drivers/md/dm-bio-prison-v1.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2012 Red Hat, Inc.
   4  *
   5  * This file is released under the GPL.
   6  */
   7
   8 #include "dm.h"
   9 #include "dm-bio-prison-v1.h"
  10 #include "dm-bio-prison-v2.h"
  11
  12 #include <linux/spinlock.h>
  13 #include <linux/mempool.h>
  14 #include <linux/module.h>
  15 #include <linux/slab.h>
  16
  17 /*----------------------------------------------------------------*/
  18
  19 #define MIN_CELLS 1024
  20
  21 struct prison_region {
  22         spinlock_t lock;
  23         struct rb_root cell;
  24 } ____cacheline_aligned_in_smp;
  25
  26 struct dm_bio_prison {
  27         mempool_t cell_pool;
  28         unsigned int num_locks;
  29         struct prison_region regions[];
  30 };
  31
  32 static struct kmem_cache *_cell_cache;
  33
  34 /*----------------------------------------------------------------*/
  35
  36 /*
  37  * @nr_cells should be the number of cells you want in use _concurrently_.
  38  * Don't confuse it with the number of distinct keys.
  39  */
  40 struct dm_bio_prison *dm_bio_prison_create(void)
  41 {
  42         int ret;
  43         unsigned int i, num_locks;
  44         struct dm_bio_prison *prison;
  45
  46         num_locks = dm_num_hash_locks();
  47         prison = kzalloc(struct_size(prison, regions, num_locks), GFP_KERNEL);
  48         if (!prison)
  49                 return NULL;
  50         prison->num_locks = num_locks;
  51
  52         for (i = 0; i < prison->num_locks; i++) {
  53                 spin_lock_init(&prison->regions[i].lock);
  54                 prison->regions[i].cell = RB_ROOT;
  55         }
  56
  57         ret = mempool_init_slab_pool(&prison->cell_pool, MIN_CELLS, _cell_cache);
  58         if (ret) {
  59                 kfree(prison);
  60                 return NULL;
  61         }
  62
  63         return prison;
  64 }
  65 EXPORT_SYMBOL_GPL(dm_bio_prison_create);
  66
  67 void dm_bio_prison_destroy(struct dm_bio_prison *prison)
  68 {
  69         mempool_exit(&prison->cell_pool);
  70         kfree(prison);
  71 }
  72 EXPORT_SYMBOL_GPL(dm_bio_prison_destroy);
  73
  74 struct dm_bio_prison_cell *dm_bio_prison_alloc_cell(struct dm_bio_prison *prison, gfp_t gfp)
  75 {
  76         return mempool_alloc(&prison->cell_pool, gfp);
  77 }
  78 EXPORT_SYMBOL_GPL(dm_bio_prison_alloc_cell);
  79
  80 void dm_bio_prison_free_cell(struct dm_bio_prison *prison,
  81                              struct dm_bio_prison_cell *cell)
  82 {
  83         mempool_free(cell, &prison->cell_pool);
  84 }
  85 EXPORT_SYMBOL_GPL(dm_bio_prison_free_cell);
  86
  87 static void __setup_new_cell(struct dm_cell_key *key,
  88                              struct bio *holder,
  89                              struct dm_bio_prison_cell *cell)
  90 {
  91         memcpy(&cell->key, key, sizeof(cell->key));
  92         cell->holder = holder;
  93         bio_list_init(&cell->bios);
  94 }
  95
  96 static int cmp_keys(struct dm_cell_key *lhs,
  97                     struct dm_cell_key *rhs)
  98 {
  99         if (lhs->virtual < rhs->virtual)
 100                 return -1;
 101
 102         if (lhs->virtual > rhs->virtual)
 103                 return 1;
 104
 105         if (lhs->dev < rhs->dev)
 106                 return -1;
 107
 108         if (lhs->dev > rhs->dev)
 109                 return 1;
 110
 111         if (lhs->block_end <= rhs->block_begin)
 112                 return -1;
 113
 114         if (lhs->block_begin >= rhs->block_end)
 115                 return 1;
 116
 117         return 0;
 118 }
 119
 120 static inline unsigned int lock_nr(struct dm_cell_key *key, unsigned int num_locks)
 121 {
 122         return dm_hash_locks_index((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT),
 123                                    num_locks);
 124 }
 125
 126 bool dm_cell_key_has_valid_range(struct dm_cell_key *key)
 127 {
 128         if (WARN_ON_ONCE(key->block_end - key->block_begin > BIO_PRISON_MAX_RANGE))
 129                 return false;
 130         if (WARN_ON_ONCE((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) !=
 131                          (key->block_end - 1) >> BIO_PRISON_MAX_RANGE_SHIFT))
 132                 return false;
 133
 134         return true;
 135 }
 136 EXPORT_SYMBOL(dm_cell_key_has_valid_range);
 137
 138 static int __bio_detain(struct rb_root *root,
 139                         struct dm_cell_key *key,
 140                         struct bio *inmate,
 141                         struct dm_bio_prison_cell *cell_prealloc,
 142                         struct dm_bio_prison_cell **cell_result)
 143 {
 144         int r;
 145         struct rb_node **new = &root->rb_node, *parent = NULL;
 146
 147         while (*new) {
 148                 struct dm_bio_prison_cell *cell =
 149                         rb_entry(*new, struct dm_bio_prison_cell, node);
 150
 151                 r = cmp_keys(key, &cell->key);
 152
 153                 parent = *new;
 154                 if (r < 0)
 155                         new = &((*new)->rb_left);
 156                 else if (r > 0)
 157                         new = &((*new)->rb_right);
 158                 else {
 159                         if (inmate)
 160                                 bio_list_add(&cell->bios, inmate);
 161                         *cell_result = cell;
 162                         return 1;
 163                 }
 164         }
 165
 166         __setup_new_cell(key, inmate, cell_prealloc);
 167         *cell_result = cell_prealloc;
 168
 169         rb_link_node(&cell_prealloc->node, parent, new);
 170         rb_insert_color(&cell_prealloc->node, root);
 171
 172         return 0;
 173 }
 174
 175 static int bio_detain(struct dm_bio_prison *prison,
 176                       struct dm_cell_key *key,
 177                       struct bio *inmate,
 178                       struct dm_bio_prison_cell *cell_prealloc,
 179                       struct dm_bio_prison_cell **cell_result)
 180 {
 181         int r;
 182         unsigned l = lock_nr(key, prison->num_locks);
 183
 184         spin_lock_irq(&prison->regions[l].lock);
 185         r = __bio_detain(&prison->regions[l].cell, key, inmate, cell_prealloc, cell_result);
 186         spin_unlock_irq(&prison->regions[l].lock);
 187
 188         return r;
 189 }
 190
 191 int dm_bio_detain(struct dm_bio_prison *prison,
 192                   struct dm_cell_key *key,
 193                   struct bio *inmate,
 194                   struct dm_bio_prison_cell *cell_prealloc,
 195                   struct dm_bio_prison_cell **cell_result)
 196 {
 197         return bio_detain(prison, key, inmate, cell_prealloc, cell_result);
 198 }
 199 EXPORT_SYMBOL_GPL(dm_bio_detain);
 200
 201 int dm_get_cell(struct dm_bio_prison *prison,
 202                 struct dm_cell_key *key,
 203                 struct dm_bio_prison_cell *cell_prealloc,
 204                 struct dm_bio_prison_cell **cell_result)
 205 {
 206         return bio_detain(prison, key, NULL, cell_prealloc, cell_result);
 207 }
 208 EXPORT_SYMBOL_GPL(dm_get_cell);
 209
 210 /*
 211  * @inmates must have been initialised prior to this call
 212  */
 213 static void __cell_release(struct rb_root *root,
 214                            struct dm_bio_prison_cell *cell,
 215                            struct bio_list *inmates)
 216 {
 217         rb_erase(&cell->node, root);
 218
 219         if (inmates) {
 220                 if (cell->holder)
 221                         bio_list_add(inmates, cell->holder);
 222                 bio_list_merge(inmates, &cell->bios);
 223         }
 224 }
 225
 226 void dm_cell_release(struct dm_bio_prison *prison,
 227                      struct dm_bio_prison_cell *cell,
 228                      struct bio_list *bios)
 229 {
 230         unsigned l = lock_nr(&cell->key, prison->num_locks);
 231
 232         spin_lock_irq(&prison->regions[l].lock);
 233         __cell_release(&prison->regions[l].cell, cell, bios);
 234         spin_unlock_irq(&prison->regions[l].lock);
 235 }
 236 EXPORT_SYMBOL_GPL(dm_cell_release);
 237
 238 /*
 239  * Sometimes we don't want the holder, just the additional bios.
 240  */
 241 static void __cell_release_no_holder(struct rb_root *root,
 242                                      struct dm_bio_prison_cell *cell,
 243                                      struct bio_list *inmates)
 244 {
 245         rb_erase(&cell->node, root);
 246         bio_list_merge(inmates, &cell->bios);
 247 }
 248
 249 void dm_cell_release_no_holder(struct dm_bio_prison *prison,
 250                                struct dm_bio_prison_cell *cell,
 251                                struct bio_list *inmates)
 252 {
 253         unsigned l = lock_nr(&cell->key, prison->num_locks);
 254         unsigned long flags;
 255
 256         spin_lock_irqsave(&prison->regions[l].lock, flags);
 257         __cell_release_no_holder(&prison->regions[l].cell, cell, inmates);
 258         spin_unlock_irqrestore(&prison->regions[l].lock, flags);
 259 }
 260 EXPORT_SYMBOL_GPL(dm_cell_release_no_holder);
 261
 262 void dm_cell_error(struct dm_bio_prison *prison,
 263                    struct dm_bio_prison_cell *cell, blk_status_t error)
 264 {
 265         struct bio_list bios;
 266         struct bio *bio;
 267
 268         bio_list_init(&bios);
 269         dm_cell_release(prison, cell, &bios);
 270
 271         while ((bio = bio_list_pop(&bios))) {
 272                 bio->bi_status = error;
 273                 bio_endio(bio);
 274         }
 275 }
 276 EXPORT_SYMBOL_GPL(dm_cell_error);
 277
 278 void dm_cell_visit_release(struct dm_bio_prison *prison,
 279                            void (*visit_fn)(void *, struct dm_bio_prison_cell *),
 280                            void *context,
 281                            struct dm_bio_prison_cell *cell)
 282 {
 283         unsigned l = lock_nr(&cell->key, prison->num_locks);
 284         spin_lock_irq(&prison->regions[l].lock);
 285         visit_fn(context, cell);
 286         rb_erase(&cell->node, &prison->regions[l].cell);
 287         spin_unlock_irq(&prison->regions[l].lock);
 288 }
 289 EXPORT_SYMBOL_GPL(dm_cell_visit_release);
 290
 291 static int __promote_or_release(struct rb_root *root,
 292                                 struct dm_bio_prison_cell *cell)
 293 {
 294         if (bio_list_empty(&cell->bios)) {
 295                 rb_erase(&cell->node, root);
 296                 return 1;
 297         }
 298
 299         cell->holder = bio_list_pop(&cell->bios);
 300         return 0;
 301 }
 302
 303 int dm_cell_promote_or_release(struct dm_bio_prison *prison,
 304                                struct dm_bio_prison_cell *cell)
 305 {
 306         int r;
 307         unsigned l = lock_nr(&cell->key, prison->num_locks);
 308
 309         spin_lock_irq(&prison->regions[l].lock);
 310         r = __promote_or_release(&prison->regions[l].cell, cell);
 311         spin_unlock_irq(&prison->regions[l].lock);
 312
 313         return r;
 314 }
 315 EXPORT_SYMBOL_GPL(dm_cell_promote_or_release);
 316
 317 /*----------------------------------------------------------------*/
 318
 319 #define DEFERRED_SET_SIZE 64
 320
 321 struct dm_deferred_entry {
 322         struct dm_deferred_set *ds;
 323         unsigned int count;
 324         struct list_head work_items;
 325 };
 326
 327 struct dm_deferred_set {
 328         spinlock_t lock;
 329         unsigned int current_entry;
 330         unsigned int sweeper;
 331         struct dm_deferred_entry entries[DEFERRED_SET_SIZE];
 332 };
 333
 334 struct dm_deferred_set *dm_deferred_set_create(void)
 335 {
 336         int i;
 337         struct dm_deferred_set *ds;
 338
 339         ds = kmalloc(sizeof(*ds), GFP_KERNEL);
 340         if (!ds)
 341                 return NULL;
 342
 343         spin_lock_init(&ds->lock);
 344         ds->current_entry = 0;
 345         ds->sweeper = 0;
 346         for (i = 0; i < DEFERRED_SET_SIZE; i++) {
 347                 ds->entries[i].ds = ds;
 348                 ds->entries[i].count = 0;
 349                 INIT_LIST_HEAD(&ds->entries[i].work_items);
 350         }
 351
 352         return ds;
 353 }
 354 EXPORT_SYMBOL_GPL(dm_deferred_set_create);
 355
 356 void dm_deferred_set_destroy(struct dm_deferred_set *ds)
 357 {
 358         kfree(ds);
 359 }
 360 EXPORT_SYMBOL_GPL(dm_deferred_set_destroy);
 361
 362 struct dm_deferred_entry *dm_deferred_entry_inc(struct dm_deferred_set *ds)
 363 {
 364         unsigned long flags;
 365         struct dm_deferred_entry *entry;
 366
 367         spin_lock_irqsave(&ds->lock, flags);
 368         entry = ds->entries + ds->current_entry;
 369         entry->count++;
 370         spin_unlock_irqrestore(&ds->lock, flags);
 371
 372         return entry;
 373 }
 374 EXPORT_SYMBOL_GPL(dm_deferred_entry_inc);
 375
 376 static unsigned int ds_next(unsigned int index)
 377 {
 378         return (index + 1) % DEFERRED_SET_SIZE;
 379 }
 380
 381 static void __sweep(struct dm_deferred_set *ds, struct list_head *head)
 382 {
 383         while ((ds->sweeper != ds->current_entry) &&
 384                !ds->entries[ds->sweeper].count) {
 385                 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
 386                 ds->sweeper = ds_next(ds->sweeper);
 387         }
 388
 389         if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
 390                 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
 391 }
 392
 393 void dm_deferred_entry_dec(struct dm_deferred_entry *entry, struct list_head *head)
 394 {
 395         unsigned long flags;
 396
 397         spin_lock_irqsave(&entry->ds->lock, flags);
 398         BUG_ON(!entry->count);
 399         --entry->count;
 400         __sweep(entry->ds, head);
 401         spin_unlock_irqrestore(&entry->ds->lock, flags);
 402 }
 403 EXPORT_SYMBOL_GPL(dm_deferred_entry_dec);
 404
 405 /*
 406  * Returns 1 if deferred or 0 if no pending items to delay job.
 407  */
 408 int dm_deferred_set_add_work(struct dm_deferred_set *ds, struct list_head *work)
 409 {
 410         int r = 1;
 411         unsigned int next_entry;
 412
 413         spin_lock_irq(&ds->lock);
 414         if ((ds->sweeper == ds->current_entry) &&
 415             !ds->entries[ds->current_entry].count)
 416                 r = 0;
 417         else {
 418                 list_add(work, &ds->entries[ds->current_entry].work_items);
 419                 next_entry = ds_next(ds->current_entry);
 420                 if (!ds->entries[next_entry].count)
 421                         ds->current_entry = next_entry;
 422         }
 423         spin_unlock_irq(&ds->lock);
 424
 425         return r;
 426 }
 427 EXPORT_SYMBOL_GPL(dm_deferred_set_add_work);
 428
 429 /*----------------------------------------------------------------*/
 430
 431 static int __init dm_bio_prison_init_v1(void)
 432 {
 433         _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
 434         if (!_cell_cache)
 435                 return -ENOMEM;
 436
 437         return 0;
 438 }
 439
 440 static void dm_bio_prison_exit_v1(void)
 441 {
 442         kmem_cache_destroy(_cell_cache);
 443         _cell_cache = NULL;
 444 }
 445
 446 static int (*_inits[])(void) __initdata = {
 447         dm_bio_prison_init_v1,
 448         dm_bio_prison_init_v2,
 449 };
 450
 451 static void (*_exits[])(void) = {
 452         dm_bio_prison_exit_v1,
 453         dm_bio_prison_exit_v2,
 454 };
 455
 456 static int __init dm_bio_prison_init(void)
 457 {
 458         const int count = ARRAY_SIZE(_inits);
 459
 460         int r, i;
 461
 462         for (i = 0; i < count; i++) {
 463                 r = _inits[i]();
 464                 if (r)
 465                         goto bad;
 466         }
 467
 468         return 0;
 469
 470 bad:
 471         while (i--)
 472                 _exits[i]();
 473
 474         return r;
 475 }
 476
 477 static void __exit dm_bio_prison_exit(void)
 478 {
 479         int i = ARRAY_SIZE(_exits);
 480
 481         while (i--)
 482                 _exits[i]();
 483 }
 484
 485 /*
 486  * module hooks
 487  */
 488 module_init(dm_bio_prison_init);
 489 module_exit(dm_bio_prison_exit);
 490
 491 MODULE_DESCRIPTION(DM_NAME " bio prison");
 492 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
 493 MODULE_LICENSE("GPL");