2 * Copyright (C) 2012 Red Hat, Inc.
4 * This file is released under the GPL.
8 #include "dm-bio-prison.h"
10 #include <linux/spinlock.h>
11 #include <linux/mempool.h>
12 #include <linux/module.h>
13 #include <linux/slab.h>
15 /*----------------------------------------------------------------*/
17 struct dm_bio_prison_cell {
18 struct hlist_node list;
19 struct dm_bio_prison *prison;
20 struct dm_cell_key key;
25 struct dm_bio_prison {
31 struct hlist_head *cells;
34 /*----------------------------------------------------------------*/
36 static uint32_t calc_nr_buckets(unsigned nr_cells)
41 nr_cells = min(nr_cells, 8192u);
49 static struct kmem_cache *_cell_cache;
52 * @nr_cells should be the number of cells you want in use _concurrently_.
53 * Don't confuse it with the number of distinct keys.
55 struct dm_bio_prison *dm_bio_prison_create(unsigned nr_cells)
58 uint32_t nr_buckets = calc_nr_buckets(nr_cells);
59 size_t len = sizeof(struct dm_bio_prison) +
60 (sizeof(struct hlist_head) * nr_buckets);
61 struct dm_bio_prison *prison = kmalloc(len, GFP_KERNEL);
66 spin_lock_init(&prison->lock);
67 prison->cell_pool = mempool_create_slab_pool(nr_cells, _cell_cache);
68 if (!prison->cell_pool) {
73 prison->nr_buckets = nr_buckets;
74 prison->hash_mask = nr_buckets - 1;
75 prison->cells = (struct hlist_head *) (prison + 1);
76 for (i = 0; i < nr_buckets; i++)
77 INIT_HLIST_HEAD(prison->cells + i);
81 EXPORT_SYMBOL_GPL(dm_bio_prison_create);
83 void dm_bio_prison_destroy(struct dm_bio_prison *prison)
85 mempool_destroy(prison->cell_pool);
88 EXPORT_SYMBOL_GPL(dm_bio_prison_destroy);
90 static uint32_t hash_key(struct dm_bio_prison *prison, struct dm_cell_key *key)
92 const unsigned long BIG_PRIME = 4294967291UL;
93 uint64_t hash = key->block * BIG_PRIME;
95 return (uint32_t) (hash & prison->hash_mask);
98 static int keys_equal(struct dm_cell_key *lhs, struct dm_cell_key *rhs)
100 return (lhs->virtual == rhs->virtual) &&
101 (lhs->dev == rhs->dev) &&
102 (lhs->block == rhs->block);
105 static struct dm_bio_prison_cell *__search_bucket(struct hlist_head *bucket,
106 struct dm_cell_key *key)
108 struct dm_bio_prison_cell *cell;
109 struct hlist_node *tmp;
111 hlist_for_each_entry(cell, tmp, bucket, list)
112 if (keys_equal(&cell->key, key))
119 * This may block if a new cell needs allocating. You must ensure that
120 * cells will be unlocked even if the calling thread is blocked.
122 * Returns 1 if the cell was already held, 0 if @inmate is the new holder.
124 int dm_bio_detain(struct dm_bio_prison *prison, struct dm_cell_key *key,
125 struct bio *inmate, struct dm_bio_prison_cell **ref)
129 uint32_t hash = hash_key(prison, key);
130 struct dm_bio_prison_cell *cell, *cell2;
132 BUG_ON(hash > prison->nr_buckets);
134 spin_lock_irqsave(&prison->lock, flags);
136 cell = __search_bucket(prison->cells + hash, key);
138 bio_list_add(&cell->bios, inmate);
143 * Allocate a new cell
145 spin_unlock_irqrestore(&prison->lock, flags);
146 cell2 = mempool_alloc(prison->cell_pool, GFP_NOIO);
147 spin_lock_irqsave(&prison->lock, flags);
150 * We've been unlocked, so we have to double check that
151 * nobody else has inserted this cell in the meantime.
153 cell = __search_bucket(prison->cells + hash, key);
155 mempool_free(cell2, prison->cell_pool);
156 bio_list_add(&cell->bios, inmate);
165 cell->prison = prison;
166 memcpy(&cell->key, key, sizeof(cell->key));
167 cell->holder = inmate;
168 bio_list_init(&cell->bios);
169 hlist_add_head(&cell->list, prison->cells + hash);
174 spin_unlock_irqrestore(&prison->lock, flags);
180 EXPORT_SYMBOL_GPL(dm_bio_detain);
183 * @inmates must have been initialised prior to this call
185 static void __cell_release(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
187 struct dm_bio_prison *prison = cell->prison;
189 hlist_del(&cell->list);
192 bio_list_add(inmates, cell->holder);
193 bio_list_merge(inmates, &cell->bios);
196 mempool_free(cell, prison->cell_pool);
199 void dm_cell_release(struct dm_bio_prison_cell *cell, struct bio_list *bios)
202 struct dm_bio_prison *prison = cell->prison;
204 spin_lock_irqsave(&prison->lock, flags);
205 __cell_release(cell, bios);
206 spin_unlock_irqrestore(&prison->lock, flags);
208 EXPORT_SYMBOL_GPL(dm_cell_release);
211 * Sometimes we don't want the holder, just the additional bios.
213 static void __cell_release_no_holder(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
215 struct dm_bio_prison *prison = cell->prison;
217 hlist_del(&cell->list);
218 bio_list_merge(inmates, &cell->bios);
220 mempool_free(cell, prison->cell_pool);
223 void dm_cell_release_no_holder(struct dm_bio_prison_cell *cell, struct bio_list *inmates)
226 struct dm_bio_prison *prison = cell->prison;
228 spin_lock_irqsave(&prison->lock, flags);
229 __cell_release_no_holder(cell, inmates);
230 spin_unlock_irqrestore(&prison->lock, flags);
232 EXPORT_SYMBOL_GPL(dm_cell_release_no_holder);
234 void dm_cell_error(struct dm_bio_prison_cell *cell)
236 struct dm_bio_prison *prison = cell->prison;
237 struct bio_list bios;
241 bio_list_init(&bios);
243 spin_lock_irqsave(&prison->lock, flags);
244 __cell_release(cell, &bios);
245 spin_unlock_irqrestore(&prison->lock, flags);
247 while ((bio = bio_list_pop(&bios)))
250 EXPORT_SYMBOL_GPL(dm_cell_error);
252 /*----------------------------------------------------------------*/
254 #define DEFERRED_SET_SIZE 64
256 struct dm_deferred_entry {
257 struct dm_deferred_set *ds;
259 struct list_head work_items;
262 struct dm_deferred_set {
264 unsigned current_entry;
266 struct dm_deferred_entry entries[DEFERRED_SET_SIZE];
269 struct dm_deferred_set *dm_deferred_set_create(void)
272 struct dm_deferred_set *ds;
274 ds = kmalloc(sizeof(*ds), GFP_KERNEL);
278 spin_lock_init(&ds->lock);
279 ds->current_entry = 0;
281 for (i = 0; i < DEFERRED_SET_SIZE; i++) {
282 ds->entries[i].ds = ds;
283 ds->entries[i].count = 0;
284 INIT_LIST_HEAD(&ds->entries[i].work_items);
289 EXPORT_SYMBOL_GPL(dm_deferred_set_create);
291 void dm_deferred_set_destroy(struct dm_deferred_set *ds)
295 EXPORT_SYMBOL_GPL(dm_deferred_set_destroy);
297 struct dm_deferred_entry *dm_deferred_entry_inc(struct dm_deferred_set *ds)
300 struct dm_deferred_entry *entry;
302 spin_lock_irqsave(&ds->lock, flags);
303 entry = ds->entries + ds->current_entry;
305 spin_unlock_irqrestore(&ds->lock, flags);
309 EXPORT_SYMBOL_GPL(dm_deferred_entry_inc);
311 static unsigned ds_next(unsigned index)
313 return (index + 1) % DEFERRED_SET_SIZE;
316 static void __sweep(struct dm_deferred_set *ds, struct list_head *head)
318 while ((ds->sweeper != ds->current_entry) &&
319 !ds->entries[ds->sweeper].count) {
320 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
321 ds->sweeper = ds_next(ds->sweeper);
324 if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
325 list_splice_init(&ds->entries[ds->sweeper].work_items, head);
328 void dm_deferred_entry_dec(struct dm_deferred_entry *entry, struct list_head *head)
332 spin_lock_irqsave(&entry->ds->lock, flags);
333 BUG_ON(!entry->count);
335 __sweep(entry->ds, head);
336 spin_unlock_irqrestore(&entry->ds->lock, flags);
338 EXPORT_SYMBOL_GPL(dm_deferred_entry_dec);
341 * Returns 1 if deferred or 0 if no pending items to delay job.
343 int dm_deferred_set_add_work(struct dm_deferred_set *ds, struct list_head *work)
349 spin_lock_irqsave(&ds->lock, flags);
350 if ((ds->sweeper == ds->current_entry) &&
351 !ds->entries[ds->current_entry].count)
354 list_add(work, &ds->entries[ds->current_entry].work_items);
355 next_entry = ds_next(ds->current_entry);
356 if (!ds->entries[next_entry].count)
357 ds->current_entry = next_entry;
359 spin_unlock_irqrestore(&ds->lock, flags);
363 EXPORT_SYMBOL_GPL(dm_deferred_set_add_work);
365 /*----------------------------------------------------------------*/
367 static int __init dm_bio_prison_init(void)
369 _cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
376 static void __exit dm_bio_prison_exit(void)
378 kmem_cache_destroy(_cell_cache);
385 module_init(dm_bio_prison_init);
386 module_exit(dm_bio_prison_exit);
388 MODULE_DESCRIPTION(DM_NAME " bio prison");
389 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
390 MODULE_LICENSE("GPL");