1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Written by Mark Hemment, 1996 (markhe@nextd.demon.co.uk).
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 *
246 * Any uninitialized fields of the structure are interpreted as unused. The
247 * exception is @freeptr_offset where %0 is a valid value, so
248 * @use_freeptr_offset must be also set to %true in order to interpret the field
249 * as used. For @useroffset %0 is also valid, but only with non-%0
250 * @usersize.
251 *
252 * When %NULL args is passed to kmem_cache_create(), it is equivalent to all
253 * fields unused.
254 */
255 struct kmem_cache_args {
256 /**
257 * @align: The required alignment for the objects.
258 *
259 * %0 means no specific alignment is requested.
260 */
261 unsigned int align;
262 /**
263 * @useroffset: Usercopy region offset.
264 *
265 * %0 is a valid offset, when @usersize is non-%0
266 */
267 unsigned int useroffset;
268 /**
269 * @usersize: Usercopy region size.
270 *
271 * %0 means no usercopy region is specified.
272 */
273 unsigned int usersize;
274 /**
275 * @freeptr_offset: Custom offset for the free pointer
276 * in &SLAB_TYPESAFE_BY_RCU caches
277 *
278 * By default &SLAB_TYPESAFE_BY_RCU caches place the free pointer
279 * outside of the object. This might cause the object to grow in size.
280 * Cache creators that have a reason to avoid this can specify a custom
281 * free pointer offset in their struct where the free pointer will be
282 * placed.
283 *
284 * Note that placing the free pointer inside the object requires the
285 * caller to ensure that no fields are invalidated that are required to
286 * guard against object recycling (See &SLAB_TYPESAFE_BY_RCU for
287 * details).
288 *
289 * Using %0 as a value for @freeptr_offset is valid. If @freeptr_offset
290 * is specified, %use_freeptr_offset must be set %true.
291 *
292 * Note that @ctor currently isn't supported with custom free pointers
293 * as a @ctor requires an external free pointer.
294 */
295 unsigned int freeptr_offset;
296 /**
297 * @use_freeptr_offset: Whether a @freeptr_offset is used.
298 */
299 bool use_freeptr_offset;
300 /**
301 * @ctor: A constructor for the objects.
302 *
303 * The constructor is invoked for each object in a newly allocated slab
304 * page. It is the cache user's responsibility to free object in the
305 * same state as after calling the constructor, or deal appropriately
306 * with any differences between a freshly constructed and a reallocated
307 * object.
308 *
309 * %NULL means no constructor.
310 */
311 void (*ctor)(void *);
312 };
313
314 struct kmem_cache *__kmem_cache_create_args(const char *name,
315 unsigned int object_size,
316 struct kmem_cache_args *args,
317 slab_flags_t flags);
318 static inline struct kmem_cache *
__kmem_cache_create(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))319 __kmem_cache_create(const char *name, unsigned int size, unsigned int align,
320 slab_flags_t flags, void (*ctor)(void *))
321 {
322 struct kmem_cache_args kmem_args = {
323 .align = align,
324 .ctor = ctor,
325 };
326
327 return __kmem_cache_create_args(name, size, &kmem_args, flags);
328 }
329
330 /**
331 * kmem_cache_create_usercopy - Create a kmem cache with a region suitable
332 * for copying to userspace.
333 * @name: A string which is used in /proc/slabinfo to identify this cache.
334 * @size: The size of objects to be created in this cache.
335 * @align: The required alignment for the objects.
336 * @flags: SLAB flags
337 * @useroffset: Usercopy region offset
338 * @usersize: Usercopy region size
339 * @ctor: A constructor for the objects, or %NULL.
340 *
341 * This is a legacy wrapper, new code should use either KMEM_CACHE_USERCOPY()
342 * if whitelisting a single field is sufficient, or kmem_cache_create() with
343 * the necessary parameters passed via the args parameter (see
344 * &struct kmem_cache_args)
345 *
346 * Return: a pointer to the cache on success, NULL on failure.
347 */
348 static inline struct kmem_cache *
kmem_cache_create_usercopy(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,unsigned int useroffset,unsigned int usersize,void (* ctor)(void *))349 kmem_cache_create_usercopy(const char *name, unsigned int size,
350 unsigned int align, slab_flags_t flags,
351 unsigned int useroffset, unsigned int usersize,
352 void (*ctor)(void *))
353 {
354 struct kmem_cache_args kmem_args = {
355 .align = align,
356 .ctor = ctor,
357 .useroffset = useroffset,
358 .usersize = usersize,
359 };
360
361 return __kmem_cache_create_args(name, size, &kmem_args, flags);
362 }
363
364 /* If NULL is passed for @args, use this variant with default arguments. */
365 static inline struct kmem_cache *
__kmem_cache_default_args(const char * name,unsigned int size,struct kmem_cache_args * args,slab_flags_t flags)366 __kmem_cache_default_args(const char *name, unsigned int size,
367 struct kmem_cache_args *args,
368 slab_flags_t flags)
369 {
370 struct kmem_cache_args kmem_default_args = {};
371
372 /* Make sure we don't get passed garbage. */
373 if (WARN_ON_ONCE(args))
374 return ERR_PTR(-EINVAL);
375
376 return __kmem_cache_create_args(name, size, &kmem_default_args, flags);
377 }
378
379 /**
380 * kmem_cache_create - Create a kmem cache.
381 * @__name: A string which is used in /proc/slabinfo to identify this cache.
382 * @__object_size: The size of objects to be created in this cache.
383 * @__args: Optional arguments, see &struct kmem_cache_args. Passing %NULL
384 * means defaults will be used for all the arguments.
385 *
386 * This is currently implemented as a macro using ``_Generic()`` to call
387 * either the new variant of the function, or a legacy one.
388 *
389 * The new variant has 4 parameters:
390 * ``kmem_cache_create(name, object_size, args, flags)``
391 *
392 * See __kmem_cache_create_args() which implements this.
393 *
394 * The legacy variant has 5 parameters:
395 * ``kmem_cache_create(name, object_size, align, flags, ctor)``
396 *
397 * The align and ctor parameters map to the respective fields of
398 * &struct kmem_cache_args
399 *
400 * Context: Cannot be called within a interrupt, but can be interrupted.
401 *
402 * Return: a pointer to the cache on success, NULL on failure.
403 */
404 #define kmem_cache_create(__name, __object_size, __args, ...) \
405 _Generic((__args), \
406 struct kmem_cache_args *: __kmem_cache_create_args, \
407 void *: __kmem_cache_default_args, \
408 default: __kmem_cache_create)(__name, __object_size, __args, __VA_ARGS__)
409
410 void kmem_cache_destroy(struct kmem_cache *s);
411 int kmem_cache_shrink(struct kmem_cache *s);
412
413 /*
414 * Please use this macro to create slab caches. Simply specify the
415 * name of the structure and maybe some flags that are listed above.
416 *
417 * The alignment of the struct determines object alignment. If you
418 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
419 * then the objects will be properly aligned in SMP configurations.
420 */
421 #define KMEM_CACHE(__struct, __flags) \
422 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \
423 &(struct kmem_cache_args) { \
424 .align = __alignof__(struct __struct), \
425 }, (__flags))
426
427 /*
428 * To whitelist a single field for copying to/from usercopy, use this
429 * macro instead for KMEM_CACHE() above.
430 */
431 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
432 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \
433 &(struct kmem_cache_args) { \
434 .align = __alignof__(struct __struct), \
435 .useroffset = offsetof(struct __struct, __field), \
436 .usersize = sizeof_field(struct __struct, __field), \
437 }, (__flags))
438
439 /*
440 * Common kmalloc functions provided by all allocators
441 */
442 void * __must_check krealloc_noprof(const void *objp, size_t new_size,
443 gfp_t flags) __realloc_size(2);
444 #define krealloc(...) alloc_hooks(krealloc_noprof(__VA_ARGS__))
445
446 void kfree(const void *objp);
447 void kfree_sensitive(const void *objp);
448 size_t __ksize(const void *objp);
449
450 DEFINE_FREE(kfree, void *, if (!IS_ERR_OR_NULL(_T)) kfree(_T))
451
452 /**
453 * ksize - Report actual allocation size of associated object
454 *
455 * @objp: Pointer returned from a prior kmalloc()-family allocation.
456 *
457 * This should not be used for writing beyond the originally requested
458 * allocation size. Either use krealloc() or round up the allocation size
459 * with kmalloc_size_roundup() prior to allocation. If this is used to
460 * access beyond the originally requested allocation size, UBSAN_BOUNDS
461 * and/or FORTIFY_SOURCE may trip, since they only know about the
462 * originally allocated size via the __alloc_size attribute.
463 */
464 size_t ksize(const void *objp);
465
466 #ifdef CONFIG_PRINTK
467 bool kmem_dump_obj(void *object);
468 #else
kmem_dump_obj(void * object)469 static inline bool kmem_dump_obj(void *object) { return false; }
470 #endif
471
472 /*
473 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
474 * alignment larger than the alignment of a 64-bit integer.
475 * Setting ARCH_DMA_MINALIGN in arch headers allows that.
476 */
477 #ifdef ARCH_HAS_DMA_MINALIGN
478 #if ARCH_DMA_MINALIGN > 8 && !defined(ARCH_KMALLOC_MINALIGN)
479 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
480 #endif
481 #endif
482
483 #ifndef ARCH_KMALLOC_MINALIGN
484 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
485 #elif ARCH_KMALLOC_MINALIGN > 8
486 #define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN
487 #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
488 #endif
489
490 /*
491 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
492 * Intended for arches that get misalignment faults even for 64 bit integer
493 * aligned buffers.
494 */
495 #ifndef ARCH_SLAB_MINALIGN
496 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
497 #endif
498
499 /*
500 * Arches can define this function if they want to decide the minimum slab
501 * alignment at runtime. The value returned by the function must be a power
502 * of two and >= ARCH_SLAB_MINALIGN.
503 */
504 #ifndef arch_slab_minalign
arch_slab_minalign(void)505 static inline unsigned int arch_slab_minalign(void)
506 {
507 return ARCH_SLAB_MINALIGN;
508 }
509 #endif
510
511 /*
512 * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN.
513 * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN
514 * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment.
515 */
516 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
517 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
518 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
519
520 /*
521 * Kmalloc array related definitions
522 */
523
524 /*
525 * SLUB directly allocates requests fitting in to an order-1 page
526 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
527 */
528 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
529 #define KMALLOC_SHIFT_MAX (MAX_PAGE_ORDER + PAGE_SHIFT)
530 #ifndef KMALLOC_SHIFT_LOW
531 #define KMALLOC_SHIFT_LOW 3
532 #endif
533
534 /* Maximum allocatable size */
535 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
536 /* Maximum size for which we actually use a slab cache */
537 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
538 /* Maximum order allocatable via the slab allocator */
539 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
540
541 /*
542 * Kmalloc subsystem.
543 */
544 #ifndef KMALLOC_MIN_SIZE
545 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
546 #endif
547
548 /*
549 * This restriction comes from byte sized index implementation.
550 * Page size is normally 2^12 bytes and, in this case, if we want to use
551 * byte sized index which can represent 2^8 entries, the size of the object
552 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
553 * If minimum size of kmalloc is less than 16, we use it as minimum object
554 * size and give up to use byte sized index.
555 */
556 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
557 (KMALLOC_MIN_SIZE) : 16)
558
559 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
560 #define RANDOM_KMALLOC_CACHES_NR 15 // # of cache copies
561 #else
562 #define RANDOM_KMALLOC_CACHES_NR 0
563 #endif
564
565 /*
566 * Whenever changing this, take care of that kmalloc_type() and
567 * create_kmalloc_caches() still work as intended.
568 *
569 * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP
570 * is for accounted but unreclaimable and non-dma objects. All the other
571 * kmem caches can have both accounted and unaccounted objects.
572 */
573 enum kmalloc_cache_type {
574 KMALLOC_NORMAL = 0,
575 #ifndef CONFIG_ZONE_DMA
576 KMALLOC_DMA = KMALLOC_NORMAL,
577 #endif
578 #ifndef CONFIG_MEMCG
579 KMALLOC_CGROUP = KMALLOC_NORMAL,
580 #endif
581 KMALLOC_RANDOM_START = KMALLOC_NORMAL,
582 KMALLOC_RANDOM_END = KMALLOC_RANDOM_START + RANDOM_KMALLOC_CACHES_NR,
583 #ifdef CONFIG_SLUB_TINY
584 KMALLOC_RECLAIM = KMALLOC_NORMAL,
585 #else
586 KMALLOC_RECLAIM,
587 #endif
588 #ifdef CONFIG_ZONE_DMA
589 KMALLOC_DMA,
590 #endif
591 #ifdef CONFIG_MEMCG
592 KMALLOC_CGROUP,
593 #endif
594 NR_KMALLOC_TYPES
595 };
596
597 typedef struct kmem_cache * kmem_buckets[KMALLOC_SHIFT_HIGH + 1];
598
599 extern kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES];
600
601 /*
602 * Define gfp bits that should not be set for KMALLOC_NORMAL.
603 */
604 #define KMALLOC_NOT_NORMAL_BITS \
605 (__GFP_RECLAIMABLE | \
606 (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \
607 (IS_ENABLED(CONFIG_MEMCG) ? __GFP_ACCOUNT : 0))
608
609 extern unsigned long random_kmalloc_seed;
610
kmalloc_type(gfp_t flags,unsigned long caller)611 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags, unsigned long caller)
612 {
613 /*
614 * The most common case is KMALLOC_NORMAL, so test for it
615 * with a single branch for all the relevant flags.
616 */
617 if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0))
618 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
619 /* RANDOM_KMALLOC_CACHES_NR (=15) copies + the KMALLOC_NORMAL */
620 return KMALLOC_RANDOM_START + hash_64(caller ^ random_kmalloc_seed,
621 ilog2(RANDOM_KMALLOC_CACHES_NR + 1));
622 #else
623 return KMALLOC_NORMAL;
624 #endif
625
626 /*
627 * At least one of the flags has to be set. Their priorities in
628 * decreasing order are:
629 * 1) __GFP_DMA
630 * 2) __GFP_RECLAIMABLE
631 * 3) __GFP_ACCOUNT
632 */
633 if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA))
634 return KMALLOC_DMA;
635 if (!IS_ENABLED(CONFIG_MEMCG) || (flags & __GFP_RECLAIMABLE))
636 return KMALLOC_RECLAIM;
637 else
638 return KMALLOC_CGROUP;
639 }
640
641 /*
642 * Figure out which kmalloc slab an allocation of a certain size
643 * belongs to.
644 * 0 = zero alloc
645 * 1 = 65 .. 96 bytes
646 * 2 = 129 .. 192 bytes
647 * n = 2^(n-1)+1 .. 2^n
648 *
649 * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized;
650 * typical usage is via kmalloc_index() and therefore evaluated at compile-time.
651 * Callers where !size_is_constant should only be test modules, where runtime
652 * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab().
653 */
__kmalloc_index(size_t size,bool size_is_constant)654 static __always_inline unsigned int __kmalloc_index(size_t size,
655 bool size_is_constant)
656 {
657 if (!size)
658 return 0;
659
660 if (size <= KMALLOC_MIN_SIZE)
661 return KMALLOC_SHIFT_LOW;
662
663 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
664 return 1;
665 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
666 return 2;
667 if (size <= 8) return 3;
668 if (size <= 16) return 4;
669 if (size <= 32) return 5;
670 if (size <= 64) return 6;
671 if (size <= 128) return 7;
672 if (size <= 256) return 8;
673 if (size <= 512) return 9;
674 if (size <= 1024) return 10;
675 if (size <= 2 * 1024) return 11;
676 if (size <= 4 * 1024) return 12;
677 if (size <= 8 * 1024) return 13;
678 if (size <= 16 * 1024) return 14;
679 if (size <= 32 * 1024) return 15;
680 if (size <= 64 * 1024) return 16;
681 if (size <= 128 * 1024) return 17;
682 if (size <= 256 * 1024) return 18;
683 if (size <= 512 * 1024) return 19;
684 if (size <= 1024 * 1024) return 20;
685 if (size <= 2 * 1024 * 1024) return 21;
686
687 if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant)
688 BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()");
689 else
690 BUG();
691
692 /* Will never be reached. Needed because the compiler may complain */
693 return -1;
694 }
695 static_assert(PAGE_SHIFT <= 20);
696 #define kmalloc_index(s) __kmalloc_index(s, true)
697
698 #include <linux/alloc_tag.h>
699
700 /**
701 * kmem_cache_alloc - Allocate an object
702 * @cachep: The cache to allocate from.
703 * @flags: See kmalloc().
704 *
705 * Allocate an object from this cache.
706 * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags.
707 *
708 * Return: pointer to the new object or %NULL in case of error
709 */
710 void *kmem_cache_alloc_noprof(struct kmem_cache *cachep,
711 gfp_t flags) __assume_slab_alignment __malloc;
712 #define kmem_cache_alloc(...) alloc_hooks(kmem_cache_alloc_noprof(__VA_ARGS__))
713
714 void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru,
715 gfp_t gfpflags) __assume_slab_alignment __malloc;
716 #define kmem_cache_alloc_lru(...) alloc_hooks(kmem_cache_alloc_lru_noprof(__VA_ARGS__))
717
718 /**
719 * kmem_cache_charge - memcg charge an already allocated slab memory
720 * @objp: address of the slab object to memcg charge
721 * @gfpflags: describe the allocation context
722 *
723 * kmem_cache_charge allows charging a slab object to the current memcg,
724 * primarily in cases where charging at allocation time might not be possible
725 * because the target memcg is not known (i.e. softirq context)
726 *
727 * The objp should be pointer returned by the slab allocator functions like
728 * kmalloc (with __GFP_ACCOUNT in flags) or kmem_cache_alloc. The memcg charge
729 * behavior can be controlled through gfpflags parameter, which affects how the
730 * necessary internal metadata can be allocated. Including __GFP_NOFAIL denotes
731 * that overcharging is requested instead of failure, but is not applied for the
732 * internal metadata allocation.
733 *
734 * There are several cases where it will return true even if the charging was
735 * not done:
736 * More specifically:
737 *
738 * 1. For !CONFIG_MEMCG or cgroup_disable=memory systems.
739 * 2. Already charged slab objects.
740 * 3. For slab objects from KMALLOC_NORMAL caches - allocated by kmalloc()
741 * without __GFP_ACCOUNT
742 * 4. Allocating internal metadata has failed
743 *
744 * Return: true if charge was successful otherwise false.
745 */
746 bool kmem_cache_charge(void *objp, gfp_t gfpflags);
747 void kmem_cache_free(struct kmem_cache *s, void *objp);
748
749 kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags,
750 unsigned int useroffset, unsigned int usersize,
751 void (*ctor)(void *));
752
753 /*
754 * Bulk allocation and freeing operations. These are accelerated in an
755 * allocator specific way to avoid taking locks repeatedly or building
756 * metadata structures unnecessarily.
757 *
758 * Note that interrupts must be enabled when calling these functions.
759 */
760 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p);
761
762 int kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags, size_t size, void **p);
763 #define kmem_cache_alloc_bulk(...) alloc_hooks(kmem_cache_alloc_bulk_noprof(__VA_ARGS__))
764
kfree_bulk(size_t size,void ** p)765 static __always_inline void kfree_bulk(size_t size, void **p)
766 {
767 kmem_cache_free_bulk(NULL, size, p);
768 }
769
770 void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t flags,
771 int node) __assume_slab_alignment __malloc;
772 #define kmem_cache_alloc_node(...) alloc_hooks(kmem_cache_alloc_node_noprof(__VA_ARGS__))
773
774 /*
775 * These macros allow declaring a kmem_buckets * parameter alongside size, which
776 * can be compiled out with CONFIG_SLAB_BUCKETS=n so that a large number of call
777 * sites don't have to pass NULL.
778 */
779 #ifdef CONFIG_SLAB_BUCKETS
780 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size), kmem_buckets *(_b)
781 #define PASS_BUCKET_PARAMS(_size, _b) (_size), (_b)
782 #define PASS_BUCKET_PARAM(_b) (_b)
783 #else
784 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size)
785 #define PASS_BUCKET_PARAMS(_size, _b) (_size)
786 #define PASS_BUCKET_PARAM(_b) NULL
787 #endif
788
789 /*
790 * The following functions are not to be used directly and are intended only
791 * for internal use from kmalloc() and kmalloc_node()
792 * with the exception of kunit tests
793 */
794
795 void *__kmalloc_noprof(size_t size, gfp_t flags)
796 __assume_kmalloc_alignment __alloc_size(1);
797
798 void *__kmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node)
799 __assume_kmalloc_alignment __alloc_size(1);
800
801 void *__kmalloc_cache_noprof(struct kmem_cache *s, gfp_t flags, size_t size)
802 __assume_kmalloc_alignment __alloc_size(3);
803
804 void *__kmalloc_cache_node_noprof(struct kmem_cache *s, gfp_t gfpflags,
805 int node, size_t size)
806 __assume_kmalloc_alignment __alloc_size(4);
807
808 void *__kmalloc_large_noprof(size_t size, gfp_t flags)
809 __assume_page_alignment __alloc_size(1);
810
811 void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
812 __assume_page_alignment __alloc_size(1);
813
814 /**
815 * kmalloc - allocate kernel memory
816 * @size: how many bytes of memory are required.
817 * @flags: describe the allocation context
818 *
819 * kmalloc is the normal method of allocating memory
820 * for objects smaller than page size in the kernel.
821 *
822 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
823 * bytes. For @size of power of two bytes, the alignment is also guaranteed
824 * to be at least to the size. For other sizes, the alignment is guaranteed to
825 * be at least the largest power-of-two divisor of @size.
826 *
827 * The @flags argument may be one of the GFP flags defined at
828 * include/linux/gfp_types.h and described at
829 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
830 *
831 * The recommended usage of the @flags is described at
832 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
833 *
834 * Below is a brief outline of the most useful GFP flags
835 *
836 * %GFP_KERNEL
837 * Allocate normal kernel ram. May sleep.
838 *
839 * %GFP_NOWAIT
840 * Allocation will not sleep.
841 *
842 * %GFP_ATOMIC
843 * Allocation will not sleep. May use emergency pools.
844 *
845 * Also it is possible to set different flags by OR'ing
846 * in one or more of the following additional @flags:
847 *
848 * %__GFP_ZERO
849 * Zero the allocated memory before returning. Also see kzalloc().
850 *
851 * %__GFP_HIGH
852 * This allocation has high priority and may use emergency pools.
853 *
854 * %__GFP_NOFAIL
855 * Indicate that this allocation is in no way allowed to fail
856 * (think twice before using).
857 *
858 * %__GFP_NORETRY
859 * If memory is not immediately available,
860 * then give up at once.
861 *
862 * %__GFP_NOWARN
863 * If allocation fails, don't issue any warnings.
864 *
865 * %__GFP_RETRY_MAYFAIL
866 * Try really hard to succeed the allocation but fail
867 * eventually.
868 */
kmalloc_noprof(size_t size,gfp_t flags)869 static __always_inline __alloc_size(1) void *kmalloc_noprof(size_t size, gfp_t flags)
870 {
871 if (__builtin_constant_p(size) && size) {
872 unsigned int index;
873
874 if (size > KMALLOC_MAX_CACHE_SIZE)
875 return __kmalloc_large_noprof(size, flags);
876
877 index = kmalloc_index(size);
878 return __kmalloc_cache_noprof(
879 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index],
880 flags, size);
881 }
882 return __kmalloc_noprof(size, flags);
883 }
884 #define kmalloc(...) alloc_hooks(kmalloc_noprof(__VA_ARGS__))
885
886 #define kmem_buckets_alloc(_b, _size, _flags) \
887 alloc_hooks(__kmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE))
888
889 #define kmem_buckets_alloc_track_caller(_b, _size, _flags) \
890 alloc_hooks(__kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE, _RET_IP_))
891
kmalloc_node_noprof(size_t size,gfp_t flags,int node)892 static __always_inline __alloc_size(1) void *kmalloc_node_noprof(size_t size, gfp_t flags, int node)
893 {
894 if (__builtin_constant_p(size) && size) {
895 unsigned int index;
896
897 if (size > KMALLOC_MAX_CACHE_SIZE)
898 return __kmalloc_large_node_noprof(size, flags, node);
899
900 index = kmalloc_index(size);
901 return __kmalloc_cache_node_noprof(
902 kmalloc_caches[kmalloc_type(flags, _RET_IP_)][index],
903 flags, node, size);
904 }
905 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node);
906 }
907 #define kmalloc_node(...) alloc_hooks(kmalloc_node_noprof(__VA_ARGS__))
908
909 /**
910 * kmalloc_array - allocate memory for an array.
911 * @n: number of elements.
912 * @size: element size.
913 * @flags: the type of memory to allocate (see kmalloc).
914 */
kmalloc_array_noprof(size_t n,size_t size,gfp_t flags)915 static inline __alloc_size(1, 2) void *kmalloc_array_noprof(size_t n, size_t size, gfp_t flags)
916 {
917 size_t bytes;
918
919 if (unlikely(check_mul_overflow(n, size, &bytes)))
920 return NULL;
921 if (__builtin_constant_p(n) && __builtin_constant_p(size))
922 return kmalloc_noprof(bytes, flags);
923 return kmalloc_noprof(bytes, flags);
924 }
925 #define kmalloc_array(...) alloc_hooks(kmalloc_array_noprof(__VA_ARGS__))
926
927 /**
928 * krealloc_array - reallocate memory for an array.
929 * @p: pointer to the memory chunk to reallocate
930 * @new_n: new number of elements to alloc
931 * @new_size: new size of a single member of the array
932 * @flags: the type of memory to allocate (see kmalloc)
933 *
934 * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
935 * initial memory allocation, every subsequent call to this API for the same
936 * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
937 * __GFP_ZERO is not fully honored by this API.
938 *
939 * See krealloc_noprof() for further details.
940 *
941 * In any case, the contents of the object pointed to are preserved up to the
942 * lesser of the new and old sizes.
943 */
krealloc_array_noprof(void * p,size_t new_n,size_t new_size,gfp_t flags)944 static inline __realloc_size(2, 3) void * __must_check krealloc_array_noprof(void *p,
945 size_t new_n,
946 size_t new_size,
947 gfp_t flags)
948 {
949 size_t bytes;
950
951 if (unlikely(check_mul_overflow(new_n, new_size, &bytes)))
952 return NULL;
953
954 return krealloc_noprof(p, bytes, flags);
955 }
956 #define krealloc_array(...) alloc_hooks(krealloc_array_noprof(__VA_ARGS__))
957
958 /**
959 * kcalloc - allocate memory for an array. The memory is set to zero.
960 * @n: number of elements.
961 * @size: element size.
962 * @flags: the type of memory to allocate (see kmalloc).
963 */
964 #define kcalloc(n, size, flags) kmalloc_array(n, size, (flags) | __GFP_ZERO)
965
966 void *__kmalloc_node_track_caller_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node,
967 unsigned long caller) __alloc_size(1);
968 #define kmalloc_node_track_caller_noprof(size, flags, node, caller) \
969 __kmalloc_node_track_caller_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node, caller)
970 #define kmalloc_node_track_caller(...) \
971 alloc_hooks(kmalloc_node_track_caller_noprof(__VA_ARGS__, _RET_IP_))
972
973 /*
974 * kmalloc_track_caller is a special version of kmalloc that records the
975 * calling function of the routine calling it for slab leak tracking instead
976 * of just the calling function (confusing, eh?).
977 * It's useful when the call to kmalloc comes from a widely-used standard
978 * allocator where we care about the real place the memory allocation
979 * request comes from.
980 */
981 #define kmalloc_track_caller(...) kmalloc_node_track_caller(__VA_ARGS__, NUMA_NO_NODE)
982
983 #define kmalloc_track_caller_noprof(...) \
984 kmalloc_node_track_caller_noprof(__VA_ARGS__, NUMA_NO_NODE, _RET_IP_)
985
kmalloc_array_node_noprof(size_t n,size_t size,gfp_t flags,int node)986 static inline __alloc_size(1, 2) void *kmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags,
987 int node)
988 {
989 size_t bytes;
990
991 if (unlikely(check_mul_overflow(n, size, &bytes)))
992 return NULL;
993 if (__builtin_constant_p(n) && __builtin_constant_p(size))
994 return kmalloc_node_noprof(bytes, flags, node);
995 return __kmalloc_node_noprof(PASS_BUCKET_PARAMS(bytes, NULL), flags, node);
996 }
997 #define kmalloc_array_node(...) alloc_hooks(kmalloc_array_node_noprof(__VA_ARGS__))
998
999 #define kcalloc_node(_n, _size, _flags, _node) \
1000 kmalloc_array_node(_n, _size, (_flags) | __GFP_ZERO, _node)
1001
1002 /*
1003 * Shortcuts
1004 */
1005 #define kmem_cache_zalloc(_k, _flags) kmem_cache_alloc(_k, (_flags)|__GFP_ZERO)
1006
1007 /**
1008 * kzalloc - allocate memory. The memory is set to zero.
1009 * @size: how many bytes of memory are required.
1010 * @flags: the type of memory to allocate (see kmalloc).
1011 */
kzalloc_noprof(size_t size,gfp_t flags)1012 static inline __alloc_size(1) void *kzalloc_noprof(size_t size, gfp_t flags)
1013 {
1014 return kmalloc_noprof(size, flags | __GFP_ZERO);
1015 }
1016 #define kzalloc(...) alloc_hooks(kzalloc_noprof(__VA_ARGS__))
1017 #define kzalloc_node(_size, _flags, _node) kmalloc_node(_size, (_flags)|__GFP_ZERO, _node)
1018
1019 void *__kvmalloc_node_noprof(DECL_BUCKET_PARAMS(size, b), gfp_t flags, int node) __alloc_size(1);
1020 #define kvmalloc_node_noprof(size, flags, node) \
1021 __kvmalloc_node_noprof(PASS_BUCKET_PARAMS(size, NULL), flags, node)
1022 #define kvmalloc_node(...) alloc_hooks(kvmalloc_node_noprof(__VA_ARGS__))
1023
1024 #define kvmalloc(_size, _flags) kvmalloc_node(_size, _flags, NUMA_NO_NODE)
1025 #define kvmalloc_noprof(_size, _flags) kvmalloc_node_noprof(_size, _flags, NUMA_NO_NODE)
1026 #define kvzalloc(_size, _flags) kvmalloc(_size, (_flags)|__GFP_ZERO)
1027
1028 #define kvzalloc_node(_size, _flags, _node) kvmalloc_node(_size, (_flags)|__GFP_ZERO, _node)
1029 #define kmem_buckets_valloc(_b, _size, _flags) \
1030 alloc_hooks(__kvmalloc_node_noprof(PASS_BUCKET_PARAMS(_size, _b), _flags, NUMA_NO_NODE))
1031
1032 static inline __alloc_size(1, 2) void *
kvmalloc_array_node_noprof(size_t n,size_t size,gfp_t flags,int node)1033 kvmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, int node)
1034 {
1035 size_t bytes;
1036
1037 if (unlikely(check_mul_overflow(n, size, &bytes)))
1038 return NULL;
1039
1040 return kvmalloc_node_noprof(bytes, flags, node);
1041 }
1042
1043 #define kvmalloc_array_noprof(...) kvmalloc_array_node_noprof(__VA_ARGS__, NUMA_NO_NODE)
1044 #define kvcalloc_node_noprof(_n,_s,_f,_node) kvmalloc_array_node_noprof(_n,_s,(_f)|__GFP_ZERO,_node)
1045 #define kvcalloc_noprof(...) kvcalloc_node_noprof(__VA_ARGS__, NUMA_NO_NODE)
1046
1047 #define kvmalloc_array(...) alloc_hooks(kvmalloc_array_noprof(__VA_ARGS__))
1048 #define kvcalloc_node(...) alloc_hooks(kvcalloc_node_noprof(__VA_ARGS__))
1049 #define kvcalloc(...) alloc_hooks(kvcalloc_noprof(__VA_ARGS__))
1050
1051 void *kvrealloc_noprof(const void *p, size_t size, gfp_t flags)
1052 __realloc_size(2);
1053 #define kvrealloc(...) alloc_hooks(kvrealloc_noprof(__VA_ARGS__))
1054
1055 extern void kvfree(const void *addr);
1056 DEFINE_FREE(kvfree, void *, if (!IS_ERR_OR_NULL(_T)) kvfree(_T))
1057
1058 extern void kvfree_sensitive(const void *addr, size_t len);
1059
1060 unsigned int kmem_cache_size(struct kmem_cache *s);
1061
1062 /**
1063 * kmalloc_size_roundup - Report allocation bucket size for the given size
1064 *
1065 * @size: Number of bytes to round up from.
1066 *
1067 * This returns the number of bytes that would be available in a kmalloc()
1068 * allocation of @size bytes. For example, a 126 byte request would be
1069 * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly
1070 * for the general-purpose kmalloc()-based allocations, and is not for the
1071 * pre-sized kmem_cache_alloc()-based allocations.)
1072 *
1073 * Use this to kmalloc() the full bucket size ahead of time instead of using
1074 * ksize() to query the size after an allocation.
1075 */
1076 size_t kmalloc_size_roundup(size_t size);
1077
1078 void __init kmem_cache_init_late(void);
1079
1080 #endif /* _LINUX_SLAB_H */
1081