b04d1f904d0763c43870e38c857fcd25baad4b7e
[kvmfornfv.git] / kernel / drivers / md / dm-cache-policy-cleaner.c
1 /*
2  * Copyright (C) 2012 Red Hat. All rights reserved.
3  *
4  * writeback cache policy supporting flushing out dirty cache blocks.
5  *
6  * This file is released under the GPL.
7  */
8
9 #include "dm-cache-policy.h"
10 #include "dm.h"
11
12 #include <linux/hash.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/vmalloc.h>
16
17 /*----------------------------------------------------------------*/
18
19 #define DM_MSG_PREFIX "cache cleaner"
20
21 /* Cache entry struct. */
22 struct wb_cache_entry {
23         struct list_head list;
24         struct hlist_node hlist;
25
26         dm_oblock_t oblock;
27         dm_cblock_t cblock;
28         bool dirty:1;
29         bool pending:1;
30 };
31
32 struct hash {
33         struct hlist_head *table;
34         dm_block_t hash_bits;
35         unsigned nr_buckets;
36 };
37
38 struct policy {
39         struct dm_cache_policy policy;
40         spinlock_t lock;
41
42         struct list_head free;
43         struct list_head clean;
44         struct list_head clean_pending;
45         struct list_head dirty;
46
47         /*
48          * We know exactly how many cblocks will be needed,
49          * so we can allocate them up front.
50          */
51         dm_cblock_t cache_size, nr_cblocks_allocated;
52         struct wb_cache_entry *cblocks;
53         struct hash chash;
54 };
55
56 /*----------------------------------------------------------------------------*/
57
58 /*
59  * Low-level functions.
60  */
61 static unsigned next_power(unsigned n, unsigned min)
62 {
63         return roundup_pow_of_two(max(n, min));
64 }
65
66 static struct policy *to_policy(struct dm_cache_policy *p)
67 {
68         return container_of(p, struct policy, policy);
69 }
70
71 static struct list_head *list_pop(struct list_head *q)
72 {
73         struct list_head *r = q->next;
74
75         list_del(r);
76
77         return r;
78 }
79
80 /*----------------------------------------------------------------------------*/
81
82 /* Allocate/free various resources. */
83 static int alloc_hash(struct hash *hash, unsigned elts)
84 {
85         hash->nr_buckets = next_power(elts >> 4, 16);
86         hash->hash_bits = ffs(hash->nr_buckets) - 1;
87         hash->table = vzalloc(sizeof(*hash->table) * hash->nr_buckets);
88
89         return hash->table ? 0 : -ENOMEM;
90 }
91
92 static void free_hash(struct hash *hash)
93 {
94         vfree(hash->table);
95 }
96
97 static int alloc_cache_blocks_with_hash(struct policy *p, dm_cblock_t cache_size)
98 {
99         int r = -ENOMEM;
100
101         p->cblocks = vzalloc(sizeof(*p->cblocks) * from_cblock(cache_size));
102         if (p->cblocks) {
103                 unsigned u = from_cblock(cache_size);
104
105                 while (u--)
106                         list_add(&p->cblocks[u].list, &p->free);
107
108                 p->nr_cblocks_allocated = 0;
109
110                 /* Cache entries hash. */
111                 r = alloc_hash(&p->chash, from_cblock(cache_size));
112                 if (r)
113                         vfree(p->cblocks);
114         }
115
116         return r;
117 }
118
119 static void free_cache_blocks_and_hash(struct policy *p)
120 {
121         free_hash(&p->chash);
122         vfree(p->cblocks);
123 }
124
125 static struct wb_cache_entry *alloc_cache_entry(struct policy *p)
126 {
127         struct wb_cache_entry *e;
128
129         BUG_ON(from_cblock(p->nr_cblocks_allocated) >= from_cblock(p->cache_size));
130
131         e = list_entry(list_pop(&p->free), struct wb_cache_entry, list);
132         p->nr_cblocks_allocated = to_cblock(from_cblock(p->nr_cblocks_allocated) + 1);
133
134         return e;
135 }
136
137 /*----------------------------------------------------------------------------*/
138
139 /* Hash functions (lookup, insert, remove). */
140 static struct wb_cache_entry *lookup_cache_entry(struct policy *p, dm_oblock_t oblock)
141 {
142         struct hash *hash = &p->chash;
143         unsigned h = hash_64(from_oblock(oblock), hash->hash_bits);
144         struct wb_cache_entry *cur;
145         struct hlist_head *bucket = &hash->table[h];
146
147         hlist_for_each_entry(cur, bucket, hlist) {
148                 if (cur->oblock == oblock) {
149                         /* Move upfront bucket for faster access. */
150                         hlist_del(&cur->hlist);
151                         hlist_add_head(&cur->hlist, bucket);
152                         return cur;
153                 }
154         }
155
156         return NULL;
157 }
158
159 static void insert_cache_hash_entry(struct policy *p, struct wb_cache_entry *e)
160 {
161         unsigned h = hash_64(from_oblock(e->oblock), p->chash.hash_bits);
162
163         hlist_add_head(&e->hlist, &p->chash.table[h]);
164 }
165
166 static void remove_cache_hash_entry(struct wb_cache_entry *e)
167 {
168         hlist_del(&e->hlist);
169 }
170
171 /* Public interface (see dm-cache-policy.h */
172 static int wb_map(struct dm_cache_policy *pe, dm_oblock_t oblock,
173                   bool can_block, bool can_migrate, bool discarded_oblock,
174                   struct bio *bio, struct policy_result *result)
175 {
176         struct policy *p = to_policy(pe);
177         struct wb_cache_entry *e;
178         unsigned long flags;
179
180         result->op = POLICY_MISS;
181
182         if (can_block)
183                 spin_lock_irqsave(&p->lock, flags);
184
185         else if (!spin_trylock_irqsave(&p->lock, flags))
186                 return -EWOULDBLOCK;
187
188         e = lookup_cache_entry(p, oblock);
189         if (e) {
190                 result->op = POLICY_HIT;
191                 result->cblock = e->cblock;
192
193         }
194
195         spin_unlock_irqrestore(&p->lock, flags);
196
197         return 0;
198 }
199
200 static int wb_lookup(struct dm_cache_policy *pe, dm_oblock_t oblock, dm_cblock_t *cblock)
201 {
202         int r;
203         struct policy *p = to_policy(pe);
204         struct wb_cache_entry *e;
205         unsigned long flags;
206
207         if (!spin_trylock_irqsave(&p->lock, flags))
208                 return -EWOULDBLOCK;
209
210         e = lookup_cache_entry(p, oblock);
211         if (e) {
212                 *cblock = e->cblock;
213                 r = 0;
214
215         } else
216                 r = -ENOENT;
217
218         spin_unlock_irqrestore(&p->lock, flags);
219
220         return r;
221 }
222
223 static void __set_clear_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock, bool set)
224 {
225         struct policy *p = to_policy(pe);
226         struct wb_cache_entry *e;
227
228         e = lookup_cache_entry(p, oblock);
229         BUG_ON(!e);
230
231         if (set) {
232                 if (!e->dirty) {
233                         e->dirty = true;
234                         list_move(&e->list, &p->dirty);
235                 }
236
237         } else {
238                 if (e->dirty) {
239                         e->pending = false;
240                         e->dirty = false;
241                         list_move(&e->list, &p->clean);
242                 }
243         }
244 }
245
246 static void wb_set_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock)
247 {
248         struct policy *p = to_policy(pe);
249         unsigned long flags;
250
251         spin_lock_irqsave(&p->lock, flags);
252         __set_clear_dirty(pe, oblock, true);
253         spin_unlock_irqrestore(&p->lock, flags);
254 }
255
256 static void wb_clear_dirty(struct dm_cache_policy *pe, dm_oblock_t oblock)
257 {
258         struct policy *p = to_policy(pe);
259         unsigned long flags;
260
261         spin_lock_irqsave(&p->lock, flags);
262         __set_clear_dirty(pe, oblock, false);
263         spin_unlock_irqrestore(&p->lock, flags);
264 }
265
266 static void add_cache_entry(struct policy *p, struct wb_cache_entry *e)
267 {
268         insert_cache_hash_entry(p, e);
269         if (e->dirty)
270                 list_add(&e->list, &p->dirty);
271         else
272                 list_add(&e->list, &p->clean);
273 }
274
275 static int wb_load_mapping(struct dm_cache_policy *pe,
276                            dm_oblock_t oblock, dm_cblock_t cblock,
277                            uint32_t hint, bool hint_valid)
278 {
279         int r;
280         struct policy *p = to_policy(pe);
281         struct wb_cache_entry *e = alloc_cache_entry(p);
282
283         if (e) {
284                 e->cblock = cblock;
285                 e->oblock = oblock;
286                 e->dirty = false; /* blocks default to clean */
287                 add_cache_entry(p, e);
288                 r = 0;
289
290         } else
291                 r = -ENOMEM;
292
293         return r;
294 }
295
296 static void wb_destroy(struct dm_cache_policy *pe)
297 {
298         struct policy *p = to_policy(pe);
299
300         free_cache_blocks_and_hash(p);
301         kfree(p);
302 }
303
304 static struct wb_cache_entry *__wb_force_remove_mapping(struct policy *p, dm_oblock_t oblock)
305 {
306         struct wb_cache_entry *r = lookup_cache_entry(p, oblock);
307
308         BUG_ON(!r);
309
310         remove_cache_hash_entry(r);
311         list_del(&r->list);
312
313         return r;
314 }
315
316 static void wb_remove_mapping(struct dm_cache_policy *pe, dm_oblock_t oblock)
317 {
318         struct policy *p = to_policy(pe);
319         struct wb_cache_entry *e;
320         unsigned long flags;
321
322         spin_lock_irqsave(&p->lock, flags);
323         e = __wb_force_remove_mapping(p, oblock);
324         list_add_tail(&e->list, &p->free);
325         BUG_ON(!from_cblock(p->nr_cblocks_allocated));
326         p->nr_cblocks_allocated = to_cblock(from_cblock(p->nr_cblocks_allocated) - 1);
327         spin_unlock_irqrestore(&p->lock, flags);
328 }
329
330 static void wb_force_mapping(struct dm_cache_policy *pe,
331                                 dm_oblock_t current_oblock, dm_oblock_t oblock)
332 {
333         struct policy *p = to_policy(pe);
334         struct wb_cache_entry *e;
335         unsigned long flags;
336
337         spin_lock_irqsave(&p->lock, flags);
338         e = __wb_force_remove_mapping(p, current_oblock);
339         e->oblock = oblock;
340         add_cache_entry(p, e);
341         spin_unlock_irqrestore(&p->lock, flags);
342 }
343
344 static struct wb_cache_entry *get_next_dirty_entry(struct policy *p)
345 {
346         struct list_head *l;
347         struct wb_cache_entry *r;
348
349         if (list_empty(&p->dirty))
350                 return NULL;
351
352         l = list_pop(&p->dirty);
353         r = container_of(l, struct wb_cache_entry, list);
354         list_add(l, &p->clean_pending);
355
356         return r;
357 }
358
359 static int wb_writeback_work(struct dm_cache_policy *pe,
360                              dm_oblock_t *oblock,
361                              dm_cblock_t *cblock)
362 {
363         int r = -ENOENT;
364         struct policy *p = to_policy(pe);
365         struct wb_cache_entry *e;
366         unsigned long flags;
367
368         spin_lock_irqsave(&p->lock, flags);
369
370         e = get_next_dirty_entry(p);
371         if (e) {
372                 *oblock = e->oblock;
373                 *cblock = e->cblock;
374                 r = 0;
375         }
376
377         spin_unlock_irqrestore(&p->lock, flags);
378
379         return r;
380 }
381
382 static dm_cblock_t wb_residency(struct dm_cache_policy *pe)
383 {
384         return to_policy(pe)->nr_cblocks_allocated;
385 }
386
387 /* Init the policy plugin interface function pointers. */
388 static void init_policy_functions(struct policy *p)
389 {
390         p->policy.destroy = wb_destroy;
391         p->policy.map = wb_map;
392         p->policy.lookup = wb_lookup;
393         p->policy.set_dirty = wb_set_dirty;
394         p->policy.clear_dirty = wb_clear_dirty;
395         p->policy.load_mapping = wb_load_mapping;
396         p->policy.walk_mappings = NULL;
397         p->policy.remove_mapping = wb_remove_mapping;
398         p->policy.writeback_work = wb_writeback_work;
399         p->policy.force_mapping = wb_force_mapping;
400         p->policy.residency = wb_residency;
401         p->policy.tick = NULL;
402 }
403
404 static struct dm_cache_policy *wb_create(dm_cblock_t cache_size,
405                                          sector_t origin_size,
406                                          sector_t cache_block_size)
407 {
408         int r;
409         struct policy *p = kzalloc(sizeof(*p), GFP_KERNEL);
410
411         if (!p)
412                 return NULL;
413
414         init_policy_functions(p);
415         INIT_LIST_HEAD(&p->free);
416         INIT_LIST_HEAD(&p->clean);
417         INIT_LIST_HEAD(&p->clean_pending);
418         INIT_LIST_HEAD(&p->dirty);
419
420         p->cache_size = cache_size;
421         spin_lock_init(&p->lock);
422
423         /* Allocate cache entry structs and add them to free list. */
424         r = alloc_cache_blocks_with_hash(p, cache_size);
425         if (!r)
426                 return &p->policy;
427
428         kfree(p);
429
430         return NULL;
431 }
432 /*----------------------------------------------------------------------------*/
433
434 static struct dm_cache_policy_type wb_policy_type = {
435         .name = "cleaner",
436         .version = {1, 0, 0},
437         .hint_size = 0,
438         .owner = THIS_MODULE,
439         .create = wb_create
440 };
441
442 static int __init wb_init(void)
443 {
444         int r = dm_cache_policy_register(&wb_policy_type);
445
446         if (r < 0)
447                 DMERR("register failed %d", r);
448         else
449                 DMINFO("version %u.%u.%u loaded",
450                        wb_policy_type.version[0],
451                        wb_policy_type.version[1],
452                        wb_policy_type.version[2]);
453
454         return r;
455 }
456
457 static void __exit wb_exit(void)
458 {
459         dm_cache_policy_unregister(&wb_policy_type);
460 }
461
462 module_init(wb_init);
463 module_exit(wb_exit);
464
465 MODULE_AUTHOR("Heinz Mauelshagen <dm-devel@redhat.com>");
466 MODULE_LICENSE("GPL");
467 MODULE_DESCRIPTION("cleaner cache policy");