]> git.ipfire.org Git - thirdparty/kernel/stable.git/blame - mm/slab_common.c
KVM: x86/mmu: remove unnecessary "bool shared" argument from functions
[thirdparty/kernel/stable.git] / mm / slab_common.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
039363f3
CL
2/*
3 * Slab allocator functions that are independent of the allocator strategy
4 *
5 * (C) 2012 Christoph Lameter <cl@linux.com>
6 */
7#include <linux/slab.h>
8
9#include <linux/mm.h>
10#include <linux/poison.h>
11#include <linux/interrupt.h>
12#include <linux/memory.h>
1c99ba29 13#include <linux/cache.h>
039363f3 14#include <linux/compiler.h>
d3fb45f3 15#include <linux/kfence.h>
039363f3 16#include <linux/module.h>
20cea968
CL
17#include <linux/cpu.h>
18#include <linux/uaccess.h>
b7454ad3 19#include <linux/seq_file.h>
963e84b0 20#include <linux/dma-mapping.h>
b035f5a6 21#include <linux/swiotlb.h>
b7454ad3 22#include <linux/proc_fs.h>
fcf8a1e4 23#include <linux/debugfs.h>
e86f8b09 24#include <linux/kasan.h>
039363f3
CL
25#include <asm/cacheflush.h>
26#include <asm/tlbflush.h>
27#include <asm/page.h>
2633d7a0 28#include <linux/memcontrol.h>
5cf909c5 29#include <linux/stackdepot.h>
928cec9c 30
44405099 31#include "internal.h"
97d06609
CL
32#include "slab.h"
33
b347aa7b
VA
34#define CREATE_TRACE_POINTS
35#include <trace/events/kmem.h>
36
97d06609 37enum slab_state slab_state;
18004c5d
CL
38LIST_HEAD(slab_caches);
39DEFINE_MUTEX(slab_mutex);
9b030cb8 40struct kmem_cache *kmem_cache;
97d06609 41
657dc2f9
TH
42static LIST_HEAD(slab_caches_to_rcu_destroy);
43static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
44static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
45 slab_caches_to_rcu_destroy_workfn);
46
423c929c
JK
47/*
48 * Set of flags that will prevent slab merging
49 */
50#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
5f0d5a3a 51 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
d0bf7d57 52 SLAB_FAILSLAB | SLAB_NO_MERGE | kasan_never_merge())
423c929c 53
230e9fc2 54#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
6d6ea1e9 55 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
423c929c
JK
56
57/*
58 * Merge control. If this is set then no merging of slab caches will occur.
423c929c 59 */
7660a6fd 60static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
423c929c
JK
61
62static int __init setup_slab_nomerge(char *str)
63{
7660a6fd 64 slab_nomerge = true;
423c929c
JK
65 return 1;
66}
67
82edd9d5
RA
68static int __init setup_slab_merge(char *str)
69{
70 slab_nomerge = false;
71 return 1;
72}
73
423c929c
JK
74#ifdef CONFIG_SLUB
75__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
82edd9d5 76__setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
423c929c
JK
77#endif
78
79__setup("slab_nomerge", setup_slab_nomerge);
82edd9d5 80__setup("slab_merge", setup_slab_merge);
423c929c 81
07f361b2
JK
82/*
83 * Determine the size of a slab object
84 */
85unsigned int kmem_cache_size(struct kmem_cache *s)
86{
87 return s->object_size;
88}
89EXPORT_SYMBOL(kmem_cache_size);
90
77be4b13 91#ifdef CONFIG_DEBUG_VM
f4957d5b 92static int kmem_cache_sanity_check(const char *name, unsigned int size)
039363f3 93{
74c1d3e0 94 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
77be4b13
SK
95 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
96 return -EINVAL;
039363f3 97 }
b920536a 98
20cea968 99 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
77be4b13
SK
100 return 0;
101}
102#else
f4957d5b 103static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
77be4b13
SK
104{
105 return 0;
106}
20cea968
CL
107#endif
108
692ae74a
BL
109/*
110 * Figure out what the alignment of the objects will be given a set of
111 * flags, a user specified alignment and the size of the objects.
112 */
f4957d5b
AD
113static unsigned int calculate_alignment(slab_flags_t flags,
114 unsigned int align, unsigned int size)
692ae74a
BL
115{
116 /*
117 * If the user wants hardware cache aligned objects then follow that
118 * suggestion if the object is sufficiently large.
119 *
120 * The hardware cache alignment cannot override the specified
121 * alignment though. If that is greater then use it.
122 */
123 if (flags & SLAB_HWCACHE_ALIGN) {
f4957d5b 124 unsigned int ralign;
692ae74a
BL
125
126 ralign = cache_line_size();
127 while (size <= ralign / 2)
128 ralign /= 2;
129 align = max(align, ralign);
130 }
131
d949a815 132 align = max(align, arch_slab_minalign());
692ae74a
BL
133
134 return ALIGN(align, sizeof(void *));
135}
136
423c929c
JK
137/*
138 * Find a mergeable slab cache
139 */
140int slab_unmergeable(struct kmem_cache *s)
141{
142 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
143 return 1;
144
423c929c
JK
145 if (s->ctor)
146 return 1;
147
346907ce 148#ifdef CONFIG_HARDENED_USERCOPY
8eb8284b
DW
149 if (s->usersize)
150 return 1;
346907ce 151#endif
8eb8284b 152
423c929c
JK
153 /*
154 * We may have set a slab to be unmergeable during bootstrap.
155 */
156 if (s->refcount < 0)
157 return 1;
158
159 return 0;
160}
161
f4957d5b 162struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
d50112ed 163 slab_flags_t flags, const char *name, void (*ctor)(void *))
423c929c
JK
164{
165 struct kmem_cache *s;
166
c6e28895 167 if (slab_nomerge)
423c929c
JK
168 return NULL;
169
170 if (ctor)
171 return NULL;
172
173 size = ALIGN(size, sizeof(void *));
174 align = calculate_alignment(flags, align, size);
175 size = ALIGN(size, align);
37540008 176 flags = kmem_cache_flags(size, flags, name);
423c929c 177
c6e28895
GM
178 if (flags & SLAB_NEVER_MERGE)
179 return NULL;
180
c7094406 181 list_for_each_entry_reverse(s, &slab_caches, list) {
423c929c
JK
182 if (slab_unmergeable(s))
183 continue;
184
185 if (size > s->size)
186 continue;
187
188 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
189 continue;
190 /*
191 * Check if alignment is compatible.
192 * Courtesy of Adrian Drzewiecki
193 */
194 if ((s->size & ~(align - 1)) != s->size)
195 continue;
196
197 if (s->size - size >= sizeof(void *))
198 continue;
199
95069ac8
JK
200 if (IS_ENABLED(CONFIG_SLAB) && align &&
201 (align > s->align || s->align % align))
202 continue;
203
423c929c
JK
204 return s;
205 }
206 return NULL;
207}
208
c9a77a79 209static struct kmem_cache *create_cache(const char *name,
613a5eb5 210 unsigned int object_size, unsigned int align,
7bbdb81e
AD
211 slab_flags_t flags, unsigned int useroffset,
212 unsigned int usersize, void (*ctor)(void *),
9855609b 213 struct kmem_cache *root_cache)
794b1248
VD
214{
215 struct kmem_cache *s;
216 int err;
217
8eb8284b
DW
218 if (WARN_ON(useroffset + usersize > object_size))
219 useroffset = usersize = 0;
220
794b1248
VD
221 err = -ENOMEM;
222 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
223 if (!s)
224 goto out;
225
226 s->name = name;
613a5eb5 227 s->size = s->object_size = object_size;
794b1248
VD
228 s->align = align;
229 s->ctor = ctor;
346907ce 230#ifdef CONFIG_HARDENED_USERCOPY
8eb8284b
DW
231 s->useroffset = useroffset;
232 s->usersize = usersize;
346907ce 233#endif
794b1248 234
794b1248
VD
235 err = __kmem_cache_create(s, flags);
236 if (err)
237 goto out_free_cache;
238
239 s->refcount = 1;
240 list_add(&s->list, &slab_caches);
794b1248
VD
241 return s;
242
243out_free_cache:
7c4da061 244 kmem_cache_free(kmem_cache, s);
b9dad156
ZL
245out:
246 return ERR_PTR(err);
794b1248 247}
45906855 248
f496990f
MR
249/**
250 * kmem_cache_create_usercopy - Create a cache with a region suitable
251 * for copying to userspace
77be4b13
SK
252 * @name: A string which is used in /proc/slabinfo to identify this cache.
253 * @size: The size of objects to be created in this cache.
254 * @align: The required alignment for the objects.
255 * @flags: SLAB flags
8eb8284b
DW
256 * @useroffset: Usercopy region offset
257 * @usersize: Usercopy region size
77be4b13
SK
258 * @ctor: A constructor for the objects.
259 *
77be4b13
SK
260 * Cannot be called within a interrupt, but can be interrupted.
261 * The @ctor is run when new pages are allocated by the cache.
262 *
263 * The flags are
264 *
265 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
266 * to catch references to uninitialised memory.
267 *
f496990f 268 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
77be4b13
SK
269 * for buffer overruns.
270 *
271 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
272 * cacheline. This can be beneficial if you're counting cycles as closely
273 * as davem.
f496990f
MR
274 *
275 * Return: a pointer to the cache on success, NULL on failure.
77be4b13 276 */
2633d7a0 277struct kmem_cache *
f4957d5b
AD
278kmem_cache_create_usercopy(const char *name,
279 unsigned int size, unsigned int align,
7bbdb81e
AD
280 slab_flags_t flags,
281 unsigned int useroffset, unsigned int usersize,
8eb8284b 282 void (*ctor)(void *))
77be4b13 283{
40911a79 284 struct kmem_cache *s = NULL;
3dec16ea 285 const char *cache_name;
3965fc36 286 int err;
039363f3 287
afe0c26d
VB
288#ifdef CONFIG_SLUB_DEBUG
289 /*
290 * If no slub_debug was enabled globally, the static key is not yet
291 * enabled by setup_slub_debug(). Enable it if the cache is being
292 * created with any of the debugging flags passed explicitly.
5cf909c5
OG
293 * It's also possible that this is the first cache created with
294 * SLAB_STORE_USER and we should init stack_depot for it.
afe0c26d
VB
295 */
296 if (flags & SLAB_DEBUG_FLAGS)
297 static_branch_enable(&slub_debug_enabled);
5cf909c5
OG
298 if (flags & SLAB_STORE_USER)
299 stack_depot_init();
afe0c26d
VB
300#endif
301
77be4b13 302 mutex_lock(&slab_mutex);
686d550d 303
794b1248 304 err = kmem_cache_sanity_check(name, size);
3aa24f51 305 if (err) {
3965fc36 306 goto out_unlock;
3aa24f51 307 }
686d550d 308
e70954fd
TG
309 /* Refuse requests with allocator specific flags */
310 if (flags & ~SLAB_FLAGS_PERMITTED) {
311 err = -EINVAL;
312 goto out_unlock;
313 }
314
d8843922
GC
315 /*
316 * Some allocators will constraint the set of valid flags to a subset
317 * of all flags. We expect them to define CACHE_CREATE_MASK in this
318 * case, and we'll just provide them with a sanitized version of the
319 * passed flags.
320 */
321 flags &= CACHE_CREATE_MASK;
686d550d 322
8eb8284b 323 /* Fail closed on bad usersize of useroffset values. */
346907ce
VB
324 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
325 WARN_ON(!usersize && useroffset) ||
8eb8284b
DW
326 WARN_ON(size < usersize || size - usersize < useroffset))
327 usersize = useroffset = 0;
328
329 if (!usersize)
330 s = __kmem_cache_alias(name, size, align, flags, ctor);
794b1248 331 if (s)
3965fc36 332 goto out_unlock;
2633d7a0 333
3dec16ea 334 cache_name = kstrdup_const(name, GFP_KERNEL);
794b1248
VD
335 if (!cache_name) {
336 err = -ENOMEM;
337 goto out_unlock;
338 }
7c9adf5a 339
613a5eb5 340 s = create_cache(cache_name, size,
c9a77a79 341 calculate_alignment(flags, align, size),
9855609b 342 flags, useroffset, usersize, ctor, NULL);
794b1248
VD
343 if (IS_ERR(s)) {
344 err = PTR_ERR(s);
3dec16ea 345 kfree_const(cache_name);
794b1248 346 }
3965fc36
VD
347
348out_unlock:
20cea968 349 mutex_unlock(&slab_mutex);
03afc0e2 350
ba3253c7 351 if (err) {
686d550d 352 if (flags & SLAB_PANIC)
4acaa7d5 353 panic("%s: Failed to create slab '%s'. Error %d\n",
354 __func__, name, err);
686d550d 355 else {
4acaa7d5 356 pr_warn("%s(%s) failed with error %d\n",
357 __func__, name, err);
686d550d
CL
358 dump_stack();
359 }
686d550d
CL
360 return NULL;
361 }
039363f3
CL
362 return s;
363}
8eb8284b
DW
364EXPORT_SYMBOL(kmem_cache_create_usercopy);
365
f496990f
MR
366/**
367 * kmem_cache_create - Create a cache.
368 * @name: A string which is used in /proc/slabinfo to identify this cache.
369 * @size: The size of objects to be created in this cache.
370 * @align: The required alignment for the objects.
371 * @flags: SLAB flags
372 * @ctor: A constructor for the objects.
373 *
374 * Cannot be called within a interrupt, but can be interrupted.
375 * The @ctor is run when new pages are allocated by the cache.
376 *
377 * The flags are
378 *
379 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
380 * to catch references to uninitialised memory.
381 *
382 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
383 * for buffer overruns.
384 *
385 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
386 * cacheline. This can be beneficial if you're counting cycles as closely
387 * as davem.
388 *
389 * Return: a pointer to the cache on success, NULL on failure.
390 */
8eb8284b 391struct kmem_cache *
f4957d5b 392kmem_cache_create(const char *name, unsigned int size, unsigned int align,
8eb8284b
DW
393 slab_flags_t flags, void (*ctor)(void *))
394{
6d07d1cd 395 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
8eb8284b
DW
396 ctor);
397}
794b1248 398EXPORT_SYMBOL(kmem_cache_create);
2633d7a0 399
0495e337
WL
400#ifdef SLAB_SUPPORTS_SYSFS
401/*
402 * For a given kmem_cache, kmem_cache_destroy() should only be called
403 * once or there will be a use-after-free problem. The actual deletion
404 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
405 * protection. So they are now done without holding those locks.
406 *
407 * Note that there will be a slight delay in the deletion of sysfs files
408 * if kmem_cache_release() is called indrectly from a work function.
409 */
410static void kmem_cache_release(struct kmem_cache *s)
411{
412 sysfs_slab_unlink(s);
413 sysfs_slab_release(s);
414}
415#else
416static void kmem_cache_release(struct kmem_cache *s)
417{
418 slab_kmem_cache_release(s);
419}
420#endif
421
657dc2f9 422static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
d5b3cf71 423{
657dc2f9
TH
424 LIST_HEAD(to_destroy);
425 struct kmem_cache *s, *s2;
d5b3cf71 426
657dc2f9 427 /*
5f0d5a3a 428 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
657dc2f9 429 * @slab_caches_to_rcu_destroy list. The slab pages are freed
081a06fa 430 * through RCU and the associated kmem_cache are dereferenced
657dc2f9
TH
431 * while freeing the pages, so the kmem_caches should be freed only
432 * after the pending RCU operations are finished. As rcu_barrier()
433 * is a pretty slow operation, we batch all pending destructions
434 * asynchronously.
435 */
436 mutex_lock(&slab_mutex);
437 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
438 mutex_unlock(&slab_mutex);
d5b3cf71 439
657dc2f9
TH
440 if (list_empty(&to_destroy))
441 return;
442
443 rcu_barrier();
444
445 list_for_each_entry_safe(s, s2, &to_destroy, list) {
64dd6849 446 debugfs_slab_release(s);
d3fb45f3 447 kfence_shutdown_cache(s);
0495e337 448 kmem_cache_release(s);
657dc2f9 449 }
d5b3cf71
VD
450}
451
657dc2f9 452static int shutdown_cache(struct kmem_cache *s)
d5b3cf71 453{
f9fa1d91
GT
454 /* free asan quarantined objects */
455 kasan_cache_shutdown(s);
456
657dc2f9
TH
457 if (__kmem_cache_shutdown(s) != 0)
458 return -EBUSY;
d5b3cf71 459
657dc2f9 460 list_del(&s->list);
d5b3cf71 461
5f0d5a3a 462 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
657dc2f9
TH
463 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
464 schedule_work(&slab_caches_to_rcu_destroy_work);
465 } else {
d3fb45f3 466 kfence_shutdown_cache(s);
64dd6849 467 debugfs_slab_release(s);
d5b3cf71 468 }
657dc2f9
TH
469
470 return 0;
d5b3cf71
VD
471}
472
41a21285
CL
473void slab_kmem_cache_release(struct kmem_cache *s)
474{
52b4b950 475 __kmem_cache_release(s);
3dec16ea 476 kfree_const(s->name);
41a21285
CL
477 kmem_cache_free(kmem_cache, s);
478}
479
945cf2b6
CL
480void kmem_cache_destroy(struct kmem_cache *s)
481{
46a9ea66 482 int err = -EBUSY;
d71608a8 483 bool rcu_set;
0495e337 484
bed0a9b5 485 if (unlikely(!s) || !kasan_check_byte(s))
3942d299
SS
486 return;
487
5a836bf6 488 cpus_read_lock();
945cf2b6 489 mutex_lock(&slab_mutex);
b8529907 490
d71608a8
FT
491 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
492
46a9ea66
RA
493 s->refcount--;
494 if (s->refcount)
b8529907
VD
495 goto out_unlock;
496
46a9ea66
RA
497 err = shutdown_cache(s);
498 WARN(err, "%s %s: Slab cache still has objects when called from %pS",
7302e91f 499 __func__, s->name, (void *)_RET_IP_);
b8529907
VD
500out_unlock:
501 mutex_unlock(&slab_mutex);
5a836bf6 502 cpus_read_unlock();
46a9ea66 503 if (!err && !rcu_set)
0495e337 504 kmem_cache_release(s);
945cf2b6
CL
505}
506EXPORT_SYMBOL(kmem_cache_destroy);
507
03afc0e2
VD
508/**
509 * kmem_cache_shrink - Shrink a cache.
510 * @cachep: The cache to shrink.
511 *
512 * Releases as many slabs as possible for a cache.
513 * To help debugging, a zero exit status indicates all slabs were released.
a862f68a
MR
514 *
515 * Return: %0 if all slabs were released, non-zero otherwise
03afc0e2
VD
516 */
517int kmem_cache_shrink(struct kmem_cache *cachep)
518{
55834c59 519 kasan_cache_shrink(cachep);
7e1fa93d 520
610f9c00 521 return __kmem_cache_shrink(cachep);
03afc0e2
VD
522}
523EXPORT_SYMBOL(kmem_cache_shrink);
524
fda90124 525bool slab_is_available(void)
97d06609
CL
526{
527 return slab_state >= UP;
528}
b7454ad3 529
5bb1bb35 530#ifdef CONFIG_PRINTK
2dfe63e6
ME
531static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
532{
533 if (__kfence_obj_info(kpp, object, slab))
534 return;
535 __kmem_obj_info(kpp, object, slab);
536}
537
8e7f37f2
PM
538/**
539 * kmem_dump_obj - Print available slab provenance information
540 * @object: slab object for which to find provenance information.
541 *
542 * This function uses pr_cont(), so that the caller is expected to have
543 * printed out whatever preamble is appropriate. The provenance information
544 * depends on the type of object and on how much debugging is enabled.
545 * For a slab-cache object, the fact that it is a slab object is printed,
546 * and, if available, the slab name, return address, and stack trace from
e548eaa1 547 * the allocation and last free path of that object.
8e7f37f2 548 *
6e284c55
ZL
549 * Return: %true if the pointer is to a not-yet-freed object from
550 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
551 * is to an already-freed object, and %false otherwise.
8e7f37f2 552 */
6e284c55 553bool kmem_dump_obj(void *object)
8e7f37f2
PM
554{
555 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
556 int i;
7213230a 557 struct slab *slab;
8e7f37f2
PM
558 unsigned long ptroffset;
559 struct kmem_obj_info kp = { };
560
6e284c55
ZL
561 /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
562 if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
563 return false;
7213230a 564 slab = virt_to_slab(object);
6e284c55
ZL
565 if (!slab)
566 return false;
567
7213230a 568 kmem_obj_info(&kp, object, slab);
8e7f37f2
PM
569 if (kp.kp_slab_cache)
570 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
571 else
572 pr_cont(" slab%s", cp);
2dfe63e6
ME
573 if (is_kfence_address(object))
574 pr_cont(" (kfence)");
8e7f37f2
PM
575 if (kp.kp_objp)
576 pr_cont(" start %px", kp.kp_objp);
577 if (kp.kp_data_offset)
578 pr_cont(" data offset %lu", kp.kp_data_offset);
579 if (kp.kp_objp) {
580 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
581 pr_cont(" pointer offset %lu", ptroffset);
582 }
346907ce
VB
583 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
584 pr_cont(" size %u", kp.kp_slab_cache->object_size);
8e7f37f2
PM
585 if (kp.kp_ret)
586 pr_cont(" allocated at %pS\n", kp.kp_ret);
587 else
588 pr_cont("\n");
589 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
590 if (!kp.kp_stack[i])
591 break;
592 pr_info(" %pS\n", kp.kp_stack[i]);
593 }
e548eaa1
MS
594
595 if (kp.kp_free_stack[0])
596 pr_cont(" Free path:\n");
597
598 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
599 if (!kp.kp_free_stack[i])
600 break;
601 pr_info(" %pS\n", kp.kp_free_stack[i]);
602 }
603
6e284c55 604 return true;
8e7f37f2 605}
0d3dd2c8 606EXPORT_SYMBOL_GPL(kmem_dump_obj);
5bb1bb35 607#endif
8e7f37f2 608
45530c44 609/* Create a cache during boot when no slab services are available yet */
361d575e
AD
610void __init create_boot_cache(struct kmem_cache *s, const char *name,
611 unsigned int size, slab_flags_t flags,
612 unsigned int useroffset, unsigned int usersize)
45530c44
CL
613{
614 int err;
59bb4798 615 unsigned int align = ARCH_KMALLOC_MINALIGN;
45530c44
CL
616
617 s->name = name;
618 s->size = s->object_size = size;
59bb4798
VB
619
620 /*
621 * For power of two sizes, guarantee natural alignment for kmalloc
622 * caches, regardless of SL*B debugging options.
623 */
624 if (is_power_of_2(size))
625 align = max(align, size);
626 s->align = calculate_alignment(flags, align, size);
627
346907ce 628#ifdef CONFIG_HARDENED_USERCOPY
8eb8284b
DW
629 s->useroffset = useroffset;
630 s->usersize = usersize;
346907ce 631#endif
f7ce3190 632
45530c44
CL
633 err = __kmem_cache_create(s, flags);
634
635 if (err)
361d575e 636 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
45530c44
CL
637 name, size, err);
638
639 s->refcount = -1; /* Exempt from merging for now */
640}
641
0c474d31
CM
642static struct kmem_cache *__init create_kmalloc_cache(const char *name,
643 unsigned int size,
644 slab_flags_t flags)
45530c44
CL
645{
646 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
647
648 if (!s)
649 panic("Out of memory when creating slab %s\n", name);
650
0c474d31 651 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
45530c44
CL
652 list_add(&s->list, &slab_caches);
653 s->refcount = 1;
654 return s;
655}
656
cc252eae 657struct kmem_cache *
a07057dc
AB
658kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
659{ /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
9425c58e
CL
660EXPORT_SYMBOL(kmalloc_caches);
661
3c615294
GR
662#ifdef CONFIG_RANDOM_KMALLOC_CACHES
663unsigned long random_kmalloc_seed __ro_after_init;
664EXPORT_SYMBOL(random_kmalloc_seed);
665#endif
666
2c59dd65
CL
667/*
668 * Conversion table for small slabs sizes / 8 to the index in the
669 * kmalloc array. This is necessary for slabs < 192 since we have non power
670 * of two cache sizes there. The size of larger slabs can be determined using
671 * fls.
672 */
d5f86655 673static u8 size_index[24] __ro_after_init = {
2c59dd65
CL
674 3, /* 8 */
675 4, /* 16 */
676 5, /* 24 */
677 5, /* 32 */
678 6, /* 40 */
679 6, /* 48 */
680 6, /* 56 */
681 6, /* 64 */
682 1, /* 72 */
683 1, /* 80 */
684 1, /* 88 */
685 1, /* 96 */
686 7, /* 104 */
687 7, /* 112 */
688 7, /* 120 */
689 7, /* 128 */
690 2, /* 136 */
691 2, /* 144 */
692 2, /* 152 */
693 2, /* 160 */
694 2, /* 168 */
695 2, /* 176 */
696 2, /* 184 */
697 2 /* 192 */
698};
699
ac914d08 700static inline unsigned int size_index_elem(unsigned int bytes)
2c59dd65
CL
701{
702 return (bytes - 1) / 8;
703}
704
705/*
706 * Find the kmem_cache structure that serves a given size of
707 * allocation
708 */
3c615294 709struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
2c59dd65 710{
d5f86655 711 unsigned int index;
2c59dd65
CL
712
713 if (size <= 192) {
714 if (!size)
715 return ZERO_SIZE_PTR;
716
717 index = size_index[size_index_elem(size)];
61448479 718 } else {
221d7da6 719 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
61448479 720 return NULL;
2c59dd65 721 index = fls(size - 1);
61448479 722 }
2c59dd65 723
3c615294 724 return kmalloc_caches[kmalloc_type(flags, caller)][index];
2c59dd65
CL
725}
726
05a94065
KC
727size_t kmalloc_size_roundup(size_t size)
728{
8446a4de
DL
729 if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
730 /*
731 * The flags don't matter since size_index is common to all.
732 * Neither does the caller for just getting ->object_size.
733 */
734 return kmalloc_slab(size, GFP_KERNEL, 0)->object_size;
735 }
736
05a94065 737 /* Above the smaller buckets, size is a multiple of page size. */
8446a4de 738 if (size && size <= KMALLOC_MAX_SIZE)
05a94065
KC
739 return PAGE_SIZE << get_order(size);
740
3c615294 741 /*
8446a4de
DL
742 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
743 * and very large size - kmalloc() may fail.
3c615294 744 */
8446a4de
DL
745 return size;
746
05a94065
KC
747}
748EXPORT_SYMBOL(kmalloc_size_roundup);
749
cb5d9fb3 750#ifdef CONFIG_ZONE_DMA
494c1dfe
WL
751#define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
752#else
753#define KMALLOC_DMA_NAME(sz)
754#endif
755
756#ifdef CONFIG_MEMCG_KMEM
757#define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
cb5d9fb3 758#else
494c1dfe
WL
759#define KMALLOC_CGROUP_NAME(sz)
760#endif
761
2f7c1c13
VB
762#ifndef CONFIG_SLUB_TINY
763#define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
764#else
765#define KMALLOC_RCL_NAME(sz)
766#endif
767
3c615294
GR
768#ifdef CONFIG_RANDOM_KMALLOC_CACHES
769#define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
770#define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
771#define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz,
772#define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz,
773#define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz,
774#define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz,
775#define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz,
776#define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz,
777#define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz,
778#define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz,
779#define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz,
780#define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
781#define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
782#define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
783#define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
784#define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
785#define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
786#else // CONFIG_RANDOM_KMALLOC_CACHES
787#define KMALLOC_RANDOM_NAME(N, sz)
788#endif
789
cb5d9fb3
PL
790#define INIT_KMALLOC_INFO(__size, __short_size) \
791{ \
792 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
2f7c1c13 793 KMALLOC_RCL_NAME(__short_size) \
494c1dfe
WL
794 KMALLOC_CGROUP_NAME(__short_size) \
795 KMALLOC_DMA_NAME(__short_size) \
3c615294 796 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \
cb5d9fb3
PL
797 .size = __size, \
798}
cb5d9fb3 799
4066c33d
GG
800/*
801 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
d6a71648
HY
802 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
803 * kmalloc-2M.
4066c33d 804 */
af3b5f87 805const struct kmalloc_info_struct kmalloc_info[] __initconst = {
cb5d9fb3
PL
806 INIT_KMALLOC_INFO(0, 0),
807 INIT_KMALLOC_INFO(96, 96),
808 INIT_KMALLOC_INFO(192, 192),
809 INIT_KMALLOC_INFO(8, 8),
810 INIT_KMALLOC_INFO(16, 16),
811 INIT_KMALLOC_INFO(32, 32),
812 INIT_KMALLOC_INFO(64, 64),
813 INIT_KMALLOC_INFO(128, 128),
814 INIT_KMALLOC_INFO(256, 256),
815 INIT_KMALLOC_INFO(512, 512),
816 INIT_KMALLOC_INFO(1024, 1k),
817 INIT_KMALLOC_INFO(2048, 2k),
818 INIT_KMALLOC_INFO(4096, 4k),
819 INIT_KMALLOC_INFO(8192, 8k),
820 INIT_KMALLOC_INFO(16384, 16k),
821 INIT_KMALLOC_INFO(32768, 32k),
822 INIT_KMALLOC_INFO(65536, 64k),
823 INIT_KMALLOC_INFO(131072, 128k),
824 INIT_KMALLOC_INFO(262144, 256k),
825 INIT_KMALLOC_INFO(524288, 512k),
826 INIT_KMALLOC_INFO(1048576, 1M),
d6a71648 827 INIT_KMALLOC_INFO(2097152, 2M)
4066c33d
GG
828};
829
f97d5f63 830/*
34cc6990
DS
831 * Patch up the size_index table if we have strange large alignment
832 * requirements for the kmalloc array. This is only the case for
833 * MIPS it seems. The standard arches will not generate any code here.
834 *
835 * Largest permitted alignment is 256 bytes due to the way we
836 * handle the index determination for the smaller caches.
837 *
838 * Make sure that nothing crazy happens if someone starts tinkering
839 * around with ARCH_KMALLOC_MINALIGN
f97d5f63 840 */
34cc6990 841void __init setup_kmalloc_cache_index_table(void)
f97d5f63 842{
ac914d08 843 unsigned int i;
f97d5f63 844
2c59dd65 845 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
7d6b6cc3 846 !is_power_of_2(KMALLOC_MIN_SIZE));
2c59dd65
CL
847
848 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
ac914d08 849 unsigned int elem = size_index_elem(i);
2c59dd65
CL
850
851 if (elem >= ARRAY_SIZE(size_index))
852 break;
853 size_index[elem] = KMALLOC_SHIFT_LOW;
854 }
855
856 if (KMALLOC_MIN_SIZE >= 64) {
857 /*
0b8f0d87 858 * The 96 byte sized cache is not used if the alignment
2c59dd65
CL
859 * is 64 byte.
860 */
861 for (i = 64 + 8; i <= 96; i += 8)
862 size_index[size_index_elem(i)] = 7;
863
864 }
865
866 if (KMALLOC_MIN_SIZE >= 128) {
867 /*
868 * The 192 byte sized cache is not used if the alignment
869 * is 128 byte. Redirect kmalloc to use the 256 byte cache
870 * instead.
871 */
872 for (i = 128 + 8; i <= 192; i += 8)
873 size_index[size_index_elem(i)] = 8;
874 }
34cc6990
DS
875}
876
963e84b0
CM
877static unsigned int __kmalloc_minalign(void)
878{
c15cdea5
CM
879 unsigned int minalign = dma_get_cache_alignment();
880
05ee7741
PT
881 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
882 is_swiotlb_allocated())
c15cdea5
CM
883 minalign = ARCH_KMALLOC_MINALIGN;
884
885 return max(minalign, arch_slab_minalign());
963e84b0
CM
886}
887
0c474d31 888void __init
13657d0a 889new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
a9730fca 890{
963e84b0
CM
891 unsigned int minalign = __kmalloc_minalign();
892 unsigned int aligned_size = kmalloc_info[idx].size;
893 int aligned_idx = idx;
894
2f7c1c13 895 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
1291523f 896 flags |= SLAB_RECLAIM_ACCOUNT;
494c1dfe 897 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
17c17367 898 if (mem_cgroup_kmem_disabled()) {
494c1dfe
WL
899 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
900 return;
901 }
902 flags |= SLAB_ACCOUNT;
33647783
OK
903 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
904 flags |= SLAB_CACHE_DMA;
494c1dfe 905 }
1291523f 906
3c615294
GR
907#ifdef CONFIG_RANDOM_KMALLOC_CACHES
908 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
909 flags |= SLAB_NO_MERGE;
910#endif
911
13e680fb
WL
912 /*
913 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
914 * KMALLOC_NORMAL caches.
915 */
916 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
d5bf4857
VB
917 flags |= SLAB_NO_MERGE;
918
963e84b0
CM
919 if (minalign > ARCH_KMALLOC_MINALIGN) {
920 aligned_size = ALIGN(aligned_size, minalign);
921 aligned_idx = __kmalloc_index(aligned_size, false);
922 }
923
924 if (!kmalloc_caches[type][aligned_idx])
925 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
926 kmalloc_info[aligned_idx].name[type],
927 aligned_size, flags);
928 if (idx != aligned_idx)
929 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
a9730fca
CL
930}
931
34cc6990
DS
932/*
933 * Create the kmalloc array. Some of the regular kmalloc arrays
934 * may already have been created because they were needed to
935 * enable allocations for slab creation.
936 */
d50112ed 937void __init create_kmalloc_caches(slab_flags_t flags)
34cc6990 938{
13657d0a
PL
939 int i;
940 enum kmalloc_cache_type type;
34cc6990 941
494c1dfe
WL
942 /*
943 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
944 */
33647783 945 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
1291523f
VB
946 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
947 if (!kmalloc_caches[type][i])
948 new_kmalloc_cache(i, type, flags);
f97d5f63 949
1291523f
VB
950 /*
951 * Caches that are not of the two-to-the-power-of size.
952 * These have to be created immediately after the
953 * earlier power of two caches
954 */
955 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
956 !kmalloc_caches[type][1])
957 new_kmalloc_cache(1, type, flags);
958 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
959 !kmalloc_caches[type][2])
960 new_kmalloc_cache(2, type, flags);
961 }
8a965b3b 962 }
3c615294
GR
963#ifdef CONFIG_RANDOM_KMALLOC_CACHES
964 random_kmalloc_seed = get_random_u64();
965#endif
8a965b3b 966
f97d5f63
CL
967 /* Kmalloc array is now usable */
968 slab_state = UP;
f97d5f63 969}
d6a71648
HY
970
971void free_large_kmalloc(struct folio *folio, void *object)
972{
973 unsigned int order = folio_order(folio);
974
975 if (WARN_ON_ONCE(order == 0))
976 pr_warn_once("object pointer: 0x%p\n", object);
977
978 kmemleak_free(object);
979 kasan_kfree_large(object);
27bc50fc 980 kmsan_kfree_large(object);
d6a71648
HY
981
982 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
983 -(PAGE_SIZE << order));
984 __free_pages(folio_page(folio, 0), order);
985}
b1405135
HY
986
987static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
988static __always_inline
989void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
990{
991 struct kmem_cache *s;
992 void *ret;
993
994 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
995 ret = __kmalloc_large_node(size, flags, node);
32868715 996 trace_kmalloc(caller, ret, size,
11e9734b 997 PAGE_SIZE << get_order(size), flags, node);
b1405135
HY
998 return ret;
999 }
1000
3c615294 1001 s = kmalloc_slab(size, flags, caller);
b1405135
HY
1002
1003 if (unlikely(ZERO_OR_NULL_PTR(s)))
1004 return s;
1005
1006 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
1007 ret = kasan_kmalloc(s, ret, size, flags);
32868715 1008 trace_kmalloc(caller, ret, size, s->size, flags, node);
b1405135
HY
1009 return ret;
1010}
1011
1012void *__kmalloc_node(size_t size, gfp_t flags, int node)
1013{
1014 return __do_kmalloc_node(size, flags, node, _RET_IP_);
1015}
1016EXPORT_SYMBOL(__kmalloc_node);
1017
1018void *__kmalloc(size_t size, gfp_t flags)
1019{
1020 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
1021}
1022EXPORT_SYMBOL(__kmalloc);
1023
1024void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
1025 int node, unsigned long caller)
1026{
1027 return __do_kmalloc_node(size, flags, node, caller);
1028}
1029EXPORT_SYMBOL(__kmalloc_node_track_caller);
1030
1031/**
1032 * kfree - free previously allocated memory
ae65a521 1033 * @object: pointer returned by kmalloc() or kmem_cache_alloc()
b1405135
HY
1034 *
1035 * If @object is NULL, no operation is performed.
b1405135
HY
1036 */
1037void kfree(const void *object)
1038{
1039 struct folio *folio;
1040 struct slab *slab;
1041 struct kmem_cache *s;
1042
1043 trace_kfree(_RET_IP_, object);
1044
1045 if (unlikely(ZERO_OR_NULL_PTR(object)))
1046 return;
1047
1048 folio = virt_to_folio(object);
1049 if (unlikely(!folio_test_slab(folio))) {
1050 free_large_kmalloc(folio, (void *)object);
1051 return;
1052 }
1053
1054 slab = folio_slab(folio);
1055 s = slab->slab_cache;
1056 __kmem_cache_free(s, (void *)object, _RET_IP_);
1057}
1058EXPORT_SYMBOL(kfree);
1059
445d41d7
VB
1060/**
1061 * __ksize -- Report full size of underlying allocation
a2076201 1062 * @object: pointer to the object
445d41d7
VB
1063 *
1064 * This should only be used internally to query the true size of allocations.
1065 * It is not meant to be a way to discover the usable size of an allocation
1066 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1067 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1068 * and/or FORTIFY_SOURCE.
1069 *
a2076201 1070 * Return: size of the actual memory used by @object in bytes
445d41d7 1071 */
b1405135
HY
1072size_t __ksize(const void *object)
1073{
1074 struct folio *folio;
1075
1076 if (unlikely(object == ZERO_SIZE_PTR))
1077 return 0;
1078
1079 folio = virt_to_folio(object);
1080
d5eff736
HY
1081 if (unlikely(!folio_test_slab(folio))) {
1082 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1083 return 0;
1084 if (WARN_ON(object != folio_address(folio)))
1085 return 0;
b1405135 1086 return folio_size(folio);
d5eff736 1087 }
b1405135 1088
946fa0db
FT
1089#ifdef CONFIG_SLUB_DEBUG
1090 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1091#endif
1092
b1405135
HY
1093 return slab_ksize(folio_slab(folio)->slab_cache);
1094}
26a40990 1095
26a40990
HY
1096void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1097{
1098 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1099 size, _RET_IP_);
1100
2c1d697f 1101 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
26a40990
HY
1102
1103 ret = kasan_kmalloc(s, ret, size, gfpflags);
1104 return ret;
1105}
1106EXPORT_SYMBOL(kmalloc_trace);
1107
1108void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1109 int node, size_t size)
1110{
1111 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1112
2c1d697f 1113 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
26a40990
HY
1114
1115 ret = kasan_kmalloc(s, ret, size, gfpflags);
1116 return ret;
1117}
1118EXPORT_SYMBOL(kmalloc_node_trace);
45530c44 1119
44405099
LL
1120gfp_t kmalloc_fix_flags(gfp_t flags)
1121{
1122 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1123
1124 flags &= ~GFP_SLAB_BUG_MASK;
1125 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1126 invalid_mask, &invalid_mask, flags, &flags);
1127 dump_stack();
1128
1129 return flags;
1130}
1131
cea371f4
VD
1132/*
1133 * To avoid unnecessary overhead, we pass through large allocation requests
1134 * directly to the page allocator. We use __GFP_COMP, because we will need to
1135 * know the allocation order to free the pages properly in kfree.
1136 */
45530c44 1137
b1405135 1138static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
52383431 1139{
52383431 1140 struct page *page;
a0c3b940
HY
1141 void *ptr = NULL;
1142 unsigned int order = get_order(size);
52383431 1143
44405099
LL
1144 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1145 flags = kmalloc_fix_flags(flags);
1146
52383431 1147 flags |= __GFP_COMP;
a0c3b940
HY
1148 page = alloc_pages_node(node, flags, order);
1149 if (page) {
1150 ptr = page_address(page);
96403bfe
MS
1151 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1152 PAGE_SIZE << order);
6a486c0a 1153 }
a0c3b940
HY
1154
1155 ptr = kasan_kmalloc_large(ptr, size, flags);
1156 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1157 kmemleak_alloc(ptr, size, 1, flags);
27bc50fc 1158 kmsan_kmalloc_large(ptr, size, flags);
a0c3b940
HY
1159
1160 return ptr;
1161}
bf37d791 1162
c4cab557
HY
1163void *kmalloc_large(size_t size, gfp_t flags)
1164{
b1405135 1165 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
c4cab557 1166
2c1d697f
HY
1167 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1168 flags, NUMA_NO_NODE);
52383431
VD
1169 return ret;
1170}
c4cab557 1171EXPORT_SYMBOL(kmalloc_large);
52383431 1172
bf37d791 1173void *kmalloc_large_node(size_t size, gfp_t flags, int node)
f1b6eb6e 1174{
b1405135 1175 void *ret = __kmalloc_large_node(size, flags, node);
bf37d791 1176
2c1d697f
HY
1177 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1178 flags, node);
f1b6eb6e
CL
1179 return ret;
1180}
a0c3b940 1181EXPORT_SYMBOL(kmalloc_large_node);
45530c44 1182
7c00fce9
TG
1183#ifdef CONFIG_SLAB_FREELIST_RANDOM
1184/* Randomize a generic freelist */
ffe4dfe0 1185static void freelist_randomize(unsigned int *list,
302d55d5 1186 unsigned int count)
7c00fce9 1187{
7c00fce9 1188 unsigned int rand;
302d55d5 1189 unsigned int i;
7c00fce9
TG
1190
1191 for (i = 0; i < count; i++)
1192 list[i] = i;
1193
1194 /* Fisher-Yates shuffle */
1195 for (i = count - 1; i > 0; i--) {
ffe4dfe0 1196 rand = get_random_u32_below(i + 1);
7c00fce9
TG
1197 swap(list[i], list[rand]);
1198 }
1199}
1200
1201/* Create a random sequence per cache */
1202int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1203 gfp_t gfp)
1204{
7c00fce9
TG
1205
1206 if (count < 2 || cachep->random_seq)
1207 return 0;
1208
1209 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1210 if (!cachep->random_seq)
1211 return -ENOMEM;
1212
ffe4dfe0 1213 freelist_randomize(cachep->random_seq, count);
7c00fce9
TG
1214 return 0;
1215}
1216
1217/* Destroy the per-cache random freelist sequence */
1218void cache_random_seq_destroy(struct kmem_cache *cachep)
1219{
1220 kfree(cachep->random_seq);
1221 cachep->random_seq = NULL;
1222}
1223#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1224
5b365771 1225#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
e9b4db2b 1226#ifdef CONFIG_SLAB
0825a6f9 1227#define SLABINFO_RIGHTS (0600)
e9b4db2b 1228#else
0825a6f9 1229#define SLABINFO_RIGHTS (0400)
e9b4db2b
WL
1230#endif
1231
b047501c 1232static void print_slabinfo_header(struct seq_file *m)
bcee6e2a
GC
1233{
1234 /*
1235 * Output format version, so at least we can change it
1236 * without _too_ many complaints.
1237 */
1238#ifdef CONFIG_DEBUG_SLAB
1239 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1240#else
1241 seq_puts(m, "slabinfo - version: 2.1\n");
1242#endif
756a025f 1243 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
bcee6e2a
GC
1244 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1245 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1246#ifdef CONFIG_DEBUG_SLAB
756a025f 1247 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
bcee6e2a
GC
1248 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1249#endif
1250 seq_putc(m, '\n');
1251}
1252
c29b5b3d 1253static void *slab_start(struct seq_file *m, loff_t *pos)
b7454ad3 1254{
b7454ad3 1255 mutex_lock(&slab_mutex);
c7094406 1256 return seq_list_start(&slab_caches, *pos);
b7454ad3
GC
1257}
1258
c29b5b3d 1259static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
b7454ad3 1260{
c7094406 1261 return seq_list_next(p, &slab_caches, pos);
b7454ad3
GC
1262}
1263
c29b5b3d 1264static void slab_stop(struct seq_file *m, void *p)
b7454ad3
GC
1265{
1266 mutex_unlock(&slab_mutex);
1267}
1268
b047501c 1269static void cache_show(struct kmem_cache *s, struct seq_file *m)
b7454ad3 1270{
0d7561c6
GC
1271 struct slabinfo sinfo;
1272
1273 memset(&sinfo, 0, sizeof(sinfo));
1274 get_slabinfo(s, &sinfo);
1275
1276 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
10befea9 1277 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
0d7561c6
GC
1278 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1279
1280 seq_printf(m, " : tunables %4u %4u %4u",
1281 sinfo.limit, sinfo.batchcount, sinfo.shared);
1282 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1283 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1284 slabinfo_show_stats(m, s);
1285 seq_putc(m, '\n');
b7454ad3
GC
1286}
1287
1df3b26f 1288static int slab_show(struct seq_file *m, void *p)
749c5415 1289{
c7094406 1290 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
749c5415 1291
c7094406 1292 if (p == slab_caches.next)
1df3b26f 1293 print_slabinfo_header(m);
10befea9 1294 cache_show(s, m);
b047501c
VD
1295 return 0;
1296}
1297
852d8be0
YS
1298void dump_unreclaimable_slab(void)
1299{
7714304f 1300 struct kmem_cache *s;
852d8be0
YS
1301 struct slabinfo sinfo;
1302
1303 /*
1304 * Here acquiring slab_mutex is risky since we don't prefer to get
1305 * sleep in oom path. But, without mutex hold, it may introduce a
1306 * risk of crash.
1307 * Use mutex_trylock to protect the list traverse, dump nothing
1308 * without acquiring the mutex.
1309 */
1310 if (!mutex_trylock(&slab_mutex)) {
1311 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1312 return;
1313 }
1314
1315 pr_info("Unreclaimable slab info:\n");
1316 pr_info("Name Used Total\n");
1317
7714304f 1318 list_for_each_entry(s, &slab_caches, list) {
10befea9 1319 if (s->flags & SLAB_RECLAIM_ACCOUNT)
852d8be0
YS
1320 continue;
1321
1322 get_slabinfo(s, &sinfo);
1323
1324 if (sinfo.num_objs > 0)
10befea9 1325 pr_info("%-17s %10luKB %10luKB\n", s->name,
852d8be0
YS
1326 (sinfo.active_objs * s->size) / 1024,
1327 (sinfo.num_objs * s->size) / 1024);
1328 }
1329 mutex_unlock(&slab_mutex);
1330}
1331
b7454ad3
GC
1332/*
1333 * slabinfo_op - iterator that generates /proc/slabinfo
1334 *
1335 * Output layout:
1336 * cache-name
1337 * num-active-objs
1338 * total-objs
1339 * object size
1340 * num-active-slabs
1341 * total-slabs
1342 * num-pages-per-slab
1343 * + further values on SMP and with statistics enabled
1344 */
1345static const struct seq_operations slabinfo_op = {
1df3b26f 1346 .start = slab_start,
276a2439
WL
1347 .next = slab_next,
1348 .stop = slab_stop,
1df3b26f 1349 .show = slab_show,
b7454ad3
GC
1350};
1351
1352static int slabinfo_open(struct inode *inode, struct file *file)
1353{
1354 return seq_open(file, &slabinfo_op);
1355}
1356
97a32539 1357static const struct proc_ops slabinfo_proc_ops = {
d919b33d 1358 .proc_flags = PROC_ENTRY_PERMANENT,
97a32539
AD
1359 .proc_open = slabinfo_open,
1360 .proc_read = seq_read,
1361 .proc_write = slabinfo_write,
1362 .proc_lseek = seq_lseek,
1363 .proc_release = seq_release,
b7454ad3
GC
1364};
1365
1366static int __init slab_proc_init(void)
1367{
97a32539 1368 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
b7454ad3
GC
1369 return 0;
1370}
1371module_init(slab_proc_init);
fcf8a1e4 1372
5b365771 1373#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
928cec9c 1374
9ed9cac1
KC
1375static __always_inline __realloc_size(2) void *
1376__do_krealloc(const void *p, size_t new_size, gfp_t flags)
928cec9c
AR
1377{
1378 void *ret;
fa9ba3aa 1379 size_t ks;
928cec9c 1380
38931d89 1381 /* Check for double-free before calling ksize. */
d12d9ad8
AK
1382 if (likely(!ZERO_OR_NULL_PTR(p))) {
1383 if (!kasan_check_byte(p))
1384 return NULL;
38931d89 1385 ks = ksize(p);
d12d9ad8
AK
1386 } else
1387 ks = 0;
928cec9c 1388
d12d9ad8 1389 /* If the object still fits, repoison it precisely. */
0316bec2 1390 if (ks >= new_size) {
0116523c 1391 p = kasan_krealloc((void *)p, new_size, flags);
928cec9c 1392 return (void *)p;
0316bec2 1393 }
928cec9c
AR
1394
1395 ret = kmalloc_track_caller(new_size, flags);
d12d9ad8
AK
1396 if (ret && p) {
1397 /* Disable KASAN checks as the object's redzone is accessed. */
1398 kasan_disable_current();
1399 memcpy(ret, kasan_reset_tag(p), ks);
1400 kasan_enable_current();
1401 }
928cec9c
AR
1402
1403 return ret;
1404}
1405
928cec9c
AR
1406/**
1407 * krealloc - reallocate memory. The contents will remain unchanged.
1408 * @p: object to reallocate memory for.
1409 * @new_size: how many bytes of memory are required.
1410 * @flags: the type of memory to allocate.
1411 *
1412 * The contents of the object pointed to are preserved up to the
15d5de49
BG
1413 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1414 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1415 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
a862f68a
MR
1416 *
1417 * Return: pointer to the allocated memory or %NULL in case of error
928cec9c
AR
1418 */
1419void *krealloc(const void *p, size_t new_size, gfp_t flags)
1420{
1421 void *ret;
1422
1423 if (unlikely(!new_size)) {
1424 kfree(p);
1425 return ZERO_SIZE_PTR;
1426 }
1427
1428 ret = __do_krealloc(p, new_size, flags);
772a2fa5 1429 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
928cec9c
AR
1430 kfree(p);
1431
1432 return ret;
1433}
1434EXPORT_SYMBOL(krealloc);
1435
1436/**
453431a5 1437 * kfree_sensitive - Clear sensitive information in memory before freeing
928cec9c
AR
1438 * @p: object to free memory of
1439 *
1440 * The memory of the object @p points to is zeroed before freed.
453431a5 1441 * If @p is %NULL, kfree_sensitive() does nothing.
928cec9c
AR
1442 *
1443 * Note: this function zeroes the whole allocated buffer which can be a good
1444 * deal bigger than the requested buffer size passed to kmalloc(). So be
1445 * careful when using this function in performance sensitive code.
1446 */
453431a5 1447void kfree_sensitive(const void *p)
928cec9c
AR
1448{
1449 size_t ks;
1450 void *mem = (void *)p;
1451
928cec9c 1452 ks = ksize(mem);
38931d89
KC
1453 if (ks) {
1454 kasan_unpoison_range(mem, ks);
fa9ba3aa 1455 memzero_explicit(mem, ks);
38931d89 1456 }
928cec9c
AR
1457 kfree(mem);
1458}
453431a5 1459EXPORT_SYMBOL(kfree_sensitive);
928cec9c 1460
10d1f8cb
ME
1461size_t ksize(const void *objp)
1462{
0d4ca4c9 1463 /*
38931d89
KC
1464 * We need to first check that the pointer to the object is valid.
1465 * The KASAN report printed from ksize() is more useful, then when
1466 * it's printed later when the behaviour could be undefined due to
1467 * a potential use-after-free or double-free.
0d4ca4c9 1468 *
611806b4
AK
1469 * We use kasan_check_byte(), which is supported for the hardware
1470 * tag-based KASAN mode, unlike kasan_check_read/write().
1471 *
1472 * If the pointed to memory is invalid, we return 0 to avoid users of
0d4ca4c9
ME
1473 * ksize() writing to and potentially corrupting the memory region.
1474 *
1475 * We want to perform the check before __ksize(), to avoid potentially
1476 * crashing in __ksize() due to accessing invalid metadata.
1477 */
611806b4 1478 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
0d4ca4c9
ME
1479 return 0;
1480
38931d89 1481 return kfence_ksize(objp) ?: __ksize(objp);
10d1f8cb
ME
1482}
1483EXPORT_SYMBOL(ksize);
1484
928cec9c
AR
1485/* Tracepoints definitions. */
1486EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1487EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
928cec9c
AR
1488EXPORT_TRACEPOINT_SYMBOL(kfree);
1489EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
4f6923fb
HM
1490
1491int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1492{
1493 if (__should_failslab(s, gfpflags))
1494 return -ENOMEM;
1495 return 0;
1496}
1497ALLOW_ERROR_INJECTION(should_failslab, ERRNO);