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 | SLAB_KASAN)
  39
  40 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
  41                          SLAB_NOTRACK | SLAB_ACCOUNT)
  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 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
 109 {
 110         size_t i;
 111
 112         for (i = 0; i < nr; i++) {
 113                 if (s)
 114                         kmem_cache_free(s, p[i]);
 115                 else
 116                         kfree(p[i]);
 117         }
 118 }
 119
 120 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
 121                                                                 void **p)
 122 {
 123         size_t i;
 124
 125         for (i = 0; i < nr; i++) {
 126                 void *x = p[i] = kmem_cache_alloc(s, flags);
 127                 if (!x) {
 128                         __kmem_cache_free_bulk(s, i, p);
 129                         return 0;
 130                 }
 131         }
 132         return i;
 133 }
 134
 135 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 136 void slab_init_memcg_params(struct kmem_cache *s)
 137 {
 138         s->memcg_params.is_root_cache = true;
 139         INIT_LIST_HEAD(&s->memcg_params.list);
 140         RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL);
 141 }
 142
 143 static int init_memcg_params(struct kmem_cache *s,
 144                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 145 {
 146         struct memcg_cache_array *arr;
 147
 148         if (memcg) {
 149                 s->memcg_params.is_root_cache = false;
 150                 s->memcg_params.memcg = memcg;
 151                 s->memcg_params.root_cache = root_cache;
 152                 return 0;
 153         }
 154
 155         slab_init_memcg_params(s);
 156
 157         if (!memcg_nr_cache_ids)
 158                 return 0;
 159
 160         arr = kzalloc(sizeof(struct memcg_cache_array) +
 161                       memcg_nr_cache_ids * sizeof(void *),
 162                       GFP_KERNEL);
 163         if (!arr)
 164                 return -ENOMEM;
 165
 166         RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr);
 167         return 0;
 168 }
 169
 170 static void destroy_memcg_params(struct kmem_cache *s)
 171 {
 172         if (is_root_cache(s))
 173                 kfree(rcu_access_pointer(s->memcg_params.memcg_caches));
 174 }
 175
 176 static int update_memcg_params(struct kmem_cache *s, int new_array_size)
 177 {
 178         struct memcg_cache_array *old, *new;
 179
 180         if (!is_root_cache(s))
 181                 return 0;
 182
 183         new = kzalloc(sizeof(struct memcg_cache_array) +
 184                       new_array_size * sizeof(void *), GFP_KERNEL);
 185         if (!new)
 186                 return -ENOMEM;
 187
 188         old = rcu_dereference_protected(s->memcg_params.memcg_caches,
 189                                         lockdep_is_held(&slab_mutex));
 190         if (old)
 191                 memcpy(new->entries, old->entries,
 192                        memcg_nr_cache_ids * sizeof(void *));
 193
 194         rcu_assign_pointer(s->memcg_params.memcg_caches, new);
 195         if (old)
 196                 kfree_rcu(old, rcu);
 197         return 0;
 198 }
 199
 200 int memcg_update_all_caches(int num_memcgs)
 201 {
 202         struct kmem_cache *s;
 203         int ret = 0;
 204
 205         mutex_lock(&slab_mutex);
 206         list_for_each_entry(s, &slab_caches, list) {
 207                 ret = update_memcg_params(s, num_memcgs);
 208                 /*
 209                  * Instead of freeing the memory, we'll just leave the caches
 210                  * up to this point in an updated state.
 211                  */
 212                 if (ret)
 213                         break;
 214         }
 215         mutex_unlock(&slab_mutex);
 216         return ret;
 217 }
 218 #else
 219 static inline int init_memcg_params(struct kmem_cache *s,
 220                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 221 {
 222         return 0;
 223 }
 224
 225 static inline void destroy_memcg_params(struct kmem_cache *s)
 226 {
 227 }
 228 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 229
 230 /*
 231  * Find a mergeable slab cache
 232  */
 233 int slab_unmergeable(struct kmem_cache *s)
 234 {
 235         if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
 236                 return 1;
 237
 238         if (!is_root_cache(s))
 239                 return 1;
 240
 241         if (s->ctor)
 242                 return 1;
 243
 244         /*
 245          * We may have set a slab to be unmergeable during bootstrap.
 246          */
 247         if (s->refcount < 0)
 248                 return 1;
 249
 250         return 0;
 251 }
 252
 253 struct kmem_cache *find_mergeable(size_t size, size_t align,
 254                 unsigned long flags, const char *name, void (*ctor)(void *))
 255 {
 256         struct kmem_cache *s;
 257
 258         if (slab_nomerge || (flags & SLAB_NEVER_MERGE))
 259                 return NULL;
 260
 261         if (ctor)
 262                 return NULL;
 263
 264         size = ALIGN(size, sizeof(void *));
 265         align = calculate_alignment(flags, align, size);
 266         size = ALIGN(size, align);
 267         flags = kmem_cache_flags(size, flags, name, NULL);
 268
 269         list_for_each_entry_reverse(s, &slab_caches, list) {
 270                 if (slab_unmergeable(s))
 271                         continue;
 272
 273                 if (size > s->size)
 274                         continue;
 275
 276                 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
 277                         continue;
 278                 /*
 279                  * Check if alignment is compatible.
 280                  * Courtesy of Adrian Drzewiecki
 281                  */
 282                 if ((s->size & ~(align - 1)) != s->size)
 283                         continue;
 284
 285                 if (s->size - size >= sizeof(void *))
 286                         continue;
 287
 288                 if (IS_ENABLED(CONFIG_SLAB) && align &&
 289                         (align > s->align || s->align % align))
 290                         continue;
 291
 292                 return s;
 293         }
 294         return NULL;
 295 }
 296
 297 /*
 298  * Figure out what the alignment of the objects will be given a set of
 299  * flags, a user specified alignment and the size of the objects.
 300  */
 301 unsigned long calculate_alignment(unsigned long flags,
 302                 unsigned long align, unsigned long size)
 303 {
 304         /*
 305          * If the user wants hardware cache aligned objects then follow that
 306          * suggestion if the object is sufficiently large.
 307          *
 308          * The hardware cache alignment cannot override the specified
 309          * alignment though. If that is greater then use it.
 310          */
 311         if (flags & SLAB_HWCACHE_ALIGN) {
 312                 unsigned long ralign = cache_line_size();
 313                 while (size <= ralign / 2)
 314                         ralign /= 2;
 315                 align = max(align, ralign);
 316         }
 317
 318         if (align < ARCH_SLAB_MINALIGN)
 319                 align = ARCH_SLAB_MINALIGN;
 320
 321         return ALIGN(align, sizeof(void *));
 322 }
 323
 324 static struct kmem_cache *create_cache(const char *name,
 325                 size_t object_size, size_t size, size_t align,
 326                 unsigned long flags, void (*ctor)(void *),
 327                 struct mem_cgroup *memcg, struct kmem_cache *root_cache)
 328 {
 329         struct kmem_cache *s;
 330         int err;
 331
 332         err = -ENOMEM;
 333         s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
 334         if (!s)
 335                 goto out;
 336
 337         s->name = name;
 338         s->object_size = object_size;
 339         s->size = size;
 340         s->align = align;
 341         s->ctor = ctor;
 342
 343         err = init_memcg_params(s, memcg, root_cache);
 344         if (err)
 345                 goto out_free_cache;
 346
 347         err = __kmem_cache_create(s, flags);
 348         if (err)
 349                 goto out_free_cache;
 350
 351         s->refcount = 1;
 352         list_add(&s->list, &slab_caches);
 353 out:
 354         if (err)
 355                 return ERR_PTR(err);
 356         return s;
 357
 358 out_free_cache:
 359         destroy_memcg_params(s);
 360         kmem_cache_free(kmem_cache, s);
 361         goto out;
 362 }
 363
 364 /*
 365  * kmem_cache_create - Create a cache.
 366  * @name: A string which is used in /proc/slabinfo to identify this cache.
 367  * @size: The size of objects to be created in this cache.
 368  * @align: The required alignment for the objects.
 369  * @flags: SLAB flags
 370  * @ctor: A constructor for the objects.
 371  *
 372  * Returns a ptr to the cache on success, NULL on failure.
 373  * Cannot be called within a interrupt, but can be interrupted.
 374  * The @ctor is run when new pages are allocated by the cache.
 375  *
 376  * The flags are
 377  *
 378  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
 379  * to catch references to uninitialised memory.
 380  *
 381  * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
 382  * for buffer overruns.
 383  *
 384  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
 385  * cacheline.  This can be beneficial if you're counting cycles as closely
 386  * as davem.
 387  */
 388 struct kmem_cache *
 389 kmem_cache_create(const char *name, size_t size, size_t align,
 390                   unsigned long flags, void (*ctor)(void *))
 391 {
 392         struct kmem_cache *s = NULL;
 393         const char *cache_name;
 394         int err;
 395
 396         get_online_cpus();
 397         get_online_mems();
 398         memcg_get_cache_ids();
 399
 400         mutex_lock(&slab_mutex);
 401
 402         err = kmem_cache_sanity_check(name, size);
 403         if (err) {
 404                 goto out_unlock;
 405         }
 406
 407         /*
 408          * Some allocators will constraint the set of valid flags to a subset
 409          * of all flags. We expect them to define CACHE_CREATE_MASK in this
 410          * case, and we'll just provide them with a sanitized version of the
 411          * passed flags.
 412          */
 413         flags &= CACHE_CREATE_MASK;
 414
 415         s = __kmem_cache_alias(name, size, align, flags, ctor);
 416         if (s)
 417                 goto out_unlock;
 418
 419         cache_name = kstrdup_const(name, GFP_KERNEL);
 420         if (!cache_name) {
 421                 err = -ENOMEM;
 422                 goto out_unlock;
 423         }
 424
 425         s = create_cache(cache_name, size, size,
 426                          calculate_alignment(flags, align, size),
 427                          flags, ctor, NULL, NULL);
 428         if (IS_ERR(s)) {
 429                 err = PTR_ERR(s);
 430                 kfree_const(cache_name);
 431         }
 432
 433 out_unlock:
 434         mutex_unlock(&slab_mutex);
 435
 436         memcg_put_cache_ids();
 437         put_online_mems();
 438         put_online_cpus();
 439
 440         if (err) {
 441                 if (flags & SLAB_PANIC)
 442                         panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
 443                                 name, err);
 444                 else {
 445                         pr_warn("kmem_cache_create(%s) failed with error %d\n",
 446                                 name, err);
 447                         dump_stack();
 448                 }
 449                 return NULL;
 450         }
 451         return s;
 452 }
 453 EXPORT_SYMBOL(kmem_cache_create);
 454
 455 static int shutdown_cache(struct kmem_cache *s,
 456                 struct list_head *release, bool *need_rcu_barrier)
 457 {
 458         if (__kmem_cache_shutdown(s) != 0)
 459                 return -EBUSY;
 460
 461         if (s->flags & SLAB_DESTROY_BY_RCU)
 462                 *need_rcu_barrier = true;
 463
 464         list_move(&s->list, release);
 465         return 0;
 466 }
 467
 468 static void release_caches(struct list_head *release, bool need_rcu_barrier)
 469 {
 470         struct kmem_cache *s, *s2;
 471
 472         if (need_rcu_barrier)
 473                 rcu_barrier();
 474
 475         list_for_each_entry_safe(s, s2, release, list) {
 476 #ifdef SLAB_SUPPORTS_SYSFS
 477                 sysfs_slab_remove(s);
 478 #else
 479                 slab_kmem_cache_release(s);
 480 #endif
 481         }
 482 }
 483
 484 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
 485 /*
 486  * memcg_create_kmem_cache - Create a cache for a memory cgroup.
 487  * @memcg: The memory cgroup the new cache is for.
 488  * @root_cache: The parent of the new cache.
 489  *
 490  * This function attempts to create a kmem cache that will serve allocation
 491  * requests going from @memcg to @root_cache. The new cache inherits properties
 492  * from its parent.
 493  */
 494 void memcg_create_kmem_cache(struct mem_cgroup *memcg,
 495                              struct kmem_cache *root_cache)
 496 {
 497         static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */
 498         struct cgroup_subsys_state *css = &memcg->css;
 499         struct memcg_cache_array *arr;
 500         struct kmem_cache *s = NULL;
 501         char *cache_name;
 502         int idx;
 503
 504         get_online_cpus();
 505         get_online_mems();
 506
 507         mutex_lock(&slab_mutex);
 508
 509         /*
 510          * The memory cgroup could have been offlined while the cache
 511          * creation work was pending.
 512          */
 513         if (memcg->kmem_state != KMEM_ONLINE)
 514                 goto out_unlock;
 515
 516         idx = memcg_cache_id(memcg);
 517         arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches,
 518                                         lockdep_is_held(&slab_mutex));
 519
 520         /*
 521          * Since per-memcg caches are created asynchronously on first
 522          * allocation (see memcg_kmem_get_cache()), several threads can try to
 523          * create the same cache, but only one of them may succeed.
 524          */
 525         if (arr->entries[idx])
 526                 goto out_unlock;
 527
 528         cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf));
 529         cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name,
 530                                css->serial_nr, memcg_name_buf);
 531         if (!cache_name)
 532                 goto out_unlock;
 533
 534         s = create_cache(cache_name, root_cache->object_size,
 535                          root_cache->size, root_cache->align,
 536                          root_cache->flags, root_cache->ctor,
 537                          memcg, root_cache);
 538         /*
 539          * If we could not create a memcg cache, do not complain, because
 540          * that's not critical at all as we can always proceed with the root
 541          * cache.
 542          */
 543         if (IS_ERR(s)) {
 544                 kfree(cache_name);
 545                 goto out_unlock;
 546         }
 547
 548         list_add(&s->memcg_params.list, &root_cache->memcg_params.list);
 549
 550         /*
 551          * Since readers won't lock (see cache_from_memcg_idx()), we need a
 552          * barrier here to ensure nobody will see the kmem_cache partially
 553          * initialized.
 554          */
 555         smp_wmb();
 556         arr->entries[idx] = s;
 557
 558 out_unlock:
 559         mutex_unlock(&slab_mutex);
 560
 561         put_online_mems();
 562         put_online_cpus();
 563 }
 564
 565 void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg)
 566 {
 567         int idx;
 568         struct memcg_cache_array *arr;
 569         struct kmem_cache *s, *c;
 570
 571         idx = memcg_cache_id(memcg);
 572
 573         get_online_cpus();
 574         get_online_mems();
 575
 576         mutex_lock(&slab_mutex);
 577         list_for_each_entry(s, &slab_caches, list) {
 578                 if (!is_root_cache(s))
 579                         continue;
 580
 581                 arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 582                                                 lockdep_is_held(&slab_mutex));
 583                 c = arr->entries[idx];
 584                 if (!c)
 585                         continue;
 586
 587                 __kmem_cache_shrink(c, true);
 588                 arr->entries[idx] = NULL;
 589         }
 590         mutex_unlock(&slab_mutex);
 591
 592         put_online_mems();
 593         put_online_cpus();
 594 }
 595
 596 static int __shutdown_memcg_cache(struct kmem_cache *s,
 597                 struct list_head *release, bool *need_rcu_barrier)
 598 {
 599         BUG_ON(is_root_cache(s));
 600
 601         if (shutdown_cache(s, release, need_rcu_barrier))
 602                 return -EBUSY;
 603
 604         list_del(&s->memcg_params.list);
 605         return 0;
 606 }
 607
 608 void memcg_destroy_kmem_caches(struct mem_cgroup *memcg)
 609 {
 610         LIST_HEAD(release);
 611         bool need_rcu_barrier = false;
 612         struct kmem_cache *s, *s2;
 613
 614         get_online_cpus();
 615         get_online_mems();
 616
 617         mutex_lock(&slab_mutex);
 618         list_for_each_entry_safe(s, s2, &slab_caches, list) {
 619                 if (is_root_cache(s) || s->memcg_params.memcg != memcg)
 620                         continue;
 621                 /*
 622                  * The cgroup is about to be freed and therefore has no charges
 623                  * left. Hence, all its caches must be empty by now.
 624                  */
 625                 BUG_ON(__shutdown_memcg_cache(s, &release, &need_rcu_barrier));
 626         }
 627         mutex_unlock(&slab_mutex);
 628
 629         put_online_mems();
 630         put_online_cpus();
 631
 632         release_caches(&release, need_rcu_barrier);
 633 }
 634
 635 static int shutdown_memcg_caches(struct kmem_cache *s,
 636                 struct list_head *release, bool *need_rcu_barrier)
 637 {
 638         struct memcg_cache_array *arr;
 639         struct kmem_cache *c, *c2;
 640         LIST_HEAD(busy);
 641         int i;
 642
 643         BUG_ON(!is_root_cache(s));
 644
 645         /*
 646          * First, shutdown active caches, i.e. caches that belong to online
 647          * memory cgroups.
 648          */
 649         arr = rcu_dereference_protected(s->memcg_params.memcg_caches,
 650                                         lockdep_is_held(&slab_mutex));
 651         for_each_memcg_cache_index(i) {
 652                 c = arr->entries[i];
 653                 if (!c)
 654                         continue;
 655                 if (__shutdown_memcg_cache(c, release, need_rcu_barrier))
 656                         /*
 657                          * The cache still has objects. Move it to a temporary
 658                          * list so as not to try to destroy it for a second
 659                          * time while iterating over inactive caches below.
 660                          */
 661                         list_move(&c->memcg_params.list, &busy);
 662                 else
 663                         /*
 664                          * The cache is empty and will be destroyed soon. Clear
 665                          * the pointer to it in the memcg_caches array so that
 666                          * it will never be accessed even if the root cache
 667                          * stays alive.
 668                          */
 669                         arr->entries[i] = NULL;
 670         }
 671
 672         /*
 673          * Second, shutdown all caches left from memory cgroups that are now
 674          * offline.
 675          */
 676         list_for_each_entry_safe(c, c2, &s->memcg_params.list,
 677                                  memcg_params.list)
 678                 __shutdown_memcg_cache(c, release, need_rcu_barrier);
 679
 680         list_splice(&busy, &s->memcg_params.list);
 681
 682         /*
 683          * A cache being destroyed must be empty. In particular, this means
 684          * that all per memcg caches attached to it must be empty too.
 685          */
 686         if (!list_empty(&s->memcg_params.list))
 687                 return -EBUSY;
 688         return 0;
 689 }
 690 #else
 691 static inline int shutdown_memcg_caches(struct kmem_cache *s,
 692                 struct list_head *release, bool *need_rcu_barrier)
 693 {
 694         return 0;
 695 }
 696 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
 697
 698 void slab_kmem_cache_release(struct kmem_cache *s)
 699 {
 700         __kmem_cache_release(s);
 701         destroy_memcg_params(s);
 702         kfree_const(s->name);
 703         kmem_cache_free(kmem_cache, s);
 704 }
 705
 706 void kmem_cache_destroy(struct kmem_cache *s)
 707 {
 708         LIST_HEAD(release);
 709         bool need_rcu_barrier = false;
 710         int err;
 711
 712         if (unlikely(!s))
 713                 return;
 714
 715         get_online_cpus();
 716         get_online_mems();
 717
 718         kasan_cache_destroy(s);
 719         mutex_lock(&slab_mutex);
 720
 721         s->refcount--;
 722         if (s->refcount)
 723                 goto out_unlock;
 724
 725         err = shutdown_memcg_caches(s, &release, &need_rcu_barrier);
 726         if (!err)
 727                 err = shutdown_cache(s, &release, &need_rcu_barrier);
 728
 729         if (err) {
 730                 pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
 731                        s->name);
 732                 dump_stack();
 733         }
 734 out_unlock:
 735         mutex_unlock(&slab_mutex);
 736
 737         put_online_mems();
 738         put_online_cpus();
 739
 740         release_caches(&release, need_rcu_barrier);
 741 }
 742 EXPORT_SYMBOL(kmem_cache_destroy);
 743
 744 /**
 745  * kmem_cache_shrink - Shrink a cache.
 746  * @cachep: The cache to shrink.
 747  *
 748  * Releases as many slabs as possible for a cache.
 749  * To help debugging, a zero exit status indicates all slabs were released.
 750  */
 751 int kmem_cache_shrink(struct kmem_cache *cachep)
 752 {
 753         int ret;
 754
 755         get_online_cpus();
 756         get_online_mems();
 757         kasan_cache_shrink(cachep);
 758         ret = __kmem_cache_shrink(cachep, false);
 759         put_online_mems();
 760         put_online_cpus();
 761         return ret;
 762 }
 763 EXPORT_SYMBOL(kmem_cache_shrink);
 764
 765 bool slab_is_available(void)
 766 {
 767         return slab_state >= UP;
 768 }
 769
 770 #ifndef CONFIG_SLOB
 771 /* Create a cache during boot when no slab services are available yet */
 772 void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size,
 773                 unsigned long flags)
 774 {
 775         int err;
 776
 777         s->name = name;
 778         s->size = s->object_size = size;
 779         s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size);
 780
 781         slab_init_memcg_params(s);
 782
 783         err = __kmem_cache_create(s, flags);
 784
 785         if (err)
 786                 panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n",
 787                                         name, size, err);
 788
 789         s->refcount = -1;       /* Exempt from merging for now */
 790 }
 791
 792 struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size,
 793                                 unsigned long flags)
 794 {
 795         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
 796
 797         if (!s)
 798                 panic("Out of memory when creating slab %s\n", name);
 799
 800         create_boot_cache(s, name, size, flags);
 801         list_add(&s->list, &slab_caches);
 802         s->refcount = 1;
 803         return s;
 804 }
 805
 806 struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1];
 807 EXPORT_SYMBOL(kmalloc_caches);
 808
 809 #ifdef CONFIG_ZONE_DMA
 810 struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1];
 811 EXPORT_SYMBOL(kmalloc_dma_caches);
 812 #endif
 813
 814 /*
 815  * Conversion table for small slabs sizes / 8 to the index in the
 816  * kmalloc array. This is necessary for slabs < 192 since we have non power
 817  * of two cache sizes there. The size of larger slabs can be determined using
 818  * fls.
 819  */
 820 static s8 size_index[24] = {
 821         3,      /* 8 */
 822         4,      /* 16 */
 823         5,      /* 24 */
 824         5,      /* 32 */
 825         6,      /* 40 */
 826         6,      /* 48 */
 827         6,      /* 56 */
 828         6,      /* 64 */
 829         1,      /* 72 */
 830         1,      /* 80 */
 831         1,      /* 88 */
 832         1,      /* 96 */
 833         7,      /* 104 */
 834         7,      /* 112 */
 835         7,      /* 120 */
 836         7,      /* 128 */
 837         2,      /* 136 */
 838         2,      /* 144 */
 839         2,      /* 152 */
 840         2,      /* 160 */
 841         2,      /* 168 */
 842         2,      /* 176 */
 843         2,      /* 184 */
 844         2       /* 192 */
 845 };
 846
 847 static inline int size_index_elem(size_t bytes)
 848 {
 849         return (bytes - 1) / 8;
 850 }
 851
 852 /*
 853  * Find the kmem_cache structure that serves a given size of
 854  * allocation
 855  */
 856 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
 857 {
 858         int index;
 859
 860         if (unlikely(size > KMALLOC_MAX_SIZE)) {
 861                 WARN_ON_ONCE(!(flags & __GFP_NOWARN));
 862                 return NULL;
 863         }
 864
 865         if (size <= 192) {
 866                 if (!size)
 867                         return ZERO_SIZE_PTR;
 868
 869                 index = size_index[size_index_elem(size)];
 870         } else
 871                 index = fls(size - 1);
 872
 873 #ifdef CONFIG_ZONE_DMA
 874         if (unlikely((flags & GFP_DMA)))
 875                 return kmalloc_dma_caches[index];
 876
 877 #endif
 878         return kmalloc_caches[index];
 879 }
 880
 881 /*
 882  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
 883  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
 884  * kmalloc-67108864.
 885  */
 886 static struct {
 887         const char *name;
 888         unsigned long size;
 889 } const kmalloc_info[] __initconst = {
 890         {NULL,                      0},         {"kmalloc-96",             96},
 891         {"kmalloc-192",           192},         {"kmalloc-8",               8},
 892         {"kmalloc-16",             16},         {"kmalloc-32",             32},
 893         {"kmalloc-64",             64},         {"kmalloc-128",           128},
 894         {"kmalloc-256",           256},         {"kmalloc-512",           512},
 895         {"kmalloc-1024",         1024},         {"kmalloc-2048",         2048},
 896         {"kmalloc-4096",         4096},         {"kmalloc-8192",         8192},
 897         {"kmalloc-16384",       16384},         {"kmalloc-32768",       32768},
 898         {"kmalloc-65536",       65536},         {"kmalloc-131072",     131072},
 899         {"kmalloc-262144",     262144},         {"kmalloc-524288",     524288},
 900         {"kmalloc-1048576",   1048576},         {"kmalloc-2097152",   2097152},
 901         {"kmalloc-4194304",   4194304},         {"kmalloc-8388608",   8388608},
 902         {"kmalloc-16777216", 16777216},         {"kmalloc-33554432", 33554432},
 903         {"kmalloc-67108864", 67108864}
 904 };
 905
 906 /*
 907  * Patch up the size_index table if we have strange large alignment
 908  * requirements for the kmalloc array. This is only the case for
 909  * MIPS it seems. The standard arches will not generate any code here.
 910  *
 911  * Largest permitted alignment is 256 bytes due to the way we
 912  * handle the index determination for the smaller caches.
 913  *
 914  * Make sure that nothing crazy happens if someone starts tinkering
 915  * around with ARCH_KMALLOC_MINALIGN
 916  */
 917 void __init setup_kmalloc_cache_index_table(void)
 918 {
 919         int i;
 920
 921         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
 922                 (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
 923
 924         for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
 925                 int elem = size_index_elem(i);
 926
 927                 if (elem >= ARRAY_SIZE(size_index))
 928                         break;
 929                 size_index[elem] = KMALLOC_SHIFT_LOW;
 930         }
 931
 932         if (KMALLOC_MIN_SIZE >= 64) {
 933                 /*
 934                  * The 96 byte size cache is not used if the alignment
 935                  * is 64 byte.
 936                  */
 937                 for (i = 64 + 8; i <= 96; i += 8)
 938                         size_index[size_index_elem(i)] = 7;
 939
 940         }
 941
 942         if (KMALLOC_MIN_SIZE >= 128) {
 943                 /*
 944                  * The 192 byte sized cache is not used if the alignment
 945                  * is 128 byte. Redirect kmalloc to use the 256 byte cache
 946                  * instead.
 947                  */
 948                 for (i = 128 + 8; i <= 192; i += 8)
 949                         size_index[size_index_elem(i)] = 8;
 950         }
 951 }
 952
 953 static void __init new_kmalloc_cache(int idx, unsigned long flags)
 954 {
 955         kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name,
 956                                         kmalloc_info[idx].size, flags);
 957 }
 958
 959 /*
 960  * Create the kmalloc array. Some of the regular kmalloc arrays
 961  * may already have been created because they were needed to
 962  * enable allocations for slab creation.
 963  */
 964 void __init create_kmalloc_caches(unsigned long flags)
 965 {
 966         int i;
 967
 968         for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
 969                 if (!kmalloc_caches[i])
 970                         new_kmalloc_cache(i, flags);
 971
 972                 /*
 973                  * Caches that are not of the two-to-the-power-of size.
 974                  * These have to be created immediately after the
 975                  * earlier power of two caches
 976                  */
 977                 if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6)
 978                         new_kmalloc_cache(1, flags);
 979                 if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7)
 980                         new_kmalloc_cache(2, flags);
 981         }
 982
 983         /* Kmalloc array is now usable */
 984         slab_state = UP;
 985
 986 #ifdef CONFIG_ZONE_DMA
 987         for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
 988                 struct kmem_cache *s = kmalloc_caches[i];
 989
 990                 if (s) {
 991                         int size = kmalloc_size(i);
 992                         char *n = kasprintf(GFP_NOWAIT,
 993                                  "dma-kmalloc-%d", size);
 994
 995                         BUG_ON(!n);
 996                         kmalloc_dma_caches[i] = create_kmalloc_cache(n,
 997                                 size, SLAB_CACHE_DMA | flags);
 998                 }
 999         }
1000 #endif
1001 }
1002 #endif /* !CONFIG_SLOB */
1003
1004 /*
1005  * To avoid unnecessary overhead, we pass through large allocation requests
1006  * directly to the page allocator. We use __GFP_COMP, because we will need to
1007  * know the allocation order to free the pages properly in kfree.
1008  */
1009 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
1010 {
1011         void *ret;
1012         struct page *page;
1013
1014         flags |= __GFP_COMP;
1015         page = alloc_pages(flags, order);
1016         ret = page ? page_address(page) : NULL;
1017         kmemleak_alloc(ret, size, 1, flags);
1018         kasan_kmalloc_large(ret, size, flags);
1019         return ret;
1020 }
1021 EXPORT_SYMBOL(kmalloc_order);
1022
1023 #ifdef CONFIG_TRACING
1024 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
1025 {
1026         void *ret = kmalloc_order(size, flags, order);
1027         trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
1028         return ret;
1029 }
1030 EXPORT_SYMBOL(kmalloc_order_trace);
1031 #endif
1032
1033 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1034 /* Randomize a generic freelist */
1035 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1036                         size_t count)
1037 {
1038         size_t i;
1039         unsigned int rand;
1040
1041         for (i = 0; i < count; i++)
1042                 list[i] = i;
1043
1044         /* Fisher-Yates shuffle */
1045         for (i = count - 1; i > 0; i--) {
1046                 rand = prandom_u32_state(state);
1047                 rand %= (i + 1);
1048                 swap(list[i], list[rand]);
1049         }
1050 }
1051
1052 /* Create a random sequence per cache */
1053 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1054                                     gfp_t gfp)
1055 {
1056         struct rnd_state state;
1057
1058         if (count < 2 || cachep->random_seq)
1059                 return 0;
1060
1061         cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1062         if (!cachep->random_seq)
1063                 return -ENOMEM;
1064
1065         /* Get best entropy at this stage of boot */
1066         prandom_seed_state(&state, get_random_long());
1067
1068         freelist_randomize(&state, cachep->random_seq, count);
1069         return 0;
1070 }
1071
1072 /* Destroy the per-cache random freelist sequence */
1073 void cache_random_seq_destroy(struct kmem_cache *cachep)
1074 {
1075         kfree(cachep->random_seq);
1076         cachep->random_seq = NULL;
1077 }
1078 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1079
1080 #ifdef CONFIG_SLABINFO
1081
1082 #ifdef CONFIG_SLAB
1083 #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR)
1084 #else
1085 #define SLABINFO_RIGHTS S_IRUSR
1086 #endif
1087
1088 static void print_slabinfo_header(struct seq_file *m)
1089 {
1090         /*
1091          * Output format version, so at least we can change it
1092          * without _too_ many complaints.
1093          */
1094 #ifdef CONFIG_DEBUG_SLAB
1095         seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1096 #else
1097         seq_puts(m, "slabinfo - version: 2.1\n");
1098 #endif
1099         seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1100         seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1101         seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1102 #ifdef CONFIG_DEBUG_SLAB
1103         seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1104         seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1105 #endif
1106         seq_putc(m, '\n');
1107 }
1108
1109 void *slab_start(struct seq_file *m, loff_t *pos)
1110 {
1111         mutex_lock(&slab_mutex);
1112         return seq_list_start(&slab_caches, *pos);
1113 }
1114
1115 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1116 {
1117         return seq_list_next(p, &slab_caches, pos);
1118 }
1119
1120 void slab_stop(struct seq_file *m, void *p)
1121 {
1122         mutex_unlock(&slab_mutex);
1123 }
1124
1125 static void
1126 memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info)
1127 {
1128         struct kmem_cache *c;
1129         struct slabinfo sinfo;
1130
1131         if (!is_root_cache(s))
1132                 return;
1133
1134         for_each_memcg_cache(c, s) {
1135                 memset(&sinfo, 0, sizeof(sinfo));
1136                 get_slabinfo(c, &sinfo);
1137
1138                 info->active_slabs += sinfo.active_slabs;
1139                 info->num_slabs += sinfo.num_slabs;
1140                 info->shared_avail += sinfo.shared_avail;
1141                 info->active_objs += sinfo.active_objs;
1142                 info->num_objs += sinfo.num_objs;
1143         }
1144 }
1145
1146 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1147 {
1148         struct slabinfo sinfo;
1149
1150         memset(&sinfo, 0, sizeof(sinfo));
1151         get_slabinfo(s, &sinfo);
1152
1153         memcg_accumulate_slabinfo(s, &sinfo);
1154
1155         seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1156                    cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size,
1157                    sinfo.objects_per_slab, (1 << sinfo.cache_order));
1158
1159         seq_printf(m, " : tunables %4u %4u %4u",
1160                    sinfo.limit, sinfo.batchcount, sinfo.shared);
1161         seq_printf(m, " : slabdata %6lu %6lu %6lu",
1162                    sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1163         slabinfo_show_stats(m, s);
1164         seq_putc(m, '\n');
1165 }
1166
1167 static int slab_show(struct seq_file *m, void *p)
1168 {
1169         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1170
1171         if (p == slab_caches.next)
1172                 print_slabinfo_header(m);
1173         if (is_root_cache(s))
1174                 cache_show(s, m);
1175         return 0;
1176 }
1177
1178 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
1179 int memcg_slab_show(struct seq_file *m, void *p)
1180 {
1181         struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1182         struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m));
1183
1184         if (p == slab_caches.next)
1185                 print_slabinfo_header(m);
1186         if (!is_root_cache(s) && s->memcg_params.memcg == memcg)
1187                 cache_show(s, m);
1188         return 0;
1189 }
1190 #endif
1191
1192 /*
1193  * slabinfo_op - iterator that generates /proc/slabinfo
1194  *
1195  * Output layout:
1196  * cache-name
1197  * num-active-objs
1198  * total-objs
1199  * object size
1200  * num-active-slabs
1201  * total-slabs
1202  * num-pages-per-slab
1203  * + further values on SMP and with statistics enabled
1204  */
1205 static const struct seq_operations slabinfo_op = {
1206         .start = slab_start,
1207         .next = slab_next,
1208         .stop = slab_stop,
1209         .show = slab_show,
1210 };
1211
1212 static int slabinfo_open(struct inode *inode, struct file *file)
1213 {
1214         return seq_open(file, &slabinfo_op);
1215 }
1216
1217 static const struct file_operations proc_slabinfo_operations = {
1218         .open           = slabinfo_open,
1219         .read           = seq_read,
1220         .write          = slabinfo_write,
1221         .llseek         = seq_lseek,
1222         .release        = seq_release,
1223 };
1224
1225 static int __init slab_proc_init(void)
1226 {
1227         proc_create("slabinfo", SLABINFO_RIGHTS, NULL,
1228                                                 &proc_slabinfo_operations);
1229         return 0;
1230 }
1231 module_init(slab_proc_init);
1232 #endif /* CONFIG_SLABINFO */
1233
1234 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1235                                            gfp_t flags)
1236 {
1237         void *ret;
1238         size_t ks = 0;
1239
1240         if (p)
1241                 ks = ksize(p);
1242
1243         if (ks >= new_size) {
1244                 kasan_krealloc((void *)p, new_size, flags);
1245                 return (void *)p;
1246         }
1247
1248         ret = kmalloc_track_caller(new_size, flags);
1249         if (ret && p)
1250                 memcpy(ret, p, ks);
1251
1252         return ret;
1253 }
1254
1255 /**
1256  * __krealloc - like krealloc() but don't free @p.
1257  * @p: object to reallocate memory for.
1258  * @new_size: how many bytes of memory are required.
1259  * @flags: the type of memory to allocate.
1260  *
1261  * This function is like krealloc() except it never frees the originally
1262  * allocated buffer. Use this if you don't want to free the buffer immediately
1263  * like, for example, with RCU.
1264  */
1265 void *__krealloc(const void *p, size_t new_size, gfp_t flags)
1266 {
1267         if (unlikely(!new_size))
1268                 return ZERO_SIZE_PTR;
1269
1270         return __do_krealloc(p, new_size, flags);
1271
1272 }
1273 EXPORT_SYMBOL(__krealloc);
1274
1275 /**
1276  * krealloc - reallocate memory. The contents will remain unchanged.
1277  * @p: object to reallocate memory for.
1278  * @new_size: how many bytes of memory are required.
1279  * @flags: the type of memory to allocate.
1280  *
1281  * The contents of the object pointed to are preserved up to the
1282  * lesser of the new and old sizes.  If @p is %NULL, krealloc()
1283  * behaves exactly like kmalloc().  If @new_size is 0 and @p is not a
1284  * %NULL pointer, the object pointed to is freed.
1285  */
1286 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1287 {
1288         void *ret;
1289
1290         if (unlikely(!new_size)) {
1291                 kfree(p);
1292                 return ZERO_SIZE_PTR;
1293         }
1294
1295         ret = __do_krealloc(p, new_size, flags);
1296         if (ret && p != ret)
1297                 kfree(p);
1298
1299         return ret;
1300 }
1301 EXPORT_SYMBOL(krealloc);
1302
1303 /**
1304  * kzfree - like kfree but zero memory
1305  * @p: object to free memory of
1306  *
1307  * The memory of the object @p points to is zeroed before freed.
1308  * If @p is %NULL, kzfree() does nothing.
1309  *
1310  * Note: this function zeroes the whole allocated buffer which can be a good
1311  * deal bigger than the requested buffer size passed to kmalloc(). So be
1312  * careful when using this function in performance sensitive code.
1313  */
1314 void kzfree(const void *p)
1315 {
1316         size_t ks;
1317         void *mem = (void *)p;
1318
1319         if (unlikely(ZERO_OR_NULL_PTR(mem)))
1320                 return;
1321         ks = ksize(mem);
1322         memset(mem, 0, ks);
1323         kfree(mem);
1324 }
1325 EXPORT_SYMBOL(kzfree);
1326
1327 /* Tracepoints definitions. */
1328 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1329 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1330 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1331 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1332 EXPORT_TRACEPOINT_SYMBOL(kfree);
1333 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);