slab.c 108 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
/*
 * linux/mm/slab.c
 * Written by Mark Hemment, 1996/97.
 * (markhe@nextd.demon.co.uk)
 *
 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
 *
 * Major cleanup, different bufctl logic, per-cpu arrays
 *	(c) 2000 Manfred Spraul
 *
 * Cleanup, make the head arrays unconditional, preparation for NUMA
 * 	(c) 2002 Manfred Spraul
 *
 * An implementation of the Slab Allocator as described in outline in;
 *	UNIX Internals: The New Frontiers by Uresh Vahalia
 *	Pub: Prentice Hall	ISBN 0-13-101908-2
 * or with a little more detail in;
 *	The Slab Allocator: An Object-Caching Kernel Memory Allocator
 *	Jeff Bonwick (Sun Microsystems).
 *	Presented at: USENIX Summer 1994 Technical Conference
 *
 * The memory is organized in caches, one cache for each object type.
 * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
 * Each cache consists out of many slabs (they are small (usually one
 * page long) and always contiguous), and each slab contains multiple
 * initialized objects.
 *
 * This means, that your constructor is used only for newly allocated
Simon Arlott's avatar
Simon Arlott committed
29
 * slabs and you must pass objects with the same initializations to
Linus Torvalds's avatar
Linus Torvalds committed
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
 * kmem_cache_free.
 *
 * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
 * normal). If you need a special memory type, then must create a new
 * cache for that memory type.
 *
 * In order to reduce fragmentation, the slabs are sorted in 3 groups:
 *   full slabs with 0 free objects
 *   partial slabs
 *   empty slabs with no allocated objects
 *
 * If partial slabs exist, then new allocations come from these slabs,
 * otherwise from empty slabs or new slabs are allocated.
 *
 * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
 * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
 *
 * Each cache has a short per-cpu head array, most allocs
 * and frees go into that array, and if that array overflows, then 1/2
 * of the entries in the array are given back into the global cache.
 * The head array is strictly LIFO and should improve the cache hit rates.
 * On SMP, it additionally reduces the spinlock operations.
 *
Andrew Morton's avatar
Andrew Morton committed
53
 * The c_cpuarray may not be read with enabled local interrupts -
Linus Torvalds's avatar
Linus Torvalds committed
54
55
56
57
 * it's changed with a smp_call_function().
 *
 * SMP synchronization:
 *  constructors and destructors are called without any locking.
58
 *  Several members in struct kmem_cache and struct slab never change, they
Linus Torvalds's avatar
Linus Torvalds committed
59
60
61
62
63
64
65
66
67
68
69
70
 *	are accessed without any locking.
 *  The per-cpu arrays are never accessed from the wrong cpu, no locking,
 *  	and local interrupts are disabled so slab code is preempt-safe.
 *  The non-constant members are protected with a per-cache irq spinlock.
 *
 * Many thanks to Mark Hemment, who wrote another per-cpu slab patch
 * in 2000 - many ideas in the current implementation are derived from
 * his patch.
 *
 * Further notes from the original documentation:
 *
 * 11 April '97.  Started multi-threading - markhe
71
 *	The global cache-chain is protected by the mutex 'slab_mutex'.
Linus Torvalds's avatar
Linus Torvalds committed
72
73
74
75
76
77
 *	The sem is only needed when accessing/extending the cache-chain, which
 *	can never happen inside an interrupt (kmem_cache_create(),
 *	kmem_cache_shrink() and kmem_cache_reap()).
 *
 *	At present, each engine can be growing a cache.  This should be blocked.
 *
78
79
80
81
82
83
84
85
86
 * 15 March 2005. NUMA slab allocator.
 *	Shai Fultheim <shai@scalex86.org>.
 *	Shobhit Dayal <shobhit@calsoftinc.com>
 *	Alok N Kataria <alokk@calsoftinc.com>
 *	Christoph Lameter <christoph@lameter.com>
 *
 *	Modified the slab allocator to be node aware on NUMA systems.
 *	Each node has its own list of partial, free and full slabs.
 *	All object allocations for a node occur from node specific slab lists.
Linus Torvalds's avatar
Linus Torvalds committed
87
88
89
90
 */

#include	<linux/slab.h>
#include	<linux/mm.h>
91
#include	<linux/poison.h>
Linus Torvalds's avatar
Linus Torvalds committed
92
93
94
95
96
#include	<linux/swap.h>
#include	<linux/cache.h>
#include	<linux/interrupt.h>
#include	<linux/init.h>
#include	<linux/compiler.h>
97
#include	<linux/cpuset.h>
98
#include	<linux/proc_fs.h>
Linus Torvalds's avatar
Linus Torvalds committed
99
100
101
102
103
104
105
#include	<linux/seq_file.h>
#include	<linux/notifier.h>
#include	<linux/kallsyms.h>
#include	<linux/cpu.h>
#include	<linux/sysctl.h>
#include	<linux/module.h>
#include	<linux/rcupdate.h>
106
#include	<linux/string.h>
107
#include	<linux/uaccess.h>
108
#include	<linux/nodemask.h>
109
#include	<linux/kmemleak.h>
110
#include	<linux/mempolicy.h>
Ingo Molnar's avatar
Ingo Molnar committed
111
#include	<linux/mutex.h>
112
#include	<linux/fault-inject.h>
Ingo Molnar's avatar
Ingo Molnar committed
113
#include	<linux/rtmutex.h>
114
#include	<linux/reciprocal_div.h>
115
#include	<linux/debugobjects.h>
Pekka Enberg's avatar
Pekka Enberg committed
116
#include	<linux/kmemcheck.h>
117
#include	<linux/memory.h>
118
#include	<linux/prefetch.h>
Linus Torvalds's avatar
Linus Torvalds committed
119

120
121
#include	<net/sock.h>

Linus Torvalds's avatar
Linus Torvalds committed
122
123
124
125
#include	<asm/cacheflush.h>
#include	<asm/tlbflush.h>
#include	<asm/page.h>

126
127
#include <trace/events/kmem.h>

128
129
#include	"internal.h"

130
131
#include	"slab.h"

Linus Torvalds's avatar
Linus Torvalds committed
132
/*
133
 * DEBUG	- 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
Linus Torvalds's avatar
Linus Torvalds committed
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * STATS	- 1 to collect stats for /proc/slabinfo.
 *		  0 for faster, smaller code (especially in the critical paths).
 *
 * FORCED_DEBUG	- 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
 */

#ifdef CONFIG_DEBUG_SLAB
#define	DEBUG		1
#define	STATS		1
#define	FORCED_DEBUG	1
#else
#define	DEBUG		0
#define	STATS		0
#define	FORCED_DEBUG	0
#endif

/* Shouldn't this be in a header file somewhere? */
#define	BYTES_PER_WORD		sizeof(void *)
David Woodhouse's avatar
David Woodhouse committed
154
#define	REDZONE_ALIGN		max(BYTES_PER_WORD, __alignof__(unsigned long long))
Linus Torvalds's avatar
Linus Torvalds committed
155
156
157
158
159

#ifndef ARCH_KMALLOC_FLAGS
#define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
#endif

160
161
162
163
164
165
166
167
168
#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
				<= SLAB_OBJ_MIN_SIZE) ? 1 : 0)

#if FREELIST_BYTE_INDEX
typedef unsigned char freelist_idx_t;
#else
typedef unsigned short freelist_idx_t;
#endif

169
#define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
170

171
172
173
174
175
176
/*
 * true if a page was allocated from pfmemalloc reserves for network-based
 * swap
 */
static bool pfmemalloc_active __read_mostly;

Linus Torvalds's avatar
Linus Torvalds committed
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
/*
 * struct array_cache
 *
 * Purpose:
 * - LIFO ordering, to hand out cache-warm objects from _alloc
 * - reduce the number of linked list operations
 * - reduce spinlock operations
 *
 * The limit is stored in the per-cpu structure to reduce the data cache
 * footprint.
 *
 */
struct array_cache {
	unsigned int avail;
	unsigned int limit;
	unsigned int batchcount;
	unsigned int touched;
194
	void *entry[];	/*
Andrew Morton's avatar
Andrew Morton committed
195
196
197
			 * Must have this definition in here for the proper
			 * alignment of array_cache. Also simplifies accessing
			 * the entries.
198
199
200
201
			 *
			 * Entries should not be directly dereferenced as
			 * entries belonging to slabs marked pfmemalloc will
			 * have the lower bits set SLAB_OBJ_PFMEMALLOC
Andrew Morton's avatar
Andrew Morton committed
202
			 */
Linus Torvalds's avatar
Linus Torvalds committed
203
204
};

Joonsoo Kim's avatar
Joonsoo Kim committed
205
206
207
208
209
struct alien_cache {
	spinlock_t lock;
	struct array_cache ac;
};

210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
#define SLAB_OBJ_PFMEMALLOC	1
static inline bool is_obj_pfmemalloc(void *objp)
{
	return (unsigned long)objp & SLAB_OBJ_PFMEMALLOC;
}

static inline void set_obj_pfmemalloc(void **objp)
{
	*objp = (void *)((unsigned long)*objp | SLAB_OBJ_PFMEMALLOC);
	return;
}

static inline void clear_obj_pfmemalloc(void **objp)
{
	*objp = (void *)((unsigned long)*objp & ~SLAB_OBJ_PFMEMALLOC);
}

227
228
229
/*
 * Need this for bootstrapping a per node allocator.
 */
230
#define NUM_INIT_LISTS (2 * MAX_NUMNODES)
231
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
232
#define	CACHE_CACHE 0
233
#define	SIZE_NODE (MAX_NUMNODES)
234

235
static int drain_freelist(struct kmem_cache *cache,
236
			struct kmem_cache_node *n, int tofree);
237
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
238
239
			int node, struct list_head *list);
static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
240
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
241
static void cache_reap(struct work_struct *unused);
242

243
244
static int slab_early_init = 1;

245
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Linus Torvalds's avatar
Linus Torvalds committed
246

247
static void kmem_cache_node_init(struct kmem_cache_node *parent)
248
249
250
251
252
253
{
	INIT_LIST_HEAD(&parent->slabs_full);
	INIT_LIST_HEAD(&parent->slabs_partial);
	INIT_LIST_HEAD(&parent->slabs_free);
	parent->shared = NULL;
	parent->alien = NULL;
254
	parent->colour_next = 0;
255
256
257
258
259
	spin_lock_init(&parent->list_lock);
	parent->free_objects = 0;
	parent->free_touched = 0;
}

Andrew Morton's avatar
Andrew Morton committed
260
261
262
#define MAKE_LIST(cachep, listp, slab, nodeid)				\
	do {								\
		INIT_LIST_HEAD(listp);					\
263
		list_splice(&get_node(cachep, nodeid)->slab, listp);	\
264
265
	} while (0)

Andrew Morton's avatar
Andrew Morton committed
266
267
#define	MAKE_ALL_LISTS(cachep, ptr, nodeid)				\
	do {								\
268
269
270
271
	MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid);	\
	MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
	MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid);	\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
272
273
274

#define CFLGS_OFF_SLAB		(0x80000000UL)
#define	OFF_SLAB(x)	((x)->flags & CFLGS_OFF_SLAB)
275
#define OFF_SLAB_MIN_SIZE (max_t(size_t, PAGE_SIZE >> 5, KMALLOC_MIN_SIZE + 1))
Linus Torvalds's avatar
Linus Torvalds committed
276
277

#define BATCHREFILL_LIMIT	16
Andrew Morton's avatar
Andrew Morton committed
278
279
280
/*
 * Optimization question: fewer reaps means less probability for unnessary
 * cpucache drain/refill cycles.
Linus Torvalds's avatar
Linus Torvalds committed
281
 *
Adrian Bunk's avatar
Adrian Bunk committed
282
 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds's avatar
Linus Torvalds committed
283
284
 * which could lock up otherwise freeable slabs.
 */
285
286
#define REAPTIMEOUT_AC		(2*HZ)
#define REAPTIMEOUT_NODE	(4*HZ)
Linus Torvalds's avatar
Linus Torvalds committed
287
288
289
290
291
292

#if STATS
#define	STATS_INC_ACTIVE(x)	((x)->num_active++)
#define	STATS_DEC_ACTIVE(x)	((x)->num_active--)
#define	STATS_INC_ALLOCED(x)	((x)->num_allocations++)
#define	STATS_INC_GROWN(x)	((x)->grown++)
293
#define	STATS_ADD_REAPED(x,y)	((x)->reaped += (y))
Andrew Morton's avatar
Andrew Morton committed
294
295
296
297
298
#define	STATS_SET_HIGH(x)						\
	do {								\
		if ((x)->num_active > (x)->high_mark)			\
			(x)->high_mark = (x)->num_active;		\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
299
300
#define	STATS_INC_ERR(x)	((x)->errors++)
#define	STATS_INC_NODEALLOCS(x)	((x)->node_allocs++)
301
#define	STATS_INC_NODEFREES(x)	((x)->node_frees++)
302
#define STATS_INC_ACOVERFLOW(x)   ((x)->node_overflow++)
Andrew Morton's avatar
Andrew Morton committed
303
304
305
306
307
#define	STATS_SET_FREEABLE(x, i)					\
	do {								\
		if ((x)->max_freeable < i)				\
			(x)->max_freeable = i;				\
	} while (0)
Linus Torvalds's avatar
Linus Torvalds committed
308
309
310
311
312
313
314
315
316
#define STATS_INC_ALLOCHIT(x)	atomic_inc(&(x)->allochit)
#define STATS_INC_ALLOCMISS(x)	atomic_inc(&(x)->allocmiss)
#define STATS_INC_FREEHIT(x)	atomic_inc(&(x)->freehit)
#define STATS_INC_FREEMISS(x)	atomic_inc(&(x)->freemiss)
#else
#define	STATS_INC_ACTIVE(x)	do { } while (0)
#define	STATS_DEC_ACTIVE(x)	do { } while (0)
#define	STATS_INC_ALLOCED(x)	do { } while (0)
#define	STATS_INC_GROWN(x)	do { } while (0)
317
#define	STATS_ADD_REAPED(x,y)	do { (void)(y); } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
318
319
320
#define	STATS_SET_HIGH(x)	do { } while (0)
#define	STATS_INC_ERR(x)	do { } while (0)
#define	STATS_INC_NODEALLOCS(x)	do { } while (0)
321
#define	STATS_INC_NODEFREES(x)	do { } while (0)
322
#define STATS_INC_ACOVERFLOW(x)   do { } while (0)
Andrew Morton's avatar
Andrew Morton committed
323
#define	STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds's avatar
Linus Torvalds committed
324
325
326
327
328
329
330
331
#define STATS_INC_ALLOCHIT(x)	do { } while (0)
#define STATS_INC_ALLOCMISS(x)	do { } while (0)
#define STATS_INC_FREEHIT(x)	do { } while (0)
#define STATS_INC_FREEMISS(x)	do { } while (0)
#endif

#if DEBUG

Andrew Morton's avatar
Andrew Morton committed
332
333
/*
 * memory layout of objects:
Linus Torvalds's avatar
Linus Torvalds committed
334
 * 0		: objp
335
 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds's avatar
Linus Torvalds committed
336
337
 * 		the end of an object is aligned with the end of the real
 * 		allocation. Catches writes behind the end of the allocation.
338
 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds's avatar
Linus Torvalds committed
339
 * 		redzone word.
340
 * cachep->obj_offset: The real object.
341
342
 * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
 * cachep->size - 1* BYTES_PER_WORD: last caller address
Andrew Morton's avatar
Andrew Morton committed
343
 *					[BYTES_PER_WORD long]
Linus Torvalds's avatar
Linus Torvalds committed
344
 */
345
static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
346
{
347
	return cachep->obj_offset;
Linus Torvalds's avatar
Linus Torvalds committed
348
349
}

350
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
351
352
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
353
354
	return (unsigned long long*) (objp + obj_offset(cachep) -
				      sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
355
356
}

357
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
358
359
360
{
	BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
	if (cachep->flags & SLAB_STORE_USER)
361
		return (unsigned long long *)(objp + cachep->size -
362
					      sizeof(unsigned long long) -
David Woodhouse's avatar
David Woodhouse committed
363
					      REDZONE_ALIGN);
364
	return (unsigned long long *) (objp + cachep->size -
365
				       sizeof(unsigned long long));
Linus Torvalds's avatar
Linus Torvalds committed
366
367
}

368
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds's avatar
Linus Torvalds committed
369
370
{
	BUG_ON(!(cachep->flags & SLAB_STORE_USER));
371
	return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds's avatar
Linus Torvalds committed
372
373
374
375
}

#else

376
#define obj_offset(x)			0
377
378
#define dbg_redzone1(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
#define dbg_redzone2(cachep, objp)	({BUG(); (unsigned long long *)NULL;})
Linus Torvalds's avatar
Linus Torvalds committed
379
380
381
382
#define dbg_userword(cachep, objp)	({BUG(); (void **)NULL;})

#endif

383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
#define OBJECT_FREE (0)
#define OBJECT_ACTIVE (1)

#ifdef CONFIG_DEBUG_SLAB_LEAK

static void set_obj_status(struct page *page, int idx, int val)
{
	int freelist_size;
	char *status;
	struct kmem_cache *cachep = page->slab_cache;

	freelist_size = cachep->num * sizeof(freelist_idx_t);
	status = (char *)page->freelist + freelist_size;
	status[idx] = val;
}

static inline unsigned int get_obj_status(struct page *page, int idx)
{
	int freelist_size;
	char *status;
	struct kmem_cache *cachep = page->slab_cache;

	freelist_size = cachep->num * sizeof(freelist_idx_t);
	status = (char *)page->freelist + freelist_size;

	return status[idx];
}

#else
static inline void set_obj_status(struct page *page, int idx, int val) {}

#endif

Linus Torvalds's avatar
Linus Torvalds committed
416
/*
417
418
 * Do not go above this order unless 0 objects fit into the slab or
 * overridden on the command line.
Linus Torvalds's avatar
Linus Torvalds committed
419
 */
420
421
422
#define	SLAB_MAX_ORDER_HI	1
#define	SLAB_MAX_ORDER_LO	0
static int slab_max_order = SLAB_MAX_ORDER_LO;
423
static bool slab_max_order_set __initdata;
Linus Torvalds's avatar
Linus Torvalds committed
424

425
426
static inline struct kmem_cache *virt_to_cache(const void *obj)
{
427
	struct page *page = virt_to_head_page(obj);
428
	return page->slab_cache;
429
430
}

431
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
432
433
				 unsigned int idx)
{
434
	return page->s_mem + cache->size * idx;
435
436
}

437
/*
438
439
440
 * We want to avoid an expensive divide : (offset / cache->size)
 *   Using the fact that size is a constant for a particular cache,
 *   we can replace (offset / cache->size) by
441
442
443
 *   reciprocal_divide(offset, cache->reciprocal_buffer_size)
 */
static inline unsigned int obj_to_index(const struct kmem_cache *cache,
444
					const struct page *page, void *obj)
445
{
446
	u32 offset = (obj - page->s_mem);
447
	return reciprocal_divide(offset, cache->reciprocal_buffer_size);
448
449
}

450
#define BOOT_CPUCACHE_ENTRIES	1
Linus Torvalds's avatar
Linus Torvalds committed
451
/* internal cache of cache description objs */
452
static struct kmem_cache kmem_cache_boot = {
453
454
455
	.batchcount = 1,
	.limit = BOOT_CPUCACHE_ENTRIES,
	.shared = 1,
456
	.size = sizeof(struct kmem_cache),
457
	.name = "kmem_cache",
Linus Torvalds's avatar
Linus Torvalds committed
458
459
};

460
461
#define BAD_ALIEN_MAGIC 0x01020304ul

462
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds's avatar
Linus Torvalds committed
463

464
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds's avatar
Linus Torvalds committed
465
{
466
	return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds's avatar
Linus Torvalds committed
467
468
}

469
470
471
472
473
474
475
476
477
478
479
480
481
482
static size_t calculate_freelist_size(int nr_objs, size_t align)
{
	size_t freelist_size;

	freelist_size = nr_objs * sizeof(freelist_idx_t);
	if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK))
		freelist_size += nr_objs * sizeof(char);

	if (align)
		freelist_size = ALIGN(freelist_size, align);

	return freelist_size;
}

483
484
static int calculate_nr_objs(size_t slab_size, size_t buffer_size,
				size_t idx_size, size_t align)
Linus Torvalds's avatar
Linus Torvalds committed
485
{
486
	int nr_objs;
487
	size_t remained_size;
488
	size_t freelist_size;
489
	int extra_space = 0;
490

491
492
	if (IS_ENABLED(CONFIG_DEBUG_SLAB_LEAK))
		extra_space = sizeof(char);
493
494
495
496
497
498
499
500
	/*
	 * Ignore padding for the initial guess. The padding
	 * is at most @align-1 bytes, and @buffer_size is at
	 * least @align. In the worst case, this result will
	 * be one greater than the number of objects that fit
	 * into the memory allocation when taking the padding
	 * into account.
	 */
501
	nr_objs = slab_size / (buffer_size + idx_size + extra_space);
502
503
504
505
506

	/*
	 * This calculated number will be either the right
	 * amount, or one greater than what we want.
	 */
507
508
509
	remained_size = slab_size - nr_objs * buffer_size;
	freelist_size = calculate_freelist_size(nr_objs, align);
	if (remained_size < freelist_size)
510
511
512
		nr_objs--;

	return nr_objs;
513
}
Linus Torvalds's avatar
Linus Torvalds committed
514

Andrew Morton's avatar
Andrew Morton committed
515
516
517
/*
 * Calculate the number of objects and left-over bytes for a given buffer size.
 */
518
519
520
521
522
523
524
static void cache_estimate(unsigned long gfporder, size_t buffer_size,
			   size_t align, int flags, size_t *left_over,
			   unsigned int *num)
{
	int nr_objs;
	size_t mgmt_size;
	size_t slab_size = PAGE_SIZE << gfporder;
Linus Torvalds's avatar
Linus Torvalds committed
525

526
527
528
529
530
	/*
	 * The slab management structure can be either off the slab or
	 * on it. For the latter case, the memory allocated for a
	 * slab is used for:
	 *
Joonsoo Kim's avatar
Joonsoo Kim committed
531
	 * - One freelist_idx_t for each object
532
533
534
535
536
537
538
539
540
541
542
543
544
	 * - Padding to respect alignment of @align
	 * - @buffer_size bytes for each object
	 *
	 * If the slab management structure is off the slab, then the
	 * alignment will already be calculated into the size. Because
	 * the slabs are all pages aligned, the objects will be at the
	 * correct alignment when allocated.
	 */
	if (flags & CFLGS_OFF_SLAB) {
		mgmt_size = 0;
		nr_objs = slab_size / buffer_size;

	} else {
545
		nr_objs = calculate_nr_objs(slab_size, buffer_size,
546
					sizeof(freelist_idx_t), align);
547
		mgmt_size = calculate_freelist_size(nr_objs, align);
548
549
550
	}
	*num = nr_objs;
	*left_over = slab_size - nr_objs*buffer_size - mgmt_size;
Linus Torvalds's avatar
Linus Torvalds committed
551
552
}

553
#if DEBUG
554
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds's avatar
Linus Torvalds committed
555

Andrew Morton's avatar
Andrew Morton committed
556
557
static void __slab_error(const char *function, struct kmem_cache *cachep,
			char *msg)
Linus Torvalds's avatar
Linus Torvalds committed
558
559
{
	printk(KERN_ERR "slab error in %s(): cache `%s': %s\n",
560
	       function, cachep->name, msg);
Linus Torvalds's avatar
Linus Torvalds committed
561
	dump_stack();
562
	add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds's avatar
Linus Torvalds committed
563
}
564
#endif
Linus Torvalds's avatar
Linus Torvalds committed
565

566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
/*
 * By default on NUMA we use alien caches to stage the freeing of
 * objects allocated from other nodes. This causes massive memory
 * inefficiencies when using fake NUMA setup to split memory into a
 * large number of small nodes, so it can be disabled on the command
 * line
  */

static int use_alien_caches __read_mostly = 1;
static int __init noaliencache_setup(char *s)
{
	use_alien_caches = 0;
	return 1;
}
__setup("noaliencache", noaliencache_setup);

582
583
584
585
586
587
588
589
590
591
592
static int __init slab_max_order_setup(char *str)
{
	get_option(&str, &slab_max_order);
	slab_max_order = slab_max_order < 0 ? 0 :
				min(slab_max_order, MAX_ORDER - 1);
	slab_max_order_set = true;

	return 1;
}
__setup("slab_max_order=", slab_max_order_setup);

593
594
595
596
597
598
599
#ifdef CONFIG_NUMA
/*
 * Special reaping functions for NUMA systems called from cache_reap().
 * These take care of doing round robin flushing of alien caches (containing
 * objects freed on different nodes from which they were allocated) and the
 * flushing of remote pcps by calling drain_node_pages.
 */
600
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
601
602
603
604
605

static void init_reap_node(int cpu)
{
	int node;

606
	node = next_node(cpu_to_mem(cpu), node_online_map);
607
	if (node == MAX_NUMNODES)
608
		node = first_node(node_online_map);
609

610
	per_cpu(slab_reap_node, cpu) = node;
611
612
613
614
}

static void next_reap_node(void)
{
615
	int node = __this_cpu_read(slab_reap_node);
616
617
618
619

	node = next_node(node, node_online_map);
	if (unlikely(node >= MAX_NUMNODES))
		node = first_node(node_online_map);
620
	__this_cpu_write(slab_reap_node, node);
621
622
623
624
625
626
627
}

#else
#define init_reap_node(cpu) do { } while (0)
#define next_reap_node(void) do { } while (0)
#endif

Linus Torvalds's avatar
Linus Torvalds committed
628
629
630
631
632
633
634
/*
 * Initiate the reap timer running on the target CPU.  We run at around 1 to 2Hz
 * via the workqueue/eventd.
 * Add the CPU number into the expiration time to minimize the possibility of
 * the CPUs getting into lockstep and contending for the global cache chain
 * lock.
 */
635
static void start_cpu_timer(int cpu)
Linus Torvalds's avatar
Linus Torvalds committed
636
{
637
	struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds's avatar
Linus Torvalds committed
638
639
640
641
642
643

	/*
	 * When this gets called from do_initcalls via cpucache_init(),
	 * init_workqueues() has already run, so keventd will be setup
	 * at that time.
	 */
644
	if (keventd_up() && reap_work->work.func == NULL) {
645
		init_reap_node(cpu);
646
		INIT_DEFERRABLE_WORK(reap_work, cache_reap);
647
648
		schedule_delayed_work_on(cpu, reap_work,
					__round_jiffies_relative(HZ, cpu));
Linus Torvalds's avatar
Linus Torvalds committed
649
650
651
	}
}

652
static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds's avatar
Linus Torvalds committed
653
{
654
655
	/*
	 * The array_cache structures contain pointers to free object.
Lucas De Marchi's avatar
Lucas De Marchi committed
656
	 * However, when such objects are allocated or transferred to another
657
658
659
660
	 * cache the pointers are not cleared and they could be counted as
	 * valid references during a kmemleak scan. Therefore, kmemleak must
	 * not scan such objects.
	 */
661
662
663
664
665
666
	kmemleak_no_scan(ac);
	if (ac) {
		ac->avail = 0;
		ac->limit = limit;
		ac->batchcount = batch;
		ac->touched = 0;
Linus Torvalds's avatar
Linus Torvalds committed
667
	}
668
669
670
671
672
}

static struct array_cache *alloc_arraycache(int node, int entries,
					    int batchcount, gfp_t gfp)
{
673
	size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
674
675
676
677
678
	struct array_cache *ac = NULL;

	ac = kmalloc_node(memsize, gfp, node);
	init_arraycache(ac, entries, batchcount);
	return ac;
Linus Torvalds's avatar
Linus Torvalds committed
679
680
}

681
static inline bool is_slab_pfmemalloc(struct page *page)
682
683
684
685
686
687
688
689
{
	return PageSlabPfmemalloc(page);
}

/* Clears pfmemalloc_active if no slabs have pfmalloc set */
static void recheck_pfmemalloc_active(struct kmem_cache *cachep,
						struct array_cache *ac)
{
690
	struct kmem_cache_node *n = get_node(cachep, numa_mem_id());
691
	struct page *page;
692
693
694
695
696
	unsigned long flags;

	if (!pfmemalloc_active)
		return;

697
	spin_lock_irqsave(&n->list_lock, flags);
698
699
	list_for_each_entry(page, &n->slabs_full, lru)
		if (is_slab_pfmemalloc(page))
700
701
			goto out;

702
703
	list_for_each_entry(page, &n->slabs_partial, lru)
		if (is_slab_pfmemalloc(page))
704
705
			goto out;

706
707
	list_for_each_entry(page, &n->slabs_free, lru)
		if (is_slab_pfmemalloc(page))
708
709
710
711
			goto out;

	pfmemalloc_active = false;
out:
712
	spin_unlock_irqrestore(&n->list_lock, flags);
713
714
}

715
static void *__ac_get_obj(struct kmem_cache *cachep, struct array_cache *ac,
716
717
718
719
720
721
722
						gfp_t flags, bool force_refill)
{
	int i;
	void *objp = ac->entry[--ac->avail];

	/* Ensure the caller is allowed to use objects from PFMEMALLOC slab */
	if (unlikely(is_obj_pfmemalloc(objp))) {
723
		struct kmem_cache_node *n;
724
725
726
727
728
729
730

		if (gfp_pfmemalloc_allowed(flags)) {
			clear_obj_pfmemalloc(&objp);
			return objp;
		}

		/* The caller cannot use PFMEMALLOC objects, find another one */
731
		for (i = 0; i < ac->avail; i++) {
732
733
734
735
736
737
738
739
740
741
742
743
744
			/* If a !PFMEMALLOC object is found, swap them */
			if (!is_obj_pfmemalloc(ac->entry[i])) {
				objp = ac->entry[i];
				ac->entry[i] = ac->entry[ac->avail];
				ac->entry[ac->avail] = objp;
				return objp;
			}
		}

		/*
		 * If there are empty slabs on the slabs_free list and we are
		 * being forced to refill the cache, mark this one !pfmemalloc.
		 */
745
		n = get_node(cachep, numa_mem_id());
746
		if (!list_empty(&n->slabs_free) && force_refill) {
747
			struct page *page = virt_to_head_page(objp);
748
			ClearPageSlabPfmemalloc(page);
749
750
751
752
753
754
755
756
757
758
759
760
761
			clear_obj_pfmemalloc(&objp);
			recheck_pfmemalloc_active(cachep, ac);
			return objp;
		}

		/* No !PFMEMALLOC objects available */
		ac->avail++;
		objp = NULL;
	}

	return objp;
}

762
763
764
765
766
767
768
769
770
771
772
773
774
static inline void *ac_get_obj(struct kmem_cache *cachep,
			struct array_cache *ac, gfp_t flags, bool force_refill)
{
	void *objp;

	if (unlikely(sk_memalloc_socks()))
		objp = __ac_get_obj(cachep, ac, flags, force_refill);
	else
		objp = ac->entry[--ac->avail];

	return objp;
}

Joonsoo Kim's avatar
Joonsoo Kim committed
775
776
static noinline void *__ac_put_obj(struct kmem_cache *cachep,
			struct array_cache *ac, void *objp)
777
778
779
{
	if (unlikely(pfmemalloc_active)) {
		/* Some pfmemalloc slabs exist, check if this is one */
780
		struct page *page = virt_to_head_page(objp);
781
782
783
784
		if (PageSlabPfmemalloc(page))
			set_obj_pfmemalloc(&objp);
	}

785
786
787
788
789
790
791
792
793
	return objp;
}

static inline void ac_put_obj(struct kmem_cache *cachep, struct array_cache *ac,
								void *objp)
{
	if (unlikely(sk_memalloc_socks()))
		objp = __ac_put_obj(cachep, ac, objp);

794
795
796
	ac->entry[ac->avail++] = objp;
}

797
798
799
800
801
802
803
804
805
806
/*
 * Transfer objects in one arraycache to another.
 * Locking must be handled by the caller.
 *
 * Return the number of entries transferred.
 */
static int transfer_objects(struct array_cache *to,
		struct array_cache *from, unsigned int max)
{
	/* Figure out how many entries to transfer */
807
	int nr = min3(from->avail, max, to->limit - to->avail);
808
809
810
811
812
813
814
815
816
817
818
819

	if (!nr)
		return 0;

	memcpy(to->entry + to->avail, from->entry + from->avail -nr,
			sizeof(void *) *nr);

	from->avail -= nr;
	to->avail += nr;
	return nr;
}

820
821
822
#ifndef CONFIG_NUMA

#define drain_alien_cache(cachep, alien) do { } while (0)
823
#define reap_alien(cachep, n) do { } while (0)
824

Joonsoo Kim's avatar
Joonsoo Kim committed
825
826
static inline struct alien_cache **alloc_alien_cache(int node,
						int limit, gfp_t gfp)
827
{
828
	return (struct alien_cache **)BAD_ALIEN_MAGIC;
829
830
}

Joonsoo Kim's avatar
Joonsoo Kim committed
831
static inline void free_alien_cache(struct alien_cache **ac_ptr)
832
833
834
835
836
837
838
839
840
841
842
843
844
845
{
}

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	return 0;
}

static inline void *alternate_node_alloc(struct kmem_cache *cachep,
		gfp_t flags)
{
	return NULL;
}

846
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
847
848
849
850
851
		 gfp_t flags, int nodeid)
{
	return NULL;
}

David Rientjes's avatar
David Rientjes committed
852
853
854
855
856
static inline gfp_t gfp_exact_node(gfp_t flags)
{
	return flags;
}

857
858
#else	/* CONFIG_NUMA */

859
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
860
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
861

Joonsoo Kim's avatar
Joonsoo Kim committed
862
863
864
static struct alien_cache *__alloc_alien_cache(int node, int entries,
						int batch, gfp_t gfp)
{
865
	size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kim's avatar
Joonsoo Kim committed
866
867
868
869
	struct alien_cache *alc = NULL;

	alc = kmalloc_node(memsize, gfp, node);
	init_arraycache(&alc->ac, entries, batch);
870
	spin_lock_init(&alc->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
871
872
873
874
	return alc;
}

static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
875
{
Joonsoo Kim's avatar
Joonsoo Kim committed
876
	struct alien_cache **alc_ptr;
877
	size_t memsize = sizeof(void *) * nr_node_ids;
878
879
880
881
	int i;

	if (limit > 1)
		limit = 12;
Joonsoo Kim's avatar
Joonsoo Kim committed
882
883
884
885
886
887
888
889
890
891
892
893
894
	alc_ptr = kzalloc_node(memsize, gfp, node);
	if (!alc_ptr)
		return NULL;

	for_each_node(i) {
		if (i == node || !node_online(i))
			continue;
		alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
		if (!alc_ptr[i]) {
			for (i--; i >= 0; i--)
				kfree(alc_ptr[i]);
			kfree(alc_ptr);
			return NULL;
895
896
		}
	}
Joonsoo Kim's avatar
Joonsoo Kim committed
897
	return alc_ptr;
898
899
}

Joonsoo Kim's avatar
Joonsoo Kim committed
900
static void free_alien_cache(struct alien_cache **alc_ptr)
901
902
903
{
	int i;

Joonsoo Kim's avatar
Joonsoo Kim committed
904
	if (!alc_ptr)
905
906
		return;
	for_each_node(i)
Joonsoo Kim's avatar
Joonsoo Kim committed
907
908
	    kfree(alc_ptr[i]);
	kfree(alc_ptr);
909
910
}

911
static void __drain_alien_cache(struct kmem_cache *cachep,
912
913
				struct array_cache *ac, int node,
				struct list_head *list)
914
{
915
	struct kmem_cache_node *n = get_node(cachep, node);
916
917

	if (ac->avail) {
918
		spin_lock(&n->list_lock);
919
920
921
922
923
		/*
		 * Stuff objects into the remote nodes shared array first.
		 * That way we could avoid the overhead of putting the objects
		 * into the free lists and getting them back later.
		 */
924
925
		if (n->shared)
			transfer_objects(n->shared, ac, ac->limit);
926

927
		free_block(cachep, ac->entry, ac->avail, node, list);
928
		ac->avail = 0;
929
		spin_unlock(&n->list_lock);
930
931
932
	}
}

933
934
935
/*
 * Called from cache_reap() to regularly drain alien caches round robin.
 */
936
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
937
{
938
	int node = __this_cpu_read(slab_reap_node);
939

940
	if (n->alien) {
Joonsoo Kim's avatar
Joonsoo Kim committed
941
942
943
944
945
		struct alien_cache *alc = n->alien[node];
		struct array_cache *ac;

		if (alc) {
			ac = &alc->ac;
946
			if (ac->avail && spin_trylock_irq(&alc->lock)) {
947
948
949
				LIST_HEAD(list);

				__drain_alien_cache(cachep, ac, node, &list);
950
				spin_unlock_irq(&alc->lock);
951
				slabs_destroy(cachep, &list);
Joonsoo Kim's avatar
Joonsoo Kim committed
952
			}
953
954
955
956
		}
	}
}

Andrew Morton's avatar
Andrew Morton committed
957
static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim's avatar
Joonsoo Kim committed
958
				struct alien_cache **alien)
959
{
960
	int i = 0;
Joonsoo Kim's avatar
Joonsoo Kim committed
961
	struct alien_cache *alc;
962
963
964
965
	struct array_cache *ac;
	unsigned long flags;

	for_each_online_node(i) {
Joonsoo Kim's avatar
Joonsoo Kim committed
966
967
		alc = alien[i];
		if (alc) {
968
969
			LIST_HEAD(list);

Joonsoo Kim's avatar
Joonsoo Kim committed
970
			ac = &alc->ac;
971
			spin_lock_irqsave(&alc->lock, flags);
972
			__drain_alien_cache(cachep, ac, i, &list);
973
			spin_unlock_irqrestore(&alc->lock, flags);
974
			slabs_destroy(cachep, &list);
975
976
977
		}
	}
}
978

979
980
static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
				int node, int page_node)
981
{
982
	struct kmem_cache_node *n;
Joonsoo Kim's avatar
Joonsoo Kim committed
983
984
	struct alien_cache *alien = NULL;
	struct array_cache *ac;
985
	LIST_HEAD(list);
986

987
	n = get_node(cachep, node);
988
	STATS_INC_NODEFREES(cachep);
989
990
	if (n->alien && n->alien[page_node]) {
		alien = n->alien[page_node];
Joonsoo Kim's avatar
Joonsoo Kim committed
991
		ac = &alien->ac;
992
		spin_lock(&alien->lock);
Joonsoo Kim's avatar
Joonsoo Kim committed
993
		if (unlikely(ac->avail == ac->limit)) {
994
			STATS_INC_ACOVERFLOW(cachep);
995
			__drain_alien_cache(cachep, ac, page_node, &list);
996
		}
Joonsoo Kim's avatar
Joonsoo Kim committed
997
		ac_put_obj(cachep, ac, objp);
998
		spin_unlock(&alien->lock);
999
		slabs_destroy(cachep, &list);
1000
	} else {
1001
		n = get_node(cachep, page_node);
1002
		spin_lock(&n->list_lock);
1003
		free_block(cachep, &objp, 1, page_node, &list);
1004
		spin_unlock(&n->list_lock);
1005
		slabs_destroy(cachep, &list);
1006
1007
1008
	}
	return 1;
}
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022

static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
	int page_node = page_to_nid(virt_to_page(objp));
	int node = numa_mem_id();
	/*
	 * Make sure we are not freeing a object from another node to the array
	 * cache on this cpu.
	 */
	if (likely(node == page_node))
		return 0;

	return __cache_free_alien(cachep, objp, node, page_node);
}
David Rientjes's avatar
David Rientjes committed
1023
1024

/*
1025
1026
 * Construct gfp mask to allocate from a specific node but do not direct reclaim
 * or warn about failures. kswapd may still wake to reclaim in the background.
David Rientjes's avatar
David Rientjes committed
1027
1028
1029
 */
static inline gfp_t gfp_exact_node(gfp_t flags)
{
1030
	return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~__GFP_DIRECT_RECLAIM;
David Rientjes's avatar
David Rientjes committed
1031
}
1032
1033
#endif

1034
/*
1035
 * Allocates and initializes node for a node on each slab cache, used for
1036
 * either memory or cpu hotplug.  If memory is being hot-added, the kmem_cache_node
1037
 * will be allocated off-node since memory is not yet online for the new node.
1038
 * When hotplugging memory or a cpu, existing node are not replaced if
1039
1040
 * already in use.
 *
1041
 * Must hold slab_mutex.
1042
 */
1043
static int init_cache_node_node(int node)
1044
1045
{
	struct kmem_cache *cachep;
1046
	struct kmem_cache_node *n;
1047
	const size_t memsize = sizeof(struct kmem_cache_node);
1048

1049
	list_for_each_entry(cachep, &slab_caches, list) {
1050
		/*
1051
		 * Set up the kmem_cache_node for cpu before we can
1052
1053
1054
		 * begin anything. Make sure some other cpu on this
		 * node has not already allocated this
		 */
1055
1056
		n = get_node(cachep, node);
		if (!n) {
1057
1058
			n = kmalloc_node(memsize, GFP_KERNEL, node);
			if (!n)
1059
				return -ENOMEM;
1060
			kmem_cache_node_init(n);
1061
1062
			n->next_reap = jiffies + REAPTIMEOUT_NODE +
			    ((unsigned long)cachep) % REAPTIMEOUT_NODE;
1063
1064

			/*
1065
1066
			 * The kmem_cache_nodes don't come and go as CPUs
			 * come and go.  slab_mutex is sufficient
1067
1068
			 * protection here.
			 */
1069
			cachep->node[node] = n;
1070
1071
		}

1072
1073
		spin_lock_irq(&n->list_lock);
		n->free_limit =
1074
1075
			(1 + nr_cpus_node(node)) *
			cachep->batchcount + cachep->num;
1076
		spin_unlock_irq(&n->list_lock);
1077
1078
1079
1080
	}
	return 0;
}

1081
1082
1083
1084
1085
1086
static inline int slabs_tofree(struct kmem_cache *cachep,
						struct kmem_cache_node *n)
{
	return (n->free_objects + cachep->num - 1) / cachep->num;
}

1087
static void cpuup_canceled(long cpu)
1088
1089
{
	struct kmem_cache *cachep;
1090
	struct kmem_cache_node *n = NULL;
1091
	int node = cpu_to_mem(cpu);
1092
	const struct cpumask *mask = cpumask_of_node(node);
1093

1094
	list_for_each_entry(cachep, &slab_caches, list) {
1095
1096
		struct array_cache *nc;
		struct array_cache *shared;
Joonsoo Kim's avatar
Joonsoo Kim committed
1097
		struct alien_cache **alien;
1098
		LIST_HEAD(list);
1099

1100
		n = get_node(cachep, node);
1101
		if (!n)
1102
			continue;
1103

1104
		spin_lock_irq(&n->list_lock);
1105

1106
1107
		/* Free limit for this kmem_cache_node */
		n->free_limit -= cachep->batchcount;
1108
1109
1110
1111

		/* cpu is dead; no one can alloc from it. */
		nc = per_cpu_ptr(cachep->cpu_cache, cpu);
		if (nc) {
Joonsoo Kim's avatar