mm/slab_common.c

   1 /*
   2  * Slab allocator functions that are independent of the allocator strategy
   3  *
   4  * (C) 2012 Christoph Lameter <cl@linux.com>
   5  */
   6 #include <linux/slab.h>
   7
   8 #include <linux/mm.h>
   9 #include <linux/poison.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/memory.h>
  12 #include <linux/compiler.h>
  13 #include <linux/module.h>
  14 #include <linux/cpu.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/seq_file.h>
  17 #include <linux/proc_fs.h>
  18 #include <asm/cacheflush.h>
  19 #include <asm/tlbflush.h>
  20 #include <asm/page.h>
  21 #include <linux/memcontrol.h>
  22
  23 #define CREATE_TRACE_POINTS
  24 #include <trace/events/kmem.h>
  25
  26 #include "slab.h"
  27
  28 enum slab_state slab_state;
  29 LIST_HEAD(slab_caches);
  30 DEFINE_MUTEX(slab_mutex);
  31 struct kmem_cache *kmem_cache;
  32
  33 /*
  34  * Set of flags that will prevent slab merging
  35  */
  36 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
  37                 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
  38                 SLAB_FAILSLAB)
  39
  40 #define SLAB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
  41                 SLAB_CACHE_DMA | SLAB_NOTRACK)
  42
  43 /*
  44  * Merge control. If this is set then no merging of slab caches will occur.
  45  * (Could be removed. This was introduced to pacify the merge skeptics.)
  46  */
  47 static int slab_nomerge;
  48
  49 static int __init setup_slab_nomerge(char *str)
  50 {
  51         slab_nomerge = 1;
  52         return 1;
  53 }
  54
  55 #ifdef CONFIG_SLUB
  56 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
  57 #endif
  58
  59 __setup("slab_nomerge", setup_slab_nomerge);
  60
  61 /*
  62  * Determine the size of a slab object
  63  */
  64 unsigned int kmem_cache_size(struct kmem_cache *s)
  65 {
  66         return s->object_size;
  67 }
  68 EXPORT_SYMBOL(kmem_cache_size);
  69
  70 #ifdef CONFIG_DEBUG_VM
  71 static int kmem_cache_sanity_check(const char *name, size_t size)
  72 {
  73         struct kmem_cache *s = NULL;
  74
  75         if (!name || in_interrupt() || size < sizeof(void *) ||
  76                 size > KMALLOC_MAX_SIZE) {
  77                 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
  78                 return -EINVAL;
  79         }
  80
  81         list_for_each_entry(s, &slab_caches, list) {
  82                 char tmp;
  83                 int res;
  84
  85                 /*
  86                  * This happens when the module gets unloaded and doesn't
  87                  * destroy its slab cache and no-one else reuses the vmalloc
  88                  * area of the module.  Print a warning.
  89                  */
  90                 res = probe_kernel_address(s->name, tmp);
  91                 if (res) {
  92                         pr_err("Slab cache with size %d has lost its name\n",
  93                                s->object_size);
  94                         continue;
  95                 }
  96
  97 #if !defined(CONFIG_SLUB)
  98                 if (!strcmp(s->name, name)) {
  99                         pr_err("%s (%s): Cache name already exists.\n",
 100                                __func__, name);
 101                         dump_stack();
 102                         s = NULL;
 103                         return -EINVAL;
 104                 }
 105 #endif
 106         }
 107
 108         WARN_ON(strchr(name, ' '));     /* It confuses parsers */
 109         return 0;
 110 }
 111 #else
 112 static inline int kmem_cache_sanity_check(const char *name, size_t size)
 113 {
 114         return 0;
 115 }
 116 #endif
 117
 118 #ifdef CONFIG_MEMCG_KMEM
 119 static int memcg_alloc_cache_params(struct mem_cgroup *memcg,
 120                 struct kmem_cache *s, struct kmem_cache *root_cache)
 121 {
 122         size_t size;
 123
 124         if (!memcg_kmem_enabled())
 125                 return 0;
 126
 127         if (!memcg) {
 128                 size = offsetof(struct memcg_cache_params, memcg_caches);
 129                 size += memcg_limited_groups_array_size * sizeof(void *);
 130         } else
 131                 size = sizeof(struct memcg_cache_params);
 132
 133         s->memcg_params = kzalloc(size, GFP_KERNEL);
 134         if (!s->memcg_params)
 135                 return -ENOMEM;
 136
 137         if (memcg) {
 138                 s->memcg_params->memcg = memcg;
 139                 s->memcg_params->root_cache = root_cache;
 140         } else
 141                 s->memcg_params->is_root_cache = true;
 142
 143         return 0;
 144 }
 145
 146 static void memcg_free_cache_params(struct kmem_cache *s)
 147 {
 148         kfree(s->memcg_params);
 149 }
 150
 151 static int memcg_update_cache_params(struct kmem_cache *s, int num_memcgs)
 152 {
 153         int size;
 154         struct memcg_cache_params *new_params, *cur_params;
 155
 156         BUG_ON(!is_root_cache(s));
 157
 158         size = offsetof(struct memcg_cache_params, memcg_caches);
 159         size += num_memcgs * sizeof(void *);
 160
 161         new_params = kzalloc(size, GFP_KERNEL);
 162         if (!new_params)
 163                 return -ENOMEM;
 164
 165         cur_params = s->memcg_params;
 166         memcpy(new_params->memcg_caches, cur_params->memcg_caches,
 167                memcg_limited_groups_array_size * sizeof(void *));
 168
 169         new_params->is_root_cache = true;
 170
 171         rcu_assign_pointer(s->memcg_params, new_params);
 172         if (cur_params)
 173                 kfree_rcu(cur_params, rcu_head);
 174
 175         return 0;
 176 }
 177
 178 int memcg_update_all_caches(int num_memcgs)
 179 {
 180         struct kmem_cache *s;
 181         int ret = 0;
 182         mutex_lock(&slab_mutex);
 183
 184         list_for_each_entry(s, &slab_caches, list) {
 185                 if (!is_root_cache(s))
 186                         continue;
 187
 188                 ret = memcg_update_cache_params(s, num_memcgs);
 189                 /*
 190                  * Instead of freeing the memory, we'll just leave the caches
 191                  * up to this point in an updated state.
 192                  */
 193                 if (ret)
 194                         goto out;
 195         }
 196
 197         memcg_update_array_size(num_memcgs);
 198 out:
 199         mutex_unlock(&slab_mutex);
 200         return ret;
 201 }
 202 #else
 203 static inline int memcg_alloc_cache_params(struct mem_cgroup *memcg,
 204                 struct kmem_cache *s, struct kmem_cache *root_cache)
 205 {
 206         return 0;
 207 }
 208
 209 static inline void memcg_free_cache_params(struct kmem_cache *s)
 210 {
 211 }
 212 #endif /* CONFIG_MEMCG_KMEM */
 213
 214 /*
 215  * Find a mergeable slab cache
 216  */
 217 int slab_unmergeable(struct kmem_cache *s)
 218 {
 219         if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 220                 return 1;
 221
 222         if (!is_root_cache(s))
 223                 return 1;
 224
 225         if (s->ctor)
 226                 return 1;
 227
 228         /*
 229          * We may have set a slab to be unmergeable during bootstrap.
 230          */
 231         if (s->refcount < 0)
 232                 return 1;
 233
 234         return 0;
 235 }
 236
 237 struct kmem_cache *find_mergeable(size_t size, size_t align,
 238                 unsigned long flags, const char *name, void (*ctor)(void *))
 239 {
 240         struct kmem_cache *s;
 241
 242         if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
 243                 return NULL;
 244
 245         if (ctor)
 246                 return NULL;
 247
 248         size = ALIGN(size, sizeof(void *));
 249         align = calculate_alignment(flags, align, size);
 250         size = ALIGN(size, align);
 251         flags = kmem_cache_flags(size, flags, name, NULL);
 252
 253         list_for_each_entry(s, &slab_caches, list) {
 254                 if (slab_unmergeable(s))
 255                         continue;
 256
 257                 if (size > s->size)
 258                         continue;
 259
 260                 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 261                         continue;
 262                 /*
 263                  * Check if alignment is compatible.
 264                  * Courtesy of Adrian Drzewiecki
 265                  */
 266                 if ((s->size & ~(align - 1)) != s->size)
 267                         continue;
 268
 269                 if (s->size - size >= sizeof(void *))
 270                         continue;
 271
 272                 return s;
 273         }
 274         return NULL;
 275 }
 276
 277 /*
 278  * Figure out what the alignment of the objects will be given a set of
 279  * flags, a user specified alignment and the size of the objects.
 280  */
 281 unsigned long calculate_alignment(unsigned long flags,
 282                 unsigned long align, unsigned long size)
 283 {
 284         /*
 285          * If the user wants hardware cache aligned objects then follow that
 286          * suggestion if the object is sufficiently large.
 287          *
 288          * The hardware cache alignment cannot override the specified
 289          * alignment though. If that is greater then use it.
 290          */
 291         if (flags & SLAB_HWCACHE_ALIGN) {
 292                 unsigned long ralign = cache_line_size();
 293                 while (size <= ralign / 2)
 294                         ralign /= 2;
 295                 align = max(align, ralign);
 296         }
 297
 298         if (align < ARCH_SLAB_MINALIGN)
 299                 align = ARCH_SLAB_MINALIGN;
 300
 301         return ALIGN(align, sizeof(void *));
 302 }
 303
 304 static struct kmem_cache *
 305 do_kmem_cache_create(char *name, size_t object_size, size_t size, size_t align,
 306                      unsigned long flags, void (*ctor)(void *),
 307                      struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 308 {
 309         struct kmem_cache *s;
 310         int err;
 311
 312         err = -ENOMEM;
 313         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 314         if (!s)
 315                 goto out;
 316
 317         s->name = name;
 318         s->object_size = object_size;
 319         s->size = size;
 320         s->align = align;
 321         s->ctor = ctor;
 322
 323         err = memcg_alloc_cache_params(memcg, s, root_cache);
 324         if (err)
 325                 goto out_free_cache;
 326
 327         err = __kmem_cache_create(s, flags);
 328         if (err)
 329                 goto out_free_cache;
 330
 331         s->refcount = 1;
 332         list_add(&s->list, &slab_caches);
 333 out:
 334         if (err)
 335                 return ERR_PTR(err);
 336         return s;
 337
 338 out_free_cache:
 339         memcg_free_cache_params(s);
 340         kfree(s);
 341         goto out;
 342 }
 343
 344 /*
 345  * kmem_cache_create - Create a cache.
 346  * @name: A string which is used in /proc/slabinfo to identify this cache.
 347  * @size: The size of objects to be created in this cache.
 348  * @align: The required alignment for the objects.
 349  * @flags: SLAB flags
 350  * @ctor: A constructor for the objects.
 351  *
 352  * Returns a ptr to the cache on success, NULL on failure.
 353  * Cannot be called within a interrupt, but can be interrupted.
 354  * The @ctor is run when new pages are allocated by the cache.
 355  *
 356  * The flags are
 357  *
 358  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 359  * to catch references to uninitialised memory.
 360  *
 361  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 362  * for buffer overruns.
 363  *
 364  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 365  * cacheline.  This can be beneficial if you're counting cycles as closely
 366  * as davem.
 367  */
 368 struct kmem_cache *
 369 kmem_cache_create(const char *name, size_t size, size_t align,
 370                   unsigned long flags, void (*ctor)(void *))
 371 {
 372         struct kmem_cache *s;
 373         char *cache_name;
 374         int err;
 375
 376         get_online_cpus();
 377         get_online_mems();
 378
 379         mutex_lock(&slab_mutex);
 380
 381         err = kmem_cache_sanity_check(name, size);
 382         if (err) {
 383                 s = NULL;       /* suppress uninit var warning */
 384                 goto out_unlock;
 385         }
 386
 387         /*
 388          * Some allocators will constraint the set of valid flags to a subset
 389          * of all flags. We expect them to define CACHE_CREATE_MASK in this
 390          * case, and we'll just provide them with a sanitized version of the
 391          * passed flags.
 392          */
 393         flags &= CACHE_CREATE_MASK;
 394
 395         s = __kmem_cache_alias(name, size, align, flags, ctor);
 396         if (s)
 397                 goto out_unlock;
 398
 399         cache_name = kstrdup(name, GFP_KERNEL);
 400         if (!cache_name) {
 401                 err = -ENOMEM;
 402                 goto out_unlock;
 403         }
 404
 405         s = do_kmem_cache_create(cache_name, size, size,
 406                                  calculate_alignment(flags, align, size),
 407                                  flags, ctor, NULL, NULL);
 408         if (IS_ERR(s)) {
 409                 err = PTR_ERR(s);
 410                 kfree(cache_name);
 411         }
 412
 413 out_unlock:
 414         mutex_unlock(&slab_mutex);
 415
 416         put_online_mems();
 417         put_online_cpus();
 418
 419         if (err) {
 420                 if (flags & SLAB_PANIC)
 421                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 422                                 name, err);
 423                 else {
 424                         printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
 425                                 name, err);
 426                         dump_stack();
 427                 }
 428                 return NULL;
 429         }
 430         return s;
 431 }
 432 EXPORT_SYMBOL(kmem_cache_create);
 433
 434 #ifdef CONFIG_MEMCG_KMEM
 435 /*
 436  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
 437  * @memcg: The memory cgroup the new cache is for.
 438  * @root_cache: The parent of the new cache.
 439  * @memcg_name: The name of the memory cgroup (used for naming the new cache).
 440  *
 441  * This function attempts to create a kmem cache that will serve allocation
 442  * requests going from @memcg to @root_cache. The new cache inherits properties
 443  * from its parent.
 444  */
 445 struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
 446                                            struct kmem_cache *root_cache,
 447                                            const char *memcg_name)
 448 {
 449         struct kmem_cache *s = NULL;
 450         char *cache_name;
 451
 452         get_online_cpus();
 453         get_online_mems();
 454
 455         mutex_lock(&slab_mutex);
 456
 457         cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
 458                                memcg_cache_id(memcg), memcg_name);
 459         if (!cache_name)
 460                 goto out_unlock;
 461
 462         s = do_kmem_cache_create(cache_name, root_cache->object_size,
 463                                  root_cache->size, root_cache->align,
 464                                  root_cache->flags, root_cache->ctor,
 465                                  memcg, root_cache);
 466         if (IS_ERR(s)) {
 467                 kfree(cache_name);
 468                 s = NULL;
 469         }
 470
 471 out_unlock:
 472         mutex_unlock(&slab_mutex);
 473
 474         put_online_mems();
 475         put_online_cpus();
 476
 477         return s;
 478 }
 479
 480 static int memcg_cleanup_cache_params(struct kmem_cache *s)
 481 {
 482         int rc;
 483
 484         if (!s->memcg_params ||
 485             !s->memcg_params->is_root_cache)
 486                 return 0;
 487
 488         mutex_unlock(&slab_mutex);
 489         rc = __memcg_cleanup_cache_params(s);
 490         mutex_lock(&slab_mutex);
 491
 492         return rc;
 493 }
 494 #else
 495 static int memcg_cleanup_cache_params(struct kmem_cache *s)
 496 {
 497         return 0;
 498 }
 499 #endif /* CONFIG_MEMCG_KMEM */
 500
 501 void slab_kmem_cache_release(struct kmem_cache *s)
 502 {
 503         kfree(s->name);
 504         kmem_cache_free(kmem_cache, s);
 505 }
 506
 507 void kmem_cache_destroy(struct kmem_cache *s)
 508 {
 509         get_online_cpus();
 510         get_online_mems();
 511
 512         mutex_lock(&slab_mutex);
 513
 514         s->refcount--;
 515         if (s->refcount)
 516                 goto out_unlock;
 517
 518         if (memcg_cleanup_cache_params(s) != 0)
 519                 goto out_unlock;
 520
 521         if (__kmem_cache_shutdown(s) != 0) {
 522                 printk(KERN_ERR "kmem_cache_destroy %s: "
 523                        "Slab cache still has objects\n", s->name);
 524                 dump_stack();
 525                 goto out_unlock;
 526         }
 527
 528         list_del(&s->list);
 529
 530         mutex_unlock(&slab_mutex);
 531         if (s->flags & SLAB_DESTROY_BY_RCU)
 532                 rcu_barrier();
 533
 534         memcg_free_cache_params(s);
 535 #ifdef SLAB_SUPPORTS_SYSFS
 536         sysfs_slab_remove(s);
 537 #else
 538         slab_kmem_cache_release(s);
 539 #endif
 540         goto out;
 541
 542 out_unlock:
 543         mutex_unlock(&slab_mutex);
 544 out:
 545         put_online_mems();
 546         put_online_cpus();
 547 }
 548 EXPORT_SYMBOL(kmem_cache_destroy);
 549
 550 /**
 551  * kmem_cache_shrink - Shrink a cache.
 552  * @cachep: The cache to shrink.
 553  *
 554  * Releases as many slabs as possible for a cache.
 555  * To help debugging, a zero exit status indicates all slabs were released.
 556  */
 557 int kmem_cache_shrink(struct kmem_cache *cachep)
 558 {
 559         int ret;
 560
 561         get_online_cpus();
 562         get_online_mems();
 563         ret = __kmem_cache_shrink(cachep);
 564         put_online_mems();
 565         put_online_cpus();
 566         return ret;
 567 }
 568 EXPORT_SYMBOL(kmem_cache_shrink);
 569
 570 int slab_is_available(void)
 571 {
 572         return slab_state >= UP;
 573 }
 574
 575 #ifndef CONFIG_SLOB
 576 /* Create a cache during boot when no slab services are available yet */
 577 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
 578                 unsigned long flags)
 579 {
 580         int err;
 581
 582         s->name = name;
 583         s->size = s->object_size = size;
 584         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 585         err = __kmem_cache_create(s, flags);
 586
 587         if (err)
 588                 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
 589                                         name, size, err);
 590
 591         s->refcount = -1;       /* Exempt from merging for now */
 592 }
 593
 594 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
 595                                 unsigned long flags)
 596 {
 597         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 598
 599         if (!s)
 600                 panic("Out of memory when creating slab %s\n", name);
 601
 602         create_boot_cache(s, name, size, flags);
 603         list_add(&s->list, &slab_caches);
 604         s->refcount = 1;
 605         return s;
 606 }
 607
 608 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 609 EXPORT_SYMBOL(kmalloc_caches);
 610
 611 #ifdef CONFIG_ZONE_DMA
 612 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 613 EXPORT_SYMBOL(kmalloc_dma_caches);
 614 #endif
 615
 616 /*
 617  * Conversion table for small slabs sizes / 8 to the index in the
 618  * kmalloc array. This is necessary for slabs < 192 since we have non power
 619  * of two cache sizes there. The size of larger slabs can be determined using
 620  * fls.
 621  */
 622 static s8 size_index[24] = {
 623         3,      /* 8 */
 624         4,      /* 16 */
 625         5,      /* 24 */
 626         5,      /* 32 */
 627         6,      /* 40 */
 628         6,      /* 48 */
 629         6,      /* 56 */
 630         6,      /* 64 */
 631         1,      /* 72 */
 632         1,      /* 80 */
 633         1,      /* 88 */
 634         1,      /* 96 */
 635         7,      /* 104 */
 636         7,      /* 112 */
 637         7,      /* 120 */
 638         7,      /* 128 */
 639         2,      /* 136 */
 640         2,      /* 144 */
 641         2,      /* 152 */
 642         2,      /* 160 */
 643         2,      /* 168 */
 644         2,      /* 176 */
 645         2,      /* 184 */
 646         2       /* 192 */
 647 };
 648
 649 static inline int size_index_elem(size_t bytes)
 650 {
 651         return (bytes - 1) / 8;
 652 }
 653
 654 /*
 655  * Find the kmem_cache structure that serves a given size of
 656  * allocation
 657  */
 658 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 659 {
 660         int index;
 661
 662         if (unlikely(size > KMALLOC_MAX_SIZE)) {
 663                 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 664                 return NULL;
 665         }
 666
 667         if (size <= 192) {
 668                 if (!size)
 669                         return ZERO_SIZE_PTR;
 670
 671                 index = size_index[size_index_elem(size)];
 672         } else
 673                 index = fls(size - 1);
 674
 675 #ifdef CONFIG_ZONE_DMA
 676         if (unlikely((flags & GFP_DMA)))
 677                 return kmalloc_dma_caches[index];
 678
 679 #endif
 680         return kmalloc_caches[index];
 681 }
 682
 683 /*
 684  * Create the kmalloc array. Some of the regular kmalloc arrays
 685  * may already have been created because they were needed to
 686  * enable allocations for slab creation.
 687  */
 688 void __init create_kmalloc_caches(unsigned long flags)
 689 {
 690         int i;
 691
 692         /*
 693          * Patch up the size_index table if we have strange large alignment
 694          * requirements for the kmalloc array. This is only the case for
 695          * MIPS it seems. The standard arches will not generate any code here.
 696          *
 697          * Largest permitted alignment is 256 bytes due to the way we
 698          * handle the index determination for the smaller caches.
 699          *
 700          * Make sure that nothing crazy happens if someone starts tinkering
 701          * around with ARCH_KMALLOC_MINALIGN
 702          */
 703         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 704                 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
 705
 706         for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 707                 int elem = size_index_elem(i);
 708
 709                 if (elem >= ARRAY_SIZE(size_index))
 710                         break;
 711                 size_index[elem] = KMALLOC_SHIFT_LOW;
 712         }
 713
 714         if (KMALLOC_MIN_SIZE >= 64) {
 715                 /*
 716                  * The 96 byte size cache is not used if the alignment
 717                  * is 64 byte.
 718                  */
 719                 for (i = 64 + 8; i <= 96; i += 8)
 720                         size_index[size_index_elem(i)] = 7;
 721
 722         }
 723
 724         if (KMALLOC_MIN_SIZE >= 128) {
 725                 /*
 726                  * The 192 byte sized cache is not used if the alignment
 727                  * is 128 byte. Redirect kmalloc to use the 256 byte cache
 728                  * instead.
 729                  */
 730                 for (i = 128 + 8; i <= 192; i += 8)
 731                         size_index[size_index_elem(i)] = 8;
 732         }
 733         for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
 734                 if (!kmalloc_caches[i]) {
 735                         kmalloc_caches[i] = create_kmalloc_cache(NULL,
 736                                                         1 << i, flags);
 737                 }
 738
 739                 /*
 740                  * Caches that are not of the two-to-the-power-of size.
 741                  * These have to be created immediately after the
 742                  * earlier power of two caches
 743                  */
 744                 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
 745                         kmalloc_caches[1] = create_kmalloc_cache(NULL, 96, flags);
 746
 747                 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
 748                         kmalloc_caches[2] = create_kmalloc_cache(NULL, 192, flags);
 749         }
 750
 751         /* Kmalloc array is now usable */
 752         slab_state = UP;
 753
 754         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
 755                 struct kmem_cache *s = kmalloc_caches[i];
 756                 char *n;
 757
 758                 if (s) {
 759                         n = kasprintf(GFP_NOWAIT, "kmalloc-%d", kmalloc_size(i));
 760
 761                         BUG_ON(!n);
 762                         s->name = n;
 763                 }
 764         }
 765
 766 #ifdef CONFIG_ZONE_DMA
 767         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
 768                 struct kmem_cache *s = kmalloc_caches[i];
 769
 770                 if (s) {
 771                         int size = kmalloc_size(i);
 772                         char *n = kasprintf(GFP_NOWAIT,
 773                                  "dma-kmalloc-%d", size);
 774
 775                         BUG_ON(!n);
 776                         kmalloc_dma_caches[i] = create_kmalloc_cache(n,
 777                                 size, SLAB_CACHE_DMA | flags);
 778                 }
 779         }
 780 #endif
 781 }
 782 #endif /* !CONFIG_SLOB */
 783
 784 /*
 785  * To avoid unnecessary overhead, we pass through large allocation requests
 786  * directly to the page allocator. We use __GFP_COMP, because we will need to
 787  * know the allocation order to free the pages properly in kfree.
 788  */
 789 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
 790 {
 791         void *ret;
 792         struct page *page;
 793
 794         flags |= __GFP_COMP;
 795         page = alloc_kmem_pages(flags, order);
 796         ret = page ? page_address(page) : NULL;
 797         kmemleak_alloc(ret, size, 1, flags);
 798         return ret;
 799 }
 800 EXPORT_SYMBOL(kmalloc_order);
 801
 802 #ifdef CONFIG_TRACING
 803 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
 804 {
 805         void *ret = kmalloc_order(size, flags, order);
 806         trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
 807         return ret;
 808 }
 809 EXPORT_SYMBOL(kmalloc_order_trace);
 810 #endif
 811
 812 #ifdef CONFIG_SLABINFO
 813
 814 #ifdef CONFIG_SLAB
 815 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
 816 #else
 817 #define SLABINFO_RIGHTS S_IRUSR
 818 #endif
 819
 820 void print_slabinfo_header(struct seq_file *m)
 821 {
 822         /*
 823          * Output format version, so at least we can change it
 824          * without _too_ many complaints.
 825          */
 826 #ifdef CONFIG_DEBUG_SLAB
 827         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
 828 #else
 829         seq_puts(m, "slabinfo - version: 2.1\n");
 830 #endif
 831         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> "
 832                  "<objperslab> <pagesperslab>");
 833         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
 834         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
 835 #ifdef CONFIG_DEBUG_SLAB
 836         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
 837                  "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
 838         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
 839 #endif
 840         seq_putc(m, '\n');
 841 }
 842
 843 static void *s_start(struct seq_file *m, loff_t *pos)
 844 {
 845         loff_t n = *pos;
 846
 847         mutex_lock(&slab_mutex);
 848         if (!n)
 849                 print_slabinfo_header(m);
 850
 851         return seq_list_start(&slab_caches, *pos);
 852 }
 853
 854 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
 855 {
 856         return seq_list_next(p, &slab_caches, pos);
 857 }
 858
 859 void slab_stop(struct seq_file *m, void *p)
 860 {
 861         mutex_unlock(&slab_mutex);
 862 }
 863
 864 static void
 865 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
 866 {
 867         struct kmem_cache *c;
 868         struct slabinfo sinfo;
 869         int i;
 870
 871         if (!is_root_cache(s))
 872                 return;
 873
 874         for_each_memcg_cache_index(i) {
 875                 c = cache_from_memcg_idx(s, i);
 876                 if (!c)
 877                         continue;
 878
 879                 memset(&sinfo, 0, sizeof(sinfo));
 880                 get_slabinfo(c, &sinfo);
 881
 882                 info->active_slabs += sinfo.active_slabs;
 883                 info->num_slabs += sinfo.num_slabs;
 884                 info->shared_avail += sinfo.shared_avail;
 885                 info->active_objs += sinfo.active_objs;
 886                 info->num_objs += sinfo.num_objs;
 887         }
 888 }
 889
 890 int cache_show(struct kmem_cache *s, struct seq_file *m)
 891 {
 892         struct slabinfo sinfo;
 893
 894         memset(&sinfo, 0, sizeof(sinfo));
 895         get_slabinfo(s, &sinfo);
 896
 897         memcg_accumulate_slabinfo(s, &sinfo);
 898
 899         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
 900                    cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
 901                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
 902
 903         seq_printf(m, " : tunables %4u %4u %4u",
 904                    sinfo.limit, sinfo.batchcount, sinfo.shared);
 905         seq_printf(m, " : slabdata %6lu %6lu %6lu",
 906                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
 907         slabinfo_show_stats(m, s);
 908         seq_putc(m, '\n');
 909         return 0;
 910 }
 911
 912 static int s_show(struct seq_file *m, void *p)
 913 {
 914         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
 915
 916         if (!is_root_cache(s))
 917                 return 0;
 918         return cache_show(s, m);
 919 }
 920
 921 /*
 922  * slabinfo_op - iterator that generates /proc/slabinfo
 923  *
 924  * Output layout:
 925  * cache-name
 926  * num-active-objs
 927  * total-objs
 928  * object size
 929  * num-active-slabs
 930  * total-slabs
 931  * num-pages-per-slab
 932  * + further values on SMP and with statistics enabled
 933  */
 934 static const struct seq_operations slabinfo_op = {
 935         .start = s_start,
 936         .next = slab_next,
 937         .stop = slab_stop,
 938         .show = s_show,
 939 };
 940
 941 static int slabinfo_open(struct inode *inode, struct file *file)
 942 {
 943         return seq_open(file, &slabinfo_op);
 944 }
 945
 946 static const struct file_operations proc_slabinfo_operations = {
 947         .open           = slabinfo_open,
 948         .read           = seq_read,
 949         .write          = slabinfo_write,
 950         .llseek         = seq_lseek,
 951         .release        = seq_release,
 952 };
 953
 954 static int __init slab_proc_init(void)
 955 {
 956         proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
 957                                                 &proc_slabinfo_operations);
 958         return 0;
 959 }
 960 module_init(slab_proc_init);
 961 #endif /* CONFIG_SLABINFO */
 962
 963 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
 964                                            gfp_t flags)
 965 {
 966         void *ret;
 967         size_t ks = 0;
 968
 969         if (p)
 970                 ks = ksize(p);
 971
 972         if (ks >= new_size)
 973                 return (void *)p;
 974
 975         ret = kmalloc_track_caller(new_size, flags);
 976         if (ret && p)
 977                 memcpy(ret, p, ks);
 978
 979         return ret;
 980 }
 981
 982 /**
 983  * __krealloc - like krealloc() but don't free @p.
 984  * @p: object to reallocate memory for.
 985  * @new_size: how many bytes of memory are required.
 986  * @flags: the type of memory to allocate.
 987  *
 988  * This function is like krealloc() except it never frees the originally
 989  * allocated buffer. Use this if you don't want to free the buffer immediately
 990  * like, for example, with RCU.
 991  */
 992 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
 993 {
 994         if (unlikely(!new_size))
 995                 return ZERO_SIZE_PTR;
 996
 997         return __do_krealloc(p, new_size, flags);
 998
 999 }
1000 EXPORT_SYMBOL(__krealloc);
1001
1002 /**
1003  * krealloc - reallocate memory. The contents will remain unchanged.
1004  * @p: object to reallocate memory for.
1005  * @new_size: how many bytes of memory are required.
1006  * @flags: the type of memory to allocate.
1007  *
1008  * The contents of the object pointed to are preserved up to the
1009  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1010  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1011  * %NULL pointer, the object pointed to is freed.
1012  */
1013 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1014 {
1015         void *ret;
1016
1017         if (unlikely(!new_size)) {
1018                 kfree(p);
1019                 return ZERO_SIZE_PTR;
1020         }
1021
1022         ret = __do_krealloc(p, new_size, flags);
1023         if (ret && p != ret)
1024                 kfree(p);
1025
1026         return ret;
1027 }
1028 EXPORT_SYMBOL(krealloc);
1029
1030 /**
1031  * kzfree - like kfree but zero memory
1032  * @p: object to free memory of
1033  *
1034  * The memory of the object @p points to is zeroed before freed.
1035  * If @p is %NULL, kzfree() does nothing.
1036  *
1037  * Note: this function zeroes the whole allocated buffer which can be a good
1038  * deal bigger than the requested buffer size passed to kmalloc(). So be
1039  * careful when using this function in performance sensitive code.
1040  */
1041 void kzfree(const void *p)
1042 {
1043         size_t ks;
1044         void *mem = (void *)p;
1045
1046         if (unlikely(ZERO_OR_NULL_PTR(mem)))
1047                 return;
1048         ks = ksize(mem);
1049         memset(mem, 0, ks);
1050         kfree(mem);
1051 }
1052 EXPORT_SYMBOL(kzfree);
1053
1054 /* Tracepoints definitions. */
1055 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1056 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1057 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1058 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1059 EXPORT_TRACEPOINT_SYMBOL(kfree);
1060 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);