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 | */ | |
d5d2c02a | 7 | void __init kmem_cache_init(void); |
97d06609 | 8 | |
d122019b MWO |
9 | /* Reuses the bits in struct page */ |
10 | struct slab { | |
11 | unsigned long __page_flags; | |
401fb12c VB |
12 | |
13 | #if defined(CONFIG_SLAB) | |
14 | ||
130d4df5 | 15 | struct kmem_cache *slab_cache; |
d122019b | 16 | union { |
130d4df5 VB |
17 | struct { |
18 | struct list_head slab_list; | |
19 | void *freelist; /* array of free object indexes */ | |
20 | void *s_mem; /* first object */ | |
21 | }; | |
401fb12c VB |
22 | struct rcu_head rcu_head; |
23 | }; | |
401fb12c VB |
24 | unsigned int active; |
25 | ||
26 | #elif defined(CONFIG_SLUB) | |
27 | ||
401fb12c | 28 | struct kmem_cache *slab_cache; |
d122019b | 29 | union { |
401fb12c | 30 | struct { |
130d4df5 VB |
31 | union { |
32 | struct list_head slab_list; | |
33 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
34 | struct { | |
35 | struct slab *next; | |
36 | int slabs; /* Nr of slabs left */ | |
37 | }; | |
38 | #endif | |
39 | }; | |
40 | /* Double-word boundary */ | |
41 | void *freelist; /* first free object */ | |
42 | union { | |
43 | unsigned long counters; | |
44 | struct { | |
45 | unsigned inuse:16; | |
46 | unsigned objects:15; | |
47 | unsigned frozen:1; | |
48 | }; | |
49 | }; | |
d122019b | 50 | }; |
130d4df5 | 51 | struct rcu_head rcu_head; |
d122019b | 52 | }; |
401fb12c VB |
53 | unsigned int __unused; |
54 | ||
401fb12c VB |
55 | #else |
56 | #error "Unexpected slab allocator configured" | |
57 | #endif | |
d122019b | 58 | |
d122019b MWO |
59 | atomic_t __page_refcount; |
60 | #ifdef CONFIG_MEMCG | |
61 | unsigned long memcg_data; | |
62 | #endif | |
63 | }; | |
64 | ||
65 | #define SLAB_MATCH(pg, sl) \ | |
66 | static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl)) | |
67 | SLAB_MATCH(flags, __page_flags); | |
130d4df5 | 68 | SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */ |
d122019b MWO |
69 | SLAB_MATCH(_refcount, __page_refcount); |
70 | #ifdef CONFIG_MEMCG | |
71 | SLAB_MATCH(memcg_data, memcg_data); | |
72 | #endif | |
73 | #undef SLAB_MATCH | |
74 | static_assert(sizeof(struct slab) <= sizeof(struct page)); | |
130d4df5 VB |
75 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && defined(CONFIG_SLUB) |
76 | static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *))); | |
77 | #endif | |
d122019b MWO |
78 | |
79 | /** | |
80 | * folio_slab - Converts from folio to slab. | |
81 | * @folio: The folio. | |
82 | * | |
83 | * Currently struct slab is a different representation of a folio where | |
84 | * folio_test_slab() is true. | |
85 | * | |
86 | * Return: The slab which contains this folio. | |
87 | */ | |
88 | #define folio_slab(folio) (_Generic((folio), \ | |
89 | const struct folio *: (const struct slab *)(folio), \ | |
90 | struct folio *: (struct slab *)(folio))) | |
91 | ||
92 | /** | |
93 | * slab_folio - The folio allocated for a slab | |
94 | * @slab: The slab. | |
95 | * | |
96 | * Slabs are allocated as folios that contain the individual objects and are | |
97 | * using some fields in the first struct page of the folio - those fields are | |
98 | * now accessed by struct slab. It is occasionally necessary to convert back to | |
99 | * a folio in order to communicate with the rest of the mm. Please use this | |
100 | * helper function instead of casting yourself, as the implementation may change | |
101 | * in the future. | |
102 | */ | |
103 | #define slab_folio(s) (_Generic((s), \ | |
104 | const struct slab *: (const struct folio *)s, \ | |
105 | struct slab *: (struct folio *)s)) | |
106 | ||
107 | /** | |
108 | * page_slab - Converts from first struct page to slab. | |
109 | * @p: The first (either head of compound or single) page of slab. | |
110 | * | |
111 | * A temporary wrapper to convert struct page to struct slab in situations where | |
112 | * we know the page is the compound head, or single order-0 page. | |
113 | * | |
114 | * Long-term ideally everything would work with struct slab directly or go | |
115 | * through folio to struct slab. | |
116 | * | |
117 | * Return: The slab which contains this page | |
118 | */ | |
119 | #define page_slab(p) (_Generic((p), \ | |
120 | const struct page *: (const struct slab *)(p), \ | |
121 | struct page *: (struct slab *)(p))) | |
122 | ||
123 | /** | |
124 | * slab_page - The first struct page allocated for a slab | |
125 | * @slab: The slab. | |
126 | * | |
127 | * A convenience wrapper for converting slab to the first struct page of the | |
128 | * underlying folio, to communicate with code not yet converted to folio or | |
129 | * struct slab. | |
130 | */ | |
131 | #define slab_page(s) folio_page(slab_folio(s), 0) | |
132 | ||
133 | /* | |
134 | * If network-based swap is enabled, sl*b must keep track of whether pages | |
135 | * were allocated from pfmemalloc reserves. | |
136 | */ | |
137 | static inline bool slab_test_pfmemalloc(const struct slab *slab) | |
138 | { | |
139 | return folio_test_active((struct folio *)slab_folio(slab)); | |
140 | } | |
141 | ||
142 | static inline void slab_set_pfmemalloc(struct slab *slab) | |
143 | { | |
144 | folio_set_active(slab_folio(slab)); | |
145 | } | |
146 | ||
147 | static inline void slab_clear_pfmemalloc(struct slab *slab) | |
148 | { | |
149 | folio_clear_active(slab_folio(slab)); | |
150 | } | |
151 | ||
152 | static inline void __slab_clear_pfmemalloc(struct slab *slab) | |
153 | { | |
154 | __folio_clear_active(slab_folio(slab)); | |
155 | } | |
156 | ||
157 | static inline void *slab_address(const struct slab *slab) | |
158 | { | |
159 | return folio_address(slab_folio(slab)); | |
160 | } | |
161 | ||
162 | static inline int slab_nid(const struct slab *slab) | |
163 | { | |
164 | return folio_nid(slab_folio(slab)); | |
165 | } | |
166 | ||
167 | static inline pg_data_t *slab_pgdat(const struct slab *slab) | |
168 | { | |
169 | return folio_pgdat(slab_folio(slab)); | |
170 | } | |
171 | ||
172 | static inline struct slab *virt_to_slab(const void *addr) | |
173 | { | |
174 | struct folio *folio = virt_to_folio(addr); | |
175 | ||
176 | if (!folio_test_slab(folio)) | |
177 | return NULL; | |
178 | ||
179 | return folio_slab(folio); | |
180 | } | |
181 | ||
182 | static inline int slab_order(const struct slab *slab) | |
183 | { | |
184 | return folio_order((struct folio *)slab_folio(slab)); | |
185 | } | |
186 | ||
187 | static inline size_t slab_size(const struct slab *slab) | |
188 | { | |
189 | return PAGE_SIZE << slab_order(slab); | |
190 | } | |
191 | ||
07f361b2 JK |
192 | #ifdef CONFIG_SLAB |
193 | #include <linux/slab_def.h> | |
194 | #endif | |
195 | ||
196 | #ifdef CONFIG_SLUB | |
197 | #include <linux/slub_def.h> | |
198 | #endif | |
199 | ||
200 | #include <linux/memcontrol.h> | |
11c7aec2 | 201 | #include <linux/fault-inject.h> |
11c7aec2 JDB |
202 | #include <linux/kasan.h> |
203 | #include <linux/kmemleak.h> | |
7c00fce9 | 204 | #include <linux/random.h> |
d92a8cfc | 205 | #include <linux/sched/mm.h> |
88f2ef73 | 206 | #include <linux/list_lru.h> |
07f361b2 | 207 | |
97d06609 CL |
208 | /* |
209 | * State of the slab allocator. | |
210 | * | |
211 | * This is used to describe the states of the allocator during bootup. | |
212 | * Allocators use this to gradually bootstrap themselves. Most allocators | |
213 | * have the problem that the structures used for managing slab caches are | |
214 | * allocated from slab caches themselves. | |
215 | */ | |
216 | enum slab_state { | |
217 | DOWN, /* No slab functionality yet */ | |
218 | PARTIAL, /* SLUB: kmem_cache_node available */ | |
ce8eb6c4 | 219 | PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */ |
97d06609 CL |
220 | UP, /* Slab caches usable but not all extras yet */ |
221 | FULL /* Everything is working */ | |
222 | }; | |
223 | ||
224 | extern enum slab_state slab_state; | |
225 | ||
18004c5d CL |
226 | /* The slab cache mutex protects the management structures during changes */ |
227 | extern struct mutex slab_mutex; | |
9b030cb8 CL |
228 | |
229 | /* The list of all slab caches on the system */ | |
18004c5d CL |
230 | extern struct list_head slab_caches; |
231 | ||
9b030cb8 CL |
232 | /* The slab cache that manages slab cache information */ |
233 | extern struct kmem_cache *kmem_cache; | |
234 | ||
af3b5f87 VB |
235 | /* A table of kmalloc cache names and sizes */ |
236 | extern const struct kmalloc_info_struct { | |
cb5d9fb3 | 237 | const char *name[NR_KMALLOC_TYPES]; |
55de8b9c | 238 | unsigned int size; |
af3b5f87 VB |
239 | } kmalloc_info[]; |
240 | ||
f97d5f63 | 241 | /* Kmalloc array related functions */ |
34cc6990 | 242 | void setup_kmalloc_cache_index_table(void); |
d50112ed | 243 | void create_kmalloc_caches(slab_flags_t); |
2c59dd65 CL |
244 | |
245 | /* Find the kmalloc slab corresponding for a certain size */ | |
246 | struct kmem_cache *kmalloc_slab(size_t, gfp_t); | |
ed4cd17e HY |
247 | |
248 | void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, | |
249 | int node, size_t orig_size, | |
250 | unsigned long caller); | |
251 | void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller); | |
f97d5f63 | 252 | |
44405099 | 253 | gfp_t kmalloc_fix_flags(gfp_t flags); |
f97d5f63 | 254 | |
9b030cb8 | 255 | /* Functions provided by the slab allocators */ |
d50112ed | 256 | int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags); |
97d06609 | 257 | |
55de8b9c AD |
258 | struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size, |
259 | slab_flags_t flags, unsigned int useroffset, | |
260 | unsigned int usersize); | |
45530c44 | 261 | extern void create_boot_cache(struct kmem_cache *, const char *name, |
361d575e AD |
262 | unsigned int size, slab_flags_t flags, |
263 | unsigned int useroffset, unsigned int usersize); | |
45530c44 | 264 | |
423c929c | 265 | int slab_unmergeable(struct kmem_cache *s); |
f4957d5b | 266 | struct kmem_cache *find_mergeable(unsigned size, unsigned align, |
d50112ed | 267 | slab_flags_t flags, const char *name, void (*ctor)(void *)); |
2633d7a0 | 268 | struct kmem_cache * |
f4957d5b | 269 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 270 | slab_flags_t flags, void (*ctor)(void *)); |
423c929c | 271 | |
0293d1fd | 272 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 273 | slab_flags_t flags, const char *name); |
cbb79694 | 274 | |
bb944290 FT |
275 | static inline bool is_kmalloc_cache(struct kmem_cache *s) |
276 | { | |
bb944290 | 277 | return (s->flags & SLAB_KMALLOC); |
bb944290 | 278 | } |
cbb79694 | 279 | |
d8843922 | 280 | /* Legal flag mask for kmem_cache_create(), for various configurations */ |
6d6ea1e9 NB |
281 | #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \ |
282 | SLAB_CACHE_DMA32 | SLAB_PANIC | \ | |
5f0d5a3a | 283 | SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS ) |
d8843922 GC |
284 | |
285 | #if defined(CONFIG_DEBUG_SLAB) | |
286 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) | |
287 | #elif defined(CONFIG_SLUB_DEBUG) | |
288 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
becfda68 | 289 | SLAB_TRACE | SLAB_CONSISTENCY_CHECKS) |
d8843922 GC |
290 | #else |
291 | #define SLAB_DEBUG_FLAGS (0) | |
292 | #endif | |
293 | ||
294 | #if defined(CONFIG_SLAB) | |
295 | #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \ | |
230e9fc2 | 296 | SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \ |
75f296d9 | 297 | SLAB_ACCOUNT) |
d8843922 GC |
298 | #elif defined(CONFIG_SLUB) |
299 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \ | |
6cd6d33c FT |
300 | SLAB_TEMPORARY | SLAB_ACCOUNT | \ |
301 | SLAB_NO_USER_FLAGS | SLAB_KMALLOC) | |
d8843922 | 302 | #else |
34dbc3aa | 303 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE) |
d8843922 GC |
304 | #endif |
305 | ||
e70954fd | 306 | /* Common flags available with current configuration */ |
d8843922 GC |
307 | #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS) |
308 | ||
e70954fd TG |
309 | /* Common flags permitted for kmem_cache_create */ |
310 | #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \ | |
311 | SLAB_RED_ZONE | \ | |
312 | SLAB_POISON | \ | |
313 | SLAB_STORE_USER | \ | |
314 | SLAB_TRACE | \ | |
315 | SLAB_CONSISTENCY_CHECKS | \ | |
316 | SLAB_MEM_SPREAD | \ | |
317 | SLAB_NOLEAKTRACE | \ | |
318 | SLAB_RECLAIM_ACCOUNT | \ | |
319 | SLAB_TEMPORARY | \ | |
a285909f | 320 | SLAB_ACCOUNT | \ |
6cd6d33c | 321 | SLAB_KMALLOC | \ |
a285909f | 322 | SLAB_NO_USER_FLAGS) |
e70954fd | 323 | |
f9e13c0a | 324 | bool __kmem_cache_empty(struct kmem_cache *); |
945cf2b6 | 325 | int __kmem_cache_shutdown(struct kmem_cache *); |
52b4b950 | 326 | void __kmem_cache_release(struct kmem_cache *); |
c9fc5864 | 327 | int __kmem_cache_shrink(struct kmem_cache *); |
41a21285 | 328 | void slab_kmem_cache_release(struct kmem_cache *); |
945cf2b6 | 329 | |
b7454ad3 GC |
330 | struct seq_file; |
331 | struct file; | |
b7454ad3 | 332 | |
0d7561c6 GC |
333 | struct slabinfo { |
334 | unsigned long active_objs; | |
335 | unsigned long num_objs; | |
336 | unsigned long active_slabs; | |
337 | unsigned long num_slabs; | |
338 | unsigned long shared_avail; | |
339 | unsigned int limit; | |
340 | unsigned int batchcount; | |
341 | unsigned int shared; | |
342 | unsigned int objects_per_slab; | |
343 | unsigned int cache_order; | |
344 | }; | |
345 | ||
346 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo); | |
347 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s); | |
b7454ad3 GC |
348 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
349 | size_t count, loff_t *ppos); | |
ba6c496e | 350 | |
1a984c4e | 351 | static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s) |
6cea1d56 RG |
352 | { |
353 | return (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
d42f3245 | 354 | NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B; |
6cea1d56 RG |
355 | } |
356 | ||
e42f174e VB |
357 | #ifdef CONFIG_SLUB_DEBUG |
358 | #ifdef CONFIG_SLUB_DEBUG_ON | |
359 | DECLARE_STATIC_KEY_TRUE(slub_debug_enabled); | |
360 | #else | |
361 | DECLARE_STATIC_KEY_FALSE(slub_debug_enabled); | |
362 | #endif | |
363 | extern void print_tracking(struct kmem_cache *s, void *object); | |
1f9f78b1 | 364 | long validate_slab_cache(struct kmem_cache *s); |
0d4a062a ME |
365 | static inline bool __slub_debug_enabled(void) |
366 | { | |
367 | return static_branch_unlikely(&slub_debug_enabled); | |
368 | } | |
e42f174e VB |
369 | #else |
370 | static inline void print_tracking(struct kmem_cache *s, void *object) | |
371 | { | |
372 | } | |
0d4a062a ME |
373 | static inline bool __slub_debug_enabled(void) |
374 | { | |
375 | return false; | |
376 | } | |
e42f174e VB |
377 | #endif |
378 | ||
379 | /* | |
380 | * Returns true if any of the specified slub_debug flags is enabled for the | |
381 | * cache. Use only for flags parsed by setup_slub_debug() as it also enables | |
382 | * the static key. | |
383 | */ | |
384 | static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags) | |
385 | { | |
0d4a062a ME |
386 | if (IS_ENABLED(CONFIG_SLUB_DEBUG)) |
387 | VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS)); | |
388 | if (__slub_debug_enabled()) | |
e42f174e | 389 | return s->flags & flags; |
e42f174e VB |
390 | return false; |
391 | } | |
392 | ||
84c07d11 | 393 | #ifdef CONFIG_MEMCG_KMEM |
4b5f8d9a VB |
394 | /* |
395 | * slab_objcgs - get the object cgroups vector associated with a slab | |
396 | * @slab: a pointer to the slab struct | |
397 | * | |
398 | * Returns a pointer to the object cgroups vector associated with the slab, | |
399 | * or NULL if no such vector has been associated yet. | |
400 | */ | |
401 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) | |
402 | { | |
403 | unsigned long memcg_data = READ_ONCE(slab->memcg_data); | |
404 | ||
405 | VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), | |
406 | slab_page(slab)); | |
407 | VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab)); | |
408 | ||
409 | return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); | |
410 | } | |
411 | ||
412 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, | |
413 | gfp_t gfp, bool new_slab); | |
fdbcb2a6 WL |
414 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
415 | enum node_stat_item idx, int nr); | |
286e04b8 | 416 | |
4b5f8d9a | 417 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 | 418 | { |
4b5f8d9a VB |
419 | kfree(slab_objcgs(slab)); |
420 | slab->memcg_data = 0; | |
286e04b8 RG |
421 | } |
422 | ||
f2fe7b09 RG |
423 | static inline size_t obj_full_size(struct kmem_cache *s) |
424 | { | |
425 | /* | |
426 | * For each accounted object there is an extra space which is used | |
427 | * to store obj_cgroup membership. Charge it too. | |
428 | */ | |
429 | return s->size + sizeof(struct obj_cgroup *); | |
430 | } | |
431 | ||
becaba65 RG |
432 | /* |
433 | * Returns false if the allocation should fail. | |
434 | */ | |
435 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 436 | struct list_lru *lru, |
becaba65 RG |
437 | struct obj_cgroup **objcgp, |
438 | size_t objects, gfp_t flags) | |
f2fe7b09 | 439 | { |
9855609b RG |
440 | struct obj_cgroup *objcg; |
441 | ||
f7a449f7 | 442 | if (!memcg_kmem_online()) |
becaba65 RG |
443 | return true; |
444 | ||
445 | if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT)) | |
446 | return true; | |
447 | ||
9855609b RG |
448 | objcg = get_obj_cgroup_from_current(); |
449 | if (!objcg) | |
becaba65 | 450 | return true; |
9855609b | 451 | |
88f2ef73 MS |
452 | if (lru) { |
453 | int ret; | |
454 | struct mem_cgroup *memcg; | |
455 | ||
456 | memcg = get_mem_cgroup_from_objcg(objcg); | |
457 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
458 | css_put(&memcg->css); | |
459 | ||
460 | if (ret) | |
461 | goto out; | |
f2fe7b09 RG |
462 | } |
463 | ||
88f2ef73 MS |
464 | if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) |
465 | goto out; | |
466 | ||
becaba65 RG |
467 | *objcgp = objcg; |
468 | return true; | |
88f2ef73 MS |
469 | out: |
470 | obj_cgroup_put(objcg); | |
471 | return false; | |
f2fe7b09 RG |
472 | } |
473 | ||
964d4bd3 RG |
474 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
475 | struct obj_cgroup *objcg, | |
10befea9 RG |
476 | gfp_t flags, size_t size, |
477 | void **p) | |
964d4bd3 | 478 | { |
4b5f8d9a | 479 | struct slab *slab; |
964d4bd3 RG |
480 | unsigned long off; |
481 | size_t i; | |
482 | ||
f7a449f7 | 483 | if (!memcg_kmem_online() || !objcg) |
10befea9 RG |
484 | return; |
485 | ||
964d4bd3 RG |
486 | for (i = 0; i < size; i++) { |
487 | if (likely(p[i])) { | |
4b5f8d9a | 488 | slab = virt_to_slab(p[i]); |
10befea9 | 489 | |
4b5f8d9a VB |
490 | if (!slab_objcgs(slab) && |
491 | memcg_alloc_slab_cgroups(slab, s, flags, | |
2e9bd483 | 492 | false)) { |
10befea9 RG |
493 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
494 | continue; | |
495 | } | |
496 | ||
4b5f8d9a | 497 | off = obj_to_index(s, slab, p[i]); |
964d4bd3 | 498 | obj_cgroup_get(objcg); |
4b5f8d9a VB |
499 | slab_objcgs(slab)[off] = objcg; |
500 | mod_objcg_state(objcg, slab_pgdat(slab), | |
f2fe7b09 RG |
501 | cache_vmstat_idx(s), obj_full_size(s)); |
502 | } else { | |
503 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
964d4bd3 RG |
504 | } |
505 | } | |
506 | obj_cgroup_put(objcg); | |
964d4bd3 RG |
507 | } |
508 | ||
b77d5b1b | 509 | static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, |
d1b2cf6c | 510 | void **p, int objects) |
964d4bd3 | 511 | { |
270c6a71 | 512 | struct obj_cgroup **objcgs; |
d1b2cf6c | 513 | int i; |
964d4bd3 | 514 | |
f7a449f7 | 515 | if (!memcg_kmem_online()) |
10befea9 RG |
516 | return; |
517 | ||
b77d5b1b MS |
518 | objcgs = slab_objcgs(slab); |
519 | if (!objcgs) | |
520 | return; | |
f2fe7b09 | 521 | |
b77d5b1b MS |
522 | for (i = 0; i < objects; i++) { |
523 | struct obj_cgroup *objcg; | |
524 | unsigned int off; | |
10befea9 | 525 | |
4b5f8d9a | 526 | off = obj_to_index(s, slab, p[i]); |
270c6a71 | 527 | objcg = objcgs[off]; |
d1b2cf6c BR |
528 | if (!objcg) |
529 | continue; | |
f2fe7b09 | 530 | |
270c6a71 | 531 | objcgs[off] = NULL; |
d1b2cf6c | 532 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
4b5f8d9a | 533 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), |
d1b2cf6c BR |
534 | -obj_full_size(s)); |
535 | obj_cgroup_put(objcg); | |
536 | } | |
964d4bd3 RG |
537 | } |
538 | ||
84c07d11 | 539 | #else /* CONFIG_MEMCG_KMEM */ |
4b5f8d9a VB |
540 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) |
541 | { | |
542 | return NULL; | |
543 | } | |
544 | ||
9855609b | 545 | static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr) |
4d96ba35 RG |
546 | { |
547 | return NULL; | |
548 | } | |
549 | ||
4b5f8d9a | 550 | static inline int memcg_alloc_slab_cgroups(struct slab *slab, |
2e9bd483 | 551 | struct kmem_cache *s, gfp_t gfp, |
4b5f8d9a | 552 | bool new_slab) |
286e04b8 RG |
553 | { |
554 | return 0; | |
555 | } | |
556 | ||
4b5f8d9a | 557 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 RG |
558 | { |
559 | } | |
560 | ||
becaba65 | 561 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, |
88f2ef73 | 562 | struct list_lru *lru, |
becaba65 RG |
563 | struct obj_cgroup **objcgp, |
564 | size_t objects, gfp_t flags) | |
f2fe7b09 | 565 | { |
becaba65 | 566 | return true; |
f2fe7b09 RG |
567 | } |
568 | ||
964d4bd3 RG |
569 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
570 | struct obj_cgroup *objcg, | |
10befea9 RG |
571 | gfp_t flags, size_t size, |
572 | void **p) | |
964d4bd3 RG |
573 | { |
574 | } | |
575 | ||
b77d5b1b | 576 | static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, |
d1b2cf6c | 577 | void **p, int objects) |
964d4bd3 RG |
578 | { |
579 | } | |
84c07d11 | 580 | #endif /* CONFIG_MEMCG_KMEM */ |
b9ce5ef4 | 581 | |
a64b5378 KC |
582 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
583 | { | |
82c1775d | 584 | struct slab *slab; |
a64b5378 | 585 | |
82c1775d MWO |
586 | slab = virt_to_slab(obj); |
587 | if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n", | |
a64b5378 KC |
588 | __func__)) |
589 | return NULL; | |
82c1775d | 590 | return slab->slab_cache; |
a64b5378 KC |
591 | } |
592 | ||
b918653b MWO |
593 | static __always_inline void account_slab(struct slab *slab, int order, |
594 | struct kmem_cache *s, gfp_t gfp) | |
6cea1d56 | 595 | { |
f7a449f7 | 596 | if (memcg_kmem_online() && (s->flags & SLAB_ACCOUNT)) |
4b5f8d9a | 597 | memcg_alloc_slab_cgroups(slab, s, gfp, true); |
2e9bd483 | 598 | |
b918653b | 599 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 600 | PAGE_SIZE << order); |
6cea1d56 RG |
601 | } |
602 | ||
b918653b MWO |
603 | static __always_inline void unaccount_slab(struct slab *slab, int order, |
604 | struct kmem_cache *s) | |
6cea1d56 | 605 | { |
f7a449f7 | 606 | if (memcg_kmem_online()) |
4b5f8d9a | 607 | memcg_free_slab_cgroups(slab); |
9855609b | 608 | |
b918653b | 609 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 610 | -(PAGE_SIZE << order)); |
6cea1d56 RG |
611 | } |
612 | ||
e42f174e VB |
613 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) |
614 | { | |
615 | struct kmem_cache *cachep; | |
616 | ||
617 | if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) && | |
e42f174e VB |
618 | !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) |
619 | return s; | |
620 | ||
621 | cachep = virt_to_cache(x); | |
10befea9 | 622 | if (WARN(cachep && cachep != s, |
e42f174e VB |
623 | "%s: Wrong slab cache. %s but object is from %s\n", |
624 | __func__, s->name, cachep->name)) | |
625 | print_tracking(cachep, x); | |
626 | return cachep; | |
627 | } | |
d6a71648 HY |
628 | |
629 | void free_large_kmalloc(struct folio *folio, void *object); | |
630 | ||
8dfa9d55 HY |
631 | size_t __ksize(const void *objp); |
632 | ||
11c7aec2 JDB |
633 | static inline size_t slab_ksize(const struct kmem_cache *s) |
634 | { | |
635 | #ifndef CONFIG_SLUB | |
636 | return s->object_size; | |
637 | ||
638 | #else /* CONFIG_SLUB */ | |
639 | # ifdef CONFIG_SLUB_DEBUG | |
640 | /* | |
641 | * Debugging requires use of the padding between object | |
642 | * and whatever may come after it. | |
643 | */ | |
644 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
645 | return s->object_size; | |
646 | # endif | |
80a9201a AP |
647 | if (s->flags & SLAB_KASAN) |
648 | return s->object_size; | |
11c7aec2 JDB |
649 | /* |
650 | * If we have the need to store the freelist pointer | |
651 | * back there or track user information then we can | |
652 | * only use the space before that information. | |
653 | */ | |
5f0d5a3a | 654 | if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER)) |
11c7aec2 JDB |
655 | return s->inuse; |
656 | /* | |
657 | * Else we can use all the padding etc for the allocation | |
658 | */ | |
659 | return s->size; | |
660 | #endif | |
661 | } | |
662 | ||
663 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 664 | struct list_lru *lru, |
964d4bd3 RG |
665 | struct obj_cgroup **objcgp, |
666 | size_t size, gfp_t flags) | |
11c7aec2 JDB |
667 | { |
668 | flags &= gfp_allowed_mask; | |
d92a8cfc | 669 | |
95d6c701 | 670 | might_alloc(flags); |
11c7aec2 | 671 | |
fab9963a | 672 | if (should_failslab(s, flags)) |
11c7aec2 JDB |
673 | return NULL; |
674 | ||
88f2ef73 | 675 | if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags)) |
becaba65 | 676 | return NULL; |
45264778 VD |
677 | |
678 | return s; | |
11c7aec2 JDB |
679 | } |
680 | ||
964d4bd3 | 681 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
da844b78 | 682 | struct obj_cgroup *objcg, gfp_t flags, |
9ce67395 FT |
683 | size_t size, void **p, bool init, |
684 | unsigned int orig_size) | |
11c7aec2 | 685 | { |
9ce67395 | 686 | unsigned int zero_size = s->object_size; |
11c7aec2 JDB |
687 | size_t i; |
688 | ||
689 | flags &= gfp_allowed_mask; | |
da844b78 | 690 | |
9ce67395 FT |
691 | /* |
692 | * For kmalloc object, the allocated memory size(object_size) is likely | |
693 | * larger than the requested size(orig_size). If redzone check is | |
694 | * enabled for the extra space, don't zero it, as it will be redzoned | |
695 | * soon. The redzone operation for this extra space could be seen as a | |
696 | * replacement of current poisoning under certain debug option, and | |
697 | * won't break other sanity checks. | |
698 | */ | |
699 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) && | |
700 | (s->flags & SLAB_KMALLOC)) | |
701 | zero_size = orig_size; | |
702 | ||
da844b78 AK |
703 | /* |
704 | * As memory initialization might be integrated into KASAN, | |
705 | * kasan_slab_alloc and initialization memset must be | |
706 | * kept together to avoid discrepancies in behavior. | |
707 | * | |
708 | * As p[i] might get tagged, memset and kmemleak hook come after KASAN. | |
709 | */ | |
11c7aec2 | 710 | for (i = 0; i < size; i++) { |
da844b78 AK |
711 | p[i] = kasan_slab_alloc(s, p[i], flags, init); |
712 | if (p[i] && init && !kasan_has_integrated_init()) | |
9ce67395 | 713 | memset(p[i], 0, zero_size); |
53128245 | 714 | kmemleak_alloc_recursive(p[i], s->object_size, 1, |
11c7aec2 | 715 | s->flags, flags); |
68ef169a | 716 | kmsan_slab_alloc(s, p[i], flags); |
11c7aec2 | 717 | } |
45264778 | 718 | |
becaba65 | 719 | memcg_slab_post_alloc_hook(s, objcg, flags, size, p); |
11c7aec2 JDB |
720 | } |
721 | ||
ca34956b CL |
722 | /* |
723 | * The slab lists for all objects. | |
724 | */ | |
725 | struct kmem_cache_node { | |
ca34956b | 726 | #ifdef CONFIG_SLAB |
b539ce9f | 727 | raw_spinlock_t list_lock; |
ca34956b CL |
728 | struct list_head slabs_partial; /* partial list first, better asm code */ |
729 | struct list_head slabs_full; | |
730 | struct list_head slabs_free; | |
bf00bd34 DR |
731 | unsigned long total_slabs; /* length of all slab lists */ |
732 | unsigned long free_slabs; /* length of free slab list only */ | |
ca34956b CL |
733 | unsigned long free_objects; |
734 | unsigned int free_limit; | |
735 | unsigned int colour_next; /* Per-node cache coloring */ | |
736 | struct array_cache *shared; /* shared per node */ | |
c8522a3a | 737 | struct alien_cache **alien; /* on other nodes */ |
ca34956b CL |
738 | unsigned long next_reap; /* updated without locking */ |
739 | int free_touched; /* updated without locking */ | |
740 | #endif | |
741 | ||
742 | #ifdef CONFIG_SLUB | |
b539ce9f | 743 | spinlock_t list_lock; |
ca34956b CL |
744 | unsigned long nr_partial; |
745 | struct list_head partial; | |
746 | #ifdef CONFIG_SLUB_DEBUG | |
747 | atomic_long_t nr_slabs; | |
748 | atomic_long_t total_objects; | |
749 | struct list_head full; | |
750 | #endif | |
751 | #endif | |
752 | ||
753 | }; | |
e25839f6 | 754 | |
44c5356f CL |
755 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
756 | { | |
757 | return s->node[node]; | |
758 | } | |
759 | ||
760 | /* | |
761 | * Iterator over all nodes. The body will be executed for each node that has | |
762 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
763 | */ | |
764 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
9163582c MP |
765 | for (__node = 0; __node < nr_node_ids; __node++) \ |
766 | if ((__n = get_node(__s, __node))) | |
44c5356f | 767 | |
44c5356f | 768 | |
852d8be0 YS |
769 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
770 | void dump_unreclaimable_slab(void); | |
771 | #else | |
772 | static inline void dump_unreclaimable_slab(void) | |
773 | { | |
774 | } | |
775 | #endif | |
776 | ||
55834c59 AP |
777 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr); |
778 | ||
7c00fce9 TG |
779 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
780 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
781 | gfp_t gfp); | |
782 | void cache_random_seq_destroy(struct kmem_cache *cachep); | |
783 | #else | |
784 | static inline int cache_random_seq_create(struct kmem_cache *cachep, | |
785 | unsigned int count, gfp_t gfp) | |
786 | { | |
787 | return 0; | |
788 | } | |
789 | static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { } | |
790 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
791 | ||
6471384a AP |
792 | static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c) |
793 | { | |
51cba1eb KC |
794 | if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, |
795 | &init_on_alloc)) { | |
6471384a AP |
796 | if (c->ctor) |
797 | return false; | |
798 | if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) | |
799 | return flags & __GFP_ZERO; | |
800 | return true; | |
801 | } | |
802 | return flags & __GFP_ZERO; | |
803 | } | |
804 | ||
805 | static inline bool slab_want_init_on_free(struct kmem_cache *c) | |
806 | { | |
51cba1eb KC |
807 | if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON, |
808 | &init_on_free)) | |
6471384a AP |
809 | return !(c->ctor || |
810 | (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))); | |
811 | return false; | |
812 | } | |
813 | ||
64dd6849 FM |
814 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
815 | void debugfs_slab_release(struct kmem_cache *); | |
816 | #else | |
817 | static inline void debugfs_slab_release(struct kmem_cache *s) { } | |
818 | #endif | |
819 | ||
5bb1bb35 | 820 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
821 | #define KS_ADDRS_COUNT 16 |
822 | struct kmem_obj_info { | |
823 | void *kp_ptr; | |
7213230a | 824 | struct slab *kp_slab; |
8e7f37f2 PM |
825 | void *kp_objp; |
826 | unsigned long kp_data_offset; | |
827 | struct kmem_cache *kp_slab_cache; | |
828 | void *kp_ret; | |
829 | void *kp_stack[KS_ADDRS_COUNT]; | |
e548eaa1 | 830 | void *kp_free_stack[KS_ADDRS_COUNT]; |
8e7f37f2 | 831 | }; |
2dfe63e6 | 832 | void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab); |
5bb1bb35 | 833 | #endif |
8e7f37f2 | 834 | |
0b3eb091 MWO |
835 | #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR |
836 | void __check_heap_object(const void *ptr, unsigned long n, | |
837 | const struct slab *slab, bool to_user); | |
838 | #else | |
839 | static inline | |
840 | void __check_heap_object(const void *ptr, unsigned long n, | |
841 | const struct slab *slab, bool to_user) | |
842 | { | |
843 | } | |
844 | #endif | |
845 | ||
946fa0db FT |
846 | #ifdef CONFIG_SLUB_DEBUG |
847 | void skip_orig_size_check(struct kmem_cache *s, const void *object); | |
848 | #endif | |
849 | ||
5240ab40 | 850 | #endif /* MM_SLAB_H */ |