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/bug.h>
16 #include <linux/cache.h>
17 #include <linux/gfp.h>
18 #include <linux/overflow.h>
19 #include <linux/types.h>
20 #include <linux/rcupdate.h>
21 #include <linux/workqueue.h>
22 #include <linux/percpu-refcount.h>
23 #include <linux/cleanup.h>
24 #include <linux/hash.h>
25
26 enum _slab_flag_bits {
27 _SLAB_CONSISTENCY_CHECKS,
28 _SLAB_RED_ZONE,
29 _SLAB_POISON,
30 _SLAB_KMALLOC,
31 _SLAB_HWCACHE_ALIGN,
32 _SLAB_CACHE_DMA,
33 _SLAB_CACHE_DMA32,
34 _SLAB_STORE_USER,
35 _SLAB_PANIC,
36 _SLAB_TYPESAFE_BY_RCU,
37 _SLAB_TRACE,
38 #ifdef CONFIG_DEBUG_OBJECTS
39 _SLAB_DEBUG_OBJECTS,
40 #endif
41 _SLAB_NOLEAKTRACE,
42 _SLAB_NO_MERGE,
43 #ifdef CONFIG_FAILSLAB
44 _SLAB_FAILSLAB,
45 #endif
46 #ifdef CONFIG_MEMCG
47 _SLAB_ACCOUNT,
48 #endif
49 #ifdef CONFIG_KASAN_GENERIC
50 _SLAB_KASAN,
51 #endif
52 _SLAB_NO_USER_FLAGS,
53 #ifdef CONFIG_KFENCE
54 _SLAB_SKIP_KFENCE,
55 #endif
56 #ifndef CONFIG_SLUB_TINY
57 _SLAB_RECLAIM_ACCOUNT,
58 #endif
59 _SLAB_OBJECT_POISON,
60 _SLAB_CMPXCHG_DOUBLE,
61 #ifdef CONFIG_SLAB_OBJ_EXT
62 _SLAB_NO_OBJ_EXT,
63 #ifdef CONFIG_64BIT
64 _SLAB_OBJ_EXT_IN_OBJ,
65 #endif
66 #endif
67 _SLAB_NO_SHEAVES,
68 _SLAB_FLAGS_LAST_BIT
69 };
70
71 #define __SLAB_FLAG_BIT(nr) ((slab_flags_t __force)(1U << (nr)))
72 #define __SLAB_FLAG_UNUSED ((slab_flags_t __force)(0U))
73
74 /*
75 * Flags to pass to kmem_cache_create().
76 * The ones marked DEBUG need CONFIG_SLUB_DEBUG enabled, otherwise are no-op
77 */
78 /* DEBUG: Perform (expensive) checks on alloc/free */
79 #define SLAB_CONSISTENCY_CHECKS __SLAB_FLAG_BIT(_SLAB_CONSISTENCY_CHECKS)
80 /* DEBUG: Red zone objs in a cache */
81 #define SLAB_RED_ZONE __SLAB_FLAG_BIT(_SLAB_RED_ZONE)
82 /* DEBUG: Poison objects */
83 #define SLAB_POISON __SLAB_FLAG_BIT(_SLAB_POISON)
84 /* Indicate a kmalloc slab */
85 #define SLAB_KMALLOC __SLAB_FLAG_BIT(_SLAB_KMALLOC)
86 /**
87 * define SLAB_HWCACHE_ALIGN - Align objects on cache line boundaries.
88 *
89 * Sufficiently large objects are aligned on cache line boundary. For object
90 * size smaller than a half of cache line size, the alignment is on the half of
91 * cache line size. In general, if object size is smaller than 1/2^n of cache
92 * line size, the alignment is adjusted to 1/2^n.
93 *
94 * If explicit alignment is also requested by the respective
95 * &struct kmem_cache_args field, the greater of both is alignments is applied.
96 */
97 #define SLAB_HWCACHE_ALIGN __SLAB_FLAG_BIT(_SLAB_HWCACHE_ALIGN)
98 /* Use GFP_DMA memory */
99 #define SLAB_CACHE_DMA __SLAB_FLAG_BIT(_SLAB_CACHE_DMA)
100 /* Use GFP_DMA32 memory */
101 #define SLAB_CACHE_DMA32 __SLAB_FLAG_BIT(_SLAB_CACHE_DMA32)
102 /* DEBUG: Store the last owner for bug hunting */
103 #define SLAB_STORE_USER __SLAB_FLAG_BIT(_SLAB_STORE_USER)
104 /* Panic if kmem_cache_create() fails */
105 #define SLAB_PANIC __SLAB_FLAG_BIT(_SLAB_PANIC)
106 /**
107 * define SLAB_TYPESAFE_BY_RCU - **WARNING** READ THIS!
108 *
109 * This delays freeing the SLAB page by a grace period, it does _NOT_
110 * delay object freeing. This means that if you do kmem_cache_free()
111 * that memory location is free to be reused at any time. Thus it may
112 * be possible to see another object there in the same RCU grace period.
113 *
114 * This feature only ensures the memory location backing the object
115 * stays valid, the trick to using this is relying on an independent
116 * object validation pass. Something like:
117 *
118 * ::
119 *
120 * begin:
121 * rcu_read_lock();
122 * obj = lockless_lookup(key);
123 * if (obj) {
124 * if (!try_get_ref(obj)) // might fail for free objects
125 * rcu_read_unlock();
126 * goto begin;
127 *
128 * if (obj->key != key) { // not the object we expected
129 * put_ref(obj);
130 * rcu_read_unlock();
131 * goto begin;
132 * }
133 * }
134 * rcu_read_unlock();
135 *
136 * This is useful if we need to approach a kernel structure obliquely,
137 * from its address obtained without the usual locking. We can lock
138 * the structure to stabilize it and check it's still at the given address,
139 * only if we can be sure that the memory has not been meanwhile reused
140 * for some other kind of object (which our subsystem's lock might corrupt).
141 *
142 * rcu_read_lock before reading the address, then rcu_read_unlock after
143 * taking the spinlock within the structure expected at that address.
144 *
145 * Note that object identity check has to be done *after* acquiring a
146 * reference, therefore user has to ensure proper ordering for loads.
147 * Similarly, when initializing objects allocated with SLAB_TYPESAFE_BY_RCU,
148 * the newly allocated object has to be fully initialized *before* its
149 * refcount gets initialized and proper ordering for stores is required.
150 * refcount_{add|inc}_not_zero_acquire() and refcount_set_release() are
151 * designed with the proper fences required for reference counting objects
152 * allocated with SLAB_TYPESAFE_BY_RCU.
153 *
154 * Note that it is not possible to acquire a lock within a structure
155 * allocated with SLAB_TYPESAFE_BY_RCU without first acquiring a reference
156 * as described above. The reason is that SLAB_TYPESAFE_BY_RCU pages
157 * are not zeroed before being given to the slab, which means that any
158 * locks must be initialized after each and every kmem_struct_alloc().
159 * Alternatively, make the ctor passed to kmem_cache_create() initialize
160 * the locks at page-allocation time, as is done in __i915_request_ctor(),
161 * sighand_ctor(), and anon_vma_ctor(). Such a ctor permits readers
162 * to safely acquire those ctor-initialized locks under rcu_read_lock()
163 * protection.
164 *
165 * Note that SLAB_TYPESAFE_BY_RCU was originally named SLAB_DESTROY_BY_RCU.
166 */
167 #define SLAB_TYPESAFE_BY_RCU __SLAB_FLAG_BIT(_SLAB_TYPESAFE_BY_RCU)
168 /* Trace allocations and frees */
169 #define SLAB_TRACE __SLAB_FLAG_BIT(_SLAB_TRACE)
170
171 /* Flag to prevent checks on free */
172 #ifdef CONFIG_DEBUG_OBJECTS
173 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_BIT(_SLAB_DEBUG_OBJECTS)
174 #else
175 # define SLAB_DEBUG_OBJECTS __SLAB_FLAG_UNUSED
176 #endif
177
178 /* Avoid kmemleak tracing */
179 #define SLAB_NOLEAKTRACE __SLAB_FLAG_BIT(_SLAB_NOLEAKTRACE)
180
181 /*
182 * Prevent merging with compatible kmem caches. This flag should be used
183 * cautiously. Valid use cases:
184 *
185 * - caches created for self-tests (e.g. kunit)
186 * - general caches created and used by a subsystem, only when a
187 * (subsystem-specific) debug option is enabled
188 * - performance critical caches, should be very rare and consulted with slab
189 * maintainers, and not used together with CONFIG_SLUB_TINY
190 */
191 #define SLAB_NO_MERGE __SLAB_FLAG_BIT(_SLAB_NO_MERGE)
192
193 /* Fault injection mark */
194 #ifdef CONFIG_FAILSLAB
195 # define SLAB_FAILSLAB __SLAB_FLAG_BIT(_SLAB_FAILSLAB)
196 #else
197 # define SLAB_FAILSLAB __SLAB_FLAG_UNUSED
198 #endif
199 /**
200 * define SLAB_ACCOUNT - Account allocations to memcg.
201 *
202 * All object allocations from this cache will be memcg accounted, regardless of
203 * __GFP_ACCOUNT being or not being passed to individual allocations.
204 */
205 #ifdef CONFIG_MEMCG
206 # define SLAB_ACCOUNT __SLAB_FLAG_BIT(_SLAB_ACCOUNT)
207 #else
208 # define SLAB_ACCOUNT __SLAB_FLAG_UNUSED
209 #endif
210
211 #ifdef CONFIG_KASAN_GENERIC
212 #define SLAB_KASAN __SLAB_FLAG_BIT(_SLAB_KASAN)
213 #else
214 #define SLAB_KASAN __SLAB_FLAG_UNUSED
215 #endif
216
217 /*
218 * Ignore user specified debugging flags.
219 * Intended for caches created for self-tests so they have only flags
220 * specified in the code and other flags are ignored.
221 */
222 #define SLAB_NO_USER_FLAGS __SLAB_FLAG_BIT(_SLAB_NO_USER_FLAGS)
223
224 #ifdef CONFIG_KFENCE
225 #define SLAB_SKIP_KFENCE __SLAB_FLAG_BIT(_SLAB_SKIP_KFENCE)
226 #else
227 #define SLAB_SKIP_KFENCE __SLAB_FLAG_UNUSED
228 #endif
229
230 /* The following flags affect the page allocator grouping pages by mobility */
231 /**
232 * define SLAB_RECLAIM_ACCOUNT - Objects are reclaimable.
233 *
234 * Use this flag for caches that have an associated shrinker. As a result, slab
235 * pages are allocated with __GFP_RECLAIMABLE, which affects grouping pages by
236 * mobility, and are accounted in SReclaimable counter in /proc/meminfo
237 */
238 #ifndef CONFIG_SLUB_TINY
239 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_BIT(_SLAB_RECLAIM_ACCOUNT)
240 #else
241 #define SLAB_RECLAIM_ACCOUNT __SLAB_FLAG_UNUSED
242 #endif
243 #define SLAB_TEMPORARY SLAB_RECLAIM_ACCOUNT /* Objects are short-lived */
244
245 /* Slab caches without obj_exts array */
246 #ifdef CONFIG_SLAB_OBJ_EXT
247 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_BIT(_SLAB_NO_OBJ_EXT)
248 #else
249 #define SLAB_NO_OBJ_EXT __SLAB_FLAG_UNUSED
250 #endif
251
252 #define SLAB_NO_SHEAVES __SLAB_FLAG_BIT(_SLAB_NO_SHEAVES)
253
254 #if defined(CONFIG_SLAB_OBJ_EXT) && defined(CONFIG_64BIT)
255 #define SLAB_OBJ_EXT_IN_OBJ __SLAB_FLAG_BIT(_SLAB_OBJ_EXT_IN_OBJ)
256 #else
257 #define SLAB_OBJ_EXT_IN_OBJ __SLAB_FLAG_UNUSED
258 #endif
259
260 /*
261 * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
262 *
263 * Dereferencing ZERO_SIZE_PTR will lead to a distinct access fault.
264 *
265 * ZERO_SIZE_PTR can be passed to kfree though in the same way that NULL can.
266 * Both make kfree a no-op.
267 */
268 #define ZERO_SIZE_PTR ((void *)16)
269
270 #define ZERO_OR_NULL_PTR(x) ((unsigned long)(x) <= \
271 (unsigned long)ZERO_SIZE_PTR)
272
273 #include <linux/kasan.h>
274
275 struct list_lru;
276 struct mem_cgroup;
277 /*
278 * struct kmem_cache related prototypes
279 */
280 bool slab_is_available(void);
281
282 /**
283 * struct kmem_cache_args - Less common arguments for kmem_cache_create()
284 *
285 * Any uninitialized fields of the structure are interpreted as unused. The
286 * exception is @freeptr_offset where %0 is a valid value, so
287 * @use_freeptr_offset must be also set to %true in order to interpret the field
288 * as used. For @useroffset %0 is also valid, but only with non-%0
289 * @usersize.
290 *
291 * When %NULL args is passed to kmem_cache_create(), it is equivalent to all
292 * fields unused.
293 */
294 struct kmem_cache_args {
295 /**
296 * @align: The required alignment for the objects.
297 *
298 * %0 means no specific alignment is requested.
299 */
300 unsigned int align;
301 /**
302 * @useroffset: Usercopy region offset.
303 *
304 * %0 is a valid offset, when @usersize is non-%0
305 */
306 unsigned int useroffset;
307 /**
308 * @usersize: Usercopy region size.
309 *
310 * %0 means no usercopy region is specified.
311 */
312 unsigned int usersize;
313 /**
314 * @freeptr_offset: Custom offset for the free pointer
315 * in caches with &SLAB_TYPESAFE_BY_RCU or @ctor
316 *
317 * By default, &SLAB_TYPESAFE_BY_RCU and @ctor caches place the free
318 * pointer outside of the object. This might cause the object to grow
319 * in size. Cache creators that have a reason to avoid this can specify
320 * a custom free pointer offset in their data structure where the free
321 * pointer will be placed.
322 *
323 * For caches with &SLAB_TYPESAFE_BY_RCU, the caller must ensure that
324 * the free pointer does not overlay fields required to guard against
325 * object recycling (See &SLAB_TYPESAFE_BY_RCU for details).
326 *
327 * For caches with @ctor, the caller must ensure that the free pointer
328 * does not overlay fields initialized by the constructor.
329 *
330 * Currently, only caches with &SLAB_TYPESAFE_BY_RCU or @ctor
331 * may specify @freeptr_offset.
332 *
333 * Using %0 as a value for @freeptr_offset is valid. If @freeptr_offset
334 * is specified, @use_freeptr_offset must be set %true.
335 */
336 unsigned int freeptr_offset;
337 /**
338 * @use_freeptr_offset: Whether a @freeptr_offset is used.
339 */
340 bool use_freeptr_offset;
341 /**
342 * @ctor: A constructor for the objects.
343 *
344 * The constructor is invoked for each object in a newly allocated slab
345 * page. It is the cache user's responsibility to free object in the
346 * same state as after calling the constructor, or deal appropriately
347 * with any differences between a freshly constructed and a reallocated
348 * object.
349 *
350 * %NULL means no constructor.
351 */
352 void (*ctor)(void *);
353 /**
354 * @sheaf_capacity: Enable sheaves of given capacity for the cache.
355 *
356 * With a non-zero value, allocations from the cache go through caching
357 * arrays called sheaves. Each cpu has a main sheaf that's always
358 * present, and a spare sheaf that may be not present. When both become
359 * empty, there's an attempt to replace an empty sheaf with a full sheaf
360 * from the per-node barn.
361 *
362 * When no full sheaf is available, and gfp flags allow blocking, a
363 * sheaf is allocated and filled from slab(s) using bulk allocation.
364 * Otherwise the allocation falls back to the normal operation
365 * allocating a single object from a slab.
366 *
367 * Analogically when freeing and both percpu sheaves are full, the barn
368 * may replace it with an empty sheaf, unless it's over capacity. In
369 * that case a sheaf is bulk freed to slab pages.
370 *
371 * The sheaves do not enforce NUMA placement of objects, so allocations
372 * via kmem_cache_alloc_node() with a node specified other than
373 * NUMA_NO_NODE will bypass them.
374 *
375 * Bulk allocation and free operations also try to use the cpu sheaves
376 * and barn, but fallback to using slab pages directly.
377 *
378 * When slub_debug is enabled for the cache, the sheaf_capacity argument
379 * is ignored.
380 *
381 * %0 means no sheaves will be created.
382 */
383 unsigned int sheaf_capacity;
384 };
385
386 struct kmem_cache *__kmem_cache_create_args(const char *name,
387 unsigned int object_size,
388 struct kmem_cache_args *args,
389 slab_flags_t flags);
390 static inline struct kmem_cache *
__kmem_cache_create(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))391 __kmem_cache_create(const char *name, unsigned int size, unsigned int align,
392 slab_flags_t flags, void (*ctor)(void *))
393 {
394 struct kmem_cache_args kmem_args = {
395 .align = align,
396 .ctor = ctor,
397 };
398
399 return __kmem_cache_create_args(name, size, &kmem_args, flags);
400 }
401
402 /**
403 * kmem_cache_create_usercopy - Create a kmem cache with a region suitable
404 * for copying to userspace.
405 * @name: A string which is used in /proc/slabinfo to identify this cache.
406 * @size: The size of objects to be created in this cache.
407 * @align: The required alignment for the objects.
408 * @flags: SLAB flags
409 * @useroffset: Usercopy region offset
410 * @usersize: Usercopy region size
411 * @ctor: A constructor for the objects, or %NULL.
412 *
413 * This is a legacy wrapper, new code should use either KMEM_CACHE_USERCOPY()
414 * if whitelisting a single field is sufficient, or kmem_cache_create() with
415 * the necessary parameters passed via the args parameter (see
416 * &struct kmem_cache_args)
417 *
418 * Return: a pointer to the cache on success, NULL on failure.
419 */
420 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 *))421 kmem_cache_create_usercopy(const char *name, unsigned int size,
422 unsigned int align, slab_flags_t flags,
423 unsigned int useroffset, unsigned int usersize,
424 void (*ctor)(void *))
425 {
426 struct kmem_cache_args kmem_args = {
427 .align = align,
428 .ctor = ctor,
429 .useroffset = useroffset,
430 .usersize = usersize,
431 };
432
433 return __kmem_cache_create_args(name, size, &kmem_args, flags);
434 }
435
436 /* If NULL is passed for @args, use this variant with default arguments. */
437 static inline struct kmem_cache *
__kmem_cache_default_args(const char * name,unsigned int size,struct kmem_cache_args * args,slab_flags_t flags)438 __kmem_cache_default_args(const char *name, unsigned int size,
439 struct kmem_cache_args *args,
440 slab_flags_t flags)
441 {
442 struct kmem_cache_args kmem_default_args = {};
443
444 /* Make sure we don't get passed garbage. */
445 if (WARN_ON_ONCE(args))
446 return ERR_PTR(-EINVAL);
447
448 return __kmem_cache_create_args(name, size, &kmem_default_args, flags);
449 }
450
451 /**
452 * kmem_cache_create - Create a kmem cache.
453 * @__name: A string which is used in /proc/slabinfo to identify this cache.
454 * @__object_size: The size of objects to be created in this cache.
455 * @__args: Optional arguments, see &struct kmem_cache_args. Passing %NULL
456 * means defaults will be used for all the arguments.
457 *
458 * This is currently implemented as a macro using ``_Generic()`` to call
459 * either the new variant of the function, or a legacy one.
460 *
461 * The new variant has 4 parameters:
462 * ``kmem_cache_create(name, object_size, args, flags)``
463 *
464 * See __kmem_cache_create_args() which implements this.
465 *
466 * The legacy variant has 5 parameters:
467 * ``kmem_cache_create(name, object_size, align, flags, ctor)``
468 *
469 * The align and ctor parameters map to the respective fields of
470 * &struct kmem_cache_args
471 *
472 * Context: Cannot be called within a interrupt, but can be interrupted.
473 *
474 * Return: a pointer to the cache on success, NULL on failure.
475 */
476 #define kmem_cache_create(__name, __object_size, __args, ...) \
477 _Generic((__args), \
478 struct kmem_cache_args *: __kmem_cache_create_args, \
479 void *: __kmem_cache_default_args, \
480 default: __kmem_cache_create)(__name, __object_size, __args, __VA_ARGS__)
481
482 void kmem_cache_destroy(struct kmem_cache *s);
483 int kmem_cache_shrink(struct kmem_cache *s);
484
485 /*
486 * Please use this macro to create slab caches. Simply specify the
487 * name of the structure and maybe some flags that are listed above.
488 *
489 * The alignment of the struct determines object alignment. If you
490 * f.e. add ____cacheline_aligned_in_smp to the struct declaration
491 * then the objects will be properly aligned in SMP configurations.
492 */
493 #define KMEM_CACHE(__struct, __flags) \
494 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \
495 &(struct kmem_cache_args) { \
496 .align = __alignof__(struct __struct), \
497 }, (__flags))
498
499 /*
500 * To whitelist a single field for copying to/from usercopy, use this
501 * macro instead for KMEM_CACHE() above.
502 */
503 #define KMEM_CACHE_USERCOPY(__struct, __flags, __field) \
504 __kmem_cache_create_args(#__struct, sizeof(struct __struct), \
505 &(struct kmem_cache_args) { \
506 .align = __alignof__(struct __struct), \
507 .useroffset = offsetof(struct __struct, __field), \
508 .usersize = sizeof_field(struct __struct, __field), \
509 }, (__flags))
510
511 #ifdef CONFIG_KMALLOC_PARTITION_CACHES
512 typedef struct { unsigned long v; } kmalloc_token_t;
513 #ifdef CONFIG_KMALLOC_PARTITION_RANDOM
514 extern unsigned long random_kmalloc_seed;
515 #define __kmalloc_token(...) ((kmalloc_token_t){ .v = _CODE_LOCATION_ })
516 #elif defined(CONFIG_KMALLOC_PARTITION_TYPED)
517 #ifdef __CHECKER__
518 #define __kmalloc_token(...) ((kmalloc_token_t){ .v = 0 })
519 #else /* !__CHECKER__ */
520 #define __kmalloc_token(...) ((kmalloc_token_t){ .v = __builtin_infer_alloc_token(__VA_ARGS__) })
521 #endif /* __CHECKER__ */
522 #endif /* CONFIG_KMALLOC_PARTITION_TYPED */
523 #define DECL_TOKEN_PARAM(_token) , kmalloc_token_t (_token)
524 #define _PASS_TOKEN_PARAM(_token) , (_token)
525 #define PASS_TOKEN_PARAM(_token) (_token)
526 #define DECL_TOKEN_PARAMS(_size, _token) size_t (_size), kmalloc_token_t (_token)
527 #define PASS_TOKEN_PARAMS(_size, _token) (_size), (_token)
528 #else /* !CONFIG_KMALLOC_PARTITION_CACHES */
529 typedef struct {} kmalloc_token_t;
530 #define __kmalloc_token(...) ((kmalloc_token_t){}) /* no-op */
531 #define DECL_TOKEN_PARAM(_token)
532 #define _PASS_TOKEN_PARAM(_token)
533 #define PASS_TOKEN_PARAM(_token) ((kmalloc_token_t){})
534 #define DECL_TOKEN_PARAMS(_size, _token) size_t (_size)
535 #define PASS_TOKEN_PARAMS(_size, _token) (_size)
536 #endif /* CONFIG_KMALLOC_PARTITION_CACHES */
537
538 /*
539 * Common kmalloc functions provided by all allocators
540 */
541 void * __must_check krealloc_node_align_noprof(const void *objp,
542 DECL_TOKEN_PARAMS(new_size, token),
543 unsigned long align,
544 gfp_t flags, int nid) __realloc_size(2);
545 #define krealloc_noprof(_o, _s, _f) krealloc_node_align_noprof(_o, PASS_TOKEN_PARAMS(_s, __kmalloc_token(_s)), 1, _f, NUMA_NO_NODE)
546 #if 0 /* kernel-doc */
547 /**
548 * krealloc_node_align - reallocate memory. The contents will remain unchanged.
549 * @p: object to reallocate memory for.
550 * @new_size: how many bytes of memory are required.
551 * @align: desired alignment.
552 * @flags: the type of memory to allocate.
553 * @nid: NUMA node or NUMA_NO_NODE
554 *
555 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
556 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
557 *
558 * Only alignments up to those guaranteed by kmalloc() will be honored. Please see
559 * Documentation/core-api/memory-allocation.rst for more details.
560 *
561 * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
562 * initial memory allocation, every subsequent call to this API for the same
563 * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
564 * __GFP_ZERO is not fully honored by this API.
565 *
566 * When slub_debug_orig_size() is off, krealloc() only knows about the bucket
567 * size of an allocation (but not the exact size it was allocated with) and
568 * hence implements the following semantics for shrinking and growing buffers
569 * with __GFP_ZERO::
570 *
571 * new bucket
572 * 0 size size
573 * |--------|----------------|
574 * | keep | zero |
575 *
576 * Otherwise, the original allocation size 'orig_size' could be used to
577 * precisely clear the requested size, and the new size will also be stored
578 * as the new 'orig_size'.
579 *
580 * In any case, the contents of the object pointed to are preserved up to the
581 * lesser of the new and old sizes.
582 *
583 * Return: pointer to the allocated memory or %NULL in case of error
584 */
585 void *krealloc_node_align(const void *p, size_t new_size, unsigned long align, gfp_t flags, int nid);
586 #endif
587 #define krealloc_node_align(p, new_size, align, flags, nid) \
588 alloc_hooks(krealloc_node_align_noprof(p, PASS_TOKEN_PARAMS(new_size, __kmalloc_token(new_size)), align, flags, nid))
589 #define krealloc_node(_o, _s, _f, _n) krealloc_node_align(_o, _s, 1, _f, _n)
590 #define krealloc(...) krealloc_node(__VA_ARGS__, NUMA_NO_NODE)
591
592 void kfree(const void *objp);
593 void kfree_nolock(const void *objp);
594 void kfree_sensitive(const void *objp);
595
596 DEFINE_FREE(kfree, void *, if (!IS_ERR_OR_NULL(_T)) kfree(_T))
597 DEFINE_FREE(kfree_sensitive, void *, if (_T) kfree_sensitive(_T))
598
599 size_t ksize(const void *objp);
600
601 #ifdef CONFIG_PRINTK
602 bool kmem_dump_obj(void *object);
603 #else
kmem_dump_obj(void * object)604 static inline bool kmem_dump_obj(void *object) { return false; }
605 #endif
606
607 /*
608 * Some archs want to perform DMA into kmalloc caches and need a guaranteed
609 * alignment larger than the alignment of a 64-bit integer.
610 * Setting ARCH_DMA_MINALIGN in arch headers allows that.
611 */
612 #ifdef ARCH_HAS_DMA_MINALIGN
613 #if ARCH_DMA_MINALIGN > 8 && !defined(ARCH_KMALLOC_MINALIGN)
614 #define ARCH_KMALLOC_MINALIGN ARCH_DMA_MINALIGN
615 #endif
616 #endif
617
618 #ifndef ARCH_KMALLOC_MINALIGN
619 #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
620 #elif ARCH_KMALLOC_MINALIGN > 8
621 #define KMALLOC_MIN_SIZE ARCH_KMALLOC_MINALIGN
622 #define KMALLOC_SHIFT_LOW ilog2(KMALLOC_MIN_SIZE)
623 #endif
624
625 /*
626 * Setting ARCH_SLAB_MINALIGN in arch headers allows a different alignment.
627 * Intended for arches that get misalignment faults even for 64 bit integer
628 * aligned buffers.
629 */
630 #ifndef ARCH_SLAB_MINALIGN
631 #define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
632 #endif
633
634 /*
635 * Arches can define this function if they want to decide the minimum slab
636 * alignment at runtime. The value returned by the function must be a power
637 * of two and >= ARCH_SLAB_MINALIGN.
638 */
639 #ifndef arch_slab_minalign
arch_slab_minalign(void)640 static inline unsigned int arch_slab_minalign(void)
641 {
642 return ARCH_SLAB_MINALIGN;
643 }
644 #endif
645
646 /*
647 * kmem_cache_alloc and friends return pointers aligned to ARCH_SLAB_MINALIGN.
648 * kmalloc and friends return pointers aligned to both ARCH_KMALLOC_MINALIGN
649 * and ARCH_SLAB_MINALIGN, but here we only assume the former alignment.
650 */
651 #define __assume_kmalloc_alignment __assume_aligned(ARCH_KMALLOC_MINALIGN)
652 #define __assume_slab_alignment __assume_aligned(ARCH_SLAB_MINALIGN)
653 #define __assume_page_alignment __assume_aligned(PAGE_SIZE)
654
655 /*
656 * Kmalloc array related definitions
657 */
658
659 /*
660 * SLUB directly allocates requests fitting in to an order-1 page
661 * (PAGE_SIZE*2). Larger requests are passed to the page allocator.
662 */
663 #define KMALLOC_SHIFT_HIGH (PAGE_SHIFT + 1)
664 #define KMALLOC_SHIFT_MAX (MAX_PAGE_ORDER + PAGE_SHIFT)
665 #ifndef KMALLOC_SHIFT_LOW
666 #define KMALLOC_SHIFT_LOW 3
667 #endif
668
669 /* Maximum allocatable size */
670 #define KMALLOC_MAX_SIZE (1UL << KMALLOC_SHIFT_MAX)
671 /* Maximum size for which we actually use a slab cache */
672 #define KMALLOC_MAX_CACHE_SIZE (1UL << KMALLOC_SHIFT_HIGH)
673 /* Maximum order allocatable via the slab allocator */
674 #define KMALLOC_MAX_ORDER (KMALLOC_SHIFT_MAX - PAGE_SHIFT)
675
676 /*
677 * Kmalloc subsystem.
678 */
679 #ifndef KMALLOC_MIN_SIZE
680 #define KMALLOC_MIN_SIZE (1 << KMALLOC_SHIFT_LOW)
681 #endif
682
683 /*
684 * This restriction comes from byte sized index implementation.
685 * Page size is normally 2^12 bytes and, in this case, if we want to use
686 * byte sized index which can represent 2^8 entries, the size of the object
687 * should be equal or greater to 2^12 / 2^8 = 2^4 = 16.
688 * If minimum size of kmalloc is less than 16, we use it as minimum object
689 * size and give up to use byte sized index.
690 */
691 #define SLAB_OBJ_MIN_SIZE (KMALLOC_MIN_SIZE < 16 ? \
692 (KMALLOC_MIN_SIZE) : 16)
693
694 #ifdef CONFIG_KMALLOC_PARTITION_CACHES
695 #define KMALLOC_PARTITION_CACHES_NR 15 // # of cache copies
696 #else
697 #define KMALLOC_PARTITION_CACHES_NR 0
698 #endif
699
700 /*
701 * Whenever changing this, take care of that kmalloc_type() and
702 * create_kmalloc_caches() still work as intended.
703 *
704 * KMALLOC_NORMAL can contain only unaccounted objects whereas KMALLOC_CGROUP
705 * is for accounted but unreclaimable and non-dma objects. All the other
706 * kmem caches can have both accounted and unaccounted objects.
707 */
708 enum kmalloc_cache_type {
709 KMALLOC_NORMAL = 0,
710 #ifndef CONFIG_ZONE_DMA
711 KMALLOC_DMA = KMALLOC_NORMAL,
712 #endif
713 #ifndef CONFIG_MEMCG
714 KMALLOC_CGROUP = KMALLOC_NORMAL,
715 #endif
716 #ifndef CONFIG_SLAB_OBJ_EXT
717 KMALLOC_NO_OBJ_EXT = KMALLOC_NORMAL,
718 #endif
719 KMALLOC_PARTITION_START = KMALLOC_NORMAL,
720 KMALLOC_PARTITION_END = KMALLOC_PARTITION_START + KMALLOC_PARTITION_CACHES_NR,
721 #ifdef CONFIG_SLUB_TINY
722 KMALLOC_RECLAIM = KMALLOC_NORMAL,
723 #else
724 KMALLOC_RECLAIM,
725 #endif
726 #ifdef CONFIG_ZONE_DMA
727 KMALLOC_DMA,
728 #endif
729 #ifdef CONFIG_MEMCG
730 KMALLOC_CGROUP,
731 #endif
732 #ifdef CONFIG_SLAB_OBJ_EXT
733 KMALLOC_NO_OBJ_EXT,
734 #endif
735 NR_KMALLOC_TYPES
736 };
737
738 typedef struct kmem_cache * kmem_buckets[KMALLOC_SHIFT_HIGH + 1];
739
740 extern kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES];
741
742 /*
743 * Define gfp bits that should not be set for KMALLOC_NORMAL.
744 */
745 #define KMALLOC_NOT_NORMAL_BITS \
746 (__GFP_RECLAIMABLE | \
747 (IS_ENABLED(CONFIG_ZONE_DMA) ? __GFP_DMA : 0) | \
748 (IS_ENABLED(CONFIG_MEMCG) ? __GFP_ACCOUNT : 0))
749
kmalloc_type(gfp_t flags,kmalloc_token_t token)750 static __always_inline enum kmalloc_cache_type kmalloc_type(gfp_t flags, kmalloc_token_t token)
751 {
752 /*
753 * The most common case is KMALLOC_NORMAL, so test for it
754 * with a single branch for all the relevant flags.
755 */
756 if (likely((flags & KMALLOC_NOT_NORMAL_BITS) == 0))
757 #ifdef CONFIG_KMALLOC_PARTITION_RANDOM
758 /* KMALLOC_PARTITION_CACHES_NR (=15) copies + the KMALLOC_NORMAL */
759 return KMALLOC_PARTITION_START + hash_64(token.v ^ random_kmalloc_seed,
760 ilog2(KMALLOC_PARTITION_CACHES_NR + 1));
761 #elif defined(CONFIG_KMALLOC_PARTITION_TYPED)
762 return KMALLOC_PARTITION_START + token.v;
763 #else
764 return KMALLOC_NORMAL;
765 #endif
766
767 /*
768 * At least one of the flags has to be set. Their priorities in
769 * decreasing order are:
770 * 1) __GFP_DMA
771 * 2) __GFP_RECLAIMABLE
772 * 3) __GFP_ACCOUNT
773 */
774 if (IS_ENABLED(CONFIG_ZONE_DMA) && (flags & __GFP_DMA))
775 return KMALLOC_DMA;
776 if (!IS_ENABLED(CONFIG_MEMCG) || (flags & __GFP_RECLAIMABLE))
777 return KMALLOC_RECLAIM;
778 else
779 return KMALLOC_CGROUP;
780 }
781
782 /*
783 * Figure out which kmalloc slab an allocation of a certain size
784 * belongs to.
785 * 0 = zero alloc
786 * 1 = 65 .. 96 bytes
787 * 2 = 129 .. 192 bytes
788 * n = 2^(n-1)+1 .. 2^n
789 *
790 * Note: __kmalloc_index() is compile-time optimized, and not runtime optimized;
791 * typical usage is via kmalloc_index() and therefore evaluated at compile-time.
792 * Callers where !size_is_constant should only be test modules, where runtime
793 * overheads of __kmalloc_index() can be tolerated. Also see kmalloc_slab().
794 */
__kmalloc_index(size_t size,bool size_is_constant)795 static __always_inline unsigned int __kmalloc_index(size_t size,
796 bool size_is_constant)
797 {
798 if (!size)
799 return 0;
800
801 if (size <= KMALLOC_MIN_SIZE)
802 return KMALLOC_SHIFT_LOW;
803
804 if (KMALLOC_MIN_SIZE <= 32 && size > 64 && size <= 96)
805 return 1;
806 if (KMALLOC_MIN_SIZE <= 64 && size > 128 && size <= 192)
807 return 2;
808 if (size <= 8) return 3;
809 if (size <= 16) return 4;
810 if (size <= 32) return 5;
811 if (size <= 64) return 6;
812 if (size <= 128) return 7;
813 if (size <= 256) return 8;
814 if (size <= 512) return 9;
815 if (size <= 1024) return 10;
816 if (size <= 2 * 1024) return 11;
817 if (size <= 4 * 1024) return 12;
818 if (size <= 8 * 1024) return 13;
819 if (size <= 16 * 1024) return 14;
820 if (size <= 32 * 1024) return 15;
821 if (size <= 64 * 1024) return 16;
822 if (size <= 128 * 1024) return 17;
823 if (size <= 256 * 1024) return 18;
824 if (size <= 512 * 1024) return 19;
825 if (size <= 1024 * 1024) return 20;
826 if (size <= 2 * 1024 * 1024) return 21;
827
828 if (!IS_ENABLED(CONFIG_PROFILE_ALL_BRANCHES) && size_is_constant)
829 BUILD_BUG_ON_MSG(1, "unexpected size in kmalloc_index()");
830 else
831 BUG();
832
833 /* Will never be reached. Needed because the compiler may complain */
834 return -1;
835 }
836 static_assert(PAGE_SHIFT <= 20);
837 #define kmalloc_index(s) __kmalloc_index(s, true)
838
839 #include <linux/alloc_tag.h>
840
841 /**
842 * kmem_cache_alloc - Allocate an object
843 * @cachep: The cache to allocate from.
844 * @flags: See kmalloc().
845 *
846 * Allocate an object from this cache.
847 * See kmem_cache_zalloc() for a shortcut of adding __GFP_ZERO to flags.
848 *
849 * Return: pointer to the new object or %NULL in case of error
850 */
851 void *kmem_cache_alloc_noprof(struct kmem_cache *cachep,
852 gfp_t flags) __assume_slab_alignment __malloc;
853 #define kmem_cache_alloc(...) alloc_hooks(kmem_cache_alloc_noprof(__VA_ARGS__))
854
855 void *kmem_cache_alloc_lru_noprof(struct kmem_cache *s, struct list_lru *lru,
856 gfp_t gfpflags) __assume_slab_alignment __malloc;
857 #define kmem_cache_alloc_lru(...) alloc_hooks(kmem_cache_alloc_lru_noprof(__VA_ARGS__))
858
859 /**
860 * kmem_cache_charge - memcg charge an already allocated slab memory
861 * @objp: address of the slab object to memcg charge
862 * @gfpflags: describe the allocation context
863 *
864 * kmem_cache_charge allows charging a slab object to the current memcg,
865 * primarily in cases where charging at allocation time might not be possible
866 * because the target memcg is not known (i.e. softirq context)
867 *
868 * The objp should be pointer returned by the slab allocator functions like
869 * kmalloc (with __GFP_ACCOUNT in flags) or kmem_cache_alloc. The memcg charge
870 * behavior can be controlled through gfpflags parameter, which affects how the
871 * necessary internal metadata can be allocated. Including __GFP_NOFAIL denotes
872 * that overcharging is requested instead of failure, but is not applied for the
873 * internal metadata allocation.
874 *
875 * There are several cases where it will return true even if the charging was
876 * not done:
877 * More specifically:
878 *
879 * 1. For !CONFIG_MEMCG or cgroup_disable=memory systems.
880 * 2. Already charged slab objects.
881 * 3. For slab objects from KMALLOC_NORMAL caches - allocated by kmalloc()
882 * without __GFP_ACCOUNT
883 * 4. Allocating internal metadata has failed
884 *
885 * Return: true if charge was successful otherwise false.
886 */
887 bool kmem_cache_charge(void *objp, gfp_t gfpflags);
888 void kmem_cache_free(struct kmem_cache *s, void *objp);
889
890 kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags,
891 unsigned int useroffset, unsigned int usersize,
892 void (*ctor)(void *));
893
894 /*
895 * Bulk allocation and freeing operations. These are accelerated in an
896 * allocator specific way to avoid taking locks repeatedly or building
897 * metadata structures unnecessarily.
898 *
899 * Note that interrupts must be enabled when calling these functions.
900 */
901 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p);
902
903 bool kmem_cache_alloc_bulk_noprof(struct kmem_cache *s, gfp_t flags,
904 size_t size, void **p);
905 #define kmem_cache_alloc_bulk(...) \
906 alloc_hooks(kmem_cache_alloc_bulk_noprof(__VA_ARGS__))
907
kfree_bulk(size_t size,void ** p)908 static __always_inline void kfree_bulk(size_t size, void **p)
909 {
910 kmem_cache_free_bulk(NULL, size, p);
911 }
912
913 void *kmem_cache_alloc_node_noprof(struct kmem_cache *s, gfp_t flags,
914 int node) __assume_slab_alignment __malloc;
915 #define kmem_cache_alloc_node(...) alloc_hooks(kmem_cache_alloc_node_noprof(__VA_ARGS__))
916
917 struct slab_sheaf *
918 kmem_cache_prefill_sheaf(struct kmem_cache *s, gfp_t gfp, unsigned int size);
919
920 int kmem_cache_refill_sheaf(struct kmem_cache *s, gfp_t gfp,
921 struct slab_sheaf **sheafp, unsigned int size);
922
923 void kmem_cache_return_sheaf(struct kmem_cache *s, gfp_t gfp,
924 struct slab_sheaf *sheaf);
925
926 void *kmem_cache_alloc_from_sheaf_noprof(struct kmem_cache *cachep, gfp_t gfp,
927 struct slab_sheaf *sheaf) __assume_slab_alignment __malloc;
928 #define kmem_cache_alloc_from_sheaf(...) \
929 alloc_hooks(kmem_cache_alloc_from_sheaf_noprof(__VA_ARGS__))
930
931 unsigned int kmem_cache_sheaf_size(struct slab_sheaf *sheaf);
932
933 /*
934 * These macros allow declaring a kmem_buckets * parameter alongside size, which
935 * can be compiled out with CONFIG_SLAB_BUCKETS=n so that a large number of call
936 * sites don't have to pass NULL.
937 */
938 #ifdef CONFIG_SLAB_BUCKETS
939 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size), kmem_buckets *(_b)
940 #define PASS_BUCKET_PARAMS(_size, _b) (_size), (_b)
941 #define PASS_BUCKET_PARAM(_b) (_b)
942 #else
943 #define DECL_BUCKET_PARAMS(_size, _b) size_t (_size)
944 #define PASS_BUCKET_PARAMS(_size, _b) (_size)
945 #define PASS_BUCKET_PARAM(_b) NULL
946 #endif
947
948 #define DECL_KMALLOC_PARAMS(_size, _b, _token) DECL_BUCKET_PARAMS(_size, _b) \
949 DECL_TOKEN_PARAM(_token)
950
951 #define PASS_KMALLOC_PARAMS(_size, _b, _token) PASS_BUCKET_PARAMS(_size, _b) \
952 _PASS_TOKEN_PARAM(_token)
953
954 /*
955 * The following functions are not to be used directly and are intended only
956 * for internal use from kmalloc() and kmalloc_node()
957 * with the exception of kunit tests
958 */
959
960 void *__kmalloc_noprof(DECL_TOKEN_PARAMS(size, token), gfp_t flags)
961 __assume_kmalloc_alignment __alloc_size(1);
962
963 void *__kmalloc_node_noprof(DECL_KMALLOC_PARAMS(size, b, token), gfp_t flags, int node)
964 __assume_kmalloc_alignment __alloc_size(1);
965
966 void *__kmalloc_cache_noprof(struct kmem_cache *s, gfp_t flags, size_t size)
967 __assume_kmalloc_alignment __alloc_size(3);
968
969 void *__kmalloc_cache_node_noprof(struct kmem_cache *s, gfp_t gfpflags,
970 int node, size_t size)
971 __assume_kmalloc_alignment __alloc_size(4);
972
973 void *__kmalloc_large_noprof(size_t size, gfp_t flags)
974 __assume_page_alignment __alloc_size(1);
975
976 void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
977 __assume_page_alignment __alloc_size(1);
978
_kmalloc_noprof(size_t size,gfp_t flags,kmalloc_token_t token)979 static __always_inline __alloc_size(1) void *_kmalloc_noprof(size_t size, gfp_t flags, kmalloc_token_t token)
980 {
981 if (__builtin_constant_p(size) && size) {
982 unsigned int index;
983
984 if (size > KMALLOC_MAX_CACHE_SIZE)
985 return __kmalloc_large_noprof(size, flags);
986
987 index = kmalloc_index(size);
988 return __kmalloc_cache_noprof(
989 kmalloc_caches[kmalloc_type(flags, token)][index],
990 flags, size);
991 }
992 return __kmalloc_noprof(PASS_TOKEN_PARAMS(size, token), flags);
993 }
994 #define kmalloc_noprof(...) _kmalloc_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
995 #if 0 /* kernel-doc */
996 /**
997 * kmalloc - allocate kernel memory
998 * @size: how many bytes of memory are required.
999 * @flags: describe the allocation context
1000 *
1001 * kmalloc is the normal method of allocating memory
1002 * for objects smaller than page size in the kernel.
1003 *
1004 * The allocated object address is aligned to at least ARCH_KMALLOC_MINALIGN
1005 * bytes. For @size of power of two bytes, the alignment is also guaranteed
1006 * to be at least to the size. For other sizes, the alignment is guaranteed to
1007 * be at least the largest power-of-two divisor of @size.
1008 *
1009 * The @flags argument may be one of the GFP flags defined at
1010 * include/linux/gfp_types.h and described at
1011 * :ref:`Documentation/core-api/mm-api.rst <mm-api-gfp-flags>`
1012 *
1013 * The recommended usage of the @flags is described at
1014 * :ref:`Documentation/core-api/memory-allocation.rst <memory_allocation>`
1015 *
1016 * Below is a brief outline of the most useful GFP flags
1017 *
1018 * %GFP_KERNEL
1019 * Allocate normal kernel ram. May sleep.
1020 *
1021 * %GFP_NOWAIT
1022 * Allocation will not sleep.
1023 *
1024 * %GFP_ATOMIC
1025 * Allocation will not sleep. May use emergency pools.
1026 *
1027 * Also it is possible to set different flags by OR'ing
1028 * in one or more of the following additional @flags:
1029 *
1030 * %__GFP_ZERO
1031 * Zero the allocated memory before returning. Also see kzalloc().
1032 *
1033 * %__GFP_HIGH
1034 * This allocation has high priority and may use emergency pools.
1035 *
1036 * %__GFP_NOFAIL
1037 * Indicate that this allocation is in no way allowed to fail
1038 * (think twice before using).
1039 *
1040 * %__GFP_NORETRY
1041 * If memory is not immediately available,
1042 * then give up at once.
1043 *
1044 * %__GFP_NOWARN
1045 * If allocation fails, don't issue any warnings.
1046 *
1047 * %__GFP_RETRY_MAYFAIL
1048 * Try really hard to succeed the allocation but fail
1049 * eventually.
1050 */
1051 void *kmalloc(size_t size, gfp_t flags);
1052 #endif
1053 #define kmalloc(size, flags) alloc_hooks(kmalloc_noprof(size, flags))
1054
1055 void *_kmalloc_nolock_noprof(DECL_TOKEN_PARAMS(size, token), gfp_t gfp_flags, int node);
1056 #define kmalloc_nolock_noprof(_s, _f, _n) _kmalloc_nolock_noprof(PASS_TOKEN_PARAMS(_s, __kmalloc_token(_s)), _f, _n)
1057 #if 0 /* kernel-doc */
1058 /**
1059 * kmalloc_nolock - Allocate an object of given size from any context.
1060 * @size: size to allocate
1061 * @gfp_flags: GFP flags. Only __GFP_ACCOUNT and __GFP_ZERO allowed. Also
1062 * __GFP_NOWARN and __GFP_NOMEMALLOC are allowed but added internally thus not
1063 * necessary.
1064 * @node: node number of the target node.
1065 *
1066 * Return: pointer to the new object or NULL in case of error.
1067 * NULL does not mean EBUSY or EAGAIN. It means ENOMEM.
1068 * There is no reason to call it again and expect !NULL.
1069 */
1070 void *kmalloc_nolock(size_t size, gfp_t gfp_flags, int node);
1071 #endif
1072 #define kmalloc_nolock(size, gfp_flags, node) alloc_hooks(kmalloc_nolock_noprof(size, gfp_flags, node))
1073
1074 /**
1075 * __alloc_objs - Allocate objects of a given type using
1076 * @KMALLOC: which size-based kmalloc wrapper to allocate with.
1077 * @GFP: GFP flags for the allocation.
1078 * @TYPE: type to allocate space for.
1079 * @COUNT: how many @TYPE objects to allocate.
1080 *
1081 * Returns: Newly allocated pointer to (first) @TYPE of @COUNT-many
1082 * allocated @TYPE objects, or NULL on failure.
1083 */
1084 #define __alloc_objs(KMALLOC, GFP, TYPE, COUNT) \
1085 ({ \
1086 const size_t __obj_size = size_mul(sizeof(TYPE), COUNT); \
1087 (TYPE *)KMALLOC(__obj_size, GFP); \
1088 })
1089
1090 /**
1091 * __alloc_flex - Allocate an object that has a trailing flexible array
1092 * @KMALLOC: kmalloc wrapper function to use for allocation.
1093 * @GFP: GFP flags for the allocation.
1094 * @TYPE: type of structure to allocate space for.
1095 * @FAM: The name of the flexible array member of @TYPE structure.
1096 * @COUNT: how many @FAM elements to allocate space for.
1097 *
1098 * Returns: Newly allocated pointer to @TYPE with @COUNT-many trailing
1099 * @FAM elements, or NULL on failure or if @COUNT cannot be represented
1100 * by the member of @TYPE that counts the @FAM elements (annotated via
1101 * __counted_by()).
1102 */
1103 #define __alloc_flex(KMALLOC, GFP, TYPE, FAM, COUNT) \
1104 ({ \
1105 const size_t __count = (COUNT); \
1106 const size_t __obj_size = struct_size_t(TYPE, FAM, __count); \
1107 TYPE *__obj_ptr = KMALLOC(__obj_size, GFP); \
1108 if (__obj_ptr) \
1109 __set_flex_counter(__obj_ptr->FAM, __count); \
1110 __obj_ptr; \
1111 })
1112
1113 /**
1114 * kmalloc_obj - Allocate a single instance of the given type
1115 * @VAR_OR_TYPE: Variable or type to allocate.
1116 * @...: optional GFP flags for the allocation (GFP_KERNEL when not specified).
1117 *
1118 * Returns: newly allocated pointer to a @VAR_OR_TYPE on success, or NULL
1119 * on failure.
1120 */
1121 #define kmalloc_obj(VAR_OR_TYPE, ...) \
1122 __alloc_objs(kmalloc, default_gfp(__VA_ARGS__), typeof(VAR_OR_TYPE), 1)
1123
1124 /**
1125 * kmalloc_objs - Allocate an array of the given type
1126 * @VAR_OR_TYPE: Variable or type to allocate an array of.
1127 * @COUNT: How many elements in the array.
1128 * @...: optional GFP flags for the allocation (GFP_KERNEL when not specified).
1129 *
1130 * Returns: newly allocated pointer to array of @VAR_OR_TYPE on success,
1131 * or NULL on failure.
1132 */
1133 #define kmalloc_objs(VAR_OR_TYPE, COUNT, ...) \
1134 __alloc_objs(kmalloc, default_gfp(__VA_ARGS__), typeof(VAR_OR_TYPE), COUNT)
1135
1136 /**
1137 * kmalloc_flex - Allocate a single instance of the given flexible structure
1138 * @VAR_OR_TYPE: Variable or type to allocate (with its flex array).
1139 * @FAM: The name of the flexible array member of the structure.
1140 * @COUNT: How many flexible array member elements are desired.
1141 * @...: optional GFP flags for the allocation (GFP_KERNEL when not specified).
1142 *
1143 * Returns: newly allocated pointer to @VAR_OR_TYPE on success, NULL on
1144 * failure. If @FAM has been annotated with __counted_by(), the allocation
1145 * will immediately fail if @COUNT is larger than what the type of the
1146 * struct's counter variable can represent.
1147 */
1148 #define kmalloc_flex(VAR_OR_TYPE, FAM, COUNT, ...) \
1149 __alloc_flex(kmalloc, default_gfp(__VA_ARGS__), typeof(VAR_OR_TYPE), FAM, COUNT)
1150
1151 /* All kzalloc aliases for kmalloc_(obj|objs|flex). */
1152 #define kzalloc_obj(P, ...) \
1153 __alloc_objs(kzalloc, default_gfp(__VA_ARGS__), typeof(P), 1)
1154 #define kzalloc_objs(P, COUNT, ...) \
1155 __alloc_objs(kzalloc, default_gfp(__VA_ARGS__), typeof(P), COUNT)
1156 #define kzalloc_flex(P, FAM, COUNT, ...) \
1157 __alloc_flex(kzalloc, default_gfp(__VA_ARGS__), typeof(P), FAM, COUNT)
1158
1159 /* All kvmalloc aliases for kmalloc_(obj|objs|flex). */
1160 #define kvmalloc_obj(P, ...) \
1161 __alloc_objs(kvmalloc, default_gfp(__VA_ARGS__), typeof(P), 1)
1162 #define kvmalloc_objs(P, COUNT, ...) \
1163 __alloc_objs(kvmalloc, default_gfp(__VA_ARGS__), typeof(P), COUNT)
1164 #define kvmalloc_flex(P, FAM, COUNT, ...) \
1165 __alloc_flex(kvmalloc, default_gfp(__VA_ARGS__), typeof(P), FAM, COUNT)
1166
1167 /* All kvzalloc aliases for kmalloc_(obj|objs|flex). */
1168 #define kvzalloc_obj(P, ...) \
1169 __alloc_objs(kvzalloc, default_gfp(__VA_ARGS__), typeof(P), 1)
1170 #define kvzalloc_objs(P, COUNT, ...) \
1171 __alloc_objs(kvzalloc, default_gfp(__VA_ARGS__), typeof(P), COUNT)
1172 #define kvzalloc_flex(P, FAM, COUNT, ...) \
1173 __alloc_flex(kvzalloc, default_gfp(__VA_ARGS__), typeof(P), FAM, COUNT)
1174
1175 #define kmem_buckets_alloc(_b, _size, _flags) \
1176 alloc_hooks(__kmalloc_node_noprof(PASS_KMALLOC_PARAMS(_size, _b, __kmalloc_token(_size)), _flags, NUMA_NO_NODE))
1177
1178 #define kmem_buckets_alloc_node_track_caller(_b, _size, _flags, _node) \
1179 alloc_hooks(__kmalloc_node_track_caller_noprof(PASS_KMALLOC_PARAMS(_size, _b, __kmalloc_token(_size)), _flags, _node, _RET_IP_))
1180
1181 #define kmem_buckets_alloc_track_caller(_b, _size, _flags) \
1182 kmem_buckets_alloc_node_track_caller(_b, _size, _flags, NUMA_NO_NODE)
1183
_kmalloc_node_noprof(size_t size,gfp_t flags,int node,kmalloc_token_t token)1184 static __always_inline __alloc_size(1) void *_kmalloc_node_noprof(size_t size, gfp_t flags, int node, kmalloc_token_t token)
1185 {
1186 if (__builtin_constant_p(size) && size) {
1187 unsigned int index;
1188
1189 if (size > KMALLOC_MAX_CACHE_SIZE)
1190 return __kmalloc_large_node_noprof(size, flags, node);
1191
1192 index = kmalloc_index(size);
1193 return __kmalloc_cache_node_noprof(
1194 kmalloc_caches[kmalloc_type(flags, token)][index],
1195 flags, node, size);
1196 }
1197 return __kmalloc_node_noprof(PASS_KMALLOC_PARAMS(size, NULL, token), flags, node);
1198 }
1199 #define kmalloc_node_noprof(...) _kmalloc_node_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1200 #define kmalloc_node(...) alloc_hooks(kmalloc_node_noprof(__VA_ARGS__))
1201
_kmalloc_array_noprof(size_t n,size_t size,gfp_t flags,kmalloc_token_t token)1202 static inline __alloc_size(1, 2) void *_kmalloc_array_noprof(size_t n, size_t size, gfp_t flags, kmalloc_token_t token)
1203 {
1204 size_t bytes;
1205
1206 if (unlikely(check_mul_overflow(n, size, &bytes)))
1207 return NULL;
1208 return _kmalloc_noprof(bytes, flags, token);
1209 }
1210 #define kmalloc_array_noprof(...) _kmalloc_array_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1211 #if 0 /* kernel-doc */
1212 /**
1213 * kmalloc_array - allocate memory for an array.
1214 * @n: number of elements.
1215 * @size: element size.
1216 * @flags: the type of memory to allocate (see kmalloc).
1217 */
1218 void *kmalloc_array(size_t n, size_t size, gfp_t flags);
1219 #endif
1220 #define kmalloc_array(n, size, flags) alloc_hooks(kmalloc_array_noprof(n, size, flags))
1221
_krealloc_array_noprof(void * p,size_t new_n,size_t new_size,gfp_t flags,kmalloc_token_t token)1222 static inline __realloc_size(2, 3) void * __must_check _krealloc_array_noprof(void *p,
1223 size_t new_n,
1224 size_t new_size,
1225 gfp_t flags, kmalloc_token_t token)
1226 {
1227 size_t bytes;
1228
1229 if (unlikely(check_mul_overflow(new_n, new_size, &bytes)))
1230 return NULL;
1231
1232 return krealloc_node_align_noprof(p, PASS_TOKEN_PARAMS(bytes, token), 1, flags, NUMA_NO_NODE);
1233 }
1234 #define krealloc_array_noprof(...) _krealloc_array_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1235 #if 0 /* kernel-doc */
1236 /**
1237 * krealloc_array - reallocate memory for an array.
1238 * @p: pointer to the memory chunk to reallocate
1239 * @new_n: new number of elements to alloc
1240 * @new_size: new size of a single member of the array
1241 * @flags: the type of memory to allocate (see kmalloc)
1242 *
1243 * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
1244 * initial memory allocation, every subsequent call to this API for the same
1245 * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
1246 * __GFP_ZERO is not fully honored by this API.
1247 *
1248 * See krealloc_noprof() for further details.
1249 *
1250 * In any case, the contents of the object pointed to are preserved up to the
1251 * lesser of the new and old sizes.
1252 */
1253 void *krealloc_array(void *p, size_t new_n, size_t new_size, gfp_t flags);
1254 #endif
1255 #define krealloc_array(p, new_n, new_size, flags) alloc_hooks(krealloc_array_noprof(p, new_n, new_size, flags))
1256
1257 /**
1258 * kcalloc - allocate memory for an array. The memory is set to zero.
1259 * @n: number of elements.
1260 * @size: element size.
1261 * @flags: the type of memory to allocate (see kmalloc).
1262 */
1263 #define kcalloc(n, size, flags) kmalloc_array(n, size, (flags) | __GFP_ZERO)
1264
1265 void *__kmalloc_node_track_caller_noprof(DECL_KMALLOC_PARAMS(size, b, token), gfp_t flags, int node,
1266 unsigned long caller) __alloc_size(1);
1267 #define kmalloc_node_track_caller_noprof(size, flags, node, caller) \
1268 __kmalloc_node_track_caller_noprof(PASS_KMALLOC_PARAMS(size, NULL, __kmalloc_token(size)), flags, node, caller)
1269 #define kmalloc_node_track_caller(...) \
1270 alloc_hooks(kmalloc_node_track_caller_noprof(__VA_ARGS__, _RET_IP_))
1271
1272 /*
1273 * kmalloc_track_caller is a special version of kmalloc that records the
1274 * calling function of the routine calling it for slab leak tracking instead
1275 * of just the calling function (confusing, eh?).
1276 * It's useful when the call to kmalloc comes from a widely-used standard
1277 * allocator where we care about the real place the memory allocation
1278 * request comes from.
1279 */
1280 #define kmalloc_track_caller(...) kmalloc_node_track_caller(__VA_ARGS__, NUMA_NO_NODE)
1281
1282 #define kmalloc_track_caller_noprof(...) \
1283 kmalloc_node_track_caller_noprof(__VA_ARGS__, NUMA_NO_NODE, _RET_IP_)
1284
_kmalloc_array_node_noprof(size_t n,size_t size,gfp_t flags,int node,kmalloc_token_t token)1285 static inline __alloc_size(1, 2) void *_kmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags,
1286 int node, kmalloc_token_t token)
1287 {
1288 size_t bytes;
1289
1290 if (unlikely(check_mul_overflow(n, size, &bytes)))
1291 return NULL;
1292 if (__builtin_constant_p(n) && __builtin_constant_p(size))
1293 return _kmalloc_node_noprof(bytes, flags, node, token);
1294 return __kmalloc_node_noprof(PASS_KMALLOC_PARAMS(bytes, NULL, token), flags, node);
1295 }
1296 #define kmalloc_array_node_noprof(...) _kmalloc_array_node_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1297 #define kmalloc_array_node(...) alloc_hooks(kmalloc_array_node_noprof(__VA_ARGS__))
1298
1299 #define kcalloc_node(_n, _size, _flags, _node) \
1300 kmalloc_array_node(_n, _size, (_flags) | __GFP_ZERO, _node)
1301
1302 /*
1303 * Shortcuts
1304 */
1305 #define kmem_cache_zalloc(_k, _flags) kmem_cache_alloc(_k, (_flags)|__GFP_ZERO)
1306
_kzalloc_noprof(size_t size,gfp_t flags,kmalloc_token_t token)1307 static inline __alloc_size(1) void *_kzalloc_noprof(size_t size, gfp_t flags, kmalloc_token_t token)
1308 {
1309 return _kmalloc_noprof(size, flags | __GFP_ZERO, token);
1310 }
1311 #define kzalloc_noprof(...) _kzalloc_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1312 #if 0 /* kernel-doc */
1313 /**
1314 * kzalloc - allocate memory. The memory is set to zero.
1315 * @size: how many bytes of memory are required.
1316 * @flags: the type of memory to allocate (see kmalloc).
1317 */
1318 void *kzalloc(size_t size, gfp_t flags);
1319 #endif
1320 #define kzalloc(size, flags) alloc_hooks(kzalloc_noprof(size, flags))
1321 #define kzalloc_node(_size, _flags, _node) kmalloc_node(_size, (_flags)|__GFP_ZERO, _node)
1322
1323 void *__kvmalloc_node_noprof(DECL_KMALLOC_PARAMS(size, b, token), unsigned long align,
1324 gfp_t flags, int node) __alloc_size(1);
1325 #define kvmalloc_node_align_noprof(_size, _align, _flags, _node) \
1326 __kvmalloc_node_noprof(PASS_KMALLOC_PARAMS(_size, NULL, __kmalloc_token(_size)), _align, _flags, _node)
1327 #define kvmalloc_node_align(...) \
1328 alloc_hooks(kvmalloc_node_align_noprof(__VA_ARGS__))
1329 #if 0 /* kernel-doc */
1330 /**
1331 * kvmalloc_node - attempt to allocate physically contiguous memory, but upon
1332 * failure, fall back to non-contiguous (vmalloc) allocation.
1333 * @size: size of the request.
1334 * @flags: gfp mask for the allocation - must be compatible (superset) with GFP_KERNEL.
1335 * @node: numa node to allocate from
1336 *
1337 * Only alignments up to those guaranteed by kmalloc() will be honored. Please see
1338 * Documentation/core-api/memory-allocation.rst for more details.
1339 *
1340 * Uses kmalloc to get the memory but if the allocation fails then falls back
1341 * to the vmalloc allocator. Use kvfree for freeing the memory.
1342 *
1343 * GFP_NOWAIT and GFP_ATOMIC are supported, the __GFP_NORETRY modifier is not.
1344 * __GFP_RETRY_MAYFAIL is supported, and it should be used only if kmalloc is
1345 * preferable to the vmalloc fallback, due to visible performance drawbacks.
1346 *
1347 * Return: pointer to the allocated memory of %NULL in case of failure
1348 */
1349 void *kvmalloc_node(size_t size, gfp_t flags, int node);
1350 #endif
1351 #define kvmalloc_node(size, flags, node) kvmalloc_node_align(size, 1, flags, node)
1352 #define kvmalloc_node_noprof(size, flags, node) \
1353 kvmalloc_node_align_noprof(size, 1, flags, node)
1354 #define kvmalloc(...) kvmalloc_node(__VA_ARGS__, NUMA_NO_NODE)
1355 #define kvmalloc_noprof(_size, _flags) kvmalloc_node_noprof(_size, _flags, NUMA_NO_NODE)
1356 #define kvzalloc(_size, _flags) kvmalloc(_size, (_flags)|__GFP_ZERO)
1357
1358 #define kvzalloc_node(_size, _flags, _node) kvmalloc_node(_size, (_flags)|__GFP_ZERO, _node)
1359
1360 #define kmem_buckets_valloc(_b, _size, _flags) \
1361 alloc_hooks(__kvmalloc_node_noprof(PASS_KMALLOC_PARAMS(_size, _b, __kmalloc_token(_size)), 1, _flags, NUMA_NO_NODE))
1362
1363 static inline __alloc_size(1, 2) void *
_kvmalloc_array_node_noprof(size_t n,size_t size,gfp_t flags,int node,kmalloc_token_t token)1364 _kvmalloc_array_node_noprof(size_t n, size_t size, gfp_t flags, int node, kmalloc_token_t token)
1365 {
1366 size_t bytes;
1367
1368 if (unlikely(check_mul_overflow(n, size, &bytes)))
1369 return NULL;
1370
1371 return __kvmalloc_node_noprof(PASS_KMALLOC_PARAMS(bytes, NULL, token), 1, flags, node);
1372 }
1373 #define kvmalloc_array_node_noprof(...) _kvmalloc_array_node_noprof(__VA_ARGS__, __kmalloc_token(__VA_ARGS__))
1374 #define kvmalloc_array_noprof(...) kvmalloc_array_node_noprof(__VA_ARGS__, NUMA_NO_NODE)
1375 #define kvcalloc_node_noprof(_n,_s,_f,_node) kvmalloc_array_node_noprof(_n,_s,(_f)|__GFP_ZERO,_node)
1376 #define kvcalloc_noprof(...) kvcalloc_node_noprof(__VA_ARGS__, NUMA_NO_NODE)
1377
1378 #define kvmalloc_array(...) alloc_hooks(kvmalloc_array_noprof(__VA_ARGS__))
1379 #define kvcalloc_node(...) alloc_hooks(kvcalloc_node_noprof(__VA_ARGS__))
1380 #define kvcalloc(...) alloc_hooks(kvcalloc_noprof(__VA_ARGS__))
1381
1382 void *kvrealloc_node_align_noprof(const void *p, DECL_TOKEN_PARAMS(size, token), unsigned long align,
1383 gfp_t flags, int nid) __realloc_size(2);
1384 #if 0 /* kernel-doc */
1385 /**
1386 * kvrealloc_node_align - reallocate memory; contents remain unchanged
1387 * @p: object to reallocate memory for
1388 * @size: the size to reallocate
1389 * @align: desired alignment
1390 * @flags: the flags for the page level allocator
1391 * @nid: NUMA node id
1392 *
1393 * If @p is %NULL, kvrealloc() behaves exactly like kvmalloc(). If @size is 0
1394 * and @p is not a %NULL pointer, the object pointed to is freed.
1395 *
1396 * Only alignments up to those guaranteed by kmalloc() will be honored. Please see
1397 * Documentation/core-api/memory-allocation.rst for more details.
1398 *
1399 * If __GFP_ZERO logic is requested, callers must ensure that, starting with the
1400 * initial memory allocation, every subsequent call to this API for the same
1401 * memory allocation is flagged with __GFP_ZERO. Otherwise, it is possible that
1402 * __GFP_ZERO is not fully honored by this API.
1403 *
1404 * In any case, the contents of the object pointed to are preserved up to the
1405 * lesser of the new and old sizes.
1406 *
1407 * This function must not be called concurrently with itself or kvfree() for the
1408 * same memory allocation.
1409 *
1410 * Return: pointer to the allocated memory or %NULL in case of error
1411 */
1412 void *kvrealloc_node_align(const void *p, size_t size, unsigned long align, gfp_t flags, int nid);
1413 #endif
1414 #define kvrealloc_node_align(p, size, align, flags, nid) \
1415 alloc_hooks(kvrealloc_node_align_noprof(p, PASS_TOKEN_PARAMS(size, __kmalloc_token(size)), align, flags, nid))
1416 #define kvrealloc_node(_p, _s, _f, _n) kvrealloc_node_align(_p, _s, 1, _f, _n)
1417 #define kvrealloc(...) kvrealloc_node(__VA_ARGS__, NUMA_NO_NODE)
1418
1419 extern void kvfree(const void *addr);
1420 DEFINE_FREE(kvfree, void *, if (!IS_ERR_OR_NULL(_T)) kvfree(_T))
1421
1422 extern void kvfree_atomic(const void *addr);
1423 DEFINE_FREE(kvfree_atomic, void *, if (!IS_ERR_OR_NULL(_T)) kvfree_atomic(_T))
1424
1425 extern void kvfree_sensitive(const void *addr, size_t len);
1426
1427 unsigned int kmem_cache_size(struct kmem_cache *s);
1428
1429 #ifndef CONFIG_KVFREE_RCU_BATCHED
kvfree_rcu_barrier(void)1430 static inline void kvfree_rcu_barrier(void)
1431 {
1432 rcu_barrier();
1433 }
1434
kvfree_rcu_barrier_on_cache(struct kmem_cache * s)1435 static inline void kvfree_rcu_barrier_on_cache(struct kmem_cache *s)
1436 {
1437 rcu_barrier();
1438 }
1439
kfree_rcu_scheduler_running(void)1440 static inline void kfree_rcu_scheduler_running(void) { }
1441 #else
1442 void kvfree_rcu_barrier(void);
1443
1444 void kvfree_rcu_barrier_on_cache(struct kmem_cache *s);
1445
1446 void kfree_rcu_scheduler_running(void);
1447 #endif
1448
1449 /**
1450 * kmalloc_size_roundup - Report allocation bucket size for the given size
1451 *
1452 * @size: Number of bytes to round up from.
1453 *
1454 * This returns the number of bytes that would be available in a kmalloc()
1455 * allocation of @size bytes. For example, a 126 byte request would be
1456 * rounded up to the next sized kmalloc bucket, 128 bytes. (This is strictly
1457 * for the general-purpose kmalloc()-based allocations, and is not for the
1458 * pre-sized kmem_cache_alloc()-based allocations.)
1459 *
1460 * Use this to kmalloc() the full bucket size ahead of time instead of using
1461 * ksize() to query the size after an allocation.
1462 */
1463 size_t kmalloc_size_roundup(size_t size);
1464
1465 void __init kmem_cache_init_late(void);
1466 void __init kvfree_rcu_init(void);
1467
1468 #endif /* _LINUX_SLAB_H */
1469