xref: /linux/drivers/md/dm-bio-prison-v1.c (revision 9ad8d22f2f3fad7a366c9772362795ef6d6a2d51)
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[] __counted_by(num_locks);
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 /*
202  * @inmates must have been initialised prior to this call
203  */
204 static void __cell_release(struct rb_root *root,
205 			   struct dm_bio_prison_cell *cell,
206 			   struct bio_list *inmates)
207 {
208 	rb_erase(&cell->node, root);
209 
210 	if (inmates) {
211 		if (cell->holder)
212 			bio_list_add(inmates, cell->holder);
213 		bio_list_merge(inmates, &cell->bios);
214 	}
215 }
216 
217 void dm_cell_release(struct dm_bio_prison *prison,
218 		     struct dm_bio_prison_cell *cell,
219 		     struct bio_list *bios)
220 {
221 	unsigned l = lock_nr(&cell->key, prison->num_locks);
222 
223 	spin_lock_irq(&prison->regions[l].lock);
224 	__cell_release(&prison->regions[l].cell, cell, bios);
225 	spin_unlock_irq(&prison->regions[l].lock);
226 }
227 EXPORT_SYMBOL_GPL(dm_cell_release);
228 
229 /*
230  * Sometimes we don't want the holder, just the additional bios.
231  */
232 static void __cell_release_no_holder(struct rb_root *root,
233 				     struct dm_bio_prison_cell *cell,
234 				     struct bio_list *inmates)
235 {
236 	rb_erase(&cell->node, root);
237 	bio_list_merge(inmates, &cell->bios);
238 }
239 
240 void dm_cell_release_no_holder(struct dm_bio_prison *prison,
241 			       struct dm_bio_prison_cell *cell,
242 			       struct bio_list *inmates)
243 {
244 	unsigned l = lock_nr(&cell->key, prison->num_locks);
245 	unsigned long flags;
246 
247 	spin_lock_irqsave(&prison->regions[l].lock, flags);
248 	__cell_release_no_holder(&prison->regions[l].cell, cell, inmates);
249 	spin_unlock_irqrestore(&prison->regions[l].lock, flags);
250 }
251 EXPORT_SYMBOL_GPL(dm_cell_release_no_holder);
252 
253 void dm_cell_error(struct dm_bio_prison *prison,
254 		   struct dm_bio_prison_cell *cell, blk_status_t error)
255 {
256 	struct bio_list bios;
257 	struct bio *bio;
258 
259 	bio_list_init(&bios);
260 	dm_cell_release(prison, cell, &bios);
261 
262 	while ((bio = bio_list_pop(&bios))) {
263 		bio->bi_status = error;
264 		bio_endio(bio);
265 	}
266 }
267 EXPORT_SYMBOL_GPL(dm_cell_error);
268 
269 void dm_cell_visit_release(struct dm_bio_prison *prison,
270 			   void (*visit_fn)(void *, struct dm_bio_prison_cell *),
271 			   void *context,
272 			   struct dm_bio_prison_cell *cell)
273 {
274 	unsigned l = lock_nr(&cell->key, prison->num_locks);
275 	spin_lock_irq(&prison->regions[l].lock);
276 	visit_fn(context, cell);
277 	rb_erase(&cell->node, &prison->regions[l].cell);
278 	spin_unlock_irq(&prison->regions[l].lock);
279 }
280 EXPORT_SYMBOL_GPL(dm_cell_visit_release);
281 
282 /*----------------------------------------------------------------*/
283 
284 #define DEFERRED_SET_SIZE 64
285 
286 struct dm_deferred_entry {
287 	struct dm_deferred_set *ds;
288 	unsigned int count;
289 	struct list_head work_items;
290 };
291 
292 struct dm_deferred_set {
293 	spinlock_t lock;
294 	unsigned int current_entry;
295 	unsigned int sweeper;
296 	struct dm_deferred_entry entries[DEFERRED_SET_SIZE];
297 };
298 
299 struct dm_deferred_set *dm_deferred_set_create(void)
300 {
301 	int i;
302 	struct dm_deferred_set *ds;
303 
304 	ds = kmalloc(sizeof(*ds), GFP_KERNEL);
305 	if (!ds)
306 		return NULL;
307 
308 	spin_lock_init(&ds->lock);
309 	ds->current_entry = 0;
310 	ds->sweeper = 0;
311 	for (i = 0; i < DEFERRED_SET_SIZE; i++) {
312 		ds->entries[i].ds = ds;
313 		ds->entries[i].count = 0;
314 		INIT_LIST_HEAD(&ds->entries[i].work_items);
315 	}
316 
317 	return ds;
318 }
319 EXPORT_SYMBOL_GPL(dm_deferred_set_create);
320 
321 void dm_deferred_set_destroy(struct dm_deferred_set *ds)
322 {
323 	kfree(ds);
324 }
325 EXPORT_SYMBOL_GPL(dm_deferred_set_destroy);
326 
327 struct dm_deferred_entry *dm_deferred_entry_inc(struct dm_deferred_set *ds)
328 {
329 	unsigned long flags;
330 	struct dm_deferred_entry *entry;
331 
332 	spin_lock_irqsave(&ds->lock, flags);
333 	entry = ds->entries + ds->current_entry;
334 	entry->count++;
335 	spin_unlock_irqrestore(&ds->lock, flags);
336 
337 	return entry;
338 }
339 EXPORT_SYMBOL_GPL(dm_deferred_entry_inc);
340 
341 static unsigned int ds_next(unsigned int index)
342 {
343 	return (index + 1) % DEFERRED_SET_SIZE;
344 }
345 
346 static void __sweep(struct dm_deferred_set *ds, struct list_head *head)
347 {
348 	while ((ds->sweeper != ds->current_entry) &&
349 	       !ds->entries[ds->sweeper].count) {
350 		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
351 		ds->sweeper = ds_next(ds->sweeper);
352 	}
353 
354 	if ((ds->sweeper == ds->current_entry) && !ds->entries[ds->sweeper].count)
355 		list_splice_init(&ds->entries[ds->sweeper].work_items, head);
356 }
357 
358 void dm_deferred_entry_dec(struct dm_deferred_entry *entry, struct list_head *head)
359 {
360 	unsigned long flags;
361 
362 	spin_lock_irqsave(&entry->ds->lock, flags);
363 	BUG_ON(!entry->count);
364 	--entry->count;
365 	__sweep(entry->ds, head);
366 	spin_unlock_irqrestore(&entry->ds->lock, flags);
367 }
368 EXPORT_SYMBOL_GPL(dm_deferred_entry_dec);
369 
370 /*
371  * Returns 1 if deferred or 0 if no pending items to delay job.
372  */
373 int dm_deferred_set_add_work(struct dm_deferred_set *ds, struct list_head *work)
374 {
375 	int r = 1;
376 	unsigned int next_entry;
377 
378 	spin_lock_irq(&ds->lock);
379 	if ((ds->sweeper == ds->current_entry) &&
380 	    !ds->entries[ds->current_entry].count)
381 		r = 0;
382 	else {
383 		list_add(work, &ds->entries[ds->current_entry].work_items);
384 		next_entry = ds_next(ds->current_entry);
385 		if (!ds->entries[next_entry].count)
386 			ds->current_entry = next_entry;
387 	}
388 	spin_unlock_irq(&ds->lock);
389 
390 	return r;
391 }
392 EXPORT_SYMBOL_GPL(dm_deferred_set_add_work);
393 
394 /*----------------------------------------------------------------*/
395 
396 static int __init dm_bio_prison_init_v1(void)
397 {
398 	_cell_cache = KMEM_CACHE(dm_bio_prison_cell, 0);
399 	if (!_cell_cache)
400 		return -ENOMEM;
401 
402 	return 0;
403 }
404 
405 static void dm_bio_prison_exit_v1(void)
406 {
407 	kmem_cache_destroy(_cell_cache);
408 	_cell_cache = NULL;
409 }
410 
411 static int (*_inits[])(void) __initdata = {
412 	dm_bio_prison_init_v1,
413 	dm_bio_prison_init_v2,
414 };
415 
416 static void (*_exits[])(void) = {
417 	dm_bio_prison_exit_v1,
418 	dm_bio_prison_exit_v2,
419 };
420 
421 static int __init dm_bio_prison_init(void)
422 {
423 	const int count = ARRAY_SIZE(_inits);
424 
425 	int r, i;
426 
427 	for (i = 0; i < count; i++) {
428 		r = _inits[i]();
429 		if (r)
430 			goto bad;
431 	}
432 
433 	return 0;
434 
435 bad:
436 	while (i--)
437 		_exits[i]();
438 
439 	return r;
440 }
441 
442 static void __exit dm_bio_prison_exit(void)
443 {
444 	int i = ARRAY_SIZE(_exits);
445 
446 	while (i--)
447 		_exits[i]();
448 }
449 
450 /*
451  * module hooks
452  */
453 module_init(dm_bio_prison_init);
454 module_exit(dm_bio_prison_exit);
455 
456 MODULE_DESCRIPTION(DM_NAME " bio prison");
457 MODULE_AUTHOR("Joe Thornber <dm-devel@lists.linux.dev>");
458 MODULE_LICENSE("GPL");
459