1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Written by Mark Hemment, 1996 ([email protected]). 4 * 5 * (C) SGI 2006, Christoph Lameter 6 * Cleaned up and restructured to ease the addition of alternative 7 * implementations of SLAB allocators. 8 * (C) Linux Foundation 2008-2013 9 * Unified interface for all slab allocators 10 */ 11 12 #ifndef _LINUX_SLAB_H 13 #define _LINUX_SLAB_H 14 15 #include <linux/cache.h> 16 #include <linux/gfp.h> 17 #include <linux/overflow.h> 18 #include <linux/types.h> 19 #include <linux/workqueue.h> 20 #include <linux/percpu-refcount.h> 21 #include <linux/cleanup.h> 22 #include <linux/hash.h> 23 24 enum _slab_flag_bits { 25 _SLAB_CONSISTENCY_CHECKS, 26 _SLAB_RED_ZONE, 27 _SLAB_POISON, 28 _SLAB_KMALLOC, 29 _SLAB_HWCACHE_ALIGN, 30 _SLAB_CACHE_DMA, 31 _SLAB_CACHE_DMA32, 32 _SLAB_STORE_USER, 33 _SLAB_PANIC, 34 _SLAB_TYPESAFE_BY_RCU, 35 _SLAB_TRACE, 36 #ifdef CONFIG_DEBUG_OBJECTS 37 _SLAB_DEBUG_OBJECTS, 38 #endif 39 _SLAB_NOLEAKTRACE, 40 _SLAB_NO_MERGE, 41 #ifdef CONFIG_FAILSLAB 42 _SLAB_FAILSLAB, 43 #endif 44 #ifdef CONFIG_MEMCG 45 _SLAB_ACCOUNT, 46 #endif 47 #ifdef CONFIG_KASAN_GENERIC 48 _SLAB_KASAN, 49 #endif 50 _SLAB_NO_USER_FLAGS, 51 #ifdef CONFIG_KFENCE 52 _SLAB_SKIP_KFENCE, 53 #endif 54 #ifndef CONFIG_SLUB_TINY 55 _SLAB_RECLAIM_ACCOUNT, 56 #endif 57 _SLAB_OBJECT_POISON, 58 _SLAB_CMPXCHG_DOUBLE, 59 #ifdef CONFIG_SLAB_OBJ_EXT 60 _SLAB_NO_OBJ_EXT, 61 #endif 62 _SLAB_FLAGS_LAST_BIT 63 }; 64 65 #define __SLAB_FLAG_BIT(nr) ((slab_flags_t __force)(1U << (nr))) 66 #define __SLAB_FLAG_UNUSED ((slab_flags_t __force)(0U)) 67 68 /* 69 * Flags to pass to kmem_cache_create(). 70 * The ones marked DEBUG need CONFIG_SLUB_DEBUG enabled, otherwise are no-op 71 */ 72 /* DEBUG: Perform (expensive) checks on alloc/free */ 73 #define SLAB_CONSISTENCY_CHECKS __SLAB_FLAG_BIT(_SLAB_CONSISTENCY_CHECKS) 74 /* DEBUG: Red zone objs in a cache */ 75 #define SLAB_RED_ZONE __SLAB_FLAG_BIT(_SLAB_RED_ZONE) 76 /* DEBUG: Poison objects */ 77 #define SLAB_POISON __SLAB_FLAG_BIT(_SLAB_POISON) 78 /* Indicate a kmalloc slab */ 79 #define SLAB_KMALLOC __SLAB_FLAG_BIT(_SLAB_KMALLOC) 80 /* Align objs on cache lines */ 81 #define SLAB_HWCACHE_ALIGN __SLAB_FLAG_BIT(_SLAB_HWCACHE_ALIGN) 82 /* Use GFP_DMA memory */ 83 #define SLAB_CACHE_DMA __SLAB_FLAG_BIT(_SLAB_CACHE_DMA) 84 /* Use GFP_DMA32 memory */ 85 #define SLAB_CACHE_DMA32 __SLAB_FLAG_BIT(_SLAB_CACHE_DMA32) 86 /* DEBUG: Store the last owner for bug hunting */ 87 #define SLAB_STORE_USER __SLAB_FLAG_BIT(_SLAB_STORE_USER) 88 /* Panic if kmem_cache_create() fails */ 89 #define SLAB_PANIC __SLAB_FLAG_BIT(_SLAB_PANIC) 90 /* 91 * SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS! 92 * 93 * This delays freeing the SLAB page by a grace period, it does _NOT_ 94 * delay object freeing. This means that if you do kmem_cache_free() 95 * that memory location is free to be reused at any time. Thus it may 96 * be possible to see another object there in the same RCU grace period. 97 * 98 * This feature only ensures the memory location backing the object 99 * stays valid, the trick to using this is relying on an independent 100 * object validation pass. Something like: 101 * 102 * begin: 103 * rcu_read_lock(); 104 * obj = lockless_lookup(key); 105 * if (obj) { 106 * if (!try_get_ref(obj)) // might fail for free objects 107 * rcu_read_unlock(); 108 * goto begin; 109 * 110 * if (obj->key != key) { // not the object we expected 111 * put_ref(obj); 112 * rcu_read_unlock(); 113 * goto begin; 114 * } 115 * } 116 * rcu_read_unlock(); 117 * 118 * This is useful if we need to approach a kernel structure obliquely, 119 * from its address obtained without the usual locking. We can lock 120 * the structure to stabilize it and check it's still at the given address, 121 * only if we can be sure that the memory has not been meanwhile reused 122 * for some other kind of object (which our subsystem's lock might corrupt). 123 * 124 * rcu_read_lock before reading the address, then rcu_read_unlock after 125 * taking the spinlock within the structure expected at that address. 126 * 127 * Note that it is not possible to acquire a lock within a structure 128 * allocated with SLAB_TYPESAFE_BY_RCU without first acquiring a reference 129 * as described above. The reason is that SLAB_TYPESAFE_BY_RCU pages 130 * are not zeroed before being given to the slab, which means that any 131 * locks must be initialized after each and every kmem_struct_alloc(). 132 * Alternatively, make the ctor passed to kmem_cache_create() initialize 133 * the locks at page-allocation time, as is done in __i915_request_ctor(), 134 * sighand_ctor(), and anon_vma_ctor(). Such a ctor permits readers 135 * to safely acquire those ctor-initialized locks under rcu_read_lock() 136 * protection. 137 * 138 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU. 139 */ 140 /* Defer freeing slabs to RCU */ 141 #define SLAB_TYPESAFE_BY_RCU __SLAB_FLAG_BIT(_SLAB_TYPESAFE_BY_RCU) 142 /* Trace allocations and frees */ 143 #define SLAB_TRACE __SLAB_FLAG_BIT(_SLAB_TRACE) 144 145 /* Flag to prevent checks on free */ 146 #ifdef CONFIG_DEBUG_OBJECTS 147 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_BIT(_SLAB_DEBUG_OBJECTS) 148 #else 149 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_UNUSED 150 #endif 151 152 /* Avoid kmemleak tracing */ 153 #define SLAB_NOLEAKTRACE __SLAB_FLAG_BIT(_SLAB_NOLEAKTRACE) 154 155 /* 156 * Prevent merging with compatible kmem caches. This flag should be used 157 * cautiously. Valid use cases: 158 * 159 * - caches created for self-tests (e.g. kunit) 160 * - general caches created and used by a subsystem, only when a 161 * (subsystem-specific) debug option is enabled 162 * - performance critical caches, should be very rare and consulted with slab 163 * maintainers, and not used together with CONFIG_SLUB_TINY 164 */ 165 #define SLAB_NO_MERGE __SLAB_FLAG_BIT(_SLAB_NO_MERGE) 166 167 /* Fault injection mark */ 168 #ifdef CONFIG_FAILSLAB 169 # define SLAB_FAILSLAB __SLAB_FLAG_BIT(_SLAB_FAILSLAB) 170 #else 171 # define SLAB_FAILSLAB __SLAB_FLAG_UNUSED 172 #endif 173 /* Account to memcg */ 174 #ifdef CONFIG_MEMCG 175 # define SLAB_ACCOUNT __SLAB_FLAG_BIT(_SLAB_ACCOUNT) 176 #else 177 # define SLAB_ACCOUNT __SLAB_FLAG_UNUSED 178 #endif 179 180 #ifdef CONFIG_KASAN_GENERIC 181 #define SLAB_KASAN __SLAB_FLAG_BIT(_SLAB_KASAN) 182 #else 183 #define SLAB_KASAN __SLAB_FLAG_UNUSED 184 #endif 185 186 /* 187 * Ignore user specified debugging flags. 188 * Intended for caches created for self-tests so they have only flags 189 * specified in the code and other flags are ignored. 190 */ 191 #define SLAB_NO_USER_FLAGS __SLAB_FLAG_BIT(_SLAB_NO_USER_FLAGS) 192 193 #ifdef CONFIG_KFENCE 194 #define SLAB_SKIP_KFENCE __SLAB_FLAG_BIT(_SLAB_SKIP_KFENCE) 195 #else 196 #define SLAB_SKIP_KFENCE __SLAB_FLAG_UNUSED 197 #endif 198 199 /* The following flags affect the page allocator grouping pages by mobility */ 200 /* Objects are reclaimable */ 201 #ifndef CONFIG_SLUB_TINY 202 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_BIT(_SLAB_RECLAIM_ACCOUNT) 203 #else 204 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_UNUSED 205 #endif 206 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */ 207 208 /* Slab created using create_boot_cache */ 209 #ifdef CONFIG_SLAB_OBJ_EXT 210 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_BIT(_SLAB_NO_OBJ_EXT) 211 #else 212 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_UNUSED 213 #endif 214 215 /* 216 * freeptr_t represents a SLUB freelist pointer, which might be encoded 217 * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled. 218 */ 219 typedef struct { unsigned long v; } freeptr_t; 220 221 /* 222 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests. 223 * 224 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault. 225 * 226 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can. 227 * Both make kfree a no-op. 228 */ 229 #define ZERO_SIZE_PTR ((void *)16) 230 231 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \ 232 (unsigned long)ZERO_SIZE_PTR) 233 234 #include <linux/kasan.h> 235 236 struct list_lru; 237 struct mem_cgroup; 238 /* 239 * struct kmem_cache related prototypes 240 */ 241 bool slab_is_available(void); 242 243 /** 244 * struct kmem_cache_args - Less common arguments for kmem_cache_create() 245 * @align: The required alignment for the objects. 246 * @useroffset: Usercopy region offset 247 * @usersize: Usercopy region size 248 * @freeptr_offset: Custom offset for the free pointer in RCU caches 249 * @use_freeptr_offset: Whether a @freeptr_offset is used 250 * @ctor: A constructor for the objects. 251 */ 252 struct kmem_cache_args { 253 unsigned int align; 254 unsigned int useroffset; 255 unsigned int usersize; 256 unsigned int freeptr_offset; 257 bool use_freeptr_offset; 258 void (*ctor)(void *); 259 }; 260 261 struct kmem_cache *__kmem_cache_create_args(const char *name, 262 unsigned int object_size, 263 struct kmem_cache_args *args, 264 slab_flags_t flags); 265 266 struct kmem_cache *__kmem_cache_create(const char *name, unsigned int size, 267 unsigned int align, slab_flags_t flags, 268 void (*ctor)(void *)); 269 struct kmem_cache *kmem_cache_create_usercopy(const char *name, 270 unsigned int size, unsigned int align, 271 slab_flags_t flags, 272 unsigned int useroffset, unsigned int usersize, 273 void (*ctor)(void *)); 274 struct kmem_cache *kmem_cache_create_rcu(const char *name, unsigned int size, 275 unsigned int freeptr_offset, 276 slab_flags_t flags); 277 278 /* If NULL is passed for @args, use this variant with default arguments. */ 279 static inline struct kmem_cache * 280 __kmem_cache_default_args(const char *name, unsigned int size, 281 struct kmem_cache_args *args, 282 slab_flags_t flags) 283 { 284 struct kmem_cache_args kmem_default_args = {}; 285 286 /* Make sure we don't get passed garbage. */ 287 if (WARN_ON_ONCE(args)) 288 return ERR_PTR(-EINVAL); 289 290 return __kmem_cache_create_args(name, size, &kmem_default_args, flags); 291 } 292 293 #define kmem_cache_create(__name, __object_size, __args, ...) \ 294 _Generic((__args), \ 295 struct kmem_cache_args *: __kmem_cache_create_args, \ 296 void *: __kmem_cache_default_args, \ 297 default: __kmem_cache_create)(__name, __object_size, __args, __VA_ARGS__) 298 299 void kmem_cache_destroy(struct kmem_cache *s); 300 int kmem_cache_shrink(struct kmem_cache *s); 301 302 /* 303 * Please use this macro to create slab caches. Simply specify the 304 * name of the structure and maybe some flags that are listed above. 305 * 306 * The alignment of the struct determines object alignment. If you 307 * f.e. add ____cacheline_aligned_in_smp to the struct declaration 308 * then the objects will be properly aligned in SMP configurations. 309 */ 310 #define KMEM_CACHE(__struct, __flags) \ 311 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \ 312 &(struct kmem_cache_args) { \ 313 .align = __alignof__(struct __struct), \ 314 }, (__flags)) 315 316 /* 317 * To whitelist a single field for copying to/from usercopy, use this 318 * macro instead for KMEM_CACHE() above. 319 */ 320 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \ 321 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \ 322 &(struct kmem_cache_args) { \ 323 .align = __alignof__(struct __struct), \ 324 .useroffset = offsetof(struct __struct, __field), \ 325 .usersize = sizeof_field(struct __struct, __field), \ 326 }, (__flags)) 327 328 /* 329 * Common kmalloc functions provided by all allocators 330 */ 331 void * __must_check krealloc_noprof(const void *objp, size_t new_size, 332 gfp_t flags) __realloc_size(2); 333 #define krealloc(...) alloc_hooks(krealloc_noprof(__VA_ARGS__)) 334 335 void kfree(const void *objp); 336 void kfree_sensitive(const void *objp); 337 size_t __ksize(const void *objp); 338 339 DEFINE_FREE(kfree, void *, if (!IS_ERR_OR_NULL(_T)) kfree(_T)) 340 341 /** 342 * ksize - Report actual allocation size of associated object 343 * 344 * @objp: Pointer returned from a prior kmalloc()-family allocation. 345 * 346 * This should not be used for writing beyond the originally requested 347 * allocation size. Either use krealloc() or round up the allocation size 348 * with kmalloc_size_roundup() prior to allocation. If this is used to 349 * access beyond the originally requested allocation size, UBSAN_BOUNDS 350 * and/or FORTIFY_SOURCE may trip, since they only know about the 351 * originally allocated size via the __alloc_size attribute. 352 */ 353 size_t ksize(const void *objp); 354 355 #ifdef CONFIG_PRINTK 356 bool kmem_dump_obj(void *object); 357 #else 358 static inline bool kmem_dump_obj(void *object) { return false; } 359 #endif 360 361 /* 362 * Some archs want to perform DMA into kmalloc caches and need a guaranteed 363 * alignment larger than the alignment of a 64-bit integer. 364 * Setting ARCH_DMA_MINALIGN in arch headers allows that. 365 */ 366 #ifdef ARCH_HAS_DMA_MINALIGN 367 #if ARCH_DMA_MINALIGN > 8 && !defined(ARCH_KMALLOC_MINALIGN) 368 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN 369 #endif 370 #endif 371 372 #ifndef ARCH_KMALLOC_MINALIGN 373 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long) 374 #elif ARCH_KMALLOC_MINALIGN > 8 375 #define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN 376 #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE) 377 #endif 378 379 /* 380 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment. 381 * Intended for arches that get misalignment faults even for 64 bit integer 382 * aligned buffers. 383 */ 384 #ifndef ARCH_SLAB_MINALIGN 385 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long) 386 #endif 387 388 /* 389 * Arches can define this function if they want to decide the minimum slab 390 * alignment at runtime. The value returned by the function must be a power 391 * of two and >= ARCH_SLAB_MINALIGN. 392 */ 393 #ifndef arch_slab_minalign 394 static inline unsigned int arch_slab_minalign(void) 395 { 396 return ARCH_SLAB_MINALIGN; 397 } 398 #endif 399 400 /* 401 * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN. 402 * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN 403 * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment. 404 */ 405 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN) 406 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN) 407 #define __assume_page_alignment __assume_aligned(PAGE_SIZE) 408 409 /* 410 * Kmalloc array related definitions 411 */ 412 413 /* 414 * SLUB directly allocates requests fitting in to an order-1 page 415 * (PAGE_SIZE*2). Larger requests are passed to the page allocator. 416 */ 417 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1) 418 #define KMALLOC_SHIFT_MAX (MAX_PAGE_ORDER + PAGE_SHIFT) 419 #ifndef KMALLOC_SHIFT_LOW 420 #define KMALLOC_SHIFT_LOW 3 421 #endif 422 423 /* Maximum allocatable size */ 424 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX) 425 /* Maximum size for which we actually use a slab cache */ 426 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH) 427 /* Maximum order allocatable via the slab allocator */ 428 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT) 429 430 /* 431 * Kmalloc subsystem. 432 */ 433 #ifndef KMALLOC_MIN_SIZE 434 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW) 435 #endif 436 437 /* 438 * This restriction comes from byte sized index implementation. 439 * Page size is normally 2^12 bytes and, in this case, if we want to use 440 * byte sized index which can represent 2^8 entries, the size of the object 441 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16. 442 * If minimum size of kmalloc is less than 16, we use it as minimum object 443 * size and give up to use byte sized index. 444 */ 445 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \ 446 (KMALLOC_MIN_SIZE) : 16) 447 448 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 449 #define RANDOM_KMALLOC_CACHES_NR 15 // # of cache copies 450 #else 451 #define RANDOM_KMALLOC_CACHES_NR 0 452 #endif 453 454 /* 455 * Whenever changing this, take care of that kmalloc_type() and 456 * create_kmalloc_caches() still work as intended. 457 * 458 * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP 459 * is for accounted but unreclaimable and non-dma objects. All the other 460 * kmem caches can have both accounted and unaccounted objects. 461 */ 462 enum kmalloc_cache_type { 463 KMALLOC_NORMAL = 0, 464 #ifndef CONFIG_ZONE_DMA 465 KMALLOC_DMA = KMALLOC_NORMAL, 466 #endif 467 #ifndef CONFIG_MEMCG 468 KMALLOC_CGROUP = KMALLOC_NORMAL, 469 #endif 470 KMALLOC_RANDOM_START = KMALLOC_NORMAL, 471 KMALLOC_RANDOM_END = KMALLOC_RANDOM_START + RANDOM_KMALLOC_CACHES_NR, 472 #ifdef CONFIG_SLUB_TINY 473 KMALLOC_RECLAIM = KMALLOC_NORMAL, 474 #else 475 KMALLOC_RECLAIM, 476 #endif 477 #ifdef CONFIG_ZONE_DMA 478 KMALLOC_DMA, 479 #endif 480 #ifdef CONFIG_MEMCG 481 KMALLOC_CGROUP, 482 #endif 483 NR_KMALLOC_TYPES 484 }; 485 486 typedef struct kmem_cache * kmem_buckets[KMALLOC_SHIFT_HIGH + 1]; 487 488 extern kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES]; 489 490 /* 491 * Define gfp bits that should not be set for KMALLOC_NORMAL. 492 */ 493 #define KMALLOC_NOT_NORMAL_BITS \ 494 (__GFP_RECLAIMABLE | \ 495 (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \ 496 (IS_ENABLED(CONFIG_MEMCG) ? __GFP_ACCOUNT : 0)) 497 498 extern unsigned long random_kmalloc_seed; 499 500 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags, unsigned long caller) 501 { 502 /* 503 * The most common case is KMALLOC_NORMAL, so test for it 504 * with a single branch for all the relevant flags. 505 */ 506 if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0)) 507 #ifdef CONFIG_RANDOM_KMALLOC_CACHES 508 /* RANDOM_KMALLOC_CACHES_NR (=15) copies + the KMALLOC_NORMAL */ 509 return KMALLOC_RANDOM_START + hash_64(caller ^ random_kmalloc_seed, 510 ilog2(RANDOM_KMALLOC_CACHES_NR + 1)); 511 #else 512 return KMALLOC_NORMAL; 513 #endif 514 515 /* 516 * At least one of the flags has to be set. Their priorities in 517 * decreasing order are: 518 * 1) __GFP_DMA 519 * 2) __GFP_RECLAIMABLE 520 * 3) __GFP_ACCOUNT 521 */ 522 if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA)) 523 return KMALLOC_DMA; 524 if (!IS_ENABLED(CONFIG_MEMCG) || (flags & __GFP_RECLAIMABLE)) 525 return KMALLOC_RECLAIM; 526 else 527 return KMALLOC_CGROUP; 528 } 529 530 /* 531 * Figure out which kmalloc slab an allocation of a certain size 532 * belongs to. 533 * 0 = zero alloc 534 * 1 = 65 .. 96 bytes 535 * 2 = 129 .. 192 bytes 536 * n = 2^(n-1)+1 .. 2^n 537 * 538 * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized; 539 * typical usage is via kmalloc_index() and therefore evaluated at compile-time. 540 * Callers where !size_is_constant should only be test modules, where runtime 541 * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab(). 542 */ 543 static __always_inline unsigned int __kmalloc_index(size_t size, 544 bool size_is_constant) 545 { 546 if (!size) 547 return 0; 548 549 if (size <= KMALLOC_MIN_SIZE) 550 return KMALLOC_SHIFT_LOW; 551 552 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96) 553 return 1; 554 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192) 555 return 2; 556 if (size <= 8) return 3; 557 if (size <= 16) return 4; 558 if (size <= 32) return 5; 559 if (size <= 64) return 6; 560 if (size <= 128) return 7; 561 if (size <= 256) return 8; 562 if (size <= 512) return 9; 563 if (size <= 1024) return 10; 564 if (size <= 2 * 1024) return 11; 565 if (size <= 4 * 1024) return 12; 566 if (size <= 8 * 1024) return 13; 567 if (size <= 16 * 1024) return 14; 568 if (size <= 32 * 1024) return 15; 569 if (size <= 64 * 1024) return 16; 570 if (size <= 128 * 1024) return 17; 571 if (size <= 256 * 1024) return 18; 572 if (size <= 512 * 1024) return 19; 573 if (size <= 1024 * 1024) return 20; 574 if (size <= 2 * 1024 * 1024) return 21; 575 576 if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant) 577 BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()"); 578 else 579 BUG(); 580 581 /* Will never be reached. Needed because the compiler may complain */ 582 return -1; 583 } 584 static_assert(PAGE_SHIFT <= 20); 585 #define kmalloc_index(s) __kmalloc_index(s, true) 586 587 #include <linux/alloc_tag.h> 588 589 /** 590 * kmem_cache_alloc - Allocate an object 591 * @cachep: The cache to allocate from. 592 * @flags: See kmalloc(). 593 * 594 * Allocate an object from this cache. 595 * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags. 596 * 597 * Return: pointer to the new object or %NULL in case of error 598 */ 599 void *kmem_cache_alloc_noprof(struct kmem_cache *cachep, 600 gfp_t flags) __assume_slab_alignment __malloc; 601 #define kmem_cache_alloc(...) alloc_hooks(kmem_cache_alloc_noprof(__VA_ARGS__)) 602 603 void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru, 604 gfp_t gfpflags) __assume_slab_alignment __malloc; 605 #define kmem_cache_alloc_lru(...) alloc_hooks(kmem_cache_alloc_lru_noprof(__VA_ARGS__)) 606 607 void kmem_cache_free(struct kmem_cache *s, void *objp); 608 609 kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags, 610 unsigned int useroffset, unsigned int usersize, 611 void (*ctor)(void *)); 612 613 /* 614 * Bulk allocation and freeing operations. These are accelerated in an 615 * allocator specific way to avoid taking locks repeatedly or building 616 * metadata structures unnecessarily. 617 * 618 * Note that interrupts must be enabled when calling these functions. 619 */ 620 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p); 621 622 int kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags, size_t size, void **p); 623 #define kmem_cache_alloc_bulk(...) alloc_hooks(kmem_cache_alloc_bulk_noprof(__VA_ARGS__)) 624 625 static __always_inline void kfree_bulk(size_t size, void **p) 626 { 627 kmem_cache_free_bulk(NULL, size, p); 628 } 629 630 void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t flags, 631 int node) __assume_slab_alignment __malloc; 632 #define kmem_cache_alloc_node(...) alloc_hooks(kmem_cache_alloc_node_noprof(__VA_ARGS__)) 633 634 /* 635 * These macros allow declaring a kmem_buckets * parameter alongside size, which 636 * can be compiled out with CONFIG_SLAB_BUCKETS=n so that a large number of call 637 * sites don't have to pass NULL. 638 */ 639 #ifdef CONFIG_SLAB_BUCKETS 640 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size), kmem_buckets *(_b) 641 #define PASS_BUCKET_PARAMS(_size, _b) (_size), (_b) 642 #define PASS_BUCKET_PARAM(_b) (_b) 643 #else 644 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size) 645 #define PASS_BUCKET_PARAMS(_size, _b) (_size) 646 #define PASS_BUCKET_PARAM(_b) NULL 647 #endif 648 649 /* 650 * The following functions are not to be used directly and are intended only 651 * for internal use from kmalloc() and kmalloc_node() 652 * with the exception of kunit tests 653 */ 654 655 void *__kmalloc_noprof(size_t size, gfp_t flags) 656 __assume_kmalloc_alignment __alloc_size(1); 657 658 void *__kmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node) 659 __assume_kmalloc_alignment __alloc_size(1); 660 661 void *__kmalloc_cache_noprof(struct kmem_cache *s, gfp_t flags, size_t size) 662 __assume_kmalloc_alignment __alloc_size(3); 663 664 void *__kmalloc_cache_node_noprof(struct kmem_cache *s, gfp_t gfpflags, 665 int node, size_t size) 666 __assume_kmalloc_alignment __alloc_size(4); 667 668 void *__kmalloc_large_noprof(size_t size, gfp_t flags) 669 __assume_page_alignment __alloc_size(1); 670 671 void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node) 672 __assume_page_alignment __alloc_size(1); 673 674 /** 675 * kmalloc - allocate kernel memory 676 * @size: how many bytes of memory are required. 677 * @flags: describe the allocation context 678 * 679 * kmalloc is the normal method of allocating memory 680 * for objects smaller than page size in the kernel. 681 * 682 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN 683 * bytes. For @size of power of two bytes, the alignment is also guaranteed 684 * to be at least to the size. For other sizes, the alignment is guaranteed to 685 * be at least the largest power-of-two divisor of @size. 686 * 687 * The @flags argument may be one of the GFP flags defined at 688 * include/linux/gfp_types.h and described at 689 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>` 690 * 691 * The recommended usage of the @flags is described at 692 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>` 693 * 694 * Below is a brief outline of the most useful GFP flags 695 * 696 * %GFP_KERNEL 697 * Allocate normal kernel ram. May sleep. 698 * 699 * %GFP_NOWAIT 700 * Allocation will not sleep. 701 * 702 * %GFP_ATOMIC 703 * Allocation will not sleep. May use emergency pools. 704 * 705 * Also it is possible to set different flags by OR'ing 706 * in one or more of the following additional @flags: 707 * 708 * %__GFP_ZERO 709 * Zero the allocated memory before returning. Also see kzalloc(). 710 * 711 * %__GFP_HIGH 712 * This allocation has high priority and may use emergency pools. 713 * 714 * %__GFP_NOFAIL 715 * Indicate that this allocation is in no way allowed to fail 716 * (think twice before using). 717 * 718 * %__GFP_NORETRY 719 * If memory is not immediately available, 720 * then give up at once. 721 * 722 * %__GFP_NOWARN 723 * If allocation fails, don't issue any warnings. 724 * 725 * %__GFP_RETRY_MAYFAIL 726 * Try really hard to succeed the allocation but fail 727 * eventually. 728 */ 729 static __always_inline __alloc_size(1) void *kmalloc_noprof(size_t size, gfp_t flags) 730 { 731 if (__builtin_constant_p(size) && size) { 732 unsigned int index; 733 734 if (size > KMALLOC_MAX_CACHE_SIZE) 735 return __kmalloc_large_noprof(size, flags); 736 737 index = kmalloc_index(size); 738 return __kmalloc_cache_noprof( 739 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index], 740 flags, size); 741 } 742 return __kmalloc_noprof(size, flags); 743 } 744 #define kmalloc(...) alloc_hooks(kmalloc_noprof(__VA_ARGS__)) 745 746 #define kmem_buckets_alloc(_b, _size, _flags) \ 747 alloc_hooks(__kmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE)) 748 749 #define kmem_buckets_alloc_track_caller(_b, _size, _flags) \ 750 alloc_hooks(__kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE, _RET_IP_)) 751 752 static __always_inline __alloc_size(1) void *kmalloc_node_noprof(size_t size, gfp_t flags, int node) 753 { 754 if (__builtin_constant_p(size) && size) { 755 unsigned int index; 756 757 if (size > KMALLOC_MAX_CACHE_SIZE) 758 return __kmalloc_large_node_noprof(size, flags, node); 759 760 index = kmalloc_index(size); 761 return __kmalloc_cache_node_noprof( 762 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index], 763 flags, node, size); 764 } 765 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node); 766 } 767 #define kmalloc_node(...) alloc_hooks(kmalloc_node_noprof(__VA_ARGS__)) 768 769 /** 770 * kmalloc_array - allocate memory for an array. 771 * @n: number of elements. 772 * @size: element size. 773 * @flags: the type of memory to allocate (see kmalloc). 774 */ 775 static inline __alloc_size(1, 2) void *kmalloc_array_noprof(size_t n, size_t size, gfp_t flags) 776 { 777 size_t bytes; 778 779 if (unlikely(check_mul_overflow(n, size, &bytes))) 780 return NULL; 781 if (__builtin_constant_p(n) && __builtin_constant_p(size)) 782 return kmalloc_noprof(bytes, flags); 783 return kmalloc_noprof(bytes, flags); 784 } 785 #define kmalloc_array(...) alloc_hooks(kmalloc_array_noprof(__VA_ARGS__)) 786 787 /** 788 * krealloc_array - reallocate memory for an array. 789 * @p: pointer to the memory chunk to reallocate 790 * @new_n: new number of elements to alloc 791 * @new_size: new size of a single member of the array 792 * @flags: the type of memory to allocate (see kmalloc) 793 */ 794 static inline __realloc_size(2, 3) void * __must_check krealloc_array_noprof(void *p, 795 size_t new_n, 796 size_t new_size, 797 gfp_t flags) 798 { 799 size_t bytes; 800 801 if (unlikely(check_mul_overflow(new_n, new_size, &bytes))) 802 return NULL; 803 804 return krealloc_noprof(p, bytes, flags); 805 } 806 #define krealloc_array(...) alloc_hooks(krealloc_array_noprof(__VA_ARGS__)) 807 808 /** 809 * kcalloc - allocate memory for an array. The memory is set to zero. 810 * @n: number of elements. 811 * @size: element size. 812 * @flags: the type of memory to allocate (see kmalloc). 813 */ 814 #define kcalloc(n, size, flags) kmalloc_array(n, size, (flags) | __GFP_ZERO) 815 816 void *__kmalloc_node_track_caller_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node, 817 unsigned long caller) __alloc_size(1); 818 #define kmalloc_node_track_caller_noprof(size, flags, node, caller) \ 819 __kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node, caller) 820 #define kmalloc_node_track_caller(...) \ 821 alloc_hooks(kmalloc_node_track_caller_noprof(__VA_ARGS__, _RET_IP_)) 822 823 /* 824 * kmalloc_track_caller is a special version of kmalloc that records the 825 * calling function of the routine calling it for slab leak tracking instead 826 * of just the calling function (confusing, eh?). 827 * It's useful when the call to kmalloc comes from a widely-used standard 828 * allocator where we care about the real place the memory allocation 829 * request comes from. 830 */ 831 #define kmalloc_track_caller(...) kmalloc_node_track_caller(__VA_ARGS__, NUMA_NO_NODE) 832 833 #define kmalloc_track_caller_noprof(...) \ 834 kmalloc_node_track_caller_noprof(__VA_ARGS__, NUMA_NO_NODE, _RET_IP_) 835 836 static inline __alloc_size(1, 2) void *kmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, 837 int node) 838 { 839 size_t bytes; 840 841 if (unlikely(check_mul_overflow(n, size, &bytes))) 842 return NULL; 843 if (__builtin_constant_p(n) && __builtin_constant_p(size)) 844 return kmalloc_node_noprof(bytes, flags, node); 845 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(bytes, NULL), flags, node); 846 } 847 #define kmalloc_array_node(...) alloc_hooks(kmalloc_array_node_noprof(__VA_ARGS__)) 848 849 #define kcalloc_node(_n, _size, _flags, _node) \ 850 kmalloc_array_node(_n, _size, (_flags) | __GFP_ZERO, _node) 851 852 /* 853 * Shortcuts 854 */ 855 #define kmem_cache_zalloc(_k, _flags) kmem_cache_alloc(_k, (_flags)|__GFP_ZERO) 856 857 /** 858 * kzalloc - allocate memory. The memory is set to zero. 859 * @size: how many bytes of memory are required. 860 * @flags: the type of memory to allocate (see kmalloc). 861 */ 862 static inline __alloc_size(1) void *kzalloc_noprof(size_t size, gfp_t flags) 863 { 864 return kmalloc_noprof(size, flags | __GFP_ZERO); 865 } 866 #define kzalloc(...) alloc_hooks(kzalloc_noprof(__VA_ARGS__)) 867 #define kzalloc_node(_size, _flags, _node) kmalloc_node(_size, (_flags)|__GFP_ZERO, _node) 868 869 void *__kvmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node) __alloc_size(1); 870 #define kvmalloc_node_noprof(size, flags, node) \ 871 __kvmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node) 872 #define kvmalloc_node(...) alloc_hooks(kvmalloc_node_noprof(__VA_ARGS__)) 873 874 #define kvmalloc(_size, _flags) kvmalloc_node(_size, _flags, NUMA_NO_NODE) 875 #define kvmalloc_noprof(_size, _flags) kvmalloc_node_noprof(_size, _flags, NUMA_NO_NODE) 876 #define kvzalloc(_size, _flags) kvmalloc(_size, (_flags)|__GFP_ZERO) 877 878 #define kvzalloc_node(_size, _flags, _node) kvmalloc_node(_size, (_flags)|__GFP_ZERO, _node) 879 #define kmem_buckets_valloc(_b, _size, _flags) \ 880 alloc_hooks(__kvmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE)) 881 882 static inline __alloc_size(1, 2) void * 883 kvmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, int node) 884 { 885 size_t bytes; 886 887 if (unlikely(check_mul_overflow(n, size, &bytes))) 888 return NULL; 889 890 return kvmalloc_node_noprof(bytes, flags, node); 891 } 892 893 #define kvmalloc_array_noprof(...) kvmalloc_array_node_noprof(__VA_ARGS__, NUMA_NO_NODE) 894 #define kvcalloc_node_noprof(_n,_s,_f,_node) kvmalloc_array_node_noprof(_n,_s,(_f)|__GFP_ZERO,_node) 895 #define kvcalloc_noprof(...) kvcalloc_node_noprof(__VA_ARGS__, NUMA_NO_NODE) 896 897 #define kvmalloc_array(...) alloc_hooks(kvmalloc_array_noprof(__VA_ARGS__)) 898 #define kvcalloc_node(...) alloc_hooks(kvcalloc_node_noprof(__VA_ARGS__)) 899 #define kvcalloc(...) alloc_hooks(kvcalloc_noprof(__VA_ARGS__)) 900 901 extern void *kvrealloc_noprof(const void *p, size_t oldsize, size_t newsize, gfp_t flags) 902 __realloc_size(3); 903 #define kvrealloc(...) alloc_hooks(kvrealloc_noprof(__VA_ARGS__)) 904 905 extern void kvfree(const void *addr); 906 DEFINE_FREE(kvfree, void *, if (!IS_ERR_OR_NULL(_T)) kvfree(_T)) 907 908 extern void kvfree_sensitive(const void *addr, size_t len); 909 910 unsigned int kmem_cache_size(struct kmem_cache *s); 911 912 /** 913 * kmalloc_size_roundup - Report allocation bucket size for the given size 914 * 915 * @size: Number of bytes to round up from. 916 * 917 * This returns the number of bytes that would be available in a kmalloc() 918 * allocation of @size bytes. For example, a 126 byte request would be 919 * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly 920 * for the general-purpose kmalloc()-based allocations, and is not for the 921 * pre-sized kmem_cache_alloc()-based allocations.) 922 * 923 * Use this to kmalloc() the full bucket size ahead of time instead of using 924 * ksize() to query the size after an allocation. 925 */ 926 size_t kmalloc_size_roundup(size_t size); 927 928 void __init kmem_cache_init_late(void); 929 930 #endif /* _LINUX_SLAB_H */ 931