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/writeback.h>
21#include "slab.h"
22
23#ifdef CONFIG_SLUB_DEBUG_ON
24static void poison_error(mempool_t *pool, void *element, size_t size,
25 size_t byte)
26{
27 const int nr = pool->curr_nr;
28 const int start = max_t(int, byte - (BITS_PER_LONG / 8), 0);
29 const int end = min_t(int, byte + (BITS_PER_LONG / 8), size);
30 int i;
31
32 pr_err("BUG: mempool element poison mismatch\n");
33 pr_err("Mempool %p size %zu\n", pool, size);
34 pr_err(" nr=%d @ %p: %s0x", nr, element, start > 0 ? "... " : "");
35 for (i = start; i < end; i++)
36 pr_cont("%x ", *(u8 *)(element + i));
37 pr_cont("%s\n", end < size ? "..." : "");
38 dump_stack();
39}
40
41static void __check_element(mempool_t *pool, void *element, size_t size)
42{
43 u8 *obj = element;
44 size_t i;
45
46 for (i = 0; i < size; i++) {
47 u8 exp = (i < size - 1) ? POISON_FREE : POISON_END;
48
49 if (obj[i] != exp) {
50 poison_error(pool, element, size, i);
51 return;
52 }
53 }
54 memset(obj, POISON_INUSE, size);
55}
56
57static void check_element(mempool_t *pool, void *element)
58{
59 /* Skip checking: KASAN might save its metadata in the element. */
60 if (kasan_enabled())
61 return;
62
63 /* Mempools backed by slab allocator */
64 if (pool->free == mempool_kfree) {
65 __check_element(pool, element, (size_t)pool->pool_data);
66 } else if (pool->free == mempool_free_slab) {
67 __check_element(pool, element, kmem_cache_size(pool->pool_data));
68 } else if (pool->free == mempool_free_pages) {
69 /* Mempools backed by page allocator */
70 int order = (int)(long)pool->pool_data;
71 void *addr = kmap_local_page((struct page *)element);
72
73 __check_element(pool, addr, 1UL << (PAGE_SHIFT + order));
74 kunmap_local(addr);
75 }
76}
77
78static void __poison_element(void *element, size_t size)
79{
80 u8 *obj = element;
81
82 memset(obj, POISON_FREE, size - 1);
83 obj[size - 1] = POISON_END;
84}
85
86static void poison_element(mempool_t *pool, void *element)
87{
88 /* Skip poisoning: KASAN might save its metadata in the element. */
89 if (kasan_enabled())
90 return;
91
92 /* Mempools backed by slab allocator */
93 if (pool->alloc == mempool_kmalloc) {
94 __poison_element(element, (size_t)pool->pool_data);
95 } else if (pool->alloc == mempool_alloc_slab) {
96 __poison_element(element, kmem_cache_size(pool->pool_data));
97 } else if (pool->alloc == mempool_alloc_pages) {
98 /* Mempools backed by page allocator */
99 int order = (int)(long)pool->pool_data;
100 void *addr = kmap_local_page((struct page *)element);
101
102 __poison_element(addr, 1UL << (PAGE_SHIFT + order));
103 kunmap_local(addr);
104 }
105}
106#else /* CONFIG_SLUB_DEBUG_ON */
107static inline void check_element(mempool_t *pool, void *element)
108{
109}
110static inline void poison_element(mempool_t *pool, void *element)
111{
112}
113#endif /* CONFIG_SLUB_DEBUG_ON */
114
115static __always_inline bool kasan_poison_element(mempool_t *pool, void *element)
116{
117 if (pool->alloc == mempool_alloc_slab || pool->alloc == mempool_kmalloc)
118 return kasan_mempool_poison_object(ptr: element);
119 else if (pool->alloc == mempool_alloc_pages)
120 return kasan_mempool_poison_pages(page: element,
121 order: (unsigned long)pool->pool_data);
122 return true;
123}
124
125static void kasan_unpoison_element(mempool_t *pool, void *element)
126{
127 if (pool->alloc == mempool_kmalloc)
128 kasan_mempool_unpoison_object(ptr: element, size: (size_t)pool->pool_data);
129 else if (pool->alloc == mempool_alloc_slab)
130 kasan_mempool_unpoison_object(ptr: element,
131 size: kmem_cache_size(s: pool->pool_data));
132 else if (pool->alloc == mempool_alloc_pages)
133 kasan_mempool_unpoison_pages(page: element,
134 order: (unsigned long)pool->pool_data);
135}
136
137static __always_inline void add_element(mempool_t *pool, void *element)
138{
139 BUG_ON(pool->min_nr != 0 && pool->curr_nr >= pool->min_nr);
140 poison_element(pool, element);
141 if (kasan_poison_element(pool, element))
142 pool->elements[pool->curr_nr++] = element;
143}
144
145static void *remove_element(mempool_t *pool)
146{
147 void *element = pool->elements[--pool->curr_nr];
148
149 BUG_ON(pool->curr_nr < 0);
150 kasan_unpoison_element(pool, element);
151 check_element(pool, element);
152 return element;
153}
154
155/**
156 * mempool_exit - exit a mempool initialized with mempool_init()
157 * @pool: pointer to the memory pool which was initialized with
158 * mempool_init().
159 *
160 * Free all reserved elements in @pool and @pool itself. This function
161 * only sleeps if the free_fn() function sleeps.
162 *
163 * May be called on a zeroed but uninitialized mempool (i.e. allocated with
164 * kzalloc()).
165 */
166void mempool_exit(mempool_t *pool)
167{
168 while (pool->curr_nr) {
169 void *element = remove_element(pool);
170 pool->free(element, pool->pool_data);
171 }
172 kfree(objp: pool->elements);
173 pool->elements = NULL;
174}
175EXPORT_SYMBOL(mempool_exit);
176
177/**
178 * mempool_destroy - deallocate a memory pool
179 * @pool: pointer to the memory pool which was allocated via
180 * mempool_create().
181 *
182 * Free all reserved elements in @pool and @pool itself. This function
183 * only sleeps if the free_fn() function sleeps.
184 */
185void mempool_destroy(mempool_t *pool)
186{
187 if (unlikely(!pool))
188 return;
189
190 mempool_exit(pool);
191 kfree(objp: pool);
192}
193EXPORT_SYMBOL(mempool_destroy);
194
195int mempool_init_node(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
196 mempool_free_t *free_fn, void *pool_data,
197 gfp_t gfp_mask, int node_id)
198{
199 spin_lock_init(&pool->lock);
200 pool->min_nr = min_nr;
201 pool->pool_data = pool_data;
202 pool->alloc = alloc_fn;
203 pool->free = free_fn;
204 init_waitqueue_head(&pool->wait);
205 /*
206 * max() used here to ensure storage for at least 1 element to support
207 * zero minimum pool
208 */
209 pool->elements = kmalloc_array_node(max(1, min_nr), sizeof(void *),
210 gfp_mask, node_id);
211 if (!pool->elements)
212 return -ENOMEM;
213
214 /*
215 * First pre-allocate the guaranteed number of buffers,
216 * also pre-allocate 1 element for zero minimum pool.
217 */
218 while (pool->curr_nr < max(1, pool->min_nr)) {
219 void *element;
220
221 element = pool->alloc(gfp_mask, pool->pool_data);
222 if (unlikely(!element)) {
223 mempool_exit(pool);
224 return -ENOMEM;
225 }
226 add_element(pool, element);
227 }
228
229 return 0;
230}
231EXPORT_SYMBOL(mempool_init_node);
232
233/**
234 * mempool_init - initialize a memory pool
235 * @pool: pointer to the memory pool that should be initialized
236 * @min_nr: the minimum number of elements guaranteed to be
237 * allocated for this pool.
238 * @alloc_fn: user-defined element-allocation function.
239 * @free_fn: user-defined element-freeing function.
240 * @pool_data: optional private data available to the user-defined functions.
241 *
242 * Like mempool_create(), but initializes the pool in (i.e. embedded in another
243 * structure).
244 *
245 * Return: %0 on success, negative error code otherwise.
246 */
247int mempool_init_noprof(mempool_t *pool, int min_nr, mempool_alloc_t *alloc_fn,
248 mempool_free_t *free_fn, void *pool_data)
249{
250 return mempool_init_node(pool, min_nr, alloc_fn, free_fn,
251 pool_data, GFP_KERNEL, NUMA_NO_NODE);
252
253}
254EXPORT_SYMBOL(mempool_init_noprof);
255
256/**
257 * mempool_create_node - create a memory pool
258 * @min_nr: the minimum number of elements guaranteed to be
259 * allocated for this pool.
260 * @alloc_fn: user-defined element-allocation function.
261 * @free_fn: user-defined element-freeing function.
262 * @pool_data: optional private data available to the user-defined functions.
263 * @gfp_mask: memory allocation flags
264 * @node_id: numa node to allocate on
265 *
266 * this function creates and allocates a guaranteed size, preallocated
267 * memory pool. The pool can be used from the mempool_alloc() and mempool_free()
268 * functions. This function might sleep. Both the alloc_fn() and the free_fn()
269 * functions might sleep - as long as the mempool_alloc() function is not called
270 * from IRQ contexts.
271 *
272 * Return: pointer to the created memory pool object or %NULL on error.
273 */
274mempool_t *mempool_create_node_noprof(int min_nr, mempool_alloc_t *alloc_fn,
275 mempool_free_t *free_fn, void *pool_data,
276 gfp_t gfp_mask, int node_id)
277{
278 mempool_t *pool;
279
280 pool = kmalloc_node_noprof(size: sizeof(*pool), flags: gfp_mask | __GFP_ZERO, node: node_id);
281 if (!pool)
282 return NULL;
283
284 if (mempool_init_node(pool, min_nr, alloc_fn, free_fn, pool_data,
285 gfp_mask, node_id)) {
286 kfree(objp: pool);
287 return NULL;
288 }
289
290 return pool;
291}
292EXPORT_SYMBOL(mempool_create_node_noprof);
293
294/**
295 * mempool_resize - resize an existing memory pool
296 * @pool: pointer to the memory pool which was allocated via
297 * mempool_create().
298 * @new_min_nr: the new minimum number of elements guaranteed to be
299 * allocated for this pool.
300 *
301 * This function shrinks/grows the pool. In the case of growing,
302 * it cannot be guaranteed that the pool will be grown to the new
303 * size immediately, but new mempool_free() calls will refill it.
304 * This function may sleep.
305 *
306 * Note, the caller must guarantee that no mempool_destroy is called
307 * while this function is running. mempool_alloc() & mempool_free()
308 * might be called (eg. from IRQ contexts) while this function executes.
309 *
310 * Return: %0 on success, negative error code otherwise.
311 */
312int mempool_resize(mempool_t *pool, int new_min_nr)
313{
314 void *element;
315 void **new_elements;
316 unsigned long flags;
317
318 BUG_ON(new_min_nr <= 0);
319 might_sleep();
320
321 spin_lock_irqsave(&pool->lock, flags);
322 if (new_min_nr <= pool->min_nr) {
323 while (new_min_nr < pool->curr_nr) {
324 element = remove_element(pool);
325 spin_unlock_irqrestore(lock: &pool->lock, flags);
326 pool->free(element, pool->pool_data);
327 spin_lock_irqsave(&pool->lock, flags);
328 }
329 pool->min_nr = new_min_nr;
330 goto out_unlock;
331 }
332 spin_unlock_irqrestore(lock: &pool->lock, flags);
333
334 /* Grow the pool */
335 new_elements = kmalloc_array(new_min_nr, sizeof(*new_elements),
336 GFP_KERNEL);
337 if (!new_elements)
338 return -ENOMEM;
339
340 spin_lock_irqsave(&pool->lock, flags);
341 if (unlikely(new_min_nr <= pool->min_nr)) {
342 /* Raced, other resize will do our work */
343 spin_unlock_irqrestore(lock: &pool->lock, flags);
344 kfree(objp: new_elements);
345 goto out;
346 }
347 memcpy(to: new_elements, from: pool->elements,
348 len: pool->curr_nr * sizeof(*new_elements));
349 kfree(objp: pool->elements);
350 pool->elements = new_elements;
351 pool->min_nr = new_min_nr;
352
353 while (pool->curr_nr < pool->min_nr) {
354 spin_unlock_irqrestore(lock: &pool->lock, flags);
355 element = pool->alloc(GFP_KERNEL, pool->pool_data);
356 if (!element)
357 goto out;
358 spin_lock_irqsave(&pool->lock, flags);
359 if (pool->curr_nr < pool->min_nr) {
360 add_element(pool, element);
361 } else {
362 spin_unlock_irqrestore(lock: &pool->lock, flags);
363 pool->free(element, pool->pool_data); /* Raced */
364 goto out;
365 }
366 }
367out_unlock:
368 spin_unlock_irqrestore(lock: &pool->lock, flags);
369out:
370 return 0;
371}
372EXPORT_SYMBOL(mempool_resize);
373
374/**
375 * mempool_alloc - allocate an element from a specific memory pool
376 * @pool: pointer to the memory pool which was allocated via
377 * mempool_create().
378 * @gfp_mask: the usual allocation bitmask.
379 *
380 * this function only sleeps if the alloc_fn() function sleeps or
381 * returns NULL. Note that due to preallocation, this function
382 * *never* fails when called from process contexts. (it might
383 * fail if called from an IRQ context.)
384 * Note: using __GFP_ZERO is not supported.
385 *
386 * Return: pointer to the allocated element or %NULL on error.
387 */
388void *mempool_alloc_noprof(mempool_t *pool, gfp_t gfp_mask)
389{
390 void *element;
391 unsigned long flags;
392 wait_queue_entry_t wait;
393 gfp_t gfp_temp;
394
395 VM_WARN_ON_ONCE(gfp_mask & __GFP_ZERO);
396 might_alloc(gfp_mask);
397
398 gfp_mask |= __GFP_NOMEMALLOC; /* don't allocate emergency reserves */
399 gfp_mask |= __GFP_NORETRY; /* don't loop in __alloc_pages */
400 gfp_mask |= __GFP_NOWARN; /* failures are OK */
401
402 gfp_temp = gfp_mask & ~(__GFP_DIRECT_RECLAIM|__GFP_IO);
403
404repeat_alloc:
405
406 element = pool->alloc(gfp_temp, pool->pool_data);
407 if (likely(element != NULL))
408 return element;
409
410 spin_lock_irqsave(&pool->lock, flags);
411 if (likely(pool->curr_nr)) {
412 element = remove_element(pool);
413 spin_unlock_irqrestore(lock: &pool->lock, flags);
414 /* paired with rmb in mempool_free(), read comment there */
415 smp_wmb();
416 /*
417 * Update the allocation stack trace as this is more useful
418 * for debugging.
419 */
420 kmemleak_update_trace(ptr: element);
421 return element;
422 }
423
424 /*
425 * We use gfp mask w/o direct reclaim or IO for the first round. If
426 * alloc failed with that and @pool was empty, retry immediately.
427 */
428 if (gfp_temp != gfp_mask) {
429 spin_unlock_irqrestore(lock: &pool->lock, flags);
430 gfp_temp = gfp_mask;
431 goto repeat_alloc;
432 }
433
434 /* We must not sleep if !__GFP_DIRECT_RECLAIM */
435 if (!(gfp_mask & __GFP_DIRECT_RECLAIM)) {
436 spin_unlock_irqrestore(lock: &pool->lock, flags);
437 return NULL;
438 }
439
440 /* Let's wait for someone else to return an element to @pool */
441 init_wait(&wait);
442 prepare_to_wait(wq_head: &pool->wait, wq_entry: &wait, TASK_UNINTERRUPTIBLE);
443
444 spin_unlock_irqrestore(lock: &pool->lock, flags);
445
446 /*
447 * FIXME: this should be io_schedule(). The timeout is there as a
448 * workaround for some DM problems in 2.6.18.
449 */
450 io_schedule_timeout(timeout: 5*HZ);
451
452 finish_wait(wq_head: &pool->wait, wq_entry: &wait);
453 goto repeat_alloc;
454}
455EXPORT_SYMBOL(mempool_alloc_noprof);
456
457/**
458 * mempool_alloc_preallocated - allocate an element from preallocated elements
459 * belonging to a specific memory pool
460 * @pool: pointer to the memory pool which was allocated via
461 * mempool_create().
462 *
463 * This function is similar to mempool_alloc, but it only attempts allocating
464 * an element from the preallocated elements. It does not sleep and immediately
465 * returns if no preallocated elements are available.
466 *
467 * Return: pointer to the allocated element or %NULL if no elements are
468 * available.
469 */
470void *mempool_alloc_preallocated(mempool_t *pool)
471{
472 void *element;
473 unsigned long flags;
474
475 spin_lock_irqsave(&pool->lock, flags);
476 if (likely(pool->curr_nr)) {
477 element = remove_element(pool);
478 spin_unlock_irqrestore(lock: &pool->lock, flags);
479 /* paired with rmb in mempool_free(), read comment there */
480 smp_wmb();
481 /*
482 * Update the allocation stack trace as this is more useful
483 * for debugging.
484 */
485 kmemleak_update_trace(ptr: element);
486 return element;
487 }
488 spin_unlock_irqrestore(lock: &pool->lock, flags);
489
490 return NULL;
491}
492EXPORT_SYMBOL(mempool_alloc_preallocated);
493
494/**
495 * mempool_free - return an element to the pool.
496 * @element: pool element pointer.
497 * @pool: pointer to the memory pool which was allocated via
498 * mempool_create().
499 *
500 * this function only sleeps if the free_fn() function sleeps.
501 */
502void mempool_free(void *element, mempool_t *pool)
503{
504 unsigned long flags;
505
506 if (unlikely(element == NULL))
507 return;
508
509 /*
510 * Paired with the wmb in mempool_alloc(). The preceding read is
511 * for @element and the following @pool->curr_nr. This ensures
512 * that the visible value of @pool->curr_nr is from after the
513 * allocation of @element. This is necessary for fringe cases
514 * where @element was passed to this task without going through
515 * barriers.
516 *
517 * For example, assume @p is %NULL at the beginning and one task
518 * performs "p = mempool_alloc(...);" while another task is doing
519 * "while (!p) cpu_relax(); mempool_free(p, ...);". This function
520 * may end up using curr_nr value which is from before allocation
521 * of @p without the following rmb.
522 */
523 smp_rmb();
524
525 /*
526 * For correctness, we need a test which is guaranteed to trigger
527 * if curr_nr + #allocated == min_nr. Testing curr_nr < min_nr
528 * without locking achieves that and refilling as soon as possible
529 * is desirable.
530 *
531 * Because curr_nr visible here is always a value after the
532 * allocation of @element, any task which decremented curr_nr below
533 * min_nr is guaranteed to see curr_nr < min_nr unless curr_nr gets
534 * incremented to min_nr afterwards. If curr_nr gets incremented
535 * to min_nr after the allocation of @element, the elements
536 * allocated after that are subject to the same guarantee.
537 *
538 * Waiters happen iff curr_nr is 0 and the above guarantee also
539 * ensures that there will be frees which return elements to the
540 * pool waking up the waiters.
541 */
542 if (unlikely(READ_ONCE(pool->curr_nr) < pool->min_nr)) {
543 spin_lock_irqsave(&pool->lock, flags);
544 if (likely(pool->curr_nr < pool->min_nr)) {
545 add_element(pool, element);
546 spin_unlock_irqrestore(lock: &pool->lock, flags);
547 if (wq_has_sleeper(wq_head: &pool->wait))
548 wake_up(&pool->wait);
549 return;
550 }
551 spin_unlock_irqrestore(lock: &pool->lock, flags);
552 }
553
554 /*
555 * Handle the min_nr = 0 edge case:
556 *
557 * For zero-minimum pools, curr_nr < min_nr (0 < 0) never succeeds,
558 * so waiters sleeping on pool->wait would never be woken by the
559 * wake-up path of previous test. This explicit check ensures the
560 * allocation of element when both min_nr and curr_nr are 0, and
561 * any active waiters are properly awakened.
562 */
563 if (unlikely(pool->min_nr == 0 &&
564 READ_ONCE(pool->curr_nr) == 0)) {
565 spin_lock_irqsave(&pool->lock, flags);
566 if (likely(pool->curr_nr == 0)) {
567 add_element(pool, element);
568 spin_unlock_irqrestore(lock: &pool->lock, flags);
569 if (wq_has_sleeper(wq_head: &pool->wait))
570 wake_up(&pool->wait);
571 return;
572 }
573 spin_unlock_irqrestore(lock: &pool->lock, flags);
574 }
575
576 pool->free(element, pool->pool_data);
577}
578EXPORT_SYMBOL(mempool_free);
579
580/*
581 * A commonly used alloc and free fn.
582 */
583void *mempool_alloc_slab(gfp_t gfp_mask, void *pool_data)
584{
585 struct kmem_cache *mem = pool_data;
586 VM_BUG_ON(mem->ctor);
587 return kmem_cache_alloc_noprof(cachep: mem, flags: gfp_mask);
588}
589EXPORT_SYMBOL(mempool_alloc_slab);
590
591void mempool_free_slab(void *element, void *pool_data)
592{
593 struct kmem_cache *mem = pool_data;
594 kmem_cache_free(s: mem, objp: element);
595}
596EXPORT_SYMBOL(mempool_free_slab);
597
598/*
599 * A commonly used alloc and free fn that kmalloc/kfrees the amount of memory
600 * specified by pool_data
601 */
602void *mempool_kmalloc(gfp_t gfp_mask, void *pool_data)
603{
604 size_t size = (size_t)pool_data;
605 return kmalloc_noprof(size, flags: gfp_mask);
606}
607EXPORT_SYMBOL(mempool_kmalloc);
608
609void mempool_kfree(void *element, void *pool_data)
610{
611 kfree(objp: element);
612}
613EXPORT_SYMBOL(mempool_kfree);
614
615void *mempool_kvmalloc(gfp_t gfp_mask, void *pool_data)
616{
617 size_t size = (size_t)pool_data;
618 return kvmalloc(size, gfp_mask);
619}
620EXPORT_SYMBOL(mempool_kvmalloc);
621
622void mempool_kvfree(void *element, void *pool_data)
623{
624 kvfree(addr: element);
625}
626EXPORT_SYMBOL(mempool_kvfree);
627
628/*
629 * A simple mempool-backed page allocator that allocates pages
630 * of the order specified by pool_data.
631 */
632void *mempool_alloc_pages(gfp_t gfp_mask, void *pool_data)
633{
634 int order = (int)(long)pool_data;
635 return alloc_pages_noprof(gfp: gfp_mask, order);
636}
637EXPORT_SYMBOL(mempool_alloc_pages);
638
639void mempool_free_pages(void *element, void *pool_data)
640{
641 int order = (int)(long)pool_data;
642 __free_pages(page: element, order);
643}
644EXPORT_SYMBOL(mempool_free_pages);
645