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);
v6.2
   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/proc_fs.h>
  21#include <linux/debugfs.h>
  22#include <linux/kasan.h>
  23#include <asm/cacheflush.h>
  24#include <asm/tlbflush.h>
  25#include <asm/page.h>
  26#include <linux/memcontrol.h>
  27#include <linux/stackdepot.h>
  28
  29#include "internal.h"
  30#include "slab.h"
  31
  32#define CREATE_TRACE_POINTS
  33#include <trace/events/kmem.h>
  34
 
 
  35enum slab_state slab_state;
  36LIST_HEAD(slab_caches);
  37DEFINE_MUTEX(slab_mutex);
  38struct kmem_cache *kmem_cache;
  39
 
 
 
 
 
 
 
 
  40static LIST_HEAD(slab_caches_to_rcu_destroy);
  41static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
  42static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
  43		    slab_caches_to_rcu_destroy_workfn);
  44
  45/*
  46 * Set of flags that will prevent slab merging
  47 */
  48#define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  49		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
  50		SLAB_FAILSLAB | kasan_never_merge())
  51
  52#define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
  53			 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
  54
  55/*
  56 * Merge control. If this is set then no merging of slab caches will occur.
  57 */
  58static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
  59
  60static int __init setup_slab_nomerge(char *str)
  61{
  62	slab_nomerge = true;
  63	return 1;
  64}
  65
  66static int __init setup_slab_merge(char *str)
  67{
  68	slab_nomerge = false;
  69	return 1;
  70}
  71
  72#ifdef CONFIG_SLUB
  73__setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
  74__setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
  75#endif
  76
  77__setup("slab_nomerge", setup_slab_nomerge);
  78__setup("slab_merge", setup_slab_merge);
  79
  80/*
  81 * Determine the size of a slab object
  82 */
  83unsigned int kmem_cache_size(struct kmem_cache *s)
  84{
  85	return s->object_size;
  86}
  87EXPORT_SYMBOL(kmem_cache_size);
  88
  89#ifdef CONFIG_DEBUG_VM
  90static int kmem_cache_sanity_check(const char *name, unsigned int size)
  91{
  92	if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
 
  93		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
  94		return -EINVAL;
  95	}
  96
  97	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
  98	return 0;
  99}
 100#else
 101static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
 102{
 103	return 0;
 104}
 105#endif
 106
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 107/*
 108 * Figure out what the alignment of the objects will be given a set of
 109 * flags, a user specified alignment and the size of the objects.
 110 */
 111static unsigned int calculate_alignment(slab_flags_t flags,
 112		unsigned int align, unsigned int size)
 113{
 114	/*
 115	 * If the user wants hardware cache aligned objects then follow that
 116	 * suggestion if the object is sufficiently large.
 117	 *
 118	 * The hardware cache alignment cannot override the specified
 119	 * alignment though. If that is greater then use it.
 120	 */
 121	if (flags & SLAB_HWCACHE_ALIGN) {
 122		unsigned int ralign;
 123
 124		ralign = cache_line_size();
 125		while (size <= ralign / 2)
 126			ralign /= 2;
 127		align = max(align, ralign);
 128	}
 129
 130	align = max(align, arch_slab_minalign());
 
 131
 132	return ALIGN(align, sizeof(void *));
 133}
 134
 135/*
 136 * Find a mergeable slab cache
 137 */
 138int slab_unmergeable(struct kmem_cache *s)
 139{
 140	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 141		return 1;
 142
 
 
 
 143	if (s->ctor)
 144		return 1;
 145
 146#ifdef CONFIG_HARDENED_USERCOPY
 147	if (s->usersize)
 148		return 1;
 149#endif
 150
 151	/*
 152	 * We may have set a slab to be unmergeable during bootstrap.
 153	 */
 154	if (s->refcount < 0)
 155		return 1;
 156
 157	return 0;
 158}
 159
 160struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
 161		slab_flags_t flags, const char *name, void (*ctor)(void *))
 162{
 163	struct kmem_cache *s;
 164
 165	if (slab_nomerge)
 166		return NULL;
 167
 168	if (ctor)
 169		return NULL;
 170
 171	size = ALIGN(size, sizeof(void *));
 172	align = calculate_alignment(flags, align, size);
 173	size = ALIGN(size, align);
 174	flags = kmem_cache_flags(size, flags, name);
 175
 176	if (flags & SLAB_NEVER_MERGE)
 177		return NULL;
 178
 179	list_for_each_entry_reverse(s, &slab_caches, list) {
 180		if (slab_unmergeable(s))
 181			continue;
 182
 183		if (size > s->size)
 184			continue;
 185
 186		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 187			continue;
 188		/*
 189		 * Check if alignment is compatible.
 190		 * Courtesy of Adrian Drzewiecki
 191		 */
 192		if ((s->size & ~(align - 1)) != s->size)
 193			continue;
 194
 195		if (s->size - size >= sizeof(void *))
 196			continue;
 197
 198		if (IS_ENABLED(CONFIG_SLAB) && align &&
 199			(align > s->align || s->align % align))
 200			continue;
 201
 202		return s;
 203	}
 204	return NULL;
 205}
 206
 207static struct kmem_cache *create_cache(const char *name,
 208		unsigned int object_size, unsigned int align,
 209		slab_flags_t flags, unsigned int useroffset,
 210		unsigned int usersize, void (*ctor)(void *),
 211		struct kmem_cache *root_cache)
 212{
 213	struct kmem_cache *s;
 214	int err;
 215
 216	if (WARN_ON(useroffset + usersize > object_size))
 217		useroffset = usersize = 0;
 218
 219	err = -ENOMEM;
 220	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 221	if (!s)
 222		goto out;
 223
 224	s->name = name;
 225	s->size = s->object_size = object_size;
 226	s->align = align;
 227	s->ctor = ctor;
 228#ifdef CONFIG_HARDENED_USERCOPY
 229	s->useroffset = useroffset;
 230	s->usersize = usersize;
 231#endif
 
 
 
 232
 233	err = __kmem_cache_create(s, flags);
 234	if (err)
 235		goto out_free_cache;
 236
 237	s->refcount = 1;
 238	list_add(&s->list, &slab_caches);
 
 239out:
 240	if (err)
 241		return ERR_PTR(err);
 242	return s;
 243
 244out_free_cache:
 
 245	kmem_cache_free(kmem_cache, s);
 246	goto out;
 247}
 248
 249/**
 250 * kmem_cache_create_usercopy - Create a cache with a region suitable
 251 * for copying to userspace
 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
 256 * @useroffset: Usercopy region offset
 257 * @usersize: Usercopy region size
 258 * @ctor: A constructor for the objects.
 259 *
 
 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 *
 268 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
 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.
 274 *
 275 * Return: a pointer to the cache on success, NULL on failure.
 276 */
 277struct kmem_cache *
 278kmem_cache_create_usercopy(const char *name,
 279		  unsigned int size, unsigned int align,
 280		  slab_flags_t flags,
 281		  unsigned int useroffset, unsigned int usersize,
 282		  void (*ctor)(void *))
 283{
 284	struct kmem_cache *s = NULL;
 285	const char *cache_name;
 286	int err;
 287
 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.
 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.
 295	 */
 296	if (flags & SLAB_DEBUG_FLAGS)
 297		static_branch_enable(&slub_debug_enabled);
 298	if (flags & SLAB_STORE_USER)
 299		stack_depot_init();
 300#endif
 301
 302	mutex_lock(&slab_mutex);
 303
 304	err = kmem_cache_sanity_check(name, size);
 305	if (err) {
 306		goto out_unlock;
 307	}
 308
 309	/* Refuse requests with allocator specific flags */
 310	if (flags & ~SLAB_FLAGS_PERMITTED) {
 311		err = -EINVAL;
 312		goto out_unlock;
 313	}
 314
 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;
 322
 323	/* Fail closed on bad usersize of useroffset values. */
 324	if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
 325	    WARN_ON(!usersize && useroffset) ||
 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);
 331	if (s)
 332		goto out_unlock;
 333
 334	cache_name = kstrdup_const(name, GFP_KERNEL);
 335	if (!cache_name) {
 336		err = -ENOMEM;
 337		goto out_unlock;
 338	}
 339
 340	s = create_cache(cache_name, size,
 341			 calculate_alignment(flags, align, size),
 342			 flags, useroffset, usersize, ctor, NULL);
 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_usercopy);
 365
 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 */
 391struct kmem_cache *
 392kmem_cache_create(const char *name, unsigned int size, unsigned int align,
 393		slab_flags_t flags, void (*ctor)(void *))
 394{
 395	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
 396					  ctor);
 397}
 398EXPORT_SYMBOL(kmem_cache_create);
 399
 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
 422static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
 423{
 424	LIST_HEAD(to_destroy);
 425	struct kmem_cache *s, *s2;
 426
 427	/*
 428	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
 429	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
 430	 * through RCU and the associated kmem_cache are dereferenced
 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);
 439
 440	if (list_empty(&to_destroy))
 441		return;
 442
 443	rcu_barrier();
 444
 445	list_for_each_entry_safe(s, s2, &to_destroy, list) {
 446		debugfs_slab_release(s);
 447		kfence_shutdown_cache(s);
 448		kmem_cache_release(s);
 
 
 449	}
 450}
 451
 452static int shutdown_cache(struct kmem_cache *s)
 453{
 454	/* free asan quarantined objects */
 455	kasan_cache_shutdown(s);
 456
 457	if (__kmem_cache_shutdown(s) != 0)
 458		return -EBUSY;
 459
 
 460	list_del(&s->list);
 461
 462	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
 463		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
 464		schedule_work(&slab_caches_to_rcu_destroy_work);
 465	} else {
 466		kfence_shutdown_cache(s);
 467		debugfs_slab_release(s);
 
 
 
 468	}
 469
 470	return 0;
 471}
 472
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 473void slab_kmem_cache_release(struct kmem_cache *s)
 474{
 475	__kmem_cache_release(s);
 
 476	kfree_const(s->name);
 477	kmem_cache_free(kmem_cache, s);
 478}
 479
 480void kmem_cache_destroy(struct kmem_cache *s)
 481{
 482	int refcnt;
 483	bool rcu_set;
 484
 485	if (unlikely(!s) || !kasan_check_byte(s))
 486		return;
 487
 488	cpus_read_lock();
 
 
 489	mutex_lock(&slab_mutex);
 490
 491	rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
 
 
 492
 493	refcnt = --s->refcount;
 494	if (refcnt)
 495		goto out_unlock;
 496
 497	WARN(shutdown_cache(s),
 498	     "%s %s: Slab cache still has objects when called from %pS",
 499	     __func__, s->name, (void *)_RET_IP_);
 
 
 500out_unlock:
 501	mutex_unlock(&slab_mutex);
 502	cpus_read_unlock();
 503	if (!refcnt && !rcu_set)
 504		kmem_cache_release(s);
 505}
 506EXPORT_SYMBOL(kmem_cache_destroy);
 507
 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.
 514 *
 515 * Return: %0 if all slabs were released, non-zero otherwise
 516 */
 517int kmem_cache_shrink(struct kmem_cache *cachep)
 518{
 
 
 
 
 519	kasan_cache_shrink(cachep);
 520
 521	return __kmem_cache_shrink(cachep);
 
 
 522}
 523EXPORT_SYMBOL(kmem_cache_shrink);
 524
 525bool slab_is_available(void)
 526{
 527	return slab_state >= UP;
 528}
 529
 530#ifdef CONFIG_PRINTK
 531/**
 532 * kmem_valid_obj - does the pointer reference a valid slab object?
 533 * @object: pointer to query.
 534 *
 535 * Return: %true if the pointer is to a not-yet-freed object from
 536 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
 537 * is to an already-freed object, and %false otherwise.
 538 */
 539bool kmem_valid_obj(void *object)
 540{
 541	struct folio *folio;
 542
 543	/* Some arches consider ZERO_SIZE_PTR to be a valid address. */
 544	if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
 545		return false;
 546	folio = virt_to_folio(object);
 547	return folio_test_slab(folio);
 548}
 549EXPORT_SYMBOL_GPL(kmem_valid_obj);
 550
 551static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
 552{
 553	if (__kfence_obj_info(kpp, object, slab))
 554		return;
 555	__kmem_obj_info(kpp, object, slab);
 556}
 557
 558/**
 559 * kmem_dump_obj - Print available slab provenance information
 560 * @object: slab object for which to find provenance information.
 561 *
 562 * This function uses pr_cont(), so that the caller is expected to have
 563 * printed out whatever preamble is appropriate.  The provenance information
 564 * depends on the type of object and on how much debugging is enabled.
 565 * For a slab-cache object, the fact that it is a slab object is printed,
 566 * and, if available, the slab name, return address, and stack trace from
 567 * the allocation and last free path of that object.
 568 *
 569 * This function will splat if passed a pointer to a non-slab object.
 570 * If you are not sure what type of object you have, you should instead
 571 * use mem_dump_obj().
 572 */
 573void kmem_dump_obj(void *object)
 574{
 575	char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
 576	int i;
 577	struct slab *slab;
 578	unsigned long ptroffset;
 579	struct kmem_obj_info kp = { };
 580
 581	if (WARN_ON_ONCE(!virt_addr_valid(object)))
 582		return;
 583	slab = virt_to_slab(object);
 584	if (WARN_ON_ONCE(!slab)) {
 585		pr_cont(" non-slab memory.\n");
 586		return;
 587	}
 588	kmem_obj_info(&kp, object, slab);
 589	if (kp.kp_slab_cache)
 590		pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
 591	else
 592		pr_cont(" slab%s", cp);
 593	if (is_kfence_address(object))
 594		pr_cont(" (kfence)");
 595	if (kp.kp_objp)
 596		pr_cont(" start %px", kp.kp_objp);
 597	if (kp.kp_data_offset)
 598		pr_cont(" data offset %lu", kp.kp_data_offset);
 599	if (kp.kp_objp) {
 600		ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
 601		pr_cont(" pointer offset %lu", ptroffset);
 602	}
 603	if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
 604		pr_cont(" size %u", kp.kp_slab_cache->object_size);
 605	if (kp.kp_ret)
 606		pr_cont(" allocated at %pS\n", kp.kp_ret);
 607	else
 608		pr_cont("\n");
 609	for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
 610		if (!kp.kp_stack[i])
 611			break;
 612		pr_info("    %pS\n", kp.kp_stack[i]);
 613	}
 614
 615	if (kp.kp_free_stack[0])
 616		pr_cont(" Free path:\n");
 617
 618	for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
 619		if (!kp.kp_free_stack[i])
 620			break;
 621		pr_info("    %pS\n", kp.kp_free_stack[i]);
 622	}
 623
 624}
 625EXPORT_SYMBOL_GPL(kmem_dump_obj);
 626#endif
 627
 628#ifndef CONFIG_SLOB
 629/* Create a cache during boot when no slab services are available yet */
 630void __init create_boot_cache(struct kmem_cache *s, const char *name,
 631		unsigned int size, slab_flags_t flags,
 632		unsigned int useroffset, unsigned int usersize)
 633{
 634	int err;
 635	unsigned int align = ARCH_KMALLOC_MINALIGN;
 636
 637	s->name = name;
 638	s->size = s->object_size = size;
 639
 640	/*
 641	 * For power of two sizes, guarantee natural alignment for kmalloc
 642	 * caches, regardless of SL*B debugging options.
 643	 */
 644	if (is_power_of_2(size))
 645		align = max(align, size);
 646	s->align = calculate_alignment(flags, align, size);
 647
 648#ifdef CONFIG_HARDENED_USERCOPY
 649	s->useroffset = useroffset;
 650	s->usersize = usersize;
 651#endif
 
 652
 653	err = __kmem_cache_create(s, flags);
 654
 655	if (err)
 656		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
 657					name, size, err);
 658
 659	s->refcount = -1;	/* Exempt from merging for now */
 660}
 661
 662struct kmem_cache *__init create_kmalloc_cache(const char *name,
 663		unsigned int size, slab_flags_t flags,
 664		unsigned int useroffset, unsigned int usersize)
 665{
 666	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 667
 668	if (!s)
 669		panic("Out of memory when creating slab %s\n", name);
 670
 671	create_boot_cache(s, name, size, flags | SLAB_KMALLOC, useroffset,
 672								usersize);
 673	kasan_cache_create_kmalloc(s);
 674	list_add(&s->list, &slab_caches);
 
 675	s->refcount = 1;
 676	return s;
 677}
 678
 679struct kmem_cache *
 680kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
 681{ /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
 682EXPORT_SYMBOL(kmalloc_caches);
 683
 
 
 
 
 
 684/*
 685 * Conversion table for small slabs sizes / 8 to the index in the
 686 * kmalloc array. This is necessary for slabs < 192 since we have non power
 687 * of two cache sizes there. The size of larger slabs can be determined using
 688 * fls.
 689 */
 690static u8 size_index[24] __ro_after_init = {
 691	3,	/* 8 */
 692	4,	/* 16 */
 693	5,	/* 24 */
 694	5,	/* 32 */
 695	6,	/* 40 */
 696	6,	/* 48 */
 697	6,	/* 56 */
 698	6,	/* 64 */
 699	1,	/* 72 */
 700	1,	/* 80 */
 701	1,	/* 88 */
 702	1,	/* 96 */
 703	7,	/* 104 */
 704	7,	/* 112 */
 705	7,	/* 120 */
 706	7,	/* 128 */
 707	2,	/* 136 */
 708	2,	/* 144 */
 709	2,	/* 152 */
 710	2,	/* 160 */
 711	2,	/* 168 */
 712	2,	/* 176 */
 713	2,	/* 184 */
 714	2	/* 192 */
 715};
 716
 717static inline unsigned int size_index_elem(unsigned int bytes)
 718{
 719	return (bytes - 1) / 8;
 720}
 721
 722/*
 723 * Find the kmem_cache structure that serves a given size of
 724 * allocation
 725 */
 726struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 727{
 728	unsigned int index;
 729
 
 
 
 
 
 730	if (size <= 192) {
 731		if (!size)
 732			return ZERO_SIZE_PTR;
 733
 734		index = size_index[size_index_elem(size)];
 735	} else {
 736		if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
 737			return NULL;
 738		index = fls(size - 1);
 739	}
 740
 741	return kmalloc_caches[kmalloc_type(flags)][index];
 742}
 743
 744size_t kmalloc_size_roundup(size_t size)
 745{
 746	struct kmem_cache *c;
 747
 748	/* Short-circuit the 0 size case. */
 749	if (unlikely(size == 0))
 750		return 0;
 751	/* Short-circuit saturated "too-large" case. */
 752	if (unlikely(size == SIZE_MAX))
 753		return SIZE_MAX;
 754	/* Above the smaller buckets, size is a multiple of page size. */
 755	if (size > KMALLOC_MAX_CACHE_SIZE)
 756		return PAGE_SIZE << get_order(size);
 757
 758	/* The flags don't matter since size_index is common to all. */
 759	c = kmalloc_slab(size, GFP_KERNEL);
 760	return c ? c->object_size : 0;
 761}
 762EXPORT_SYMBOL(kmalloc_size_roundup);
 763
 764#ifdef CONFIG_ZONE_DMA
 765#define KMALLOC_DMA_NAME(sz)	.name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
 766#else
 767#define KMALLOC_DMA_NAME(sz)
 768#endif
 769
 770#ifdef CONFIG_MEMCG_KMEM
 771#define KMALLOC_CGROUP_NAME(sz)	.name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
 772#else
 773#define KMALLOC_CGROUP_NAME(sz)
 774#endif
 775
 776#ifndef CONFIG_SLUB_TINY
 777#define KMALLOC_RCL_NAME(sz)	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
 778#else
 779#define KMALLOC_RCL_NAME(sz)
 780#endif
 781
 782#define INIT_KMALLOC_INFO(__size, __short_size)			\
 783{								\
 784	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
 785	KMALLOC_RCL_NAME(__short_size)				\
 786	KMALLOC_CGROUP_NAME(__short_size)			\
 787	KMALLOC_DMA_NAME(__short_size)				\
 788	.size = __size,						\
 789}
 790
 791/*
 792 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
 793 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
 794 * kmalloc-2M.
 795 */
 796const struct kmalloc_info_struct kmalloc_info[] __initconst = {
 797	INIT_KMALLOC_INFO(0, 0),
 798	INIT_KMALLOC_INFO(96, 96),
 799	INIT_KMALLOC_INFO(192, 192),
 800	INIT_KMALLOC_INFO(8, 8),
 801	INIT_KMALLOC_INFO(16, 16),
 802	INIT_KMALLOC_INFO(32, 32),
 803	INIT_KMALLOC_INFO(64, 64),
 804	INIT_KMALLOC_INFO(128, 128),
 805	INIT_KMALLOC_INFO(256, 256),
 806	INIT_KMALLOC_INFO(512, 512),
 807	INIT_KMALLOC_INFO(1024, 1k),
 808	INIT_KMALLOC_INFO(2048, 2k),
 809	INIT_KMALLOC_INFO(4096, 4k),
 810	INIT_KMALLOC_INFO(8192, 8k),
 811	INIT_KMALLOC_INFO(16384, 16k),
 812	INIT_KMALLOC_INFO(32768, 32k),
 813	INIT_KMALLOC_INFO(65536, 64k),
 814	INIT_KMALLOC_INFO(131072, 128k),
 815	INIT_KMALLOC_INFO(262144, 256k),
 816	INIT_KMALLOC_INFO(524288, 512k),
 817	INIT_KMALLOC_INFO(1048576, 1M),
 818	INIT_KMALLOC_INFO(2097152, 2M)
 819};
 820
 821/*
 822 * Patch up the size_index table if we have strange large alignment
 823 * requirements for the kmalloc array. This is only the case for
 824 * MIPS it seems. The standard arches will not generate any code here.
 825 *
 826 * Largest permitted alignment is 256 bytes due to the way we
 827 * handle the index determination for the smaller caches.
 828 *
 829 * Make sure that nothing crazy happens if someone starts tinkering
 830 * around with ARCH_KMALLOC_MINALIGN
 831 */
 832void __init setup_kmalloc_cache_index_table(void)
 833{
 834	unsigned int i;
 835
 836	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 837		!is_power_of_2(KMALLOC_MIN_SIZE));
 838
 839	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 840		unsigned int elem = size_index_elem(i);
 841
 842		if (elem >= ARRAY_SIZE(size_index))
 843			break;
 844		size_index[elem] = KMALLOC_SHIFT_LOW;
 845	}
 846
 847	if (KMALLOC_MIN_SIZE >= 64) {
 848		/*
 849		 * The 96 byte sized cache is not used if the alignment
 850		 * is 64 byte.
 851		 */
 852		for (i = 64 + 8; i <= 96; i += 8)
 853			size_index[size_index_elem(i)] = 7;
 854
 855	}
 856
 857	if (KMALLOC_MIN_SIZE >= 128) {
 858		/*
 859		 * The 192 byte sized cache is not used if the alignment
 860		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
 861		 * instead.
 862		 */
 863		for (i = 128 + 8; i <= 192; i += 8)
 864			size_index[size_index_elem(i)] = 8;
 865	}
 866}
 867
 868static void __init
 869new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
 870{
 871	if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
 872		flags |= SLAB_RECLAIM_ACCOUNT;
 873	} else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
 874		if (mem_cgroup_kmem_disabled()) {
 875			kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
 876			return;
 877		}
 878		flags |= SLAB_ACCOUNT;
 879	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
 880		flags |= SLAB_CACHE_DMA;
 881	}
 882
 883	kmalloc_caches[type][idx] = create_kmalloc_cache(
 884					kmalloc_info[idx].name[type],
 885					kmalloc_info[idx].size, flags, 0,
 886					kmalloc_info[idx].size);
 887
 888	/*
 889	 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
 890	 * KMALLOC_NORMAL caches.
 891	 */
 892	if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
 893		kmalloc_caches[type][idx]->refcount = -1;
 894}
 895
 896/*
 897 * Create the kmalloc array. Some of the regular kmalloc arrays
 898 * may already have been created because they were needed to
 899 * enable allocations for slab creation.
 900 */
 901void __init create_kmalloc_caches(slab_flags_t flags)
 902{
 903	int i;
 904	enum kmalloc_cache_type type;
 905
 906	/*
 907	 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
 908	 */
 909	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
 910		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
 911			if (!kmalloc_caches[type][i])
 912				new_kmalloc_cache(i, type, flags);
 913
 914			/*
 915			 * Caches that are not of the two-to-the-power-of size.
 916			 * These have to be created immediately after the
 917			 * earlier power of two caches
 918			 */
 919			if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
 920					!kmalloc_caches[type][1])
 921				new_kmalloc_cache(1, type, flags);
 922			if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
 923					!kmalloc_caches[type][2])
 924				new_kmalloc_cache(2, type, flags);
 925		}
 926	}
 927
 928	/* Kmalloc array is now usable */
 929	slab_state = UP;
 930}
 931
 932void free_large_kmalloc(struct folio *folio, void *object)
 933{
 934	unsigned int order = folio_order(folio);
 935
 936	if (WARN_ON_ONCE(order == 0))
 937		pr_warn_once("object pointer: 0x%p\n", object);
 938
 939	kmemleak_free(object);
 940	kasan_kfree_large(object);
 941	kmsan_kfree_large(object);
 942
 943	mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
 944			      -(PAGE_SIZE << order));
 945	__free_pages(folio_page(folio, 0), order);
 946}
 947
 948static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
 949static __always_inline
 950void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
 951{
 952	struct kmem_cache *s;
 953	void *ret;
 954
 955	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
 956		ret = __kmalloc_large_node(size, flags, node);
 957		trace_kmalloc(caller, ret, size,
 958			      PAGE_SIZE << get_order(size), flags, node);
 959		return ret;
 960	}
 961
 962	s = kmalloc_slab(size, flags);
 963
 964	if (unlikely(ZERO_OR_NULL_PTR(s)))
 965		return s;
 966
 967	ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
 968	ret = kasan_kmalloc(s, ret, size, flags);
 969	trace_kmalloc(caller, ret, size, s->size, flags, node);
 970	return ret;
 971}
 972
 973void *__kmalloc_node(size_t size, gfp_t flags, int node)
 974{
 975	return __do_kmalloc_node(size, flags, node, _RET_IP_);
 976}
 977EXPORT_SYMBOL(__kmalloc_node);
 978
 979void *__kmalloc(size_t size, gfp_t flags)
 980{
 981	return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
 982}
 983EXPORT_SYMBOL(__kmalloc);
 984
 985void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
 986				  int node, unsigned long caller)
 987{
 988	return __do_kmalloc_node(size, flags, node, caller);
 989}
 990EXPORT_SYMBOL(__kmalloc_node_track_caller);
 991
 992/**
 993 * kfree - free previously allocated memory
 994 * @object: pointer returned by kmalloc.
 995 *
 996 * If @object is NULL, no operation is performed.
 997 *
 998 * Don't free memory not originally allocated by kmalloc()
 999 * or you will run into trouble.
1000 */
1001void kfree(const void *object)
1002{
1003	struct folio *folio;
1004	struct slab *slab;
1005	struct kmem_cache *s;
1006
1007	trace_kfree(_RET_IP_, object);
1008
1009	if (unlikely(ZERO_OR_NULL_PTR(object)))
1010		return;
1011
1012	folio = virt_to_folio(object);
1013	if (unlikely(!folio_test_slab(folio))) {
1014		free_large_kmalloc(folio, (void *)object);
1015		return;
1016	}
1017
1018	slab = folio_slab(folio);
1019	s = slab->slab_cache;
1020	__kmem_cache_free(s, (void *)object, _RET_IP_);
1021}
1022EXPORT_SYMBOL(kfree);
1023
1024/**
1025 * __ksize -- Report full size of underlying allocation
1026 * @object: pointer to the object
1027 *
1028 * This should only be used internally to query the true size of allocations.
1029 * It is not meant to be a way to discover the usable size of an allocation
1030 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1031 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1032 * and/or FORTIFY_SOURCE.
1033 *
1034 * Return: size of the actual memory used by @object in bytes
1035 */
1036size_t __ksize(const void *object)
1037{
1038	struct folio *folio;
1039
1040	if (unlikely(object == ZERO_SIZE_PTR))
1041		return 0;
1042
1043	folio = virt_to_folio(object);
1044
1045	if (unlikely(!folio_test_slab(folio))) {
1046		if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1047			return 0;
1048		if (WARN_ON(object != folio_address(folio)))
1049			return 0;
1050		return folio_size(folio);
1051	}
1052
1053#ifdef CONFIG_SLUB_DEBUG
1054	skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1055#endif
1056
1057	return slab_ksize(folio_slab(folio)->slab_cache);
1058}
1059
1060void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1061{
1062	void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1063					    size, _RET_IP_);
1064
1065	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1066
1067	ret = kasan_kmalloc(s, ret, size, gfpflags);
1068	return ret;
1069}
1070EXPORT_SYMBOL(kmalloc_trace);
1071
1072void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1073			 int node, size_t size)
1074{
1075	void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1076
1077	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1078
1079	ret = kasan_kmalloc(s, ret, size, gfpflags);
1080	return ret;
1081}
1082EXPORT_SYMBOL(kmalloc_node_trace);
1083#endif /* !CONFIG_SLOB */
1084
1085gfp_t kmalloc_fix_flags(gfp_t flags)
1086{
1087	gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1088
1089	flags &= ~GFP_SLAB_BUG_MASK;
1090	pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1091			invalid_mask, &invalid_mask, flags, &flags);
1092	dump_stack();
1093
1094	return flags;
1095}
1096
1097/*
1098 * To avoid unnecessary overhead, we pass through large allocation requests
1099 * directly to the page allocator. We use __GFP_COMP, because we will need to
1100 * know the allocation order to free the pages properly in kfree.
1101 */
1102
1103static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1104{
 
1105	struct page *page;
1106	void *ptr = NULL;
1107	unsigned int order = get_order(size);
1108
1109	if (unlikely(flags & GFP_SLAB_BUG_MASK))
1110		flags = kmalloc_fix_flags(flags);
1111
1112	flags |= __GFP_COMP;
1113	page = alloc_pages_node(node, flags, order);
1114	if (page) {
1115		ptr = page_address(page);
1116		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1117				      PAGE_SIZE << order);
1118	}
1119
1120	ptr = kasan_kmalloc_large(ptr, size, flags);
1121	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1122	kmemleak_alloc(ptr, size, 1, flags);
1123	kmsan_kmalloc_large(ptr, size, flags);
1124
1125	return ptr;
1126}
1127
1128void *kmalloc_large(size_t size, gfp_t flags)
1129{
1130	void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1131
1132	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1133		      flags, NUMA_NO_NODE);
1134	return ret;
1135}
1136EXPORT_SYMBOL(kmalloc_large);
1137
1138void *kmalloc_large_node(size_t size, gfp_t flags, int node)
 
1139{
1140	void *ret = __kmalloc_large_node(size, flags, node);
1141
1142	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1143		      flags, node);
1144	return ret;
1145}
1146EXPORT_SYMBOL(kmalloc_large_node);
 
1147
1148#ifdef CONFIG_SLAB_FREELIST_RANDOM
1149/* Randomize a generic freelist */
1150static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1151			       unsigned int count)
1152{
1153	unsigned int rand;
1154	unsigned int i;
1155
1156	for (i = 0; i < count; i++)
1157		list[i] = i;
1158
1159	/* Fisher-Yates shuffle */
1160	for (i = count - 1; i > 0; i--) {
1161		rand = prandom_u32_state(state);
1162		rand %= (i + 1);
1163		swap(list[i], list[rand]);
1164	}
1165}
1166
1167/* Create a random sequence per cache */
1168int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1169				    gfp_t gfp)
1170{
1171	struct rnd_state state;
1172
1173	if (count < 2 || cachep->random_seq)
1174		return 0;
1175
1176	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1177	if (!cachep->random_seq)
1178		return -ENOMEM;
1179
1180	/* Get best entropy at this stage of boot */
1181	prandom_seed_state(&state, get_random_long());
1182
1183	freelist_randomize(&state, cachep->random_seq, count);
1184	return 0;
1185}
1186
1187/* Destroy the per-cache random freelist sequence */
1188void cache_random_seq_destroy(struct kmem_cache *cachep)
1189{
1190	kfree(cachep->random_seq);
1191	cachep->random_seq = NULL;
1192}
1193#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1194
1195#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1196#ifdef CONFIG_SLAB
1197#define SLABINFO_RIGHTS (0600)
1198#else
1199#define SLABINFO_RIGHTS (0400)
1200#endif
1201
1202static void print_slabinfo_header(struct seq_file *m)
1203{
1204	/*
1205	 * Output format version, so at least we can change it
1206	 * without _too_ many complaints.
1207	 */
1208#ifdef CONFIG_DEBUG_SLAB
1209	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1210#else
1211	seq_puts(m, "slabinfo - version: 2.1\n");
1212#endif
1213	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1214	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1215	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1216#ifdef CONFIG_DEBUG_SLAB
1217	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1218	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1219#endif
1220	seq_putc(m, '\n');
1221}
1222
1223static void *slab_start(struct seq_file *m, loff_t *pos)
1224{
1225	mutex_lock(&slab_mutex);
1226	return seq_list_start(&slab_caches, *pos);
1227}
1228
1229static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1230{
1231	return seq_list_next(p, &slab_caches, pos);
1232}
1233
1234static void slab_stop(struct seq_file *m, void *p)
1235{
1236	mutex_unlock(&slab_mutex);
1237}
1238
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1239static void cache_show(struct kmem_cache *s, struct seq_file *m)
1240{
1241	struct slabinfo sinfo;
1242
1243	memset(&sinfo, 0, sizeof(sinfo));
1244	get_slabinfo(s, &sinfo);
1245
 
 
1246	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1247		   s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1248		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1249
1250	seq_printf(m, " : tunables %4u %4u %4u",
1251		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1252	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1253		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1254	slabinfo_show_stats(m, s);
1255	seq_putc(m, '\n');
1256}
1257
1258static int slab_show(struct seq_file *m, void *p)
1259{
1260	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1261
1262	if (p == slab_caches.next)
1263		print_slabinfo_header(m);
1264	cache_show(s, m);
1265	return 0;
1266}
1267
1268void dump_unreclaimable_slab(void)
1269{
1270	struct kmem_cache *s;
1271	struct slabinfo sinfo;
1272
1273	/*
1274	 * Here acquiring slab_mutex is risky since we don't prefer to get
1275	 * sleep in oom path. But, without mutex hold, it may introduce a
1276	 * risk of crash.
1277	 * Use mutex_trylock to protect the list traverse, dump nothing
1278	 * without acquiring the mutex.
1279	 */
1280	if (!mutex_trylock(&slab_mutex)) {
1281		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1282		return;
1283	}
1284
1285	pr_info("Unreclaimable slab info:\n");
1286	pr_info("Name                      Used          Total\n");
1287
1288	list_for_each_entry(s, &slab_caches, list) {
1289		if (s->flags & SLAB_RECLAIM_ACCOUNT)
1290			continue;
1291
1292		get_slabinfo(s, &sinfo);
1293
1294		if (sinfo.num_objs > 0)
1295			pr_info("%-17s %10luKB %10luKB\n", s->name,
1296				(sinfo.active_objs * s->size) / 1024,
1297				(sinfo.num_objs * s->size) / 1024);
1298	}
1299	mutex_unlock(&slab_mutex);
1300}
1301
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1302/*
1303 * slabinfo_op - iterator that generates /proc/slabinfo
1304 *
1305 * Output layout:
1306 * cache-name
1307 * num-active-objs
1308 * total-objs
1309 * object size
1310 * num-active-slabs
1311 * total-slabs
1312 * num-pages-per-slab
1313 * + further values on SMP and with statistics enabled
1314 */
1315static const struct seq_operations slabinfo_op = {
1316	.start = slab_start,
1317	.next = slab_next,
1318	.stop = slab_stop,
1319	.show = slab_show,
1320};
1321
1322static int slabinfo_open(struct inode *inode, struct file *file)
1323{
1324	return seq_open(file, &slabinfo_op);
1325}
1326
1327static const struct proc_ops slabinfo_proc_ops = {
1328	.proc_flags	= PROC_ENTRY_PERMANENT,
1329	.proc_open	= slabinfo_open,
1330	.proc_read	= seq_read,
1331	.proc_write	= slabinfo_write,
1332	.proc_lseek	= seq_lseek,
1333	.proc_release	= seq_release,
1334};
1335
1336static int __init slab_proc_init(void)
1337{
1338	proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
 
1339	return 0;
1340}
1341module_init(slab_proc_init);
1342
1343#endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1344
1345static __always_inline __realloc_size(2) void *
1346__do_krealloc(const void *p, size_t new_size, gfp_t flags)
1347{
1348	void *ret;
1349	size_t ks;
1350
1351	/* Check for double-free before calling ksize. */
1352	if (likely(!ZERO_OR_NULL_PTR(p))) {
1353		if (!kasan_check_byte(p))
1354			return NULL;
1355		ks = ksize(p);
1356	} else
1357		ks = 0;
1358
1359	/* If the object still fits, repoison it precisely. */
1360	if (ks >= new_size) {
1361		p = kasan_krealloc((void *)p, new_size, flags);
1362		return (void *)p;
1363	}
1364
1365	ret = kmalloc_track_caller(new_size, flags);
1366	if (ret && p) {
1367		/* Disable KASAN checks as the object's redzone is accessed. */
1368		kasan_disable_current();
1369		memcpy(ret, kasan_reset_tag(p), ks);
1370		kasan_enable_current();
1371	}
1372
1373	return ret;
1374}
1375
1376/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1377 * krealloc - reallocate memory. The contents will remain unchanged.
1378 * @p: object to reallocate memory for.
1379 * @new_size: how many bytes of memory are required.
1380 * @flags: the type of memory to allocate.
1381 *
1382 * The contents of the object pointed to are preserved up to the
1383 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1384 * If @p is %NULL, krealloc() behaves exactly like kmalloc().  If @new_size
1385 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1386 *
1387 * Return: pointer to the allocated memory or %NULL in case of error
1388 */
1389void *krealloc(const void *p, size_t new_size, gfp_t flags)
1390{
1391	void *ret;
1392
1393	if (unlikely(!new_size)) {
1394		kfree(p);
1395		return ZERO_SIZE_PTR;
1396	}
1397
1398	ret = __do_krealloc(p, new_size, flags);
1399	if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1400		kfree(p);
1401
1402	return ret;
1403}
1404EXPORT_SYMBOL(krealloc);
1405
1406/**
1407 * kfree_sensitive - Clear sensitive information in memory before freeing
1408 * @p: object to free memory of
1409 *
1410 * The memory of the object @p points to is zeroed before freed.
1411 * If @p is %NULL, kfree_sensitive() does nothing.
1412 *
1413 * Note: this function zeroes the whole allocated buffer which can be a good
1414 * deal bigger than the requested buffer size passed to kmalloc(). So be
1415 * careful when using this function in performance sensitive code.
1416 */
1417void kfree_sensitive(const void *p)
1418{
1419	size_t ks;
1420	void *mem = (void *)p;
1421
 
 
1422	ks = ksize(mem);
1423	if (ks) {
1424		kasan_unpoison_range(mem, ks);
1425		memzero_explicit(mem, ks);
1426	}
1427	kfree(mem);
1428}
1429EXPORT_SYMBOL(kfree_sensitive);
1430
1431size_t ksize(const void *objp)
1432{
1433	/*
1434	 * We need to first check that the pointer to the object is valid.
1435	 * The KASAN report printed from ksize() is more useful, then when
1436	 * it's printed later when the behaviour could be undefined due to
1437	 * a potential use-after-free or double-free.
1438	 *
1439	 * We use kasan_check_byte(), which is supported for the hardware
1440	 * tag-based KASAN mode, unlike kasan_check_read/write().
1441	 *
1442	 * If the pointed to memory is invalid, we return 0 to avoid users of
1443	 * ksize() writing to and potentially corrupting the memory region.
1444	 *
1445	 * We want to perform the check before __ksize(), to avoid potentially
1446	 * crashing in __ksize() due to accessing invalid metadata.
1447	 */
1448	if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1449		return 0;
1450
1451	return kfence_ksize(objp) ?: __ksize(objp);
1452}
1453EXPORT_SYMBOL(ksize);
1454
1455/* Tracepoints definitions. */
1456EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1457EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
 
 
1458EXPORT_TRACEPOINT_SYMBOL(kfree);
1459EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1460
1461int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1462{
1463	if (__should_failslab(s, gfpflags))
1464		return -ENOMEM;
1465	return 0;
1466}
1467ALLOW_ERROR_INJECTION(should_failslab, ERRNO);