
 719         synchronize_rcu_expedited();
 720 }
 721 
 722 /*
 723  * A fast alternative to dm_get_live_table/dm_put_live_table.
 724  * The caller must not block between these two functions.
 725  */
 726 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
 727 {
 728         rcu_read_lock();
 729         return rcu_dereference(md->map);
 730 }
 731 
 732 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
 733 {
 734         rcu_read_unlock();
 735 }
 736 
 737 static char *_dm_claim_ptr = "I belong to device-mapper";
 738 
 739 /*
 740  * Open a table device so we can use it as a map destination.
 741  */
 742 static int open_table_device(struct table_device *td, dev_t dev,
 743                              struct mapped_device *md)
 744 {
 745         struct block_device *bdev;
 746 
 747         int r;
 748 
 749         BUG_ON(td->dm_dev.bdev);
 750 
 751         bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
 752         if (IS_ERR(bdev))
 753                 return PTR_ERR(bdev);
 754 
 755         r = bd_link_disk_holder(bdev, dm_disk(md));
 756         if (r) {
 757                 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
 758                 return r;
 759         }
 760 
 761         td->dm_dev.bdev = bdev;
 762         td->dm_dev.dax_dev = dax_get_by_host(bdev->bd_disk->disk_name);
 763         return 0;
 764 }
 765 
 766 /*
 767  * Close a table device that we've been using.
 768  */
 769 static void close_table_device(struct table_device *td, struct mapped_device *md)
 770 {
 771         if (!td->dm_dev.bdev)
 772                 return;
 773 
 774         bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
 775         blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
 776         put_dax(td->dm_dev.dax_dev);
 777         td->dm_dev.bdev = NULL;
 778         td->dm_dev.dax_dev = NULL;
 779 }
 780 
 781 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
 782                                               fmode_t mode)
 783 {
 784         struct table_device *td;
 785 
 786         list_for_each_entry(td, l, list)
 787                 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
 788                         return td;
 789 
 790         return NULL;
 791 }
 792 
 793 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
 794                         struct dm_dev **result)
 795 {
 796         int r;
 797         struct table_device *td;
 798 
 799         mutex_lock(&md->table_devices_lock);
 800         td = find_table_device(&md->table_devices, dev, mode);
 801         if (!td) {
 802                 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
 803                 if (!td) {
 804                         mutex_unlock(&md->table_devices_lock);
 805                         return -ENOMEM;
 806                 }
 807 
 808                 td->dm_dev.mode = mode;
 809                 td->dm_dev.bdev = NULL;
 810 
 811                 if ((r = open_table_device(td, dev, md))) {
 812                         mutex_unlock(&md->table_devices_lock);
 813                         kfree(td);
 814                         return r;
 815                 }
 816 
 817                 format_dev_t(td->dm_dev.name, dev);
 818 
 819                 refcount_set(&td->count, 1);
 820                 list_add(&td->list, &md->table_devices);
 821         } else {
 822                 refcount_inc(&td->count);
 823         }
 824         mutex_unlock(&md->table_devices_lock);
 825 
 826         *result = &td->dm_dev;
 827         return 0;
 828 }
 829 EXPORT_SYMBOL_GPL(dm_get_table_device);
 830 
 831 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
 832 {
 833         struct table_device *td = container_of(d, struct table_device, dm_dev);
 834 
 835         mutex_lock(&md->table_devices_lock);
 836         if (refcount_dec_and_test(&td->count)) {
 837                 close_table_device(td, md);
 838                 list_del(&td->list);
 839                 kfree(td);
 840         }
 841         mutex_unlock(&md->table_devices_lock);
 842 }
 843 EXPORT_SYMBOL(dm_put_table_device);
 844 
 845 static void free_table_devices(struct list_head *devices)
 846 {
 847         struct list_head *tmp, *next;
 848 
 849         list_for_each_safe(tmp, next, devices) {
 850                 struct table_device *td = list_entry(tmp, struct table_device, list);
 851 
 852                 DMWARN("dm_destroy: %s still exists with %d references",
 853                        td->dm_dev.name, refcount_read(&td->count));
 854                 kfree(td);
 855         }
 856 }
 857 
 858 /*
 859  * Get the geometry associated with a dm device
 860  */
 861 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
 862 {
 863         *geo = md->geometry;
 864 
 865         return 0;
 866 }
 867 
 868 /*
 869  * Set the geometry of a device.
 870  */
 871 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
 872 {
 873         sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
 874 
 875         if (geo->start > sz) {
 876                 DMWARN("Start sector is beyond the geometry limits.");
 877                 return -EINVAL;
 878         }
 879 
 880         md->geometry = *geo;
 881 
 882         return 0;
 883 }
 884 
 885 static int __noflush_suspending(struct mapped_device *md)
 886 {
 887         return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
 888 }
 889 
 890 /*
 891  * Decrements the number of outstanding ios that a bio has been
 892  * cloned into, completing the original io if necc.
 893  */
 894 static void dec_pending(struct dm_io *io, blk_status_t error)
 895 {
 896         unsigned long flags;
 897         blk_status_t io_error;
 898         struct bio *bio;
 899         struct mapped_device *md = io->md;
 900 
 901         /* Push-back supersedes any I/O errors */
 902         if (unlikely(error)) {
 903                 spin_lock_irqsave(&io->endio_lock, flags);
 904                 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
 905                         io->status = error;
 906                 spin_unlock_irqrestore(&io->endio_lock, flags);
 907         }
 908 
 909         if (atomic_dec_and_test(&io->io_count)) {
 910                 if (io->status == BLK_STS_DM_REQUEUE) {
 911                         /*
 912                          * Target requested pushing back the I/O.
 913                          */
 914                         spin_lock_irqsave(&md->deferred_lock, flags);
 915                         if (__noflush_suspending(md))
 916                                 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
 917                                 bio_list_add_head(&md->deferred, io->orig_bio);
 918                         else
 919                                 /* noflush suspend was interrupted. */
 920                                 io->status = BLK_STS_IOERR;
 921                         spin_unlock_irqrestore(&md->deferred_lock, flags);
 922                 }
 923 
 924                 io_error = io->status;
 925                 bio = io->orig_bio;
 926                 end_io_acct(io);
 927                 free_io(md, io);
 928 
 929                 if (io_error == BLK_STS_DM_REQUEUE)
 930                         return;
 931 
 932                 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
 933                         /*
 934                          * Preflush done for flush with data, reissue
 935                          * without REQ_PREFLUSH.
 936                          */
 937                         bio->bi_opf &= ~REQ_PREFLUSH;
 938                         queue_io(md, bio);
 939                 } else {
 940                         /* done with normal IO or empty flush */
 941                         if (io_error)
 942                                 bio->bi_status = io_error;
 943                         bio_endio(bio);
 944                 }
 945         }
 946 }
 947 
 948 void disable_discard(struct mapped_device *md)
 949 {
 950         struct queue_limits *limits = dm_get_queue_limits(md);
 951 
 952         /* device doesn't really support DISCARD, disable it */
 953         limits->max_discard_sectors = 0;
 954         blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
 955 }
 956 
 957 void disable_write_same(struct mapped_device *md)
 958 {
 959         struct queue_limits *limits = dm_get_queue_limits(md);
 960 
 961         /* device doesn't really support WRITE SAME, disable it */
 962         limits->max_write_same_sectors = 0;
 963 }
 964 
 965 void disable_write_zeroes(struct mapped_device *md)
 966 {
 967         struct queue_limits *limits = dm_get_queue_limits(md);
 968 
 969         /* device doesn't really support WRITE ZEROES, disable it */
 970         limits->max_write_zeroes_sectors = 0;
 971 }
 972 
 973 static void clone_endio(struct bio *bio)
 974 {
 975         blk_status_t error = bio->bi_status;
 976         struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
 977         struct dm_io *io = tio->io;
 978         struct mapped_device *md = tio->io->md;
 979         dm_endio_fn endio = tio->ti->type->end_io;
 980 
 981         if (unlikely(error == BLK_STS_TARGET) && md->type != DM_TYPE_NVME_BIO_BASED) {
 982                 if (bio_op(bio) == REQ_OP_DISCARD &&
 983                     !bio->bi_disk->queue->limits.max_discard_sectors)
 984                         disable_discard(md);
 985                 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
 986                          !bio->bi_disk->queue->limits.max_write_same_sectors)
 987                         disable_write_same(md);
 988                 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
 989                          !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
 990                         disable_write_zeroes(md);
 991         }
 992 
 993         if (endio) {
 994                 int r = endio(tio->ti, bio, &error);
 995                 switch (r) {
 996                 case DM_ENDIO_REQUEUE:
 997                         error = BLK_STS_DM_REQUEUE;
 998                         /*FALLTHRU*/
 999                 case DM_ENDIO_DONE:
1000                         break;
1001                 case DM_ENDIO_INCOMPLETE:
1002                         /* The target will handle the io */
1003                         return;
1004                 default:
1005                         DMWARN("unimplemented target endio return value: %d", r);
1006                         BUG();
1007                 }
1008         }
1009 
1010         free_tio(tio);
1011         dec_pending(io, error);
1012 }
1013 
1014 /*
1015  * Return maximum size of I/O possible at the supplied sector up to the current
1016  * target boundary.
1017  */
1018 static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1019 {
1020         sector_t target_offset = dm_target_offset(ti, sector);
1021 
1022         return ti->len - target_offset;
1023 }
1024 
1025 static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1026 {
1027         sector_t len = max_io_len_target_boundary(sector, ti);
1028         sector_t offset, max_len;
1029 
1030         /*
1031          * Does the target need to split even further?
1032          */
1033         if (ti->max_io_len) {
1034                 offset = dm_target_offset(ti, sector);
1035                 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1036                         max_len = sector_div(offset, ti->max_io_len);
1037                 else
1038                         max_len = offset & (ti->max_io_len - 1);
1039                 max_len = ti->max_io_len - max_len;
1040 
1041                 if (len > max_len)
1042                         len = max_len;
1043         }
1044 
1045         return len;
1046 }
1047 
1048 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1049 {
1050         if (len > UINT_MAX) {
1051                 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1052                       (unsigned long long)len, UINT_MAX);
1053                 ti->error = "Maximum size of target IO is too large";
1054                 return -EINVAL;
1055         }
1056 
1057         ti->max_io_len = (uint32_t) len;
1058 
1059         return 0;
1060 }
1061 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1062 
1063 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1064                                                 sector_t sector, int *srcu_idx)
1065         __acquires(md->io_barrier)
1066 {
1067         struct dm_table *map;
1068         struct dm_target *ti;
1069 
1070         map = dm_get_live_table(md, srcu_idx);
1071         if (!map)
1072                 return NULL;
1073 
1074         ti = dm_table_find_target(map, sector);
1075         if (!ti)
1076                 return NULL;
1077 
1078         return ti;
1079 }
1080 
1081 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1082                                  long nr_pages, void **kaddr, pfn_t *pfn)
1083 {
1084         struct mapped_device *md = dax_get_private(dax_dev);
1085         sector_t sector = pgoff * PAGE_SECTORS;
1086         struct dm_target *ti;
1087         long len, ret = -EIO;
1088         int srcu_idx;
1089 
1090         ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1091 
1092         if (!ti)
1093                 goto out;
1094         if (!ti->type->direct_access)
1095                 goto out;
1096         len = max_io_len(sector, ti) / PAGE_SECTORS;
1097         if (len < 1)
1098                 goto out;
1099         nr_pages = min(len, nr_pages);
1100         ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1101 
1102  out:
1103         dm_put_live_table(md, srcu_idx);
1104 
1105         return ret;
1106 }
1107 
1108 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1109                 int blocksize, sector_t start, sector_t len)
1110 {
1111         struct mapped_device *md = dax_get_private(dax_dev);
1112         struct dm_table *map;
1113         int srcu_idx;
1114         bool ret;
1115 
1116         map = dm_get_live_table(md, &srcu_idx);
1117         if (!map)
1118                 return false;
1119 
1120         ret = dm_table_supports_dax(map, device_supports_dax, &blocksize);
1121 
1122         dm_put_live_table(md, srcu_idx);
1123 
1124         return ret;
1125 }
1126 
1127 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1128                                     void *addr, size_t bytes, struct iov_iter *i)
1129 {
1130         struct mapped_device *md = dax_get_private(dax_dev);
1131         sector_t sector = pgoff * PAGE_SECTORS;
1132         struct dm_target *ti;
1133         long ret = 0;
1134         int srcu_idx;
1135 
1136         ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1137 
1138         if (!ti)
1139                 goto out;
1140         if (!ti->type->dax_copy_from_iter) {
1141                 ret = copy_from_iter(addr, bytes, i);
1142                 goto out;
1143         }
1144         ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1145  out:
1146         dm_put_live_table(md, srcu_idx);
1147 
1148         return ret;
1149 }
1150 
1151 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1152                 void *addr, size_t bytes, struct iov_iter *i)
1153 {
1154         struct mapped_device *md = dax_get_private(dax_dev);
1155         sector_t sector = pgoff * PAGE_SECTORS;
1156         struct dm_target *ti;
1157         long ret = 0;
1158         int srcu_idx;
1159 
1160         ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1161 
1162         if (!ti)
1163                 goto out;
1164         if (!ti->type->dax_copy_to_iter) {
1165                 ret = copy_to_iter(addr, bytes, i);
1166                 goto out;
1167         }
1168         ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1169  out:
1170         dm_put_live_table(md, srcu_idx);
1171 
1172         return ret;
1173 }
1174 
1175 /*
1176  * A target may call dm_accept_partial_bio only from the map routine.  It is
1177  * allowed for all bio types except REQ_PREFLUSH and REQ_OP_ZONE_RESET.
1178  *
1179  * dm_accept_partial_bio informs the dm that the target only wants to process
1180  * additional n_sectors sectors of the bio and the rest of the data should be
1181  * sent in a next bio.
1182  *
1183  * A diagram that explains the arithmetics:
1184  * +--------------------+---------------+-------+
1185  * |         1          |       2       |   3   |
1186  * +--------------------+---------------+-------+
1187  *
1188  * <-------------- *tio->len_ptr --------------->
1189  *                      <------- bi_size ------->
1190  *                      <-- n_sectors -->
1191  *
1192  * Region 1 was already iterated over with bio_advance or similar function.
1193  *      (it may be empty if the target doesn't use bio_advance)
1194  * Region 2 is the remaining bio size that the target wants to process.
1195  *      (it may be empty if region 1 is non-empty, although there is no reason
1196  *       to make it empty)
1197  * The target requires that region 3 is to be sent in the next bio.
1198  *
1199  * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1200  * the partially processed part (the sum of regions 1+2) must be the same for all
1201  * copies of the bio.
1202  */
1203 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1204 {
1205         struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1206         unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1207         BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1208         BUG_ON(bi_size > *tio->len_ptr);
1209         BUG_ON(n_sectors > bi_size);
1210         *tio->len_ptr -= bi_size - n_sectors;
1211         bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1212 }
1213 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1214 
1215 /*
1216  * The zone descriptors obtained with a zone report indicate
1217  * zone positions within the underlying device of the target. The zone
1218  * descriptors must be remapped to match their position within the dm device.
1219  * The caller target should obtain the zones information using
1220  * blkdev_report_zones() to ensure that remapping for partition offset is
1221  * already handled.
1222  */
1223 void dm_remap_zone_report(struct dm_target *ti, sector_t start,
1224                           struct blk_zone *zones, unsigned int *nr_zones)
1225 {
1226 #ifdef CONFIG_BLK_DEV_ZONED
1227         struct blk_zone *zone;
1228         unsigned int nrz = *nr_zones;
1229         int i;
1230 
1231         /*
1232          * Remap the start sector and write pointer position of the zones in
1233          * the array. Since we may have obtained from the target underlying
1234          * device more zones that the target size, also adjust the number
1235          * of zones.
1236          */
1237         for (i = 0; i < nrz; i++) {
1238                 zone = zones + i;
1239                 if (zone->start >= start + ti->len) {
1240                         memset(zone, 0, sizeof(struct blk_zone) * (nrz - i));
1241                         break;
1242                 }
1243 
1244                 zone->start = zone->start + ti->begin - start;
1245                 if (zone->type == BLK_ZONE_TYPE_CONVENTIONAL)
1246                         continue;
1247 
1248                 if (zone->cond == BLK_ZONE_COND_FULL)
1249                         zone->wp = zone->start + zone->len;
1250                 else if (zone->cond == BLK_ZONE_COND_EMPTY)
1251                         zone->wp = zone->start;
1252                 else
1253                         zone->wp = zone->wp + ti->begin - start;
1254         }
1255 
1256         *nr_zones = i;
1257 #else /* !CONFIG_BLK_DEV_ZONED */
1258         *nr_zones = 0;
1259 #endif
1260 }
1261 EXPORT_SYMBOL_GPL(dm_remap_zone_report);
1262 
1263 static blk_qc_t __map_bio(struct dm_target_io *tio)
1264 {
1265         int r;
1266         sector_t sector;
1267         struct bio *clone = &tio->clone;
1268         struct dm_io *io = tio->io;
1269         struct mapped_device *md = io->md;
1270         struct dm_target *ti = tio->ti;
1271         blk_qc_t ret = BLK_QC_T_NONE;
1272 
1273         clone->bi_end_io = clone_endio;
1274 
1275         /*
1276          * Map the clone.  If r == 0 we don't need to do
1277          * anything, the target has assumed ownership of
1278          * this io.
1279          */
1280         atomic_inc(&io->io_count);
1281         sector = clone->bi_iter.bi_sector;
1282 
1283         r = ti->type->map(ti, clone);
1284         switch (r) {
1285         case DM_MAPIO_SUBMITTED:
1286                 break;
1287         case DM_MAPIO_REMAPPED:
1288                 /* the bio has been remapped so dispatch it */
1289                 trace_block_bio_remap(clone->bi_disk->queue, clone,
1290                                       bio_dev(io->orig_bio), sector);
1291                 if (md->type == DM_TYPE_NVME_BIO_BASED)
1292                         ret = direct_make_request(clone);
1293                 else
1294                         ret = generic_make_request(clone);
1295                 break;
1296         case DM_MAPIO_KILL:
1297                 free_tio(tio);
1298                 dec_pending(io, BLK_STS_IOERR);
1299                 break;
1300         case DM_MAPIO_REQUEUE:
1301                 free_tio(tio);
1302                 dec_pending(io, BLK_STS_DM_REQUEUE);
1303                 break;
1304         default:
1305                 DMWARN("unimplemented target map return value: %d", r);
1306                 BUG();
1307         }
1308 
1309         return ret;
1310 }
1311 
1312 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1313 {
1314         bio->bi_iter.bi_sector = sector;
1315         bio->bi_iter.bi_size = to_bytes(len);
1316 }
1317 
1318 /*
1319  * Creates a bio that consists of range of complete bvecs.
1320  */
1321 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1322                      sector_t sector, unsigned len)
1323 {
1324         struct bio *clone = &tio->clone;
1325 
1326         __bio_clone_fast(clone, bio);
1327 
1328         if (bio_integrity(bio)) {
1329                 int r;
1330 
1331                 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1332                              !dm_target_passes_integrity(tio->ti->type))) {
1333                         DMWARN("%s: the target %s doesn't support integrity data.",
1334                                 dm_device_name(tio->io->md),
1335                                 tio->ti->type->name);
1336                         return -EIO;
1337                 }
1338 
1339                 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1340                 if (r < 0)
1341                         return r;
1342         }
1343 
1344         bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1345         clone->bi_iter.bi_size = to_bytes(len);
1346 
1347         if (bio_integrity(bio))
1348                 bio_integrity_trim(clone);
1349 
1350         return 0;
1351 }
1352 
1353 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1354                                 struct dm_target *ti, unsigned num_bios)
1355 {
1356         struct dm_target_io *tio;
1357         int try;
1358 
1359         if (!num_bios)
1360                 return;
1361 
1362         if (num_bios == 1) {
1363                 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1364                 bio_list_add(blist, &tio->clone);
1365                 return;
1366         }
1367 
1368         for (try = 0; try < 2; try++) {
1369                 int bio_nr;
1370                 struct bio *bio;
1371 
1372                 if (try)
1373                         mutex_lock(&ci->io->md->table_devices_lock);
1374                 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1375                         tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1376                         if (!tio)
1377                                 break;
1378 
1379                         bio_list_add(blist, &tio->clone);
1380                 }
1381                 if (try)
1382                         mutex_unlock(&ci->io->md->table_devices_lock);
1383                 if (bio_nr == num_bios)
1384                         return;
1385 
1386                 while ((bio = bio_list_pop(blist))) {
1387                         tio = container_of(bio, struct dm_target_io, clone);
1388                         free_tio(tio);
1389                 }
1390         }
1391 }
1392 
1393 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1394                                            struct dm_target_io *tio, unsigned *len)
1395 {
1396         struct bio *clone = &tio->clone;
1397 
1398         tio->len_ptr = len;
1399 
1400         __bio_clone_fast(clone, ci->bio);
1401         if (len)
1402                 bio_setup_sector(clone, ci->sector, *len);
1403 
1404         return __map_bio(tio);
1405 }
1406 
1407 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1408                                   unsigned num_bios, unsigned *len)
1409 {
1410         struct bio_list blist = BIO_EMPTY_LIST;
1411         struct bio *bio;
1412         struct dm_target_io *tio;
1413 
1414         alloc_multiple_bios(&blist, ci, ti, num_bios);
1415 
1416         while ((bio = bio_list_pop(&blist))) {
1417                 tio = container_of(bio, struct dm_target_io, clone);
1418                 (void) __clone_and_map_simple_bio(ci, tio, len);
1419         }
1420 }
1421 
1422 static int __send_empty_flush(struct clone_info *ci)
1423 {
1424         unsigned target_nr = 0;
1425         struct dm_target *ti;
1426 
1427         /*
1428          * Empty flush uses a statically initialized bio, as the base for
1429          * cloning.  However, blkg association requires that a bdev is
1430          * associated with a gendisk, which doesn't happen until the bdev is
1431          * opened.  So, blkg association is done at issue time of the flush
1432          * rather than when the device is created in alloc_dev().
1433          */
1434         bio_set_dev(ci->bio, ci->io->md->bdev);
1435 
1436         BUG_ON(bio_has_data(ci->bio));
1437         while ((ti = dm_table_get_target(ci->map, target_nr++)))
1438                 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1439 
1440         bio_disassociate_blkg(ci->bio);
1441 
1442         return 0;
1443 }
1444 
1445 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1446                                     sector_t sector, unsigned *len)
1447 {
1448         struct bio *bio = ci->bio;
1449         struct dm_target_io *tio;
1450         int r;
1451 
1452         tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1453         tio->len_ptr = len;
1454         r = clone_bio(tio, bio, sector, *len);
1455         if (r < 0) {
1456                 free_tio(tio);
1457                 return r;
1458         }
1459         (void) __map_bio(tio);
1460 
1461         return 0;
1462 }
1463 
1464 typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1465 
1466 static unsigned get_num_discard_bios(struct dm_target *ti)
1467 {
1468         return ti->num_discard_bios;
1469 }
1470 
1471 static unsigned get_num_secure_erase_bios(struct dm_target *ti)
1472 {
1473         return ti->num_secure_erase_bios;
1474 }
1475 
1476 static unsigned get_num_write_same_bios(struct dm_target *ti)
1477 {
1478         return ti->num_write_same_bios;
1479 }
1480 
1481 static unsigned get_num_write_zeroes_bios(struct dm_target *ti)
1482 {
1483         return ti->num_write_zeroes_bios;
1484 }
1485 
1486 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1487                                        unsigned num_bios)
1488 {
1489         unsigned len;
1490 
1491         /*
1492          * Even though the device advertised support for this type of
1493          * request, that does not mean every target supports it, and
1494          * reconfiguration might also have changed that since the
1495          * check was performed.
1496          */
1497         if (!num_bios)
1498                 return -EOPNOTSUPP;
1499 
1500         len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1501 
1502         __send_duplicate_bios(ci, ti, num_bios, &len);
1503 
1504         ci->sector += len;
1505         ci->sector_count -= len;
1506 
1507         return 0;
1508 }
1509 
1510 static int __send_discard(struct clone_info *ci, struct dm_target *ti)
1511 {
1512         return __send_changing_extent_only(ci, ti, get_num_discard_bios(ti));
1513 }
1514 
1515 static int __send_secure_erase(struct clone_info *ci, struct dm_target *ti)
1516 {
1517         return __send_changing_extent_only(ci, ti, get_num_secure_erase_bios(ti));
1518 }
1519 
1520 static int __send_write_same(struct clone_info *ci, struct dm_target *ti)
1521 {
1522         return __send_changing_extent_only(ci, ti, get_num_write_same_bios(ti));
1523 }
1524 
1525 static int __send_write_zeroes(struct clone_info *ci, struct dm_target *ti)
1526 {
1527         return __send_changing_extent_only(ci, ti, get_num_write_zeroes_bios(ti));
1528 }
1529 
1530 static bool is_abnormal_io(struct bio *bio)
1531 {
1532         bool r = false;
1533 
1534         switch (bio_op(bio)) {
1535         case REQ_OP_DISCARD:
1536         case REQ_OP_SECURE_ERASE:
1537         case REQ_OP_WRITE_SAME:
1538         case REQ_OP_WRITE_ZEROES:
1539                 r = true;
1540                 break;
1541         }
1542 
1543         return r;
1544 }
1545 
1546 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1547                                   int *result)
1548 {
1549         struct bio *bio = ci->bio;
1550 
1551         if (bio_op(bio) == REQ_OP_DISCARD)
1552                 *result = __send_discard(ci, ti);
1553         else if (bio_op(bio) == REQ_OP_SECURE_ERASE)
1554                 *result = __send_secure_erase(ci, ti);
1555         else if (bio_op(bio) == REQ_OP_WRITE_SAME)
1556                 *result = __send_write_same(ci, ti);
1557         else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
1558                 *result = __send_write_zeroes(ci, ti);
1559         else
1560                 return false;
1561 
1562         return true;
1563 }
1564 
1565 /*
1566  * Select the correct strategy for processing a non-flush bio.
1567  */
1568 static int __split_and_process_non_flush(struct clone_info *ci)
1569 {
1570         struct dm_target *ti;
1571         unsigned len;
1572         int r;
1573 
1574         ti = dm_table_find_target(ci->map, ci->sector);
1575         if (!ti)
1576                 return -EIO;
1577 
1578         if (__process_abnormal_io(ci, ti, &r))
1579                 return r;
1580 
1581         len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1582 
1583         r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1584         if (r < 0)
1585                 return r;
1586 
1587         ci->sector += len;
1588         ci->sector_count -= len;
1589 
1590         return 0;
1591 }
1592 
1593 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1594                             struct dm_table *map, struct bio *bio)
1595 {
1596         ci->map = map;
1597         ci->io = alloc_io(md, bio);
1598         ci->sector = bio->bi_iter.bi_sector;
1599 }
1600 
1601 #define __dm_part_stat_sub(part, field, subnd)  \
1602         (part_stat_get(part, field) -= (subnd))
1603 
1604 /*
1605  * Entry point to split a bio into clones and submit them to the targets.
1606  */
1607 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1608                                         struct dm_table *map, struct bio *bio)
1609 {
1610         struct clone_info ci;
1611         blk_qc_t ret = BLK_QC_T_NONE;
1612         int error = 0;
1613 
1614         init_clone_info(&ci, md, map, bio);
1615 
1616         if (bio->bi_opf & REQ_PREFLUSH) {
1617                 struct bio flush_bio;
1618 
1619                 /*
1620                  * Use an on-stack bio for this, it's safe since we don't
1621                  * need to reference it after submit. It's just used as
1622                  * the basis for the clone(s).
1623                  */
1624                 bio_init(&flush_bio, NULL, 0);
1625                 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1626                 ci.bio = &flush_bio;
1627                 ci.sector_count = 0;
1628                 error = __send_empty_flush(&ci);
1629                 /* dec_pending submits any data associated with flush */
1630         } else if (bio_op(bio) == REQ_OP_ZONE_RESET) {
1631                 ci.bio = bio;
1632                 ci.sector_count = 0;
1633                 error = __split_and_process_non_flush(&ci);
1634         } else {
1635                 ci.bio = bio;
1636                 ci.sector_count = bio_sectors(bio);
1637                 while (ci.sector_count && !error) {
1638                         error = __split_and_process_non_flush(&ci);
1639                         if (current->bio_list && ci.sector_count && !error) {
1640                                 /*
1641                                  * Remainder must be passed to generic_make_request()
1642                                  * so that it gets handled *after* bios already submitted
1643                                  * have been completely processed.
1644                                  * We take a clone of the original to store in
1645                                  * ci.io->orig_bio to be used by end_io_acct() and
1646                                  * for dec_pending to use for completion handling.
1647                                  */
1648                                 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1649                                                           GFP_NOIO, &md->queue->bio_split);
1650                                 ci.io->orig_bio = b;
1651 
1652                                 /*
1653                                  * Adjust IO stats for each split, otherwise upon queue
1654                                  * reentry there will be redundant IO accounting.
1655                                  * NOTE: this is a stop-gap fix, a proper fix involves
1656                                  * significant refactoring of DM core's bio splitting
1657                                  * (by eliminating DM's splitting and just using bio_split)
1658                                  */
1659                                 part_stat_lock();
1660                                 __dm_part_stat_sub(&dm_disk(md)->part0,
1661                                                    sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1662                                 part_stat_unlock();
1663 
1664                                 bio_chain(b, bio);
1665                                 trace_block_split(md->queue, b, bio->bi_iter.bi_sector);
1666                                 ret = generic_make_request(bio);
1667                                 break;
1668                         }
1669                 }
1670         }
1671 
1672         /* drop the extra reference count */
1673         dec_pending(ci.io, errno_to_blk_status(error));
1674         return ret;
1675 }
1676 
1677 /*
1678  * Optimized variant of __split_and_process_bio that leverages the
1679  * fact that targets that use it do _not_ have a need to split bios.
1680  */
1681 static blk_qc_t __process_bio(struct mapped_device *md, struct dm_table *map,
1682                               struct bio *bio, struct dm_target *ti)
1683 {
1684         struct clone_info ci;
1685         blk_qc_t ret = BLK_QC_T_NONE;
1686         int error = 0;
1687 
1688         init_clone_info(&ci, md, map, bio);
1689 
1690         if (bio->bi_opf & REQ_PREFLUSH) {
1691                 struct bio flush_bio;
1692 
1693                 /*
1694                  * Use an on-stack bio for this, it's safe since we don't
1695                  * need to reference it after submit. It's just used as
1696                  * the basis for the clone(s).
1697                  */
1698                 bio_init(&flush_bio, NULL, 0);
1699                 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1700                 ci.bio = &flush_bio;
1701                 ci.sector_count = 0;
1702                 error = __send_empty_flush(&ci);
1703                 /* dec_pending submits any data associated with flush */
1704         } else {
1705                 struct dm_target_io *tio;
1706 
1707                 ci.bio = bio;
1708                 ci.sector_count = bio_sectors(bio);
1709                 if (__process_abnormal_io(&ci, ti, &error))
1710                         goto out;
1711 
1712                 tio = alloc_tio(&ci, ti, 0, GFP_NOIO);
1713                 ret = __clone_and_map_simple_bio(&ci, tio, NULL);
1714         }
1715 out:
1716         /* drop the extra reference count */
1717         dec_pending(ci.io, errno_to_blk_status(error));
1718         return ret;
1719 }
1720 
1721 static void dm_queue_split(struct mapped_device *md, struct dm_target *ti, struct bio **bio)
1722 {
1723         unsigned len, sector_count;
1724 
1725         sector_count = bio_sectors(*bio);
1726         len = min_t(sector_t, max_io_len((*bio)->bi_iter.bi_sector, ti), sector_count);
1727 
1728         if (sector_count > len) {
1729                 struct bio *split = bio_split(*bio, len, GFP_NOIO, &md->queue->bio_split);
1730 
1731                 bio_chain(split, *bio);
1732                 trace_block_split(md->queue, split, (*bio)->bi_iter.bi_sector);
1733                 generic_make_request(*bio);
1734                 *bio = split;
1735         }
1736 }
1737 
1738 static blk_qc_t dm_process_bio(struct mapped_device *md,
1739                                struct dm_table *map, struct bio *bio)
1740 {
1741         blk_qc_t ret = BLK_QC_T_NONE;
1742         struct dm_target *ti = md->immutable_target;
1743 
1744         if (unlikely(!map)) {
1745                 bio_io_error(bio);
1746                 return ret;
1747         }
1748 
1749         if (!ti) {
1750                 ti = dm_table_find_target(map, bio->bi_iter.bi_sector);
1751                 if (unlikely(!ti)) {
1752                         bio_io_error(bio);
1753                         return ret;
1754                 }
1755         }
1756 
1757         /*
1758          * If in ->make_request_fn we need to use blk_queue_split(), otherwise
1759          * queue_limits for abnormal requests (e.g. discard, writesame, etc)
1760          * won't be imposed.
1761          */
1762         if (current->bio_list) {
1763                 if (is_abnormal_io(bio))
1764                         blk_queue_split(md->queue, &bio);
1765                 else
1766                         dm_queue_split(md, ti, &bio);
1767         }
1768 
1769         if (dm_get_md_type(md) == DM_TYPE_NVME_BIO_BASED)
1770                 return __process_bio(md, map, bio, ti);
1771         else
1772                 return __split_and_process_bio(md, map, bio);
1773 }
1774 
1775 static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio)
1776 {
1777         struct mapped_device *md = q->queuedata;
1778         blk_qc_t ret = BLK_QC_T_NONE;
1779         int srcu_idx;
1780         struct dm_table *map;
1781 
1782         map = dm_get_live_table(md, &srcu_idx);
1783 
1784         /* if we're suspended, we have to queue this io for later */
1785         if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1786                 dm_put_live_table(md, srcu_idx);
1787 
1788                 if (!(bio->bi_opf & REQ_RAHEAD))
1789                         queue_io(md, bio);
1790                 else
1791                         bio_io_error(bio);
1792                 return ret;
1793         }
1794 
1795         ret = dm_process_bio(md, map, bio);
1796 
1797         dm_put_live_table(md, srcu_idx);
1798         return ret;
1799 }
1800 
1801 static int dm_any_congested(void *congested_data, int bdi_bits)
1802 {
1803         int r = bdi_bits;
1804         struct mapped_device *md = congested_data;
1805         struct dm_table *map;
1806 
1807         if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1808                 if (dm_request_based(md)) {
1809                         /*
1810                          * With request-based DM we only need to check the
1811                          * top-level queue for congestion.
1812                          */
1813                         struct backing_dev_info *bdi = md->queue->backing_dev_info;
1814                         r = bdi->wb.congested->state & bdi_bits;
1815                 } else {
1816                         map = dm_get_live_table_fast(md);
1817                         if (map)
1818                                 r = dm_table_any_congested(map, bdi_bits);
1819                         dm_put_live_table_fast(md);
1820                 }
1821         }
1822 
1823         return r;
1824 }
1825 
1826 /*-----------------------------------------------------------------
1827  * An IDR is used to keep track of allocated minor numbers.
1828  *---------------------------------------------------------------*/
1829 static void free_minor(int minor)
1830 {
1831         spin_lock(&_minor_lock);
1832         idr_remove(&_minor_idr, minor);
1833         spin_unlock(&_minor_lock);
1834 }
1835 
1836 /*
1837  * See if the device with a specific minor # is free.
1838  */
1839 static int specific_minor(int minor)
1840 {
1841         int r;
1842 
1843         if (minor >= (1 << MINORBITS))
1844                 return -EINVAL;
1845 
1846         idr_preload(GFP_KERNEL);
1847         spin_lock(&_minor_lock);
1848 
1849         r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1850 
1851         spin_unlock(&_minor_lock);
1852         idr_preload_end();
1853         if (r < 0)
1854                 return r == -ENOSPC ? -EBUSY : r;
1855         return 0;
1856 }
1857 
1858 static int next_free_minor(int *minor)
1859 {
1860         int r;
1861 
1862         idr_preload(GFP_KERNEL);
1863         spin_lock(&_minor_lock);
1864 
1865         r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1866 
1867         spin_unlock(&_minor_lock);
1868         idr_preload_end();
1869         if (r < 0)
1870                 return r;
1871         *minor = r;
1872         return 0;
1873 }
1874 
1875 static const struct block_device_operations dm_blk_dops;
1876 static const struct dax_operations dm_dax_ops;
1877 
1878 static void dm_wq_work(struct work_struct *work);
1879 
1880 static void cleanup_mapped_device(struct mapped_device *md)
1881 {
1882         if (md->wq)
1883                 destroy_workqueue(md->wq);
1884         bioset_exit(&md->bs);
1885         bioset_exit(&md->io_bs);
1886 
1887         if (md->dax_dev) {
1888                 kill_dax(md->dax_dev);
1889                 put_dax(md->dax_dev);
1890                 md->dax_dev = NULL;
1891         }
1892 
1893         if (md->disk) {
1894                 spin_lock(&_minor_lock);
1895                 md->disk->private_data = NULL;
1896                 spin_unlock(&_minor_lock);
1897                 del_gendisk(md->disk);
1898                 put_disk(md->disk);
1899         }
1900 
1901         if (md->queue)
1902                 blk_cleanup_queue(md->queue);
1903 
1904         cleanup_srcu_struct(&md->io_barrier);
1905 
1906         if (md->bdev) {
1907                 bdput(md->bdev);
1908                 md->bdev = NULL;
1909         }
1910 
1911         mutex_destroy(&md->suspend_lock);
1912         mutex_destroy(&md->type_lock);
1913         mutex_destroy(&md->table_devices_lock);
1914 
1915         dm_mq_cleanup_mapped_device(md);
1916 }
1917 
1918 /*
1919  * Allocate and initialise a blank device with a given minor.
1920  */
1921 static struct mapped_device *alloc_dev(int minor)
1922 {
1923         int r, numa_node_id = dm_get_numa_node();
1924         struct mapped_device *md;
1925         void *old_md;
1926 
1927         md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1928         if (!md) {
1929                 DMWARN("unable to allocate device, out of memory.");
1930                 return NULL;
1931         }
1932 
1933         if (!try_module_get(THIS_MODULE))
1934                 goto bad_module_get;
1935 
1936         /* get a minor number for the dev */
1937         if (minor == DM_ANY_MINOR)
1938                 r = next_free_minor(&minor);
1939         else
1940                 r = specific_minor(minor);
1941         if (r < 0)
1942                 goto bad_minor;
1943 
1944         r = init_srcu_struct(&md->io_barrier);
1945         if (r < 0)
1946                 goto bad_io_barrier;
1947 
1948         md->numa_node_id = numa_node_id;
1949         md->init_tio_pdu = false;
1950         md->type = DM_TYPE_NONE;
1951         mutex_init(&md->suspend_lock);
1952         mutex_init(&md->type_lock);
1953         mutex_init(&md->table_devices_lock);
1954         spin_lock_init(&md->deferred_lock);
1955         atomic_set(&md->holders, 1);
1956         atomic_set(&md->open_count, 0);
1957         atomic_set(&md->event_nr, 0);
1958         atomic_set(&md->uevent_seq, 0);
1959         INIT_LIST_HEAD(&md->uevent_list);
1960         INIT_LIST_HEAD(&md->table_devices);
1961         spin_lock_init(&md->uevent_lock);
1962 
1963         md->queue = blk_alloc_queue_node(GFP_KERNEL, numa_node_id);
1964         if (!md->queue)
1965                 goto bad;
1966         md->queue->queuedata = md;
1967         /*
1968          * default to bio-based required ->make_request_fn until DM
1969          * table is loaded and md->type established. If request-based
1970          * table is loaded: blk-mq will override accordingly.
1971          */
1972         blk_queue_make_request(md->queue, dm_make_request);
1973 
1974         md->disk = alloc_disk_node(1, md->numa_node_id);
1975         if (!md->disk)
1976                 goto bad;
1977 
1978         init_waitqueue_head(&md->wait);
1979         INIT_WORK(&md->work, dm_wq_work);
1980         init_waitqueue_head(&md->eventq);
1981         init_completion(&md->kobj_holder.completion);
1982 
1983         md->disk->major = _major;
1984         md->disk->first_minor = minor;
1985         md->disk->fops = &dm_blk_dops;
1986         md->disk->queue = md->queue;
1987         md->disk->private_data = md;
1988         sprintf(md->disk->disk_name, "dm-%d", minor);
1989 
1990         if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1991                 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1992                                         &dm_dax_ops, 0);
1993                 if (!md->dax_dev)
1994                         goto bad;
1995         }
1996 
1997         add_disk_no_queue_reg(md->disk);
1998         format_dev_t(md->name, MKDEV(_major, minor));
1999 
2000         md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
2001         if (!md->wq)
2002                 goto bad;
2003 
2004         md->bdev = bdget_disk(md->disk, 0);
2005         if (!md->bdev)
2006                 goto bad;
2007 
2008         dm_stats_init(&md->stats);
2009 
2010         /* Populate the mapping, nobody knows we exist yet */
2011         spin_lock(&_minor_lock);
2012         old_md = idr_replace(&_minor_idr, md, minor);
2013         spin_unlock(&_minor_lock);
2014 
2015         BUG_ON(old_md != MINOR_ALLOCED);
2016 
2017         return md;
2018 
2019 bad:
2020         cleanup_mapped_device(md);
2021 bad_io_barrier:
2022         free_minor(minor);
2023 bad_minor:
2024         module_put(THIS_MODULE);
2025 bad_module_get:
2026         kvfree(md);
2027         return NULL;
2028 }
2029 
2030 static void unlock_fs(struct mapped_device *md);
2031 
2032 static void free_dev(struct mapped_device *md)
2033 {
2034         int minor = MINOR(disk_devt(md->disk));
2035 
2036         unlock_fs(md);
2037 
2038         cleanup_mapped_device(md);
2039 
2040         free_table_devices(&md->table_devices);
2041         dm_stats_cleanup(&md->stats);
2042         free_minor(minor);
2043 
2044         module_put(THIS_MODULE);
2045         kvfree(md);
2046 }
2047 
2048 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
2049 {
2050         struct dm_md_mempools *p = dm_table_get_md_mempools(t);
2051         int ret = 0;
2052 
2053         if (dm_table_bio_based(t)) {
2054                 /*
2055                  * The md may already have mempools that need changing.
2056                  * If so, reload bioset because front_pad may have changed
2057                  * because a different table was loaded.
2058                  */
2059                 bioset_exit(&md->bs);
2060                 bioset_exit(&md->io_bs);
2061 
2062         } else if (bioset_initialized(&md->bs)) {
2063                 /*
2064                  * There's no need to reload with request-based dm
2065                  * because the size of front_pad doesn't change.
2066                  * Note for future: If you are to reload bioset,
2067                  * prep-ed requests in the queue may refer
2068                  * to bio from the old bioset, so you must walk
2069                  * through the queue to unprep.
2070                  */
2071                 goto out;
2072         }
2073 
2074         BUG_ON(!p ||
2075                bioset_initialized(&md->bs) ||
2076                bioset_initialized(&md->io_bs));
2077 
2078         ret = bioset_init_from_src(&md->bs, &p->bs);
2079         if (ret)
2080                 goto out;
2081         ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
2082         if (ret)
2083                 bioset_exit(&md->bs);
2084 out:
2085         /* mempool bind completed, no longer need any mempools in the table */
2086         dm_table_free_md_mempools(t);
2087         return ret;
2088 }
2089 
2090 /*
2091  * Bind a table to the device.
2092  */
2093 static void event_callback(void *context)
2094 {
2095         unsigned long flags;
2096         LIST_HEAD(uevents);
2097         struct mapped_device *md = (struct mapped_device *) context;
2098 
2099         spin_lock_irqsave(&md->uevent_lock, flags);
2100         list_splice_init(&md->uevent_list, &uevents);
2101         spin_unlock_irqrestore(&md->uevent_lock, flags);
2102 
2103         dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2104 
2105         atomic_inc(&md->event_nr);
2106         wake_up(&md->eventq);
2107         dm_issue_global_event();
2108 }
2109 
2110 /*
2111  * Protected by md->suspend_lock obtained by dm_swap_table().
2112  */
2113 static void __set_size(struct mapped_device *md, sector_t size)
2114 {
2115         lockdep_assert_held(&md->suspend_lock);
2116 
2117         set_capacity(md->disk, size);
2118 
2119         i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2120 }
2121 
2122 /*
2123  * Returns old map, which caller must destroy.
2124  */
2125 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2126                                struct queue_limits *limits)
2127 {
2128         struct dm_table *old_map;
2129         struct request_queue *q = md->queue;
2130         bool request_based = dm_table_request_based(t);
2131         sector_t size;
2132         int ret;
2133 
2134         lockdep_assert_held(&md->suspend_lock);
2135 
2136         size = dm_table_get_size(t);
2137 
2138         /*
2139          * Wipe any geometry if the size of the table changed.
2140          */
2141         if (size != dm_get_size(md))
2142                 memset(&md->geometry, 0, sizeof(md->geometry));
2143 
2144         __set_size(md, size);
2145 
2146         dm_table_event_callback(t, event_callback, md);
2147 
2148         /*
2149          * The queue hasn't been stopped yet, if the old table type wasn't
2150          * for request-based during suspension.  So stop it to prevent
2151          * I/O mapping before resume.
2152          * This must be done before setting the queue restrictions,
2153          * because request-based dm may be run just after the setting.
2154          */
2155         if (request_based)
2156                 dm_stop_queue(q);
2157 
2158         if (request_based || md->type == DM_TYPE_NVME_BIO_BASED) {
2159                 /*
2160                  * Leverage the fact that request-based DM targets and
2161                  * NVMe bio based targets are immutable singletons
2162                  * - used to optimize both dm_request_fn and dm_mq_queue_rq;
2163                  *   and __process_bio.
2164                  */
2165                 md->immutable_target = dm_table_get_immutable_target(t);
2166         }
2167 
2168         ret = __bind_mempools(md, t);
2169         if (ret) {
2170                 old_map = ERR_PTR(ret);
2171                 goto out;
2172         }
2173 
2174         old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2175         rcu_assign_pointer(md->map, (void *)t);
2176         md->immutable_target_type = dm_table_get_immutable_target_type(t);
2177 
2178         dm_table_set_restrictions(t, q, limits);
2179         if (old_map)
2180                 dm_sync_table(md);
2181 
2182 out:
2183         return old_map;
2184 }
2185 
2186 /*
2187  * Returns unbound table for the caller to free.
2188  */
2189 static struct dm_table *__unbind(struct mapped_device *md)
2190 {
2191         struct dm_table *map = rcu_dereference_protected(md->map, 1);
2192 
2193         if (!map)
2194                 return NULL;
2195 
2196         dm_table_event_callback(map, NULL, NULL);
2197         RCU_INIT_POINTER(md->map, NULL);
2198         dm_sync_table(md);
2199 
2200         return map;
2201 }
2202 
2203 /*
2204  * Constructor for a new device.
2205  */
2206 int dm_create(int minor, struct mapped_device **result)
2207 {
2208         int r;
2209         struct mapped_device *md;
2210 
2211         md = alloc_dev(minor);
2212         if (!md)
2213                 return -ENXIO;
2214 
2215         r = dm_sysfs_init(md);
2216         if (r) {
2217                 free_dev(md);
2218                 return r;
2219         }
2220 
2221         *result = md;
2222         return 0;
2223 }
2224 
2225 /*
2226  * Functions to manage md->type.
2227  * All are required to hold md->type_lock.
2228  */
2229 void dm_lock_md_type(struct mapped_device *md)
2230 {
2231         mutex_lock(&md->type_lock);
2232 }
2233 
2234 void dm_unlock_md_type(struct mapped_device *md)
2235 {
2236         mutex_unlock(&md->type_lock);
2237 }
2238 
2239 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2240 {
2241         BUG_ON(!mutex_is_locked(&md->type_lock));
2242         md->type = type;
2243 }
2244 
2245 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2246 {
2247         return md->type;
2248 }
2249 
2250 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2251 {
2252         return md->immutable_target_type;
2253 }
2254 
2255 /*
2256  * The queue_limits are only valid as long as you have a reference
2257  * count on 'md'.
2258  */
2259 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2260 {
2261         BUG_ON(!atomic_read(&md->holders));
2262         return &md->queue->limits;
2263 }
2264 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2265 
2266 static void dm_init_congested_fn(struct mapped_device *md)
2267 {
2268         md->queue->backing_dev_info->congested_data = md;
2269         md->queue->backing_dev_info->congested_fn = dm_any_congested;
2270 }
2271 
2272 /*
2273  * Setup the DM device's queue based on md's type
2274  */
2275 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2276 {
2277         int r;
2278         struct queue_limits limits;
2279         enum dm_queue_mode type = dm_get_md_type(md);
2280 
2281         switch (type) {
2282         case DM_TYPE_REQUEST_BASED:
2283                 r = dm_mq_init_request_queue(md, t);
2284                 if (r) {
2285                         DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2286                         return r;
2287                 }
2288                 dm_init_congested_fn(md);
2289                 break;
2290         case DM_TYPE_BIO_BASED:
2291         case DM_TYPE_DAX_BIO_BASED:
2292         case DM_TYPE_NVME_BIO_BASED:
2293                 dm_init_congested_fn(md);
2294                 break;
2295         case DM_TYPE_NONE:
2296                 WARN_ON_ONCE(true);
2297                 break;
2298         }
2299 
2300         r = dm_calculate_queue_limits(t, &limits);
2301         if (r) {
2302                 DMERR("Cannot calculate initial queue limits");
2303                 return r;
2304         }
2305         dm_table_set_restrictions(t, md->queue, &limits);
2306         blk_register_queue(md->disk);
2307 
2308         return 0;
2309 }
2310 
2311 struct mapped_device *dm_get_md(dev_t dev)
2312 {
2313         struct mapped_device *md;
2314         unsigned minor = MINOR(dev);
2315 
2316         if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2317                 return NULL;
2318 
2319         spin_lock(&_minor_lock);
2320 
2321         md = idr_find(&_minor_idr, minor);
2322         if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2323             test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2324                 md = NULL;
2325                 goto out;
2326         }
2327         dm_get(md);
2328 out:
2329         spin_unlock(&_minor_lock);
2330 
2331         return md;
2332 }
2333 EXPORT_SYMBOL_GPL(dm_get_md);
2334 
2335 void *dm_get_mdptr(struct mapped_device *md)
2336 {
2337         return md->interface_ptr;
2338 }
2339 
2340 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2341 {
2342         md->interface_ptr = ptr;
2343 }
2344 
2345 void dm_get(struct mapped_device *md)
2346 {
2347         atomic_inc(&md->holders);
2348         BUG_ON(test_bit(DMF_FREEING, &md->flags));
2349 }
2350 
2351 int dm_hold(struct mapped_device *md)
2352 {
2353         spin_lock(&_minor_lock);
2354         if (test_bit(DMF_FREEING, &md->flags)) {
2355                 spin_unlock(&_minor_lock);
2356                 return -EBUSY;
2357         }
2358         dm_get(md);
2359         spin_unlock(&_minor_lock);
2360         return 0;
2361 }
2362 EXPORT_SYMBOL_GPL(dm_hold);
2363 
2364 const char *dm_device_name(struct mapped_device *md)
2365 {
2366         return md->name;
2367 }
2368 EXPORT_SYMBOL_GPL(dm_device_name);
2369 
2370 static void __dm_destroy(struct mapped_device *md, bool wait)
2371 {
2372         struct dm_table *map;
2373         int srcu_idx;
2374 
2375         might_sleep();
2376 
2377         spin_lock(&_minor_lock);
2378         idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2379         set_bit(DMF_FREEING, &md->flags);
2380         spin_unlock(&_minor_lock);
2381 
2382         blk_set_queue_dying(md->queue);
2383 
2384         /*
2385          * Take suspend_lock so that presuspend and postsuspend methods
2386          * do not race with internal suspend.
2387          */
2388         mutex_lock(&md->suspend_lock);
2389         map = dm_get_live_table(md, &srcu_idx);
2390         if (!dm_suspended_md(md)) {
2391                 dm_table_presuspend_targets(map);
2392                 set_bit(DMF_SUSPENDED, &md->flags);
2393                 dm_table_postsuspend_targets(map);
2394         }
2395         /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2396         dm_put_live_table(md, srcu_idx);
2397         mutex_unlock(&md->suspend_lock);
2398 
2399         /*
2400          * Rare, but there may be I/O requests still going to complete,
2401          * for example.  Wait for all references to disappear.
2402          * No one should increment the reference count of the mapped_device,
2403          * after the mapped_device state becomes DMF_FREEING.
2404          */
2405         if (wait)
2406                 while (atomic_read(&md->holders))
2407                         msleep(1);
2408         else if (atomic_read(&md->holders))
2409                 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2410                        dm_device_name(md), atomic_read(&md->holders));
2411 
2412         dm_sysfs_exit(md);
2413         dm_table_destroy(__unbind(md));
2414         free_dev(md);
2415 }
2416 
2417 void dm_destroy(struct mapped_device *md)
2418 {
2419         __dm_destroy(md, true);
2420 }
2421 
2422 void dm_destroy_immediate(struct mapped_device *md)
2423 {
2424         __dm_destroy(md, false);
2425 }
2426 
2427 void dm_put(struct mapped_device *md)
2428 {
2429         atomic_dec(&md->holders);
2430 }
2431 EXPORT_SYMBOL_GPL(dm_put);
2432 
2433 static int dm_wait_for_completion(struct mapped_device *md, long task_state)
2434 {
2435         int r = 0;
2436         DEFINE_WAIT(wait);
2437 
2438         while (1) {
2439                 prepare_to_wait(&md->wait, &wait, task_state);
2440 
2441                 if (!md_in_flight(md))
2442                         break;
2443 
2444                 if (signal_pending_state(task_state, current)) {
2445                         r = -EINTR;
2446                         break;
2447                 }
2448 
2449                 io_schedule();
2450         }
2451         finish_wait(&md->wait, &wait);
2452 
2453         return r;
2454 }
2455 
2456 /*
2457  * Process the deferred bios
2458  */
2459 static void dm_wq_work(struct work_struct *work)
2460 {
2461         struct mapped_device *md = container_of(work, struct mapped_device,
2462                                                 work);
2463         struct bio *c;
2464         int srcu_idx;
2465         struct dm_table *map;
2466 
2467         map = dm_get_live_table(md, &srcu_idx);
2468 
2469         while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2470                 spin_lock_irq(&md->deferred_lock);
2471                 c = bio_list_pop(&md->deferred);
2472                 spin_unlock_irq(&md->deferred_lock);
2473 
2474                 if (!c)
2475                         break;
2476 
2477                 if (dm_request_based(md))
2478                         (void) generic_make_request(c);
2479                 else
2480                         (void) dm_process_bio(md, map, c);
2481         }
2482 
2483         dm_put_live_table(md, srcu_idx);
2484 }
2485 
2486 static void dm_queue_flush(struct mapped_device *md)
2487 {
2488         clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2489         smp_mb__after_atomic();
2490         queue_work(md->wq, &md->work);
2491 }
2492 
2493 /*
2494  * Swap in a new table, returning the old one for the caller to destroy.
2495  */
2496 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2497 {
2498         struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2499         struct queue_limits limits;
2500         int r;
2501 
2502         mutex_lock(&md->suspend_lock);
2503 
2504         /* device must be suspended */
2505         if (!dm_suspended_md(md))
2506                 goto out;
2507 
2508         /*
2509          * If the new table has no data devices, retain the existing limits.
2510          * This helps multipath with queue_if_no_path if all paths disappear,
2511          * then new I/O is queued based on these limits, and then some paths
2512          * reappear.
2513          */
2514         if (dm_table_has_no_data_devices(table)) {
2515                 live_map = dm_get_live_table_fast(md);
2516                 if (live_map)
2517                         limits = md->queue->limits;
2518                 dm_put_live_table_fast(md);
2519         }
2520 
2521         if (!live_map) {
2522                 r = dm_calculate_queue_limits(table, &limits);
2523                 if (r) {
2524                         map = ERR_PTR(r);
2525                         goto out;
2526                 }
2527         }
2528 
2529         map = __bind(md, table, &limits);
2530         dm_issue_global_event();
2531 
2532 out:
2533         mutex_unlock(&md->suspend_lock);
2534         return map;
2535 }
2536 
2537 /*
2538  * Functions to lock and unlock any filesystem running on the
2539  * device.
2540  */
2541 static int lock_fs(struct mapped_device *md)
2542 {
2543         int r;
2544 
2545         WARN_ON(md->frozen_sb);
2546 
2547         md->frozen_sb = freeze_bdev(md->bdev);
2548         if (IS_ERR(md->frozen_sb)) {
2549                 r = PTR_ERR(md->frozen_sb);
2550                 md->frozen_sb = NULL;
2551                 return r;
2552         }
2553 
2554         set_bit(DMF_FROZEN, &md->flags);
2555 
2556         return 0;
2557 }
2558 
2559 static void unlock_fs(struct mapped_device *md)
2560 {
2561         if (!test_bit(DMF_FROZEN, &md->flags))
2562                 return;
2563 
2564         thaw_bdev(md->bdev, md->frozen_sb);
2565         md->frozen_sb = NULL;
2566         clear_bit(DMF_FROZEN, &md->flags);
2567 }
2568 
2569 /*
2570  * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2571  * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2572  * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2573  *
2574  * If __dm_suspend returns 0, the device is completely quiescent
2575  * now. There is no request-processing activity. All new requests
2576  * are being added to md->deferred list.
2577  */
2578 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2579                         unsigned suspend_flags, long task_state,
2580                         int dmf_suspended_flag)
2581 {
2582         bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2583         bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2584         int r;
2585 
2586         lockdep_assert_held(&md->suspend_lock);
2587 
2588         /*
2589          * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2590          * This flag is cleared before dm_suspend returns.
2591          */
2592         if (noflush)
2593                 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2594         else
2595                 pr_debug("%s: suspending with flush\n", dm_device_name(md));
2596 
2597         /*
2598          * This gets reverted if there's an error later and the targets
2599          * provide the .presuspend_undo hook.
2600          */
2601         dm_table_presuspend_targets(map);
2602 
2603         /*
2604          * Flush I/O to the device.
2605          * Any I/O submitted after lock_fs() may not be flushed.
2606          * noflush takes precedence over do_lockfs.
2607          * (lock_fs() flushes I/Os and waits for them to complete.)
2608          */
2609         if (!noflush && do_lockfs) {
2610                 r = lock_fs(md);
2611                 if (r) {
2612                         dm_table_presuspend_undo_targets(map);
2613                         return r;
2614                 }
2615         }
2616 
2617         /*
2618          * Here we must make sure that no processes are submitting requests
2619          * to target drivers i.e. no one may be executing
2620          * __split_and_process_bio. This is called from dm_request and
2621          * dm_wq_work.
2622          *
2623          * To get all processes out of __split_and_process_bio in dm_request,
2624          * we take the write lock. To prevent any process from reentering
2625          * __split_and_process_bio from dm_request and quiesce the thread
2626          * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2627          * flush_workqueue(md->wq).
2628          */
2629         set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2630         if (map)
2631                 synchronize_srcu(&md->io_barrier);
2632 
2633         /*
2634          * Stop md->queue before flushing md->wq in case request-based
2635          * dm defers requests to md->wq from md->queue.
2636          */
2637         if (dm_request_based(md))
2638                 dm_stop_queue(md->queue);
2639 
2640         flush_workqueue(md->wq);
2641 
2642         /*
2643          * At this point no more requests are entering target request routines.
2644          * We call dm_wait_for_completion to wait for all existing requests
2645          * to finish.
2646          */
2647         r = dm_wait_for_completion(md, task_state);
2648         if (!r)
2649                 set_bit(dmf_suspended_flag, &md->flags);
2650 
2651         if (noflush)
2652                 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2653         if (map)
2654                 synchronize_srcu(&md->io_barrier);
2655 
2656         /* were we interrupted ? */
2657         if (r < 0) {
2658                 dm_queue_flush(md);
2659 
2660                 if (dm_request_based(md))
2661                         dm_start_queue(md->queue);
2662 
2663                 unlock_fs(md);
2664                 dm_table_presuspend_undo_targets(map);
2665                 /* pushback list is already flushed, so skip flush */
2666         }
2667 
2668         return r;
2669 }
2670 
2671 /*
2672  * We need to be able to change a mapping table under a mounted
2673  * filesystem.  For example we might want to move some data in
2674  * the background.  Before the table can be swapped with
2675  * dm_bind_table, dm_suspend must be called to flush any in
2676  * flight bios and ensure that any further io gets deferred.
2677  */
2678 /*
2679  * Suspend mechanism in request-based dm.
2680  *
2681  * 1. Flush all I/Os by lock_fs() if needed.
2682  * 2. Stop dispatching any I/O by stopping the request_queue.
2683  * 3. Wait for all in-flight I/Os to be completed or requeued.
2684  *
2685  * To abort suspend, start the request_queue.
2686  */
2687 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2688 {
2689         struct dm_table *map = NULL;
2690         int r = 0;
2691 
2692 retry:
2693         mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2694 
2695         if (dm_suspended_md(md)) {
2696                 r = -EINVAL;
2697                 goto out_unlock;
2698         }
2699 
2700         if (dm_suspended_internally_md(md)) {
2701                 /* already internally suspended, wait for internal resume */
2702                 mutex_unlock(&md->suspend_lock);
2703                 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2704                 if (r)
2705                         return r;
2706                 goto retry;
2707         }
2708 
2709         map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2710 
2711         r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2712         if (r)
2713                 goto out_unlock;
2714 
2715         dm_table_postsuspend_targets(map);
2716 
2717 out_unlock:
2718         mutex_unlock(&md->suspend_lock);
2719         return r;
2720 }
2721 
2722 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2723 {
2724         if (map) {
2725                 int r = dm_table_resume_targets(map);
2726                 if (r)
2727                         return r;
2728         }
2729 
2730         dm_queue_flush(md);
2731 
2732         /*
2733          * Flushing deferred I/Os must be done after targets are resumed
2734          * so that mapping of targets can work correctly.
2735          * Request-based dm is queueing the deferred I/Os in its request_queue.
2736          */
2737         if (dm_request_based(md))
2738                 dm_start_queue(md->queue);
2739 
2740         unlock_fs(md);
2741 
2742         return 0;
2743 }
2744 
2745 int dm_resume(struct mapped_device *md)
2746 {
2747         int r;
2748         struct dm_table *map = NULL;
2749 
2750 retry:
2751         r = -EINVAL;
2752         mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2753 
2754         if (!dm_suspended_md(md))
2755                 goto out;
2756 
2757         if (dm_suspended_internally_md(md)) {
2758                 /* already internally suspended, wait for internal resume */
2759                 mutex_unlock(&md->suspend_lock);
2760                 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2761                 if (r)
2762                         return r;
2763                 goto retry;
2764         }
2765 
2766         map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2767         if (!map || !dm_table_get_size(map))
2768                 goto out;
2769 
2770         r = __dm_resume(md, map);
2771         if (r)
2772                 goto out;
2773 
2774         clear_bit(DMF_SUSPENDED, &md->flags);
2775 out:
2776         mutex_unlock(&md->suspend_lock);
2777 
2778         return r;
2779 }
2780 
2781 /*
2782  * Internal suspend/resume works like userspace-driven suspend. It waits
2783  * until all bios finish and prevents issuing new bios to the target drivers.
2784  * It may be used only from the kernel.
2785  */
2786 
2787 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2788 {
2789         struct dm_table *map = NULL;
2790 
2791         lockdep_assert_held(&md->suspend_lock);
2792 
2793         if (md->internal_suspend_count++)
2794                 return; /* nested internal suspend */
2795 
2796         if (dm_suspended_md(md)) {
2797                 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2798                 return; /* nest suspend */
2799         }
2800 
2801         map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2802 
2803         /*
2804          * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2805          * supported.  Properly supporting a TASK_INTERRUPTIBLE internal suspend
2806          * would require changing .presuspend to return an error -- avoid this
2807          * until there is a need for more elaborate variants of internal suspend.
2808          */
2809         (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2810                             DMF_SUSPENDED_INTERNALLY);
2811 
2812         dm_table_postsuspend_targets(map);
2813 }
2814 
2815 static void __dm_internal_resume(struct mapped_device *md)
2816 {
2817         BUG_ON(!md->internal_suspend_count);
2818 
2819         if (--md->internal_suspend_count)
2820                 return; /* resume from nested internal suspend */
2821 
2822         if (dm_suspended_md(md))
2823                 goto done; /* resume from nested suspend */
2824 
2825         /*
2826          * NOTE: existing callers don't need to call dm_table_resume_targets
2827          * (which may fail -- so best to avoid it for now by passing NULL map)
2828          */
2829         (void) __dm_resume(md, NULL);
2830 
2831 done:
2832         clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2833         smp_mb__after_atomic();
2834         wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2835 }
2836 
2837 void dm_internal_suspend_noflush(struct mapped_device *md)
2838 {
2839         mutex_lock(&md->suspend_lock);
2840         __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2841         mutex_unlock(&md->suspend_lock);
2842 }
2843 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2844 
2845 void dm_internal_resume(struct mapped_device *md)
2846 {
2847         mutex_lock(&md->suspend_lock);
2848         __dm_internal_resume(md);
2849         mutex_unlock(&md->suspend_lock);
2850 }
2851 EXPORT_SYMBOL_GPL(dm_internal_resume);
2852 
2853 /*
2854  * Fast variants of internal suspend/resume hold md->suspend_lock,
2855  * which prevents interaction with userspace-driven suspend.
2856  */
2857 
2858 void dm_internal_suspend_fast(struct mapped_device *md)
2859 {
2860         mutex_lock(&md->suspend_lock);
2861         if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2862                 return;
2863 
2864         set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2865         synchronize_srcu(&md->io_barrier);
2866         flush_workqueue(md->wq);
2867         dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2868 }
2869 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2870 
2871 void dm_internal_resume_fast(struct mapped_device *md)
2872 {
2873         if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2874                 goto done;
2875 
2876         dm_queue_flush(md);
2877 
2878 done:
2879         mutex_unlock(&md->suspend_lock);
2880 }
2881 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2882 
2883 /*-----------------------------------------------------------------
2884  * Event notification.
2885  *---------------------------------------------------------------*/
2886 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2887                        unsigned cookie)
2888 {
2889         char udev_cookie[DM_COOKIE_LENGTH];
2890         char *envp[] = { udev_cookie, NULL };
2891 
2892         if (!cookie)
2893                 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2894         else {
2895                 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2896                          DM_COOKIE_ENV_VAR_NAME, cookie);
2897                 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2898                                           action, envp);
2899         }
2900 }
2901 
2902 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2903 {
2904         return atomic_add_return(1, &md->uevent_seq);
2905 }
2906 
2907 uint32_t dm_get_event_nr(struct mapped_device *md)
2908 {
2909         return atomic_read(&md->event_nr);
2910 }
2911 
2912 int dm_wait_event(struct mapped_device *md, int event_nr)
2913 {
2914         return wait_event_interruptible(md->eventq,
2915                         (event_nr != atomic_read(&md->event_nr)));
2916 }
2917 
2918 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2919 {
2920         unsigned long flags;
2921 
2922         spin_lock_irqsave(&md->uevent_lock, flags);
2923         list_add(elist, &md->uevent_list);
2924         spin_unlock_irqrestore(&md->uevent_lock, flags);
2925 }
2926 
2927 /*
2928  * The gendisk is only valid as long as you have a reference
2929  * count on 'md'.
2930  */
2931 struct gendisk *dm_disk(struct mapped_device *md)
2932 {
2933         return md->disk;
2934 }
2935 EXPORT_SYMBOL_GPL(dm_disk);
2936 
2937 struct kobject *dm_kobject(struct mapped_device *md)
2938 {
2939         return &md->kobj_holder.kobj;
2940 }
2941 
2942 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2943 {
2944         struct mapped_device *md;
2945 
2946         md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2947 
2948         spin_lock(&_minor_lock);
2949         if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2950                 md = NULL;
2951                 goto out;
2952         }
2953         dm_get(md);
2954 out:
2955         spin_unlock(&_minor_lock);
2956 
2957         return md;
2958 }
2959 
2960 int dm_suspended_md(struct mapped_device *md)
2961 {
2962         return test_bit(DMF_SUSPENDED, &md->flags);
2963 }
2964 
2965 int dm_suspended_internally_md(struct mapped_device *md)
2966 {
2967         return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2968 }
2969 
2970 int dm_test_deferred_remove_flag(struct mapped_device *md)
2971 {
2972         return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2973 }
2974 
2975 int dm_suspended(struct dm_target *ti)
2976 {
2977         return dm_suspended_md(dm_table_get_md(ti->table));
2978 }
2979 EXPORT_SYMBOL_GPL(dm_suspended);
2980 
2981 int dm_noflush_suspending(struct dm_target *ti)
2982 {
2983         return __noflush_suspending(dm_table_get_md(ti->table));
2984 }
2985 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2986 
2987 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2988                                             unsigned integrity, unsigned per_io_data_size,
2989                                             unsigned min_pool_size)
2990 {
2991         struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2992         unsigned int pool_size = 0;
2993         unsigned int front_pad, io_front_pad;
2994         int ret;
2995 
2996         if (!pools)
2997                 return NULL;
2998 
2999         switch (type) {
3000         case DM_TYPE_BIO_BASED:
3001         case DM_TYPE_DAX_BIO_BASED:
3002         case DM_TYPE_NVME_BIO_BASED:
3003                 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
3004                 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
3005                 io_front_pad = roundup(front_pad,  __alignof__(struct dm_io)) + offsetof(struct dm_io, tio);
3006                 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
3007                 if (ret)
3008                         goto out;
3009                 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
3010                         goto out;
3011                 break;
3012         case DM_TYPE_REQUEST_BASED:
3013                 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
3014                 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
3015                 /* per_io_data_size is used for blk-mq pdu at queue allocation */
3016                 break;
3017         default:
3018                 BUG();
3019         }
3020 
3021         ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
3022         if (ret)
3023                 goto out;
3024 
3025         if (integrity && bioset_integrity_create(&pools->bs, pool_size))
3026                 goto out;
3027 
3028         return pools;
3029 
3030 out:
3031         dm_free_md_mempools(pools);
3032 
3033         return NULL;
3034 }
3035 
3036 void dm_free_md_mempools(struct dm_md_mempools *pools)
3037 {
3038         if (!pools)
3039                 return;
3040 
3041         bioset_exit(&pools->bs);
3042         bioset_exit(&pools->io_bs);
3043 
3044         kfree(pools);
3045 }
3046 
3047 struct dm_pr {
3048         u64     old_key;
3049         u64     new_key;
3050         u32     flags;
3051         bool    fail_early;
3052 };
3053 
3054 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3055                       void *data)
3056 {
3057         struct mapped_device *md = bdev->bd_disk->private_data;
3058         struct dm_table *table;
3059         struct dm_target *ti;
3060         int ret = -ENOTTY, srcu_idx;
3061 
3062         table = dm_get_live_table(md, &srcu_idx);
3063         if (!table || !dm_table_get_size(table))
3064                 goto out;
3065 
3066         /* We only support devices that have a single target */
3067         if (dm_table_get_num_targets(table) != 1)
3068                 goto out;
3069         ti = dm_table_get_target(table, 0);
3070 
3071         ret = -EINVAL;
3072         if (!ti->type->iterate_devices)
3073                 goto out;
3074 
3075         ret = ti->type->iterate_devices(ti, fn, data);
3076 out:
3077         dm_put_live_table(md, srcu_idx);
3078         return ret;
3079 }
3080 
3081 /*
3082  * For register / unregister we need to manually call out to every path.
3083  */
3084 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3085                             sector_t start, sector_t len, void *data)
3086 {
3087         struct dm_pr *pr = data;
3088         const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3089 
3090         if (!ops || !ops->pr_register)
3091                 return -EOPNOTSUPP;
3092         return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3093 }
3094 
3095 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3096                           u32 flags)
3097 {
3098         struct dm_pr pr = {
3099                 .old_key        = old_key,
3100                 .new_key        = new_key,
3101                 .flags          = flags,
3102                 .fail_early     = true,
3103         };
3104         int ret;
3105 
3106         ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3107         if (ret && new_key) {
3108                 /* unregister all paths if we failed to register any path */
3109                 pr.old_key = new_key;
3110                 pr.new_key = 0;
3111                 pr.flags = 0;
3112                 pr.fail_early = false;
3113                 dm_call_pr(bdev, __dm_pr_register, &pr);
3114         }
3115 
3116         return ret;
3117 }
3118 
3119 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3120                          u32 flags)
3121 {
3122         struct mapped_device *md = bdev->bd_disk->private_data;
3123         const struct pr_ops *ops;
3124         int r, srcu_idx;
3125 
3126         r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3127         if (r < 0)
3128                 goto out;
3129 
3130         ops = bdev->bd_disk->fops->pr_ops;
3131         if (ops && ops->pr_reserve)
3132                 r = ops->pr_reserve(bdev, key, type, flags);
3133         else
3134                 r = -EOPNOTSUPP;
3135 out:
3136         dm_unprepare_ioctl(md, srcu_idx);
3137         return r;
3138 }
3139 
3140 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3141 {
3142         struct mapped_device *md = bdev->bd_disk->private_data;
3143         const struct pr_ops *ops;
3144         int r, srcu_idx;
3145 
3146         r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3147         if (r < 0)
3148                 goto out;
3149 
3150         ops = bdev->bd_disk->fops->pr_ops;
3151         if (ops && ops->pr_release)
3152                 r = ops->pr_release(bdev, key, type);
3153         else
3154                 r = -EOPNOTSUPP;
3155 out:
3156         dm_unprepare_ioctl(md, srcu_idx);
3157         return r;
3158 }
3159 
3160 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3161                          enum pr_type type, bool abort)
3162 {
3163         struct mapped_device *md = bdev->bd_disk->private_data;
3164         const struct pr_ops *ops;
3165         int r, srcu_idx;
3166 
3167         r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3168         if (r < 0)
3169                 goto out;
3170 
3171         ops = bdev->bd_disk->fops->pr_ops;
3172         if (ops && ops->pr_preempt)
3173                 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3174         else
3175                 r = -EOPNOTSUPP;
3176 out:
3177         dm_unprepare_ioctl(md, srcu_idx);
3178         return r;
3179 }
3180 
3181 static int dm_pr_clear(struct block_device *bdev, u64 key)
3182 {
3183         struct mapped_device *md = bdev->bd_disk->private_data;
3184         const struct pr_ops *ops;
3185         int r, srcu_idx;
3186 
3187         r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3188         if (r < 0)
3189                 goto out;
3190 
3191         ops = bdev->bd_disk->fops->pr_ops;
3192         if (ops && ops->pr_clear)
3193                 r = ops->pr_clear(bdev, key);
3194         else
3195                 r = -EOPNOTSUPP;
3196 out:
3197         dm_unprepare_ioctl(md, srcu_idx);
3198         return r;
3199 }
3200 
3201 static const struct pr_ops dm_pr_ops = {
3202         .pr_register    = dm_pr_register,
3203         .pr_reserve     = dm_pr_reserve,
3204         .pr_release     = dm_pr_release,
3205         .pr_preempt     = dm_pr_preempt,
3206         .pr_clear       = dm_pr_clear,
3207 };
3208 
3209 static const struct block_device_operations dm_blk_dops = {
3210         .open = dm_blk_open,
3211         .release = dm_blk_close,
3212         .ioctl = dm_blk_ioctl,
3213         .getgeo = dm_blk_getgeo,
3214         .report_zones = dm_blk_report_zones,
3215         .pr_ops = &dm_pr_ops,
3216         .owner = THIS_MODULE
3217 };
3218 
3219 static const struct dax_operations dm_dax_ops = {
3220         .direct_access = dm_dax_direct_access,
3221         .dax_supported = dm_dax_supported,
3222         .copy_from_iter = dm_dax_copy_from_iter,
3223         .copy_to_iter = dm_dax_copy_to_iter,
3224 };
3225 
3226 /*
3227  * module hooks
3228  */
3229 module_init(dm_init);
3230 module_exit(dm_exit);
3231 
3232 module_param(major, uint, 0);
3233 MODULE_PARM_DESC(major, "The major number of the device mapper");
3234 
3235 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3236 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3237 
3238 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3239 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3240 
3241 MODULE_DESCRIPTION(DM_NAME " driver");
3242 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3243 MODULE_LICENSE("GPL");