Linux Audio

Check our new training course

Loading...
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   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>
  13#include <linux/cache.h>
  14#include <linux/compiler.h>
 
  15#include <linux/module.h>
  16#include <linux/cpu.h>
  17#include <linux/uaccess.h>
  18#include <linux/seq_file.h>
 
 
  19#include <linux/proc_fs.h>
 
 
 
  20#include <asm/cacheflush.h>
  21#include <asm/tlbflush.h>
  22#include <asm/page.h>
  23#include <linux/memcontrol.h>
 
 
 
 
  24
  25#define CREATE_TRACE_POINTS
  26#include <trace/events/kmem.h>
  27
  28#include "slab.h"
  29
  30enum slab_state slab_state;
  31LIST_HEAD(slab_caches);
  32DEFINE_MUTEX(slab_mutex);
  33struct kmem_cache *kmem_cache;
  34
  35#ifdef CONFIG_HARDENED_USERCOPY
  36bool usercopy_fallback __ro_after_init =
  37		IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK);
  38module_param(usercopy_fallback, bool, 0400);
  39MODULE_PARM_DESC(usercopy_fallback,
  40		"WARN instead of reject usercopy whitelist violations");
  41#endif
  42
  43static LIST_HEAD(slab_caches_to_rcu_destroy);
  44static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
  45static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
  46		    slab_caches_to_rcu_destroy_workfn);
  47
  48/*
  49 * Set of flags that will prevent slab merging
  50 */
  51#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  52		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
  53		SLAB_FAILSLAB | SLAB_KASAN)
  54
  55#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
  56			 SLAB_ACCOUNT)
  57
  58/*
  59 * Merge control. If this is set then no merging of slab caches will occur.
  60 */
  61static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
  62
  63static int __init setup_slab_nomerge(char *str)
  64{
  65	slab_nomerge = true;
  66	return 1;
  67}
  68
  69#ifdef CONFIG_SLUB
 
 
 
 
 
  70__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
  71#endif
  72
  73__setup("slab_nomerge", setup_slab_nomerge);
 
  74
  75/*
  76 * Determine the size of a slab object
  77 */
  78unsigned int kmem_cache_size(struct kmem_cache *s)
  79{
  80	return s->object_size;
  81}
  82EXPORT_SYMBOL(kmem_cache_size);
  83
  84#ifdef CONFIG_DEBUG_VM
  85static int kmem_cache_sanity_check(const char *name, unsigned int size)
  86{
  87	if (!name || in_interrupt() || size < sizeof(void *) ||
  88		size > KMALLOC_MAX_SIZE) {
  89		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
  90		return -EINVAL;
  91	}
  92
  93	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
  94	return 0;
  95}
  96#else
  97static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
  98{
  99	return 0;
 100}
 101#endif
 102
 103void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
 104{
 105	size_t i;
 106
 107	for (i = 0; i < nr; i++) {
 108		if (s)
 109			kmem_cache_free(s, p[i]);
 110		else
 111			kfree(p[i]);
 112	}
 113}
 114
 115int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
 116								void **p)
 117{
 118	size_t i;
 119
 120	for (i = 0; i < nr; i++) {
 121		void *x = p[i] = kmem_cache_alloc(s, flags);
 122		if (!x) {
 123			__kmem_cache_free_bulk(s, i, p);
 124			return 0;
 125		}
 126	}
 127	return i;
 128}
 129
 130#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 131
 132LIST_HEAD(slab_root_caches);
 133
 134void slab_init_memcg_params(struct kmem_cache *s)
 135{
 136	s->memcg_params.root_cache = NULL;
 137	RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
 138	INIT_LIST_HEAD(&s->memcg_params.children);
 139}
 140
 141static int init_memcg_params(struct kmem_cache *s,
 142		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 143{
 144	struct memcg_cache_array *arr;
 145
 146	if (root_cache) {
 147		s->memcg_params.root_cache = root_cache;
 148		s->memcg_params.memcg = memcg;
 149		INIT_LIST_HEAD(&s->memcg_params.children_node);
 150		INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node);
 151		return 0;
 152	}
 153
 154	slab_init_memcg_params(s);
 155
 156	if (!memcg_nr_cache_ids)
 157		return 0;
 158
 159	arr = kvzalloc(sizeof(struct memcg_cache_array) +
 160		       memcg_nr_cache_ids * sizeof(void *),
 161		       GFP_KERNEL);
 162	if (!arr)
 163		return -ENOMEM;
 164
 165	RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
 166	return 0;
 167}
 168
 169static void destroy_memcg_params(struct kmem_cache *s)
 170{
 171	if (is_root_cache(s))
 172		kvfree(rcu_access_pointer(s->memcg_params.memcg_caches));
 173}
 174
 175static void free_memcg_params(struct rcu_head *rcu)
 176{
 177	struct memcg_cache_array *old;
 178
 179	old = container_of(rcu, struct memcg_cache_array, rcu);
 180	kvfree(old);
 181}
 182
 183static int update_memcg_params(struct kmem_cache *s, int new_array_size)
 184{
 185	struct memcg_cache_array *old, *new;
 186
 187	new = kvzalloc(sizeof(struct memcg_cache_array) +
 188		       new_array_size * sizeof(void *), GFP_KERNEL);
 189	if (!new)
 190		return -ENOMEM;
 191
 192	old = rcu_dereference_protected(s->memcg_params.memcg_caches,
 193					lockdep_is_held(&slab_mutex));
 194	if (old)
 195		memcpy(new->entries, old->entries,
 196		       memcg_nr_cache_ids * sizeof(void *));
 197
 198	rcu_assign_pointer(s->memcg_params.memcg_caches, new);
 199	if (old)
 200		call_rcu(&old->rcu, free_memcg_params);
 201	return 0;
 202}
 203
 204int memcg_update_all_caches(int num_memcgs)
 205{
 206	struct kmem_cache *s;
 207	int ret = 0;
 208
 209	mutex_lock(&slab_mutex);
 210	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
 211		ret = update_memcg_params(s, num_memcgs);
 212		/*
 213		 * Instead of freeing the memory, we'll just leave the caches
 214		 * up to this point in an updated state.
 215		 */
 216		if (ret)
 217			break;
 218	}
 219	mutex_unlock(&slab_mutex);
 220	return ret;
 221}
 222
 223void memcg_link_cache(struct kmem_cache *s)
 224{
 225	if (is_root_cache(s)) {
 226		list_add(&s->root_caches_node, &slab_root_caches);
 227	} else {
 228		list_add(&s->memcg_params.children_node,
 229			 &s->memcg_params.root_cache->memcg_params.children);
 230		list_add(&s->memcg_params.kmem_caches_node,
 231			 &s->memcg_params.memcg->kmem_caches);
 232	}
 233}
 234
 235static void memcg_unlink_cache(struct kmem_cache *s)
 236{
 237	if (is_root_cache(s)) {
 238		list_del(&s->root_caches_node);
 239	} else {
 240		list_del(&s->memcg_params.children_node);
 241		list_del(&s->memcg_params.kmem_caches_node);
 242	}
 243}
 244#else
 245static inline int init_memcg_params(struct kmem_cache *s,
 246		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 247{
 248	return 0;
 249}
 250
 251static inline void destroy_memcg_params(struct kmem_cache *s)
 252{
 253}
 254
 255static inline void memcg_unlink_cache(struct kmem_cache *s)
 256{
 257}
 258#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 259
 260/*
 261 * Figure out what the alignment of the objects will be given a set of
 262 * flags, a user specified alignment and the size of the objects.
 263 */
 264static unsigned int calculate_alignment(slab_flags_t flags,
 265		unsigned int align, unsigned int size)
 266{
 267	/*
 268	 * If the user wants hardware cache aligned objects then follow that
 269	 * suggestion if the object is sufficiently large.
 270	 *
 271	 * The hardware cache alignment cannot override the specified
 272	 * alignment though. If that is greater then use it.
 273	 */
 274	if (flags & SLAB_HWCACHE_ALIGN) {
 275		unsigned int ralign;
 276
 277		ralign = cache_line_size();
 278		while (size <= ralign / 2)
 279			ralign /= 2;
 280		align = max(align, ralign);
 281	}
 282
 283	if (align < ARCH_SLAB_MINALIGN)
 284		align = ARCH_SLAB_MINALIGN;
 285
 286	return ALIGN(align, sizeof(void *));
 287}
 288
 289/*
 290 * Find a mergeable slab cache
 291 */
 292int slab_unmergeable(struct kmem_cache *s)
 293{
 294	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 295		return 1;
 296
 297	if (!is_root_cache(s))
 298		return 1;
 299
 300	if (s->ctor)
 301		return 1;
 302
 
 303	if (s->usersize)
 304		return 1;
 
 305
 306	/*
 307	 * We may have set a slab to be unmergeable during bootstrap.
 308	 */
 309	if (s->refcount < 0)
 310		return 1;
 311
 312	return 0;
 313}
 314
 315struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
 316		slab_flags_t flags, const char *name, void (*ctor)(void *))
 317{
 318	struct kmem_cache *s;
 319
 320	if (slab_nomerge)
 321		return NULL;
 322
 323	if (ctor)
 324		return NULL;
 325
 326	size = ALIGN(size, sizeof(void *));
 327	align = calculate_alignment(flags, align, size);
 328	size = ALIGN(size, align);
 329	flags = kmem_cache_flags(size, flags, name, NULL);
 330
 331	if (flags & SLAB_NEVER_MERGE)
 332		return NULL;
 333
 334	list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) {
 
 
 
 
 335		if (slab_unmergeable(s))
 336			continue;
 337
 338		if (size > s->size)
 339			continue;
 340
 341		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 342			continue;
 343		/*
 344		 * Check if alignment is compatible.
 345		 * Courtesy of Adrian Drzewiecki
 346		 */
 347		if ((s->size & ~(align - 1)) != s->size)
 348			continue;
 349
 350		if (s->size - size >= sizeof(void *))
 351			continue;
 352
 353		if (IS_ENABLED(CONFIG_SLAB) && align &&
 354			(align > s->align || s->align % align))
 355			continue;
 356
 357		return s;
 358	}
 359	return NULL;
 360}
 361
 362static struct kmem_cache *create_cache(const char *name,
 363		unsigned int object_size, unsigned int align,
 364		slab_flags_t flags, unsigned int useroffset,
 365		unsigned int usersize, void (*ctor)(void *),
 366		struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 367{
 368	struct kmem_cache *s;
 369	int err;
 370
 371	if (WARN_ON(useroffset + usersize > object_size))
 372		useroffset = usersize = 0;
 
 
 
 
 
 373
 374	err = -ENOMEM;
 375	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 376	if (!s)
 377		goto out;
 378
 379	s->name = name;
 380	s->size = s->object_size = object_size;
 381	s->align = align;
 382	s->ctor = ctor;
 383	s->useroffset = useroffset;
 384	s->usersize = usersize;
 385
 386	err = init_memcg_params(s, memcg, root_cache);
 387	if (err)
 388		goto out_free_cache;
 389
 390	err = __kmem_cache_create(s, flags);
 391	if (err)
 392		goto out_free_cache;
 393
 394	s->refcount = 1;
 395	list_add(&s->list, &slab_caches);
 396	memcg_link_cache(s);
 397out:
 398	if (err)
 399		return ERR_PTR(err);
 400	return s;
 401
 402out_free_cache:
 403	destroy_memcg_params(s);
 404	kmem_cache_free(kmem_cache, s);
 405	goto out;
 
 406}
 407
 408/*
 409 * kmem_cache_create_usercopy - Create a cache.
 410 * @name: A string which is used in /proc/slabinfo to identify this cache.
 411 * @size: The size of objects to be created in this cache.
 412 * @align: The required alignment for the objects.
 413 * @flags: SLAB flags
 414 * @useroffset: Usercopy region offset
 415 * @usersize: Usercopy region size
 416 * @ctor: A constructor for the objects.
 417 *
 418 * Returns a ptr to the cache on success, NULL on failure.
 419 * Cannot be called within a interrupt, but can be interrupted.
 420 * The @ctor is run when new pages are allocated by the cache.
 421 *
 422 * The flags are
 423 *
 424 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 425 * to catch references to uninitialised memory.
 426 *
 427 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 428 * for buffer overruns.
 429 *
 430 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 431 * cacheline.  This can be beneficial if you're counting cycles as closely
 432 * as davem.
 
 433 */
 434struct kmem_cache *
 435kmem_cache_create_usercopy(const char *name,
 436		  unsigned int size, unsigned int align,
 437		  slab_flags_t flags,
 438		  unsigned int useroffset, unsigned int usersize,
 439		  void (*ctor)(void *))
 440{
 441	struct kmem_cache *s = NULL;
 442	const char *cache_name;
 443	int err;
 444
 445	get_online_cpus();
 446	get_online_mems();
 447	memcg_get_cache_ids();
 
 
 
 
 
 
 
 
 
 
 448
 449	mutex_lock(&slab_mutex);
 450
 451	err = kmem_cache_sanity_check(name, size);
 452	if (err) {
 453		goto out_unlock;
 454	}
 455
 456	/* Refuse requests with allocator specific flags */
 457	if (flags & ~SLAB_FLAGS_PERMITTED) {
 458		err = -EINVAL;
 459		goto out_unlock;
 460	}
 461
 462	/*
 463	 * Some allocators will constraint the set of valid flags to a subset
 464	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
 465	 * case, and we'll just provide them with a sanitized version of the
 466	 * passed flags.
 467	 */
 468	flags &= CACHE_CREATE_MASK;
 469
 470	/* Fail closed on bad usersize of useroffset values. */
 471	if (WARN_ON(!usersize && useroffset) ||
 472	    WARN_ON(size < usersize || size - usersize < useroffset))
 473		usersize = useroffset = 0;
 474
 475	if (!usersize)
 476		s = __kmem_cache_alias(name, size, align, flags, ctor);
 
 
 
 477	if (s)
 478		goto out_unlock;
 479
 480	cache_name = kstrdup_const(name, GFP_KERNEL);
 481	if (!cache_name) {
 482		err = -ENOMEM;
 483		goto out_unlock;
 484	}
 485
 486	s = create_cache(cache_name, size,
 487			 calculate_alignment(flags, align, size),
 488			 flags, useroffset, usersize, ctor, NULL, NULL);
 489	if (IS_ERR(s)) {
 490		err = PTR_ERR(s);
 491		kfree_const(cache_name);
 492	}
 493
 494out_unlock:
 495	mutex_unlock(&slab_mutex);
 496
 497	memcg_put_cache_ids();
 498	put_online_mems();
 499	put_online_cpus();
 500
 501	if (err) {
 502		if (flags & SLAB_PANIC)
 503			panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 504				name, err);
 505		else {
 506			pr_warn("kmem_cache_create(%s) failed with error %d\n",
 507				name, err);
 508			dump_stack();
 509		}
 510		return NULL;
 511	}
 512	return s;
 513}
 514EXPORT_SYMBOL(kmem_cache_create_usercopy);
 515
 516struct kmem_cache *
 517kmem_cache_create(const char *name, unsigned int size, unsigned int align,
 518		slab_flags_t flags, void (*ctor)(void *))
 519{
 520	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
 521					  ctor);
 522}
 523EXPORT_SYMBOL(kmem_cache_create);
 524
 525static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
 526{
 527	LIST_HEAD(to_destroy);
 528	struct kmem_cache *s, *s2;
 529
 530	/*
 531	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
 532	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
 533	 * through RCU and and the associated kmem_cache are dereferenced
 534	 * while freeing the pages, so the kmem_caches should be freed only
 535	 * after the pending RCU operations are finished.  As rcu_barrier()
 536	 * is a pretty slow operation, we batch all pending destructions
 537	 * asynchronously.
 538	 */
 539	mutex_lock(&slab_mutex);
 540	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
 541	mutex_unlock(&slab_mutex);
 542
 543	if (list_empty(&to_destroy))
 544		return;
 545
 546	rcu_barrier();
 547
 548	list_for_each_entry_safe(s, s2, &to_destroy, list) {
 549#ifdef SLAB_SUPPORTS_SYSFS
 550		sysfs_slab_release(s);
 551#else
 552		slab_kmem_cache_release(s);
 553#endif
 554	}
 555}
 556
 557static int shutdown_cache(struct kmem_cache *s)
 558{
 559	/* free asan quarantined objects */
 560	kasan_cache_shutdown(s);
 561
 562	if (__kmem_cache_shutdown(s) != 0)
 563		return -EBUSY;
 564
 565	memcg_unlink_cache(s);
 566	list_del(&s->list);
 567
 568	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
 569		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
 570		schedule_work(&slab_caches_to_rcu_destroy_work);
 571	} else {
 572#ifdef SLAB_SUPPORTS_SYSFS
 573		sysfs_slab_release(s);
 574#else
 575		slab_kmem_cache_release(s);
 576#endif
 577	}
 578
 579	return 0;
 580}
 581
 582#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 583/*
 584 * memcg_create_kmem_cache - Create a cache for a memory cgroup.
 585 * @memcg: The memory cgroup the new cache is for.
 586 * @root_cache: The parent of the new cache.
 587 *
 588 * This function attempts to create a kmem cache that will serve allocation
 589 * requests going from @memcg to @root_cache. The new cache inherits properties
 590 * from its parent.
 
 
 
 
 
 
 591 */
 592void memcg_create_kmem_cache(struct mem_cgroup *memcg,
 593			     struct kmem_cache *root_cache)
 594{
 595	static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
 596	struct cgroup_subsys_state *css = &memcg->css;
 597	struct memcg_cache_array *arr;
 598	struct kmem_cache *s = NULL;
 599	char *cache_name;
 600	int idx;
 601
 602	get_online_cpus();
 603	get_online_mems();
 604
 605	mutex_lock(&slab_mutex);
 606
 607	/*
 608	 * The memory cgroup could have been offlined while the cache
 609	 * creation work was pending.
 610	 */
 611	if (memcg->kmem_state != KMEM_ONLINE)
 612		goto out_unlock;
 613
 614	idx = memcg_cache_id(memcg);
 615	arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
 616					lockdep_is_held(&slab_mutex));
 617
 618	/*
 619	 * Since per-memcg caches are created asynchronously on first
 620	 * allocation (see memcg_kmem_get_cache()), several threads can try to
 621	 * create the same cache, but only one of them may succeed.
 622	 */
 623	if (arr->entries[idx])
 624		goto out_unlock;
 625
 626	cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
 627	cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
 628			       css->serial_nr, memcg_name_buf);
 629	if (!cache_name)
 630		goto out_unlock;
 631
 632	s = create_cache(cache_name, root_cache->object_size,
 633			 root_cache->align,
 634			 root_cache->flags & CACHE_CREATE_MASK,
 635			 root_cache->useroffset, root_cache->usersize,
 636			 root_cache->ctor, memcg, root_cache);
 637	/*
 638	 * If we could not create a memcg cache, do not complain, because
 639	 * that's not critical at all as we can always proceed with the root
 640	 * cache.
 641	 */
 642	if (IS_ERR(s)) {
 643		kfree(cache_name);
 644		goto out_unlock;
 645	}
 646
 647	/*
 648	 * Since readers won't lock (see cache_from_memcg_idx()), we need a
 649	 * barrier here to ensure nobody will see the kmem_cache partially
 650	 * initialized.
 651	 */
 652	smp_wmb();
 653	arr->entries[idx] = s;
 654
 655out_unlock:
 656	mutex_unlock(&slab_mutex);
 657
 658	put_online_mems();
 659	put_online_cpus();
 660}
 661
 662static void kmemcg_deactivate_workfn(struct work_struct *work)
 663{
 664	struct kmem_cache *s = container_of(work, struct kmem_cache,
 665					    memcg_params.deact_work);
 666
 667	get_online_cpus();
 668	get_online_mems();
 669
 670	mutex_lock(&slab_mutex);
 671
 672	s->memcg_params.deact_fn(s);
 673
 674	mutex_unlock(&slab_mutex);
 675
 676	put_online_mems();
 677	put_online_cpus();
 678
 679	/* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */
 680	css_put(&s->memcg_params.memcg->css);
 681}
 682
 683static void kmemcg_deactivate_rcufn(struct rcu_head *head)
 684{
 685	struct kmem_cache *s = container_of(head, struct kmem_cache,
 686					    memcg_params.deact_rcu_head);
 687
 688	/*
 689	 * We need to grab blocking locks.  Bounce to ->deact_work.  The
 690	 * work item shares the space with the RCU head and can't be
 691	 * initialized eariler.
 692	 */
 693	INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn);
 694	queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work);
 695}
 696
 697/**
 698 * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a
 699 *					   sched RCU grace period
 700 * @s: target kmem_cache
 701 * @deact_fn: deactivation function to call
 702 *
 703 * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex
 704 * held after a sched RCU grace period.  The slab is guaranteed to stay
 705 * alive until @deact_fn is finished.  This is to be used from
 706 * __kmemcg_cache_deactivate().
 707 */
 708void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
 709					   void (*deact_fn)(struct kmem_cache *))
 710{
 711	if (WARN_ON_ONCE(is_root_cache(s)) ||
 712	    WARN_ON_ONCE(s->memcg_params.deact_fn))
 713		return;
 714
 715	/* pin memcg so that @s doesn't get destroyed in the middle */
 716	css_get(&s->memcg_params.memcg->css);
 717
 718	s->memcg_params.deact_fn = deact_fn;
 719	call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn);
 720}
 721
 722void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
 723{
 724	int idx;
 725	struct memcg_cache_array *arr;
 726	struct kmem_cache *s, *c;
 727
 728	idx = memcg_cache_id(memcg);
 729
 730	get_online_cpus();
 731	get_online_mems();
 
 
 732
 733	mutex_lock(&slab_mutex);
 734	list_for_each_entry(s, &slab_root_caches, root_caches_node) {
 735		arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 736						lockdep_is_held(&slab_mutex));
 737		c = arr->entries[idx];
 738		if (!c)
 739			continue;
 740
 741		__kmemcg_cache_deactivate(c);
 742		arr->entries[idx] = NULL;
 743	}
 744	mutex_unlock(&slab_mutex);
 745
 746	put_online_mems();
 747	put_online_cpus();
 748}
 749
 750void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
 751{
 752	struct kmem_cache *s, *s2;
 753
 754	get_online_cpus();
 755	get_online_mems();
 756
 757	mutex_lock(&slab_mutex);
 758	list_for_each_entry_safe(s, s2, &memcg->kmem_caches,
 759				 memcg_params.kmem_caches_node) {
 760		/*
 761		 * The cgroup is about to be freed and therefore has no charges
 762		 * left. Hence, all its caches must be empty by now.
 763		 */
 764		BUG_ON(shutdown_cache(s));
 765	}
 766	mutex_unlock(&slab_mutex);
 767
 768	put_online_mems();
 769	put_online_cpus();
 770}
 771
 772static int shutdown_memcg_caches(struct kmem_cache *s)
 773{
 774	struct memcg_cache_array *arr;
 775	struct kmem_cache *c, *c2;
 776	LIST_HEAD(busy);
 777	int i;
 778
 779	BUG_ON(!is_root_cache(s));
 
 
 
 
 
 
 
 
 
 
 780
 781	/*
 782	 * First, shutdown active caches, i.e. caches that belong to online
 783	 * memory cgroups.
 784	 */
 785	arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 786					lockdep_is_held(&slab_mutex));
 787	for_each_memcg_cache_index(i) {
 788		c = arr->entries[i];
 789		if (!c)
 790			continue;
 791		if (shutdown_cache(c))
 792			/*
 793			 * The cache still has objects. Move it to a temporary
 794			 * list so as not to try to destroy it for a second
 795			 * time while iterating over inactive caches below.
 796			 */
 797			list_move(&c->memcg_params.children_node, &busy);
 798		else
 799			/*
 800			 * The cache is empty and will be destroyed soon. Clear
 801			 * the pointer to it in the memcg_caches array so that
 802			 * it will never be accessed even if the root cache
 803			 * stays alive.
 804			 */
 805			arr->entries[i] = NULL;
 806	}
 807
 808	/*
 809	 * Second, shutdown all caches left from memory cgroups that are now
 810	 * offline.
 811	 */
 812	list_for_each_entry_safe(c, c2, &s->memcg_params.children,
 813				 memcg_params.children_node)
 814		shutdown_cache(c);
 815
 816	list_splice(&busy, &s->memcg_params.children);
 
 
 
 817
 818	/*
 819	 * A cache being destroyed must be empty. In particular, this means
 820	 * that all per memcg caches attached to it must be empty too.
 821	 */
 822	if (!list_empty(&s->memcg_params.children))
 823		return -EBUSY;
 824	return 0;
 825}
 826#else
 827static inline int shutdown_memcg_caches(struct kmem_cache *s)
 
 
 
 
 
 
 
 828{
 829	return 0;
 
 
 
 
 830}
 831#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 832
 833void slab_kmem_cache_release(struct kmem_cache *s)
 834{
 835	__kmem_cache_release(s);
 836	destroy_memcg_params(s);
 837	kfree_const(s->name);
 838	kmem_cache_free(kmem_cache, s);
 839}
 840
 841void kmem_cache_destroy(struct kmem_cache *s)
 842{
 843	int err;
 844
 845	if (unlikely(!s))
 846		return;
 847
 848	get_online_cpus();
 849	get_online_mems();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 850
 
 851	mutex_lock(&slab_mutex);
 852
 853	s->refcount--;
 854	if (s->refcount)
 855		goto out_unlock;
 
 
 
 856
 857	err = shutdown_memcg_caches(s);
 858	if (!err)
 859		err = shutdown_cache(s);
 
 
 
 
 
 
 860
 861	if (err) {
 862		pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
 863		       s->name);
 864		dump_stack();
 865	}
 866out_unlock:
 867	mutex_unlock(&slab_mutex);
 
 
 
 
 
 868
 869	put_online_mems();
 870	put_online_cpus();
 
 
 
 
 
 871}
 872EXPORT_SYMBOL(kmem_cache_destroy);
 873
 874/**
 875 * kmem_cache_shrink - Shrink a cache.
 876 * @cachep: The cache to shrink.
 877 *
 878 * Releases as many slabs as possible for a cache.
 879 * To help debugging, a zero exit status indicates all slabs were released.
 
 
 880 */
 881int kmem_cache_shrink(struct kmem_cache *cachep)
 882{
 883	int ret;
 884
 885	get_online_cpus();
 886	get_online_mems();
 887	kasan_cache_shrink(cachep);
 888	ret = __kmem_cache_shrink(cachep);
 889	put_online_mems();
 890	put_online_cpus();
 891	return ret;
 892}
 893EXPORT_SYMBOL(kmem_cache_shrink);
 894
 895bool slab_is_available(void)
 896{
 897	return slab_state >= UP;
 898}
 899
 900#ifndef CONFIG_SLOB
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 901/* Create a cache during boot when no slab services are available yet */
 902void __init create_boot_cache(struct kmem_cache *s, const char *name,
 903		unsigned int size, slab_flags_t flags,
 904		unsigned int useroffset, unsigned int usersize)
 905{
 906	int err;
 
 
 907
 908	s->name = name;
 909	s->size = s->object_size = size;
 910	s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 911	s->useroffset = useroffset;
 912	s->usersize = usersize;
 
 
 
 913
 914	slab_init_memcg_params(s);
 
 
 
 915
 916	err = __kmem_cache_create(s, flags);
 917
 918	if (err)
 919		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
 920					name, size, err);
 921
 922	s->refcount = -1;	/* Exempt from merging for now */
 923}
 924
 925struct kmem_cache *__init create_kmalloc_cache(const char *name,
 926		unsigned int size, slab_flags_t flags,
 927		unsigned int useroffset, unsigned int usersize)
 928{
 929	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 930
 931	if (!s)
 932		panic("Out of memory when creating slab %s\n", name);
 933
 934	create_boot_cache(s, name, size, flags, useroffset, usersize);
 935	list_add(&s->list, &slab_caches);
 936	memcg_link_cache(s);
 937	s->refcount = 1;
 938	return s;
 939}
 940
 941struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __ro_after_init;
 
 942EXPORT_SYMBOL(kmalloc_caches);
 943
 944#ifdef CONFIG_ZONE_DMA
 945struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1] __ro_after_init;
 946EXPORT_SYMBOL(kmalloc_dma_caches);
 947#endif
 948
 949/*
 950 * Conversion table for small slabs sizes / 8 to the index in the
 951 * kmalloc array. This is necessary for slabs < 192 since we have non power
 952 * of two cache sizes there. The size of larger slabs can be determined using
 953 * fls.
 954 */
 955static u8 size_index[24] __ro_after_init = {
 956	3,	/* 8 */
 957	4,	/* 16 */
 958	5,	/* 24 */
 959	5,	/* 32 */
 960	6,	/* 40 */
 961	6,	/* 48 */
 962	6,	/* 56 */
 963	6,	/* 64 */
 964	1,	/* 72 */
 965	1,	/* 80 */
 966	1,	/* 88 */
 967	1,	/* 96 */
 968	7,	/* 104 */
 969	7,	/* 112 */
 970	7,	/* 120 */
 971	7,	/* 128 */
 972	2,	/* 136 */
 973	2,	/* 144 */
 974	2,	/* 152 */
 975	2,	/* 160 */
 976	2,	/* 168 */
 977	2,	/* 176 */
 978	2,	/* 184 */
 979	2	/* 192 */
 980};
 981
 982static inline unsigned int size_index_elem(unsigned int bytes)
 983{
 984	return (bytes - 1) / 8;
 985}
 986
 987/*
 988 * Find the kmem_cache structure that serves a given size of
 989 * allocation
 990 */
 991struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 992{
 993	unsigned int index;
 994
 995	if (unlikely(size > KMALLOC_MAX_SIZE)) {
 996		WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 997		return NULL;
 
 998	}
 999
1000	if (size <= 192) {
1001		if (!size)
1002			return ZERO_SIZE_PTR;
 
 
 
 
 
 
1003
1004		index = size_index[size_index_elem(size)];
1005	} else
1006		index = fls(size - 1);
1007
1008#ifdef CONFIG_ZONE_DMA
1009	if (unlikely((flags & GFP_DMA)))
1010		return kmalloc_dma_caches[index];
 
 
1011
 
 
 
 
1012#endif
1013	return kmalloc_caches[index];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1014}
1015
1016/*
1017 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
1018 * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
1019 * kmalloc-67108864.
1020 */
1021const struct kmalloc_info_struct kmalloc_info[] __initconst = {
1022	{NULL,                      0},		{"kmalloc-96",             96},
1023	{"kmalloc-192",           192},		{"kmalloc-8",               8},
1024	{"kmalloc-16",             16},		{"kmalloc-32",             32},
1025	{"kmalloc-64",             64},		{"kmalloc-128",           128},
1026	{"kmalloc-256",           256},		{"kmalloc-512",           512},
1027	{"kmalloc-1024",         1024},		{"kmalloc-2048",         2048},
1028	{"kmalloc-4096",         4096},		{"kmalloc-8192",         8192},
1029	{"kmalloc-16384",       16384},		{"kmalloc-32768",       32768},
1030	{"kmalloc-65536",       65536},		{"kmalloc-131072",     131072},
1031	{"kmalloc-262144",     262144},		{"kmalloc-524288",     524288},
1032	{"kmalloc-1048576",   1048576},		{"kmalloc-2097152",   2097152},
1033	{"kmalloc-4194304",   4194304},		{"kmalloc-8388608",   8388608},
1034	{"kmalloc-16777216", 16777216},		{"kmalloc-33554432", 33554432},
1035	{"kmalloc-67108864", 67108864}
 
 
 
 
 
 
 
 
1036};
1037
1038/*
1039 * Patch up the size_index table if we have strange large alignment
1040 * requirements for the kmalloc array. This is only the case for
1041 * MIPS it seems. The standard arches will not generate any code here.
1042 *
1043 * Largest permitted alignment is 256 bytes due to the way we
1044 * handle the index determination for the smaller caches.
1045 *
1046 * Make sure that nothing crazy happens if someone starts tinkering
1047 * around with ARCH_KMALLOC_MINALIGN
1048 */
1049void __init setup_kmalloc_cache_index_table(void)
1050{
1051	unsigned int i;
1052
1053	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
1054		(KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
1055
1056	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
1057		unsigned int elem = size_index_elem(i);
1058
1059		if (elem >= ARRAY_SIZE(size_index))
1060			break;
1061		size_index[elem] = KMALLOC_SHIFT_LOW;
1062	}
1063
1064	if (KMALLOC_MIN_SIZE >= 64) {
1065		/*
1066		 * The 96 byte size cache is not used if the alignment
1067		 * is 64 byte.
1068		 */
1069		for (i = 64 + 8; i <= 96; i += 8)
1070			size_index[size_index_elem(i)] = 7;
1071
1072	}
1073
1074	if (KMALLOC_MIN_SIZE >= 128) {
1075		/*
1076		 * The 192 byte sized cache is not used if the alignment
1077		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
1078		 * instead.
1079		 */
1080		for (i = 128 + 8; i <= 192; i += 8)
1081			size_index[size_index_elem(i)] = 8;
1082	}
1083}
1084
1085static void __init new_kmalloc_cache(int idx, slab_flags_t flags)
1086{
1087	kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
1088					kmalloc_info[idx].size, flags, 0,
1089					kmalloc_info[idx].size);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1090}
1091
1092/*
1093 * Create the kmalloc array. Some of the regular kmalloc arrays
1094 * may already have been created because they were needed to
1095 * enable allocations for slab creation.
1096 */
1097void __init create_kmalloc_caches(slab_flags_t flags)
1098{
1099	int i;
 
1100
1101	for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
1102		if (!kmalloc_caches[i])
1103			new_kmalloc_cache(i, flags);
1104
1105		/*
1106		 * Caches that are not of the two-to-the-power-of size.
1107		 * These have to be created immediately after the
1108		 * earlier power of two caches
1109		 */
1110		if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
1111			new_kmalloc_cache(1, flags);
1112		if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
1113			new_kmalloc_cache(2, flags);
1114	}
 
 
 
1115
1116	/* Kmalloc array is now usable */
1117	slab_state = UP;
1118
1119#ifdef CONFIG_ZONE_DMA
1120	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
1121		struct kmem_cache *s = kmalloc_caches[i];
1122
1123		if (s) {
1124			unsigned int size = kmalloc_size(i);
1125			char *n = kasprintf(GFP_NOWAIT,
1126				 "dma-kmalloc-%u", size);
1127
1128			BUG_ON(!n);
1129			kmalloc_dma_caches[i] = create_kmalloc_cache(n,
1130				size, SLAB_CACHE_DMA | flags, 0, 0);
1131		}
1132	}
1133#endif
1134}
1135#endif /* !CONFIG_SLOB */
1136
1137/*
1138 * To avoid unnecessary overhead, we pass through large allocation requests
1139 * directly to the page allocator. We use __GFP_COMP, because we will need to
1140 * know the allocation order to free the pages properly in kfree.
 
 
 
 
 
 
 
1141 */
1142void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1143{
1144	void *ret;
1145	struct page *page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1146
1147	flags |= __GFP_COMP;
1148	page = alloc_pages(flags, order);
1149	ret = page ? page_address(page) : NULL;
1150	kmemleak_alloc(ret, size, 1, flags);
1151	kasan_kmalloc_large(ret, size, flags);
1152	return ret;
1153}
1154EXPORT_SYMBOL(kmalloc_order);
1155
1156#ifdef CONFIG_TRACING
1157void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1158{
1159	void *ret = kmalloc_order(size, flags, order);
1160	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1161	return ret;
 
 
 
 
 
1162}
1163EXPORT_SYMBOL(kmalloc_order_trace);
1164#endif
1165
1166#ifdef CONFIG_SLAB_FREELIST_RANDOM
1167/* Randomize a generic freelist */
1168static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1169			       unsigned int count)
1170{
1171	unsigned int rand;
1172	unsigned int i;
1173
1174	for (i = 0; i < count; i++)
1175		list[i] = i;
1176
1177	/* Fisher-Yates shuffle */
1178	for (i = count - 1; i > 0; i--) {
1179		rand = prandom_u32_state(state);
1180		rand %= (i + 1);
1181		swap(list[i], list[rand]);
1182	}
1183}
1184
1185/* Create a random sequence per cache */
1186int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1187				    gfp_t gfp)
1188{
1189	struct rnd_state state;
1190
1191	if (count < 2 || cachep->random_seq)
1192		return 0;
1193
1194	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1195	if (!cachep->random_seq)
1196		return -ENOMEM;
1197
1198	/* Get best entropy at this stage of boot */
1199	prandom_seed_state(&state, get_random_long());
1200
1201	freelist_randomize(&state, cachep->random_seq, count);
1202	return 0;
1203}
1204
1205/* Destroy the per-cache random freelist sequence */
1206void cache_random_seq_destroy(struct kmem_cache *cachep)
1207{
1208	kfree(cachep->random_seq);
1209	cachep->random_seq = NULL;
1210}
1211#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1212
1213#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1214#ifdef CONFIG_SLAB
1215#define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1216#else
1217#define SLABINFO_RIGHTS S_IRUSR
1218#endif
1219
1220static void print_slabinfo_header(struct seq_file *m)
1221{
1222	/*
1223	 * Output format version, so at least we can change it
1224	 * without _too_ many complaints.
1225	 */
1226#ifdef CONFIG_DEBUG_SLAB
1227	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1228#else
1229	seq_puts(m, "slabinfo - version: 2.1\n");
1230#endif
1231	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1232	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1233	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1234#ifdef CONFIG_DEBUG_SLAB
1235	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1236	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1237#endif
1238	seq_putc(m, '\n');
1239}
1240
1241void *slab_start(struct seq_file *m, loff_t *pos)
1242{
1243	mutex_lock(&slab_mutex);
1244	return seq_list_start(&slab_root_caches, *pos);
1245}
1246
1247void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1248{
1249	return seq_list_next(p, &slab_root_caches, pos);
1250}
1251
1252void slab_stop(struct seq_file *m, void *p)
1253{
1254	mutex_unlock(&slab_mutex);
1255}
1256
1257static void
1258memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1259{
1260	struct kmem_cache *c;
1261	struct slabinfo sinfo;
1262
1263	if (!is_root_cache(s))
1264		return;
1265
1266	for_each_memcg_cache(c, s) {
1267		memset(&sinfo, 0, sizeof(sinfo));
1268		get_slabinfo(c, &sinfo);
1269
1270		info->active_slabs += sinfo.active_slabs;
1271		info->num_slabs += sinfo.num_slabs;
1272		info->shared_avail += sinfo.shared_avail;
1273		info->active_objs += sinfo.active_objs;
1274		info->num_objs += sinfo.num_objs;
1275	}
1276}
1277
1278static void cache_show(struct kmem_cache *s, struct seq_file *m)
1279{
1280	struct slabinfo sinfo;
1281
1282	memset(&sinfo, 0, sizeof(sinfo));
1283	get_slabinfo(s, &sinfo);
1284
1285	memcg_accumulate_slabinfo(s, &sinfo);
1286
1287	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1288		   cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1289		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1290
1291	seq_printf(m, " : tunables %4u %4u %4u",
1292		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1293	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1294		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1295	slabinfo_show_stats(m, s);
1296	seq_putc(m, '\n');
1297}
1298
1299static int slab_show(struct seq_file *m, void *p)
1300{
1301	struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node);
1302
1303	if (p == slab_root_caches.next)
1304		print_slabinfo_header(m);
1305	cache_show(s, m);
1306	return 0;
1307}
1308
1309void dump_unreclaimable_slab(void)
1310{
1311	struct kmem_cache *s, *s2;
1312	struct slabinfo sinfo;
1313
1314	/*
1315	 * Here acquiring slab_mutex is risky since we don't prefer to get
1316	 * sleep in oom path. But, without mutex hold, it may introduce a
1317	 * risk of crash.
1318	 * Use mutex_trylock to protect the list traverse, dump nothing
1319	 * without acquiring the mutex.
1320	 */
1321	if (!mutex_trylock(&slab_mutex)) {
1322		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1323		return;
1324	}
1325
1326	pr_info("Unreclaimable slab info:\n");
1327	pr_info("Name                      Used          Total\n");
1328
1329	list_for_each_entry_safe(s, s2, &slab_caches, list) {
1330		if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT))
1331			continue;
1332
1333		get_slabinfo(s, &sinfo);
1334
1335		if (sinfo.num_objs > 0)
1336			pr_info("%-17s %10luKB %10luKB\n", cache_name(s),
1337				(sinfo.active_objs * s->size) / 1024,
1338				(sinfo.num_objs * s->size) / 1024);
1339	}
1340	mutex_unlock(&slab_mutex);
1341}
1342
1343#if defined(CONFIG_MEMCG)
1344void *memcg_slab_start(struct seq_file *m, loff_t *pos)
1345{
1346	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1347
1348	mutex_lock(&slab_mutex);
1349	return seq_list_start(&memcg->kmem_caches, *pos);
1350}
1351
1352void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos)
1353{
1354	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1355
1356	return seq_list_next(p, &memcg->kmem_caches, pos);
1357}
1358
1359void memcg_slab_stop(struct seq_file *m, void *p)
1360{
1361	mutex_unlock(&slab_mutex);
1362}
1363
1364int memcg_slab_show(struct seq_file *m, void *p)
1365{
1366	struct kmem_cache *s = list_entry(p, struct kmem_cache,
1367					  memcg_params.kmem_caches_node);
1368	struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1369
1370	if (p == memcg->kmem_caches.next)
1371		print_slabinfo_header(m);
1372	cache_show(s, m);
1373	return 0;
1374}
1375#endif
1376
1377/*
1378 * slabinfo_op - iterator that generates /proc/slabinfo
1379 *
1380 * Output layout:
1381 * cache-name
1382 * num-active-objs
1383 * total-objs
1384 * object size
1385 * num-active-slabs
1386 * total-slabs
1387 * num-pages-per-slab
1388 * + further values on SMP and with statistics enabled
1389 */
1390static const struct seq_operations slabinfo_op = {
1391	.start = slab_start,
1392	.next = slab_next,
1393	.stop = slab_stop,
1394	.show = slab_show,
1395};
1396
1397static int slabinfo_open(struct inode *inode, struct file *file)
1398{
1399	return seq_open(file, &slabinfo_op);
1400}
1401
1402static const struct file_operations proc_slabinfo_operations = {
1403	.open		= slabinfo_open,
1404	.read		= seq_read,
1405	.write          = slabinfo_write,
1406	.llseek		= seq_lseek,
1407	.release	= seq_release,
1408};
1409
1410static int __init slab_proc_init(void)
1411{
1412	proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1413						&proc_slabinfo_operations);
1414	return 0;
1415}
1416module_init(slab_proc_init);
1417#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1418
1419static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1420					   gfp_t flags)
1421{
1422	void *ret;
1423	size_t ks = 0;
1424
1425	if (p)
1426		ks = ksize(p);
1427
1428	if (ks >= new_size) {
1429		kasan_krealloc((void *)p, new_size, flags);
1430		return (void *)p;
1431	}
1432
1433	ret = kmalloc_track_caller(new_size, flags);
1434	if (ret && p)
1435		memcpy(ret, p, ks);
1436
1437	return ret;
1438}
1439
1440/**
1441 * __krealloc - like krealloc() but don't free @p.
1442 * @p: object to reallocate memory for.
1443 * @new_size: how many bytes of memory are required.
1444 * @flags: the type of memory to allocate.
1445 *
1446 * This function is like krealloc() except it never frees the originally
1447 * allocated buffer. Use this if you don't want to free the buffer immediately
1448 * like, for example, with RCU.
1449 */
1450void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1451{
1452	if (unlikely(!new_size))
1453		return ZERO_SIZE_PTR;
1454
1455	return __do_krealloc(p, new_size, flags);
1456
1457}
1458EXPORT_SYMBOL(__krealloc);
1459
1460/**
1461 * krealloc - reallocate memory. The contents will remain unchanged.
1462 * @p: object to reallocate memory for.
1463 * @new_size: how many bytes of memory are required.
1464 * @flags: the type of memory to allocate.
1465 *
1466 * The contents of the object pointed to are preserved up to the
1467 * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1468 * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1469 * %NULL pointer, the object pointed to is freed.
1470 */
1471void *krealloc(const void *p, size_t new_size, gfp_t flags)
1472{
1473	void *ret;
1474
1475	if (unlikely(!new_size)) {
1476		kfree(p);
1477		return ZERO_SIZE_PTR;
1478	}
1479
1480	ret = __do_krealloc(p, new_size, flags);
1481	if (ret && p != ret)
1482		kfree(p);
1483
1484	return ret;
1485}
1486EXPORT_SYMBOL(krealloc);
1487
1488/**
1489 * kzfree - like kfree but zero memory
1490 * @p: object to free memory of
1491 *
1492 * The memory of the object @p points to is zeroed before freed.
1493 * If @p is %NULL, kzfree() does nothing.
1494 *
1495 * Note: this function zeroes the whole allocated buffer which can be a good
1496 * deal bigger than the requested buffer size passed to kmalloc(). So be
1497 * careful when using this function in performance sensitive code.
1498 */
1499void kzfree(const void *p)
1500{
1501	size_t ks;
1502	void *mem = (void *)p;
1503
1504	if (unlikely(ZERO_OR_NULL_PTR(mem)))
1505		return;
1506	ks = ksize(mem);
1507	memset(mem, 0, ks);
 
 
 
1508	kfree(mem);
1509}
1510EXPORT_SYMBOL(kzfree);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1511
1512/* Tracepoints definitions. */
1513EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1514EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1515EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1516EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1517EXPORT_TRACEPOINT_SYMBOL(kfree);
1518EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1519
1520int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1521{
1522	if (__should_failslab(s, gfpflags))
1523		return -ENOMEM;
1524	return 0;
1525}
1526ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   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>
  13#include <linux/cache.h>
  14#include <linux/compiler.h>
  15#include <linux/kfence.h>
  16#include <linux/module.h>
  17#include <linux/cpu.h>
  18#include <linux/uaccess.h>
  19#include <linux/seq_file.h>
  20#include <linux/dma-mapping.h>
  21#include <linux/swiotlb.h>
  22#include <linux/proc_fs.h>
  23#include <linux/debugfs.h>
  24#include <linux/kmemleak.h>
  25#include <linux/kasan.h>
  26#include <asm/cacheflush.h>
  27#include <asm/tlbflush.h>
  28#include <asm/page.h>
  29#include <linux/memcontrol.h>
  30#include <linux/stackdepot.h>
  31
  32#include "internal.h"
  33#include "slab.h"
  34
  35#define CREATE_TRACE_POINTS
  36#include <trace/events/kmem.h>
  37
 
 
  38enum slab_state slab_state;
  39LIST_HEAD(slab_caches);
  40DEFINE_MUTEX(slab_mutex);
  41struct kmem_cache *kmem_cache;
  42
 
 
 
 
 
 
 
 
 
 
 
 
 
  43/*
  44 * Set of flags that will prevent slab merging
  45 */
  46#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  47		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
  48		SLAB_FAILSLAB | SLAB_NO_MERGE)
  49
  50#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
  51			 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
  52
  53/*
  54 * Merge control. If this is set then no merging of slab caches will occur.
  55 */
  56static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
  57
  58static int __init setup_slab_nomerge(char *str)
  59{
  60	slab_nomerge = true;
  61	return 1;
  62}
  63
  64static int __init setup_slab_merge(char *str)
  65{
  66	slab_nomerge = false;
  67	return 1;
  68}
  69
  70__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
  71__setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
  72
  73__setup("slab_nomerge", setup_slab_nomerge);
  74__setup("slab_merge", setup_slab_merge);
  75
  76/*
  77 * Determine the size of a slab object
  78 */
  79unsigned int kmem_cache_size(struct kmem_cache *s)
  80{
  81	return s->object_size;
  82}
  83EXPORT_SYMBOL(kmem_cache_size);
  84
  85#ifdef CONFIG_DEBUG_VM
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  86
  87static bool kmem_cache_is_duplicate_name(const char *name)
 
  88{
  89	struct kmem_cache *s;
  90
  91	list_for_each_entry(s, &slab_caches, list) {
  92		if (!strcmp(s->name, name))
  93			return true;
 
 
 
  94	}
 
 
  95
  96	return false;
 
 
 
 
 
 
 
 
  97}
  98
  99static int kmem_cache_sanity_check(const char *name, unsigned int size)
 
 100{
 101	if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
 102		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
 103		return -EINVAL;
 
 
 
 
 
 104	}
 105
 106	/* Duplicate names will confuse slabtop, et al */
 107	WARN(kmem_cache_is_duplicate_name(name),
 108			"kmem_cache of name '%s' already exists\n", name);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 109
 110	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
 
 
 
 
 
 
 
 
 111	return 0;
 112}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 113#else
 114static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
 
 115{
 116	return 0;
 117}
 118#endif
 
 
 
 
 
 
 
 
 119
 120/*
 121 * Figure out what the alignment of the objects will be given a set of
 122 * flags, a user specified alignment and the size of the objects.
 123 */
 124static unsigned int calculate_alignment(slab_flags_t flags,
 125		unsigned int align, unsigned int size)
 126{
 127	/*
 128	 * If the user wants hardware cache aligned objects then follow that
 129	 * suggestion if the object is sufficiently large.
 130	 *
 131	 * The hardware cache alignment cannot override the specified
 132	 * alignment though. If that is greater then use it.
 133	 */
 134	if (flags & SLAB_HWCACHE_ALIGN) {
 135		unsigned int ralign;
 136
 137		ralign = cache_line_size();
 138		while (size <= ralign / 2)
 139			ralign /= 2;
 140		align = max(align, ralign);
 141	}
 142
 143	align = max(align, arch_slab_minalign());
 
 144
 145	return ALIGN(align, sizeof(void *));
 146}
 147
 148/*
 149 * Find a mergeable slab cache
 150 */
 151int slab_unmergeable(struct kmem_cache *s)
 152{
 153	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 154		return 1;
 155
 
 
 
 156	if (s->ctor)
 157		return 1;
 158
 159#ifdef CONFIG_HARDENED_USERCOPY
 160	if (s->usersize)
 161		return 1;
 162#endif
 163
 164	/*
 165	 * We may have set a slab to be unmergeable during bootstrap.
 166	 */
 167	if (s->refcount < 0)
 168		return 1;
 169
 170	return 0;
 171}
 172
 173struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
 174		slab_flags_t flags, const char *name, void (*ctor)(void *))
 175{
 176	struct kmem_cache *s;
 177
 178	if (slab_nomerge)
 179		return NULL;
 180
 181	if (ctor)
 182		return NULL;
 183
 184	flags = kmem_cache_flags(flags, name);
 
 
 
 185
 186	if (flags & SLAB_NEVER_MERGE)
 187		return NULL;
 188
 189	size = ALIGN(size, sizeof(void *));
 190	align = calculate_alignment(flags, align, size);
 191	size = ALIGN(size, align);
 192
 193	list_for_each_entry_reverse(s, &slab_caches, list) {
 194		if (slab_unmergeable(s))
 195			continue;
 196
 197		if (size > s->size)
 198			continue;
 199
 200		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 201			continue;
 202		/*
 203		 * Check if alignment is compatible.
 204		 * Courtesy of Adrian Drzewiecki
 205		 */
 206		if ((s->size & ~(align - 1)) != s->size)
 207			continue;
 208
 209		if (s->size - size >= sizeof(void *))
 210			continue;
 211
 
 
 
 
 212		return s;
 213	}
 214	return NULL;
 215}
 216
 217static struct kmem_cache *create_cache(const char *name,
 218				       unsigned int object_size,
 219				       struct kmem_cache_args *args,
 220				       slab_flags_t flags)
 
 221{
 222	struct kmem_cache *s;
 223	int err;
 224
 225	/* If a custom freelist pointer is requested make sure it's sane. */
 226	err = -EINVAL;
 227	if (args->use_freeptr_offset &&
 228	    (args->freeptr_offset >= object_size ||
 229	     !(flags & SLAB_TYPESAFE_BY_RCU) ||
 230	     !IS_ALIGNED(args->freeptr_offset, __alignof__(freeptr_t))))
 231		goto out;
 232
 233	err = -ENOMEM;
 234	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 235	if (!s)
 236		goto out;
 237	err = do_kmem_cache_create(s, name, object_size, args, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 238	if (err)
 239		goto out_free_cache;
 240
 241	s->refcount = 1;
 242	list_add(&s->list, &slab_caches);
 
 
 
 
 243	return s;
 244
 245out_free_cache:
 
 246	kmem_cache_free(kmem_cache, s);
 247out:
 248	return ERR_PTR(err);
 249}
 250
 251/**
 252 * __kmem_cache_create_args - Create a kmem cache.
 253 * @name: A string which is used in /proc/slabinfo to identify this cache.
 254 * @object_size: The size of objects to be created in this cache.
 255 * @args: Additional arguments for the cache creation (see
 256 *        &struct kmem_cache_args).
 257 * @flags: See the desriptions of individual flags. The common ones are listed
 258 *         in the description below.
 259 *
 260 * Not to be called directly, use the kmem_cache_create() wrapper with the same
 261 * parameters.
 262 *
 263 * Commonly used @flags:
 264 *
 265 * &SLAB_ACCOUNT - Account allocations to memcg.
 266 *
 267 * &SLAB_HWCACHE_ALIGN - Align objects on cache line boundaries.
 268 *
 269 * &SLAB_RECLAIM_ACCOUNT - Objects are reclaimable.
 270 *
 271 * &SLAB_TYPESAFE_BY_RCU - Slab page (not individual objects) freeing delayed
 272 * by a grace period - see the full description before using.
 273 *
 274 * Context: Cannot be called within a interrupt, but can be interrupted.
 275 *
 276 * Return: a pointer to the cache on success, NULL on failure.
 277 */
 278struct kmem_cache *__kmem_cache_create_args(const char *name,
 279					    unsigned int object_size,
 280					    struct kmem_cache_args *args,
 281					    slab_flags_t flags)
 
 
 282{
 283	struct kmem_cache *s = NULL;
 284	const char *cache_name;
 285	int err;
 286
 287#ifdef CONFIG_SLUB_DEBUG
 288	/*
 289	 * If no slab_debug was enabled globally, the static key is not yet
 290	 * enabled by setup_slub_debug(). Enable it if the cache is being
 291	 * created with any of the debugging flags passed explicitly.
 292	 * It's also possible that this is the first cache created with
 293	 * SLAB_STORE_USER and we should init stack_depot for it.
 294	 */
 295	if (flags & SLAB_DEBUG_FLAGS)
 296		static_branch_enable(&slub_debug_enabled);
 297	if (flags & SLAB_STORE_USER)
 298		stack_depot_init();
 299#endif
 300
 301	mutex_lock(&slab_mutex);
 302
 303	err = kmem_cache_sanity_check(name, object_size);
 304	if (err) {
 305		goto out_unlock;
 306	}
 307
 308	/* Refuse requests with allocator specific flags */
 309	if (flags & ~SLAB_FLAGS_PERMITTED) {
 310		err = -EINVAL;
 311		goto out_unlock;
 312	}
 313
 314	/*
 315	 * Some allocators will constraint the set of valid flags to a subset
 316	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
 317	 * case, and we'll just provide them with a sanitized version of the
 318	 * passed flags.
 319	 */
 320	flags &= CACHE_CREATE_MASK;
 321
 322	/* Fail closed on bad usersize of useroffset values. */
 323	if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
 324	    WARN_ON(!args->usersize && args->useroffset) ||
 325	    WARN_ON(object_size < args->usersize ||
 326		    object_size - args->usersize < args->useroffset))
 327		args->usersize = args->useroffset = 0;
 328
 329	if (!args->usersize)
 330		s = __kmem_cache_alias(name, object_size, args->align, flags,
 331				       args->ctor);
 332	if (s)
 333		goto out_unlock;
 334
 335	cache_name = kstrdup_const(name, GFP_KERNEL);
 336	if (!cache_name) {
 337		err = -ENOMEM;
 338		goto out_unlock;
 339	}
 340
 341	args->align = calculate_alignment(flags, args->align, object_size);
 342	s = create_cache(cache_name, object_size, args, flags);
 
 343	if (IS_ERR(s)) {
 344		err = PTR_ERR(s);
 345		kfree_const(cache_name);
 346	}
 347
 348out_unlock:
 349	mutex_unlock(&slab_mutex);
 350
 
 
 
 
 351	if (err) {
 352		if (flags & SLAB_PANIC)
 353			panic("%s: Failed to create slab '%s'. Error %d\n",
 354				__func__, name, err);
 355		else {
 356			pr_warn("%s(%s) failed with error %d\n",
 357				__func__, name, err);
 358			dump_stack();
 359		}
 360		return NULL;
 361	}
 362	return s;
 363}
 364EXPORT_SYMBOL(__kmem_cache_create_args);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 365
 366static struct kmem_cache *kmem_buckets_cache __ro_after_init;
 
 
 
 
 
 
 
 
 
 
 
 
 367
 368/**
 369 * kmem_buckets_create - Create a set of caches that handle dynamic sized
 370 *			 allocations via kmem_buckets_alloc()
 371 * @name: A prefix string which is used in /proc/slabinfo to identify this
 372 *	  cache. The individual caches with have their sizes as the suffix.
 373 * @flags: SLAB flags (see kmem_cache_create() for details).
 374 * @useroffset: Starting offset within an allocation that may be copied
 375 *		to/from userspace.
 376 * @usersize: How many bytes, starting at @useroffset, may be copied
 377 *		to/from userspace.
 378 * @ctor: A constructor for the objects, run when new allocations are made.
 379 *
 380 * Cannot be called within an interrupt, but can be interrupted.
 381 *
 382 * Return: a pointer to the cache on success, NULL on failure. When
 383 * CONFIG_SLAB_BUCKETS is not enabled, ZERO_SIZE_PTR is returned, and
 384 * subsequent calls to kmem_buckets_alloc() will fall back to kmalloc().
 385 * (i.e. callers only need to check for NULL on failure.)
 386 */
 387kmem_buckets *kmem_buckets_create(const char *name, slab_flags_t flags,
 388				  unsigned int useroffset,
 389				  unsigned int usersize,
 390				  void (*ctor)(void *))
 391{
 392	unsigned long mask = 0;
 393	unsigned int idx;
 394	kmem_buckets *b;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 395
 396	BUILD_BUG_ON(ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]) > BITS_PER_LONG);
 
 
 
 
 
 
 
 
 
 
 
 
 397
 398	/*
 399	 * When the separate buckets API is not built in, just return
 400	 * a non-NULL value for the kmem_buckets pointer, which will be
 401	 * unused when performing allocations.
 402	 */
 403	if (!IS_ENABLED(CONFIG_SLAB_BUCKETS))
 404		return ZERO_SIZE_PTR;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 405
 406	if (WARN_ON(!kmem_buckets_cache))
 407		return NULL;
 
 408
 409	b = kmem_cache_alloc(kmem_buckets_cache, GFP_KERNEL|__GFP_ZERO);
 410	if (WARN_ON(!b))
 411		return NULL;
 
 
 412
 413	flags |= SLAB_NO_MERGE;
 414
 415	for (idx = 0; idx < ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]); idx++) {
 416		char *short_size, *cache_name;
 417		unsigned int cache_useroffset, cache_usersize;
 418		unsigned int size, aligned_idx;
 419
 420		if (!kmalloc_caches[KMALLOC_NORMAL][idx])
 
 
 
 
 
 421			continue;
 422
 423		size = kmalloc_caches[KMALLOC_NORMAL][idx]->object_size;
 424		if (!size)
 425			continue;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 426
 427		short_size = strchr(kmalloc_caches[KMALLOC_NORMAL][idx]->name, '-');
 428		if (WARN_ON(!short_size))
 429			goto fail;
 430
 431		if (useroffset >= size) {
 432			cache_useroffset = 0;
 433			cache_usersize = 0;
 434		} else {
 435			cache_useroffset = useroffset;
 436			cache_usersize = min(size - cache_useroffset, usersize);
 437		}
 438
 439		aligned_idx = __kmalloc_index(size, false);
 440		if (!(*b)[aligned_idx]) {
 441			cache_name = kasprintf(GFP_KERNEL, "%s-%s", name, short_size + 1);
 442			if (WARN_ON(!cache_name))
 443				goto fail;
 444			(*b)[aligned_idx] = kmem_cache_create_usercopy(cache_name, size,
 445					0, flags, cache_useroffset,
 446					cache_usersize, ctor);
 447			kfree(cache_name);
 448			if (WARN_ON(!(*b)[aligned_idx]))
 449				goto fail;
 450			set_bit(aligned_idx, &mask);
 451		}
 452		if (idx != aligned_idx)
 453			(*b)[idx] = (*b)[aligned_idx];
 
 
 
 
 
 
 
 
 
 
 454	}
 455
 456	return b;
 
 
 
 
 
 
 457
 458fail:
 459	for_each_set_bit(idx, &mask, ARRAY_SIZE(kmalloc_caches[KMALLOC_NORMAL]))
 460		kmem_cache_destroy((*b)[idx]);
 461	kmem_cache_free(kmem_buckets_cache, b);
 462
 463	return NULL;
 
 
 
 
 
 
 464}
 465EXPORT_SYMBOL(kmem_buckets_create);
 466
 467/*
 468 * For a given kmem_cache, kmem_cache_destroy() should only be called
 469 * once or there will be a use-after-free problem. The actual deletion
 470 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
 471 * protection. So they are now done without holding those locks.
 472 */
 473static void kmem_cache_release(struct kmem_cache *s)
 474{
 475	kfence_shutdown_cache(s);
 476	if (__is_defined(SLAB_SUPPORTS_SYSFS) && slab_state >= FULL)
 477		sysfs_slab_release(s);
 478	else
 479		slab_kmem_cache_release(s);
 480}
 
 481
 482void slab_kmem_cache_release(struct kmem_cache *s)
 483{
 484	__kmem_cache_release(s);
 
 485	kfree_const(s->name);
 486	kmem_cache_free(kmem_cache, s);
 487}
 488
 489void kmem_cache_destroy(struct kmem_cache *s)
 490{
 491	int err;
 492
 493	if (unlikely(!s) || !kasan_check_byte(s))
 494		return;
 495
 496	/* in-flight kfree_rcu()'s may include objects from our cache */
 497	kvfree_rcu_barrier();
 498
 499	if (IS_ENABLED(CONFIG_SLUB_RCU_DEBUG) &&
 500	    (s->flags & SLAB_TYPESAFE_BY_RCU)) {
 501		/*
 502		 * Under CONFIG_SLUB_RCU_DEBUG, when objects in a
 503		 * SLAB_TYPESAFE_BY_RCU slab are freed, SLUB will internally
 504		 * defer their freeing with call_rcu().
 505		 * Wait for such call_rcu() invocations here before actually
 506		 * destroying the cache.
 507		 *
 508		 * It doesn't matter that we haven't looked at the slab refcount
 509		 * yet - slabs with SLAB_TYPESAFE_BY_RCU can't be merged, so
 510		 * the refcount should be 1 here.
 511		 */
 512		rcu_barrier();
 513	}
 514
 515	cpus_read_lock();
 516	mutex_lock(&slab_mutex);
 517
 518	s->refcount--;
 519	if (s->refcount) {
 520		mutex_unlock(&slab_mutex);
 521		cpus_read_unlock();
 522		return;
 523	}
 524
 525	/* free asan quarantined objects */
 526	kasan_cache_shutdown(s);
 527
 528	err = __kmem_cache_shutdown(s);
 529	if (!slab_in_kunit_test())
 530		WARN(err, "%s %s: Slab cache still has objects when called from %pS",
 531		     __func__, s->name, (void *)_RET_IP_);
 532
 533	list_del(&s->list);
 534
 
 
 
 
 
 
 535	mutex_unlock(&slab_mutex);
 536	cpus_read_unlock();
 537
 538	if (slab_state >= FULL)
 539		sysfs_slab_unlink(s);
 540	debugfs_slab_release(s);
 541
 542	if (err)
 543		return;
 544
 545	if (s->flags & SLAB_TYPESAFE_BY_RCU)
 546		rcu_barrier();
 547
 548	kmem_cache_release(s);
 549}
 550EXPORT_SYMBOL(kmem_cache_destroy);
 551
 552/**
 553 * kmem_cache_shrink - Shrink a cache.
 554 * @cachep: The cache to shrink.
 555 *
 556 * Releases as many slabs as possible for a cache.
 557 * To help debugging, a zero exit status indicates all slabs were released.
 558 *
 559 * Return: %0 if all slabs were released, non-zero otherwise
 560 */
 561int kmem_cache_shrink(struct kmem_cache *cachep)
 562{
 
 
 
 
 563	kasan_cache_shrink(cachep);
 564
 565	return __kmem_cache_shrink(cachep);
 
 
 566}
 567EXPORT_SYMBOL(kmem_cache_shrink);
 568
 569bool slab_is_available(void)
 570{
 571	return slab_state >= UP;
 572}
 573
 574#ifdef CONFIG_PRINTK
 575static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
 576{
 577	if (__kfence_obj_info(kpp, object, slab))
 578		return;
 579	__kmem_obj_info(kpp, object, slab);
 580}
 581
 582/**
 583 * kmem_dump_obj - Print available slab provenance information
 584 * @object: slab object for which to find provenance information.
 585 *
 586 * This function uses pr_cont(), so that the caller is expected to have
 587 * printed out whatever preamble is appropriate.  The provenance information
 588 * depends on the type of object and on how much debugging is enabled.
 589 * For a slab-cache object, the fact that it is a slab object is printed,
 590 * and, if available, the slab name, return address, and stack trace from
 591 * the allocation and last free path of that object.
 592 *
 593 * Return: %true if the pointer is to a not-yet-freed object from
 594 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
 595 * is to an already-freed object, and %false otherwise.
 596 */
 597bool kmem_dump_obj(void *object)
 598{
 599	char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
 600	int i;
 601	struct slab *slab;
 602	unsigned long ptroffset;
 603	struct kmem_obj_info kp = { };
 604
 605	/* Some arches consider ZERO_SIZE_PTR to be a valid address. */
 606	if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
 607		return false;
 608	slab = virt_to_slab(object);
 609	if (!slab)
 610		return false;
 611
 612	kmem_obj_info(&kp, object, slab);
 613	if (kp.kp_slab_cache)
 614		pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
 615	else
 616		pr_cont(" slab%s", cp);
 617	if (is_kfence_address(object))
 618		pr_cont(" (kfence)");
 619	if (kp.kp_objp)
 620		pr_cont(" start %px", kp.kp_objp);
 621	if (kp.kp_data_offset)
 622		pr_cont(" data offset %lu", kp.kp_data_offset);
 623	if (kp.kp_objp) {
 624		ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
 625		pr_cont(" pointer offset %lu", ptroffset);
 626	}
 627	if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
 628		pr_cont(" size %u", kp.kp_slab_cache->object_size);
 629	if (kp.kp_ret)
 630		pr_cont(" allocated at %pS\n", kp.kp_ret);
 631	else
 632		pr_cont("\n");
 633	for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
 634		if (!kp.kp_stack[i])
 635			break;
 636		pr_info("    %pS\n", kp.kp_stack[i]);
 637	}
 638
 639	if (kp.kp_free_stack[0])
 640		pr_cont(" Free path:\n");
 641
 642	for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
 643		if (!kp.kp_free_stack[i])
 644			break;
 645		pr_info("    %pS\n", kp.kp_free_stack[i]);
 646	}
 647
 648	return true;
 649}
 650EXPORT_SYMBOL_GPL(kmem_dump_obj);
 651#endif
 652
 653/* Create a cache during boot when no slab services are available yet */
 654void __init create_boot_cache(struct kmem_cache *s, const char *name,
 655		unsigned int size, slab_flags_t flags,
 656		unsigned int useroffset, unsigned int usersize)
 657{
 658	int err;
 659	unsigned int align = ARCH_KMALLOC_MINALIGN;
 660	struct kmem_cache_args kmem_args = {};
 661
 662	/*
 663	 * kmalloc caches guarantee alignment of at least the largest
 664	 * power-of-two divisor of the size. For power-of-two sizes,
 665	 * it is the size itself.
 666	 */
 667	if (flags & SLAB_KMALLOC)
 668		align = max(align, 1U << (ffs(size) - 1));
 669	kmem_args.align = calculate_alignment(flags, align, size);
 670
 671#ifdef CONFIG_HARDENED_USERCOPY
 672	kmem_args.useroffset = useroffset;
 673	kmem_args.usersize = usersize;
 674#endif
 675
 676	err = do_kmem_cache_create(s, name, size, &kmem_args, flags);
 677
 678	if (err)
 679		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
 680					name, size, err);
 681
 682	s->refcount = -1;	/* Exempt from merging for now */
 683}
 684
 685static struct kmem_cache *__init create_kmalloc_cache(const char *name,
 686						      unsigned int size,
 687						      slab_flags_t flags)
 688{
 689	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 690
 691	if (!s)
 692		panic("Out of memory when creating slab %s\n", name);
 693
 694	create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
 695	list_add(&s->list, &slab_caches);
 
 696	s->refcount = 1;
 697	return s;
 698}
 699
 700kmem_buckets kmalloc_caches[NR_KMALLOC_TYPES] __ro_after_init =
 701{ /* initialization for https://llvm.org/pr42570 */ };
 702EXPORT_SYMBOL(kmalloc_caches);
 703
 704#ifdef CONFIG_RANDOM_KMALLOC_CACHES
 705unsigned long random_kmalloc_seed __ro_after_init;
 706EXPORT_SYMBOL(random_kmalloc_seed);
 707#endif
 708
 709/*
 710 * Conversion table for small slabs sizes / 8 to the index in the
 711 * kmalloc array. This is necessary for slabs < 192 since we have non power
 712 * of two cache sizes there. The size of larger slabs can be determined using
 713 * fls.
 714 */
 715u8 kmalloc_size_index[24] __ro_after_init = {
 716	3,	/* 8 */
 717	4,	/* 16 */
 718	5,	/* 24 */
 719	5,	/* 32 */
 720	6,	/* 40 */
 721	6,	/* 48 */
 722	6,	/* 56 */
 723	6,	/* 64 */
 724	1,	/* 72 */
 725	1,	/* 80 */
 726	1,	/* 88 */
 727	1,	/* 96 */
 728	7,	/* 104 */
 729	7,	/* 112 */
 730	7,	/* 120 */
 731	7,	/* 128 */
 732	2,	/* 136 */
 733	2,	/* 144 */
 734	2,	/* 152 */
 735	2,	/* 160 */
 736	2,	/* 168 */
 737	2,	/* 176 */
 738	2,	/* 184 */
 739	2	/* 192 */
 740};
 741
 742size_t kmalloc_size_roundup(size_t size)
 
 
 
 
 
 
 
 
 
 743{
 744	if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
 745		/*
 746		 * The flags don't matter since size_index is common to all.
 747		 * Neither does the caller for just getting ->object_size.
 748		 */
 749		return kmalloc_slab(size, NULL, GFP_KERNEL, 0)->object_size;
 750	}
 751
 752	/* Above the smaller buckets, size is a multiple of page size. */
 753	if (size && size <= KMALLOC_MAX_SIZE)
 754		return PAGE_SIZE << get_order(size);
 755
 756	/*
 757	 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
 758	 * and very large size - kmalloc() may fail.
 759	 */
 760	return size;
 761
 762}
 763EXPORT_SYMBOL(kmalloc_size_roundup);
 
 764
 765#ifdef CONFIG_ZONE_DMA
 766#define KMALLOC_DMA_NAME(sz)	.name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
 767#else
 768#define KMALLOC_DMA_NAME(sz)
 769#endif
 770
 771#ifdef CONFIG_MEMCG
 772#define KMALLOC_CGROUP_NAME(sz)	.name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
 773#else
 774#define KMALLOC_CGROUP_NAME(sz)
 775#endif
 776
 777#ifndef CONFIG_SLUB_TINY
 778#define KMALLOC_RCL_NAME(sz)	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
 779#else
 780#define KMALLOC_RCL_NAME(sz)
 781#endif
 782
 783#ifdef CONFIG_RANDOM_KMALLOC_CACHES
 784#define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
 785#define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
 786#define KMA_RAND_1(sz)                  .name[KMALLOC_RANDOM_START +  1] = "kmalloc-rnd-01-" #sz,
 787#define KMA_RAND_2(sz)  KMA_RAND_1(sz)  .name[KMALLOC_RANDOM_START +  2] = "kmalloc-rnd-02-" #sz,
 788#define KMA_RAND_3(sz)  KMA_RAND_2(sz)  .name[KMALLOC_RANDOM_START +  3] = "kmalloc-rnd-03-" #sz,
 789#define KMA_RAND_4(sz)  KMA_RAND_3(sz)  .name[KMALLOC_RANDOM_START +  4] = "kmalloc-rnd-04-" #sz,
 790#define KMA_RAND_5(sz)  KMA_RAND_4(sz)  .name[KMALLOC_RANDOM_START +  5] = "kmalloc-rnd-05-" #sz,
 791#define KMA_RAND_6(sz)  KMA_RAND_5(sz)  .name[KMALLOC_RANDOM_START +  6] = "kmalloc-rnd-06-" #sz,
 792#define KMA_RAND_7(sz)  KMA_RAND_6(sz)  .name[KMALLOC_RANDOM_START +  7] = "kmalloc-rnd-07-" #sz,
 793#define KMA_RAND_8(sz)  KMA_RAND_7(sz)  .name[KMALLOC_RANDOM_START +  8] = "kmalloc-rnd-08-" #sz,
 794#define KMA_RAND_9(sz)  KMA_RAND_8(sz)  .name[KMALLOC_RANDOM_START +  9] = "kmalloc-rnd-09-" #sz,
 795#define KMA_RAND_10(sz) KMA_RAND_9(sz)  .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
 796#define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
 797#define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
 798#define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
 799#define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
 800#define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
 801#else // CONFIG_RANDOM_KMALLOC_CACHES
 802#define KMALLOC_RANDOM_NAME(N, sz)
 803#endif
 804
 805#define INIT_KMALLOC_INFO(__size, __short_size)			\
 806{								\
 807	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
 808	KMALLOC_RCL_NAME(__short_size)				\
 809	KMALLOC_CGROUP_NAME(__short_size)			\
 810	KMALLOC_DMA_NAME(__short_size)				\
 811	KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size)	\
 812	.size = __size,						\
 813}
 814
 815/*
 816 * kmalloc_info[] is to make slab_debug=,kmalloc-xx option work at boot time.
 817 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
 818 * kmalloc-2M.
 819 */
 820const struct kmalloc_info_struct kmalloc_info[] __initconst = {
 821	INIT_KMALLOC_INFO(0, 0),
 822	INIT_KMALLOC_INFO(96, 96),
 823	INIT_KMALLOC_INFO(192, 192),
 824	INIT_KMALLOC_INFO(8, 8),
 825	INIT_KMALLOC_INFO(16, 16),
 826	INIT_KMALLOC_INFO(32, 32),
 827	INIT_KMALLOC_INFO(64, 64),
 828	INIT_KMALLOC_INFO(128, 128),
 829	INIT_KMALLOC_INFO(256, 256),
 830	INIT_KMALLOC_INFO(512, 512),
 831	INIT_KMALLOC_INFO(1024, 1k),
 832	INIT_KMALLOC_INFO(2048, 2k),
 833	INIT_KMALLOC_INFO(4096, 4k),
 834	INIT_KMALLOC_INFO(8192, 8k),
 835	INIT_KMALLOC_INFO(16384, 16k),
 836	INIT_KMALLOC_INFO(32768, 32k),
 837	INIT_KMALLOC_INFO(65536, 64k),
 838	INIT_KMALLOC_INFO(131072, 128k),
 839	INIT_KMALLOC_INFO(262144, 256k),
 840	INIT_KMALLOC_INFO(524288, 512k),
 841	INIT_KMALLOC_INFO(1048576, 1M),
 842	INIT_KMALLOC_INFO(2097152, 2M)
 843};
 844
 845/*
 846 * Patch up the size_index table if we have strange large alignment
 847 * requirements for the kmalloc array. This is only the case for
 848 * MIPS it seems. The standard arches will not generate any code here.
 849 *
 850 * Largest permitted alignment is 256 bytes due to the way we
 851 * handle the index determination for the smaller caches.
 852 *
 853 * Make sure that nothing crazy happens if someone starts tinkering
 854 * around with ARCH_KMALLOC_MINALIGN
 855 */
 856void __init setup_kmalloc_cache_index_table(void)
 857{
 858	unsigned int i;
 859
 860	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 861		!is_power_of_2(KMALLOC_MIN_SIZE));
 862
 863	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 864		unsigned int elem = size_index_elem(i);
 865
 866		if (elem >= ARRAY_SIZE(kmalloc_size_index))
 867			break;
 868		kmalloc_size_index[elem] = KMALLOC_SHIFT_LOW;
 869	}
 870
 871	if (KMALLOC_MIN_SIZE >= 64) {
 872		/*
 873		 * The 96 byte sized cache is not used if the alignment
 874		 * is 64 byte.
 875		 */
 876		for (i = 64 + 8; i <= 96; i += 8)
 877			kmalloc_size_index[size_index_elem(i)] = 7;
 878
 879	}
 880
 881	if (KMALLOC_MIN_SIZE >= 128) {
 882		/*
 883		 * The 192 byte sized cache is not used if the alignment
 884		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
 885		 * instead.
 886		 */
 887		for (i = 128 + 8; i <= 192; i += 8)
 888			kmalloc_size_index[size_index_elem(i)] = 8;
 889	}
 890}
 891
 892static unsigned int __kmalloc_minalign(void)
 893{
 894	unsigned int minalign = dma_get_cache_alignment();
 895
 896	if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
 897	    is_swiotlb_allocated())
 898		minalign = ARCH_KMALLOC_MINALIGN;
 899
 900	return max(minalign, arch_slab_minalign());
 901}
 902
 903static void __init
 904new_kmalloc_cache(int idx, enum kmalloc_cache_type type)
 905{
 906	slab_flags_t flags = 0;
 907	unsigned int minalign = __kmalloc_minalign();
 908	unsigned int aligned_size = kmalloc_info[idx].size;
 909	int aligned_idx = idx;
 910
 911	if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
 912		flags |= SLAB_RECLAIM_ACCOUNT;
 913	} else if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_CGROUP)) {
 914		if (mem_cgroup_kmem_disabled()) {
 915			kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
 916			return;
 917		}
 918		flags |= SLAB_ACCOUNT;
 919	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
 920		flags |= SLAB_CACHE_DMA;
 921	}
 922
 923#ifdef CONFIG_RANDOM_KMALLOC_CACHES
 924	if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
 925		flags |= SLAB_NO_MERGE;
 926#endif
 927
 928	/*
 929	 * If CONFIG_MEMCG is enabled, disable cache merging for
 930	 * KMALLOC_NORMAL caches.
 931	 */
 932	if (IS_ENABLED(CONFIG_MEMCG) && (type == KMALLOC_NORMAL))
 933		flags |= SLAB_NO_MERGE;
 934
 935	if (minalign > ARCH_KMALLOC_MINALIGN) {
 936		aligned_size = ALIGN(aligned_size, minalign);
 937		aligned_idx = __kmalloc_index(aligned_size, false);
 938	}
 939
 940	if (!kmalloc_caches[type][aligned_idx])
 941		kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
 942					kmalloc_info[aligned_idx].name[type],
 943					aligned_size, flags);
 944	if (idx != aligned_idx)
 945		kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
 946}
 947
 948/*
 949 * Create the kmalloc array. Some of the regular kmalloc arrays
 950 * may already have been created because they were needed to
 951 * enable allocations for slab creation.
 952 */
 953void __init create_kmalloc_caches(void)
 954{
 955	int i;
 956	enum kmalloc_cache_type type;
 957
 958	/*
 959	 * Including KMALLOC_CGROUP if CONFIG_MEMCG defined
 960	 */
 961	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
 962		/* Caches that are NOT of the two-to-the-power-of size. */
 963		if (KMALLOC_MIN_SIZE <= 32)
 964			new_kmalloc_cache(1, type);
 965		if (KMALLOC_MIN_SIZE <= 64)
 966			new_kmalloc_cache(2, type);
 967
 968		/* Caches that are of the two-to-the-power-of size. */
 969		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
 970			new_kmalloc_cache(i, type);
 971	}
 972#ifdef CONFIG_RANDOM_KMALLOC_CACHES
 973	random_kmalloc_seed = get_random_u64();
 974#endif
 975
 976	/* Kmalloc array is now usable */
 977	slab_state = UP;
 978
 979	if (IS_ENABLED(CONFIG_SLAB_BUCKETS))
 980		kmem_buckets_cache = kmem_cache_create("kmalloc_buckets",
 981						       sizeof(kmem_buckets),
 982						       0, SLAB_NO_MERGE, NULL);
 
 
 
 
 
 
 
 
 
 
 
 983}
 
 984
 985/**
 986 * __ksize -- Report full size of underlying allocation
 987 * @object: pointer to the object
 988 *
 989 * This should only be used internally to query the true size of allocations.
 990 * It is not meant to be a way to discover the usable size of an allocation
 991 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
 992 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
 993 * and/or FORTIFY_SOURCE.
 994 *
 995 * Return: size of the actual memory used by @object in bytes
 996 */
 997size_t __ksize(const void *object)
 998{
 999	struct folio *folio;
1000
1001	if (unlikely(object == ZERO_SIZE_PTR))
1002		return 0;
1003
1004	folio = virt_to_folio(object);
1005
1006	if (unlikely(!folio_test_slab(folio))) {
1007		if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1008			return 0;
1009		if (WARN_ON(object != folio_address(folio)))
1010			return 0;
1011		return folio_size(folio);
1012	}
1013
1014#ifdef CONFIG_SLUB_DEBUG
1015	skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1016#endif
1017
1018	return slab_ksize(folio_slab(folio)->slab_cache);
 
 
 
 
 
1019}
 
1020
1021gfp_t kmalloc_fix_flags(gfp_t flags)
 
1022{
1023	gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1024
1025	flags &= ~GFP_SLAB_BUG_MASK;
1026	pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1027			invalid_mask, &invalid_mask, flags, &flags);
1028	dump_stack();
1029
1030	return flags;
1031}
 
 
1032
1033#ifdef CONFIG_SLAB_FREELIST_RANDOM
1034/* Randomize a generic freelist */
1035static void freelist_randomize(unsigned int *list,
1036			       unsigned int count)
1037{
1038	unsigned int rand;
1039	unsigned int i;
1040
1041	for (i = 0; i < count; i++)
1042		list[i] = i;
1043
1044	/* Fisher-Yates shuffle */
1045	for (i = count - 1; i > 0; i--) {
1046		rand = get_random_u32_below(i + 1);
 
1047		swap(list[i], list[rand]);
1048	}
1049}
1050
1051/* Create a random sequence per cache */
1052int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1053				    gfp_t gfp)
1054{
 
1055
1056	if (count < 2 || cachep->random_seq)
1057		return 0;
1058
1059	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1060	if (!cachep->random_seq)
1061		return -ENOMEM;
1062
1063	freelist_randomize(cachep->random_seq, count);
 
 
 
1064	return 0;
1065}
1066
1067/* Destroy the per-cache random freelist sequence */
1068void cache_random_seq_destroy(struct kmem_cache *cachep)
1069{
1070	kfree(cachep->random_seq);
1071	cachep->random_seq = NULL;
1072}
1073#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1074
1075#ifdef CONFIG_SLUB_DEBUG
1076#define SLABINFO_RIGHTS (0400)
 
 
 
 
1077
1078static void print_slabinfo_header(struct seq_file *m)
1079{
1080	/*
1081	 * Output format version, so at least we can change it
1082	 * without _too_ many complaints.
1083	 */
 
 
 
1084	seq_puts(m, "slabinfo - version: 2.1\n");
 
1085	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1086	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1087	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
 
 
 
 
1088	seq_putc(m, '\n');
1089}
1090
1091static void *slab_start(struct seq_file *m, loff_t *pos)
1092{
1093	mutex_lock(&slab_mutex);
1094	return seq_list_start(&slab_caches, *pos);
1095}
1096
1097static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1098{
1099	return seq_list_next(p, &slab_caches, pos);
1100}
1101
1102static void slab_stop(struct seq_file *m, void *p)
1103{
1104	mutex_unlock(&slab_mutex);
1105}
1106
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1107static void cache_show(struct kmem_cache *s, struct seq_file *m)
1108{
1109	struct slabinfo sinfo;
1110
1111	memset(&sinfo, 0, sizeof(sinfo));
1112	get_slabinfo(s, &sinfo);
1113
 
 
1114	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1115		   s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1116		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1117
1118	seq_printf(m, " : tunables %4u %4u %4u",
1119		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1120	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1121		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
 
1122	seq_putc(m, '\n');
1123}
1124
1125static int slab_show(struct seq_file *m, void *p)
1126{
1127	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1128
1129	if (p == slab_caches.next)
1130		print_slabinfo_header(m);
1131	cache_show(s, m);
1132	return 0;
1133}
1134
1135void dump_unreclaimable_slab(void)
1136{
1137	struct kmem_cache *s;
1138	struct slabinfo sinfo;
1139
1140	/*
1141	 * Here acquiring slab_mutex is risky since we don't prefer to get
1142	 * sleep in oom path. But, without mutex hold, it may introduce a
1143	 * risk of crash.
1144	 * Use mutex_trylock to protect the list traverse, dump nothing
1145	 * without acquiring the mutex.
1146	 */
1147	if (!mutex_trylock(&slab_mutex)) {
1148		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1149		return;
1150	}
1151
1152	pr_info("Unreclaimable slab info:\n");
1153	pr_info("Name                      Used          Total\n");
1154
1155	list_for_each_entry(s, &slab_caches, list) {
1156		if (s->flags & SLAB_RECLAIM_ACCOUNT)
1157			continue;
1158
1159		get_slabinfo(s, &sinfo);
1160
1161		if (sinfo.num_objs > 0)
1162			pr_info("%-17s %10luKB %10luKB\n", s->name,
1163				(sinfo.active_objs * s->size) / 1024,
1164				(sinfo.num_objs * s->size) / 1024);
1165	}
1166	mutex_unlock(&slab_mutex);
1167}
1168
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1169/*
1170 * slabinfo_op - iterator that generates /proc/slabinfo
1171 *
1172 * Output layout:
1173 * cache-name
1174 * num-active-objs
1175 * total-objs
1176 * object size
1177 * num-active-slabs
1178 * total-slabs
1179 * num-pages-per-slab
1180 * + further values on SMP and with statistics enabled
1181 */
1182static const struct seq_operations slabinfo_op = {
1183	.start = slab_start,
1184	.next = slab_next,
1185	.stop = slab_stop,
1186	.show = slab_show,
1187};
1188
1189static int slabinfo_open(struct inode *inode, struct file *file)
1190{
1191	return seq_open(file, &slabinfo_op);
1192}
1193
1194static const struct proc_ops slabinfo_proc_ops = {
1195	.proc_flags	= PROC_ENTRY_PERMANENT,
1196	.proc_open	= slabinfo_open,
1197	.proc_read	= seq_read,
1198	.proc_lseek	= seq_lseek,
1199	.proc_release	= seq_release,
1200};
1201
1202static int __init slab_proc_init(void)
1203{
1204	proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
 
1205	return 0;
1206}
1207module_init(slab_proc_init);
 
1208
1209#endif /* CONFIG_SLUB_DEBUG */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1210
1211/**
1212 * kfree_sensitive - Clear sensitive information in memory before freeing
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1213 * @p: object to free memory of
1214 *
1215 * The memory of the object @p points to is zeroed before freed.
1216 * If @p is %NULL, kfree_sensitive() does nothing.
1217 *
1218 * Note: this function zeroes the whole allocated buffer which can be a good
1219 * deal bigger than the requested buffer size passed to kmalloc(). So be
1220 * careful when using this function in performance sensitive code.
1221 */
1222void kfree_sensitive(const void *p)
1223{
1224	size_t ks;
1225	void *mem = (void *)p;
1226
 
 
1227	ks = ksize(mem);
1228	if (ks) {
1229		kasan_unpoison_range(mem, ks);
1230		memzero_explicit(mem, ks);
1231	}
1232	kfree(mem);
1233}
1234EXPORT_SYMBOL(kfree_sensitive);
1235
1236size_t ksize(const void *objp)
1237{
1238	/*
1239	 * We need to first check that the pointer to the object is valid.
1240	 * The KASAN report printed from ksize() is more useful, then when
1241	 * it's printed later when the behaviour could be undefined due to
1242	 * a potential use-after-free or double-free.
1243	 *
1244	 * We use kasan_check_byte(), which is supported for the hardware
1245	 * tag-based KASAN mode, unlike kasan_check_read/write().
1246	 *
1247	 * If the pointed to memory is invalid, we return 0 to avoid users of
1248	 * ksize() writing to and potentially corrupting the memory region.
1249	 *
1250	 * We want to perform the check before __ksize(), to avoid potentially
1251	 * crashing in __ksize() due to accessing invalid metadata.
1252	 */
1253	if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1254		return 0;
1255
1256	return kfence_ksize(objp) ?: __ksize(objp);
1257}
1258EXPORT_SYMBOL(ksize);
1259
1260#ifdef CONFIG_BPF_SYSCALL
1261#include <linux/btf.h>
1262
1263__bpf_kfunc_start_defs();
1264
1265__bpf_kfunc struct kmem_cache *bpf_get_kmem_cache(u64 addr)
1266{
1267	struct slab *slab;
1268
1269	if (!virt_addr_valid((void *)(long)addr))
1270		return NULL;
1271
1272	slab = virt_to_slab((void *)(long)addr);
1273	return slab ? slab->slab_cache : NULL;
1274}
1275
1276__bpf_kfunc_end_defs();
1277#endif /* CONFIG_BPF_SYSCALL */
1278
1279/* Tracepoints definitions. */
1280EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1281EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
 
 
1282EXPORT_TRACEPOINT_SYMBOL(kfree);
1283EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1284