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 unsigned lock_nr(struct dm_cell_key *key, unsigned int num_locks)
 121 {
 122         return (key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) & (num_locks - 1);
 123 }
 124
 125 bool dm_cell_key_has_valid_range(struct dm_cell_key *key)
 126 {
 127         if (WARN_ON_ONCE(key->block_end - key->block_begin > BIO_PRISON_MAX_RANGE))
 128                 return false;
 129         if (WARN_ON_ONCE((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) !=
 130                          (key->block_end - 1) >> BIO_PRISON_MAX_RANGE_SHIFT))
 131                 return false;
 132
 133         return true;
 134 }
 135 EXPORT_SYMBOL(dm_cell_key_has_valid_range);
 136
 137 static int __bio_detain(struct rb_root *root,
 138                         struct dm_cell_key *key,
 139                         struct bio *inmate,
 140                         struct dm_bio_prison_cell *cell_prealloc,
 141                         struct dm_bio_prison_cell **cell_result)
 142 {
 143         int r;
 144         struct rb_node **new = &root->rb_node, *parent = NULL;
 145
 146         while (*new) {
 147                 struct dm_bio_prison_cell *cell =
 148                         rb_entry(*new, struct dm_bio_prison_cell, node);
 149
 150                 r = cmp_keys(key, &cell->key);
 151
 152                 parent = *new;
 153                 if (r < 0)
 154                         new = &((*new)->rb_left);
 155                 else if (r > 0)
 156                         new = &((*new)->rb_right);
 157                 else {
 158                         if (inmate)
 159                                 bio_list_add(&cell->bios, inmate);
 160                         *cell_result = cell;
 161                         return 1;
 162                 }
 163         }
 164
 165         __setup_new_cell(key, inmate, cell_prealloc);
 166         *cell_result = cell_prealloc;
 167
 168         rb_link_node(&cell_prealloc->node, parent, new);
 169         rb_insert_color(&cell_prealloc->node, root);
 170
 171         return 0;
 172 }
 173
 174 static int bio_detain(struct dm_bio_prison *prison,
 175                       struct dm_cell_key *key,
 176                       struct bio *inmate,
 177                       struct dm_bio_prison_cell *cell_prealloc,
 178                       struct dm_bio_prison_cell **cell_result)
 179 {
 180         int r;
 181         unsigned l = lock_nr(key, prison->num_locks);
 182
 183         spin_lock_irq(&prison->regions[l].lock);
 184         r = __bio_detain(&prison->regions[l].cell, key, inmate, cell_prealloc, cell_result);
 185         spin_unlock_irq(&prison->regions[l].lock);
 186
 187         return r;
 188 }
 189
 190 int dm_bio_detain(struct dm_bio_prison *prison,
 191                   struct dm_cell_key *key,
 192                   struct bio *inmate,
 193                   struct dm_bio_prison_cell *cell_prealloc,
 194                   struct dm_bio_prison_cell **cell_result)
 195 {
 196         return bio_detain(prison, key, inmate, cell_prealloc, cell_result);
 197 }
 198 EXPORT_SYMBOL_GPL(dm_bio_detain);
 199
 200 int dm_get_cell(struct dm_bio_prison *prison,
 201                 struct dm_cell_key *key,
 202                 struct dm_bio_prison_cell *cell_prealloc,
 203                 struct dm_bio_prison_cell **cell_result)
 204 {
 205         return bio_detain(prison, key, NULL, cell_prealloc, cell_result);
 206 }
 207 EXPORT_SYMBOL_GPL(dm_get_cell);
 208
 209 /*
 210  * @inmates must have been initialised prior to this call
 211  */
 212 static void __cell_release(struct rb_root *root,
 213                            struct dm_bio_prison_cell *cell,
 214                            struct bio_list *inmates)
 215 {
 216         rb_erase(&cell->node, root);
 217
 218         if (inmates) {
 219                 if (cell->holder)
 220                         bio_list_add(inmates, cell->holder);
 221                 bio_list_merge(inmates, &cell->bios);
 222         }
 223 }
 224
 225 void dm_cell_release(struct dm_bio_prison *prison,
 226                      struct dm_bio_prison_cell *cell,
 227                      struct bio_list *bios)
 228 {
 229         unsigned l = lock_nr(&cell->key, prison->num_locks);
 230
 231         spin_lock_irq(&prison->regions[l].lock);
 232         __cell_release(&prison->regions[l].cell, cell, bios);
 233         spin_unlock_irq(&prison->regions[l].lock);
 234 }
 235 EXPORT_SYMBOL_GPL(dm_cell_release);
 236
 237 /*
 238  * Sometimes we don't want the holder, just the additional bios.
 239  */
 240 static void __cell_release_no_holder(struct rb_root *root,
 241                                      struct dm_bio_prison_cell *cell,
 242                                      struct bio_list *inmates)
 243 {
 244         rb_erase(&cell->node, root);
 245         bio_list_merge(inmates, &cell->bios);
 246 }
 247
 248 void dm_cell_release_no_holder(struct dm_bio_prison *prison,
 249                                struct dm_bio_prison_cell *cell,
 250                                struct bio_list *inmates)
 251 {
 252         unsigned l = lock_nr(&cell->key, prison->num_locks);
 253         unsigned long flags;
 254
 255         spin_lock_irqsave(&prison->regions[l].lock, flags);
 256         __cell_release_no_holder(&prison->regions[l].cell, cell, inmates);
 257         spin_unlock_irqrestore(&prison->regions[l].lock, flags);
 258 }
 259 EXPORT_SYMBOL_GPL(dm_cell_release_no_holder);
 260
 261 void dm_cell_error(struct dm_bio_prison *prison,
 262                    struct dm_bio_prison_cell *cell, blk_status_t error)
 263 {
 264         struct bio_list bios;
 265         struct bio *bio;
 266
 267         bio_list_init(&bios);
 268         dm_cell_release(prison, cell, &bios);
 269
 270         while ((bio = bio_list_pop(&bios))) {
 271                 bio->bi_status = error;
 272                 bio_endio(bio);
 273         }
 274 }
 275 EXPORT_SYMBOL_GPL(dm_cell_error);
 276
 277 void dm_cell_visit_release(struct dm_bio_prison *prison,
 278                            void (*visit_fn)(void *, struct dm_bio_prison_cell *),
 279                            void *context,
 280                            struct dm_bio_prison_cell *cell)
 281 {
 282         unsigned l = lock_nr(&cell->key, prison->num_locks);
 283         spin_lock_irq(&prison->regions[l].lock);
 284         visit_fn(context, cell);
 285         rb_erase(&cell->node, &prison->regions[l].cell);
 286         spin_unlock_irq(&prison->regions[l].lock);
 287 }
 288 EXPORT_SYMBOL_GPL(dm_cell_visit_release);
 289
 290 static int __promote_or_release(struct rb_root *root,
 291                                 struct dm_bio_prison_cell *cell)
 292 {
 293         if (bio_list_empty(&cell->bios)) {
 294                 rb_erase(&cell->node, root);
 295                 return 1;
 296         }
 297
 298         cell->holder = bio_list_pop(&cell->bios);
 299         return 0;
 300 }
 301
 302 int dm_cell_promote_or_release(struct dm_bio_prison *prison,
 303                                struct dm_bio_prison_cell *cell)
 304 {
 305         int r;
 306         unsigned l = lock_nr(&cell->key, prison->num_locks);
 307
 308         spin_lock_irq(&prison->regions[l].lock);
 309         r = __promote_or_release(&prison->regions[l].cell, cell);
 310         spin_unlock_irq(&prison->regions[l].lock);
 311
 312         return r;
 313 }
 314 EXPORT_SYMBOL_GPL(dm_cell_promote_or_release);
 315
 316 /*----------------------------------------------------------------*/
 317
 318 #define DEFERRED_SET_SIZE 64
 319
 320 struct dm_deferred_entry {
 321         struct dm_deferred_set *ds;
 322         unsigned int count;
 323         struct list_head work_items;
 324 };
 325
 326 struct dm_deferred_set {
 327         spinlock_t lock;
 328         unsigned int current_entry;
 329         unsigned int sweeper;
 330         struct dm_deferred_entry entries[DEFERRED_SET_SIZE];
 331 };
 332
 333 struct dm_deferred_set *dm_deferred_set_create(void)
 334 {
 335         int i;
 336         struct dm_deferred_set *ds;
 337
 338         ds = kmalloc(sizeof(*ds), GFP_KERNEL);
 339         if (!ds)
 340                 return NULL;
 341
 342         spin_lock_init(&ds->lock);
 343         ds->current_entry = 0;
 344         ds->sweeper = 0;
 345         for (i = 0; i < DEFERRED_SET_SIZE; i++) {
 346                 ds->entries[i].ds = ds;
 347                 ds->entries[i].count = 0;
 348                 INIT_LIST_HEAD(&ds->entries[i].work_items);
 349         }
 350
 351         return ds;
 352 }
 353 EXPORT_SYMBOL_GPL(dm_deferred_set_create);
 354
 355 void dm_deferred_set_destroy(struct dm_deferred_set *ds)
 356 {
 357         kfree(ds);
 358 }
 359 EXPORT_SYMBOL_GPL(dm_deferred_set_destroy);
 360
 361 struct dm_deferred_entry *dm_deferred_entry_inc(struct dm_deferred_set *ds)
 362 {
 363         unsigned long flags;
 364         struct dm_deferred_entry *entry;
 365
 366         spin_lock_irqsave(&ds->lock, flags);
 367         entry = ds->entries + ds->current_entry;
 368         entry->count++;
 369         spin_unlock_irqrestore(&ds->lock, flags);
 370
 371         return entry;
 372 }
 373 EXPORT_SYMBOL_GPL(dm_deferred_entry_inc);
 374
 375 static unsigned int ds_next(unsigned int index)
 376 {
 377         return (index + 1) % DEFERRED_SET_SIZE;
 378 }
 379
 380 static void __sweep(struct dm_deferred_set *ds, struct list_head *head)
 381 {
 382         while ((ds->sweeper != ds->current_entry) &&
 383                !ds->entries[ds->sweeper].count) {
 384                 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
 385                 ds->sweeper = ds_next(ds->sweeper);
 386         }
 387
 388         if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
 389                 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
 390 }
 391
 392 void dm_deferred_entry_dec(struct dm_deferred_entry *entry, struct list_head *head)
 393 {
 394         unsigned long flags;
 395
 396         spin_lock_irqsave(&entry->ds->lock, flags);
 397         BUG_ON(!entry->count);
 398         --entry->count;
 399         __sweep(entry->ds, head);
 400         spin_unlock_irqrestore(&entry->ds->lock, flags);
 401 }
 402 EXPORT_SYMBOL_GPL(dm_deferred_entry_dec);
 403
 404 /*
 405  * Returns 1 if deferred or 0 if no pending items to delay job.
 406  */
 407 int dm_deferred_set_add_work(struct dm_deferred_set *ds, struct list_head *work)
 408 {
 409         int r = 1;
 410         unsigned int next_entry;
 411
 412         spin_lock_irq(&ds->lock);
 413         if ((ds->sweeper == ds->current_entry) &&
 414             !ds->entries[ds->current_entry].count)
 415                 r = 0;
 416         else {
 417                 list_add(work, &ds->entries[ds->current_entry].work_items);
 418                 next_entry = ds_next(ds->current_entry);
 419                 if (!ds->entries[next_entry].count)
 420                         ds->current_entry = next_entry;
 421         }
 422         spin_unlock_irq(&ds->lock);
 423
 424         return r;
 425 }
 426 EXPORT_SYMBOL_GPL(dm_deferred_set_add_work);
 427
 428 /*----------------------------------------------------------------*/
 429
 430 static int __init dm_bio_prison_init_v1(void)
 431 {
 432         _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
 433         if (!_cell_cache)
 434                 return -ENOMEM;
 435
 436         return 0;
 437 }
 438
 439 static void dm_bio_prison_exit_v1(void)
 440 {
 441         kmem_cache_destroy(_cell_cache);
 442         _cell_cache = NULL;
 443 }
 444
 445 static int (*_inits[])(void) __initdata = {
 446         dm_bio_prison_init_v1,
 447         dm_bio_prison_init_v2,
 448 };
 449
 450 static void (*_exits[])(void) = {
 451         dm_bio_prison_exit_v1,
 452         dm_bio_prison_exit_v2,
 453 };
 454
 455 static int __init dm_bio_prison_init(void)
 456 {
 457         const int count = ARRAY_SIZE(_inits);
 458
 459         int r, i;
 460
 461         for (i = 0; i < count; i++) {
 462                 r = _inits[i]();
 463                 if (r)
 464                         goto bad;
 465         }
 466
 467         return 0;
 468
 469 bad:
 470         while (i--)
 471                 _exits[i]();
 472
 473         return r;
 474 }
 475
 476 static void __exit dm_bio_prison_exit(void)
 477 {
 478         int i = ARRAY_SIZE(_exits);
 479
 480         while (i--)
 481                 _exits[i]();
 482 }
 483
 484 /*
 485  * module hooks
 486  */
 487 module_init(dm_bio_prison_init);
 488 module_exit(dm_bio_prison_exit);
 489
 490 MODULE_DESCRIPTION(DM_NAME " bio prison");
 491 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
 492 MODULE_LICENSE("GPL");