1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/mm/mempool.c
4 *
5 * memory buffer pool support. Such pools are mostly used
6 * for guaranteed, deadlock-free memory allocations during
7 * extreme VM load.
8 *
9 * started by Ingo Molnar, Copyright (C) 2001
10 * debugging by David Rientjes, Copyright (C) 2015
11 */
12
13#include <linux/mm.h>
14#include <linux/slab.h>
15#include <linux/highmem.h>
16#include <linux/kasan.h>
17#include <linux/kmemleak.h>
18#include <linux/export.h>
19#include <linux/mempool.h>
20#include <linux/blkdev.h>
21#include <linux/writeback.h>
22#include "slab.h"
23
24#if defined(CONFIG_DEBUG_SLAB) || defined(CONFIG_SLUB_DEBUG_ON)
25static void poison_error(mempool_t *pool, void *element, size_t size,
26 size_t byte)
27{
28 const int nr = pool->curr_nr;
29 const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
30 const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
31 int i;
32
33 pr_err("BUG: mempool element poison mismatch\n");
34 pr_err("Mempool %p size %zu\n", pool, size);
35 pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
36 for (i = start; i < end; i++)
37 pr_cont("%x ", *(u8 *)(element + i));
38 pr_cont("%s\n", end < size ? "..." : "");
39 dump_stack();
40}
41
42static void __check_element(mempool_t *pool, void *element, size_t size)
43{
44 u8 *obj = element;
45 size_t i;
46
47 for (i = 0; i < size; i++) {
48 u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
49
50 if (obj[i] != exp) {
51 poison_error(pool, element, size, i);
52 return;
53 }
54 }
55 memset(obj, POISON_INUSE, size);
56}
57
58static void check_element(mempool_t *pool, void *element)
59{
60 /* Mempools backed by slab allocator */
61 if (pool->free == mempool_free_slab || pool->free == mempool_kfree)
62 __check_element(pool, element, ksize(element));
63
64 /* Mempools backed by page allocator */
65 if (pool->free == mempool_free_pages) {
66 int order = (int)(long)pool->pool_data;
67 void *addr = kmap_atomic((struct page *)element);
68
69 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
70 kunmap_atomic(addr);
71 }
72}
73
74static void __poison_element(void *element, size_t size)
75{
76 u8 *obj = element;
77
78 memset(obj, POISON_FREE, size - 1);
79 obj[size - 1] = POISON_END;
80}
81
82static void poison_element(mempool_t *pool, void *element)
83{
84 /* Mempools backed by slab allocator */
85 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
86 __poison_element(element, ksize(element));
87
88 /* Mempools backed by page allocator */
89 if (pool->alloc == mempool_alloc_pages) {
90 int order = (int)(long)pool->pool_data;
91 void *addr = kmap_atomic((struct page *)element);
92
93 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
94 kunmap_atomic(addr);
95 }
96}
97#else /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
98static inline void check_element(mempool_t *pool, void *element)
99{
100}
101static inline void poison_element(mempool_t *pool, void *element)
102{
103}
104#endif /* CONFIG_DEBUG_SLAB || CONFIG_SLUB_DEBUG_ON */
105
106static __always_inline void kasan_poison_element(mempool_t *pool, void *element)
107{
108 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
109 kasan_poison_kfree(element, _RET_IP_);
110 if (pool->alloc == mempool_alloc_pages)
111 kasan_free_pages(element, (unsigned long)pool->pool_data);
112}
113
114static void kasan_unpoison_element(mempool_t *pool, void *element)
115{
116 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
117 kasan_unpoison_slab(element);
118 if (pool->alloc == mempool_alloc_pages)
119 kasan_alloc_pages(element, (unsigned long)pool->pool_data);
120}
121
122static __always_inline void add_element(mempool_t *pool, void *element)
123{
124 BUG_ON(pool->curr_nr >= pool->min_nr);
125 poison_element(pool, element);
126 kasan_poison_element(pool, element);
127 pool->elements[pool->curr_nr++] = element;
128}
129
130static void *remove_element(mempool_t *pool)
131{
132 void *element = pool->elements[--pool->curr_nr];
133
134 BUG_ON(pool->curr_nr < 0);
135 kasan_unpoison_element(pool, element);
136 check_element(pool, element);
137 return element;
138}
139
140/**
141 * mempool_exit - exit a mempool initialized with mempool_init()
142 * @pool: pointer to the memory pool which was initialized with
143 * mempool_init().
144 *
145 * Free all reserved elements in @pool and @pool itself. This function
146 * only sleeps if the free_fn() function sleeps.
147 *
148 * May be called on a zeroed but uninitialized mempool (i.e. allocated with
149 * kzalloc()).
150 */
151void mempool_exit(mempool_t *pool)
152{
153 while (pool->curr_nr) {
154 void *element = remove_element(pool);
155 pool->free(element, pool->pool_data);
156 }
157 kfree(pool->elements);
158 pool->elements = NULL;
159}
160EXPORT_SYMBOL(mempool_exit);
161
162/**
163 * mempool_destroy - deallocate a memory pool
164 * @pool: pointer to the memory pool which was allocated via
165 * mempool_create().
166 *
167 * Free all reserved elements in @pool and @pool itself. This function
168 * only sleeps if the free_fn() function sleeps.
169 */
170void mempool_destroy(mempool_t *pool)
171{
172 if (unlikely(!pool))
173 return;
174
175 mempool_exit(pool);
176 kfree(pool);
177}
178EXPORT_SYMBOL(mempool_destroy);
179
180int mempool_init_node(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
181 mempool_free_t *free_fn, void *pool_data,
182 gfp_t gfp_mask, int node_id)
183{
184 spin_lock_init(&pool->lock);
185 pool->min_nr = min_nr;
186 pool->pool_data = pool_data;
187 pool->alloc = alloc_fn;
188 pool->free = free_fn;
189 init_waitqueue_head(&pool->wait);
190
191 pool->elements = kmalloc_array_node(min_nr, sizeof(void *),
192 gfp_mask, node_id);
193 if (!pool->elements)
194 return -ENOMEM;
195
196 /*
197 * First pre-allocate the guaranteed number of buffers.
198 */
199 while (pool->curr_nr < pool->min_nr) {
200 void *element;
201
202 element = pool->alloc(gfp_mask, pool->pool_data);
203 if (unlikely(!element)) {
204 mempool_exit(pool);
205 return -ENOMEM;
206 }
207 add_element(pool, element);
208 }
209
210 return 0;
211}
212EXPORT_SYMBOL(mempool_init_node);
213
214/**
215 * mempool_init - initialize a memory pool
216 * @pool: pointer to the memory pool that should be initialized
217 * @min_nr: the minimum number of elements guaranteed to be
218 * allocated for this pool.
219 * @alloc_fn: user-defined element-allocation function.
220 * @free_fn: user-defined element-freeing function.
221 * @pool_data: optional private data available to the user-defined functions.
222 *
223 * Like mempool_create(), but initializes the pool in (i.e. embedded in another
224 * structure).
225 *
226 * Return: %0 on success, negative error code otherwise.
227 */
228int mempool_init(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
229 mempool_free_t *free_fn, void *pool_data)
230{
231 return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
232 pool_data, GFP_KERNEL, NUMA_NO_NODE);
233
234}
235EXPORT_SYMBOL(mempool_init);
236
237/**
238 * mempool_create - create a memory pool
239 * @min_nr: the minimum number of elements guaranteed to be
240 * allocated for this pool.
241 * @alloc_fn: user-defined element-allocation function.
242 * @free_fn: user-defined element-freeing function.
243 * @pool_data: optional private data available to the user-defined functions.
244 *
245 * this function creates and allocates a guaranteed size, preallocated
246 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
247 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
248 * functions might sleep - as long as the mempool_alloc() function is not called
249 * from IRQ contexts.
250 *
251 * Return: pointer to the created memory pool object or %NULL on error.
252 */
253mempool_t *mempool_create(int min_nr, mempool_alloc_t *alloc_fn,
254 mempool_free_t *free_fn, void *pool_data)
255{
256 return mempool_create_node(min_nr,alloc_fn,free_fn, pool_data,
257 GFP_KERNEL, NUMA_NO_NODE);
258}
259EXPORT_SYMBOL(mempool_create);
260
261mempool_t *mempool_create_node(int min_nr, mempool_alloc_t *alloc_fn,
262 mempool_free_t *free_fn, void *pool_data,
263 gfp_t gfp_mask, int node_id)
264{
265 mempool_t *pool;
266
267 pool = kzalloc_node(sizeof(*pool), gfp_mask, node_id);
268 if (!pool)
269 return NULL;
270
271 if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
272 gfp_mask, node_id)) {
273 kfree(pool);
274 return NULL;
275 }
276
277 return pool;
278}
279EXPORT_SYMBOL(mempool_create_node);
280
281/**
282 * mempool_resize - resize an existing memory pool
283 * @pool: pointer to the memory pool which was allocated via
284 * mempool_create().
285 * @new_min_nr: the new minimum number of elements guaranteed to be
286 * allocated for this pool.
287 *
288 * This function shrinks/grows the pool. In the case of growing,
289 * it cannot be guaranteed that the pool will be grown to the new
290 * size immediately, but new mempool_free() calls will refill it.
291 * This function may sleep.
292 *
293 * Note, the caller must guarantee that no mempool_destroy is called
294 * while this function is running. mempool_alloc() & mempool_free()
295 * might be called (eg. from IRQ contexts) while this function executes.
296 *
297 * Return: %0 on success, negative error code otherwise.
298 */
299int mempool_resize(mempool_t *pool, int new_min_nr)
300{
301 void *element;
302 void **new_elements;
303 unsigned long flags;
304
305 BUG_ON(new_min_nr <= 0);
306 might_sleep();
307
308 spin_lock_irqsave(&pool->lock, flags);
309 if (new_min_nr <= pool->min_nr) {
310 while (new_min_nr < pool->curr_nr) {
311 element = remove_element(pool);
312 spin_unlock_irqrestore(&pool->lock, flags);
313 pool->free(element, pool->pool_data);
314 spin_lock_irqsave(&pool->lock, flags);
315 }
316 pool->min_nr = new_min_nr;
317 goto out_unlock;
318 }
319 spin_unlock_irqrestore(&pool->lock, flags);
320
321 /* Grow the pool */
322 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
323 GFP_KERNEL);
324 if (!new_elements)
325 return -ENOMEM;
326
327 spin_lock_irqsave(&pool->lock, flags);
328 if (unlikely(new_min_nr <= pool->min_nr)) {
329 /* Raced, other resize will do our work */
330 spin_unlock_irqrestore(&pool->lock, flags);
331 kfree(new_elements);
332 goto out;
333 }
334 memcpy(new_elements, pool->elements,
335 pool->curr_nr * sizeof(*new_elements));
336 kfree(pool->elements);
337 pool->elements = new_elements;
338 pool->min_nr = new_min_nr;
339
340 while (pool->curr_nr < pool->min_nr) {
341 spin_unlock_irqrestore(&pool->lock, flags);
342 element = pool->alloc(GFP_KERNEL, pool->pool_data);
343 if (!element)
344 goto out;
345 spin_lock_irqsave(&pool->lock, flags);
346 if (pool->curr_nr < pool->min_nr) {
347 add_element(pool, element);
348 } else {
349 spin_unlock_irqrestore(&pool->lock, flags);
350 pool->free(element, pool->pool_data); /* Raced */
351 goto out;
352 }
353 }
354out_unlock:
355 spin_unlock_irqrestore(&pool->lock, flags);
356out:
357 return 0;
358}
359EXPORT_SYMBOL(mempool_resize);
360
361/**
362 * mempool_alloc - allocate an element from a specific memory pool
363 * @pool: pointer to the memory pool which was allocated via
364 * mempool_create().
365 * @gfp_mask: the usual allocation bitmask.
366 *
367 * this function only sleeps if the alloc_fn() function sleeps or
368 * returns NULL. Note that due to preallocation, this function
369 * *never* fails when called from process contexts. (it might
370 * fail if called from an IRQ context.)
371 * Note: using __GFP_ZERO is not supported.
372 *
373 * Return: pointer to the allocated element or %NULL on error.
374 */
375void *mempool_alloc(mempool_t *pool, gfp_t gfp_mask)
376{
377 void *element;
378 unsigned long flags;
379 wait_queue_entry_t wait;
380 gfp_t gfp_temp;
381
382 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
383 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
384
385 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
386 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
387 gfp_mask |= __GFP_NOWARN; /* failures are OK */
388
389 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
390
391repeat_alloc:
392
393 element = pool->alloc(gfp_temp, pool->pool_data);
394 if (likely(element != NULL))
395 return element;
396
397 spin_lock_irqsave(&pool->lock, flags);
398 if (likely(pool->curr_nr)) {
399 element = remove_element(pool);
400 spin_unlock_irqrestore(&pool->lock, flags);
401 /* paired with rmb in mempool_free(), read comment there */
402 smp_wmb();
403 /*
404 * Update the allocation stack trace as this is more useful
405 * for debugging.
406 */
407 kmemleak_update_trace(element);
408 return element;
409 }
410
411 /*
412 * We use gfp mask w/o direct reclaim or IO for the first round. If
413 * alloc failed with that and @pool was empty, retry immediately.
414 */
415 if (gfp_temp != gfp_mask) {
416 spin_unlock_irqrestore(&pool->lock, flags);
417 gfp_temp = gfp_mask;
418 goto repeat_alloc;
419 }
420
421 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
422 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
423 spin_unlock_irqrestore(&pool->lock, flags);
424 return NULL;
425 }
426
427 /* Let's wait for someone else to return an element to @pool */
428 init_wait(&wait);
429 prepare_to_wait(&pool->wait, &wait, TASK_UNINTERRUPTIBLE);
430
431 spin_unlock_irqrestore(&pool->lock, flags);
432
433 /*
434 * FIXME: this should be io_schedule(). The timeout is there as a
435 * workaround for some DM problems in 2.6.18.
436 */
437 io_schedule_timeout(5*HZ);
438
439 finish_wait(&pool->wait, &wait);
440 goto repeat_alloc;
441}
442EXPORT_SYMBOL(mempool_alloc);
443
444/**
445 * mempool_free - return an element to the pool.
446 * @element: pool element pointer.
447 * @pool: pointer to the memory pool which was allocated via
448 * mempool_create().
449 *
450 * this function only sleeps if the free_fn() function sleeps.
451 */
452void mempool_free(void *element, mempool_t *pool)
453{
454 unsigned long flags;
455
456 if (unlikely(element == NULL))
457 return;
458
459 /*
460 * Paired with the wmb in mempool_alloc(). The preceding read is
461 * for @element and the following @pool->curr_nr. This ensures
462 * that the visible value of @pool->curr_nr is from after the
463 * allocation of @element. This is necessary for fringe cases
464 * where @element was passed to this task without going through
465 * barriers.
466 *
467 * For example, assume @p is %NULL at the beginning and one task
468 * performs "p = mempool_alloc(...);" while another task is doing
469 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
470 * may end up using curr_nr value which is from before allocation
471 * of @p without the following rmb.
472 */
473 smp_rmb();
474
475 /*
476 * For correctness, we need a test which is guaranteed to trigger
477 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
478 * without locking achieves that and refilling as soon as possible
479 * is desirable.
480 *
481 * Because curr_nr visible here is always a value after the
482 * allocation of @element, any task which decremented curr_nr below
483 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
484 * incremented to min_nr afterwards. If curr_nr gets incremented
485 * to min_nr after the allocation of @element, the elements
486 * allocated after that are subject to the same guarantee.
487 *
488 * Waiters happen iff curr_nr is 0 and the above guarantee also
489 * ensures that there will be frees which return elements to the
490 * pool waking up the waiters.
491 */
492 if (unlikely(pool->curr_nr < pool->min_nr)) {
493 spin_lock_irqsave(&pool->lock, flags);
494 if (likely(pool->curr_nr < pool->min_nr)) {
495 add_element(pool, element);
496 spin_unlock_irqrestore(&pool->lock, flags);
497 wake_up(&pool->wait);
498 return;
499 }
500 spin_unlock_irqrestore(&pool->lock, flags);
501 }
502 pool->free(element, pool->pool_data);
503}
504EXPORT_SYMBOL(mempool_free);
505
506/*
507 * A commonly used alloc and free fn.
508 */
509void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
510{
511 struct kmem_cache *mem = pool_data;
512 VM_BUG_ON(mem->ctor);
513 return kmem_cache_alloc(mem, gfp_mask);
514}
515EXPORT_SYMBOL(mempool_alloc_slab);
516
517void mempool_free_slab(void *element, void *pool_data)
518{
519 struct kmem_cache *mem = pool_data;
520 kmem_cache_free(mem, element);
521}
522EXPORT_SYMBOL(mempool_free_slab);
523
524/*
525 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
526 * specified by pool_data
527 */
528void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
529{
530 size_t size = (size_t)pool_data;
531 return kmalloc(size, gfp_mask);
532}
533EXPORT_SYMBOL(mempool_kmalloc);
534
535void mempool_kfree(void *element, void *pool_data)
536{
537 kfree(element);
538}
539EXPORT_SYMBOL(mempool_kfree);
540
541/*
542 * A simple mempool-backed page allocator that allocates pages
543 * of the order specified by pool_data.
544 */
545void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
546{
547 int order = (int)(long)pool_data;
548 return alloc_pages(gfp_mask, order);
549}
550EXPORT_SYMBOL(mempool_alloc_pages);
551
552void mempool_free_pages(void *element, void *pool_data)
553{
554 int order = (int)(long)pool_data;
555 __free_pages(element, order);
556}
557EXPORT_SYMBOL(mempool_free_pages);
558