]>
Commit | Line | Data |
---|---|---|
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 | ||
d122019b MWO |
8 | /* Reuses the bits in struct page */ |
9 | struct slab { | |
10 | unsigned long __page_flags; | |
401fb12c VB |
11 | |
12 | #if defined(CONFIG_SLAB) | |
13 | ||
d122019b MWO |
14 | union { |
15 | struct list_head slab_list; | |
401fb12c VB |
16 | struct rcu_head rcu_head; |
17 | }; | |
18 | struct kmem_cache *slab_cache; | |
19 | void *freelist; /* array of free object indexes */ | |
20 | void *s_mem; /* first object */ | |
21 | unsigned int active; | |
22 | ||
23 | #elif defined(CONFIG_SLUB) | |
24 | ||
25 | union { | |
26 | struct list_head slab_list; | |
27 | struct rcu_head rcu_head; | |
9c01e9af | 28 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
401fb12c | 29 | struct { |
d122019b | 30 | struct slab *next; |
d122019b | 31 | int slabs; /* Nr of slabs left */ |
d122019b | 32 | }; |
9c01e9af | 33 | #endif |
d122019b | 34 | }; |
401fb12c | 35 | struct kmem_cache *slab_cache; |
d122019b MWO |
36 | /* Double-word boundary */ |
37 | void *freelist; /* first free object */ | |
38 | union { | |
401fb12c VB |
39 | unsigned long counters; |
40 | struct { | |
d122019b MWO |
41 | unsigned inuse:16; |
42 | unsigned objects:15; | |
43 | unsigned frozen:1; | |
44 | }; | |
45 | }; | |
401fb12c VB |
46 | unsigned int __unused; |
47 | ||
48 | #elif defined(CONFIG_SLOB) | |
49 | ||
50 | struct list_head slab_list; | |
51 | void *__unused_1; | |
52 | void *freelist; /* first free block */ | |
b01af5c0 HY |
53 | long units; |
54 | unsigned int __unused_2; | |
401fb12c VB |
55 | |
56 | #else | |
57 | #error "Unexpected slab allocator configured" | |
58 | #endif | |
d122019b | 59 | |
d122019b MWO |
60 | atomic_t __page_refcount; |
61 | #ifdef CONFIG_MEMCG | |
62 | unsigned long memcg_data; | |
63 | #endif | |
64 | }; | |
65 | ||
66 | #define SLAB_MATCH(pg, sl) \ | |
67 | static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl)) | |
68 | SLAB_MATCH(flags, __page_flags); | |
69 | SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */ | |
401fb12c | 70 | #ifndef CONFIG_SLOB |
d122019b | 71 | SLAB_MATCH(rcu_head, rcu_head); |
401fb12c | 72 | #endif |
d122019b MWO |
73 | SLAB_MATCH(_refcount, __page_refcount); |
74 | #ifdef CONFIG_MEMCG | |
75 | SLAB_MATCH(memcg_data, memcg_data); | |
76 | #endif | |
77 | #undef SLAB_MATCH | |
78 | static_assert(sizeof(struct slab) <= sizeof(struct page)); | |
79 | ||
80 | /** | |
81 | * folio_slab - Converts from folio to slab. | |
82 | * @folio: The folio. | |
83 | * | |
84 | * Currently struct slab is a different representation of a folio where | |
85 | * folio_test_slab() is true. | |
86 | * | |
87 | * Return: The slab which contains this folio. | |
88 | */ | |
89 | #define folio_slab(folio) (_Generic((folio), \ | |
90 | const struct folio *: (const struct slab *)(folio), \ | |
91 | struct folio *: (struct slab *)(folio))) | |
92 | ||
93 | /** | |
94 | * slab_folio - The folio allocated for a slab | |
95 | * @slab: The slab. | |
96 | * | |
97 | * Slabs are allocated as folios that contain the individual objects and are | |
98 | * using some fields in the first struct page of the folio - those fields are | |
99 | * now accessed by struct slab. It is occasionally necessary to convert back to | |
100 | * a folio in order to communicate with the rest of the mm. Please use this | |
101 | * helper function instead of casting yourself, as the implementation may change | |
102 | * in the future. | |
103 | */ | |
104 | #define slab_folio(s) (_Generic((s), \ | |
105 | const struct slab *: (const struct folio *)s, \ | |
106 | struct slab *: (struct folio *)s)) | |
107 | ||
108 | /** | |
109 | * page_slab - Converts from first struct page to slab. | |
110 | * @p: The first (either head of compound or single) page of slab. | |
111 | * | |
112 | * A temporary wrapper to convert struct page to struct slab in situations where | |
113 | * we know the page is the compound head, or single order-0 page. | |
114 | * | |
115 | * Long-term ideally everything would work with struct slab directly or go | |
116 | * through folio to struct slab. | |
117 | * | |
118 | * Return: The slab which contains this page | |
119 | */ | |
120 | #define page_slab(p) (_Generic((p), \ | |
121 | const struct page *: (const struct slab *)(p), \ | |
122 | struct page *: (struct slab *)(p))) | |
123 | ||
124 | /** | |
125 | * slab_page - The first struct page allocated for a slab | |
126 | * @slab: The slab. | |
127 | * | |
128 | * A convenience wrapper for converting slab to the first struct page of the | |
129 | * underlying folio, to communicate with code not yet converted to folio or | |
130 | * struct slab. | |
131 | */ | |
132 | #define slab_page(s) folio_page(slab_folio(s), 0) | |
133 | ||
134 | /* | |
135 | * If network-based swap is enabled, sl*b must keep track of whether pages | |
136 | * were allocated from pfmemalloc reserves. | |
137 | */ | |
138 | static inline bool slab_test_pfmemalloc(const struct slab *slab) | |
139 | { | |
140 | return folio_test_active((struct folio *)slab_folio(slab)); | |
141 | } | |
142 | ||
143 | static inline void slab_set_pfmemalloc(struct slab *slab) | |
144 | { | |
145 | folio_set_active(slab_folio(slab)); | |
146 | } | |
147 | ||
148 | static inline void slab_clear_pfmemalloc(struct slab *slab) | |
149 | { | |
150 | folio_clear_active(slab_folio(slab)); | |
151 | } | |
152 | ||
153 | static inline void __slab_clear_pfmemalloc(struct slab *slab) | |
154 | { | |
155 | __folio_clear_active(slab_folio(slab)); | |
156 | } | |
157 | ||
158 | static inline void *slab_address(const struct slab *slab) | |
159 | { | |
160 | return folio_address(slab_folio(slab)); | |
161 | } | |
162 | ||
163 | static inline int slab_nid(const struct slab *slab) | |
164 | { | |
165 | return folio_nid(slab_folio(slab)); | |
166 | } | |
167 | ||
168 | static inline pg_data_t *slab_pgdat(const struct slab *slab) | |
169 | { | |
170 | return folio_pgdat(slab_folio(slab)); | |
171 | } | |
172 | ||
173 | static inline struct slab *virt_to_slab(const void *addr) | |
174 | { | |
175 | struct folio *folio = virt_to_folio(addr); | |
176 | ||
177 | if (!folio_test_slab(folio)) | |
178 | return NULL; | |
179 | ||
180 | return folio_slab(folio); | |
181 | } | |
182 | ||
183 | static inline int slab_order(const struct slab *slab) | |
184 | { | |
185 | return folio_order((struct folio *)slab_folio(slab)); | |
186 | } | |
187 | ||
188 | static inline size_t slab_size(const struct slab *slab) | |
189 | { | |
190 | return PAGE_SIZE << slab_order(slab); | |
191 | } | |
192 | ||
07f361b2 JK |
193 | #ifdef CONFIG_SLOB |
194 | /* | |
195 | * Common fields provided in kmem_cache by all slab allocators | |
196 | * This struct is either used directly by the allocator (SLOB) | |
197 | * or the allocator must include definitions for all fields | |
198 | * provided in kmem_cache_common in their definition of kmem_cache. | |
199 | * | |
200 | * Once we can do anonymous structs (C11 standard) we could put a | |
201 | * anonymous struct definition in these allocators so that the | |
202 | * separate allocations in the kmem_cache structure of SLAB and | |
203 | * SLUB is no longer needed. | |
204 | */ | |
205 | struct kmem_cache { | |
206 | unsigned int object_size;/* The original size of the object */ | |
207 | unsigned int size; /* The aligned/padded/added on size */ | |
208 | unsigned int align; /* Alignment as calculated */ | |
d50112ed | 209 | slab_flags_t flags; /* Active flags on the slab */ |
7bbdb81e AD |
210 | unsigned int useroffset;/* Usercopy region offset */ |
211 | unsigned int usersize; /* Usercopy region size */ | |
07f361b2 JK |
212 | const char *name; /* Slab name for sysfs */ |
213 | int refcount; /* Use counter */ | |
214 | void (*ctor)(void *); /* Called on object slot creation */ | |
215 | struct list_head list; /* List of all slab caches on the system */ | |
216 | }; | |
217 | ||
218 | #endif /* CONFIG_SLOB */ | |
219 | ||
220 | #ifdef CONFIG_SLAB | |
221 | #include <linux/slab_def.h> | |
222 | #endif | |
223 | ||
224 | #ifdef CONFIG_SLUB | |
225 | #include <linux/slub_def.h> | |
226 | #endif | |
227 | ||
228 | #include <linux/memcontrol.h> | |
11c7aec2 | 229 | #include <linux/fault-inject.h> |
11c7aec2 JDB |
230 | #include <linux/kasan.h> |
231 | #include <linux/kmemleak.h> | |
7c00fce9 | 232 | #include <linux/random.h> |
d92a8cfc | 233 | #include <linux/sched/mm.h> |
88f2ef73 | 234 | #include <linux/list_lru.h> |
07f361b2 | 235 | |
97d06609 CL |
236 | /* |
237 | * State of the slab allocator. | |
238 | * | |
239 | * This is used to describe the states of the allocator during bootup. | |
240 | * Allocators use this to gradually bootstrap themselves. Most allocators | |
241 | * have the problem that the structures used for managing slab caches are | |
242 | * allocated from slab caches themselves. | |
243 | */ | |
244 | enum slab_state { | |
245 | DOWN, /* No slab functionality yet */ | |
246 | PARTIAL, /* SLUB: kmem_cache_node available */ | |
ce8eb6c4 | 247 | PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */ |
97d06609 CL |
248 | UP, /* Slab caches usable but not all extras yet */ |
249 | FULL /* Everything is working */ | |
250 | }; | |
251 | ||
252 | extern enum slab_state slab_state; | |
253 | ||
18004c5d CL |
254 | /* The slab cache mutex protects the management structures during changes */ |
255 | extern struct mutex slab_mutex; | |
9b030cb8 CL |
256 | |
257 | /* The list of all slab caches on the system */ | |
18004c5d CL |
258 | extern struct list_head slab_caches; |
259 | ||
9b030cb8 CL |
260 | /* The slab cache that manages slab cache information */ |
261 | extern struct kmem_cache *kmem_cache; | |
262 | ||
af3b5f87 VB |
263 | /* A table of kmalloc cache names and sizes */ |
264 | extern const struct kmalloc_info_struct { | |
cb5d9fb3 | 265 | const char *name[NR_KMALLOC_TYPES]; |
55de8b9c | 266 | unsigned int size; |
af3b5f87 VB |
267 | } kmalloc_info[]; |
268 | ||
f97d5f63 CL |
269 | #ifndef CONFIG_SLOB |
270 | /* Kmalloc array related functions */ | |
34cc6990 | 271 | void setup_kmalloc_cache_index_table(void); |
d50112ed | 272 | void create_kmalloc_caches(slab_flags_t); |
2c59dd65 CL |
273 | |
274 | /* Find the kmalloc slab corresponding for a certain size */ | |
275 | struct kmem_cache *kmalloc_slab(size_t, gfp_t); | |
f97d5f63 CL |
276 | #endif |
277 | ||
44405099 | 278 | gfp_t kmalloc_fix_flags(gfp_t flags); |
f97d5f63 | 279 | |
9b030cb8 | 280 | /* Functions provided by the slab allocators */ |
d50112ed | 281 | int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags); |
97d06609 | 282 | |
55de8b9c AD |
283 | struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size, |
284 | slab_flags_t flags, unsigned int useroffset, | |
285 | unsigned int usersize); | |
45530c44 | 286 | extern void create_boot_cache(struct kmem_cache *, const char *name, |
361d575e AD |
287 | unsigned int size, slab_flags_t flags, |
288 | unsigned int useroffset, unsigned int usersize); | |
45530c44 | 289 | |
423c929c | 290 | int slab_unmergeable(struct kmem_cache *s); |
f4957d5b | 291 | struct kmem_cache *find_mergeable(unsigned size, unsigned align, |
d50112ed | 292 | slab_flags_t flags, const char *name, void (*ctor)(void *)); |
12220dea | 293 | #ifndef CONFIG_SLOB |
2633d7a0 | 294 | struct kmem_cache * |
f4957d5b | 295 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 296 | slab_flags_t flags, void (*ctor)(void *)); |
423c929c | 297 | |
0293d1fd | 298 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 299 | slab_flags_t flags, const char *name); |
cbb79694 | 300 | #else |
2633d7a0 | 301 | static inline struct kmem_cache * |
f4957d5b | 302 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 303 | slab_flags_t flags, void (*ctor)(void *)) |
cbb79694 | 304 | { return NULL; } |
423c929c | 305 | |
0293d1fd | 306 | static inline slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 307 | slab_flags_t flags, const char *name) |
423c929c JK |
308 | { |
309 | return flags; | |
310 | } | |
cbb79694 CL |
311 | #endif |
312 | ||
313 | ||
d8843922 | 314 | /* Legal flag mask for kmem_cache_create(), for various configurations */ |
6d6ea1e9 NB |
315 | #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \ |
316 | SLAB_CACHE_DMA32 | SLAB_PANIC | \ | |
5f0d5a3a | 317 | SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS ) |
d8843922 GC |
318 | |
319 | #if defined(CONFIG_DEBUG_SLAB) | |
320 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) | |
321 | #elif defined(CONFIG_SLUB_DEBUG) | |
322 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
becfda68 | 323 | SLAB_TRACE | SLAB_CONSISTENCY_CHECKS) |
d8843922 GC |
324 | #else |
325 | #define SLAB_DEBUG_FLAGS (0) | |
326 | #endif | |
327 | ||
328 | #if defined(CONFIG_SLAB) | |
329 | #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \ | |
230e9fc2 | 330 | SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \ |
75f296d9 | 331 | SLAB_ACCOUNT) |
d8843922 GC |
332 | #elif defined(CONFIG_SLUB) |
333 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \ | |
75f296d9 | 334 | SLAB_TEMPORARY | SLAB_ACCOUNT) |
d8843922 | 335 | #else |
34dbc3aa | 336 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE) |
d8843922 GC |
337 | #endif |
338 | ||
e70954fd | 339 | /* Common flags available with current configuration */ |
d8843922 GC |
340 | #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS) |
341 | ||
e70954fd TG |
342 | /* Common flags permitted for kmem_cache_create */ |
343 | #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \ | |
344 | SLAB_RED_ZONE | \ | |
345 | SLAB_POISON | \ | |
346 | SLAB_STORE_USER | \ | |
347 | SLAB_TRACE | \ | |
348 | SLAB_CONSISTENCY_CHECKS | \ | |
349 | SLAB_MEM_SPREAD | \ | |
350 | SLAB_NOLEAKTRACE | \ | |
351 | SLAB_RECLAIM_ACCOUNT | \ | |
352 | SLAB_TEMPORARY | \ | |
e70954fd TG |
353 | SLAB_ACCOUNT) |
354 | ||
f9e13c0a | 355 | bool __kmem_cache_empty(struct kmem_cache *); |
945cf2b6 | 356 | int __kmem_cache_shutdown(struct kmem_cache *); |
52b4b950 | 357 | void __kmem_cache_release(struct kmem_cache *); |
c9fc5864 | 358 | int __kmem_cache_shrink(struct kmem_cache *); |
41a21285 | 359 | void slab_kmem_cache_release(struct kmem_cache *); |
945cf2b6 | 360 | |
b7454ad3 GC |
361 | struct seq_file; |
362 | struct file; | |
b7454ad3 | 363 | |
0d7561c6 GC |
364 | struct slabinfo { |
365 | unsigned long active_objs; | |
366 | unsigned long num_objs; | |
367 | unsigned long active_slabs; | |
368 | unsigned long num_slabs; | |
369 | unsigned long shared_avail; | |
370 | unsigned int limit; | |
371 | unsigned int batchcount; | |
372 | unsigned int shared; | |
373 | unsigned int objects_per_slab; | |
374 | unsigned int cache_order; | |
375 | }; | |
376 | ||
377 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo); | |
378 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s); | |
b7454ad3 GC |
379 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
380 | size_t count, loff_t *ppos); | |
ba6c496e | 381 | |
484748f0 CL |
382 | /* |
383 | * Generic implementation of bulk operations | |
384 | * These are useful for situations in which the allocator cannot | |
9f706d68 | 385 | * perform optimizations. In that case segments of the object listed |
484748f0 CL |
386 | * may be allocated or freed using these operations. |
387 | */ | |
388 | void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **); | |
865762a8 | 389 | int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **); |
484748f0 | 390 | |
1a984c4e | 391 | static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s) |
6cea1d56 RG |
392 | { |
393 | return (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
d42f3245 | 394 | NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B; |
6cea1d56 RG |
395 | } |
396 | ||
e42f174e VB |
397 | #ifdef CONFIG_SLUB_DEBUG |
398 | #ifdef CONFIG_SLUB_DEBUG_ON | |
399 | DECLARE_STATIC_KEY_TRUE(slub_debug_enabled); | |
400 | #else | |
401 | DECLARE_STATIC_KEY_FALSE(slub_debug_enabled); | |
402 | #endif | |
403 | extern void print_tracking(struct kmem_cache *s, void *object); | |
1f9f78b1 | 404 | long validate_slab_cache(struct kmem_cache *s); |
0d4a062a ME |
405 | static inline bool __slub_debug_enabled(void) |
406 | { | |
407 | return static_branch_unlikely(&slub_debug_enabled); | |
408 | } | |
e42f174e VB |
409 | #else |
410 | static inline void print_tracking(struct kmem_cache *s, void *object) | |
411 | { | |
412 | } | |
0d4a062a ME |
413 | static inline bool __slub_debug_enabled(void) |
414 | { | |
415 | return false; | |
416 | } | |
e42f174e VB |
417 | #endif |
418 | ||
419 | /* | |
420 | * Returns true if any of the specified slub_debug flags is enabled for the | |
421 | * cache. Use only for flags parsed by setup_slub_debug() as it also enables | |
422 | * the static key. | |
423 | */ | |
424 | static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags) | |
425 | { | |
0d4a062a ME |
426 | if (IS_ENABLED(CONFIG_SLUB_DEBUG)) |
427 | VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS)); | |
428 | if (__slub_debug_enabled()) | |
e42f174e | 429 | return s->flags & flags; |
e42f174e VB |
430 | return false; |
431 | } | |
432 | ||
84c07d11 | 433 | #ifdef CONFIG_MEMCG_KMEM |
4b5f8d9a VB |
434 | /* |
435 | * slab_objcgs - get the object cgroups vector associated with a slab | |
436 | * @slab: a pointer to the slab struct | |
437 | * | |
438 | * Returns a pointer to the object cgroups vector associated with the slab, | |
439 | * or NULL if no such vector has been associated yet. | |
440 | */ | |
441 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) | |
442 | { | |
443 | unsigned long memcg_data = READ_ONCE(slab->memcg_data); | |
444 | ||
445 | VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), | |
446 | slab_page(slab)); | |
447 | VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab)); | |
448 | ||
449 | return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); | |
450 | } | |
451 | ||
452 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, | |
453 | gfp_t gfp, bool new_slab); | |
fdbcb2a6 WL |
454 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
455 | enum node_stat_item idx, int nr); | |
286e04b8 | 456 | |
4b5f8d9a | 457 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 | 458 | { |
4b5f8d9a VB |
459 | kfree(slab_objcgs(slab)); |
460 | slab->memcg_data = 0; | |
286e04b8 RG |
461 | } |
462 | ||
f2fe7b09 RG |
463 | static inline size_t obj_full_size(struct kmem_cache *s) |
464 | { | |
465 | /* | |
466 | * For each accounted object there is an extra space which is used | |
467 | * to store obj_cgroup membership. Charge it too. | |
468 | */ | |
469 | return s->size + sizeof(struct obj_cgroup *); | |
470 | } | |
471 | ||
becaba65 RG |
472 | /* |
473 | * Returns false if the allocation should fail. | |
474 | */ | |
475 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 476 | struct list_lru *lru, |
becaba65 RG |
477 | struct obj_cgroup **objcgp, |
478 | size_t objects, gfp_t flags) | |
f2fe7b09 | 479 | { |
9855609b RG |
480 | struct obj_cgroup *objcg; |
481 | ||
becaba65 RG |
482 | if (!memcg_kmem_enabled()) |
483 | return true; | |
484 | ||
485 | if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT)) | |
486 | return true; | |
487 | ||
9855609b RG |
488 | objcg = get_obj_cgroup_from_current(); |
489 | if (!objcg) | |
becaba65 | 490 | return true; |
9855609b | 491 | |
88f2ef73 MS |
492 | if (lru) { |
493 | int ret; | |
494 | struct mem_cgroup *memcg; | |
495 | ||
496 | memcg = get_mem_cgroup_from_objcg(objcg); | |
497 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
498 | css_put(&memcg->css); | |
499 | ||
500 | if (ret) | |
501 | goto out; | |
f2fe7b09 RG |
502 | } |
503 | ||
88f2ef73 MS |
504 | if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) |
505 | goto out; | |
506 | ||
becaba65 RG |
507 | *objcgp = objcg; |
508 | return true; | |
88f2ef73 MS |
509 | out: |
510 | obj_cgroup_put(objcg); | |
511 | return false; | |
f2fe7b09 RG |
512 | } |
513 | ||
964d4bd3 RG |
514 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
515 | struct obj_cgroup *objcg, | |
10befea9 RG |
516 | gfp_t flags, size_t size, |
517 | void **p) | |
964d4bd3 | 518 | { |
4b5f8d9a | 519 | struct slab *slab; |
964d4bd3 RG |
520 | unsigned long off; |
521 | size_t i; | |
522 | ||
becaba65 | 523 | if (!memcg_kmem_enabled() || !objcg) |
10befea9 RG |
524 | return; |
525 | ||
964d4bd3 RG |
526 | for (i = 0; i < size; i++) { |
527 | if (likely(p[i])) { | |
4b5f8d9a | 528 | slab = virt_to_slab(p[i]); |
10befea9 | 529 | |
4b5f8d9a VB |
530 | if (!slab_objcgs(slab) && |
531 | memcg_alloc_slab_cgroups(slab, s, flags, | |
2e9bd483 | 532 | false)) { |
10befea9 RG |
533 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
534 | continue; | |
535 | } | |
536 | ||
4b5f8d9a | 537 | off = obj_to_index(s, slab, p[i]); |
964d4bd3 | 538 | obj_cgroup_get(objcg); |
4b5f8d9a VB |
539 | slab_objcgs(slab)[off] = objcg; |
540 | mod_objcg_state(objcg, slab_pgdat(slab), | |
f2fe7b09 RG |
541 | cache_vmstat_idx(s), obj_full_size(s)); |
542 | } else { | |
543 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
964d4bd3 RG |
544 | } |
545 | } | |
546 | obj_cgroup_put(objcg); | |
964d4bd3 RG |
547 | } |
548 | ||
d1b2cf6c BR |
549 | static inline void memcg_slab_free_hook(struct kmem_cache *s_orig, |
550 | void **p, int objects) | |
964d4bd3 | 551 | { |
d1b2cf6c | 552 | struct kmem_cache *s; |
270c6a71 | 553 | struct obj_cgroup **objcgs; |
964d4bd3 | 554 | struct obj_cgroup *objcg; |
4b5f8d9a | 555 | struct slab *slab; |
964d4bd3 | 556 | unsigned int off; |
d1b2cf6c | 557 | int i; |
964d4bd3 | 558 | |
10befea9 RG |
559 | if (!memcg_kmem_enabled()) |
560 | return; | |
561 | ||
d1b2cf6c BR |
562 | for (i = 0; i < objects; i++) { |
563 | if (unlikely(!p[i])) | |
564 | continue; | |
964d4bd3 | 565 | |
4b5f8d9a VB |
566 | slab = virt_to_slab(p[i]); |
567 | /* we could be given a kmalloc_large() object, skip those */ | |
568 | if (!slab) | |
569 | continue; | |
570 | ||
571 | objcgs = slab_objcgs(slab); | |
270c6a71 | 572 | if (!objcgs) |
d1b2cf6c | 573 | continue; |
f2fe7b09 | 574 | |
d1b2cf6c | 575 | if (!s_orig) |
4b5f8d9a | 576 | s = slab->slab_cache; |
d1b2cf6c BR |
577 | else |
578 | s = s_orig; | |
10befea9 | 579 | |
4b5f8d9a | 580 | off = obj_to_index(s, slab, p[i]); |
270c6a71 | 581 | objcg = objcgs[off]; |
d1b2cf6c BR |
582 | if (!objcg) |
583 | continue; | |
f2fe7b09 | 584 | |
270c6a71 | 585 | objcgs[off] = NULL; |
d1b2cf6c | 586 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
4b5f8d9a | 587 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), |
d1b2cf6c BR |
588 | -obj_full_size(s)); |
589 | obj_cgroup_put(objcg); | |
590 | } | |
964d4bd3 RG |
591 | } |
592 | ||
84c07d11 | 593 | #else /* CONFIG_MEMCG_KMEM */ |
4b5f8d9a VB |
594 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) |
595 | { | |
596 | return NULL; | |
597 | } | |
598 | ||
9855609b | 599 | static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr) |
4d96ba35 RG |
600 | { |
601 | return NULL; | |
602 | } | |
603 | ||
4b5f8d9a | 604 | static inline int memcg_alloc_slab_cgroups(struct slab *slab, |
2e9bd483 | 605 | struct kmem_cache *s, gfp_t gfp, |
4b5f8d9a | 606 | bool new_slab) |
286e04b8 RG |
607 | { |
608 | return 0; | |
609 | } | |
610 | ||
4b5f8d9a | 611 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 RG |
612 | { |
613 | } | |
614 | ||
becaba65 | 615 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, |
88f2ef73 | 616 | struct list_lru *lru, |
becaba65 RG |
617 | struct obj_cgroup **objcgp, |
618 | size_t objects, gfp_t flags) | |
f2fe7b09 | 619 | { |
becaba65 | 620 | return true; |
f2fe7b09 RG |
621 | } |
622 | ||
964d4bd3 RG |
623 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
624 | struct obj_cgroup *objcg, | |
10befea9 RG |
625 | gfp_t flags, size_t size, |
626 | void **p) | |
964d4bd3 RG |
627 | { |
628 | } | |
629 | ||
d1b2cf6c BR |
630 | static inline void memcg_slab_free_hook(struct kmem_cache *s, |
631 | void **p, int objects) | |
964d4bd3 RG |
632 | { |
633 | } | |
84c07d11 | 634 | #endif /* CONFIG_MEMCG_KMEM */ |
b9ce5ef4 | 635 | |
401fb12c | 636 | #ifndef CONFIG_SLOB |
a64b5378 KC |
637 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
638 | { | |
82c1775d | 639 | struct slab *slab; |
a64b5378 | 640 | |
82c1775d MWO |
641 | slab = virt_to_slab(obj); |
642 | if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n", | |
a64b5378 KC |
643 | __func__)) |
644 | return NULL; | |
82c1775d | 645 | return slab->slab_cache; |
a64b5378 KC |
646 | } |
647 | ||
b918653b MWO |
648 | static __always_inline void account_slab(struct slab *slab, int order, |
649 | struct kmem_cache *s, gfp_t gfp) | |
6cea1d56 | 650 | { |
2e9bd483 | 651 | if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT)) |
4b5f8d9a | 652 | memcg_alloc_slab_cgroups(slab, s, gfp, true); |
2e9bd483 | 653 | |
b918653b | 654 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 655 | PAGE_SIZE << order); |
6cea1d56 RG |
656 | } |
657 | ||
b918653b MWO |
658 | static __always_inline void unaccount_slab(struct slab *slab, int order, |
659 | struct kmem_cache *s) | |
6cea1d56 | 660 | { |
10befea9 | 661 | if (memcg_kmem_enabled()) |
4b5f8d9a | 662 | memcg_free_slab_cgroups(slab); |
9855609b | 663 | |
b918653b | 664 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 665 | -(PAGE_SIZE << order)); |
6cea1d56 RG |
666 | } |
667 | ||
e42f174e VB |
668 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) |
669 | { | |
670 | struct kmem_cache *cachep; | |
671 | ||
672 | if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) && | |
e42f174e VB |
673 | !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) |
674 | return s; | |
675 | ||
676 | cachep = virt_to_cache(x); | |
10befea9 | 677 | if (WARN(cachep && cachep != s, |
e42f174e VB |
678 | "%s: Wrong slab cache. %s but object is from %s\n", |
679 | __func__, s->name, cachep->name)) | |
680 | print_tracking(cachep, x); | |
681 | return cachep; | |
682 | } | |
401fb12c | 683 | #endif /* CONFIG_SLOB */ |
e42f174e | 684 | |
11c7aec2 JDB |
685 | static inline size_t slab_ksize(const struct kmem_cache *s) |
686 | { | |
687 | #ifndef CONFIG_SLUB | |
688 | return s->object_size; | |
689 | ||
690 | #else /* CONFIG_SLUB */ | |
691 | # ifdef CONFIG_SLUB_DEBUG | |
692 | /* | |
693 | * Debugging requires use of the padding between object | |
694 | * and whatever may come after it. | |
695 | */ | |
696 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
697 | return s->object_size; | |
698 | # endif | |
80a9201a AP |
699 | if (s->flags & SLAB_KASAN) |
700 | return s->object_size; | |
11c7aec2 JDB |
701 | /* |
702 | * If we have the need to store the freelist pointer | |
703 | * back there or track user information then we can | |
704 | * only use the space before that information. | |
705 | */ | |
5f0d5a3a | 706 | if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER)) |
11c7aec2 JDB |
707 | return s->inuse; |
708 | /* | |
709 | * Else we can use all the padding etc for the allocation | |
710 | */ | |
711 | return s->size; | |
712 | #endif | |
713 | } | |
714 | ||
715 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 716 | struct list_lru *lru, |
964d4bd3 RG |
717 | struct obj_cgroup **objcgp, |
718 | size_t size, gfp_t flags) | |
11c7aec2 JDB |
719 | { |
720 | flags &= gfp_allowed_mask; | |
d92a8cfc | 721 | |
95d6c701 | 722 | might_alloc(flags); |
11c7aec2 | 723 | |
fab9963a | 724 | if (should_failslab(s, flags)) |
11c7aec2 JDB |
725 | return NULL; |
726 | ||
88f2ef73 | 727 | if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags)) |
becaba65 | 728 | return NULL; |
45264778 VD |
729 | |
730 | return s; | |
11c7aec2 JDB |
731 | } |
732 | ||
964d4bd3 | 733 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
da844b78 AK |
734 | struct obj_cgroup *objcg, gfp_t flags, |
735 | size_t size, void **p, bool init) | |
11c7aec2 JDB |
736 | { |
737 | size_t i; | |
738 | ||
739 | flags &= gfp_allowed_mask; | |
da844b78 AK |
740 | |
741 | /* | |
742 | * As memory initialization might be integrated into KASAN, | |
743 | * kasan_slab_alloc and initialization memset must be | |
744 | * kept together to avoid discrepancies in behavior. | |
745 | * | |
746 | * As p[i] might get tagged, memset and kmemleak hook come after KASAN. | |
747 | */ | |
11c7aec2 | 748 | for (i = 0; i < size; i++) { |
da844b78 AK |
749 | p[i] = kasan_slab_alloc(s, p[i], flags, init); |
750 | if (p[i] && init && !kasan_has_integrated_init()) | |
751 | memset(p[i], 0, s->object_size); | |
53128245 | 752 | kmemleak_alloc_recursive(p[i], s->object_size, 1, |
11c7aec2 | 753 | s->flags, flags); |
11c7aec2 | 754 | } |
45264778 | 755 | |
becaba65 | 756 | memcg_slab_post_alloc_hook(s, objcg, flags, size, p); |
11c7aec2 JDB |
757 | } |
758 | ||
44c5356f | 759 | #ifndef CONFIG_SLOB |
ca34956b CL |
760 | /* |
761 | * The slab lists for all objects. | |
762 | */ | |
763 | struct kmem_cache_node { | |
764 | spinlock_t list_lock; | |
765 | ||
766 | #ifdef CONFIG_SLAB | |
767 | struct list_head slabs_partial; /* partial list first, better asm code */ | |
768 | struct list_head slabs_full; | |
769 | struct list_head slabs_free; | |
bf00bd34 DR |
770 | unsigned long total_slabs; /* length of all slab lists */ |
771 | unsigned long free_slabs; /* length of free slab list only */ | |
ca34956b CL |
772 | unsigned long free_objects; |
773 | unsigned int free_limit; | |
774 | unsigned int colour_next; /* Per-node cache coloring */ | |
775 | struct array_cache *shared; /* shared per node */ | |
c8522a3a | 776 | struct alien_cache **alien; /* on other nodes */ |
ca34956b CL |
777 | unsigned long next_reap; /* updated without locking */ |
778 | int free_touched; /* updated without locking */ | |
779 | #endif | |
780 | ||
781 | #ifdef CONFIG_SLUB | |
782 | unsigned long nr_partial; | |
783 | struct list_head partial; | |
784 | #ifdef CONFIG_SLUB_DEBUG | |
785 | atomic_long_t nr_slabs; | |
786 | atomic_long_t total_objects; | |
787 | struct list_head full; | |
788 | #endif | |
789 | #endif | |
790 | ||
791 | }; | |
e25839f6 | 792 | |
44c5356f CL |
793 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
794 | { | |
795 | return s->node[node]; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Iterator over all nodes. The body will be executed for each node that has | |
800 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
801 | */ | |
802 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
9163582c MP |
803 | for (__node = 0; __node < nr_node_ids; __node++) \ |
804 | if ((__n = get_node(__s, __node))) | |
44c5356f CL |
805 | |
806 | #endif | |
807 | ||
852d8be0 YS |
808 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
809 | void dump_unreclaimable_slab(void); | |
810 | #else | |
811 | static inline void dump_unreclaimable_slab(void) | |
812 | { | |
813 | } | |
814 | #endif | |
815 | ||
55834c59 AP |
816 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr); |
817 | ||
7c00fce9 TG |
818 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
819 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
820 | gfp_t gfp); | |
821 | void cache_random_seq_destroy(struct kmem_cache *cachep); | |
822 | #else | |
823 | static inline int cache_random_seq_create(struct kmem_cache *cachep, | |
824 | unsigned int count, gfp_t gfp) | |
825 | { | |
826 | return 0; | |
827 | } | |
828 | static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { } | |
829 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
830 | ||
6471384a AP |
831 | static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c) |
832 | { | |
51cba1eb KC |
833 | if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, |
834 | &init_on_alloc)) { | |
6471384a AP |
835 | if (c->ctor) |
836 | return false; | |
837 | if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) | |
838 | return flags & __GFP_ZERO; | |
839 | return true; | |
840 | } | |
841 | return flags & __GFP_ZERO; | |
842 | } | |
843 | ||
844 | static inline bool slab_want_init_on_free(struct kmem_cache *c) | |
845 | { | |
51cba1eb KC |
846 | if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON, |
847 | &init_on_free)) | |
6471384a AP |
848 | return !(c->ctor || |
849 | (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))); | |
850 | return false; | |
851 | } | |
852 | ||
64dd6849 FM |
853 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
854 | void debugfs_slab_release(struct kmem_cache *); | |
855 | #else | |
856 | static inline void debugfs_slab_release(struct kmem_cache *s) { } | |
857 | #endif | |
858 | ||
5bb1bb35 | 859 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
860 | #define KS_ADDRS_COUNT 16 |
861 | struct kmem_obj_info { | |
862 | void *kp_ptr; | |
7213230a | 863 | struct slab *kp_slab; |
8e7f37f2 PM |
864 | void *kp_objp; |
865 | unsigned long kp_data_offset; | |
866 | struct kmem_cache *kp_slab_cache; | |
867 | void *kp_ret; | |
868 | void *kp_stack[KS_ADDRS_COUNT]; | |
e548eaa1 | 869 | void *kp_free_stack[KS_ADDRS_COUNT]; |
8e7f37f2 | 870 | }; |
7213230a | 871 | void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab); |
5bb1bb35 | 872 | #endif |
8e7f37f2 | 873 | |
0b3eb091 MWO |
874 | #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR |
875 | void __check_heap_object(const void *ptr, unsigned long n, | |
876 | const struct slab *slab, bool to_user); | |
877 | #else | |
878 | static inline | |
879 | void __check_heap_object(const void *ptr, unsigned long n, | |
880 | const struct slab *slab, bool to_user) | |
881 | { | |
882 | } | |
883 | #endif | |
884 | ||
5240ab40 | 885 | #endif /* MM_SLAB_H */ |