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 | ||
130d4df5 | 14 | struct kmem_cache *slab_cache; |
d122019b | 15 | union { |
130d4df5 VB |
16 | struct { |
17 | struct list_head slab_list; | |
18 | void *freelist; /* array of free object indexes */ | |
19 | void *s_mem; /* first object */ | |
20 | }; | |
401fb12c VB |
21 | struct rcu_head rcu_head; |
22 | }; | |
401fb12c VB |
23 | unsigned int active; |
24 | ||
25 | #elif defined(CONFIG_SLUB) | |
26 | ||
401fb12c | 27 | struct kmem_cache *slab_cache; |
d122019b | 28 | union { |
401fb12c | 29 | struct { |
130d4df5 VB |
30 | union { |
31 | struct list_head slab_list; | |
32 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
33 | struct { | |
34 | struct slab *next; | |
35 | int slabs; /* Nr of slabs left */ | |
36 | }; | |
37 | #endif | |
38 | }; | |
39 | /* Double-word boundary */ | |
40 | void *freelist; /* first free object */ | |
41 | union { | |
42 | unsigned long counters; | |
43 | struct { | |
44 | unsigned inuse:16; | |
45 | unsigned objects:15; | |
46 | unsigned frozen:1; | |
47 | }; | |
48 | }; | |
d122019b | 49 | }; |
130d4df5 | 50 | struct rcu_head rcu_head; |
d122019b | 51 | }; |
401fb12c VB |
52 | unsigned int __unused; |
53 | ||
54 | #elif defined(CONFIG_SLOB) | |
55 | ||
56 | struct list_head slab_list; | |
57 | void *__unused_1; | |
58 | void *freelist; /* first free block */ | |
b01af5c0 HY |
59 | long units; |
60 | unsigned int __unused_2; | |
401fb12c VB |
61 | |
62 | #else | |
63 | #error "Unexpected slab allocator configured" | |
64 | #endif | |
d122019b | 65 | |
d122019b MWO |
66 | atomic_t __page_refcount; |
67 | #ifdef CONFIG_MEMCG | |
68 | unsigned long memcg_data; | |
69 | #endif | |
70 | }; | |
71 | ||
72 | #define SLAB_MATCH(pg, sl) \ | |
73 | static_assert(offsetof(struct page, pg) == offsetof(struct slab, sl)) | |
74 | SLAB_MATCH(flags, __page_flags); | |
401fb12c | 75 | #ifndef CONFIG_SLOB |
130d4df5 VB |
76 | SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */ |
77 | #else | |
78 | SLAB_MATCH(compound_head, slab_list); /* Ensure bit 0 is clear */ | |
401fb12c | 79 | #endif |
d122019b MWO |
80 | SLAB_MATCH(_refcount, __page_refcount); |
81 | #ifdef CONFIG_MEMCG | |
82 | SLAB_MATCH(memcg_data, memcg_data); | |
83 | #endif | |
84 | #undef SLAB_MATCH | |
85 | static_assert(sizeof(struct slab) <= sizeof(struct page)); | |
130d4df5 VB |
86 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && defined(CONFIG_SLUB) |
87 | static_assert(IS_ALIGNED(offsetof(struct slab, freelist), 2*sizeof(void *))); | |
88 | #endif | |
d122019b MWO |
89 | |
90 | /** | |
91 | * folio_slab - Converts from folio to slab. | |
92 | * @folio: The folio. | |
93 | * | |
94 | * Currently struct slab is a different representation of a folio where | |
95 | * folio_test_slab() is true. | |
96 | * | |
97 | * Return: The slab which contains this folio. | |
98 | */ | |
99 | #define folio_slab(folio) (_Generic((folio), \ | |
100 | const struct folio *: (const struct slab *)(folio), \ | |
101 | struct folio *: (struct slab *)(folio))) | |
102 | ||
103 | /** | |
104 | * slab_folio - The folio allocated for a slab | |
105 | * @slab: The slab. | |
106 | * | |
107 | * Slabs are allocated as folios that contain the individual objects and are | |
108 | * using some fields in the first struct page of the folio - those fields are | |
109 | * now accessed by struct slab. It is occasionally necessary to convert back to | |
110 | * a folio in order to communicate with the rest of the mm. Please use this | |
111 | * helper function instead of casting yourself, as the implementation may change | |
112 | * in the future. | |
113 | */ | |
114 | #define slab_folio(s) (_Generic((s), \ | |
115 | const struct slab *: (const struct folio *)s, \ | |
116 | struct slab *: (struct folio *)s)) | |
117 | ||
118 | /** | |
119 | * page_slab - Converts from first struct page to slab. | |
120 | * @p: The first (either head of compound or single) page of slab. | |
121 | * | |
122 | * A temporary wrapper to convert struct page to struct slab in situations where | |
123 | * we know the page is the compound head, or single order-0 page. | |
124 | * | |
125 | * Long-term ideally everything would work with struct slab directly or go | |
126 | * through folio to struct slab. | |
127 | * | |
128 | * Return: The slab which contains this page | |
129 | */ | |
130 | #define page_slab(p) (_Generic((p), \ | |
131 | const struct page *: (const struct slab *)(p), \ | |
132 | struct page *: (struct slab *)(p))) | |
133 | ||
134 | /** | |
135 | * slab_page - The first struct page allocated for a slab | |
136 | * @slab: The slab. | |
137 | * | |
138 | * A convenience wrapper for converting slab to the first struct page of the | |
139 | * underlying folio, to communicate with code not yet converted to folio or | |
140 | * struct slab. | |
141 | */ | |
142 | #define slab_page(s) folio_page(slab_folio(s), 0) | |
143 | ||
144 | /* | |
145 | * If network-based swap is enabled, sl*b must keep track of whether pages | |
146 | * were allocated from pfmemalloc reserves. | |
147 | */ | |
148 | static inline bool slab_test_pfmemalloc(const struct slab *slab) | |
149 | { | |
150 | return folio_test_active((struct folio *)slab_folio(slab)); | |
151 | } | |
152 | ||
153 | static inline void slab_set_pfmemalloc(struct slab *slab) | |
154 | { | |
155 | folio_set_active(slab_folio(slab)); | |
156 | } | |
157 | ||
158 | static inline void slab_clear_pfmemalloc(struct slab *slab) | |
159 | { | |
160 | folio_clear_active(slab_folio(slab)); | |
161 | } | |
162 | ||
163 | static inline void __slab_clear_pfmemalloc(struct slab *slab) | |
164 | { | |
165 | __folio_clear_active(slab_folio(slab)); | |
166 | } | |
167 | ||
168 | static inline void *slab_address(const struct slab *slab) | |
169 | { | |
170 | return folio_address(slab_folio(slab)); | |
171 | } | |
172 | ||
173 | static inline int slab_nid(const struct slab *slab) | |
174 | { | |
175 | return folio_nid(slab_folio(slab)); | |
176 | } | |
177 | ||
178 | static inline pg_data_t *slab_pgdat(const struct slab *slab) | |
179 | { | |
180 | return folio_pgdat(slab_folio(slab)); | |
181 | } | |
182 | ||
183 | static inline struct slab *virt_to_slab(const void *addr) | |
184 | { | |
185 | struct folio *folio = virt_to_folio(addr); | |
186 | ||
187 | if (!folio_test_slab(folio)) | |
188 | return NULL; | |
189 | ||
190 | return folio_slab(folio); | |
191 | } | |
192 | ||
193 | static inline int slab_order(const struct slab *slab) | |
194 | { | |
195 | return folio_order((struct folio *)slab_folio(slab)); | |
196 | } | |
197 | ||
198 | static inline size_t slab_size(const struct slab *slab) | |
199 | { | |
200 | return PAGE_SIZE << slab_order(slab); | |
201 | } | |
202 | ||
07f361b2 JK |
203 | #ifdef CONFIG_SLOB |
204 | /* | |
205 | * Common fields provided in kmem_cache by all slab allocators | |
206 | * This struct is either used directly by the allocator (SLOB) | |
207 | * or the allocator must include definitions for all fields | |
208 | * provided in kmem_cache_common in their definition of kmem_cache. | |
209 | * | |
210 | * Once we can do anonymous structs (C11 standard) we could put a | |
211 | * anonymous struct definition in these allocators so that the | |
212 | * separate allocations in the kmem_cache structure of SLAB and | |
213 | * SLUB is no longer needed. | |
214 | */ | |
215 | struct kmem_cache { | |
216 | unsigned int object_size;/* The original size of the object */ | |
217 | unsigned int size; /* The aligned/padded/added on size */ | |
218 | unsigned int align; /* Alignment as calculated */ | |
d50112ed | 219 | slab_flags_t flags; /* Active flags on the slab */ |
07f361b2 JK |
220 | const char *name; /* Slab name for sysfs */ |
221 | int refcount; /* Use counter */ | |
222 | void (*ctor)(void *); /* Called on object slot creation */ | |
223 | struct list_head list; /* List of all slab caches on the system */ | |
224 | }; | |
225 | ||
226 | #endif /* CONFIG_SLOB */ | |
227 | ||
228 | #ifdef CONFIG_SLAB | |
229 | #include <linux/slab_def.h> | |
230 | #endif | |
231 | ||
232 | #ifdef CONFIG_SLUB | |
233 | #include <linux/slub_def.h> | |
234 | #endif | |
235 | ||
236 | #include <linux/memcontrol.h> | |
11c7aec2 | 237 | #include <linux/fault-inject.h> |
11c7aec2 JDB |
238 | #include <linux/kasan.h> |
239 | #include <linux/kmemleak.h> | |
7c00fce9 | 240 | #include <linux/random.h> |
d92a8cfc | 241 | #include <linux/sched/mm.h> |
88f2ef73 | 242 | #include <linux/list_lru.h> |
07f361b2 | 243 | |
97d06609 CL |
244 | /* |
245 | * State of the slab allocator. | |
246 | * | |
247 | * This is used to describe the states of the allocator during bootup. | |
248 | * Allocators use this to gradually bootstrap themselves. Most allocators | |
249 | * have the problem that the structures used for managing slab caches are | |
250 | * allocated from slab caches themselves. | |
251 | */ | |
252 | enum slab_state { | |
253 | DOWN, /* No slab functionality yet */ | |
254 | PARTIAL, /* SLUB: kmem_cache_node available */ | |
ce8eb6c4 | 255 | PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */ |
97d06609 CL |
256 | UP, /* Slab caches usable but not all extras yet */ |
257 | FULL /* Everything is working */ | |
258 | }; | |
259 | ||
260 | extern enum slab_state slab_state; | |
261 | ||
18004c5d CL |
262 | /* The slab cache mutex protects the management structures during changes */ |
263 | extern struct mutex slab_mutex; | |
9b030cb8 CL |
264 | |
265 | /* The list of all slab caches on the system */ | |
18004c5d CL |
266 | extern struct list_head slab_caches; |
267 | ||
9b030cb8 CL |
268 | /* The slab cache that manages slab cache information */ |
269 | extern struct kmem_cache *kmem_cache; | |
270 | ||
af3b5f87 VB |
271 | /* A table of kmalloc cache names and sizes */ |
272 | extern const struct kmalloc_info_struct { | |
cb5d9fb3 | 273 | const char *name[NR_KMALLOC_TYPES]; |
55de8b9c | 274 | unsigned int size; |
af3b5f87 VB |
275 | } kmalloc_info[]; |
276 | ||
f97d5f63 CL |
277 | #ifndef CONFIG_SLOB |
278 | /* Kmalloc array related functions */ | |
34cc6990 | 279 | void setup_kmalloc_cache_index_table(void); |
d50112ed | 280 | void create_kmalloc_caches(slab_flags_t); |
2c59dd65 CL |
281 | |
282 | /* Find the kmalloc slab corresponding for a certain size */ | |
283 | struct kmem_cache *kmalloc_slab(size_t, gfp_t); | |
ed4cd17e HY |
284 | |
285 | void *__kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, | |
286 | int node, size_t orig_size, | |
287 | unsigned long caller); | |
288 | void __kmem_cache_free(struct kmem_cache *s, void *x, unsigned long caller); | |
f97d5f63 CL |
289 | #endif |
290 | ||
44405099 | 291 | gfp_t kmalloc_fix_flags(gfp_t flags); |
f97d5f63 | 292 | |
9b030cb8 | 293 | /* Functions provided by the slab allocators */ |
d50112ed | 294 | int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags); |
97d06609 | 295 | |
55de8b9c AD |
296 | struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size, |
297 | slab_flags_t flags, unsigned int useroffset, | |
298 | unsigned int usersize); | |
45530c44 | 299 | extern void create_boot_cache(struct kmem_cache *, const char *name, |
361d575e AD |
300 | unsigned int size, slab_flags_t flags, |
301 | unsigned int useroffset, unsigned int usersize); | |
45530c44 | 302 | |
423c929c | 303 | int slab_unmergeable(struct kmem_cache *s); |
f4957d5b | 304 | struct kmem_cache *find_mergeable(unsigned size, unsigned align, |
d50112ed | 305 | slab_flags_t flags, const char *name, void (*ctor)(void *)); |
12220dea | 306 | #ifndef CONFIG_SLOB |
2633d7a0 | 307 | struct kmem_cache * |
f4957d5b | 308 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 309 | slab_flags_t flags, void (*ctor)(void *)); |
423c929c | 310 | |
0293d1fd | 311 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 312 | slab_flags_t flags, const char *name); |
cbb79694 | 313 | #else |
2633d7a0 | 314 | static inline struct kmem_cache * |
f4957d5b | 315 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 316 | slab_flags_t flags, void (*ctor)(void *)) |
cbb79694 | 317 | { return NULL; } |
423c929c | 318 | |
0293d1fd | 319 | static inline slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 320 | slab_flags_t flags, const char *name) |
423c929c JK |
321 | { |
322 | return flags; | |
323 | } | |
cbb79694 CL |
324 | #endif |
325 | ||
326 | ||
d8843922 | 327 | /* Legal flag mask for kmem_cache_create(), for various configurations */ |
6d6ea1e9 NB |
328 | #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \ |
329 | SLAB_CACHE_DMA32 | SLAB_PANIC | \ | |
5f0d5a3a | 330 | SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS ) |
d8843922 GC |
331 | |
332 | #if defined(CONFIG_DEBUG_SLAB) | |
333 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) | |
334 | #elif defined(CONFIG_SLUB_DEBUG) | |
335 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
becfda68 | 336 | SLAB_TRACE | SLAB_CONSISTENCY_CHECKS) |
d8843922 GC |
337 | #else |
338 | #define SLAB_DEBUG_FLAGS (0) | |
339 | #endif | |
340 | ||
341 | #if defined(CONFIG_SLAB) | |
342 | #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \ | |
230e9fc2 | 343 | SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \ |
75f296d9 | 344 | SLAB_ACCOUNT) |
d8843922 GC |
345 | #elif defined(CONFIG_SLUB) |
346 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \ | |
6cd6d33c FT |
347 | SLAB_TEMPORARY | SLAB_ACCOUNT | \ |
348 | SLAB_NO_USER_FLAGS | SLAB_KMALLOC) | |
d8843922 | 349 | #else |
34dbc3aa | 350 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE) |
d8843922 GC |
351 | #endif |
352 | ||
e70954fd | 353 | /* Common flags available with current configuration */ |
d8843922 GC |
354 | #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS) |
355 | ||
e70954fd TG |
356 | /* Common flags permitted for kmem_cache_create */ |
357 | #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \ | |
358 | SLAB_RED_ZONE | \ | |
359 | SLAB_POISON | \ | |
360 | SLAB_STORE_USER | \ | |
361 | SLAB_TRACE | \ | |
362 | SLAB_CONSISTENCY_CHECKS | \ | |
363 | SLAB_MEM_SPREAD | \ | |
364 | SLAB_NOLEAKTRACE | \ | |
365 | SLAB_RECLAIM_ACCOUNT | \ | |
366 | SLAB_TEMPORARY | \ | |
a285909f | 367 | SLAB_ACCOUNT | \ |
6cd6d33c | 368 | SLAB_KMALLOC | \ |
a285909f | 369 | SLAB_NO_USER_FLAGS) |
e70954fd | 370 | |
f9e13c0a | 371 | bool __kmem_cache_empty(struct kmem_cache *); |
945cf2b6 | 372 | int __kmem_cache_shutdown(struct kmem_cache *); |
52b4b950 | 373 | void __kmem_cache_release(struct kmem_cache *); |
c9fc5864 | 374 | int __kmem_cache_shrink(struct kmem_cache *); |
41a21285 | 375 | void slab_kmem_cache_release(struct kmem_cache *); |
945cf2b6 | 376 | |
b7454ad3 GC |
377 | struct seq_file; |
378 | struct file; | |
b7454ad3 | 379 | |
0d7561c6 GC |
380 | struct slabinfo { |
381 | unsigned long active_objs; | |
382 | unsigned long num_objs; | |
383 | unsigned long active_slabs; | |
384 | unsigned long num_slabs; | |
385 | unsigned long shared_avail; | |
386 | unsigned int limit; | |
387 | unsigned int batchcount; | |
388 | unsigned int shared; | |
389 | unsigned int objects_per_slab; | |
390 | unsigned int cache_order; | |
391 | }; | |
392 | ||
393 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo); | |
394 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s); | |
b7454ad3 GC |
395 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
396 | size_t count, loff_t *ppos); | |
ba6c496e | 397 | |
1a984c4e | 398 | static inline enum node_stat_item cache_vmstat_idx(struct kmem_cache *s) |
6cea1d56 RG |
399 | { |
400 | return (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
d42f3245 | 401 | NR_SLAB_RECLAIMABLE_B : NR_SLAB_UNRECLAIMABLE_B; |
6cea1d56 RG |
402 | } |
403 | ||
e42f174e VB |
404 | #ifdef CONFIG_SLUB_DEBUG |
405 | #ifdef CONFIG_SLUB_DEBUG_ON | |
406 | DECLARE_STATIC_KEY_TRUE(slub_debug_enabled); | |
407 | #else | |
408 | DECLARE_STATIC_KEY_FALSE(slub_debug_enabled); | |
409 | #endif | |
410 | extern void print_tracking(struct kmem_cache *s, void *object); | |
1f9f78b1 | 411 | long validate_slab_cache(struct kmem_cache *s); |
0d4a062a ME |
412 | static inline bool __slub_debug_enabled(void) |
413 | { | |
414 | return static_branch_unlikely(&slub_debug_enabled); | |
415 | } | |
e42f174e VB |
416 | #else |
417 | static inline void print_tracking(struct kmem_cache *s, void *object) | |
418 | { | |
419 | } | |
0d4a062a ME |
420 | static inline bool __slub_debug_enabled(void) |
421 | { | |
422 | return false; | |
423 | } | |
e42f174e VB |
424 | #endif |
425 | ||
426 | /* | |
427 | * Returns true if any of the specified slub_debug flags is enabled for the | |
428 | * cache. Use only for flags parsed by setup_slub_debug() as it also enables | |
429 | * the static key. | |
430 | */ | |
431 | static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags) | |
432 | { | |
0d4a062a ME |
433 | if (IS_ENABLED(CONFIG_SLUB_DEBUG)) |
434 | VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS)); | |
435 | if (__slub_debug_enabled()) | |
e42f174e | 436 | return s->flags & flags; |
e42f174e VB |
437 | return false; |
438 | } | |
439 | ||
84c07d11 | 440 | #ifdef CONFIG_MEMCG_KMEM |
4b5f8d9a VB |
441 | /* |
442 | * slab_objcgs - get the object cgroups vector associated with a slab | |
443 | * @slab: a pointer to the slab struct | |
444 | * | |
445 | * Returns a pointer to the object cgroups vector associated with the slab, | |
446 | * or NULL if no such vector has been associated yet. | |
447 | */ | |
448 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) | |
449 | { | |
450 | unsigned long memcg_data = READ_ONCE(slab->memcg_data); | |
451 | ||
452 | VM_BUG_ON_PAGE(memcg_data && !(memcg_data & MEMCG_DATA_OBJCGS), | |
453 | slab_page(slab)); | |
454 | VM_BUG_ON_PAGE(memcg_data & MEMCG_DATA_KMEM, slab_page(slab)); | |
455 | ||
456 | return (struct obj_cgroup **)(memcg_data & ~MEMCG_DATA_FLAGS_MASK); | |
457 | } | |
458 | ||
459 | int memcg_alloc_slab_cgroups(struct slab *slab, struct kmem_cache *s, | |
460 | gfp_t gfp, bool new_slab); | |
fdbcb2a6 WL |
461 | void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat, |
462 | enum node_stat_item idx, int nr); | |
286e04b8 | 463 | |
4b5f8d9a | 464 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 | 465 | { |
4b5f8d9a VB |
466 | kfree(slab_objcgs(slab)); |
467 | slab->memcg_data = 0; | |
286e04b8 RG |
468 | } |
469 | ||
f2fe7b09 RG |
470 | static inline size_t obj_full_size(struct kmem_cache *s) |
471 | { | |
472 | /* | |
473 | * For each accounted object there is an extra space which is used | |
474 | * to store obj_cgroup membership. Charge it too. | |
475 | */ | |
476 | return s->size + sizeof(struct obj_cgroup *); | |
477 | } | |
478 | ||
becaba65 RG |
479 | /* |
480 | * Returns false if the allocation should fail. | |
481 | */ | |
482 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 483 | struct list_lru *lru, |
becaba65 RG |
484 | struct obj_cgroup **objcgp, |
485 | size_t objects, gfp_t flags) | |
f2fe7b09 | 486 | { |
9855609b RG |
487 | struct obj_cgroup *objcg; |
488 | ||
becaba65 RG |
489 | if (!memcg_kmem_enabled()) |
490 | return true; | |
491 | ||
492 | if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT)) | |
493 | return true; | |
494 | ||
9855609b RG |
495 | objcg = get_obj_cgroup_from_current(); |
496 | if (!objcg) | |
becaba65 | 497 | return true; |
9855609b | 498 | |
88f2ef73 MS |
499 | if (lru) { |
500 | int ret; | |
501 | struct mem_cgroup *memcg; | |
502 | ||
503 | memcg = get_mem_cgroup_from_objcg(objcg); | |
504 | ret = memcg_list_lru_alloc(memcg, lru, flags); | |
505 | css_put(&memcg->css); | |
506 | ||
507 | if (ret) | |
508 | goto out; | |
f2fe7b09 RG |
509 | } |
510 | ||
88f2ef73 MS |
511 | if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) |
512 | goto out; | |
513 | ||
becaba65 RG |
514 | *objcgp = objcg; |
515 | return true; | |
88f2ef73 MS |
516 | out: |
517 | obj_cgroup_put(objcg); | |
518 | return false; | |
f2fe7b09 RG |
519 | } |
520 | ||
964d4bd3 RG |
521 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
522 | struct obj_cgroup *objcg, | |
10befea9 RG |
523 | gfp_t flags, size_t size, |
524 | void **p) | |
964d4bd3 | 525 | { |
4b5f8d9a | 526 | struct slab *slab; |
964d4bd3 RG |
527 | unsigned long off; |
528 | size_t i; | |
529 | ||
becaba65 | 530 | if (!memcg_kmem_enabled() || !objcg) |
10befea9 RG |
531 | return; |
532 | ||
964d4bd3 RG |
533 | for (i = 0; i < size; i++) { |
534 | if (likely(p[i])) { | |
4b5f8d9a | 535 | slab = virt_to_slab(p[i]); |
10befea9 | 536 | |
4b5f8d9a VB |
537 | if (!slab_objcgs(slab) && |
538 | memcg_alloc_slab_cgroups(slab, s, flags, | |
2e9bd483 | 539 | false)) { |
10befea9 RG |
540 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
541 | continue; | |
542 | } | |
543 | ||
4b5f8d9a | 544 | off = obj_to_index(s, slab, p[i]); |
964d4bd3 | 545 | obj_cgroup_get(objcg); |
4b5f8d9a VB |
546 | slab_objcgs(slab)[off] = objcg; |
547 | mod_objcg_state(objcg, slab_pgdat(slab), | |
f2fe7b09 RG |
548 | cache_vmstat_idx(s), obj_full_size(s)); |
549 | } else { | |
550 | obj_cgroup_uncharge(objcg, obj_full_size(s)); | |
964d4bd3 RG |
551 | } |
552 | } | |
553 | obj_cgroup_put(objcg); | |
964d4bd3 RG |
554 | } |
555 | ||
b77d5b1b | 556 | static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, |
d1b2cf6c | 557 | void **p, int objects) |
964d4bd3 | 558 | { |
270c6a71 | 559 | struct obj_cgroup **objcgs; |
d1b2cf6c | 560 | int i; |
964d4bd3 | 561 | |
10befea9 RG |
562 | if (!memcg_kmem_enabled()) |
563 | return; | |
564 | ||
b77d5b1b MS |
565 | objcgs = slab_objcgs(slab); |
566 | if (!objcgs) | |
567 | return; | |
f2fe7b09 | 568 | |
b77d5b1b MS |
569 | for (i = 0; i < objects; i++) { |
570 | struct obj_cgroup *objcg; | |
571 | unsigned int off; | |
10befea9 | 572 | |
4b5f8d9a | 573 | off = obj_to_index(s, slab, p[i]); |
270c6a71 | 574 | objcg = objcgs[off]; |
d1b2cf6c BR |
575 | if (!objcg) |
576 | continue; | |
f2fe7b09 | 577 | |
270c6a71 | 578 | objcgs[off] = NULL; |
d1b2cf6c | 579 | obj_cgroup_uncharge(objcg, obj_full_size(s)); |
4b5f8d9a | 580 | mod_objcg_state(objcg, slab_pgdat(slab), cache_vmstat_idx(s), |
d1b2cf6c BR |
581 | -obj_full_size(s)); |
582 | obj_cgroup_put(objcg); | |
583 | } | |
964d4bd3 RG |
584 | } |
585 | ||
84c07d11 | 586 | #else /* CONFIG_MEMCG_KMEM */ |
4b5f8d9a VB |
587 | static inline struct obj_cgroup **slab_objcgs(struct slab *slab) |
588 | { | |
589 | return NULL; | |
590 | } | |
591 | ||
9855609b | 592 | static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr) |
4d96ba35 RG |
593 | { |
594 | return NULL; | |
595 | } | |
596 | ||
4b5f8d9a | 597 | static inline int memcg_alloc_slab_cgroups(struct slab *slab, |
2e9bd483 | 598 | struct kmem_cache *s, gfp_t gfp, |
4b5f8d9a | 599 | bool new_slab) |
286e04b8 RG |
600 | { |
601 | return 0; | |
602 | } | |
603 | ||
4b5f8d9a | 604 | static inline void memcg_free_slab_cgroups(struct slab *slab) |
286e04b8 RG |
605 | { |
606 | } | |
607 | ||
becaba65 | 608 | static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s, |
88f2ef73 | 609 | struct list_lru *lru, |
becaba65 RG |
610 | struct obj_cgroup **objcgp, |
611 | size_t objects, gfp_t flags) | |
f2fe7b09 | 612 | { |
becaba65 | 613 | return true; |
f2fe7b09 RG |
614 | } |
615 | ||
964d4bd3 RG |
616 | static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s, |
617 | struct obj_cgroup *objcg, | |
10befea9 RG |
618 | gfp_t flags, size_t size, |
619 | void **p) | |
964d4bd3 RG |
620 | { |
621 | } | |
622 | ||
b77d5b1b | 623 | static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab, |
d1b2cf6c | 624 | void **p, int objects) |
964d4bd3 RG |
625 | { |
626 | } | |
84c07d11 | 627 | #endif /* CONFIG_MEMCG_KMEM */ |
b9ce5ef4 | 628 | |
401fb12c | 629 | #ifndef CONFIG_SLOB |
a64b5378 KC |
630 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
631 | { | |
82c1775d | 632 | struct slab *slab; |
a64b5378 | 633 | |
82c1775d MWO |
634 | slab = virt_to_slab(obj); |
635 | if (WARN_ONCE(!slab, "%s: Object is not a Slab page!\n", | |
a64b5378 KC |
636 | __func__)) |
637 | return NULL; | |
82c1775d | 638 | return slab->slab_cache; |
a64b5378 KC |
639 | } |
640 | ||
b918653b MWO |
641 | static __always_inline void account_slab(struct slab *slab, int order, |
642 | struct kmem_cache *s, gfp_t gfp) | |
6cea1d56 | 643 | { |
2e9bd483 | 644 | if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT)) |
4b5f8d9a | 645 | memcg_alloc_slab_cgroups(slab, s, gfp, true); |
2e9bd483 | 646 | |
b918653b | 647 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 648 | PAGE_SIZE << order); |
6cea1d56 RG |
649 | } |
650 | ||
b918653b MWO |
651 | static __always_inline void unaccount_slab(struct slab *slab, int order, |
652 | struct kmem_cache *s) | |
6cea1d56 | 653 | { |
10befea9 | 654 | if (memcg_kmem_enabled()) |
4b5f8d9a | 655 | memcg_free_slab_cgroups(slab); |
9855609b | 656 | |
b918653b | 657 | mod_node_page_state(slab_pgdat(slab), cache_vmstat_idx(s), |
f2fe7b09 | 658 | -(PAGE_SIZE << order)); |
6cea1d56 RG |
659 | } |
660 | ||
e42f174e VB |
661 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) |
662 | { | |
663 | struct kmem_cache *cachep; | |
664 | ||
665 | if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) && | |
e42f174e VB |
666 | !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) |
667 | return s; | |
668 | ||
669 | cachep = virt_to_cache(x); | |
10befea9 | 670 | if (WARN(cachep && cachep != s, |
e42f174e VB |
671 | "%s: Wrong slab cache. %s but object is from %s\n", |
672 | __func__, s->name, cachep->name)) | |
673 | print_tracking(cachep, x); | |
674 | return cachep; | |
675 | } | |
d6a71648 HY |
676 | |
677 | void free_large_kmalloc(struct folio *folio, void *object); | |
678 | ||
401fb12c | 679 | #endif /* CONFIG_SLOB */ |
e42f174e | 680 | |
8dfa9d55 HY |
681 | size_t __ksize(const void *objp); |
682 | ||
11c7aec2 JDB |
683 | static inline size_t slab_ksize(const struct kmem_cache *s) |
684 | { | |
685 | #ifndef CONFIG_SLUB | |
686 | return s->object_size; | |
687 | ||
688 | #else /* CONFIG_SLUB */ | |
689 | # ifdef CONFIG_SLUB_DEBUG | |
690 | /* | |
691 | * Debugging requires use of the padding between object | |
692 | * and whatever may come after it. | |
693 | */ | |
694 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
695 | return s->object_size; | |
696 | # endif | |
80a9201a AP |
697 | if (s->flags & SLAB_KASAN) |
698 | return s->object_size; | |
11c7aec2 JDB |
699 | /* |
700 | * If we have the need to store the freelist pointer | |
701 | * back there or track user information then we can | |
702 | * only use the space before that information. | |
703 | */ | |
5f0d5a3a | 704 | if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER)) |
11c7aec2 JDB |
705 | return s->inuse; |
706 | /* | |
707 | * Else we can use all the padding etc for the allocation | |
708 | */ | |
709 | return s->size; | |
710 | #endif | |
711 | } | |
712 | ||
713 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, | |
88f2ef73 | 714 | struct list_lru *lru, |
964d4bd3 RG |
715 | struct obj_cgroup **objcgp, |
716 | size_t size, gfp_t flags) | |
11c7aec2 JDB |
717 | { |
718 | flags &= gfp_allowed_mask; | |
d92a8cfc | 719 | |
95d6c701 | 720 | might_alloc(flags); |
11c7aec2 | 721 | |
fab9963a | 722 | if (should_failslab(s, flags)) |
11c7aec2 JDB |
723 | return NULL; |
724 | ||
88f2ef73 | 725 | if (!memcg_slab_pre_alloc_hook(s, lru, objcgp, size, flags)) |
becaba65 | 726 | return NULL; |
45264778 VD |
727 | |
728 | return s; | |
11c7aec2 JDB |
729 | } |
730 | ||
964d4bd3 | 731 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
da844b78 | 732 | struct obj_cgroup *objcg, gfp_t flags, |
9ce67395 FT |
733 | size_t size, void **p, bool init, |
734 | unsigned int orig_size) | |
11c7aec2 | 735 | { |
9ce67395 | 736 | unsigned int zero_size = s->object_size; |
11c7aec2 JDB |
737 | size_t i; |
738 | ||
739 | flags &= gfp_allowed_mask; | |
da844b78 | 740 | |
9ce67395 FT |
741 | /* |
742 | * For kmalloc object, the allocated memory size(object_size) is likely | |
743 | * larger than the requested size(orig_size). If redzone check is | |
744 | * enabled for the extra space, don't zero it, as it will be redzoned | |
745 | * soon. The redzone operation for this extra space could be seen as a | |
746 | * replacement of current poisoning under certain debug option, and | |
747 | * won't break other sanity checks. | |
748 | */ | |
749 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER | SLAB_RED_ZONE) && | |
750 | (s->flags & SLAB_KMALLOC)) | |
751 | zero_size = orig_size; | |
752 | ||
da844b78 AK |
753 | /* |
754 | * As memory initialization might be integrated into KASAN, | |
755 | * kasan_slab_alloc and initialization memset must be | |
756 | * kept together to avoid discrepancies in behavior. | |
757 | * | |
758 | * As p[i] might get tagged, memset and kmemleak hook come after KASAN. | |
759 | */ | |
11c7aec2 | 760 | for (i = 0; i < size; i++) { |
da844b78 AK |
761 | p[i] = kasan_slab_alloc(s, p[i], flags, init); |
762 | if (p[i] && init && !kasan_has_integrated_init()) | |
9ce67395 | 763 | memset(p[i], 0, zero_size); |
53128245 | 764 | kmemleak_alloc_recursive(p[i], s->object_size, 1, |
11c7aec2 | 765 | s->flags, flags); |
68ef169a | 766 | kmsan_slab_alloc(s, p[i], flags); |
11c7aec2 | 767 | } |
45264778 | 768 | |
becaba65 | 769 | memcg_slab_post_alloc_hook(s, objcg, flags, size, p); |
11c7aec2 JDB |
770 | } |
771 | ||
44c5356f | 772 | #ifndef CONFIG_SLOB |
ca34956b CL |
773 | /* |
774 | * The slab lists for all objects. | |
775 | */ | |
776 | struct kmem_cache_node { | |
ca34956b | 777 | #ifdef CONFIG_SLAB |
b539ce9f | 778 | raw_spinlock_t list_lock; |
ca34956b CL |
779 | struct list_head slabs_partial; /* partial list first, better asm code */ |
780 | struct list_head slabs_full; | |
781 | struct list_head slabs_free; | |
bf00bd34 DR |
782 | unsigned long total_slabs; /* length of all slab lists */ |
783 | unsigned long free_slabs; /* length of free slab list only */ | |
ca34956b CL |
784 | unsigned long free_objects; |
785 | unsigned int free_limit; | |
786 | unsigned int colour_next; /* Per-node cache coloring */ | |
787 | struct array_cache *shared; /* shared per node */ | |
c8522a3a | 788 | struct alien_cache **alien; /* on other nodes */ |
ca34956b CL |
789 | unsigned long next_reap; /* updated without locking */ |
790 | int free_touched; /* updated without locking */ | |
791 | #endif | |
792 | ||
793 | #ifdef CONFIG_SLUB | |
b539ce9f | 794 | spinlock_t list_lock; |
ca34956b CL |
795 | unsigned long nr_partial; |
796 | struct list_head partial; | |
797 | #ifdef CONFIG_SLUB_DEBUG | |
798 | atomic_long_t nr_slabs; | |
799 | atomic_long_t total_objects; | |
800 | struct list_head full; | |
801 | #endif | |
802 | #endif | |
803 | ||
804 | }; | |
e25839f6 | 805 | |
44c5356f CL |
806 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
807 | { | |
808 | return s->node[node]; | |
809 | } | |
810 | ||
811 | /* | |
812 | * Iterator over all nodes. The body will be executed for each node that has | |
813 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
814 | */ | |
815 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
9163582c MP |
816 | for (__node = 0; __node < nr_node_ids; __node++) \ |
817 | if ((__n = get_node(__s, __node))) | |
44c5356f CL |
818 | |
819 | #endif | |
820 | ||
852d8be0 YS |
821 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
822 | void dump_unreclaimable_slab(void); | |
823 | #else | |
824 | static inline void dump_unreclaimable_slab(void) | |
825 | { | |
826 | } | |
827 | #endif | |
828 | ||
55834c59 AP |
829 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr); |
830 | ||
7c00fce9 TG |
831 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
832 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
833 | gfp_t gfp); | |
834 | void cache_random_seq_destroy(struct kmem_cache *cachep); | |
835 | #else | |
836 | static inline int cache_random_seq_create(struct kmem_cache *cachep, | |
837 | unsigned int count, gfp_t gfp) | |
838 | { | |
839 | return 0; | |
840 | } | |
841 | static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { } | |
842 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
843 | ||
6471384a AP |
844 | static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c) |
845 | { | |
51cba1eb KC |
846 | if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON, |
847 | &init_on_alloc)) { | |
6471384a AP |
848 | if (c->ctor) |
849 | return false; | |
850 | if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) | |
851 | return flags & __GFP_ZERO; | |
852 | return true; | |
853 | } | |
854 | return flags & __GFP_ZERO; | |
855 | } | |
856 | ||
857 | static inline bool slab_want_init_on_free(struct kmem_cache *c) | |
858 | { | |
51cba1eb KC |
859 | if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON, |
860 | &init_on_free)) | |
6471384a AP |
861 | return !(c->ctor || |
862 | (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))); | |
863 | return false; | |
864 | } | |
865 | ||
64dd6849 FM |
866 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
867 | void debugfs_slab_release(struct kmem_cache *); | |
868 | #else | |
869 | static inline void debugfs_slab_release(struct kmem_cache *s) { } | |
870 | #endif | |
871 | ||
5bb1bb35 | 872 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
873 | #define KS_ADDRS_COUNT 16 |
874 | struct kmem_obj_info { | |
875 | void *kp_ptr; | |
7213230a | 876 | struct slab *kp_slab; |
8e7f37f2 PM |
877 | void *kp_objp; |
878 | unsigned long kp_data_offset; | |
879 | struct kmem_cache *kp_slab_cache; | |
880 | void *kp_ret; | |
881 | void *kp_stack[KS_ADDRS_COUNT]; | |
e548eaa1 | 882 | void *kp_free_stack[KS_ADDRS_COUNT]; |
8e7f37f2 | 883 | }; |
2dfe63e6 | 884 | void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab); |
5bb1bb35 | 885 | #endif |
8e7f37f2 | 886 | |
0b3eb091 MWO |
887 | #ifdef CONFIG_HAVE_HARDENED_USERCOPY_ALLOCATOR |
888 | void __check_heap_object(const void *ptr, unsigned long n, | |
889 | const struct slab *slab, bool to_user); | |
890 | #else | |
891 | static inline | |
892 | void __check_heap_object(const void *ptr, unsigned long n, | |
893 | const struct slab *slab, bool to_user) | |
894 | { | |
895 | } | |
896 | #endif | |
897 | ||
946fa0db FT |
898 | #ifdef CONFIG_SLUB_DEBUG |
899 | void skip_orig_size_check(struct kmem_cache *s, const void *object); | |
900 | #endif | |
901 | ||
5240ab40 | 902 | #endif /* MM_SLAB_H */ |