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
  98         WARN_ON(strchr(name, ' '));     /* It confuses parsers */
  99         return 0;
 100 }
 101 #else
 102 static inline int kmem_cache_sanity_check(const char *name, size_t size)
 103 {
 104         return 0;
 105 }
 106 #endif
 107
 108 #ifdef CONFIG_MEMCG_KMEM
 109 void slab_init_memcg_params(struct kmem_cache *s)
 110 {
 111         s->memcg_params.is_root_cache = true;
 112         INIT_LIST_HEAD(&s->memcg_params.list);
 113         RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
 114 }
 115
 116 static int init_memcg_params(struct kmem_cache *s,
 117                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 118 {
 119         struct memcg_cache_array *arr;
 120
 121         if (memcg) {
 122                 s->memcg_params.is_root_cache = false;
 123                 s->memcg_params.memcg = memcg;
 124                 s->memcg_params.root_cache = root_cache;
 125                 return 0;
 126         }
 127
 128         slab_init_memcg_params(s);
 129
 130         if (!memcg_nr_cache_ids)
 131                 return 0;
 132
 133         arr = kzalloc(sizeof(struct memcg_cache_array) +
 134                       memcg_nr_cache_ids * sizeof(void *),
 135                       GFP_KERNEL);
 136         if (!arr)
 137                 return -ENOMEM;
 138
 139         RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
 140         return 0;
 141 }
 142
 143 static void destroy_memcg_params(struct kmem_cache *s)
 144 {
 145         if (is_root_cache(s))
 146                 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
 147 }
 148
 149 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
 150 {
 151         struct memcg_cache_array *old, *new;
 152
 153         if (!is_root_cache(s))
 154                 return 0;
 155
 156         new = kzalloc(sizeof(struct memcg_cache_array) +
 157                       new_array_size * sizeof(void *), GFP_KERNEL);
 158         if (!new)
 159                 return -ENOMEM;
 160
 161         old = rcu_dereference_protected(s->memcg_params.memcg_caches,
 162                                         lockdep_is_held(&slab_mutex));
 163         if (old)
 164                 memcpy(new->entries, old->entries,
 165                        memcg_nr_cache_ids * sizeof(void *));
 166
 167         rcu_assign_pointer(s->memcg_params.memcg_caches, new);
 168         if (old)
 169                 kfree_rcu(old, rcu);
 170         return 0;
 171 }
 172
 173 int memcg_update_all_caches(int num_memcgs)
 174 {
 175         struct kmem_cache *s;
 176         int ret = 0;
 177
 178         mutex_lock(&slab_mutex);
 179         list_for_each_entry(s, &slab_caches, list) {
 180                 ret = update_memcg_params(s, num_memcgs);
 181                 /*
 182                  * Instead of freeing the memory, we'll just leave the caches
 183                  * up to this point in an updated state.
 184                  */
 185                 if (ret)
 186                         break;
 187         }
 188         mutex_unlock(&slab_mutex);
 189         return ret;
 190 }
 191 #else
 192 static inline int init_memcg_params(struct kmem_cache *s,
 193                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 194 {
 195         return 0;
 196 }
 197
 198 static inline void destroy_memcg_params(struct kmem_cache *s)
 199 {
 200 }
 201 #endif /* CONFIG_MEMCG_KMEM */
 202
 203 /*
 204  * Find a mergeable slab cache
 205  */
 206 int slab_unmergeable(struct kmem_cache *s)
 207 {
 208         if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 209                 return 1;
 210
 211         if (!is_root_cache(s))
 212                 return 1;
 213
 214         if (s->ctor)
 215                 return 1;
 216
 217         /*
 218          * We may have set a slab to be unmergeable during bootstrap.
 219          */
 220         if (s->refcount < 0)
 221                 return 1;
 222
 223         return 0;
 224 }
 225
 226 struct kmem_cache *find_mergeable(size_t size, size_t align,
 227                 unsigned long flags, const char *name, void (*ctor)(void *))
 228 {
 229         struct kmem_cache *s;
 230
 231         if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
 232                 return NULL;
 233
 234         if (ctor)
 235                 return NULL;
 236
 237         size = ALIGN(size, sizeof(void *));
 238         align = calculate_alignment(flags, align, size);
 239         size = ALIGN(size, align);
 240         flags = kmem_cache_flags(size, flags, name, NULL);
 241
 242         list_for_each_entry_reverse(s, &slab_caches, list) {
 243                 if (slab_unmergeable(s))
 244                         continue;
 245
 246                 if (size > s->size)
 247                         continue;
 248
 249                 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 250                         continue;
 251                 /*
 252                  * Check if alignment is compatible.
 253                  * Courtesy of Adrian Drzewiecki
 254                  */
 255                 if ((s->size & ~(align - 1)) != s->size)
 256                         continue;
 257
 258                 if (s->size - size >= sizeof(void *))
 259                         continue;
 260
 261                 if (IS_ENABLED(CONFIG_SLAB) && align &&
 262                         (align > s->align || s->align % align))
 263                         continue;
 264
 265                 return s;
 266         }
 267         return NULL;
 268 }
 269
 270 /*
 271  * Figure out what the alignment of the objects will be given a set of
 272  * flags, a user specified alignment and the size of the objects.
 273  */
 274 unsigned long calculate_alignment(unsigned long flags,
 275                 unsigned long align, unsigned long size)
 276 {
 277         /*
 278          * If the user wants hardware cache aligned objects then follow that
 279          * suggestion if the object is sufficiently large.
 280          *
 281          * The hardware cache alignment cannot override the specified
 282          * alignment though. If that is greater then use it.
 283          */
 284         if (flags & SLAB_HWCACHE_ALIGN) {
 285                 unsigned long ralign = cache_line_size();
 286                 while (size <= ralign / 2)
 287                         ralign /= 2;
 288                 align = max(align, ralign);
 289         }
 290
 291         if (align < ARCH_SLAB_MINALIGN)
 292                 align = ARCH_SLAB_MINALIGN;
 293
 294         return ALIGN(align, sizeof(void *));
 295 }
 296
 297 static struct kmem_cache *
 298 do_kmem_cache_create(const char *name, size_t object_size, size_t size,
 299                      size_t align, unsigned long flags, void (*ctor)(void *),
 300                      struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 301 {
 302         struct kmem_cache *s;
 303         int err;
 304
 305         err = -ENOMEM;
 306         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 307         if (!s)
 308                 goto out;
 309
 310         s->name = name;
 311         s->object_size = object_size;
 312         s->size = size;
 313         s->align = align;
 314         s->ctor = ctor;
 315
 316         err = init_memcg_params(s, memcg, root_cache);
 317         if (err)
 318                 goto out_free_cache;
 319
 320         err = __kmem_cache_create(s, flags);
 321         if (err)
 322                 goto out_free_cache;
 323
 324         s->refcount = 1;
 325         list_add(&s->list, &slab_caches);
 326 out:
 327         if (err)
 328                 return ERR_PTR(err);
 329         return s;
 330
 331 out_free_cache:
 332         destroy_memcg_params(s);
 333         kmem_cache_free(kmem_cache, s);
 334         goto out;
 335 }
 336
 337 /*
 338  * kmem_cache_create - Create a cache.
 339  * @name: A string which is used in /proc/slabinfo to identify this cache.
 340  * @size: The size of objects to be created in this cache.
 341  * @align: The required alignment for the objects.
 342  * @flags: SLAB flags
 343  * @ctor: A constructor for the objects.
 344  *
 345  * Returns a ptr to the cache on success, NULL on failure.
 346  * Cannot be called within a interrupt, but can be interrupted.
 347  * The @ctor is run when new pages are allocated by the cache.
 348  *
 349  * The flags are
 350  *
 351  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 352  * to catch references to uninitialised memory.
 353  *
 354  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 355  * for buffer overruns.
 356  *
 357  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 358  * cacheline.  This can be beneficial if you're counting cycles as closely
 359  * as davem.
 360  */
 361 struct kmem_cache *
 362 kmem_cache_create(const char *name, size_t size, size_t align,
 363                   unsigned long flags, void (*ctor)(void *))
 364 {
 365         struct kmem_cache *s;
 366         const char *cache_name;
 367         int err;
 368
 369         get_online_cpus();
 370         get_online_mems();
 371         memcg_get_cache_ids();
 372
 373         mutex_lock(&slab_mutex);
 374
 375         err = kmem_cache_sanity_check(name, size);
 376         if (err) {
 377                 s = NULL;       /* suppress uninit var warning */
 378                 goto out_unlock;
 379         }
 380
 381         /*
 382          * Some allocators will constraint the set of valid flags to a subset
 383          * of all flags. We expect them to define CACHE_CREATE_MASK in this
 384          * case, and we'll just provide them with a sanitized version of the
 385          * passed flags.
 386          */
 387         flags &= CACHE_CREATE_MASK;
 388
 389         s = __kmem_cache_alias(name, size, align, flags, ctor);
 390         if (s)
 391                 goto out_unlock;
 392
 393         cache_name = kstrdup_const(name, GFP_KERNEL);
 394         if (!cache_name) {
 395                 err = -ENOMEM;
 396                 goto out_unlock;
 397         }
 398
 399         s = do_kmem_cache_create(cache_name, size, size,
 400                                  calculate_alignment(flags, align, size),
 401                                  flags, ctor, NULL, NULL);
 402         if (IS_ERR(s)) {
 403                 err = PTR_ERR(s);
 404                 kfree_const(cache_name);
 405         }
 406
 407 out_unlock:
 408         mutex_unlock(&slab_mutex);
 409
 410         memcg_put_cache_ids();
 411         put_online_mems();
 412         put_online_cpus();
 413
 414         if (err) {
 415                 if (flags & SLAB_PANIC)
 416                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 417                                 name, err);
 418                 else {
 419                         printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d",
 420                                 name, err);
 421                         dump_stack();
 422                 }
 423                 return NULL;
 424         }
 425         return s;
 426 }
 427 EXPORT_SYMBOL(kmem_cache_create);
 428
 429 static int do_kmem_cache_shutdown(struct kmem_cache *s,
 430                 struct list_head *release, bool *need_rcu_barrier)
 431 {
 432         if (__kmem_cache_shutdown(s) != 0) {
 433                 printk(KERN_ERR "kmem_cache_destroy %s: "
 434                        "Slab cache still has objects\n", s->name);
 435                 dump_stack();
 436                 return -EBUSY;
 437         }
 438
 439         if (s->flags & SLAB_DESTROY_BY_RCU)
 440                 *need_rcu_barrier = true;
 441
 442 #ifdef CONFIG_MEMCG_KMEM
 443         if (!is_root_cache(s))
 444                 list_del(&s->memcg_params.list);
 445 #endif
 446         list_move(&s->list, release);
 447         return 0;
 448 }
 449
 450 static void do_kmem_cache_release(struct list_head *release,
 451                                   bool need_rcu_barrier)
 452 {
 453         struct kmem_cache *s, *s2;
 454
 455         if (need_rcu_barrier)
 456                 rcu_barrier();
 457
 458         list_for_each_entry_safe(s, s2, release, list) {
 459 #ifdef SLAB_SUPPORTS_SYSFS
 460                 sysfs_slab_remove(s);
 461 #else
 462                 slab_kmem_cache_release(s);
 463 #endif
 464         }
 465 }
 466
 467 #ifdef CONFIG_MEMCG_KMEM
 468 /*
 469  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
 470  * @memcg: The memory cgroup the new cache is for.
 471  * @root_cache: The parent of the new cache.
 472  *
 473  * This function attempts to create a kmem cache that will serve allocation
 474  * requests going from @memcg to @root_cache. The new cache inherits properties
 475  * from its parent.
 476  */
 477 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
 478                              struct kmem_cache *root_cache)
 479 {
 480         static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
 481         struct cgroup_subsys_state *css = mem_cgroup_css(memcg);
 482         struct memcg_cache_array *arr;
 483         struct kmem_cache *s = NULL;
 484         char *cache_name;
 485         int idx;
 486
 487         get_online_cpus();
 488         get_online_mems();
 489
 490         mutex_lock(&slab_mutex);
 491
 492         /*
 493          * The memory cgroup could have been deactivated while the cache
 494          * creation work was pending.
 495          */
 496         if (!memcg_kmem_is_active(memcg))
 497                 goto out_unlock;
 498
 499         idx = memcg_cache_id(memcg);
 500         arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
 501                                         lockdep_is_held(&slab_mutex));
 502
 503         /*
 504          * Since per-memcg caches are created asynchronously on first
 505          * allocation (see memcg_kmem_get_cache()), several threads can try to
 506          * create the same cache, but only one of them may succeed.
 507          */
 508         if (arr->entries[idx])
 509                 goto out_unlock;
 510
 511         cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
 512         cache_name = kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name,
 513                                css->id, memcg_name_buf);
 514         if (!cache_name)
 515                 goto out_unlock;
 516
 517         s = do_kmem_cache_create(cache_name, root_cache->object_size,
 518                                  root_cache->size, root_cache->align,
 519                                  root_cache->flags, root_cache->ctor,
 520                                  memcg, root_cache);
 521         /*
 522          * If we could not create a memcg cache, do not complain, because
 523          * that's not critical at all as we can always proceed with the root
 524          * cache.
 525          */
 526         if (IS_ERR(s)) {
 527                 kfree(cache_name);
 528                 goto out_unlock;
 529         }
 530
 531         list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
 532
 533         /*
 534          * Since readers won't lock (see cache_from_memcg_idx()), we need a
 535          * barrier here to ensure nobody will see the kmem_cache partially
 536          * initialized.
 537          */
 538         smp_wmb();
 539         arr->entries[idx] = s;
 540
 541 out_unlock:
 542         mutex_unlock(&slab_mutex);
 543
 544         put_online_mems();
 545         put_online_cpus();
 546 }
 547
 548 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
 549 {
 550         int idx;
 551         struct memcg_cache_array *arr;
 552         struct kmem_cache *s, *c;
 553
 554         idx = memcg_cache_id(memcg);
 555
 556         get_online_cpus();
 557         get_online_mems();
 558
 559         mutex_lock(&slab_mutex);
 560         list_for_each_entry(s, &slab_caches, list) {
 561                 if (!is_root_cache(s))
 562                         continue;
 563
 564                 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 565                                                 lockdep_is_held(&slab_mutex));
 566                 c = arr->entries[idx];
 567                 if (!c)
 568                         continue;
 569
 570                 __kmem_cache_shrink(c, true);
 571                 arr->entries[idx] = NULL;
 572         }
 573         mutex_unlock(&slab_mutex);
 574
 575         put_online_mems();
 576         put_online_cpus();
 577 }
 578
 579 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
 580 {
 581         LIST_HEAD(release);
 582         bool need_rcu_barrier = false;
 583         struct kmem_cache *s, *s2;
 584
 585         get_online_cpus();
 586         get_online_mems();
 587
 588         mutex_lock(&slab_mutex);
 589         list_for_each_entry_safe(s, s2, &slab_caches, list) {
 590                 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
 591                         continue;
 592                 /*
 593                  * The cgroup is about to be freed and therefore has no charges
 594                  * left. Hence, all its caches must be empty by now.
 595                  */
 596                 BUG_ON(do_kmem_cache_shutdown(s, &release, &need_rcu_barrier));
 597         }
 598         mutex_unlock(&slab_mutex);
 599
 600         put_online_mems();
 601         put_online_cpus();
 602
 603         do_kmem_cache_release(&release, need_rcu_barrier);
 604 }
 605 #endif /* CONFIG_MEMCG_KMEM */
 606
 607 void slab_kmem_cache_release(struct kmem_cache *s)
 608 {
 609         destroy_memcg_params(s);
 610         kfree_const(s->name);
 611         kmem_cache_free(kmem_cache, s);
 612 }
 613
 614 void kmem_cache_destroy(struct kmem_cache *s)
 615 {
 616         struct kmem_cache *c, *c2;
 617         LIST_HEAD(release);
 618         bool need_rcu_barrier = false;
 619         bool busy = false;
 620
 621         BUG_ON(!is_root_cache(s));
 622
 623         get_online_cpus();
 624         get_online_mems();
 625
 626         mutex_lock(&slab_mutex);
 627
 628         s->refcount--;
 629         if (s->refcount)
 630                 goto out_unlock;
 631
 632         for_each_memcg_cache_safe(c, c2, s) {
 633                 if (do_kmem_cache_shutdown(c, &release, &need_rcu_barrier))
 634                         busy = true;
 635         }
 636
 637         if (!busy)
 638                 do_kmem_cache_shutdown(s, &release, &need_rcu_barrier);
 639
 640 out_unlock:
 641         mutex_unlock(&slab_mutex);
 642
 643         put_online_mems();
 644         put_online_cpus();
 645
 646         do_kmem_cache_release(&release, need_rcu_barrier);
 647 }
 648 EXPORT_SYMBOL(kmem_cache_destroy);
 649
 650 /**
 651  * kmem_cache_shrink - Shrink a cache.
 652  * @cachep: The cache to shrink.
 653  *
 654  * Releases as many slabs as possible for a cache.
 655  * To help debugging, a zero exit status indicates all slabs were released.
 656  */
 657 int kmem_cache_shrink(struct kmem_cache *cachep)
 658 {
 659         int ret;
 660
 661         get_online_cpus();
 662         get_online_mems();
 663         ret = __kmem_cache_shrink(cachep, false);
 664         put_online_mems();
 665         put_online_cpus();
 666         return ret;
 667 }
 668 EXPORT_SYMBOL(kmem_cache_shrink);
 669
 670 int slab_is_available(void)
 671 {
 672         return slab_state >= UP;
 673 }
 674
 675 #ifndef CONFIG_SLOB
 676 /* Create a cache during boot when no slab services are available yet */
 677 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
 678                 unsigned long flags)
 679 {
 680         int err;
 681
 682         s->name = name;
 683         s->size = s->object_size = size;
 684         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 685
 686         slab_init_memcg_params(s);
 687
 688         err = __kmem_cache_create(s, flags);
 689
 690         if (err)
 691                 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
 692                                         name, size, err);
 693
 694         s->refcount = -1;       /* Exempt from merging for now */
 695 }
 696
 697 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
 698                                 unsigned long flags)
 699 {
 700         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 701
 702         if (!s)
 703                 panic("Out of memory when creating slab %s\n", name);
 704
 705         create_boot_cache(s, name, size, flags);
 706         list_add(&s->list, &slab_caches);
 707         s->refcount = 1;
 708         return s;
 709 }
 710
 711 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 712 EXPORT_SYMBOL(kmalloc_caches);
 713
 714 #ifdef CONFIG_ZONE_DMA
 715 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 716 EXPORT_SYMBOL(kmalloc_dma_caches);
 717 #endif
 718
 719 /*
 720  * Conversion table for small slabs sizes / 8 to the index in the
 721  * kmalloc array. This is necessary for slabs < 192 since we have non power
 722  * of two cache sizes there. The size of larger slabs can be determined using
 723  * fls.
 724  */
 725 static s8 size_index[24] = {
 726         3,      /* 8 */
 727         4,      /* 16 */
 728         5,      /* 24 */
 729         5,      /* 32 */
 730         6,      /* 40 */
 731         6,      /* 48 */
 732         6,      /* 56 */
 733         6,      /* 64 */
 734         1,      /* 72 */
 735         1,      /* 80 */
 736         1,      /* 88 */
 737         1,      /* 96 */
 738         7,      /* 104 */
 739         7,      /* 112 */
 740         7,      /* 120 */
 741         7,      /* 128 */
 742         2,      /* 136 */
 743         2,      /* 144 */
 744         2,      /* 152 */
 745         2,      /* 160 */
 746         2,      /* 168 */
 747         2,      /* 176 */
 748         2,      /* 184 */
 749         2       /* 192 */
 750 };
 751
 752 static inline int size_index_elem(size_t bytes)
 753 {
 754         return (bytes - 1) / 8;
 755 }
 756
 757 /*
 758  * Find the kmem_cache structure that serves a given size of
 759  * allocation
 760  */
 761 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 762 {
 763         int index;
 764
 765         if (unlikely(size > KMALLOC_MAX_SIZE)) {
 766                 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 767                 return NULL;
 768         }
 769
 770         if (size <= 192) {
 771                 if (!size)
 772                         return ZERO_SIZE_PTR;
 773
 774                 index = size_index[size_index_elem(size)];
 775         } else
 776                 index = fls(size - 1);
 777
 778 #ifdef CONFIG_ZONE_DMA
 779         if (unlikely((flags & GFP_DMA)))
 780                 return kmalloc_dma_caches[index];
 781
 782 #endif
 783         return kmalloc_caches[index];
 784 }
 785
 786 /*
 787  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
 788  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
 789  * kmalloc-67108864.
 790  */
 791 static struct {
 792         const char *name;
 793         unsigned long size;
 794 } const kmalloc_info[] __initconst = {
 795         {NULL,                      0},         {"kmalloc-96",             96},
 796         {"kmalloc-192",           192},         {"kmalloc-8",               8},
 797         {"kmalloc-16",             16},         {"kmalloc-32",             32},
 798         {"kmalloc-64",             64},         {"kmalloc-128",           128},
 799         {"kmalloc-256",           256},         {"kmalloc-512",           512},
 800         {"kmalloc-1024",         1024},         {"kmalloc-2048",         2048},
 801         {"kmalloc-4096",         4096},         {"kmalloc-8192",         8192},
 802         {"kmalloc-16384",       16384},         {"kmalloc-32768",       32768},
 803         {"kmalloc-65536",       65536},         {"kmalloc-131072",     131072},
 804         {"kmalloc-262144",     262144},         {"kmalloc-524288",     524288},
 805         {"kmalloc-1048576",   1048576},         {"kmalloc-2097152",   2097152},
 806         {"kmalloc-4194304",   4194304},         {"kmalloc-8388608",   8388608},
 807         {"kmalloc-16777216", 16777216},         {"kmalloc-33554432", 33554432},
 808         {"kmalloc-67108864", 67108864}
 809 };
 810
 811 /*
 812  * Patch up the size_index table if we have strange large alignment
 813  * requirements for the kmalloc array. This is only the case for
 814  * MIPS it seems. The standard arches will not generate any code here.
 815  *
 816  * Largest permitted alignment is 256 bytes due to the way we
 817  * handle the index determination for the smaller caches.
 818  *
 819  * Make sure that nothing crazy happens if someone starts tinkering
 820  * around with ARCH_KMALLOC_MINALIGN
 821  */
 822 void __init setup_kmalloc_cache_index_table(void)
 823 {
 824         int i;
 825
 826         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 827                 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
 828
 829         for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 830                 int elem = size_index_elem(i);
 831
 832                 if (elem >= ARRAY_SIZE(size_index))
 833                         break;
 834                 size_index[elem] = KMALLOC_SHIFT_LOW;
 835         }
 836
 837         if (KMALLOC_MIN_SIZE >= 64) {
 838                 /*
 839                  * The 96 byte size cache is not used if the alignment
 840                  * is 64 byte.
 841                  */
 842                 for (i = 64 + 8; i <= 96; i += 8)
 843                         size_index[size_index_elem(i)] = 7;
 844
 845         }
 846
 847         if (KMALLOC_MIN_SIZE >= 128) {
 848                 /*
 849                  * The 192 byte sized cache is not used if the alignment
 850                  * is 128 byte. Redirect kmalloc to use the 256 byte cache
 851                  * instead.
 852                  */
 853                 for (i = 128 + 8; i <= 192; i += 8)
 854                         size_index[size_index_elem(i)] = 8;
 855         }
 856 }
 857
 858 static void __init new_kmalloc_cache(int idx, unsigned long flags)
 859 {
 860         kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
 861                                         kmalloc_info[idx].size, flags);
 862 }
 863
 864 /*
 865  * Create the kmalloc array. Some of the regular kmalloc arrays
 866  * may already have been created because they were needed to
 867  * enable allocations for slab creation.
 868  */
 869 void __init create_kmalloc_caches(unsigned long flags)
 870 {
 871         int i;
 872
 873         for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
 874                 if (!kmalloc_caches[i])
 875                         new_kmalloc_cache(i, flags);
 876
 877                 /*
 878                  * Caches that are not of the two-to-the-power-of size.
 879                  * These have to be created immediately after the
 880                  * earlier power of two caches
 881                  */
 882                 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
 883                         new_kmalloc_cache(1, flags);
 884                 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
 885                         new_kmalloc_cache(2, flags);
 886         }
 887
 888         /* Kmalloc array is now usable */
 889         slab_state = UP;
 890
 891 #ifdef CONFIG_ZONE_DMA
 892         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
 893                 struct kmem_cache *s = kmalloc_caches[i];
 894
 895                 if (s) {
 896                         int size = kmalloc_size(i);
 897                         char *n = kasprintf(GFP_NOWAIT,
 898                                  "dma-kmalloc-%d", size);
 899
 900                         BUG_ON(!n);
 901                         kmalloc_dma_caches[i] = create_kmalloc_cache(n,
 902                                 size, SLAB_CACHE_DMA | flags);
 903                 }
 904         }
 905 #endif
 906 }
 907 #endif /* !CONFIG_SLOB */
 908
 909 /*
 910  * To avoid unnecessary overhead, we pass through large allocation requests
 911  * directly to the page allocator. We use __GFP_COMP, because we will need to
 912  * know the allocation order to free the pages properly in kfree.
 913  */
 914 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
 915 {
 916         void *ret;
 917         struct page *page;
 918
 919         flags |= __GFP_COMP;
 920         page = alloc_kmem_pages(flags, order);
 921         ret = page ? page_address(page) : NULL;
 922         kmemleak_alloc(ret, size, 1, flags);
 923         kasan_kmalloc_large(ret, size);
 924         return ret;
 925 }
 926 EXPORT_SYMBOL(kmalloc_order);
 927
 928 #ifdef CONFIG_TRACING
 929 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
 930 {
 931         void *ret = kmalloc_order(size, flags, order);
 932         trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
 933         return ret;
 934 }
 935 EXPORT_SYMBOL(kmalloc_order_trace);
 936 #endif
 937
 938 #ifdef CONFIG_SLABINFO
 939
 940 #ifdef CONFIG_SLAB
 941 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
 942 #else
 943 #define SLABINFO_RIGHTS S_IRUSR
 944 #endif
 945
 946 static void print_slabinfo_header(struct seq_file *m)
 947 {
 948         /*
 949          * Output format version, so at least we can change it
 950          * without _too_ many complaints.
 951          */
 952 #ifdef CONFIG_DEBUG_SLAB
 953         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
 954 #else
 955         seq_puts(m, "slabinfo - version: 2.1\n");
 956 #endif
 957         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> "
 958                  "<objperslab> <pagesperslab>");
 959         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
 960         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
 961 #ifdef CONFIG_DEBUG_SLAB
 962         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
 963                  "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
 964         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
 965 #endif
 966         seq_putc(m, '\n');
 967 }
 968
 969 void *slab_start(struct seq_file *m, loff_t *pos)
 970 {
 971         mutex_lock(&slab_mutex);
 972         return seq_list_start(&slab_caches, *pos);
 973 }
 974
 975 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
 976 {
 977         return seq_list_next(p, &slab_caches, pos);
 978 }
 979
 980 void slab_stop(struct seq_file *m, void *p)
 981 {
 982         mutex_unlock(&slab_mutex);
 983 }
 984
 985 static void
 986 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
 987 {
 988         struct kmem_cache *c;
 989         struct slabinfo sinfo;
 990
 991         if (!is_root_cache(s))
 992                 return;
 993
 994         for_each_memcg_cache(c, s) {
 995                 memset(&sinfo, 0, sizeof(sinfo));
 996                 get_slabinfo(c, &sinfo);
 997
 998                 info->active_slabs += sinfo.active_slabs;
 999                 info->num_slabs += sinfo.num_slabs;
1000                 info->shared_avail += sinfo.shared_avail;
1001                 info->active_objs += sinfo.active_objs;
1002                 info->num_objs += sinfo.num_objs;
1003         }
1004 }
1005
1006 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1007 {
1008         struct slabinfo sinfo;
1009
1010         memset(&sinfo, 0, sizeof(sinfo));
1011         get_slabinfo(s, &sinfo);
1012
1013         memcg_accumulate_slabinfo(s, &sinfo);
1014
1015         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1016                    cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1017                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
1018
1019         seq_printf(m, " : tunables %4u %4u %4u",
1020                    sinfo.limit, sinfo.batchcount, sinfo.shared);
1021         seq_printf(m, " : slabdata %6lu %6lu %6lu",
1022                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1023         slabinfo_show_stats(m, s);
1024         seq_putc(m, '\n');
1025 }
1026
1027 static int slab_show(struct seq_file *m, void *p)
1028 {
1029         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1030
1031         if (p == slab_caches.next)
1032                 print_slabinfo_header(m);
1033         if (is_root_cache(s))
1034                 cache_show(s, m);
1035         return 0;
1036 }
1037
1038 #ifdef CONFIG_MEMCG_KMEM
1039 int memcg_slab_show(struct seq_file *m, void *p)
1040 {
1041         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1042         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1043
1044         if (p == slab_caches.next)
1045                 print_slabinfo_header(m);
1046         if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1047                 cache_show(s, m);
1048         return 0;
1049 }
1050 #endif
1051
1052 /*
1053  * slabinfo_op - iterator that generates /proc/slabinfo
1054  *
1055  * Output layout:
1056  * cache-name
1057  * num-active-objs
1058  * total-objs
1059  * object size
1060  * num-active-slabs
1061  * total-slabs
1062  * num-pages-per-slab
1063  * + further values on SMP and with statistics enabled
1064  */
1065 static const struct seq_operations slabinfo_op = {
1066         .start = slab_start,
1067         .next = slab_next,
1068         .stop = slab_stop,
1069         .show = slab_show,
1070 };
1071
1072 static int slabinfo_open(struct inode *inode, struct file *file)
1073 {
1074         return seq_open(file, &slabinfo_op);
1075 }
1076
1077 static const struct file_operations proc_slabinfo_operations = {
1078         .open           = slabinfo_open,
1079         .read           = seq_read,
1080         .write          = slabinfo_write,
1081         .llseek         = seq_lseek,
1082         .release        = seq_release,
1083 };
1084
1085 static int __init slab_proc_init(void)
1086 {
1087         proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1088                                                 &proc_slabinfo_operations);
1089         return 0;
1090 }
1091 module_init(slab_proc_init);
1092 #endif /* CONFIG_SLABINFO */
1093
1094 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1095                                            gfp_t flags)
1096 {
1097         void *ret;
1098         size_t ks = 0;
1099
1100         if (p)
1101                 ks = ksize(p);
1102
1103         if (ks >= new_size) {
1104                 kasan_krealloc((void *)p, new_size);
1105                 return (void *)p;
1106         }
1107
1108         ret = kmalloc_track_caller(new_size, flags);
1109         if (ret && p)
1110                 memcpy(ret, p, ks);
1111
1112         return ret;
1113 }
1114
1115 /**
1116  * __krealloc - like krealloc() but don't free @p.
1117  * @p: object to reallocate memory for.
1118  * @new_size: how many bytes of memory are required.
1119  * @flags: the type of memory to allocate.
1120  *
1121  * This function is like krealloc() except it never frees the originally
1122  * allocated buffer. Use this if you don't want to free the buffer immediately
1123  * like, for example, with RCU.
1124  */
1125 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1126 {
1127         if (unlikely(!new_size))
1128                 return ZERO_SIZE_PTR;
1129
1130         return __do_krealloc(p, new_size, flags);
1131
1132 }
1133 EXPORT_SYMBOL(__krealloc);
1134
1135 /**
1136  * krealloc - reallocate memory. The contents will remain unchanged.
1137  * @p: object to reallocate memory for.
1138  * @new_size: how many bytes of memory are required.
1139  * @flags: the type of memory to allocate.
1140  *
1141  * The contents of the object pointed to are preserved up to the
1142  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1143  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1144  * %NULL pointer, the object pointed to is freed.
1145  */
1146 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1147 {
1148         void *ret;
1149
1150         if (unlikely(!new_size)) {
1151                 kfree(p);
1152                 return ZERO_SIZE_PTR;
1153         }
1154
1155         ret = __do_krealloc(p, new_size, flags);
1156         if (ret && p != ret)
1157                 kfree(p);
1158
1159         return ret;
1160 }
1161 EXPORT_SYMBOL(krealloc);
1162
1163 /**
1164  * kzfree - like kfree but zero memory
1165  * @p: object to free memory of
1166  *
1167  * The memory of the object @p points to is zeroed before freed.
1168  * If @p is %NULL, kzfree() does nothing.
1169  *
1170  * Note: this function zeroes the whole allocated buffer which can be a good
1171  * deal bigger than the requested buffer size passed to kmalloc(). So be
1172  * careful when using this function in performance sensitive code.
1173  */
1174 void kzfree(const void *p)
1175 {
1176         size_t ks;
1177         void *mem = (void *)p;
1178
1179         if (unlikely(ZERO_OR_NULL_PTR(mem)))
1180                 return;
1181         ks = ksize(mem);
1182         memset(mem, 0, ks);
1183         kfree(mem);
1184 }
1185 EXPORT_SYMBOL(kzfree);
1186
1187 /* Tracepoints definitions. */
1188 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1189 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1190 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1191 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1192 EXPORT_TRACEPOINT_SYMBOL(kfree);
1193 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);