]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - mm/slab.h
mm: memcg/slab: synchronize access to kmem_cache dying flag using a spinlock
[mirror_ubuntu-hirsute-kernel.git] / mm / slab.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
97d06609
CL
2#ifndef MM_SLAB_H
3#define MM_SLAB_H
4/*
5 * Internal slab definitions
6 */
7
07f361b2
JK
8#ifdef CONFIG_SLOB
9/*
10 * Common fields provided in kmem_cache by all slab allocators
11 * This struct is either used directly by the allocator (SLOB)
12 * or the allocator must include definitions for all fields
13 * provided in kmem_cache_common in their definition of kmem_cache.
14 *
15 * Once we can do anonymous structs (C11 standard) we could put a
16 * anonymous struct definition in these allocators so that the
17 * separate allocations in the kmem_cache structure of SLAB and
18 * SLUB is no longer needed.
19 */
20struct kmem_cache {
21 unsigned int object_size;/* The original size of the object */
22 unsigned int size; /* The aligned/padded/added on size */
23 unsigned int align; /* Alignment as calculated */
d50112ed 24 slab_flags_t flags; /* Active flags on the slab */
7bbdb81e
AD
25 unsigned int useroffset;/* Usercopy region offset */
26 unsigned int usersize; /* Usercopy region size */
07f361b2
JK
27 const char *name; /* Slab name for sysfs */
28 int refcount; /* Use counter */
29 void (*ctor)(void *); /* Called on object slot creation */
30 struct list_head list; /* List of all slab caches on the system */
31};
32
33#endif /* CONFIG_SLOB */
34
35#ifdef CONFIG_SLAB
36#include <linux/slab_def.h>
37#endif
38
39#ifdef CONFIG_SLUB
40#include <linux/slub_def.h>
41#endif
42
43#include <linux/memcontrol.h>
11c7aec2 44#include <linux/fault-inject.h>
11c7aec2
JDB
45#include <linux/kasan.h>
46#include <linux/kmemleak.h>
7c00fce9 47#include <linux/random.h>
d92a8cfc 48#include <linux/sched/mm.h>
07f361b2 49
97d06609
CL
50/*
51 * State of the slab allocator.
52 *
53 * This is used to describe the states of the allocator during bootup.
54 * Allocators use this to gradually bootstrap themselves. Most allocators
55 * have the problem that the structures used for managing slab caches are
56 * allocated from slab caches themselves.
57 */
58enum slab_state {
59 DOWN, /* No slab functionality yet */
60 PARTIAL, /* SLUB: kmem_cache_node available */
ce8eb6c4 61 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
97d06609
CL
62 UP, /* Slab caches usable but not all extras yet */
63 FULL /* Everything is working */
64};
65
66extern enum slab_state slab_state;
67
18004c5d
CL
68/* The slab cache mutex protects the management structures during changes */
69extern struct mutex slab_mutex;
9b030cb8
CL
70
71/* The list of all slab caches on the system */
18004c5d
CL
72extern struct list_head slab_caches;
73
9b030cb8
CL
74/* The slab cache that manages slab cache information */
75extern struct kmem_cache *kmem_cache;
76
af3b5f87
VB
77/* A table of kmalloc cache names and sizes */
78extern const struct kmalloc_info_struct {
79 const char *name;
55de8b9c 80 unsigned int size;
af3b5f87
VB
81} kmalloc_info[];
82
f97d5f63
CL
83#ifndef CONFIG_SLOB
84/* Kmalloc array related functions */
34cc6990 85void setup_kmalloc_cache_index_table(void);
d50112ed 86void create_kmalloc_caches(slab_flags_t);
2c59dd65
CL
87
88/* Find the kmalloc slab corresponding for a certain size */
89struct kmem_cache *kmalloc_slab(size_t, gfp_t);
f97d5f63
CL
90#endif
91
92
9b030cb8 93/* Functions provided by the slab allocators */
d50112ed 94int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
97d06609 95
55de8b9c
AD
96struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
97 slab_flags_t flags, unsigned int useroffset,
98 unsigned int usersize);
45530c44 99extern void create_boot_cache(struct kmem_cache *, const char *name,
361d575e
AD
100 unsigned int size, slab_flags_t flags,
101 unsigned int useroffset, unsigned int usersize);
45530c44 102
423c929c 103int slab_unmergeable(struct kmem_cache *s);
f4957d5b 104struct kmem_cache *find_mergeable(unsigned size, unsigned align,
d50112ed 105 slab_flags_t flags, const char *name, void (*ctor)(void *));
12220dea 106#ifndef CONFIG_SLOB
2633d7a0 107struct kmem_cache *
f4957d5b 108__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 109 slab_flags_t flags, void (*ctor)(void *));
423c929c 110
0293d1fd 111slab_flags_t kmem_cache_flags(unsigned int object_size,
d50112ed 112 slab_flags_t flags, const char *name,
423c929c 113 void (*ctor)(void *));
cbb79694 114#else
2633d7a0 115static inline struct kmem_cache *
f4957d5b 116__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 117 slab_flags_t flags, void (*ctor)(void *))
cbb79694 118{ return NULL; }
423c929c 119
0293d1fd 120static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
d50112ed 121 slab_flags_t flags, const char *name,
423c929c
JK
122 void (*ctor)(void *))
123{
124 return flags;
125}
cbb79694
CL
126#endif
127
128
d8843922 129/* Legal flag mask for kmem_cache_create(), for various configurations */
6d6ea1e9
NB
130#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
131 SLAB_CACHE_DMA32 | SLAB_PANIC | \
5f0d5a3a 132 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
d8843922
GC
133
134#if defined(CONFIG_DEBUG_SLAB)
135#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
136#elif defined(CONFIG_SLUB_DEBUG)
137#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
becfda68 138 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
d8843922
GC
139#else
140#define SLAB_DEBUG_FLAGS (0)
141#endif
142
143#if defined(CONFIG_SLAB)
144#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
230e9fc2 145 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
75f296d9 146 SLAB_ACCOUNT)
d8843922
GC
147#elif defined(CONFIG_SLUB)
148#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
75f296d9 149 SLAB_TEMPORARY | SLAB_ACCOUNT)
d8843922
GC
150#else
151#define SLAB_CACHE_FLAGS (0)
152#endif
153
e70954fd 154/* Common flags available with current configuration */
d8843922
GC
155#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
156
e70954fd
TG
157/* Common flags permitted for kmem_cache_create */
158#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
159 SLAB_RED_ZONE | \
160 SLAB_POISON | \
161 SLAB_STORE_USER | \
162 SLAB_TRACE | \
163 SLAB_CONSISTENCY_CHECKS | \
164 SLAB_MEM_SPREAD | \
165 SLAB_NOLEAKTRACE | \
166 SLAB_RECLAIM_ACCOUNT | \
167 SLAB_TEMPORARY | \
e70954fd
TG
168 SLAB_ACCOUNT)
169
f9e13c0a 170bool __kmem_cache_empty(struct kmem_cache *);
945cf2b6 171int __kmem_cache_shutdown(struct kmem_cache *);
52b4b950 172void __kmem_cache_release(struct kmem_cache *);
c9fc5864
TH
173int __kmem_cache_shrink(struct kmem_cache *);
174void __kmemcg_cache_deactivate(struct kmem_cache *s);
43486694 175void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s);
41a21285 176void slab_kmem_cache_release(struct kmem_cache *);
945cf2b6 177
b7454ad3
GC
178struct seq_file;
179struct file;
b7454ad3 180
0d7561c6
GC
181struct slabinfo {
182 unsigned long active_objs;
183 unsigned long num_objs;
184 unsigned long active_slabs;
185 unsigned long num_slabs;
186 unsigned long shared_avail;
187 unsigned int limit;
188 unsigned int batchcount;
189 unsigned int shared;
190 unsigned int objects_per_slab;
191 unsigned int cache_order;
192};
193
194void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
195void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
b7454ad3
GC
196ssize_t slabinfo_write(struct file *file, const char __user *buffer,
197 size_t count, loff_t *ppos);
ba6c496e 198
484748f0
CL
199/*
200 * Generic implementation of bulk operations
201 * These are useful for situations in which the allocator cannot
9f706d68 202 * perform optimizations. In that case segments of the object listed
484748f0
CL
203 * may be allocated or freed using these operations.
204 */
205void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
865762a8 206int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
484748f0 207
6cea1d56
RG
208static inline int cache_vmstat_idx(struct kmem_cache *s)
209{
210 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
211 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE;
212}
213
84c07d11 214#ifdef CONFIG_MEMCG_KMEM
510ded33
TH
215
216/* List of all root caches. */
217extern struct list_head slab_root_caches;
218#define root_caches_node memcg_params.__root_caches_node
219
426589f5
VD
220/*
221 * Iterate over all memcg caches of the given root cache. The caller must hold
222 * slab_mutex.
223 */
224#define for_each_memcg_cache(iter, root) \
9eeadc8b
TH
225 list_for_each_entry(iter, &(root)->memcg_params.children, \
226 memcg_params.children_node)
426589f5 227
ba6c496e
GC
228static inline bool is_root_cache(struct kmem_cache *s)
229{
9eeadc8b 230 return !s->memcg_params.root_cache;
ba6c496e 231}
2633d7a0 232
b9ce5ef4 233static inline bool slab_equal_or_root(struct kmem_cache *s,
f7ce3190 234 struct kmem_cache *p)
b9ce5ef4 235{
f7ce3190 236 return p == s || p == s->memcg_params.root_cache;
b9ce5ef4 237}
749c5415
GC
238
239/*
240 * We use suffixes to the name in memcg because we can't have caches
241 * created in the system with the same name. But when we print them
242 * locally, better refer to them with the base name
243 */
244static inline const char *cache_name(struct kmem_cache *s)
245{
246 if (!is_root_cache(s))
f7ce3190 247 s = s->memcg_params.root_cache;
749c5415
GC
248 return s->name;
249}
250
f8570263
VD
251/*
252 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
f7ce3190
VD
253 * That said the caller must assure the memcg's cache won't go away by either
254 * taking a css reference to the owner cgroup, or holding the slab_mutex.
f8570263 255 */
2ade4de8
QH
256static inline struct kmem_cache *
257cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415 258{
959c8963 259 struct kmem_cache *cachep;
f7ce3190 260 struct memcg_cache_array *arr;
f8570263
VD
261
262 rcu_read_lock();
f7ce3190 263 arr = rcu_dereference(s->memcg_params.memcg_caches);
959c8963
VD
264
265 /*
266 * Make sure we will access the up-to-date value. The code updating
267 * memcg_caches issues a write barrier to match this (see
f7ce3190 268 * memcg_create_kmem_cache()).
959c8963 269 */
506458ef 270 cachep = READ_ONCE(arr->entries[idx]);
8df0c2dc
PK
271 rcu_read_unlock();
272
959c8963 273 return cachep;
749c5415 274}
943a451a
GC
275
276static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
277{
278 if (is_root_cache(s))
279 return s;
f7ce3190 280 return s->memcg_params.root_cache;
943a451a 281}
5dfb4175 282
f3ccb2c4
VD
283static __always_inline int memcg_charge_slab(struct page *page,
284 gfp_t gfp, int order,
285 struct kmem_cache *s)
5dfb4175 286{
5dfb4175
VD
287 if (is_root_cache(s))
288 return 0;
7779f212 289 return memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
27ee57c9
VD
290}
291
292static __always_inline void memcg_uncharge_slab(struct page *page, int order,
293 struct kmem_cache *s)
294{
27ee57c9 295 memcg_kmem_uncharge(page, order);
5dfb4175 296}
f7ce3190
VD
297
298extern void slab_init_memcg_params(struct kmem_cache *);
c03914b7 299extern void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg);
f7ce3190 300
84c07d11 301#else /* CONFIG_MEMCG_KMEM */
f7ce3190 302
510ded33
TH
303/* If !memcg, all caches are root. */
304#define slab_root_caches slab_caches
305#define root_caches_node list
306
426589f5
VD
307#define for_each_memcg_cache(iter, root) \
308 for ((void)(iter), (void)(root); 0; )
426589f5 309
ba6c496e
GC
310static inline bool is_root_cache(struct kmem_cache *s)
311{
312 return true;
313}
314
b9ce5ef4
GC
315static inline bool slab_equal_or_root(struct kmem_cache *s,
316 struct kmem_cache *p)
317{
598a0717 318 return s == p;
b9ce5ef4 319}
749c5415
GC
320
321static inline const char *cache_name(struct kmem_cache *s)
322{
323 return s->name;
324}
325
2ade4de8
QH
326static inline struct kmem_cache *
327cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415
GC
328{
329 return NULL;
330}
943a451a
GC
331
332static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
333{
334 return s;
335}
5dfb4175 336
f3ccb2c4
VD
337static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
338 struct kmem_cache *s)
5dfb4175
VD
339{
340 return 0;
341}
342
27ee57c9
VD
343static inline void memcg_uncharge_slab(struct page *page, int order,
344 struct kmem_cache *s)
345{
346}
347
f7ce3190
VD
348static inline void slab_init_memcg_params(struct kmem_cache *s)
349{
350}
510ded33 351
c03914b7
RG
352static inline void memcg_link_cache(struct kmem_cache *s,
353 struct mem_cgroup *memcg)
510ded33
TH
354{
355}
356
84c07d11 357#endif /* CONFIG_MEMCG_KMEM */
b9ce5ef4 358
a64b5378
KC
359static inline struct kmem_cache *virt_to_cache(const void *obj)
360{
361 struct page *page;
362
363 page = virt_to_head_page(obj);
364 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
365 __func__))
366 return NULL;
367 return page->slab_cache;
368}
369
6cea1d56
RG
370static __always_inline int charge_slab_page(struct page *page,
371 gfp_t gfp, int order,
372 struct kmem_cache *s)
373{
374 int ret = memcg_charge_slab(page, gfp, order, s);
375
376 if (!ret)
377 mod_lruvec_page_state(page, cache_vmstat_idx(s), 1 << order);
378
379 return ret;
380}
381
382static __always_inline void uncharge_slab_page(struct page *page, int order,
383 struct kmem_cache *s)
384{
385 mod_lruvec_page_state(page, cache_vmstat_idx(s), -(1 << order));
386 memcg_uncharge_slab(page, order, s);
387}
388
b9ce5ef4
GC
389static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
390{
391 struct kmem_cache *cachep;
b9ce5ef4
GC
392
393 /*
394 * When kmemcg is not being used, both assignments should return the
395 * same value. but we don't want to pay the assignment price in that
396 * case. If it is not compiled in, the compiler should be smart enough
397 * to not do even the assignment. In that case, slab_equal_or_root
398 * will also be a constant.
399 */
becfda68 400 if (!memcg_kmem_enabled() &&
598a0717 401 !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
becfda68 402 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
b9ce5ef4
GC
403 return s;
404
a64b5378
KC
405 cachep = virt_to_cache(x);
406 WARN_ONCE(cachep && !slab_equal_or_root(cachep, s),
598a0717
KC
407 "%s: Wrong slab cache. %s but object is from %s\n",
408 __func__, s->name, cachep->name);
409 return cachep;
b9ce5ef4 410}
ca34956b 411
11c7aec2
JDB
412static inline size_t slab_ksize(const struct kmem_cache *s)
413{
414#ifndef CONFIG_SLUB
415 return s->object_size;
416
417#else /* CONFIG_SLUB */
418# ifdef CONFIG_SLUB_DEBUG
419 /*
420 * Debugging requires use of the padding between object
421 * and whatever may come after it.
422 */
423 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
424 return s->object_size;
425# endif
80a9201a
AP
426 if (s->flags & SLAB_KASAN)
427 return s->object_size;
11c7aec2
JDB
428 /*
429 * If we have the need to store the freelist pointer
430 * back there or track user information then we can
431 * only use the space before that information.
432 */
5f0d5a3a 433 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
11c7aec2
JDB
434 return s->inuse;
435 /*
436 * Else we can use all the padding etc for the allocation
437 */
438 return s->size;
439#endif
440}
441
442static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
443 gfp_t flags)
444{
445 flags &= gfp_allowed_mask;
d92a8cfc
PZ
446
447 fs_reclaim_acquire(flags);
448 fs_reclaim_release(flags);
449
11c7aec2
JDB
450 might_sleep_if(gfpflags_allow_blocking(flags));
451
fab9963a 452 if (should_failslab(s, flags))
11c7aec2
JDB
453 return NULL;
454
45264778
VD
455 if (memcg_kmem_enabled() &&
456 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
457 return memcg_kmem_get_cache(s);
458
459 return s;
11c7aec2
JDB
460}
461
462static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
463 size_t size, void **p)
464{
465 size_t i;
466
467 flags &= gfp_allowed_mask;
468 for (i = 0; i < size; i++) {
53128245 469 p[i] = kasan_slab_alloc(s, p[i], flags);
a2f77575 470 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
53128245 471 kmemleak_alloc_recursive(p[i], s->object_size, 1,
11c7aec2 472 s->flags, flags);
11c7aec2 473 }
45264778
VD
474
475 if (memcg_kmem_enabled())
476 memcg_kmem_put_cache(s);
11c7aec2
JDB
477}
478
44c5356f 479#ifndef CONFIG_SLOB
ca34956b
CL
480/*
481 * The slab lists for all objects.
482 */
483struct kmem_cache_node {
484 spinlock_t list_lock;
485
486#ifdef CONFIG_SLAB
487 struct list_head slabs_partial; /* partial list first, better asm code */
488 struct list_head slabs_full;
489 struct list_head slabs_free;
bf00bd34
DR
490 unsigned long total_slabs; /* length of all slab lists */
491 unsigned long free_slabs; /* length of free slab list only */
ca34956b
CL
492 unsigned long free_objects;
493 unsigned int free_limit;
494 unsigned int colour_next; /* Per-node cache coloring */
495 struct array_cache *shared; /* shared per node */
c8522a3a 496 struct alien_cache **alien; /* on other nodes */
ca34956b
CL
497 unsigned long next_reap; /* updated without locking */
498 int free_touched; /* updated without locking */
499#endif
500
501#ifdef CONFIG_SLUB
502 unsigned long nr_partial;
503 struct list_head partial;
504#ifdef CONFIG_SLUB_DEBUG
505 atomic_long_t nr_slabs;
506 atomic_long_t total_objects;
507 struct list_head full;
508#endif
509#endif
510
511};
e25839f6 512
44c5356f
CL
513static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
514{
515 return s->node[node];
516}
517
518/*
519 * Iterator over all nodes. The body will be executed for each node that has
520 * a kmem_cache_node structure allocated (which is true for all online nodes)
521 */
522#define for_each_kmem_cache_node(__s, __node, __n) \
9163582c
MP
523 for (__node = 0; __node < nr_node_ids; __node++) \
524 if ((__n = get_node(__s, __node)))
44c5356f
CL
525
526#endif
527
1df3b26f 528void *slab_start(struct seq_file *m, loff_t *pos);
276a2439
WL
529void *slab_next(struct seq_file *m, void *p, loff_t *pos);
530void slab_stop(struct seq_file *m, void *p);
bc2791f8
TH
531void *memcg_slab_start(struct seq_file *m, loff_t *pos);
532void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
533void memcg_slab_stop(struct seq_file *m, void *p);
b047501c 534int memcg_slab_show(struct seq_file *m, void *p);
5240ab40 535
852d8be0
YS
536#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
537void dump_unreclaimable_slab(void);
538#else
539static inline void dump_unreclaimable_slab(void)
540{
541}
542#endif
543
55834c59
AP
544void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
545
7c00fce9
TG
546#ifdef CONFIG_SLAB_FREELIST_RANDOM
547int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
548 gfp_t gfp);
549void cache_random_seq_destroy(struct kmem_cache *cachep);
550#else
551static inline int cache_random_seq_create(struct kmem_cache *cachep,
552 unsigned int count, gfp_t gfp)
553{
554 return 0;
555}
556static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
557#endif /* CONFIG_SLAB_FREELIST_RANDOM */
558
5240ab40 559#endif /* MM_SLAB_H */