blob: 2eedacea439de698bdf2103f54ae936fe40de48d [file] [log] [blame]
Christoph Lameter97d06602012-07-06 15:25:11 -05001#ifndef MM_SLAB_H
2#define MM_SLAB_H
3/*
4 * Internal slab definitions
5 */
6
Joonsoo Kim07f361b2014-10-09 15:26:00 -07007#ifdef CONFIG_SLOB
8/*
9 * Common fields provided in kmem_cache by all slab allocators
10 * This struct is either used directly by the allocator (SLOB)
11 * or the allocator must include definitions for all fields
12 * provided in kmem_cache_common in their definition of kmem_cache.
13 *
14 * Once we can do anonymous structs (C11 standard) we could put a
15 * anonymous struct definition in these allocators so that the
16 * separate allocations in the kmem_cache structure of SLAB and
17 * SLUB is no longer needed.
18 */
19struct kmem_cache {
20 unsigned int object_size;/* The original size of the object */
21 unsigned int size; /* The aligned/padded/added on size */
22 unsigned int align; /* Alignment as calculated */
23 unsigned long flags; /* Active flags on the slab */
24 const char *name; /* Slab name for sysfs */
25 int refcount; /* Use counter */
26 void (*ctor)(void *); /* Called on object slot creation */
27 struct list_head list; /* List of all slab caches on the system */
28};
29
30#endif /* CONFIG_SLOB */
31
32#ifdef CONFIG_SLAB
33#include <linux/slab_def.h>
34#endif
35
36#ifdef CONFIG_SLUB
37#include <linux/slub_def.h>
38#endif
39
40#include <linux/memcontrol.h>
41
Christoph Lameter97d06602012-07-06 15:25:11 -050042/*
43 * State of the slab allocator.
44 *
45 * This is used to describe the states of the allocator during bootup.
46 * Allocators use this to gradually bootstrap themselves. Most allocators
47 * have the problem that the structures used for managing slab caches are
48 * allocated from slab caches themselves.
49 */
50enum slab_state {
51 DOWN, /* No slab functionality yet */
52 PARTIAL, /* SLUB: kmem_cache_node available */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +000053 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
Christoph Lameter97d06602012-07-06 15:25:11 -050054 UP, /* Slab caches usable but not all extras yet */
55 FULL /* Everything is working */
56};
57
58extern enum slab_state slab_state;
59
Christoph Lameter18004c52012-07-06 15:25:12 -050060/* The slab cache mutex protects the management structures during changes */
61extern struct mutex slab_mutex;
Christoph Lameter9b030cb2012-09-05 00:20:33 +000062
63/* The list of all slab caches on the system */
Christoph Lameter18004c52012-07-06 15:25:12 -050064extern struct list_head slab_caches;
65
Christoph Lameter9b030cb2012-09-05 00:20:33 +000066/* The slab cache that manages slab cache information */
67extern struct kmem_cache *kmem_cache;
68
Christoph Lameter45906852012-11-28 16:23:16 +000069unsigned long calculate_alignment(unsigned long flags,
70 unsigned long align, unsigned long size);
71
Christoph Lameterf97d5f62013-01-10 19:12:17 +000072#ifndef CONFIG_SLOB
73/* Kmalloc array related functions */
Daniel Sanders34cc6992015-06-24 16:55:57 -070074void setup_kmalloc_cache_index_table(void);
Christoph Lameterf97d5f62013-01-10 19:12:17 +000075void create_kmalloc_caches(unsigned long);
Christoph Lameter2c59dd62013-01-10 19:14:19 +000076
77/* Find the kmalloc slab corresponding for a certain size */
78struct kmem_cache *kmalloc_slab(size_t, gfp_t);
Christoph Lameterf97d5f62013-01-10 19:12:17 +000079#endif
80
81
Christoph Lameter9b030cb2012-09-05 00:20:33 +000082/* Functions provided by the slab allocators */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +000083extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
Christoph Lameter97d06602012-07-06 15:25:11 -050084
Christoph Lameter45530c42012-11-28 16:23:07 +000085extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
86 unsigned long flags);
87extern void create_boot_cache(struct kmem_cache *, const char *name,
88 size_t size, unsigned long flags);
89
Joonsoo Kim423c9292014-10-09 15:26:22 -070090int slab_unmergeable(struct kmem_cache *s);
91struct kmem_cache *find_mergeable(size_t size, size_t align,
92 unsigned long flags, const char *name, void (*ctor)(void *));
Joonsoo Kim12220de2014-10-09 15:26:24 -070093#ifndef CONFIG_SLOB
Glauber Costa2633d7a2012-12-18 14:22:34 -080094struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -070095__kmem_cache_alias(const char *name, size_t size, size_t align,
96 unsigned long flags, void (*ctor)(void *));
Joonsoo Kim423c9292014-10-09 15:26:22 -070097
98unsigned long kmem_cache_flags(unsigned long object_size,
99 unsigned long flags, const char *name,
100 void (*ctor)(void *));
Christoph Lametercbb79692012-09-05 00:18:32 +0000101#else
Glauber Costa2633d7a2012-12-18 14:22:34 -0800102static inline struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -0700103__kmem_cache_alias(const char *name, size_t size, size_t align,
104 unsigned long flags, void (*ctor)(void *))
Christoph Lametercbb79692012-09-05 00:18:32 +0000105{ return NULL; }
Joonsoo Kim423c9292014-10-09 15:26:22 -0700106
107static inline unsigned long kmem_cache_flags(unsigned long object_size,
108 unsigned long flags, const char *name,
109 void (*ctor)(void *))
110{
111 return flags;
112}
Christoph Lametercbb79692012-09-05 00:18:32 +0000113#endif
114
115
Glauber Costad8843922012-10-17 15:36:51 +0400116/* Legal flag mask for kmem_cache_create(), for various configurations */
117#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
118 SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
119
120#if defined(CONFIG_DEBUG_SLAB)
121#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
122#elif defined(CONFIG_SLUB_DEBUG)
123#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
124 SLAB_TRACE | SLAB_DEBUG_FREE)
125#else
126#define SLAB_DEBUG_FLAGS (0)
127#endif
128
129#if defined(CONFIG_SLAB)
130#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800131 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
132 SLAB_NOTRACK | SLAB_ACCOUNT)
Glauber Costad8843922012-10-17 15:36:51 +0400133#elif defined(CONFIG_SLUB)
134#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
Vladimir Davydov230e9fc2016-01-14 15:18:15 -0800135 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
Glauber Costad8843922012-10-17 15:36:51 +0400136#else
137#define SLAB_CACHE_FLAGS (0)
138#endif
139
140#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
141
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000142int __kmem_cache_shutdown(struct kmem_cache *);
Dmitry Safonov52b4b952016-02-17 13:11:37 -0800143void __kmem_cache_release(struct kmem_cache *);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -0800144int __kmem_cache_shrink(struct kmem_cache *, bool);
Christoph Lameter41a21282014-05-06 12:50:08 -0700145void slab_kmem_cache_release(struct kmem_cache *);
Christoph Lameter945cf2b2012-09-04 23:18:33 +0000146
Glauber Costab7454ad2012-10-19 18:20:25 +0400147struct seq_file;
148struct file;
Glauber Costab7454ad2012-10-19 18:20:25 +0400149
Glauber Costa0d7561c2012-10-19 18:20:27 +0400150struct slabinfo {
151 unsigned long active_objs;
152 unsigned long num_objs;
153 unsigned long active_slabs;
154 unsigned long num_slabs;
155 unsigned long shared_avail;
156 unsigned int limit;
157 unsigned int batchcount;
158 unsigned int shared;
159 unsigned int objects_per_slab;
160 unsigned int cache_order;
161};
162
163void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
164void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
Glauber Costab7454ad2012-10-19 18:20:25 +0400165ssize_t slabinfo_write(struct file *file, const char __user *buffer,
166 size_t count, loff_t *ppos);
Glauber Costaba6c4962012-12-18 14:22:27 -0800167
Christoph Lameter484748f2015-09-04 15:45:34 -0700168/*
169 * Generic implementation of bulk operations
170 * These are useful for situations in which the allocator cannot
171 * perform optimizations. In that case segments of the objecct listed
172 * may be allocated or freed using these operations.
173 */
174void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -0800175int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
Christoph Lameter484748f2015-09-04 15:45:34 -0700176
Johannes Weiner127424c2016-01-20 15:02:32 -0800177#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
Vladimir Davydov426589f2015-02-12 14:59:23 -0800178/*
179 * Iterate over all memcg caches of the given root cache. The caller must hold
180 * slab_mutex.
181 */
182#define for_each_memcg_cache(iter, root) \
183 list_for_each_entry(iter, &(root)->memcg_params.list, \
184 memcg_params.list)
185
Glauber Costaba6c4962012-12-18 14:22:27 -0800186static inline bool is_root_cache(struct kmem_cache *s)
187{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800188 return s->memcg_params.is_root_cache;
Glauber Costaba6c4962012-12-18 14:22:27 -0800189}
Glauber Costa2633d7a2012-12-18 14:22:34 -0800190
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800191static inline bool slab_equal_or_root(struct kmem_cache *s,
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800192 struct kmem_cache *p)
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800193{
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800194 return p == s || p == s->memcg_params.root_cache;
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800195}
Glauber Costa749c5412012-12-18 14:23:01 -0800196
197/*
198 * We use suffixes to the name in memcg because we can't have caches
199 * created in the system with the same name. But when we print them
200 * locally, better refer to them with the base name
201 */
202static inline const char *cache_name(struct kmem_cache *s)
203{
204 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800205 s = s->memcg_params.root_cache;
Glauber Costa749c5412012-12-18 14:23:01 -0800206 return s->name;
207}
208
Vladimir Davydovf8570262014-01-23 15:53:06 -0800209/*
210 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800211 * That said the caller must assure the memcg's cache won't go away by either
212 * taking a css reference to the owner cgroup, or holding the slab_mutex.
Vladimir Davydovf8570262014-01-23 15:53:06 -0800213 */
Qiang Huang2ade4de2013-11-12 15:08:23 -0800214static inline struct kmem_cache *
215cache_from_memcg_idx(struct kmem_cache *s, int idx)
Glauber Costa749c5412012-12-18 14:23:01 -0800216{
Vladimir Davydov959c8962014-01-23 15:52:59 -0800217 struct kmem_cache *cachep;
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800218 struct memcg_cache_array *arr;
Vladimir Davydovf8570262014-01-23 15:53:06 -0800219
220 rcu_read_lock();
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800221 arr = rcu_dereference(s->memcg_params.memcg_caches);
Vladimir Davydov959c8962014-01-23 15:52:59 -0800222
223 /*
224 * Make sure we will access the up-to-date value. The code updating
225 * memcg_caches issues a write barrier to match this (see
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800226 * memcg_create_kmem_cache()).
Vladimir Davydov959c8962014-01-23 15:52:59 -0800227 */
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800228 cachep = lockless_dereference(arr->entries[idx]);
Pranith Kumar8df0c2d2014-12-10 15:42:28 -0800229 rcu_read_unlock();
230
Vladimir Davydov959c8962014-01-23 15:52:59 -0800231 return cachep;
Glauber Costa749c5412012-12-18 14:23:01 -0800232}
Glauber Costa943a4512012-12-18 14:23:03 -0800233
234static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
235{
236 if (is_root_cache(s))
237 return s;
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800238 return s->memcg_params.root_cache;
Glauber Costa943a4512012-12-18 14:23:03 -0800239}
Vladimir Davydov5dfb4172014-06-04 16:06:38 -0700240
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -0800241static __always_inline int memcg_charge_slab(struct page *page,
242 gfp_t gfp, int order,
243 struct kmem_cache *s)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -0700244{
245 if (!memcg_kmem_enabled())
246 return 0;
247 if (is_root_cache(s))
248 return 0;
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -0800249 return __memcg_kmem_charge_memcg(page, gfp, order,
250 s->memcg_params.memcg);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -0700251}
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800252
253extern void slab_init_memcg_params(struct kmem_cache *);
254
Johannes Weiner127424c2016-01-20 15:02:32 -0800255#else /* CONFIG_MEMCG && !CONFIG_SLOB */
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800256
Vladimir Davydov426589f2015-02-12 14:59:23 -0800257#define for_each_memcg_cache(iter, root) \
258 for ((void)(iter), (void)(root); 0; )
Vladimir Davydov426589f2015-02-12 14:59:23 -0800259
Glauber Costaba6c4962012-12-18 14:22:27 -0800260static inline bool is_root_cache(struct kmem_cache *s)
261{
262 return true;
263}
264
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800265static inline bool slab_equal_or_root(struct kmem_cache *s,
266 struct kmem_cache *p)
267{
268 return true;
269}
Glauber Costa749c5412012-12-18 14:23:01 -0800270
271static inline const char *cache_name(struct kmem_cache *s)
272{
273 return s->name;
274}
275
Qiang Huang2ade4de2013-11-12 15:08:23 -0800276static inline struct kmem_cache *
277cache_from_memcg_idx(struct kmem_cache *s, int idx)
Glauber Costa749c5412012-12-18 14:23:01 -0800278{
279 return NULL;
280}
Glauber Costa943a4512012-12-18 14:23:03 -0800281
282static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
283{
284 return s;
285}
Vladimir Davydov5dfb4172014-06-04 16:06:38 -0700286
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -0800287static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
288 struct kmem_cache *s)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -0700289{
290 return 0;
291}
292
Vladimir Davydovf7ce3192015-02-12 14:59:20 -0800293static inline void slab_init_memcg_params(struct kmem_cache *s)
294{
295}
Johannes Weiner127424c2016-01-20 15:02:32 -0800296#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800297
298static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
299{
300 struct kmem_cache *cachep;
301 struct page *page;
302
303 /*
304 * When kmemcg is not being used, both assignments should return the
305 * same value. but we don't want to pay the assignment price in that
306 * case. If it is not compiled in, the compiler should be smart enough
307 * to not do even the assignment. In that case, slab_equal_or_root
308 * will also be a constant.
309 */
310 if (!memcg_kmem_enabled() && !unlikely(s->flags & SLAB_DEBUG_FREE))
311 return s;
312
313 page = virt_to_head_page(x);
314 cachep = page->slab_cache;
315 if (slab_equal_or_root(cachep, s))
316 return cachep;
317
318 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
Daniel Borkmann2d16e0fd2015-09-04 15:45:57 -0700319 __func__, s->name, cachep->name);
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800320 WARN_ON_ONCE(1);
321 return s;
322}
Christoph Lameterca349562013-01-10 19:14:19 +0000323
Christoph Lameter44c53562014-08-06 16:04:07 -0700324#ifndef CONFIG_SLOB
Christoph Lameterca349562013-01-10 19:14:19 +0000325/*
326 * The slab lists for all objects.
327 */
328struct kmem_cache_node {
329 spinlock_t list_lock;
330
331#ifdef CONFIG_SLAB
332 struct list_head slabs_partial; /* partial list first, better asm code */
333 struct list_head slabs_full;
334 struct list_head slabs_free;
335 unsigned long free_objects;
336 unsigned int free_limit;
337 unsigned int colour_next; /* Per-node cache coloring */
338 struct array_cache *shared; /* shared per node */
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700339 struct alien_cache **alien; /* on other nodes */
Christoph Lameterca349562013-01-10 19:14:19 +0000340 unsigned long next_reap; /* updated without locking */
341 int free_touched; /* updated without locking */
342#endif
343
344#ifdef CONFIG_SLUB
345 unsigned long nr_partial;
346 struct list_head partial;
347#ifdef CONFIG_SLUB_DEBUG
348 atomic_long_t nr_slabs;
349 atomic_long_t total_objects;
350 struct list_head full;
351#endif
352#endif
353
354};
Wanpeng Lie25839f2013-07-04 08:33:23 +0800355
Christoph Lameter44c53562014-08-06 16:04:07 -0700356static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
357{
358 return s->node[node];
359}
360
361/*
362 * Iterator over all nodes. The body will be executed for each node that has
363 * a kmem_cache_node structure allocated (which is true for all online nodes)
364 */
365#define for_each_kmem_cache_node(__s, __node, __n) \
Mikulas Patocka91635822014-10-09 15:26:20 -0700366 for (__node = 0; __node < nr_node_ids; __node++) \
367 if ((__n = get_node(__s, __node)))
Christoph Lameter44c53562014-08-06 16:04:07 -0700368
369#endif
370
Vladimir Davydov1df3b262014-12-10 15:42:16 -0800371void *slab_start(struct seq_file *m, loff_t *pos);
Wanpeng Li276a2432013-07-08 08:08:28 +0800372void *slab_next(struct seq_file *m, void *p, loff_t *pos);
373void slab_stop(struct seq_file *m, void *p);
Vladimir Davydovb0475012014-12-10 15:44:19 -0800374int memcg_slab_show(struct seq_file *m, void *p);
Andrey Ryabinin5240ab42014-08-06 16:04:14 -0700375
376#endif /* MM_SLAB_H */