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_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | SLAB_NOTRACK)
  41
  42 /*
  43  * Merge control. If this is set then no merging of slab caches will occur.
  44  * (Could be removed. This was introduced to pacify the merge skeptics.)
  45  */
  46 static int slab_nomerge;
  47
  48 static int __init setup_slab_nomerge(char *str)
  49 {
  50         slab_nomerge = 1;
  51         return 1;
  52 }
  53
  54 #ifdef CONFIG_SLUB
  55 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
  56 #endif
  57
  58 __setup("slab_nomerge", setup_slab_nomerge);
  59
  60 /*
  61  * Determine the size of a slab object
  62  */
  63 unsigned int kmem_cache_size(struct kmem_cache *s)
  64 {
  65         return s->object_size;
  66 }
  67 EXPORT_SYMBOL(kmem_cache_size);
  68
  69 #ifdef CONFIG_DEBUG_VM
  70 static int kmem_cache_sanity_check(const char *name, size_t size)
  71 {
  72         struct kmem_cache *s = NULL;
  73
  74         if (!name || in_interrupt() || size < sizeof(void *) ||
  75                 size > KMALLOC_MAX_SIZE) {
  76                 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
  77                 return -EINVAL;
  78         }
  79
  80         list_for_each_entry(s, &slab_caches, list) {
  81                 char tmp;
  82                 int res;
  83
  84                 /*
  85                  * This happens when the module gets unloaded and doesn't
  86                  * destroy its slab cache and no-one else reuses the vmalloc
  87                  * area of the module.  Print a warning.
  88                  */
  89                 res = probe_kernel_address(s->name, tmp);
  90                 if (res) {
  91                         pr_err("Slab cache with size %d has lost its name\n",
  92                                s->object_size);
  93                         continue;
  94                 }
  95         }
  96
  97         WARN_ON(strchr(name, ' '));     /* It confuses parsers */
  98         return 0;
  99 }
 100 #else
 101 static inline int kmem_cache_sanity_check(const char *name, size_t size)
 102 {
 103         return 0;
 104 }
 105 #endif
 106
 107 #ifdef CONFIG_MEMCG_KMEM
 108 void slab_init_memcg_params(struct kmem_cache *s)
 109 {
 110         s->memcg_params.is_root_cache = true;
 111         INIT_LIST_HEAD(&s->memcg_params.list);
 112         RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
 113 }
 114
 115 static int init_memcg_params(struct kmem_cache *s,
 116                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 117 {
 118         struct memcg_cache_array *arr;
 119
 120         if (memcg) {
 121                 s->memcg_params.is_root_cache = false;
 122                 s->memcg_params.memcg = memcg;
 123                 s->memcg_params.root_cache = root_cache;
 124                 return 0;
 125         }
 126
 127         slab_init_memcg_params(s);
 128
 129         if (!memcg_nr_cache_ids)
 130                 return 0;
 131
 132         arr = kzalloc(sizeof(struct memcg_cache_array) +
 133                       memcg_nr_cache_ids * sizeof(void *),
 134                       GFP_KERNEL);
 135         if (!arr)
 136                 return -ENOMEM;
 137
 138         RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
 139         return 0;
 140 }
 141
 142 static void destroy_memcg_params(struct kmem_cache *s)
 143 {
 144         if (is_root_cache(s))
 145                 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
 146 }
 147
 148 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
 149 {
 150         struct memcg_cache_array *old, *new;
 151
 152         if (!is_root_cache(s))
 153                 return 0;
 154
 155         new = kzalloc(sizeof(struct memcg_cache_array) +
 156                       new_array_size * sizeof(void *), GFP_KERNEL);
 157         if (!new)
 158                 return -ENOMEM;
 159
 160         old = rcu_dereference_protected(s->memcg_params.memcg_caches,
 161                                         lockdep_is_held(&slab_mutex));
 162         if (old)
 163                 memcpy(new->entries, old->entries,
 164                        memcg_nr_cache_ids * sizeof(void *));
 165
 166         rcu_assign_pointer(s->memcg_params.memcg_caches, new);
 167         if (old)
 168                 kfree_rcu(old, rcu);
 169         return 0;
 170 }
 171
 172 int memcg_update_all_caches(int num_memcgs)
 173 {
 174         struct kmem_cache *s;
 175         int ret = 0;
 176
 177         mutex_lock(&slab_mutex);
 178         list_for_each_entry(s, &slab_caches, list) {
 179                 ret = update_memcg_params(s, num_memcgs);
 180                 /*
 181                  * Instead of freeing the memory, we'll just leave the caches
 182                  * up to this point in an updated state.
 183                  */
 184                 if (ret)
 185                         break;
 186         }
 187         mutex_unlock(&slab_mutex);
 188         return ret;
 189 }
 190 #else
 191 static inline int init_memcg_params(struct kmem_cache *s,
 192                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 193 {
 194         return 0;
 195 }
 196
 197 static inline void destroy_memcg_params(struct kmem_cache *s)
 198 {
 199 }
 200 #endif /* CONFIG_MEMCG_KMEM */
 201
 202 /*
 203  * Find a mergeable slab cache
 204  */
 205 int slab_unmergeable(struct kmem_cache *s)
 206 {
 207         if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 208                 return 1;
 209
 210         if (!is_root_cache(s))
 211                 return 1;
 212
 213         if (s->ctor)
 214                 return 1;
 215
 216         /*
 217          * We may have set a slab to be unmergeable during bootstrap.
 218          */
 219         if (s->refcount < 0)
 220                 return 1;
 221
 222         return 0;
 223 }
 224
 225 struct kmem_cache *find_mergeable(size_t size, size_t align,
 226                 unsigned long flags, const char *name, void (*ctor)(void *))
 227 {
 228         struct kmem_cache *s;
 229
 230         if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
 231                 return NULL;
 232
 233         if (ctor)
 234                 return NULL;
 235
 236         size = ALIGN(size, sizeof(void *));
 237         align = calculate_alignment(flags, align, size);
 238         size = ALIGN(size, align);
 239         flags = kmem_cache_flags(size, flags, name, NULL);
 240
 241         list_for_each_entry_reverse(s, &slab_caches, list) {
 242                 if (slab_unmergeable(s))
 243                         continue;
 244
 245                 if (size > s->size)
 246                         continue;
 247
 248                 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 249                         continue;
 250                 /*
 251                  * Check if alignment is compatible.
 252                  * Courtesy of Adrian Drzewiecki
 253                  */
 254                 if ((s->size & ~(align - 1)) != s->size)
 255                         continue;
 256
 257                 if (s->size - size >= sizeof(void *))
 258                         continue;
 259
 260                 if (IS_ENABLED(CONFIG_SLAB) && align &&
 261                         (align > s->align || s->align % align))
 262                         continue;
 263
 264                 return s;
 265         }
 266         return NULL;
 267 }
 268
 269 /*
 270  * Figure out what the alignment of the objects will be given a set of
 271  * flags, a user specified alignment and the size of the objects.
 272  */
 273 unsigned long calculate_alignment(unsigned long flags,
 274                 unsigned long align, unsigned long size)
 275 {
 276         /*
 277          * If the user wants hardware cache aligned objects then follow that
 278          * suggestion if the object is sufficiently large.
 279          *
 280          * The hardware cache alignment cannot override the specified
 281          * alignment though. If that is greater then use it.
 282          */
 283         if (flags & SLAB_HWCACHE_ALIGN) {
 284                 unsigned long ralign = cache_line_size();
 285                 while (size <= ralign / 2)
 286                         ralign /= 2;
 287                 align = max(align, ralign);
 288         }
 289
 290         if (align < ARCH_SLAB_MINALIGN)
 291                 align = ARCH_SLAB_MINALIGN;
 292
 293         return ALIGN(align, sizeof(void *));
 294 }
 295
 296 static struct kmem_cache *
 297 do_kmem_cache_create(const char *name, size_t object_size, size_t size,
 298                      size_t align, unsigned long flags, void (*ctor)(void *),
 299                      struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 300 {
 301         struct kmem_cache *s;
 302         int err;
 303
 304         err = -ENOMEM;
 305         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 306         if (!s)
 307                 goto out;
 308
 309         s->name = name;
 310         s->object_size = object_size;
 311         s->size = size;
 312         s->align = align;
 313         s->ctor = ctor;
 314
 315         err = init_memcg_params(s, memcg, root_cache);
 316         if (err)
 317                 goto out_free_cache;
 318
 319         err = __kmem_cache_create(s, flags);
 320         if (err)
 321                 goto out_free_cache;
 322
 323         s->refcount = 1;
 324         list_add(&s->list, &slab_caches);
 325 out:
 326         if (err)
 327                 return ERR_PTR(err);
 328         return s;
 329
 330 out_free_cache:
 331         destroy_memcg_params(s);
 332         kmem_cache_free(kmem_cache, s);
 333         goto out;
 334 }
 335
 336 /*
 337  * kmem_cache_create - Create a cache.
 338  * @name: A string which is used in /proc/slabinfo to identify this cache.
 339  * @size: The size of objects to be created in this cache.
 340  * @align: The required alignment for the objects.
 341  * @flags: SLAB flags
 342  * @ctor: A constructor for the objects.
 343  *
 344  * Returns a ptr to the cache on success, NULL on failure.
 345  * Cannot be called within a interrupt, but can be interrupted.
 346  * The @ctor is run when new pages are allocated by the cache.
 347  *
 348  * The flags are
 349  *
 350  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 351  * to catch references to uninitialised memory.
 352  *
 353  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 354  * for buffer overruns.
 355  *
 356  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 357  * cacheline.  This can be beneficial if you're counting cycles as closely
 358  * as davem.
 359  */
 360 struct kmem_cache *
 361 kmem_cache_create(const char *name, size_t size, size_t align,
 362                   unsigned long flags, void (*ctor)(void *))
 363 {
 364         struct kmem_cache *s;
 365         const char *cache_name;
 366         int err;
 367
 368         get_online_cpus();
 369         get_online_mems();
 370         memcg_get_cache_ids();
 371
 372         mutex_lock(&slab_mutex);
 373
 374         err = kmem_cache_sanity_check(name, size);
 375         if (err) {
 376                 s = NULL;       /* suppress uninit var warning */
 377                 goto out_unlock;
 378         }
 379
 380         /*
 381          * Some allocators will constraint the set of valid flags to a subset
 382          * of all flags. We expect them to define CACHE_CREATE_MASK in this
 383          * case, and we'll just provide them with a sanitized version of the
 384          * passed flags.
 385          */
 386         flags &= CACHE_CREATE_MASK;
 387
 388         s = __kmem_cache_alias(name, size, align, flags, ctor);
 389         if (s)
 390                 goto out_unlock;
 391
 392         cache_name = kstrdup_const(name, GFP_KERNEL);
 393         if (!cache_name) {
 394                 err = -ENOMEM;
 395                 goto out_unlock;
 396         }
 397
 398         s = do_kmem_cache_create(cache_name, size, size,
 399                                  calculate_alignment(flags, align, size),
 400                                  flags, ctor, NULL, NULL);
 401         if (IS_ERR(s)) {
 402                 err = PTR_ERR(s);
 403                 kfree_const(cache_name);
 404         }
 405
 406 out_unlock:
 407         mutex_unlock(&slab_mutex);
 408
 409         memcg_put_cache_ids();
 410         put_online_mems();
 411         put_online_cpus();
 412
 413         if (err) {
 414                 if (flags & SLAB_PANIC)
 415                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 416                                 name, err);
 417                 else {
 418                         printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
 419                                 name, err);
 420                         dump_stack();
 421                 }
 422                 return NULL;
 423         }
 424         return s;
 425 }
 426 EXPORT_SYMBOL(kmem_cache_create);
 427
 428 static int do_kmem_cache_shutdown(struct kmem_cache *s,
 429                 struct list_head *release, bool *need_rcu_barrier)
 430 {
 431         if (__kmem_cache_shutdown(s) != 0) {
 432                 printk(KERN_ERR "kmem_cache_destroy %s: "
 433                        "Slab cache still has objects\n", s->name);
 434                 dump_stack();
 435                 return -EBUSY;
 436         }
 437
 438         if (s->flags & SLAB_DESTROY_BY_RCU)
 439                 *need_rcu_barrier = true;
 440
 441 #ifdef CONFIG_MEMCG_KMEM
 442         if (!is_root_cache(s))
 443                 list_del(&s->memcg_params.list);
 444 #endif
 445         list_move(&s->list, release);
 446         return 0;
 447 }
 448
 449 static void do_kmem_cache_release(struct list_head *release,
 450                                   bool need_rcu_barrier)
 451 {
 452         struct kmem_cache *s, *s2;
 453
 454         if (need_rcu_barrier)
 455                 rcu_barrier();
 456
 457         list_for_each_entry_safe(s, s2, release, list) {
 458 #ifdef SLAB_SUPPORTS_SYSFS
 459                 sysfs_slab_remove(s);
 460 #else
 461                 slab_kmem_cache_release(s);
 462 #endif
 463         }
 464 }
 465
 466 #ifdef CONFIG_MEMCG_KMEM
 467 /*
 468  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
 469  * @memcg: The memory cgroup the new cache is for.
 470  * @root_cache: The parent of the new cache.
 471  *
 472  * This function attempts to create a kmem cache that will serve allocation
 473  * requests going from @memcg to @root_cache. The new cache inherits properties
 474  * from its parent.
 475  */
 476 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
 477                              struct kmem_cache *root_cache)
 478 {
 479         static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
 480         struct cgroup_subsys_state *css = mem_cgroup_css(memcg);
 481         struct memcg_cache_array *arr;
 482         struct kmem_cache *s = NULL;
 483         char *cache_name;
 484         int idx;
 485
 486         get_online_cpus();
 487         get_online_mems();
 488
 489         mutex_lock(&slab_mutex);
 490
 491         /*
 492          * The memory cgroup could have been deactivated while the cache
 493          * creation work was pending.
 494          */
 495         if (!memcg_kmem_is_active(memcg))
 496                 goto out_unlock;
 497
 498         idx = memcg_cache_id(memcg);
 499         arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
 500                                         lockdep_is_held(&slab_mutex));
 501
 502         /*
 503          * Since per-memcg caches are created asynchronously on first
 504          * allocation (see memcg_kmem_get_cache()), several threads can try to
 505          * create the same cache, but only one of them may succeed.
 506          */
 507         if (arr->entries[idx])
 508                 goto out_unlock;
 509
 510         cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
 511         cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
 512                                css->id, memcg_name_buf);
 513         if (!cache_name)
 514                 goto out_unlock;
 515
 516         s = do_kmem_cache_create(cache_name, root_cache->object_size,
 517                                  root_cache->size, root_cache->align,
 518                                  root_cache->flags, root_cache->ctor,
 519                                  memcg, root_cache);
 520         /*
 521          * If we could not create a memcg cache, do not complain, because
 522          * that's not critical at all as we can always proceed with the root
 523          * cache.
 524          */
 525         if (IS_ERR(s)) {
 526                 kfree(cache_name);
 527                 goto out_unlock;
 528         }
 529
 530         list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
 531
 532         /*
 533          * Since readers won't lock (see cache_from_memcg_idx()), we need a
 534          * barrier here to ensure nobody will see the kmem_cache partially
 535          * initialized.
 536          */
 537         smp_wmb();
 538         arr->entries[idx] = s;
 539
 540 out_unlock:
 541         mutex_unlock(&slab_mutex);
 542
 543         put_online_mems();
 544         put_online_cpus();
 545 }
 546
 547 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
 548 {
 549         int idx;
 550         struct memcg_cache_array *arr;
 551         struct kmem_cache *s, *c;
 552
 553         idx = memcg_cache_id(memcg);
 554
 555         get_online_cpus();
 556         get_online_mems();
 557
 558         mutex_lock(&slab_mutex);
 559         list_for_each_entry(s, &slab_caches, list) {
 560                 if (!is_root_cache(s))
 561                         continue;
 562
 563                 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 564                                                 lockdep_is_held(&slab_mutex));
 565                 c = arr->entries[idx];
 566                 if (!c)
 567                         continue;
 568
 569                 __kmem_cache_shrink(c, true);
 570                 arr->entries[idx] = NULL;
 571         }
 572         mutex_unlock(&slab_mutex);
 573
 574         put_online_mems();
 575         put_online_cpus();
 576 }
 577
 578 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
 579 {
 580         LIST_HEAD(release);
 581         bool need_rcu_barrier = false;
 582         struct kmem_cache *s, *s2;
 583
 584         get_online_cpus();
 585         get_online_mems();
 586
 587         mutex_lock(&slab_mutex);
 588         list_for_each_entry_safe(s, s2, &slab_caches, list) {
 589                 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
 590                         continue;
 591                 /*
 592                  * The cgroup is about to be freed and therefore has no charges
 593                  * left. Hence, all its caches must be empty by now.
 594                  */
 595                 BUG_ON(do_kmem_cache_shutdown(s, &release, &need_rcu_barrier));
 596         }
 597         mutex_unlock(&slab_mutex);
 598
 599         put_online_mems();
 600         put_online_cpus();
 601
 602         do_kmem_cache_release(&release, need_rcu_barrier);
 603 }
 604 #endif /* CONFIG_MEMCG_KMEM */
 605
 606 void slab_kmem_cache_release(struct kmem_cache *s)
 607 {
 608         destroy_memcg_params(s);
 609         kfree_const(s->name);
 610         kmem_cache_free(kmem_cache, s);
 611 }
 612
 613 void kmem_cache_destroy(struct kmem_cache *s)
 614 {
 615         struct kmem_cache *c, *c2;
 616         LIST_HEAD(release);
 617         bool need_rcu_barrier = false;
 618         bool busy = false;
 619
 620         BUG_ON(!is_root_cache(s));
 621
 622         get_online_cpus();
 623         get_online_mems();
 624
 625         mutex_lock(&slab_mutex);
 626
 627         s->refcount--;
 628         if (s->refcount)
 629                 goto out_unlock;
 630
 631         for_each_memcg_cache_safe(c, c2, s) {
 632                 if (do_kmem_cache_shutdown(c, &release, &need_rcu_barrier))
 633                         busy = true;
 634         }
 635
 636         if (!busy)
 637                 do_kmem_cache_shutdown(s, &release, &need_rcu_barrier);
 638
 639 out_unlock:
 640         mutex_unlock(&slab_mutex);
 641
 642         put_online_mems();
 643         put_online_cpus();
 644
 645         do_kmem_cache_release(&release, need_rcu_barrier);
 646 }
 647 EXPORT_SYMBOL(kmem_cache_destroy);
 648
 649 /**
 650  * kmem_cache_shrink - Shrink a cache.
 651  * @cachep: The cache to shrink.
 652  *
 653  * Releases as many slabs as possible for a cache.
 654  * To help debugging, a zero exit status indicates all slabs were released.
 655  */
 656 int kmem_cache_shrink(struct kmem_cache *cachep)
 657 {
 658         int ret;
 659
 660         get_online_cpus();
 661         get_online_mems();
 662         ret = __kmem_cache_shrink(cachep, false);
 663         put_online_mems();
 664         put_online_cpus();
 665         return ret;
 666 }
 667 EXPORT_SYMBOL(kmem_cache_shrink);
 668
 669 int slab_is_available(void)
 670 {
 671         return slab_state >= UP;
 672 }
 673
 674 #ifndef CONFIG_SLOB
 675 /* Create a cache during boot when no slab services are available yet */
 676 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
 677                 unsigned long flags)
 678 {
 679         int err;
 680
 681         s->name = name;
 682         s->size = s->object_size = size;
 683         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 684
 685         slab_init_memcg_params(s);
 686
 687         err = __kmem_cache_create(s, flags);
 688
 689         if (err)
 690                 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
 691                                         name, size, err);
 692
 693         s->refcount = -1;       /* Exempt from merging for now */
 694 }
 695
 696 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
 697                                 unsigned long flags)
 698 {
 699         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 700
 701         if (!s)
 702                 panic("Out of memory when creating slab %s\n", name);
 703
 704         create_boot_cache(s, name, size, flags);
 705         list_add(&s->list, &slab_caches);
 706         s->refcount = 1;
 707         return s;
 708 }
 709
 710 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 711 EXPORT_SYMBOL(kmalloc_caches);
 712
 713 #ifdef CONFIG_ZONE_DMA
 714 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 715 EXPORT_SYMBOL(kmalloc_dma_caches);
 716 #endif
 717
 718 /*
 719  * Conversion table for small slabs sizes / 8 to the index in the
 720  * kmalloc array. This is necessary for slabs < 192 since we have non power
 721  * of two cache sizes there. The size of larger slabs can be determined using
 722  * fls.
 723  */
 724 static s8 size_index[24] = {
 725         3,      /* 8 */
 726         4,      /* 16 */
 727         5,      /* 24 */
 728         5,      /* 32 */
 729         6,      /* 40 */
 730         6,      /* 48 */
 731         6,      /* 56 */
 732         6,      /* 64 */
 733         1,      /* 72 */
 734         1,      /* 80 */
 735         1,      /* 88 */
 736         1,      /* 96 */
 737         7,      /* 104 */
 738         7,      /* 112 */
 739         7,      /* 120 */
 740         7,      /* 128 */
 741         2,      /* 136 */
 742         2,      /* 144 */
 743         2,      /* 152 */
 744         2,      /* 160 */
 745         2,      /* 168 */
 746         2,      /* 176 */
 747         2,      /* 184 */
 748         2       /* 192 */
 749 };
 750
 751 static inline int size_index_elem(size_t bytes)
 752 {
 753         return (bytes - 1) / 8;
 754 }
 755
 756 /*
 757  * Find the kmem_cache structure that serves a given size of
 758  * allocation
 759  */
 760 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 761 {
 762         int index;
 763
 764         if (unlikely(size > KMALLOC_MAX_SIZE)) {
 765                 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 766                 return NULL;
 767         }
 768
 769         if (size <= 192) {
 770                 if (!size)
 771                         return ZERO_SIZE_PTR;
 772
 773                 index = size_index[size_index_elem(size)];
 774         } else
 775                 index = fls(size - 1);
 776
 777 #ifdef CONFIG_ZONE_DMA
 778         if (unlikely((flags & GFP_DMA)))
 779                 return kmalloc_dma_caches[index];
 780
 781 #endif
 782         return kmalloc_caches[index];
 783 }
 784
 785 /*
 786  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
 787  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
 788  * kmalloc-67108864.
 789  */
 790 static struct {
 791         const char *name;
 792         unsigned long size;
 793 } const kmalloc_info[] __initconst = {
 794         {NULL,                      0},         {"kmalloc-96",             96},
 795         {"kmalloc-192",           192},         {"kmalloc-8",               8},
 796         {"kmalloc-16",             16},         {"kmalloc-32",             32},
 797         {"kmalloc-64",             64},         {"kmalloc-128",           128},
 798         {"kmalloc-256",           256},         {"kmalloc-512",           512},
 799         {"kmalloc-1024",         1024},         {"kmalloc-2048",         2048},
 800         {"kmalloc-4096",         4096},         {"kmalloc-8192",         8192},
 801         {"kmalloc-16384",       16384},         {"kmalloc-32768",       32768},
 802         {"kmalloc-65536",       65536},         {"kmalloc-131072",     131072},
 803         {"kmalloc-262144",     262144},         {"kmalloc-524288",     524288},
 804         {"kmalloc-1048576",   1048576},         {"kmalloc-2097152",   2097152},
 805         {"kmalloc-4194304",   4194304},         {"kmalloc-8388608",   8388608},
 806         {"kmalloc-16777216", 16777216},         {"kmalloc-33554432", 33554432},
 807         {"kmalloc-67108864", 67108864}
 808 };
 809
 810 /*
 811  * Patch up the size_index table if we have strange large alignment
 812  * requirements for the kmalloc array. This is only the case for
 813  * MIPS it seems. The standard arches will not generate any code here.
 814  *
 815  * Largest permitted alignment is 256 bytes due to the way we
 816  * handle the index determination for the smaller caches.
 817  *
 818  * Make sure that nothing crazy happens if someone starts tinkering
 819  * around with ARCH_KMALLOC_MINALIGN
 820  */
 821 void __init setup_kmalloc_cache_index_table(void)
 822 {
 823         int i;
 824
 825         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 826                 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
 827
 828         for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 829                 int elem = size_index_elem(i);
 830
 831                 if (elem >= ARRAY_SIZE(size_index))
 832                         break;
 833                 size_index[elem] = KMALLOC_SHIFT_LOW;
 834         }
 835
 836         if (KMALLOC_MIN_SIZE >= 64) {
 837                 /*
 838                  * The 96 byte size cache is not used if the alignment
 839                  * is 64 byte.
 840                  */
 841                 for (i = 64 + 8; i <= 96; i += 8)
 842                         size_index[size_index_elem(i)] = 7;
 843
 844         }
 845
 846         if (KMALLOC_MIN_SIZE >= 128) {
 847                 /*
 848                  * The 192 byte sized cache is not used if the alignment
 849                  * is 128 byte. Redirect kmalloc to use the 256 byte cache
 850                  * instead.
 851                  */
 852                 for (i = 128 + 8; i <= 192; i += 8)
 853                         size_index[size_index_elem(i)] = 8;
 854         }
 855 }
 856
 857 static void __init new_kmalloc_cache(int idx, unsigned long flags)
 858 {
 859         kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
 860                                         kmalloc_info[idx].size, flags);
 861 }
 862
 863 /*
 864  * Create the kmalloc array. Some of the regular kmalloc arrays
 865  * may already have been created because they were needed to
 866  * enable allocations for slab creation.
 867  */
 868 void __init create_kmalloc_caches(unsigned long flags)
 869 {
 870         int i;
 871
 872         for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
 873                 if (!kmalloc_caches[i])
 874                         new_kmalloc_cache(i, flags);
 875
 876                 /*
 877                  * Caches that are not of the two-to-the-power-of size.
 878                  * These have to be created immediately after the
 879                  * earlier power of two caches
 880                  */
 881                 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
 882                         new_kmalloc_cache(1, flags);
 883                 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
 884                         new_kmalloc_cache(2, flags);
 885         }
 886
 887         /* Kmalloc array is now usable */
 888         slab_state = UP;
 889
 890 #ifdef CONFIG_ZONE_DMA
 891         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
 892                 struct kmem_cache *s = kmalloc_caches[i];
 893
 894                 if (s) {
 895                         int size = kmalloc_size(i);
 896                         char *n = kasprintf(GFP_NOWAIT,
 897                                  "dma-kmalloc-%d", size);
 898
 899                         BUG_ON(!n);
 900                         kmalloc_dma_caches[i] = create_kmalloc_cache(n,
 901                                 size, SLAB_CACHE_DMA | flags);
 902                 }
 903         }
 904 #endif
 905 }
 906 #endif /* !CONFIG_SLOB */
 907
 908 /*
 909  * To avoid unnecessary overhead, we pass through large allocation requests
 910  * directly to the page allocator. We use __GFP_COMP, because we will need to
 911  * know the allocation order to free the pages properly in kfree.
 912  */
 913 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
 914 {
 915         void *ret;
 916         struct page *page;
 917
 918         flags |= __GFP_COMP;
 919         page = alloc_kmem_pages(flags, order);
 920         ret = page ? page_address(page) : NULL;
 921         kmemleak_alloc(ret, size, 1, flags);
 922         kasan_kmalloc_large(ret, size);
 923         return ret;
 924 }
 925 EXPORT_SYMBOL(kmalloc_order);
 926
 927 #ifdef CONFIG_TRACING
 928 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
 929 {
 930         void *ret = kmalloc_order(size, flags, order);
 931         trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
 932         return ret;
 933 }
 934 EXPORT_SYMBOL(kmalloc_order_trace);
 935 #endif
 936
 937 #ifdef CONFIG_SLABINFO
 938
 939 #ifdef CONFIG_SLAB
 940 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
 941 #else
 942 #define SLABINFO_RIGHTS S_IRUSR
 943 #endif
 944
 945 static void print_slabinfo_header(struct seq_file *m)
 946 {
 947         /*
 948          * Output format version, so at least we can change it
 949          * without _too_ many complaints.
 950          */
 951 #ifdef CONFIG_DEBUG_SLAB
 952         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
 953 #else
 954         seq_puts(m, "slabinfo - version: 2.1\n");
 955 #endif
 956         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> "
 957                  "<objperslab> <pagesperslab>");
 958         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
 959         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
 960 #ifdef CONFIG_DEBUG_SLAB
 961         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
 962                  "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
 963         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
 964 #endif
 965         seq_putc(m, '\n');
 966 }
 967
 968 void *slab_start(struct seq_file *m, loff_t *pos)
 969 {
 970         mutex_lock(&slab_mutex);
 971         return seq_list_start(&slab_caches, *pos);
 972 }
 973
 974 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
 975 {
 976         return seq_list_next(p, &slab_caches, pos);
 977 }
 978
 979 void slab_stop(struct seq_file *m, void *p)
 980 {
 981         mutex_unlock(&slab_mutex);
 982 }
 983
 984 static void
 985 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
 986 {
 987         struct kmem_cache *c;
 988         struct slabinfo sinfo;
 989
 990         if (!is_root_cache(s))
 991                 return;
 992
 993         for_each_memcg_cache(c, s) {
 994                 memset(&sinfo, 0, sizeof(sinfo));
 995                 get_slabinfo(c, &sinfo);
 996
 997                 info->active_slabs += sinfo.active_slabs;
 998                 info->num_slabs += sinfo.num_slabs;
 999                 info->shared_avail += sinfo.shared_avail;
1000                 info->active_objs += sinfo.active_objs;
1001                 info->num_objs += sinfo.num_objs;
1002         }
1003 }
1004
1005 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1006 {
1007         struct slabinfo sinfo;
1008
1009         memset(&sinfo, 0, sizeof(sinfo));
1010         get_slabinfo(s, &sinfo);
1011
1012         memcg_accumulate_slabinfo(s, &sinfo);
1013
1014         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1015                    cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1016                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
1017
1018         seq_printf(m, " : tunables %4u %4u %4u",
1019                    sinfo.limit, sinfo.batchcount, sinfo.shared);
1020         seq_printf(m, " : slabdata %6lu %6lu %6lu",
1021                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1022         slabinfo_show_stats(m, s);
1023         seq_putc(m, '\n');
1024 }
1025
1026 static int slab_show(struct seq_file *m, void *p)
1027 {
1028         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1029
1030         if (p == slab_caches.next)
1031                 print_slabinfo_header(m);
1032         if (is_root_cache(s))
1033                 cache_show(s, m);
1034         return 0;
1035 }
1036
1037 #ifdef CONFIG_MEMCG_KMEM
1038 int memcg_slab_show(struct seq_file *m, void *p)
1039 {
1040         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1041         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1042
1043         if (p == slab_caches.next)
1044                 print_slabinfo_header(m);
1045         if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1046                 cache_show(s, m);
1047         return 0;
1048 }
1049 #endif
1050
1051 /*
1052  * slabinfo_op - iterator that generates /proc/slabinfo
1053  *
1054  * Output layout:
1055  * cache-name
1056  * num-active-objs
1057  * total-objs
1058  * object size
1059  * num-active-slabs
1060  * total-slabs
1061  * num-pages-per-slab
1062  * + further values on SMP and with statistics enabled
1063  */
1064 static const struct seq_operations slabinfo_op = {
1065         .start = slab_start,
1066         .next = slab_next,
1067         .stop = slab_stop,
1068         .show = slab_show,
1069 };
1070
1071 static int slabinfo_open(struct inode *inode, struct file *file)
1072 {
1073         return seq_open(file, &slabinfo_op);
1074 }
1075
1076 static const struct file_operations proc_slabinfo_operations = {
1077         .open           = slabinfo_open,
1078         .read           = seq_read,
1079         .write          = slabinfo_write,
1080         .llseek         = seq_lseek,
1081         .release        = seq_release,
1082 };
1083
1084 static int __init slab_proc_init(void)
1085 {
1086         proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1087                                                 &proc_slabinfo_operations);
1088         return 0;
1089 }
1090 module_init(slab_proc_init);
1091 #endif /* CONFIG_SLABINFO */
1092
1093 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1094                                            gfp_t flags)
1095 {
1096         void *ret;
1097         size_t ks = 0;
1098
1099         if (p)
1100                 ks = ksize(p);
1101
1102         if (ks >= new_size) {
1103                 kasan_krealloc((void *)p, new_size);
1104                 return (void *)p;
1105         }
1106
1107         ret = kmalloc_track_caller(new_size, flags);
1108         if (ret && p)
1109                 memcpy(ret, p, ks);
1110
1111         return ret;
1112 }
1113
1114 /**
1115  * __krealloc - like krealloc() but don't free @p.
1116  * @p: object to reallocate memory for.
1117  * @new_size: how many bytes of memory are required.
1118  * @flags: the type of memory to allocate.
1119  *
1120  * This function is like krealloc() except it never frees the originally
1121  * allocated buffer. Use this if you don't want to free the buffer immediately
1122  * like, for example, with RCU.
1123  */
1124 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1125 {
1126         if (unlikely(!new_size))
1127                 return ZERO_SIZE_PTR;
1128
1129         return __do_krealloc(p, new_size, flags);
1130
1131 }
1132 EXPORT_SYMBOL(__krealloc);
1133
1134 /**
1135  * krealloc - reallocate memory. The contents will remain unchanged.
1136  * @p: object to reallocate memory for.
1137  * @new_size: how many bytes of memory are required.
1138  * @flags: the type of memory to allocate.
1139  *
1140  * The contents of the object pointed to are preserved up to the
1141  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1142  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1143  * %NULL pointer, the object pointed to is freed.
1144  */
1145 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1146 {
1147         void *ret;
1148
1149         if (unlikely(!new_size)) {
1150                 kfree(p);
1151                 return ZERO_SIZE_PTR;
1152         }
1153
1154         ret = __do_krealloc(p, new_size, flags);
1155         if (ret && p != ret)
1156                 kfree(p);
1157
1158         return ret;
1159 }
1160 EXPORT_SYMBOL(krealloc);
1161
1162 /**
1163  * kzfree - like kfree but zero memory
1164  * @p: object to free memory of
1165  *
1166  * The memory of the object @p points to is zeroed before freed.
1167  * If @p is %NULL, kzfree() does nothing.
1168  *
1169  * Note: this function zeroes the whole allocated buffer which can be a good
1170  * deal bigger than the requested buffer size passed to kmalloc(). So be
1171  * careful when using this function in performance sensitive code.
1172  */
1173 void kzfree(const void *p)
1174 {
1175         size_t ks;
1176         void *mem = (void *)p;
1177
1178         if (unlikely(ZERO_OR_NULL_PTR(mem)))
1179                 return;
1180         ks = ksize(mem);
1181         memset(mem, 0, ks);
1182         kfree(mem);
1183 }
1184 EXPORT_SYMBOL(kzfree);
1185
1186 /* Tracepoints definitions. */
1187 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1188 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1189 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1190 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1191 EXPORT_TRACEPOINT_SYMBOL(kfree);
1192 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);