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