root/mm/slub.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. kmem_cache_debug
  2. fixup_red_left
  3. kmem_cache_has_cpu_partial
  4. sysfs_slab_add
  5. sysfs_slab_alias
  6. memcg_propagate_slab_attrs
  7. sysfs_slab_remove
  8. freelist_ptr
  9. freelist_dereference
  10. get_freepointer
  11. prefetch_freepointer
  12. get_freepointer_safe
  13. set_freepointer
  14. slab_index
  15. order_objects
  16. oo_make
  17. oo_order
  18. oo_objects
  19. slab_lock
  20. slab_unlock
  21. __cmpxchg_double_slab
  22. cmpxchg_double_slab
  23. get_map
  24. size_from_object
  25. restore_red_left
  26. metadata_access_enable
  27. metadata_access_disable
  28. check_valid_pointer
  29. print_section
  30. get_track
  31. set_track
  32. init_tracking
  33. print_track
  34. print_tracking
  35. print_page_info
  36. slab_bug
  37. slab_fix
  38. print_trailer
  39. object_err
  40. __printf
  41. init_object
  42. restore_bytes
  43. check_bytes_and_report
  44. check_pad_bytes
  45. slab_pad_check
  46. check_object
  47. check_slab
  48. on_freelist
  49. trace
  50. add_full
  51. remove_full
  52. slabs_node
  53. node_nr_slabs
  54. inc_slabs_node
  55. dec_slabs_node
  56. setup_object_debug
  57. setup_page_debug
  58. alloc_consistency_checks
  59. alloc_debug_processing
  60. free_consistency_checks
  61. free_debug_processing
  62. setup_slub_debug
  63. kmem_cache_flags
  64. setup_object_debug
  65. setup_page_debug
  66. alloc_debug_processing
  67. free_debug_processing
  68. slab_pad_check
  69. check_object
  70. add_full
  71. remove_full
  72. kmem_cache_flags
  73. slabs_node
  74. node_nr_slabs
  75. inc_slabs_node
  76. dec_slabs_node
  77. kmalloc_large_node_hook
  78. kfree_hook
  79. slab_free_hook
  80. slab_free_freelist_hook
  81. setup_object
  82. alloc_slab_page
  83. init_cache_random_seq
  84. init_freelist_randomization
  85. next_freelist_entry
  86. shuffle_freelist
  87. init_cache_random_seq
  88. init_freelist_randomization
  89. shuffle_freelist
  90. allocate_slab
  91. new_slab
  92. __free_slab
  93. rcu_free_slab
  94. free_slab
  95. discard_slab
  96. __add_partial
  97. add_partial
  98. remove_partial
  99. acquire_slab
  100. get_partial_node
  101. get_any_partial
  102. get_partial
  103. next_tid
  104. tid_to_cpu
  105. tid_to_event
  106. init_tid
  107. note_cmpxchg_failure
  108. init_kmem_cache_cpus
  109. deactivate_slab
  110. unfreeze_partials
  111. put_cpu_partial
  112. flush_slab
  113. __flush_cpu_slab
  114. flush_cpu_slab
  115. has_cpu_slab
  116. flush_all
  117. slub_cpu_dead
  118. node_match
  119. count_free
  120. node_nr_objs
  121. count_partial
  122. slab_out_of_memory
  123. new_slab_objects
  124. pfmemalloc_match
  125. get_freelist
  126. ___slab_alloc
  127. __slab_alloc
  128. maybe_wipe_obj_freeptr
  129. slab_alloc_node
  130. slab_alloc
  131. kmem_cache_alloc
  132. kmem_cache_alloc_trace
  133. kmem_cache_alloc_node
  134. kmem_cache_alloc_node_trace
  135. __slab_free
  136. do_slab_free
  137. slab_free
  138. ___cache_free
  139. kmem_cache_free
  140. build_detached_freelist
  141. kmem_cache_free_bulk
  142. kmem_cache_alloc_bulk
  143. slab_order
  144. calculate_order
  145. init_kmem_cache_node
  146. alloc_kmem_cache_cpus
  147. early_kmem_cache_node_alloc
  148. free_kmem_cache_nodes
  149. __kmem_cache_release
  150. init_kmem_cache_nodes
  151. set_min_partial
  152. set_cpu_partial
  153. calculate_sizes
  154. kmem_cache_open
  155. list_slab_objects
  156. free_partial
  157. __kmem_cache_empty
  158. __kmem_cache_shutdown
  159. setup_slub_min_order
  160. setup_slub_max_order
  161. setup_slub_min_objects
  162. __kmalloc
  163. kmalloc_large_node
  164. __kmalloc_node
  165. __check_heap_object
  166. __ksize
  167. kfree
  168. __kmem_cache_shrink
  169. __kmemcg_cache_deactivate_after_rcu
  170. __kmemcg_cache_deactivate
  171. slab_mem_going_offline_callback
  172. slab_mem_offline_callback
  173. slab_mem_going_online_callback
  174. slab_memory_callback
  175. bootstrap
  176. kmem_cache_init
  177. kmem_cache_init_late
  178. __kmem_cache_alias
  179. __kmem_cache_create
  180. __kmalloc_track_caller
  181. __kmalloc_node_track_caller
  182. count_inuse
  183. count_total
  184. validate_slab
  185. validate_slab_slab
  186. validate_slab_node
  187. validate_slab_cache
  188. free_loc_track
  189. alloc_loc_track
  190. add_location
  191. process_slab
  192. list_locations
  193. resiliency_test
  194. resiliency_test
  195. setup_slub_memcg_sysfs
  196. show_slab_objects
  197. any_slab_objects
  198. slab_size_show
  199. align_show
  200. object_size_show
  201. objs_per_slab_show
  202. order_store
  203. order_show
  204. min_partial_show
  205. min_partial_store
  206. cpu_partial_show
  207. cpu_partial_store
  208. ctor_show
  209. aliases_show
  210. partial_show
  211. cpu_slabs_show
  212. objects_show
  213. objects_partial_show
  214. slabs_cpu_partial_show
  215. reclaim_account_show
  216. reclaim_account_store
  217. hwcache_align_show
  218. cache_dma_show
  219. usersize_show
  220. destroy_by_rcu_show
  221. slabs_show
  222. total_objects_show
  223. sanity_checks_show
  224. sanity_checks_store
  225. trace_show
  226. trace_store
  227. red_zone_show
  228. red_zone_store
  229. poison_show
  230. poison_store
  231. store_user_show
  232. store_user_store
  233. validate_show
  234. validate_store
  235. alloc_calls_show
  236. free_calls_show
  237. failslab_show
  238. failslab_store
  239. shrink_show
  240. shrink_store
  241. remote_node_defrag_ratio_show
  242. remote_node_defrag_ratio_store
  243. show_stat
  244. clear_stat
  245. slab_attr_show
  246. slab_attr_store
  247. memcg_propagate_slab_attrs
  248. kmem_cache_release
  249. uevent_filter
  250. cache_kset
  251. create_unique_id
  252. sysfs_slab_remove_workfn
  253. sysfs_slab_add
  254. sysfs_slab_remove
  255. sysfs_slab_unlink
  256. sysfs_slab_release
  257. sysfs_slab_alias
  258. slab_sysfs_init
  259. get_slabinfo
  260. slabinfo_show_stats
  261. slabinfo_write

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * SLUB: A slab allocator that limits cache line use instead of queuing
   4  * objects in per cpu and per node lists.
   5  *
   6  * The allocator synchronizes using per slab locks or atomic operatios
   7  * and only uses a centralized lock to manage a pool of partial slabs.
   8  *
   9  * (C) 2007 SGI, Christoph Lameter
  10  * (C) 2011 Linux Foundation, Christoph Lameter
  11  */
  12 
  13 #include <linux/mm.h>
  14 #include <linux/swap.h> /* struct reclaim_state */
  15 #include <linux/module.h>
  16 #include <linux/bit_spinlock.h>
  17 #include <linux/interrupt.h>
  18 #include <linux/bitops.h>
  19 #include <linux/slab.h>
  20 #include "slab.h"
  21 #include <linux/proc_fs.h>
  22 #include <linux/seq_file.h>
  23 #include <linux/kasan.h>
  24 #include <linux/cpu.h>
  25 #include <linux/cpuset.h>
  26 #include <linux/mempolicy.h>
  27 #include <linux/ctype.h>
  28 #include <linux/debugobjects.h>
  29 #include <linux/kallsyms.h>
  30 #include <linux/memory.h>
  31 #include <linux/math64.h>
  32 #include <linux/fault-inject.h>
  33 #include <linux/stacktrace.h>
  34 #include <linux/prefetch.h>
  35 #include <linux/memcontrol.h>
  36 #include <linux/random.h>
  37 
  38 #include <trace/events/kmem.h>
  39 
  40 #include "internal.h"
  41 
  42 /*
  43  * Lock order:
  44  *   1. slab_mutex (Global Mutex)
  45  *   2. node->list_lock
  46  *   3. slab_lock(page) (Only on some arches and for debugging)
  47  *
  48  *   slab_mutex
  49  *
  50  *   The role of the slab_mutex is to protect the list of all the slabs
  51  *   and to synchronize major metadata changes to slab cache structures.
  52  *
  53  *   The slab_lock is only used for debugging and on arches that do not
  54  *   have the ability to do a cmpxchg_double. It only protects:
  55  *      A. page->freelist       -> List of object free in a page
  56  *      B. page->inuse          -> Number of objects in use
  57  *      C. page->objects        -> Number of objects in page
  58  *      D. page->frozen         -> frozen state
  59  *
  60  *   If a slab is frozen then it is exempt from list management. It is not
  61  *   on any list except per cpu partial list. The processor that froze the
  62  *   slab is the one who can perform list operations on the page. Other
  63  *   processors may put objects onto the freelist but the processor that
  64  *   froze the slab is the only one that can retrieve the objects from the
  65  *   page's freelist.
  66  *
  67  *   The list_lock protects the partial and full list on each node and
  68  *   the partial slab counter. If taken then no new slabs may be added or
  69  *   removed from the lists nor make the number of partial slabs be modified.
  70  *   (Note that the total number of slabs is an atomic value that may be
  71  *   modified without taking the list lock).
  72  *
  73  *   The list_lock is a centralized lock and thus we avoid taking it as
  74  *   much as possible. As long as SLUB does not have to handle partial
  75  *   slabs, operations can continue without any centralized lock. F.e.
  76  *   allocating a long series of objects that fill up slabs does not require
  77  *   the list lock.
  78  *   Interrupts are disabled during allocation and deallocation in order to
  79  *   make the slab allocator safe to use in the context of an irq. In addition
  80  *   interrupts are disabled to ensure that the processor does not change
  81  *   while handling per_cpu slabs, due to kernel preemption.
  82  *
  83  * SLUB assigns one slab for allocation to each processor.
  84  * Allocations only occur from these slabs called cpu slabs.
  85  *
  86  * Slabs with free elements are kept on a partial list and during regular
  87  * operations no list for full slabs is used. If an object in a full slab is
  88  * freed then the slab will show up again on the partial lists.
  89  * We track full slabs for debugging purposes though because otherwise we
  90  * cannot scan all objects.
  91  *
  92  * Slabs are freed when they become empty. Teardown and setup is
  93  * minimal so we rely on the page allocators per cpu caches for
  94  * fast frees and allocs.
  95  *
  96  * Overloading of page flags that are otherwise used for LRU management.
  97  *
  98  * PageActive           The slab is frozen and exempt from list processing.
  99  *                      This means that the slab is dedicated to a purpose
 100  *                      such as satisfying allocations for a specific
 101  *                      processor. Objects may be freed in the slab while
 102  *                      it is frozen but slab_free will then skip the usual
 103  *                      list operations. It is up to the processor holding
 104  *                      the slab to integrate the slab into the slab lists
 105  *                      when the slab is no longer needed.
 106  *
 107  *                      One use of this flag is to mark slabs that are
 108  *                      used for allocations. Then such a slab becomes a cpu
 109  *                      slab. The cpu slab may be equipped with an additional
 110  *                      freelist that allows lockless access to
 111  *                      free objects in addition to the regular freelist
 112  *                      that requires the slab lock.
 113  *
 114  * PageError            Slab requires special handling due to debug
 115  *                      options set. This moves slab handling out of
 116  *                      the fast path and disables lockless freelists.
 117  */
 118 
 119 static inline int kmem_cache_debug(struct kmem_cache *s)
 120 {
 121 #ifdef CONFIG_SLUB_DEBUG
 122         return unlikely(s->flags & SLAB_DEBUG_FLAGS);
 123 #else
 124         return 0;
 125 #endif
 126 }
 127 
 128 void *fixup_red_left(struct kmem_cache *s, void *p)
 129 {
 130         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
 131                 p += s->red_left_pad;
 132 
 133         return p;
 134 }
 135 
 136 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
 137 {
 138 #ifdef CONFIG_SLUB_CPU_PARTIAL
 139         return !kmem_cache_debug(s);
 140 #else
 141         return false;
 142 #endif
 143 }
 144 
 145 /*
 146  * Issues still to be resolved:
 147  *
 148  * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
 149  *
 150  * - Variable sizing of the per node arrays
 151  */
 152 
 153 /* Enable to test recovery from slab corruption on boot */
 154 #undef SLUB_RESILIENCY_TEST
 155 
 156 /* Enable to log cmpxchg failures */
 157 #undef SLUB_DEBUG_CMPXCHG
 158 
 159 /*
 160  * Mininum number of partial slabs. These will be left on the partial
 161  * lists even if they are empty. kmem_cache_shrink may reclaim them.
 162  */
 163 #define MIN_PARTIAL 5
 164 
 165 /*
 166  * Maximum number of desirable partial slabs.
 167  * The existence of more partial slabs makes kmem_cache_shrink
 168  * sort the partial list by the number of objects in use.
 169  */
 170 #define MAX_PARTIAL 10
 171 
 172 #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
 173                                 SLAB_POISON | SLAB_STORE_USER)
 174 
 175 /*
 176  * These debug flags cannot use CMPXCHG because there might be consistency
 177  * issues when checking or reading debug information
 178  */
 179 #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
 180                                 SLAB_TRACE)
 181 
 182 
 183 /*
 184  * Debugging flags that require metadata to be stored in the slab.  These get
 185  * disabled when slub_debug=O is used and a cache's min order increases with
 186  * metadata.
 187  */
 188 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
 189 
 190 #define OO_SHIFT        16
 191 #define OO_MASK         ((1 << OO_SHIFT) - 1)
 192 #define MAX_OBJS_PER_PAGE       32767 /* since page.objects is u15 */
 193 
 194 /* Internal SLUB flags */
 195 /* Poison object */
 196 #define __OBJECT_POISON         ((slab_flags_t __force)0x80000000U)
 197 /* Use cmpxchg_double */
 198 #define __CMPXCHG_DOUBLE        ((slab_flags_t __force)0x40000000U)
 199 
 200 /*
 201  * Tracking user of a slab.
 202  */
 203 #define TRACK_ADDRS_COUNT 16
 204 struct track {
 205         unsigned long addr;     /* Called from address */
 206 #ifdef CONFIG_STACKTRACE
 207         unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
 208 #endif
 209         int cpu;                /* Was running on cpu */
 210         int pid;                /* Pid context */
 211         unsigned long when;     /* When did the operation occur */
 212 };
 213 
 214 enum track_item { TRACK_ALLOC, TRACK_FREE };
 215 
 216 #ifdef CONFIG_SYSFS
 217 static int sysfs_slab_add(struct kmem_cache *);
 218 static int sysfs_slab_alias(struct kmem_cache *, const char *);
 219 static void memcg_propagate_slab_attrs(struct kmem_cache *s);
 220 static void sysfs_slab_remove(struct kmem_cache *s);
 221 #else
 222 static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
 223 static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
 224                                                         { return 0; }
 225 static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
 226 static inline void sysfs_slab_remove(struct kmem_cache *s) { }
 227 #endif
 228 
 229 static inline void stat(const struct kmem_cache *s, enum stat_item si)
 230 {
 231 #ifdef CONFIG_SLUB_STATS
 232         /*
 233          * The rmw is racy on a preemptible kernel but this is acceptable, so
 234          * avoid this_cpu_add()'s irq-disable overhead.
 235          */
 236         raw_cpu_inc(s->cpu_slab->stat[si]);
 237 #endif
 238 }
 239 
 240 /********************************************************************
 241  *                      Core slab cache functions
 242  *******************************************************************/
 243 
 244 /*
 245  * Returns freelist pointer (ptr). With hardening, this is obfuscated
 246  * with an XOR of the address where the pointer is held and a per-cache
 247  * random number.
 248  */
 249 static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
 250                                  unsigned long ptr_addr)
 251 {
 252 #ifdef CONFIG_SLAB_FREELIST_HARDENED
 253         /*
 254          * When CONFIG_KASAN_SW_TAGS is enabled, ptr_addr might be tagged.
 255          * Normally, this doesn't cause any issues, as both set_freepointer()
 256          * and get_freepointer() are called with a pointer with the same tag.
 257          * However, there are some issues with CONFIG_SLUB_DEBUG code. For
 258          * example, when __free_slub() iterates over objects in a cache, it
 259          * passes untagged pointers to check_object(). check_object() in turns
 260          * calls get_freepointer() with an untagged pointer, which causes the
 261          * freepointer to be restored incorrectly.
 262          */
 263         return (void *)((unsigned long)ptr ^ s->random ^
 264                         swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
 265 #else
 266         return ptr;
 267 #endif
 268 }
 269 
 270 /* Returns the freelist pointer recorded at location ptr_addr. */
 271 static inline void *freelist_dereference(const struct kmem_cache *s,
 272                                          void *ptr_addr)
 273 {
 274         return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
 275                             (unsigned long)ptr_addr);
 276 }
 277 
 278 static inline void *get_freepointer(struct kmem_cache *s, void *object)
 279 {
 280         return freelist_dereference(s, object + s->offset);
 281 }
 282 
 283 static void prefetch_freepointer(const struct kmem_cache *s, void *object)
 284 {
 285         prefetch(object + s->offset);
 286 }
 287 
 288 static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
 289 {
 290         unsigned long freepointer_addr;
 291         void *p;
 292 
 293         if (!debug_pagealloc_enabled_static())
 294                 return get_freepointer(s, object);
 295 
 296         freepointer_addr = (unsigned long)object + s->offset;
 297         probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p));
 298         return freelist_ptr(s, p, freepointer_addr);
 299 }
 300 
 301 static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
 302 {
 303         unsigned long freeptr_addr = (unsigned long)object + s->offset;
 304 
 305 #ifdef CONFIG_SLAB_FREELIST_HARDENED
 306         BUG_ON(object == fp); /* naive detection of double free or corruption */
 307 #endif
 308 
 309         *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
 310 }
 311 
 312 /* Loop over all objects in a slab */
 313 #define for_each_object(__p, __s, __addr, __objects) \
 314         for (__p = fixup_red_left(__s, __addr); \
 315                 __p < (__addr) + (__objects) * (__s)->size; \
 316                 __p += (__s)->size)
 317 
 318 /* Determine object index from a given position */
 319 static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr)
 320 {
 321         return (kasan_reset_tag(p) - addr) / s->size;
 322 }
 323 
 324 static inline unsigned int order_objects(unsigned int order, unsigned int size)
 325 {
 326         return ((unsigned int)PAGE_SIZE << order) / size;
 327 }
 328 
 329 static inline struct kmem_cache_order_objects oo_make(unsigned int order,
 330                 unsigned int size)
 331 {
 332         struct kmem_cache_order_objects x = {
 333                 (order << OO_SHIFT) + order_objects(order, size)
 334         };
 335 
 336         return x;
 337 }
 338 
 339 static inline unsigned int oo_order(struct kmem_cache_order_objects x)
 340 {
 341         return x.x >> OO_SHIFT;
 342 }
 343 
 344 static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
 345 {
 346         return x.x & OO_MASK;
 347 }
 348 
 349 /*
 350  * Per slab locking using the pagelock
 351  */
 352 static __always_inline void slab_lock(struct page *page)
 353 {
 354         VM_BUG_ON_PAGE(PageTail(page), page);
 355         bit_spin_lock(PG_locked, &page->flags);
 356 }
 357 
 358 static __always_inline void slab_unlock(struct page *page)
 359 {
 360         VM_BUG_ON_PAGE(PageTail(page), page);
 361         __bit_spin_unlock(PG_locked, &page->flags);
 362 }
 363 
 364 /* Interrupts must be disabled (for the fallback code to work right) */
 365 static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
 366                 void *freelist_old, unsigned long counters_old,
 367                 void *freelist_new, unsigned long counters_new,
 368                 const char *n)
 369 {
 370         VM_BUG_ON(!irqs_disabled());
 371 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
 372     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
 373         if (s->flags & __CMPXCHG_DOUBLE) {
 374                 if (cmpxchg_double(&page->freelist, &page->counters,
 375                                    freelist_old, counters_old,
 376                                    freelist_new, counters_new))
 377                         return true;
 378         } else
 379 #endif
 380         {
 381                 slab_lock(page);
 382                 if (page->freelist == freelist_old &&
 383                                         page->counters == counters_old) {
 384                         page->freelist = freelist_new;
 385                         page->counters = counters_new;
 386                         slab_unlock(page);
 387                         return true;
 388                 }
 389                 slab_unlock(page);
 390         }
 391 
 392         cpu_relax();
 393         stat(s, CMPXCHG_DOUBLE_FAIL);
 394 
 395 #ifdef SLUB_DEBUG_CMPXCHG
 396         pr_info("%s %s: cmpxchg double redo ", n, s->name);
 397 #endif
 398 
 399         return false;
 400 }
 401 
 402 static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
 403                 void *freelist_old, unsigned long counters_old,
 404                 void *freelist_new, unsigned long counters_new,
 405                 const char *n)
 406 {
 407 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
 408     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
 409         if (s->flags & __CMPXCHG_DOUBLE) {
 410                 if (cmpxchg_double(&page->freelist, &page->counters,
 411                                    freelist_old, counters_old,
 412                                    freelist_new, counters_new))
 413                         return true;
 414         } else
 415 #endif
 416         {
 417                 unsigned long flags;
 418 
 419                 local_irq_save(flags);
 420                 slab_lock(page);
 421                 if (page->freelist == freelist_old &&
 422                                         page->counters == counters_old) {
 423                         page->freelist = freelist_new;
 424                         page->counters = counters_new;
 425                         slab_unlock(page);
 426                         local_irq_restore(flags);
 427                         return true;
 428                 }
 429                 slab_unlock(page);
 430                 local_irq_restore(flags);
 431         }
 432 
 433         cpu_relax();
 434         stat(s, CMPXCHG_DOUBLE_FAIL);
 435 
 436 #ifdef SLUB_DEBUG_CMPXCHG
 437         pr_info("%s %s: cmpxchg double redo ", n, s->name);
 438 #endif
 439 
 440         return false;
 441 }
 442 
 443 #ifdef CONFIG_SLUB_DEBUG
 444 /*
 445  * Determine a map of object in use on a page.
 446  *
 447  * Node listlock must be held to guarantee that the page does
 448  * not vanish from under us.
 449  */
 450 static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
 451 {
 452         void *p;
 453         void *addr = page_address(page);
 454 
 455         for (p = page->freelist; p; p = get_freepointer(s, p))
 456                 set_bit(slab_index(p, s, addr), map);
 457 }
 458 
 459 static inline unsigned int size_from_object(struct kmem_cache *s)
 460 {
 461         if (s->flags & SLAB_RED_ZONE)
 462                 return s->size - s->red_left_pad;
 463 
 464         return s->size;
 465 }
 466 
 467 static inline void *restore_red_left(struct kmem_cache *s, void *p)
 468 {
 469         if (s->flags & SLAB_RED_ZONE)
 470                 p -= s->red_left_pad;
 471 
 472         return p;
 473 }
 474 
 475 /*
 476  * Debug settings:
 477  */
 478 #if defined(CONFIG_SLUB_DEBUG_ON)
 479 static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
 480 #else
 481 static slab_flags_t slub_debug;
 482 #endif
 483 
 484 static char *slub_debug_slabs;
 485 static int disable_higher_order_debug;
 486 
 487 /*
 488  * slub is about to manipulate internal object metadata.  This memory lies
 489  * outside the range of the allocated object, so accessing it would normally
 490  * be reported by kasan as a bounds error.  metadata_access_enable() is used
 491  * to tell kasan that these accesses are OK.
 492  */
 493 static inline void metadata_access_enable(void)
 494 {
 495         kasan_disable_current();
 496 }
 497 
 498 static inline void metadata_access_disable(void)
 499 {
 500         kasan_enable_current();
 501 }
 502 
 503 /*
 504  * Object debugging
 505  */
 506 
 507 /* Verify that a pointer has an address that is valid within a slab page */
 508 static inline int check_valid_pointer(struct kmem_cache *s,
 509                                 struct page *page, void *object)
 510 {
 511         void *base;
 512 
 513         if (!object)
 514                 return 1;
 515 
 516         base = page_address(page);
 517         object = kasan_reset_tag(object);
 518         object = restore_red_left(s, object);
 519         if (object < base || object >= base + page->objects * s->size ||
 520                 (object - base) % s->size) {
 521                 return 0;
 522         }
 523 
 524         return 1;
 525 }
 526 
 527 static void print_section(char *level, char *text, u8 *addr,
 528                           unsigned int length)
 529 {
 530         metadata_access_enable();
 531         print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
 532                         length, 1);
 533         metadata_access_disable();
 534 }
 535 
 536 static struct track *get_track(struct kmem_cache *s, void *object,
 537         enum track_item alloc)
 538 {
 539         struct track *p;
 540 
 541         if (s->offset)
 542                 p = object + s->offset + sizeof(void *);
 543         else
 544                 p = object + s->inuse;
 545 
 546         return p + alloc;
 547 }
 548 
 549 static void set_track(struct kmem_cache *s, void *object,
 550                         enum track_item alloc, unsigned long addr)
 551 {
 552         struct track *p = get_track(s, object, alloc);
 553 
 554         if (addr) {
 555 #ifdef CONFIG_STACKTRACE
 556                 unsigned int nr_entries;
 557 
 558                 metadata_access_enable();
 559                 nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3);
 560                 metadata_access_disable();
 561 
 562                 if (nr_entries < TRACK_ADDRS_COUNT)
 563                         p->addrs[nr_entries] = 0;
 564 #endif
 565                 p->addr = addr;
 566                 p->cpu = smp_processor_id();
 567                 p->pid = current->pid;
 568                 p->when = jiffies;
 569         } else {
 570                 memset(p, 0, sizeof(struct track));
 571         }
 572 }
 573 
 574 static void init_tracking(struct kmem_cache *s, void *object)
 575 {
 576         if (!(s->flags & SLAB_STORE_USER))
 577                 return;
 578 
 579         set_track(s, object, TRACK_FREE, 0UL);
 580         set_track(s, object, TRACK_ALLOC, 0UL);
 581 }
 582 
 583 static void print_track(const char *s, struct track *t, unsigned long pr_time)
 584 {
 585         if (!t->addr)
 586                 return;
 587 
 588         pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
 589                s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
 590 #ifdef CONFIG_STACKTRACE
 591         {
 592                 int i;
 593                 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
 594                         if (t->addrs[i])
 595                                 pr_err("\t%pS\n", (void *)t->addrs[i]);
 596                         else
 597                                 break;
 598         }
 599 #endif
 600 }
 601 
 602 static void print_tracking(struct kmem_cache *s, void *object)
 603 {
 604         unsigned long pr_time = jiffies;
 605         if (!(s->flags & SLAB_STORE_USER))
 606                 return;
 607 
 608         print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
 609         print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
 610 }
 611 
 612 static void print_page_info(struct page *page)
 613 {
 614         pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
 615                page, page->objects, page->inuse, page->freelist, page->flags);
 616 
 617 }
 618 
 619 static void slab_bug(struct kmem_cache *s, char *fmt, ...)
 620 {
 621         struct va_format vaf;
 622         va_list args;
 623 
 624         va_start(args, fmt);
 625         vaf.fmt = fmt;
 626         vaf.va = &args;
 627         pr_err("=============================================================================\n");
 628         pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
 629         pr_err("-----------------------------------------------------------------------------\n\n");
 630 
 631         add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 632         va_end(args);
 633 }
 634 
 635 static void slab_fix(struct kmem_cache *s, char *fmt, ...)
 636 {
 637         struct va_format vaf;
 638         va_list args;
 639 
 640         va_start(args, fmt);
 641         vaf.fmt = fmt;
 642         vaf.va = &args;
 643         pr_err("FIX %s: %pV\n", s->name, &vaf);
 644         va_end(args);
 645 }
 646 
 647 static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
 648 {
 649         unsigned int off;       /* Offset of last byte */
 650         u8 *addr = page_address(page);
 651 
 652         print_tracking(s, p);
 653 
 654         print_page_info(page);
 655 
 656         pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
 657                p, p - addr, get_freepointer(s, p));
 658 
 659         if (s->flags & SLAB_RED_ZONE)
 660                 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
 661                               s->red_left_pad);
 662         else if (p > addr + 16)
 663                 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
 664 
 665         print_section(KERN_ERR, "Object ", p,
 666                       min_t(unsigned int, s->object_size, PAGE_SIZE));
 667         if (s->flags & SLAB_RED_ZONE)
 668                 print_section(KERN_ERR, "Redzone ", p + s->object_size,
 669                         s->inuse - s->object_size);
 670 
 671         if (s->offset)
 672                 off = s->offset + sizeof(void *);
 673         else
 674                 off = s->inuse;
 675 
 676         if (s->flags & SLAB_STORE_USER)
 677                 off += 2 * sizeof(struct track);
 678 
 679         off += kasan_metadata_size(s);
 680 
 681         if (off != size_from_object(s))
 682                 /* Beginning of the filler is the free pointer */
 683                 print_section(KERN_ERR, "Padding ", p + off,
 684                               size_from_object(s) - off);
 685 
 686         dump_stack();
 687 }
 688 
 689 void object_err(struct kmem_cache *s, struct page *page,
 690                         u8 *object, char *reason)
 691 {
 692         slab_bug(s, "%s", reason);
 693         print_trailer(s, page, object);
 694 }
 695 
 696 static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
 697                         const char *fmt, ...)
 698 {
 699         va_list args;
 700         char buf[100];
 701 
 702         va_start(args, fmt);
 703         vsnprintf(buf, sizeof(buf), fmt, args);
 704         va_end(args);
 705         slab_bug(s, "%s", buf);
 706         print_page_info(page);
 707         dump_stack();
 708 }
 709 
 710 static void init_object(struct kmem_cache *s, void *object, u8 val)
 711 {
 712         u8 *p = object;
 713 
 714         if (s->flags & SLAB_RED_ZONE)
 715                 memset(p - s->red_left_pad, val, s->red_left_pad);
 716 
 717         if (s->flags & __OBJECT_POISON) {
 718                 memset(p, POISON_FREE, s->object_size - 1);
 719                 p[s->object_size - 1] = POISON_END;
 720         }
 721 
 722         if (s->flags & SLAB_RED_ZONE)
 723                 memset(p + s->object_size, val, s->inuse - s->object_size);
 724 }
 725 
 726 static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
 727                                                 void *from, void *to)
 728 {
 729         slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
 730         memset(from, data, to - from);
 731 }
 732 
 733 static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
 734                         u8 *object, char *what,
 735                         u8 *start, unsigned int value, unsigned int bytes)
 736 {
 737         u8 *fault;
 738         u8 *end;
 739 
 740         metadata_access_enable();
 741         fault = memchr_inv(start, value, bytes);
 742         metadata_access_disable();
 743         if (!fault)
 744                 return 1;
 745 
 746         end = start + bytes;
 747         while (end > fault && end[-1] == value)
 748                 end--;
 749 
 750         slab_bug(s, "%s overwritten", what);
 751         pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
 752                                         fault, end - 1, fault[0], value);
 753         print_trailer(s, page, object);
 754 
 755         restore_bytes(s, what, value, fault, end);
 756         return 0;
 757 }
 758 
 759 /*
 760  * Object layout:
 761  *
 762  * object address
 763  *      Bytes of the object to be managed.
 764  *      If the freepointer may overlay the object then the free
 765  *      pointer is the first word of the object.
 766  *
 767  *      Poisoning uses 0x6b (POISON_FREE) and the last byte is
 768  *      0xa5 (POISON_END)
 769  *
 770  * object + s->object_size
 771  *      Padding to reach word boundary. This is also used for Redzoning.
 772  *      Padding is extended by another word if Redzoning is enabled and
 773  *      object_size == inuse.
 774  *
 775  *      We fill with 0xbb (RED_INACTIVE) for inactive objects and with
 776  *      0xcc (RED_ACTIVE) for objects in use.
 777  *
 778  * object + s->inuse
 779  *      Meta data starts here.
 780  *
 781  *      A. Free pointer (if we cannot overwrite object on free)
 782  *      B. Tracking data for SLAB_STORE_USER
 783  *      C. Padding to reach required alignment boundary or at mininum
 784  *              one word if debugging is on to be able to detect writes
 785  *              before the word boundary.
 786  *
 787  *      Padding is done using 0x5a (POISON_INUSE)
 788  *
 789  * object + s->size
 790  *      Nothing is used beyond s->size.
 791  *
 792  * If slabcaches are merged then the object_size and inuse boundaries are mostly
 793  * ignored. And therefore no slab options that rely on these boundaries
 794  * may be used with merged slabcaches.
 795  */
 796 
 797 static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
 798 {
 799         unsigned long off = s->inuse;   /* The end of info */
 800 
 801         if (s->offset)
 802                 /* Freepointer is placed after the object. */
 803                 off += sizeof(void *);
 804 
 805         if (s->flags & SLAB_STORE_USER)
 806                 /* We also have user information there */
 807                 off += 2 * sizeof(struct track);
 808 
 809         off += kasan_metadata_size(s);
 810 
 811         if (size_from_object(s) == off)
 812                 return 1;
 813 
 814         return check_bytes_and_report(s, page, p, "Object padding",
 815                         p + off, POISON_INUSE, size_from_object(s) - off);
 816 }
 817 
 818 /* Check the pad bytes at the end of a slab page */
 819 static int slab_pad_check(struct kmem_cache *s, struct page *page)
 820 {
 821         u8 *start;
 822         u8 *fault;
 823         u8 *end;
 824         u8 *pad;
 825         int length;
 826         int remainder;
 827 
 828         if (!(s->flags & SLAB_POISON))
 829                 return 1;
 830 
 831         start = page_address(page);
 832         length = page_size(page);
 833         end = start + length;
 834         remainder = length % s->size;
 835         if (!remainder)
 836                 return 1;
 837 
 838         pad = end - remainder;
 839         metadata_access_enable();
 840         fault = memchr_inv(pad, POISON_INUSE, remainder);
 841         metadata_access_disable();
 842         if (!fault)
 843                 return 1;
 844         while (end > fault && end[-1] == POISON_INUSE)
 845                 end--;
 846 
 847         slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
 848         print_section(KERN_ERR, "Padding ", pad, remainder);
 849 
 850         restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
 851         return 0;
 852 }
 853 
 854 static int check_object(struct kmem_cache *s, struct page *page,
 855                                         void *object, u8 val)
 856 {
 857         u8 *p = object;
 858         u8 *endobject = object + s->object_size;
 859 
 860         if (s->flags & SLAB_RED_ZONE) {
 861                 if (!check_bytes_and_report(s, page, object, "Redzone",
 862                         object - s->red_left_pad, val, s->red_left_pad))
 863                         return 0;
 864 
 865                 if (!check_bytes_and_report(s, page, object, "Redzone",
 866                         endobject, val, s->inuse - s->object_size))
 867                         return 0;
 868         } else {
 869                 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
 870                         check_bytes_and_report(s, page, p, "Alignment padding",
 871                                 endobject, POISON_INUSE,
 872                                 s->inuse - s->object_size);
 873                 }
 874         }
 875 
 876         if (s->flags & SLAB_POISON) {
 877                 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
 878                         (!check_bytes_and_report(s, page, p, "Poison", p,
 879                                         POISON_FREE, s->object_size - 1) ||
 880                          !check_bytes_and_report(s, page, p, "Poison",
 881                                 p + s->object_size - 1, POISON_END, 1)))
 882                         return 0;
 883                 /*
 884                  * check_pad_bytes cleans up on its own.
 885                  */
 886                 check_pad_bytes(s, page, p);
 887         }
 888 
 889         if (!s->offset && val == SLUB_RED_ACTIVE)
 890                 /*
 891                  * Object and freepointer overlap. Cannot check
 892                  * freepointer while object is allocated.
 893                  */
 894                 return 1;
 895 
 896         /* Check free pointer validity */
 897         if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
 898                 object_err(s, page, p, "Freepointer corrupt");
 899                 /*
 900                  * No choice but to zap it and thus lose the remainder
 901                  * of the free objects in this slab. May cause
 902                  * another error because the object count is now wrong.
 903                  */
 904                 set_freepointer(s, p, NULL);
 905                 return 0;
 906         }
 907         return 1;
 908 }
 909 
 910 static int check_slab(struct kmem_cache *s, struct page *page)
 911 {
 912         int maxobj;
 913 
 914         VM_BUG_ON(!irqs_disabled());
 915 
 916         if (!PageSlab(page)) {
 917                 slab_err(s, page, "Not a valid slab page");
 918                 return 0;
 919         }
 920 
 921         maxobj = order_objects(compound_order(page), s->size);
 922         if (page->objects > maxobj) {
 923                 slab_err(s, page, "objects %u > max %u",
 924                         page->objects, maxobj);
 925                 return 0;
 926         }
 927         if (page->inuse > page->objects) {
 928                 slab_err(s, page, "inuse %u > max %u",
 929                         page->inuse, page->objects);
 930                 return 0;
 931         }
 932         /* Slab_pad_check fixes things up after itself */
 933         slab_pad_check(s, page);
 934         return 1;
 935 }
 936 
 937 /*
 938  * Determine if a certain object on a page is on the freelist. Must hold the
 939  * slab lock to guarantee that the chains are in a consistent state.
 940  */
 941 static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
 942 {
 943         int nr = 0;
 944         void *fp;
 945         void *object = NULL;
 946         int max_objects;
 947 
 948         fp = page->freelist;
 949         while (fp && nr <= page->objects) {
 950                 if (fp == search)
 951                         return 1;
 952                 if (!check_valid_pointer(s, page, fp)) {
 953                         if (object) {
 954                                 object_err(s, page, object,
 955                                         "Freechain corrupt");
 956                                 set_freepointer(s, object, NULL);
 957                         } else {
 958                                 slab_err(s, page, "Freepointer corrupt");
 959                                 page->freelist = NULL;
 960                                 page->inuse = page->objects;
 961                                 slab_fix(s, "Freelist cleared");
 962                                 return 0;
 963                         }
 964                         break;
 965                 }
 966                 object = fp;
 967                 fp = get_freepointer(s, object);
 968                 nr++;
 969         }
 970 
 971         max_objects = order_objects(compound_order(page), s->size);
 972         if (max_objects > MAX_OBJS_PER_PAGE)
 973                 max_objects = MAX_OBJS_PER_PAGE;
 974 
 975         if (page->objects != max_objects) {
 976                 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
 977                          page->objects, max_objects);
 978                 page->objects = max_objects;
 979                 slab_fix(s, "Number of objects adjusted.");
 980         }
 981         if (page->inuse != page->objects - nr) {
 982                 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
 983                          page->inuse, page->objects - nr);
 984                 page->inuse = page->objects - nr;
 985                 slab_fix(s, "Object count adjusted.");
 986         }
 987         return search == NULL;
 988 }
 989 
 990 static void trace(struct kmem_cache *s, struct page *page, void *object,
 991                                                                 int alloc)
 992 {
 993         if (s->flags & SLAB_TRACE) {
 994                 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
 995                         s->name,
 996                         alloc ? "alloc" : "free",
 997                         object, page->inuse,
 998                         page->freelist);
 999 
1000                 if (!alloc)
1001                         print_section(KERN_INFO, "Object ", (void *)object,
1002                                         s->object_size);
1003 
1004                 dump_stack();
1005         }
1006 }
1007 
1008 /*
1009  * Tracking of fully allocated slabs for debugging purposes.
1010  */
1011 static void add_full(struct kmem_cache *s,
1012         struct kmem_cache_node *n, struct page *page)
1013 {
1014         if (!(s->flags & SLAB_STORE_USER))
1015                 return;
1016 
1017         lockdep_assert_held(&n->list_lock);
1018         list_add(&page->slab_list, &n->full);
1019 }
1020 
1021 static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
1022 {
1023         if (!(s->flags & SLAB_STORE_USER))
1024                 return;
1025 
1026         lockdep_assert_held(&n->list_lock);
1027         list_del(&page->slab_list);
1028 }
1029 
1030 /* Tracking of the number of slabs for debugging purposes */
1031 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1032 {
1033         struct kmem_cache_node *n = get_node(s, node);
1034 
1035         return atomic_long_read(&n->nr_slabs);
1036 }
1037 
1038 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1039 {
1040         return atomic_long_read(&n->nr_slabs);
1041 }
1042 
1043 static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
1044 {
1045         struct kmem_cache_node *n = get_node(s, node);
1046 
1047         /*
1048          * May be called early in order to allocate a slab for the
1049          * kmem_cache_node structure. Solve the chicken-egg
1050          * dilemma by deferring the increment of the count during
1051          * bootstrap (see early_kmem_cache_node_alloc).
1052          */
1053         if (likely(n)) {
1054                 atomic_long_inc(&n->nr_slabs);
1055                 atomic_long_add(objects, &n->total_objects);
1056         }
1057 }
1058 static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
1059 {
1060         struct kmem_cache_node *n = get_node(s, node);
1061 
1062         atomic_long_dec(&n->nr_slabs);
1063         atomic_long_sub(objects, &n->total_objects);
1064 }
1065 
1066 /* Object debug checks for alloc/free paths */
1067 static void setup_object_debug(struct kmem_cache *s, struct page *page,
1068                                                                 void *object)
1069 {
1070         if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1071                 return;
1072 
1073         init_object(s, object, SLUB_RED_INACTIVE);
1074         init_tracking(s, object);
1075 }
1076 
1077 static
1078 void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr)
1079 {
1080         if (!(s->flags & SLAB_POISON))
1081                 return;
1082 
1083         metadata_access_enable();
1084         memset(addr, POISON_INUSE, page_size(page));
1085         metadata_access_disable();
1086 }
1087 
1088 static inline int alloc_consistency_checks(struct kmem_cache *s,
1089                                         struct page *page, void *object)
1090 {
1091         if (!check_slab(s, page))
1092                 return 0;
1093 
1094         if (!check_valid_pointer(s, page, object)) {
1095                 object_err(s, page, object, "Freelist Pointer check fails");
1096                 return 0;
1097         }
1098 
1099         if (!check_object(s, page, object, SLUB_RED_INACTIVE))
1100                 return 0;
1101 
1102         return 1;
1103 }
1104 
1105 static noinline int alloc_debug_processing(struct kmem_cache *s,
1106                                         struct page *page,
1107                                         void *object, unsigned long addr)
1108 {
1109         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1110                 if (!alloc_consistency_checks(s, page, object))
1111                         goto bad;
1112         }
1113 
1114         /* Success perform special debug activities for allocs */
1115         if (s->flags & SLAB_STORE_USER)
1116                 set_track(s, object, TRACK_ALLOC, addr);
1117         trace(s, page, object, 1);
1118         init_object(s, object, SLUB_RED_ACTIVE);
1119         return 1;
1120 
1121 bad:
1122         if (PageSlab(page)) {
1123                 /*
1124                  * If this is a slab page then lets do the best we can
1125                  * to avoid issues in the future. Marking all objects
1126                  * as used avoids touching the remaining objects.
1127                  */
1128                 slab_fix(s, "Marking all objects used");
1129                 page->inuse = page->objects;
1130                 page->freelist = NULL;
1131         }
1132         return 0;
1133 }
1134 
1135 static inline int free_consistency_checks(struct kmem_cache *s,
1136                 struct page *page, void *object, unsigned long addr)
1137 {
1138         if (!check_valid_pointer(s, page, object)) {
1139                 slab_err(s, page, "Invalid object pointer 0x%p", object);
1140                 return 0;
1141         }
1142 
1143         if (on_freelist(s, page, object)) {
1144                 object_err(s, page, object, "Object already free");
1145                 return 0;
1146         }
1147 
1148         if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1149                 return 0;
1150 
1151         if (unlikely(s != page->slab_cache)) {
1152                 if (!PageSlab(page)) {
1153                         slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1154                                  object);
1155                 } else if (!page->slab_cache) {
1156                         pr_err("SLUB <none>: no slab for object 0x%p.\n",
1157                                object);
1158                         dump_stack();
1159                 } else
1160                         object_err(s, page, object,
1161                                         "page slab pointer corrupt.");
1162                 return 0;
1163         }
1164         return 1;
1165 }
1166 
1167 /* Supports checking bulk free of a constructed freelist */
1168 static noinline int free_debug_processing(
1169         struct kmem_cache *s, struct page *page,
1170         void *head, void *tail, int bulk_cnt,
1171         unsigned long addr)
1172 {
1173         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1174         void *object = head;
1175         int cnt = 0;
1176         unsigned long uninitialized_var(flags);
1177         int ret = 0;
1178 
1179         spin_lock_irqsave(&n->list_lock, flags);
1180         slab_lock(page);
1181 
1182         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1183                 if (!check_slab(s, page))
1184                         goto out;
1185         }
1186 
1187 next_object:
1188         cnt++;
1189 
1190         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1191                 if (!free_consistency_checks(s, page, object, addr))
1192                         goto out;
1193         }
1194 
1195         if (s->flags & SLAB_STORE_USER)
1196                 set_track(s, object, TRACK_FREE, addr);
1197         trace(s, page, object, 0);
1198         /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
1199         init_object(s, object, SLUB_RED_INACTIVE);
1200 
1201         /* Reached end of constructed freelist yet? */
1202         if (object != tail) {
1203                 object = get_freepointer(s, object);
1204                 goto next_object;
1205         }
1206         ret = 1;
1207 
1208 out:
1209         if (cnt != bulk_cnt)
1210                 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1211                          bulk_cnt, cnt);
1212 
1213         slab_unlock(page);
1214         spin_unlock_irqrestore(&n->list_lock, flags);
1215         if (!ret)
1216                 slab_fix(s, "Object at 0x%p not freed", object);
1217         return ret;
1218 }
1219 
1220 static int __init setup_slub_debug(char *str)
1221 {
1222         slub_debug = DEBUG_DEFAULT_FLAGS;
1223         if (*str++ != '=' || !*str)
1224                 /*
1225                  * No options specified. Switch on full debugging.
1226                  */
1227                 goto out;
1228 
1229         if (*str == ',')
1230                 /*
1231                  * No options but restriction on slabs. This means full
1232                  * debugging for slabs matching a pattern.
1233                  */
1234                 goto check_slabs;
1235 
1236         slub_debug = 0;
1237         if (*str == '-')
1238                 /*
1239                  * Switch off all debugging measures.
1240                  */
1241                 goto out;
1242 
1243         /*
1244          * Determine which debug features should be switched on
1245          */
1246         for (; *str && *str != ','; str++) {
1247                 switch (tolower(*str)) {
1248                 case 'f':
1249                         slub_debug |= SLAB_CONSISTENCY_CHECKS;
1250                         break;
1251                 case 'z':
1252                         slub_debug |= SLAB_RED_ZONE;
1253                         break;
1254                 case 'p':
1255                         slub_debug |= SLAB_POISON;
1256                         break;
1257                 case 'u':
1258                         slub_debug |= SLAB_STORE_USER;
1259                         break;
1260                 case 't':
1261                         slub_debug |= SLAB_TRACE;
1262                         break;
1263                 case 'a':
1264                         slub_debug |= SLAB_FAILSLAB;
1265                         break;
1266                 case 'o':
1267                         /*
1268                          * Avoid enabling debugging on caches if its minimum
1269                          * order would increase as a result.
1270                          */
1271                         disable_higher_order_debug = 1;
1272                         break;
1273                 default:
1274                         pr_err("slub_debug option '%c' unknown. skipped\n",
1275                                *str);
1276                 }
1277         }
1278 
1279 check_slabs:
1280         if (*str == ',')
1281                 slub_debug_slabs = str + 1;
1282 out:
1283         if ((static_branch_unlikely(&init_on_alloc) ||
1284              static_branch_unlikely(&init_on_free)) &&
1285             (slub_debug & SLAB_POISON))
1286                 pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
1287         return 1;
1288 }
1289 
1290 __setup("slub_debug", setup_slub_debug);
1291 
1292 /*
1293  * kmem_cache_flags - apply debugging options to the cache
1294  * @object_size:        the size of an object without meta data
1295  * @flags:              flags to set
1296  * @name:               name of the cache
1297  * @ctor:               constructor function
1298  *
1299  * Debug option(s) are applied to @flags. In addition to the debug
1300  * option(s), if a slab name (or multiple) is specified i.e.
1301  * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1302  * then only the select slabs will receive the debug option(s).
1303  */
1304 slab_flags_t kmem_cache_flags(unsigned int object_size,
1305         slab_flags_t flags, const char *name,
1306         void (*ctor)(void *))
1307 {
1308         char *iter;
1309         size_t len;
1310 
1311         /* If slub_debug = 0, it folds into the if conditional. */
1312         if (!slub_debug_slabs)
1313                 return flags | slub_debug;
1314 
1315         len = strlen(name);
1316         iter = slub_debug_slabs;
1317         while (*iter) {
1318                 char *end, *glob;
1319                 size_t cmplen;
1320 
1321                 end = strchrnul(iter, ',');
1322 
1323                 glob = strnchr(iter, end - iter, '*');
1324                 if (glob)
1325                         cmplen = glob - iter;
1326                 else
1327                         cmplen = max_t(size_t, len, (end - iter));
1328 
1329                 if (!strncmp(name, iter, cmplen)) {
1330                         flags |= slub_debug;
1331                         break;
1332                 }
1333 
1334                 if (!*end)
1335                         break;
1336                 iter = end + 1;
1337         }
1338 
1339         return flags;
1340 }
1341 #else /* !CONFIG_SLUB_DEBUG */
1342 static inline void setup_object_debug(struct kmem_cache *s,
1343                         struct page *page, void *object) {}
1344 static inline
1345 void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {}
1346 
1347 static inline int alloc_debug_processing(struct kmem_cache *s,
1348         struct page *page, void *object, unsigned long addr) { return 0; }
1349 
1350 static inline int free_debug_processing(
1351         struct kmem_cache *s, struct page *page,
1352         void *head, void *tail, int bulk_cnt,
1353         unsigned long addr) { return 0; }
1354 
1355 static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1356                         { return 1; }
1357 static inline int check_object(struct kmem_cache *s, struct page *page,
1358                         void *object, u8 val) { return 1; }
1359 static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1360                                         struct page *page) {}
1361 static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1362                                         struct page *page) {}
1363 slab_flags_t kmem_cache_flags(unsigned int object_size,
1364         slab_flags_t flags, const char *name,
1365         void (*ctor)(void *))
1366 {
1367         return flags;
1368 }
1369 #define slub_debug 0
1370 
1371 #define disable_higher_order_debug 0
1372 
1373 static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1374                                                         { return 0; }
1375 static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1376                                                         { return 0; }
1377 static inline void inc_slabs_node(struct kmem_cache *s, int node,
1378                                                         int objects) {}
1379 static inline void dec_slabs_node(struct kmem_cache *s, int node,
1380                                                         int objects) {}
1381 
1382 #endif /* CONFIG_SLUB_DEBUG */
1383 
1384 /*
1385  * Hooks for other subsystems that check memory allocations. In a typical
1386  * production configuration these hooks all should produce no code at all.
1387  */
1388 static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1389 {
1390         ptr = kasan_kmalloc_large(ptr, size, flags);
1391         /* As ptr might get tagged, call kmemleak hook after KASAN. */
1392         kmemleak_alloc(ptr, size, 1, flags);
1393         return ptr;
1394 }
1395 
1396 static __always_inline void kfree_hook(void *x)
1397 {
1398         kmemleak_free(x);
1399         kasan_kfree_large(x, _RET_IP_);
1400 }
1401 
1402 static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
1403 {
1404         kmemleak_free_recursive(x, s->flags);
1405 
1406         /*
1407          * Trouble is that we may no longer disable interrupts in the fast path
1408          * So in order to make the debug calls that expect irqs to be
1409          * disabled we need to disable interrupts temporarily.
1410          */
1411 #ifdef CONFIG_LOCKDEP
1412         {
1413                 unsigned long flags;
1414 
1415                 local_irq_save(flags);
1416                 debug_check_no_locks_freed(x, s->object_size);
1417                 local_irq_restore(flags);
1418         }
1419 #endif
1420         if (!(s->flags & SLAB_DEBUG_OBJECTS))
1421                 debug_check_no_obj_freed(x, s->object_size);
1422 
1423         /* KASAN might put x into memory quarantine, delaying its reuse */
1424         return kasan_slab_free(s, x, _RET_IP_);
1425 }
1426 
1427 static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1428                                            void **head, void **tail)
1429 {
1430 
1431         void *object;
1432         void *next = *head;
1433         void *old_tail = *tail ? *tail : *head;
1434         int rsize;
1435 
1436         /* Head and tail of the reconstructed freelist */
1437         *head = NULL;
1438         *tail = NULL;
1439 
1440         do {
1441                 object = next;
1442                 next = get_freepointer(s, object);
1443 
1444                 if (slab_want_init_on_free(s)) {
1445                         /*
1446                          * Clear the object and the metadata, but don't touch
1447                          * the redzone.
1448                          */
1449                         memset(object, 0, s->object_size);
1450                         rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad
1451                                                            : 0;
1452                         memset((char *)object + s->inuse, 0,
1453                                s->size - s->inuse - rsize);
1454 
1455                 }
1456                 /* If object's reuse doesn't have to be delayed */
1457                 if (!slab_free_hook(s, object)) {
1458                         /* Move object to the new freelist */
1459                         set_freepointer(s, object, *head);
1460                         *head = object;
1461                         if (!*tail)
1462                                 *tail = object;
1463                 }
1464         } while (object != old_tail);
1465 
1466         if (*head == *tail)
1467                 *tail = NULL;
1468 
1469         return *head != NULL;
1470 }
1471 
1472 static void *setup_object(struct kmem_cache *s, struct page *page,
1473                                 void *object)
1474 {
1475         setup_object_debug(s, page, object);
1476         object = kasan_init_slab_obj(s, object);
1477         if (unlikely(s->ctor)) {
1478                 kasan_unpoison_object_data(s, object);
1479                 s->ctor(object);
1480                 kasan_poison_object_data(s, object);
1481         }
1482         return object;
1483 }
1484 
1485 /*
1486  * Slab allocation and freeing
1487  */
1488 static inline struct page *alloc_slab_page(struct kmem_cache *s,
1489                 gfp_t flags, int node, struct kmem_cache_order_objects oo)
1490 {
1491         struct page *page;
1492         unsigned int order = oo_order(oo);
1493 
1494         if (node == NUMA_NO_NODE)
1495                 page = alloc_pages(flags, order);
1496         else
1497                 page = __alloc_pages_node(node, flags, order);
1498 
1499         if (page && charge_slab_page(page, flags, order, s)) {
1500                 __free_pages(page, order);
1501                 page = NULL;
1502         }
1503 
1504         return page;
1505 }
1506 
1507 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1508 /* Pre-initialize the random sequence cache */
1509 static int init_cache_random_seq(struct kmem_cache *s)
1510 {
1511         unsigned int count = oo_objects(s->oo);
1512         int err;
1513 
1514         /* Bailout if already initialised */
1515         if (s->random_seq)
1516                 return 0;
1517 
1518         err = cache_random_seq_create(s, count, GFP_KERNEL);
1519         if (err) {
1520                 pr_err("SLUB: Unable to initialize free list for %s\n",
1521                         s->name);
1522                 return err;
1523         }
1524 
1525         /* Transform to an offset on the set of pages */
1526         if (s->random_seq) {
1527                 unsigned int i;
1528 
1529                 for (i = 0; i < count; i++)
1530                         s->random_seq[i] *= s->size;
1531         }
1532         return 0;
1533 }
1534 
1535 /* Initialize each random sequence freelist per cache */
1536 static void __init init_freelist_randomization(void)
1537 {
1538         struct kmem_cache *s;
1539 
1540         mutex_lock(&slab_mutex);
1541 
1542         list_for_each_entry(s, &slab_caches, list)
1543                 init_cache_random_seq(s);
1544 
1545         mutex_unlock(&slab_mutex);
1546 }
1547 
1548 /* Get the next entry on the pre-computed freelist randomized */
1549 static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1550                                 unsigned long *pos, void *start,
1551                                 unsigned long page_limit,
1552                                 unsigned long freelist_count)
1553 {
1554         unsigned int idx;
1555 
1556         /*
1557          * If the target page allocation failed, the number of objects on the
1558          * page might be smaller than the usual size defined by the cache.
1559          */
1560         do {
1561                 idx = s->random_seq[*pos];
1562                 *pos += 1;
1563                 if (*pos >= freelist_count)
1564                         *pos = 0;
1565         } while (unlikely(idx >= page_limit));
1566 
1567         return (char *)start + idx;
1568 }
1569 
1570 /* Shuffle the single linked freelist based on a random pre-computed sequence */
1571 static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1572 {
1573         void *start;
1574         void *cur;
1575         void *next;
1576         unsigned long idx, pos, page_limit, freelist_count;
1577 
1578         if (page->objects < 2 || !s->random_seq)
1579                 return false;
1580 
1581         freelist_count = oo_objects(s->oo);
1582         pos = get_random_int() % freelist_count;
1583 
1584         page_limit = page->objects * s->size;
1585         start = fixup_red_left(s, page_address(page));
1586 
1587         /* First entry is used as the base of the freelist */
1588         cur = next_freelist_entry(s, page, &pos, start, page_limit,
1589                                 freelist_count);
1590         cur = setup_object(s, page, cur);
1591         page->freelist = cur;
1592 
1593         for (idx = 1; idx < page->objects; idx++) {
1594                 next = next_freelist_entry(s, page, &pos, start, page_limit,
1595                         freelist_count);
1596                 next = setup_object(s, page, next);
1597                 set_freepointer(s, cur, next);
1598                 cur = next;
1599         }
1600         set_freepointer(s, cur, NULL);
1601 
1602         return true;
1603 }
1604 #else
1605 static inline int init_cache_random_seq(struct kmem_cache *s)
1606 {
1607         return 0;
1608 }
1609 static inline void init_freelist_randomization(void) { }
1610 static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1611 {
1612         return false;
1613 }
1614 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1615 
1616 static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1617 {
1618         struct page *page;
1619         struct kmem_cache_order_objects oo = s->oo;
1620         gfp_t alloc_gfp;
1621         void *start, *p, *next;
1622         int idx;
1623         bool shuffle;
1624 
1625         flags &= gfp_allowed_mask;
1626 
1627         if (gfpflags_allow_blocking(flags))
1628                 local_irq_enable();
1629 
1630         flags |= s->allocflags;
1631 
1632         /*
1633          * Let the initial higher-order allocation fail under memory pressure
1634          * so we fall-back to the minimum order allocation.
1635          */
1636         alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1637         if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
1638                 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
1639 
1640         page = alloc_slab_page(s, alloc_gfp, node, oo);
1641         if (unlikely(!page)) {
1642                 oo = s->min;
1643                 alloc_gfp = flags;
1644                 /*
1645                  * Allocation may have failed due to fragmentation.
1646                  * Try a lower order alloc if possible
1647                  */
1648                 page = alloc_slab_page(s, alloc_gfp, node, oo);
1649                 if (unlikely(!page))
1650                         goto out;
1651                 stat(s, ORDER_FALLBACK);
1652         }
1653 
1654         page->objects = oo_objects(oo);
1655 
1656         page->slab_cache = s;
1657         __SetPageSlab(page);
1658         if (page_is_pfmemalloc(page))
1659                 SetPageSlabPfmemalloc(page);
1660 
1661         kasan_poison_slab(page);
1662 
1663         start = page_address(page);
1664 
1665         setup_page_debug(s, page, start);
1666 
1667         shuffle = shuffle_freelist(s, page);
1668 
1669         if (!shuffle) {
1670                 start = fixup_red_left(s, start);
1671                 start = setup_object(s, page, start);
1672                 page->freelist = start;
1673                 for (idx = 0, p = start; idx < page->objects - 1; idx++) {
1674                         next = p + s->size;
1675                         next = setup_object(s, page, next);
1676                         set_freepointer(s, p, next);
1677                         p = next;
1678                 }
1679                 set_freepointer(s, p, NULL);
1680         }
1681 
1682         page->inuse = page->objects;
1683         page->frozen = 1;
1684 
1685 out:
1686         if (gfpflags_allow_blocking(flags))
1687                 local_irq_disable();
1688         if (!page)
1689                 return NULL;
1690 
1691         inc_slabs_node(s, page_to_nid(page), page->objects);
1692 
1693         return page;
1694 }
1695 
1696 static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1697 {
1698         if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1699                 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1700                 flags &= ~GFP_SLAB_BUG_MASK;
1701                 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1702                                 invalid_mask, &invalid_mask, flags, &flags);
1703                 dump_stack();
1704         }
1705 
1706         return allocate_slab(s,
1707                 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1708 }
1709 
1710 static void __free_slab(struct kmem_cache *s, struct page *page)
1711 {
1712         int order = compound_order(page);
1713         int pages = 1 << order;
1714 
1715         if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1716                 void *p;
1717 
1718                 slab_pad_check(s, page);
1719                 for_each_object(p, s, page_address(page),
1720                                                 page->objects)
1721                         check_object(s, page, p, SLUB_RED_INACTIVE);
1722         }
1723 
1724         __ClearPageSlabPfmemalloc(page);
1725         __ClearPageSlab(page);
1726 
1727         page->mapping = NULL;
1728         if (current->reclaim_state)
1729                 current->reclaim_state->reclaimed_slab += pages;
1730         uncharge_slab_page(page, order, s);
1731         __free_pages(page, order);
1732 }
1733 
1734 static void rcu_free_slab(struct rcu_head *h)
1735 {
1736         struct page *page = container_of(h, struct page, rcu_head);
1737 
1738         __free_slab(page->slab_cache, page);
1739 }
1740 
1741 static void free_slab(struct kmem_cache *s, struct page *page)
1742 {
1743         if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
1744                 call_rcu(&page->rcu_head, rcu_free_slab);
1745         } else
1746                 __free_slab(s, page);
1747 }
1748 
1749 static void discard_slab(struct kmem_cache *s, struct page *page)
1750 {
1751         dec_slabs_node(s, page_to_nid(page), page->objects);
1752         free_slab(s, page);
1753 }
1754 
1755 /*
1756  * Management of partially allocated slabs.
1757  */
1758 static inline void
1759 __add_partial(struct kmem_cache_node *n, struct page *page, int tail)
1760 {
1761         n->nr_partial++;
1762         if (tail == DEACTIVATE_TO_TAIL)
1763                 list_add_tail(&page->slab_list, &n->partial);
1764         else
1765                 list_add(&page->slab_list, &n->partial);
1766 }
1767 
1768 static inline void add_partial(struct kmem_cache_node *n,
1769                                 struct page *page, int tail)
1770 {
1771         lockdep_assert_held(&n->list_lock);
1772         __add_partial(n, page, tail);
1773 }
1774 
1775 static inline void remove_partial(struct kmem_cache_node *n,
1776                                         struct page *page)
1777 {
1778         lockdep_assert_held(&n->list_lock);
1779         list_del(&page->slab_list);
1780         n->nr_partial--;
1781 }
1782 
1783 /*
1784  * Remove slab from the partial list, freeze it and
1785  * return the pointer to the freelist.
1786  *
1787  * Returns a list of objects or NULL if it fails.
1788  */
1789 static inline void *acquire_slab(struct kmem_cache *s,
1790                 struct kmem_cache_node *n, struct page *page,
1791                 int mode, int *objects)
1792 {
1793         void *freelist;
1794         unsigned long counters;
1795         struct page new;
1796 
1797         lockdep_assert_held(&n->list_lock);
1798 
1799         /*
1800          * Zap the freelist and set the frozen bit.
1801          * The old freelist is the list of objects for the
1802          * per cpu allocation list.
1803          */
1804         freelist = page->freelist;
1805         counters = page->counters;
1806         new.counters = counters;
1807         *objects = new.objects - new.inuse;
1808         if (mode) {
1809                 new.inuse = page->objects;
1810                 new.freelist = NULL;
1811         } else {
1812                 new.freelist = freelist;
1813         }
1814 
1815         VM_BUG_ON(new.frozen);
1816         new.frozen = 1;
1817 
1818         if (!__cmpxchg_double_slab(s, page,
1819                         freelist, counters,
1820                         new.freelist, new.counters,
1821                         "acquire_slab"))
1822                 return NULL;
1823 
1824         remove_partial(n, page);
1825         WARN_ON(!freelist);
1826         return freelist;
1827 }
1828 
1829 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
1830 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
1831 
1832 /*
1833  * Try to allocate a partial slab from a specific node.
1834  */
1835 static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1836                                 struct kmem_cache_cpu *c, gfp_t flags)
1837 {
1838         struct page *page, *page2;
1839         void *object = NULL;
1840         unsigned int available = 0;
1841         int objects;
1842 
1843         /*
1844          * Racy check. If we mistakenly see no partial slabs then we
1845          * just allocate an empty slab. If we mistakenly try to get a
1846          * partial slab and there is none available then get_partials()
1847          * will return NULL.
1848          */
1849         if (!n || !n->nr_partial)
1850                 return NULL;
1851 
1852         spin_lock(&n->list_lock);
1853         list_for_each_entry_safe(page, page2, &n->partial, slab_list) {
1854                 void *t;
1855 
1856                 if (!pfmemalloc_match(page, flags))
1857                         continue;
1858 
1859                 t = acquire_slab(s, n, page, object == NULL, &objects);
1860                 if (!t)
1861                         break;
1862 
1863                 available += objects;
1864                 if (!object) {
1865                         c->page = page;
1866                         stat(s, ALLOC_FROM_PARTIAL);
1867                         object = t;
1868                 } else {
1869                         put_cpu_partial(s, page, 0);
1870                         stat(s, CPU_PARTIAL_NODE);
1871                 }
1872                 if (!kmem_cache_has_cpu_partial(s)
1873                         || available > slub_cpu_partial(s) / 2)
1874                         break;
1875 
1876         }
1877         spin_unlock(&n->list_lock);
1878         return object;
1879 }
1880 
1881 /*
1882  * Get a page from somewhere. Search in increasing NUMA distances.
1883  */
1884 static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
1885                 struct kmem_cache_cpu *c)
1886 {
1887 #ifdef CONFIG_NUMA
1888         struct zonelist *zonelist;
1889         struct zoneref *z;
1890         struct zone *zone;
1891         enum zone_type high_zoneidx = gfp_zone(flags);
1892         void *object;
1893         unsigned int cpuset_mems_cookie;
1894 
1895         /*
1896          * The defrag ratio allows a configuration of the tradeoffs between
1897          * inter node defragmentation and node local allocations. A lower
1898          * defrag_ratio increases the tendency to do local allocations
1899          * instead of attempting to obtain partial slabs from other nodes.
1900          *
1901          * If the defrag_ratio is set to 0 then kmalloc() always
1902          * returns node local objects. If the ratio is higher then kmalloc()
1903          * may return off node objects because partial slabs are obtained
1904          * from other nodes and filled up.
1905          *
1906          * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1907          * (which makes defrag_ratio = 1000) then every (well almost)
1908          * allocation will first attempt to defrag slab caches on other nodes.
1909          * This means scanning over all nodes to look for partial slabs which
1910          * may be expensive if we do it every time we are trying to find a slab
1911          * with available objects.
1912          */
1913         if (!s->remote_node_defrag_ratio ||
1914                         get_cycles() % 1024 > s->remote_node_defrag_ratio)
1915                 return NULL;
1916 
1917         do {
1918                 cpuset_mems_cookie = read_mems_allowed_begin();
1919                 zonelist = node_zonelist(mempolicy_slab_node(), flags);
1920                 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1921                         struct kmem_cache_node *n;
1922 
1923                         n = get_node(s, zone_to_nid(zone));
1924 
1925                         if (n && cpuset_zone_allowed(zone, flags) &&
1926                                         n->nr_partial > s->min_partial) {
1927                                 object = get_partial_node(s, n, c, flags);
1928                                 if (object) {
1929                                         /*
1930                                          * Don't check read_mems_allowed_retry()
1931                                          * here - if mems_allowed was updated in
1932                                          * parallel, that was a harmless race
1933                                          * between allocation and the cpuset
1934                                          * update
1935                                          */
1936                                         return object;
1937                                 }
1938                         }
1939                 }
1940         } while (read_mems_allowed_retry(cpuset_mems_cookie));
1941 #endif  /* CONFIG_NUMA */
1942         return NULL;
1943 }
1944 
1945 /*
1946  * Get a partial page, lock it and return it.
1947  */
1948 static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
1949                 struct kmem_cache_cpu *c)
1950 {
1951         void *object;
1952         int searchnode = node;
1953 
1954         if (node == NUMA_NO_NODE)
1955                 searchnode = numa_mem_id();
1956 
1957         object = get_partial_node(s, get_node(s, searchnode), c, flags);
1958         if (object || node != NUMA_NO_NODE)
1959                 return object;
1960 
1961         return get_any_partial(s, flags, c);
1962 }
1963 
1964 #ifdef CONFIG_PREEMPT
1965 /*
1966  * Calculate the next globally unique transaction for disambiguiation
1967  * during cmpxchg. The transactions start with the cpu number and are then
1968  * incremented by CONFIG_NR_CPUS.
1969  */
1970 #define TID_STEP  roundup_pow_of_two(CONFIG_NR_CPUS)
1971 #else
1972 /*
1973  * No preemption supported therefore also no need to check for
1974  * different cpus.
1975  */
1976 #define TID_STEP 1
1977 #endif
1978 
1979 static inline unsigned long next_tid(unsigned long tid)
1980 {
1981         return tid + TID_STEP;
1982 }
1983 
1984 #ifdef SLUB_DEBUG_CMPXCHG
1985 static inline unsigned int tid_to_cpu(unsigned long tid)
1986 {
1987         return tid % TID_STEP;
1988 }
1989 
1990 static inline unsigned long tid_to_event(unsigned long tid)
1991 {
1992         return tid / TID_STEP;
1993 }
1994 #endif
1995 
1996 static inline unsigned int init_tid(int cpu)
1997 {
1998         return cpu;
1999 }
2000 
2001 static inline void note_cmpxchg_failure(const char *n,
2002                 const struct kmem_cache *s, unsigned long tid)
2003 {
2004 #ifdef SLUB_DEBUG_CMPXCHG
2005         unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2006 
2007         pr_info("%s %s: cmpxchg redo ", n, s->name);
2008 
2009 #ifdef CONFIG_PREEMPT
2010         if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
2011                 pr_warn("due to cpu change %d -> %d\n",
2012                         tid_to_cpu(tid), tid_to_cpu(actual_tid));
2013         else
2014 #endif
2015         if (tid_to_event(tid) != tid_to_event(actual_tid))
2016                 pr_warn("due to cpu running other code. Event %ld->%ld\n",
2017                         tid_to_event(tid), tid_to_event(actual_tid));
2018         else
2019                 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
2020                         actual_tid, tid, next_tid(tid));
2021 #endif
2022         stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
2023 }
2024 
2025 static void init_kmem_cache_cpus(struct kmem_cache *s)
2026 {
2027         int cpu;
2028 
2029         for_each_possible_cpu(cpu)
2030                 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
2031 }
2032 
2033 /*
2034  * Remove the cpu slab
2035  */
2036 static void deactivate_slab(struct kmem_cache *s, struct page *page,
2037                                 void *freelist, struct kmem_cache_cpu *c)
2038 {
2039         enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2040         struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2041         int lock = 0;
2042         enum slab_modes l = M_NONE, m = M_NONE;
2043         void *nextfree;
2044         int tail = DEACTIVATE_TO_HEAD;
2045         struct page new;
2046         struct page old;
2047 
2048         if (page->freelist) {
2049                 stat(s, DEACTIVATE_REMOTE_FREES);
2050                 tail = DEACTIVATE_TO_TAIL;
2051         }
2052 
2053         /*
2054          * Stage one: Free all available per cpu objects back
2055          * to the page freelist while it is still frozen. Leave the
2056          * last one.
2057          *
2058          * There is no need to take the list->lock because the page
2059          * is still frozen.
2060          */
2061         while (freelist && (nextfree = get_freepointer(s, freelist))) {
2062                 void *prior;
2063                 unsigned long counters;
2064 
2065                 do {
2066                         prior = page->freelist;
2067                         counters = page->counters;
2068                         set_freepointer(s, freelist, prior);
2069                         new.counters = counters;
2070                         new.inuse--;
2071                         VM_BUG_ON(!new.frozen);
2072 
2073                 } while (!__cmpxchg_double_slab(s, page,
2074                         prior, counters,
2075                         freelist, new.counters,
2076                         "drain percpu freelist"));
2077 
2078                 freelist = nextfree;
2079         }
2080 
2081         /*
2082          * Stage two: Ensure that the page is unfrozen while the
2083          * list presence reflects the actual number of objects
2084          * during unfreeze.
2085          *
2086          * We setup the list membership and then perform a cmpxchg
2087          * with the count. If there is a mismatch then the page
2088          * is not unfrozen but the page is on the wrong list.
2089          *
2090          * Then we restart the process which may have to remove
2091          * the page from the list that we just put it on again
2092          * because the number of objects in the slab may have
2093          * changed.
2094          */
2095 redo:
2096 
2097         old.freelist = page->freelist;
2098         old.counters = page->counters;
2099         VM_BUG_ON(!old.frozen);
2100 
2101         /* Determine target state of the slab */
2102         new.counters = old.counters;
2103         if (freelist) {
2104                 new.inuse--;
2105                 set_freepointer(s, freelist, old.freelist);
2106                 new.freelist = freelist;
2107         } else
2108                 new.freelist = old.freelist;
2109 
2110         new.frozen = 0;
2111 
2112         if (!new.inuse && n->nr_partial >= s->min_partial)
2113                 m = M_FREE;
2114         else if (new.freelist) {
2115                 m = M_PARTIAL;
2116                 if (!lock) {
2117                         lock = 1;
2118                         /*
2119                          * Taking the spinlock removes the possibility
2120                          * that acquire_slab() will see a slab page that
2121                          * is frozen
2122                          */
2123                         spin_lock(&n->list_lock);
2124                 }
2125         } else {
2126                 m = M_FULL;
2127                 if (kmem_cache_debug(s) && !lock) {
2128                         lock = 1;
2129                         /*
2130                          * This also ensures that the scanning of full
2131                          * slabs from diagnostic functions will not see
2132                          * any frozen slabs.
2133                          */
2134                         spin_lock(&n->list_lock);
2135                 }
2136         }
2137 
2138         if (l != m) {
2139                 if (l == M_PARTIAL)
2140                         remove_partial(n, page);
2141                 else if (l == M_FULL)
2142                         remove_full(s, n, page);
2143 
2144                 if (m == M_PARTIAL)
2145                         add_partial(n, page, tail);
2146                 else if (m == M_FULL)
2147                         add_full(s, n, page);
2148         }
2149 
2150         l = m;
2151         if (!__cmpxchg_double_slab(s, page,
2152                                 old.freelist, old.counters,
2153                                 new.freelist, new.counters,
2154                                 "unfreezing slab"))
2155                 goto redo;
2156 
2157         if (lock)
2158                 spin_unlock(&n->list_lock);
2159 
2160         if (m == M_PARTIAL)
2161                 stat(s, tail);
2162         else if (m == M_FULL)
2163                 stat(s, DEACTIVATE_FULL);
2164         else if (m == M_FREE) {
2165                 stat(s, DEACTIVATE_EMPTY);
2166                 discard_slab(s, page);
2167                 stat(s, FREE_SLAB);
2168         }
2169 
2170         c->page = NULL;
2171         c->freelist = NULL;
2172 }
2173 
2174 /*
2175  * Unfreeze all the cpu partial slabs.
2176  *
2177  * This function must be called with interrupts disabled
2178  * for the cpu using c (or some other guarantee must be there
2179  * to guarantee no concurrent accesses).
2180  */
2181 static void unfreeze_partials(struct kmem_cache *s,
2182                 struct kmem_cache_cpu *c)
2183 {
2184 #ifdef CONFIG_SLUB_CPU_PARTIAL
2185         struct kmem_cache_node *n = NULL, *n2 = NULL;
2186         struct page *page, *discard_page = NULL;
2187 
2188         while ((page = c->partial)) {
2189                 struct page new;
2190                 struct page old;
2191 
2192                 c->partial = page->next;
2193 
2194                 n2 = get_node(s, page_to_nid(page));
2195                 if (n != n2) {
2196                         if (n)
2197                                 spin_unlock(&n->list_lock);
2198 
2199                         n = n2;
2200                         spin_lock(&n->list_lock);
2201                 }
2202 
2203                 do {
2204 
2205                         old.freelist = page->freelist;
2206                         old.counters = page->counters;
2207                         VM_BUG_ON(!old.frozen);
2208 
2209                         new.counters = old.counters;
2210                         new.freelist = old.freelist;
2211 
2212                         new.frozen = 0;
2213 
2214                 } while (!__cmpxchg_double_slab(s, page,
2215                                 old.freelist, old.counters,
2216                                 new.freelist, new.counters,
2217                                 "unfreezing slab"));
2218 
2219                 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
2220                         page->next = discard_page;
2221                         discard_page = page;
2222                 } else {
2223                         add_partial(n, page, DEACTIVATE_TO_TAIL);
2224                         stat(s, FREE_ADD_PARTIAL);
2225                 }
2226         }
2227 
2228         if (n)
2229                 spin_unlock(&n->list_lock);
2230 
2231         while (discard_page) {
2232                 page = discard_page;
2233                 discard_page = discard_page->next;
2234 
2235                 stat(s, DEACTIVATE_EMPTY);
2236                 discard_slab(s, page);
2237                 stat(s, FREE_SLAB);
2238         }
2239 #endif  /* CONFIG_SLUB_CPU_PARTIAL */
2240 }
2241 
2242 /*
2243  * Put a page that was just frozen (in __slab_free|get_partial_node) into a
2244  * partial page slot if available.
2245  *
2246  * If we did not find a slot then simply move all the partials to the
2247  * per node partial list.
2248  */
2249 static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
2250 {
2251 #ifdef CONFIG_SLUB_CPU_PARTIAL
2252         struct page *oldpage;
2253         int pages;
2254         int pobjects;
2255 
2256         preempt_disable();
2257         do {
2258                 pages = 0;
2259                 pobjects = 0;
2260                 oldpage = this_cpu_read(s->cpu_slab->partial);
2261 
2262                 if (oldpage) {
2263                         pobjects = oldpage->pobjects;
2264                         pages = oldpage->pages;
2265                         if (drain && pobjects > s->cpu_partial) {
2266                                 unsigned long flags;
2267                                 /*
2268                                  * partial array is full. Move the existing
2269                                  * set to the per node partial list.
2270                                  */
2271                                 local_irq_save(flags);
2272                                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2273                                 local_irq_restore(flags);
2274                                 oldpage = NULL;
2275                                 pobjects = 0;
2276                                 pages = 0;
2277                                 stat(s, CPU_PARTIAL_DRAIN);
2278                         }
2279                 }
2280 
2281                 pages++;
2282                 pobjects += page->objects - page->inuse;
2283 
2284                 page->pages = pages;
2285                 page->pobjects = pobjects;
2286                 page->next = oldpage;
2287 
2288         } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2289                                                                 != oldpage);
2290         if (unlikely(!s->cpu_partial)) {
2291                 unsigned long flags;
2292 
2293                 local_irq_save(flags);
2294                 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2295                 local_irq_restore(flags);
2296         }
2297         preempt_enable();
2298 #endif  /* CONFIG_SLUB_CPU_PARTIAL */
2299 }
2300 
2301 static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
2302 {
2303         stat(s, CPUSLAB_FLUSH);
2304         deactivate_slab(s, c->page, c->freelist, c);
2305 
2306         c->tid = next_tid(c->tid);
2307 }
2308 
2309 /*
2310  * Flush cpu slab.
2311  *
2312  * Called from IPI handler with interrupts disabled.
2313  */
2314 static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
2315 {
2316         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2317 
2318         if (c->page)
2319                 flush_slab(s, c);
2320 
2321         unfreeze_partials(s, c);
2322 }
2323 
2324 static void flush_cpu_slab(void *d)
2325 {
2326         struct kmem_cache *s = d;
2327 
2328         __flush_cpu_slab(s, smp_processor_id());
2329 }
2330 
2331 static bool has_cpu_slab(int cpu, void *info)
2332 {
2333         struct kmem_cache *s = info;
2334         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2335 
2336         return c->page || slub_percpu_partial(c);
2337 }
2338 
2339 static void flush_all(struct kmem_cache *s)
2340 {
2341         on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
2342 }
2343 
2344 /*
2345  * Use the cpu notifier to insure that the cpu slabs are flushed when
2346  * necessary.
2347  */
2348 static int slub_cpu_dead(unsigned int cpu)
2349 {
2350         struct kmem_cache *s;
2351         unsigned long flags;
2352 
2353         mutex_lock(&slab_mutex);
2354         list_for_each_entry(s, &slab_caches, list) {
2355                 local_irq_save(flags);
2356                 __flush_cpu_slab(s, cpu);
2357                 local_irq_restore(flags);
2358         }
2359         mutex_unlock(&slab_mutex);
2360         return 0;
2361 }
2362 
2363 /*
2364  * Check if the objects in a per cpu structure fit numa
2365  * locality expectations.
2366  */
2367 static inline int node_match(struct page *page, int node)
2368 {
2369 #ifdef CONFIG_NUMA
2370         if (node != NUMA_NO_NODE && page_to_nid(page) != node)
2371                 return 0;
2372 #endif
2373         return 1;
2374 }
2375 
2376 #ifdef CONFIG_SLUB_DEBUG
2377 static int count_free(struct page *page)
2378 {
2379         return page->objects - page->inuse;
2380 }
2381 
2382 static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2383 {
2384         return atomic_long_read(&n->total_objects);
2385 }
2386 #endif /* CONFIG_SLUB_DEBUG */
2387 
2388 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2389 static unsigned long count_partial(struct kmem_cache_node *n,
2390                                         int (*get_count)(struct page *))
2391 {
2392         unsigned long flags;
2393         unsigned long x = 0;
2394         struct page *page;
2395 
2396         spin_lock_irqsave(&n->list_lock, flags);
2397         list_for_each_entry(page, &n->partial, slab_list)
2398                 x += get_count(page);
2399         spin_unlock_irqrestore(&n->list_lock, flags);
2400         return x;
2401 }
2402 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2403 
2404 static noinline void
2405 slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2406 {
2407 #ifdef CONFIG_SLUB_DEBUG
2408         static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2409                                       DEFAULT_RATELIMIT_BURST);
2410         int node;
2411         struct kmem_cache_node *n;
2412 
2413         if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2414                 return;
2415 
2416         pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2417                 nid, gfpflags, &gfpflags);
2418         pr_warn("  cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
2419                 s->name, s->object_size, s->size, oo_order(s->oo),
2420                 oo_order(s->min));
2421 
2422         if (oo_order(s->min) > get_order(s->object_size))
2423                 pr_warn("  %s debugging increased min order, use slub_debug=O to disable.\n",
2424                         s->name);
2425 
2426         for_each_kmem_cache_node(s, node, n) {
2427                 unsigned long nr_slabs;
2428                 unsigned long nr_objs;
2429                 unsigned long nr_free;
2430 
2431                 nr_free  = count_partial(n, count_free);
2432                 nr_slabs = node_nr_slabs(n);
2433                 nr_objs  = node_nr_objs(n);
2434 
2435                 pr_warn("  node %d: slabs: %ld, objs: %ld, free: %ld\n",
2436                         node, nr_slabs, nr_objs, nr_free);
2437         }
2438 #endif
2439 }
2440 
2441 static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2442                         int node, struct kmem_cache_cpu **pc)
2443 {
2444         void *freelist;
2445         struct kmem_cache_cpu *c = *pc;
2446         struct page *page;
2447 
2448         WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2449 
2450         freelist = get_partial(s, flags, node, c);
2451 
2452         if (freelist)
2453                 return freelist;
2454 
2455         page = new_slab(s, flags, node);
2456         if (page) {
2457                 c = raw_cpu_ptr(s->cpu_slab);
2458                 if (c->page)
2459                         flush_slab(s, c);
2460 
2461                 /*
2462                  * No other reference to the page yet so we can
2463                  * muck around with it freely without cmpxchg
2464                  */
2465                 freelist = page->freelist;
2466                 page->freelist = NULL;
2467 
2468                 stat(s, ALLOC_SLAB);
2469                 c->page = page;
2470                 *pc = c;
2471         }
2472 
2473         return freelist;
2474 }
2475 
2476 static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2477 {
2478         if (unlikely(PageSlabPfmemalloc(page)))
2479                 return gfp_pfmemalloc_allowed(gfpflags);
2480 
2481         return true;
2482 }
2483 
2484 /*
2485  * Check the page->freelist of a page and either transfer the freelist to the
2486  * per cpu freelist or deactivate the page.
2487  *
2488  * The page is still frozen if the return value is not NULL.
2489  *
2490  * If this function returns NULL then the page has been unfrozen.
2491  *
2492  * This function must be called with interrupt disabled.
2493  */
2494 static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2495 {
2496         struct page new;
2497         unsigned long counters;
2498         void *freelist;
2499 
2500         do {
2501                 freelist = page->freelist;
2502                 counters = page->counters;
2503 
2504                 new.counters = counters;
2505                 VM_BUG_ON(!new.frozen);
2506 
2507                 new.inuse = page->objects;
2508                 new.frozen = freelist != NULL;
2509 
2510         } while (!__cmpxchg_double_slab(s, page,
2511                 freelist, counters,
2512                 NULL, new.counters,
2513                 "get_freelist"));
2514 
2515         return freelist;
2516 }
2517 
2518 /*
2519  * Slow path. The lockless freelist is empty or we need to perform
2520  * debugging duties.
2521  *
2522  * Processing is still very fast if new objects have been freed to the
2523  * regular freelist. In that case we simply take over the regular freelist
2524  * as the lockless freelist and zap the regular freelist.
2525  *
2526  * If that is not working then we fall back to the partial lists. We take the
2527  * first element of the freelist as the object to allocate now and move the
2528  * rest of the freelist to the lockless freelist.
2529  *
2530  * And if we were unable to get a new slab from the partial slab lists then
2531  * we need to allocate a new slab. This is the slowest path since it involves
2532  * a call to the page allocator and the setup of a new slab.
2533  *
2534  * Version of __slab_alloc to use when we know that interrupts are
2535  * already disabled (which is the case for bulk allocation).
2536  */
2537 static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2538                           unsigned long addr, struct kmem_cache_cpu *c)
2539 {
2540         void *freelist;
2541         struct page *page;
2542 
2543         page = c->page;
2544         if (!page) {
2545                 /*
2546                  * if the node is not online or has no normal memory, just
2547                  * ignore the node constraint
2548                  */
2549                 if (unlikely(node != NUMA_NO_NODE &&
2550                              !node_state(node, N_NORMAL_MEMORY)))
2551                         node = NUMA_NO_NODE;
2552                 goto new_slab;
2553         }
2554 redo:
2555 
2556         if (unlikely(!node_match(page, node))) {
2557                 /*
2558                  * same as above but node_match() being false already
2559                  * implies node != NUMA_NO_NODE
2560                  */
2561                 if (!node_state(node, N_NORMAL_MEMORY)) {
2562                         node = NUMA_NO_NODE;
2563                         goto redo;
2564                 } else {
2565                         stat(s, ALLOC_NODE_MISMATCH);
2566                         deactivate_slab(s, page, c->freelist, c);
2567                         goto new_slab;
2568                 }
2569         }
2570 
2571         /*
2572          * By rights, we should be searching for a slab page that was
2573          * PFMEMALLOC but right now, we are losing the pfmemalloc
2574          * information when the page leaves the per-cpu allocator
2575          */
2576         if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2577                 deactivate_slab(s, page, c->freelist, c);
2578                 goto new_slab;
2579         }
2580 
2581         /* must check again c->freelist in case of cpu migration or IRQ */
2582         freelist = c->freelist;
2583         if (freelist)
2584                 goto load_freelist;
2585 
2586         freelist = get_freelist(s, page);
2587 
2588         if (!freelist) {
2589                 c->page = NULL;
2590                 stat(s, DEACTIVATE_BYPASS);
2591                 goto new_slab;
2592         }
2593 
2594         stat(s, ALLOC_REFILL);
2595 
2596 load_freelist:
2597         /*
2598          * freelist is pointing to the list of objects to be used.
2599          * page is pointing to the page from which the objects are obtained.
2600          * That page must be frozen for per cpu allocations to work.
2601          */
2602         VM_BUG_ON(!c->page->frozen);
2603         c->freelist = get_freepointer(s, freelist);
2604         c->tid = next_tid(c->tid);
2605         return freelist;
2606 
2607 new_slab:
2608 
2609         if (slub_percpu_partial(c)) {
2610                 page = c->page = slub_percpu_partial(c);
2611                 slub_set_percpu_partial(c, page);
2612                 stat(s, CPU_PARTIAL_ALLOC);
2613                 goto redo;
2614         }
2615 
2616         freelist = new_slab_objects(s, gfpflags, node, &c);
2617 
2618         if (unlikely(!freelist)) {
2619                 slab_out_of_memory(s, gfpflags, node);
2620                 return NULL;
2621         }
2622 
2623         page = c->page;
2624         if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
2625                 goto load_freelist;
2626 
2627         /* Only entered in the debug case */
2628         if (kmem_cache_debug(s) &&
2629                         !alloc_debug_processing(s, page, freelist, addr))
2630                 goto new_slab;  /* Slab failed checks. Next slab needed */
2631 
2632         deactivate_slab(s, page, get_freepointer(s, freelist), c);
2633         return freelist;
2634 }
2635 
2636 /*
2637  * Another one that disabled interrupt and compensates for possible
2638  * cpu changes by refetching the per cpu area pointer.
2639  */
2640 static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2641                           unsigned long addr, struct kmem_cache_cpu *c)
2642 {
2643         void *p;
2644         unsigned long flags;
2645 
2646         local_irq_save(flags);
2647 #ifdef CONFIG_PREEMPT
2648         /*
2649          * We may have been preempted and rescheduled on a different
2650          * cpu before disabling interrupts. Need to reload cpu area
2651          * pointer.
2652          */
2653         c = this_cpu_ptr(s->cpu_slab);
2654 #endif
2655 
2656         p = ___slab_alloc(s, gfpflags, node, addr, c);
2657         local_irq_restore(flags);
2658         return p;
2659 }
2660 
2661 /*
2662  * If the object has been wiped upon free, make sure it's fully initialized by
2663  * zeroing out freelist pointer.
2664  */
2665 static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
2666                                                    void *obj)
2667 {
2668         if (unlikely(slab_want_init_on_free(s)) && obj)
2669                 memset((void *)((char *)obj + s->offset), 0, sizeof(void *));
2670 }
2671 
2672 /*
2673  * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2674  * have the fastpath folded into their functions. So no function call
2675  * overhead for requests that can be satisfied on the fastpath.
2676  *
2677  * The fastpath works by first checking if the lockless freelist can be used.
2678  * If not then __slab_alloc is called for slow processing.
2679  *
2680  * Otherwise we can simply pick the next object from the lockless free list.
2681  */
2682 static __always_inline void *slab_alloc_node(struct kmem_cache *s,
2683                 gfp_t gfpflags, int node, unsigned long addr)
2684 {
2685         void *object;
2686         struct kmem_cache_cpu *c;
2687         struct page *page;
2688         unsigned long tid;
2689 
2690         s = slab_pre_alloc_hook(s, gfpflags);
2691         if (!s)
2692                 return NULL;
2693 redo:
2694         /*
2695          * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2696          * enabled. We may switch back and forth between cpus while
2697          * reading from one cpu area. That does not matter as long
2698          * as we end up on the original cpu again when doing the cmpxchg.
2699          *
2700          * We should guarantee that tid and kmem_cache are retrieved on
2701          * the same cpu. It could be different if CONFIG_PREEMPT so we need
2702          * to check if it is matched or not.
2703          */
2704         do {
2705                 tid = this_cpu_read(s->cpu_slab->tid);
2706                 c = raw_cpu_ptr(s->cpu_slab);
2707         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2708                  unlikely(tid != READ_ONCE(c->tid)));
2709 
2710         /*
2711          * Irqless object alloc/free algorithm used here depends on sequence
2712          * of fetching cpu_slab's data. tid should be fetched before anything
2713          * on c to guarantee that object and page associated with previous tid
2714          * won't be used with current tid. If we fetch tid first, object and
2715          * page could be one associated with next tid and our alloc/free
2716          * request will be failed. In this case, we will retry. So, no problem.
2717          */
2718         barrier();
2719 
2720         /*
2721          * The transaction ids are globally unique per cpu and per operation on
2722          * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2723          * occurs on the right processor and that there was no operation on the
2724          * linked list in between.
2725          */
2726 
2727         object = c->freelist;
2728         page = c->page;
2729         if (unlikely(!object || !node_match(page, node))) {
2730                 object = __slab_alloc(s, gfpflags, node, addr, c);
2731                 stat(s, ALLOC_SLOWPATH);
2732         } else {
2733                 void *next_object = get_freepointer_safe(s, object);
2734 
2735                 /*
2736                  * The cmpxchg will only match if there was no additional
2737                  * operation and if we are on the right processor.
2738                  *
2739                  * The cmpxchg does the following atomically (without lock
2740                  * semantics!)
2741                  * 1. Relocate first pointer to the current per cpu area.
2742                  * 2. Verify that tid and freelist have not been changed
2743                  * 3. If they were not changed replace tid and freelist
2744                  *
2745                  * Since this is without lock semantics the protection is only
2746                  * against code executing on this cpu *not* from access by
2747                  * other cpus.
2748                  */
2749                 if (unlikely(!this_cpu_cmpxchg_double(
2750                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2751                                 object, tid,
2752                                 next_object, next_tid(tid)))) {
2753 
2754                         note_cmpxchg_failure("slab_alloc", s, tid);
2755                         goto redo;
2756                 }
2757                 prefetch_freepointer(s, next_object);
2758                 stat(s, ALLOC_FASTPATH);
2759         }
2760 
2761         maybe_wipe_obj_freeptr(s, object);
2762 
2763         if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object)
2764                 memset(object, 0, s->object_size);
2765 
2766         slab_post_alloc_hook(s, gfpflags, 1, &object);
2767 
2768         return object;
2769 }
2770 
2771 static __always_inline void *slab_alloc(struct kmem_cache *s,
2772                 gfp_t gfpflags, unsigned long addr)
2773 {
2774         return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2775 }
2776 
2777 void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2778 {
2779         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2780 
2781         trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2782                                 s->size, gfpflags);
2783 
2784         return ret;
2785 }
2786 EXPORT_SYMBOL(kmem_cache_alloc);
2787 
2788 #ifdef CONFIG_TRACING
2789 void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2790 {
2791         void *ret = slab_alloc(s, gfpflags, _RET_IP_);
2792         trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2793         ret = kasan_kmalloc(s, ret, size, gfpflags);
2794         return ret;
2795 }
2796 EXPORT_SYMBOL(kmem_cache_alloc_trace);
2797 #endif
2798 
2799 #ifdef CONFIG_NUMA
2800 void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2801 {
2802         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2803 
2804         trace_kmem_cache_alloc_node(_RET_IP_, ret,
2805                                     s->object_size, s->size, gfpflags, node);
2806 
2807         return ret;
2808 }
2809 EXPORT_SYMBOL(kmem_cache_alloc_node);
2810 
2811 #ifdef CONFIG_TRACING
2812 void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
2813                                     gfp_t gfpflags,
2814                                     int node, size_t size)
2815 {
2816         void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
2817 
2818         trace_kmalloc_node(_RET_IP_, ret,
2819                            size, s->size, gfpflags, node);
2820 
2821         ret = kasan_kmalloc(s, ret, size, gfpflags);
2822         return ret;
2823 }
2824 EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
2825 #endif
2826 #endif  /* CONFIG_NUMA */
2827 
2828 /*
2829  * Slow path handling. This may still be called frequently since objects
2830  * have a longer lifetime than the cpu slabs in most processing loads.
2831  *
2832  * So we still attempt to reduce cache line usage. Just take the slab
2833  * lock and free the item. If there is no additional partial page
2834  * handling required then we can return immediately.
2835  */
2836 static void __slab_free(struct kmem_cache *s, struct page *page,
2837                         void *head, void *tail, int cnt,
2838                         unsigned long addr)
2839 
2840 {
2841         void *prior;
2842         int was_frozen;
2843         struct page new;
2844         unsigned long counters;
2845         struct kmem_cache_node *n = NULL;
2846         unsigned long uninitialized_var(flags);
2847 
2848         stat(s, FREE_SLOWPATH);
2849 
2850         if (kmem_cache_debug(s) &&
2851             !free_debug_processing(s, page, head, tail, cnt, addr))
2852                 return;
2853 
2854         do {
2855                 if (unlikely(n)) {
2856                         spin_unlock_irqrestore(&n->list_lock, flags);
2857                         n = NULL;
2858                 }
2859                 prior = page->freelist;
2860                 counters = page->counters;
2861                 set_freepointer(s, tail, prior);
2862                 new.counters = counters;
2863                 was_frozen = new.frozen;
2864                 new.inuse -= cnt;
2865                 if ((!new.inuse || !prior) && !was_frozen) {
2866 
2867                         if (kmem_cache_has_cpu_partial(s) && !prior) {
2868 
2869                                 /*
2870                                  * Slab was on no list before and will be
2871                                  * partially empty
2872                                  * We can defer the list move and instead
2873                                  * freeze it.
2874                                  */
2875                                 new.frozen = 1;
2876 
2877                         } else { /* Needs to be taken off a list */
2878 
2879                                 n = get_node(s, page_to_nid(page));
2880                                 /*
2881                                  * Speculatively acquire the list_lock.
2882                                  * If the cmpxchg does not succeed then we may
2883                                  * drop the list_lock without any processing.
2884                                  *
2885                                  * Otherwise the list_lock will synchronize with
2886                                  * other processors updating the list of slabs.
2887                                  */
2888                                 spin_lock_irqsave(&n->list_lock, flags);
2889 
2890                         }
2891                 }
2892 
2893         } while (!cmpxchg_double_slab(s, page,
2894                 prior, counters,
2895                 head, new.counters,
2896                 "__slab_free"));
2897 
2898         if (likely(!n)) {
2899 
2900                 /*
2901                  * If we just froze the page then put it onto the
2902                  * per cpu partial list.
2903                  */
2904                 if (new.frozen && !was_frozen) {
2905                         put_cpu_partial(s, page, 1);
2906                         stat(s, CPU_PARTIAL_FREE);
2907                 }
2908                 /*
2909                  * The list lock was not taken therefore no list
2910                  * activity can be necessary.
2911                  */
2912                 if (was_frozen)
2913                         stat(s, FREE_FROZEN);
2914                 return;
2915         }
2916 
2917         if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
2918                 goto slab_empty;
2919 
2920         /*
2921          * Objects left in the slab. If it was not on the partial list before
2922          * then add it.
2923          */
2924         if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2925                 remove_full(s, n, page);
2926                 add_partial(n, page, DEACTIVATE_TO_TAIL);
2927                 stat(s, FREE_ADD_PARTIAL);
2928         }
2929         spin_unlock_irqrestore(&n->list_lock, flags);
2930         return;
2931 
2932 slab_empty:
2933         if (prior) {
2934                 /*
2935                  * Slab on the partial list.
2936                  */
2937                 remove_partial(n, page);
2938                 stat(s, FREE_REMOVE_PARTIAL);
2939         } else {
2940                 /* Slab must be on the full list */
2941                 remove_full(s, n, page);
2942         }
2943 
2944         spin_unlock_irqrestore(&n->list_lock, flags);
2945         stat(s, FREE_SLAB);
2946         discard_slab(s, page);
2947 }
2948 
2949 /*
2950  * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2951  * can perform fastpath freeing without additional function calls.
2952  *
2953  * The fastpath is only possible if we are freeing to the current cpu slab
2954  * of this processor. This typically the case if we have just allocated
2955  * the item before.
2956  *
2957  * If fastpath is not possible then fall back to __slab_free where we deal
2958  * with all sorts of special processing.
2959  *
2960  * Bulk free of a freelist with several objects (all pointing to the
2961  * same page) possible by specifying head and tail ptr, plus objects
2962  * count (cnt). Bulk free indicated by tail pointer being set.
2963  */
2964 static __always_inline void do_slab_free(struct kmem_cache *s,
2965                                 struct page *page, void *head, void *tail,
2966                                 int cnt, unsigned long addr)
2967 {
2968         void *tail_obj = tail ? : head;
2969         struct kmem_cache_cpu *c;
2970         unsigned long tid;
2971 redo:
2972         /*
2973          * Determine the currently cpus per cpu slab.
2974          * The cpu may change afterward. However that does not matter since
2975          * data is retrieved via this pointer. If we are on the same cpu
2976          * during the cmpxchg then the free will succeed.
2977          */
2978         do {
2979                 tid = this_cpu_read(s->cpu_slab->tid);
2980                 c = raw_cpu_ptr(s->cpu_slab);
2981         } while (IS_ENABLED(CONFIG_PREEMPT) &&
2982                  unlikely(tid != READ_ONCE(c->tid)));
2983 
2984         /* Same with comment on barrier() in slab_alloc_node() */
2985         barrier();
2986 
2987         if (likely(page == c->page)) {
2988                 void **freelist = READ_ONCE(c->freelist);
2989 
2990                 set_freepointer(s, tail_obj, freelist);
2991 
2992                 if (unlikely(!this_cpu_cmpxchg_double(
2993                                 s->cpu_slab->freelist, s->cpu_slab->tid,
2994                                 freelist, tid,
2995                                 head, next_tid(tid)))) {
2996 
2997                         note_cmpxchg_failure("slab_free", s, tid);
2998                         goto redo;
2999                 }
3000                 stat(s, FREE_FASTPATH);
3001         } else
3002                 __slab_free(s, page, head, tail_obj, cnt, addr);
3003 
3004 }
3005 
3006 static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
3007                                       void *head, void *tail, int cnt,
3008                                       unsigned long addr)
3009 {
3010         /*
3011          * With KASAN enabled slab_free_freelist_hook modifies the freelist
3012          * to remove objects, whose reuse must be delayed.
3013          */
3014         if (slab_free_freelist_hook(s, &head, &tail))
3015                 do_slab_free(s, page, head, tail, cnt, addr);
3016 }
3017 
3018 #ifdef CONFIG_KASAN_GENERIC
3019 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3020 {
3021         do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
3022 }
3023 #endif
3024 
3025 void kmem_cache_free(struct kmem_cache *s, void *x)
3026 {
3027         s = cache_from_obj(s, x);
3028         if (!s)
3029                 return;
3030         slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
3031         trace_kmem_cache_free(_RET_IP_, x);
3032 }
3033 EXPORT_SYMBOL(kmem_cache_free);
3034 
3035 struct detached_freelist {
3036         struct page *page;
3037         void *tail;
3038         void *freelist;
3039         int cnt;
3040         struct kmem_cache *s;
3041 };
3042 
3043 /*
3044  * This function progressively scans the array with free objects (with
3045  * a limited look ahead) and extract objects belonging to the same
3046  * page.  It builds a detached freelist directly within the given
3047  * page/objects.  This can happen without any need for
3048  * synchronization, because the objects are owned by running process.
3049  * The freelist is build up as a single linked list in the objects.
3050  * The idea is, that this detached freelist can then be bulk
3051  * transferred to the real freelist(s), but only requiring a single
3052  * synchronization primitive.  Look ahead in the array is limited due
3053  * to performance reasons.
3054  */
3055 static inline
3056 int build_detached_freelist(struct kmem_cache *s, size_t size,
3057                             void **p, struct detached_freelist *df)
3058 {
3059         size_t first_skipped_index = 0;
3060         int lookahead = 3;
3061         void *object;
3062         struct page *page;
3063 
3064         /* Always re-init detached_freelist */
3065         df->page = NULL;
3066 
3067         do {
3068                 object = p[--size];
3069                 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
3070         } while (!object && size);
3071 
3072         if (!object)
3073                 return 0;
3074 
3075         page = virt_to_head_page(object);
3076         if (!s) {
3077                 /* Handle kalloc'ed objects */
3078                 if (unlikely(!PageSlab(page))) {
3079                         BUG_ON(!PageCompound(page));
3080                         kfree_hook(object);
3081                         __free_pages(page, compound_order(page));
3082                         p[size] = NULL; /* mark object processed */
3083                         return size;
3084                 }
3085                 /* Derive kmem_cache from object */
3086                 df->s = page->slab_cache;
3087         } else {
3088                 df->s = cache_from_obj(s, object); /* Support for memcg */
3089         }
3090 
3091         /* Start new detached freelist */
3092         df->page = page;
3093         set_freepointer(df->s, object, NULL);
3094         df->tail = object;
3095         df->freelist = object;
3096         p[size] = NULL; /* mark object processed */
3097         df->cnt = 1;
3098 
3099         while (size) {
3100                 object = p[--size];
3101                 if (!object)
3102                         continue; /* Skip processed objects */
3103 
3104                 /* df->page is always set at this point */
3105                 if (df->page == virt_to_head_page(object)) {
3106                         /* Opportunity build freelist */
3107                         set_freepointer(df->s, object, df->freelist);
3108                         df->freelist = object;
3109                         df->cnt++;
3110                         p[size] = NULL; /* mark object processed */
3111 
3112                         continue;
3113                 }
3114 
3115                 /* Limit look ahead search */
3116                 if (!--lookahead)
3117                         break;
3118 
3119                 if (!first_skipped_index)
3120                         first_skipped_index = size + 1;
3121         }
3122 
3123         return first_skipped_index;
3124 }
3125 
3126 /* Note that interrupts must be enabled when calling this function. */
3127 void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
3128 {
3129         if (WARN_ON(!size))
3130                 return;
3131 
3132         do {
3133                 struct detached_freelist df;
3134 
3135                 size = build_detached_freelist(s, size, p, &df);
3136                 if (!df.page)
3137                         continue;
3138 
3139                 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
3140         } while (likely(size));
3141 }
3142 EXPORT_SYMBOL(kmem_cache_free_bulk);
3143 
3144 /* Note that interrupts must be enabled when calling this function. */
3145 int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3146                           void **p)
3147 {
3148         struct kmem_cache_cpu *c;
3149         int i;
3150 
3151         /* memcg and kmem_cache debug support */
3152         s = slab_pre_alloc_hook(s, flags);
3153         if (unlikely(!s))
3154                 return false;
3155         /*
3156          * Drain objects in the per cpu slab, while disabling local
3157          * IRQs, which protects against PREEMPT and interrupts
3158          * handlers invoking normal fastpath.
3159          */
3160         local_irq_disable();
3161         c = this_cpu_ptr(s->cpu_slab);
3162 
3163         for (i = 0; i < size; i++) {
3164                 void *object = c->freelist;
3165 
3166                 if (unlikely(!object)) {
3167                         /*
3168                          * We may have removed an object from c->freelist using
3169                          * the fastpath in the previous iteration; in that case,
3170                          * c->tid has not been bumped yet.
3171                          * Since ___slab_alloc() may reenable interrupts while
3172                          * allocating memory, we should bump c->tid now.
3173                          */
3174                         c->tid = next_tid(c->tid);
3175 
3176                         /*
3177                          * Invoking slow path likely have side-effect
3178                          * of re-populating per CPU c->freelist
3179                          */
3180                         p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
3181                                             _RET_IP_, c);
3182                         if (unlikely(!p[i]))
3183                                 goto error;
3184 
3185                         c = this_cpu_ptr(s->cpu_slab);
3186                         maybe_wipe_obj_freeptr(s, p[i]);
3187 
3188                         continue; /* goto for-loop */
3189                 }
3190                 c->freelist = get_freepointer(s, object);
3191                 p[i] = object;
3192                 maybe_wipe_obj_freeptr(s, p[i]);
3193         }
3194         c->tid = next_tid(c->tid);
3195         local_irq_enable();
3196 
3197         /* Clear memory outside IRQ disabled fastpath loop */
3198         if (unlikely(slab_want_init_on_alloc(flags, s))) {
3199                 int j;
3200 
3201                 for (j = 0; j < i; j++)
3202                         memset(p[j], 0, s->object_size);
3203         }
3204 
3205         /* memcg and kmem_cache debug support */
3206         slab_post_alloc_hook(s, flags, size, p);
3207         return i;
3208 error:
3209         local_irq_enable();
3210         slab_post_alloc_hook(s, flags, i, p);
3211         __kmem_cache_free_bulk(s, i, p);
3212         return 0;
3213 }
3214 EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3215 
3216 
3217 /*
3218  * Object placement in a slab is made very easy because we always start at
3219  * offset 0. If we tune the size of the object to the alignment then we can
3220  * get the required alignment by putting one properly sized object after
3221  * another.
3222  *
3223  * Notice that the allocation order determines the sizes of the per cpu
3224  * caches. Each processor has always one slab available for allocations.
3225  * Increasing the allocation order reduces the number of times that slabs
3226  * must be moved on and off the partial lists and is therefore a factor in
3227  * locking overhead.
3228  */
3229 
3230 /*
3231  * Mininum / Maximum order of slab pages. This influences locking overhead
3232  * and slab fragmentation. A higher order reduces the number of partial slabs
3233  * and increases the number of allocations possible without having to
3234  * take the list_lock.
3235  */
3236 static unsigned int slub_min_order;
3237 static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3238 static unsigned int slub_min_objects;
3239 
3240 /*
3241  * Calculate the order of allocation given an slab object size.
3242  *
3243  * The order of allocation has significant impact on performance and other
3244  * system components. Generally order 0 allocations should be preferred since
3245  * order 0 does not cause fragmentation in the page allocator. Larger objects
3246  * be problematic to put into order 0 slabs because there may be too much
3247  * unused space left. We go to a higher order if more than 1/16th of the slab
3248  * would be wasted.
3249  *
3250  * In order to reach satisfactory performance we must ensure that a minimum
3251  * number of objects is in one slab. Otherwise we may generate too much
3252  * activity on the partial lists which requires taking the list_lock. This is
3253  * less a concern for large slabs though which are rarely used.
3254  *
3255  * slub_max_order specifies the order where we begin to stop considering the
3256  * number of objects in a slab as critical. If we reach slub_max_order then
3257  * we try to keep the page order as low as possible. So we accept more waste
3258  * of space in favor of a small page order.
3259  *
3260  * Higher order allocations also allow the placement of more objects in a
3261  * slab and thereby reduce object handling overhead. If the user has
3262  * requested a higher mininum order then we start with that one instead of
3263  * the smallest order which will fit the object.
3264  */
3265 static inline unsigned int slab_order(unsigned int size,
3266                 unsigned int min_objects, unsigned int max_order,
3267                 unsigned int fract_leftover)
3268 {
3269         unsigned int min_order = slub_min_order;
3270         unsigned int order;
3271 
3272         if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
3273                 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
3274 
3275         for (order = max(min_order, (unsigned int)get_order(min_objects * size));
3276                         order <= max_order; order++) {
3277 
3278                 unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3279                 unsigned int rem;
3280 
3281                 rem = slab_size % size;
3282 
3283                 if (rem <= slab_size / fract_leftover)
3284                         break;
3285         }
3286 
3287         return order;
3288 }
3289 
3290 static inline int calculate_order(unsigned int size)
3291 {
3292         unsigned int order;
3293         unsigned int min_objects;
3294         unsigned int max_objects;
3295 
3296         /*
3297          * Attempt to find best configuration for a slab. This
3298          * works by first attempting to generate a layout with
3299          * the best configuration and backing off gradually.
3300          *
3301          * First we increase the acceptable waste in a slab. Then
3302          * we reduce the minimum objects required in a slab.
3303          */
3304         min_objects = slub_min_objects;
3305         if (!min_objects)
3306                 min_objects = 4 * (fls(nr_cpu_ids) + 1);
3307         max_objects = order_objects(slub_max_order, size);
3308         min_objects = min(min_objects, max_objects);
3309 
3310         while (min_objects > 1) {
3311                 unsigned int fraction;
3312 
3313                 fraction = 16;
3314                 while (fraction >= 4) {
3315                         order = slab_order(size, min_objects,
3316                                         slub_max_order, fraction);
3317                         if (order <= slub_max_order)
3318                                 return order;
3319                         fraction /= 2;
3320                 }
3321                 min_objects--;
3322         }
3323 
3324         /*
3325          * We were unable to place multiple objects in a slab. Now
3326          * lets see if we can place a single object there.
3327          */
3328         order = slab_order(size, 1, slub_max_order, 1);
3329         if (order <= slub_max_order)
3330                 return order;
3331 
3332         /*
3333          * Doh this slab cannot be placed using slub_max_order.
3334          */
3335         order = slab_order(size, 1, MAX_ORDER, 1);
3336         if (order < MAX_ORDER)
3337                 return order;
3338         return -ENOSYS;
3339 }
3340 
3341 static void
3342 init_kmem_cache_node(struct kmem_cache_node *n)
3343 {
3344         n->nr_partial = 0;
3345         spin_lock_init(&n->list_lock);
3346         INIT_LIST_HEAD(&n->partial);
3347 #ifdef CONFIG_SLUB_DEBUG
3348         atomic_long_set(&n->nr_slabs, 0);
3349         atomic_long_set(&n->total_objects, 0);
3350         INIT_LIST_HEAD(&n->full);
3351 #endif
3352 }
3353 
3354 static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
3355 {
3356         BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
3357                         KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
3358 
3359         /*
3360          * Must align to double word boundary for the double cmpxchg
3361          * instructions to work; see __pcpu_double_call_return_bool().
3362          */
3363         s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3364                                      2 * sizeof(void *));
3365 
3366         if (!s->cpu_slab)
3367                 return 0;
3368 
3369         init_kmem_cache_cpus(s);
3370 
3371         return 1;
3372 }
3373 
3374 static struct kmem_cache *kmem_cache_node;
3375 
3376 /*
3377  * No kmalloc_node yet so do it by hand. We know that this is the first
3378  * slab on the node for this slabcache. There are no concurrent accesses
3379  * possible.
3380  *
3381  * Note that this function only works on the kmem_cache_node
3382  * when allocating for the kmem_cache_node. This is used for bootstrapping
3383  * memory on a fresh node that has no slab structures yet.
3384  */
3385 static void early_kmem_cache_node_alloc(int node)
3386 {
3387         struct page *page;
3388         struct kmem_cache_node *n;
3389 
3390         BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
3391 
3392         page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
3393 
3394         BUG_ON(!page);
3395         if (page_to_nid(page) != node) {
3396                 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3397                 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
3398         }
3399 
3400         n = page->freelist;
3401         BUG_ON(!n);
3402 #ifdef CONFIG_SLUB_DEBUG
3403         init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
3404         init_tracking(kmem_cache_node, n);
3405 #endif
3406         n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3407                       GFP_KERNEL);
3408         page->freelist = get_freepointer(kmem_cache_node, n);
3409         page->inuse = 1;
3410         page->frozen = 0;
3411         kmem_cache_node->node[node] = n;
3412         init_kmem_cache_node(n);
3413         inc_slabs_node(kmem_cache_node, node, page->objects);
3414 
3415         /*
3416          * No locks need to be taken here as it has just been
3417          * initialized and there is no concurrent access.
3418          */
3419         __add_partial(n, page, DEACTIVATE_TO_HEAD);
3420 }
3421 
3422 static void free_kmem_cache_nodes(struct kmem_cache *s)
3423 {
3424         int node;
3425         struct kmem_cache_node *n;
3426 
3427         for_each_kmem_cache_node(s, node, n) {
3428                 s->node[node] = NULL;
3429                 kmem_cache_free(kmem_cache_node, n);
3430         }
3431 }
3432 
3433 void __kmem_cache_release(struct kmem_cache *s)
3434 {
3435         cache_random_seq_destroy(s);
3436         free_percpu(s->cpu_slab);
3437         free_kmem_cache_nodes(s);
3438 }
3439 
3440 static int init_kmem_cache_nodes(struct kmem_cache *s)
3441 {
3442         int node;
3443 
3444         for_each_node_state(node, N_NORMAL_MEMORY) {
3445                 struct kmem_cache_node *n;
3446 
3447                 if (slab_state == DOWN) {
3448                         early_kmem_cache_node_alloc(node);
3449                         continue;
3450                 }
3451                 n = kmem_cache_alloc_node(kmem_cache_node,
3452                                                 GFP_KERNEL, node);
3453 
3454                 if (!n) {
3455                         free_kmem_cache_nodes(s);
3456                         return 0;
3457                 }
3458 
3459                 init_kmem_cache_node(n);
3460                 s->node[node] = n;
3461         }
3462         return 1;
3463 }
3464 
3465 static void set_min_partial(struct kmem_cache *s, unsigned long min)
3466 {
3467         if (min < MIN_PARTIAL)
3468                 min = MIN_PARTIAL;
3469         else if (min > MAX_PARTIAL)
3470                 min = MAX_PARTIAL;
3471         s->min_partial = min;
3472 }
3473 
3474 static void set_cpu_partial(struct kmem_cache *s)
3475 {
3476 #ifdef CONFIG_SLUB_CPU_PARTIAL
3477         /*
3478          * cpu_partial determined the maximum number of objects kept in the
3479          * per cpu partial lists of a processor.
3480          *
3481          * Per cpu partial lists mainly contain slabs that just have one
3482          * object freed. If they are used for allocation then they can be
3483          * filled up again with minimal effort. The slab will never hit the
3484          * per node partial lists and therefore no locking will be required.
3485          *
3486          * This setting also determines
3487          *
3488          * A) The number of objects from per cpu partial slabs dumped to the
3489          *    per node list when we reach the limit.
3490          * B) The number of objects in cpu partial slabs to extract from the
3491          *    per node list when we run out of per cpu objects. We only fetch
3492          *    50% to keep some capacity around for frees.
3493          */
3494         if (!kmem_cache_has_cpu_partial(s))
3495                 s->cpu_partial = 0;
3496         else if (s->size >= PAGE_SIZE)
3497                 s->cpu_partial = 2;
3498         else if (s->size >= 1024)
3499                 s->cpu_partial = 6;
3500         else if (s->size >= 256)
3501                 s->cpu_partial = 13;
3502         else
3503                 s->cpu_partial = 30;
3504 #endif
3505 }
3506 
3507 /*
3508  * calculate_sizes() determines the order and the distribution of data within
3509  * a slab object.
3510  */
3511 static int calculate_sizes(struct kmem_cache *s, int forced_order)
3512 {
3513         slab_flags_t flags = s->flags;
3514         unsigned int size = s->object_size;
3515         unsigned int order;
3516 
3517         /*
3518          * Round up object size to the next word boundary. We can only
3519          * place the free pointer at word boundaries and this determines
3520          * the possible location of the free pointer.
3521          */
3522         size = ALIGN(size, sizeof(void *));
3523 
3524 #ifdef CONFIG_SLUB_DEBUG
3525         /*
3526          * Determine if we can poison the object itself. If the user of
3527          * the slab may touch the object after free or before allocation
3528          * then we should never poison the object itself.
3529          */
3530         if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
3531                         !s->ctor)
3532                 s->flags |= __OBJECT_POISON;
3533         else
3534                 s->flags &= ~__OBJECT_POISON;
3535 
3536 
3537         /*
3538          * If we are Redzoning then check if there is some space between the
3539          * end of the object and the free pointer. If not then add an
3540          * additional word to have some bytes to store Redzone information.
3541          */
3542         if ((flags & SLAB_RED_ZONE) && size == s->object_size)
3543                 size += sizeof(void *);
3544 #endif
3545 
3546         /*
3547          * With that we have determined the number of bytes in actual use
3548          * by the object. This is the potential offset to the free pointer.
3549          */
3550         s->inuse = size;
3551 
3552         if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3553                 s->ctor)) {
3554                 /*
3555                  * Relocate free pointer after the object if it is not
3556                  * permitted to overwrite the first word of the object on
3557                  * kmem_cache_free.
3558                  *
3559                  * This is the case if we do RCU, have a constructor or
3560                  * destructor or are poisoning the objects.
3561                  */
3562                 s->offset = size;
3563                 size += sizeof(void *);
3564         }
3565 
3566 #ifdef CONFIG_SLUB_DEBUG
3567         if (flags & SLAB_STORE_USER)
3568                 /*
3569                  * Need to store information about allocs and frees after
3570                  * the object.
3571                  */
3572                 size += 2 * sizeof(struct track);
3573 #endif
3574 
3575         kasan_cache_create(s, &size, &s->flags);
3576 #ifdef CONFIG_SLUB_DEBUG
3577         if (flags & SLAB_RED_ZONE) {
3578                 /*
3579                  * Add some empty padding so that we can catch
3580                  * overwrites from earlier objects rather than let
3581                  * tracking information or the free pointer be
3582                  * corrupted if a user writes before the start
3583                  * of the object.
3584                  */
3585                 size += sizeof(void *);
3586 
3587                 s->red_left_pad = sizeof(void *);
3588                 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3589                 size += s->red_left_pad;
3590         }
3591 #endif
3592 
3593         /*
3594          * SLUB stores one object immediately after another beginning from
3595          * offset 0. In order to align the objects we have to simply size
3596          * each object to conform to the alignment.
3597          */
3598         size = ALIGN(size, s->align);
3599         s->size = size;
3600         if (forced_order >= 0)
3601                 order = forced_order;
3602         else
3603                 order = calculate_order(size);
3604 
3605         if ((int)order < 0)
3606                 return 0;
3607 
3608         s->allocflags = 0;
3609         if (order)
3610                 s->allocflags |= __GFP_COMP;
3611 
3612         if (s->flags & SLAB_CACHE_DMA)
3613                 s->allocflags |= GFP_DMA;
3614 
3615         if (s->flags & SLAB_CACHE_DMA32)
3616                 s->allocflags |= GFP_DMA32;
3617 
3618         if (s->flags & SLAB_RECLAIM_ACCOUNT)
3619                 s->allocflags |= __GFP_RECLAIMABLE;
3620 
3621         /*
3622          * Determine the number of objects per slab
3623          */
3624         s->oo = oo_make(order, size);
3625         s->min = oo_make(get_order(size), size);
3626         if (oo_objects(s->oo) > oo_objects(s->max))
3627                 s->max = s->oo;
3628 
3629         return !!oo_objects(s->oo);
3630 }
3631 
3632 static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
3633 {
3634         s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
3635 #ifdef CONFIG_SLAB_FREELIST_HARDENED
3636         s->random = get_random_long();
3637 #endif
3638 
3639         if (!calculate_sizes(s, -1))
3640                 goto error;
3641         if (disable_higher_order_debug) {
3642                 /*
3643                  * Disable debugging flags that store metadata if the min slab
3644                  * order increased.
3645                  */
3646                 if (get_order(s->size) > get_order(s->object_size)) {
3647                         s->flags &= ~DEBUG_METADATA_FLAGS;
3648                         s->offset = 0;
3649                         if (!calculate_sizes(s, -1))
3650                                 goto error;
3651                 }
3652         }
3653 
3654 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3655     defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3656         if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
3657                 /* Enable fast mode */
3658                 s->flags |= __CMPXCHG_DOUBLE;
3659 #endif
3660 
3661         /*
3662          * The larger the object size is, the more pages we want on the partial
3663          * list to avoid pounding the page allocator excessively.
3664          */
3665         set_min_partial(s, ilog2(s->size) / 2);
3666 
3667         set_cpu_partial(s);
3668 
3669 #ifdef CONFIG_NUMA
3670         s->remote_node_defrag_ratio = 1000;
3671 #endif
3672 
3673         /* Initialize the pre-computed randomized freelist if slab is up */
3674         if (slab_state >= UP) {
3675                 if (init_cache_random_seq(s))
3676                         goto error;
3677         }
3678 
3679         if (!init_kmem_cache_nodes(s))
3680                 goto error;
3681 
3682         if (alloc_kmem_cache_cpus(s))
3683                 return 0;
3684 
3685         free_kmem_cache_nodes(s);
3686 error:
3687         return -EINVAL;
3688 }
3689 
3690 static void list_slab_objects(struct kmem_cache *s, struct page *page,
3691                                                         const char *text)
3692 {
3693 #ifdef CONFIG_SLUB_DEBUG
3694         void *addr = page_address(page);
3695         void *p;
3696         unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC);
3697         if (!map)
3698                 return;
3699         slab_err(s, page, text, s->name);
3700         slab_lock(page);
3701 
3702         get_map(s, page, map);
3703         for_each_object(p, s, addr, page->objects) {
3704 
3705                 if (!test_bit(slab_index(p, s, addr), map)) {
3706                         pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
3707                         print_tracking(s, p);
3708                 }
3709         }
3710         slab_unlock(page);
3711         bitmap_free(map);
3712 #endif
3713 }
3714 
3715 /*
3716  * Attempt to free all partial slabs on a node.
3717  * This is called from __kmem_cache_shutdown(). We must take list_lock
3718  * because sysfs file might still access partial list after the shutdowning.
3719  */
3720 static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
3721 {
3722         LIST_HEAD(discard);
3723         struct page *page, *h;
3724 
3725         BUG_ON(irqs_disabled());
3726         spin_lock_irq(&n->list_lock);
3727         list_for_each_entry_safe(page, h, &n->partial, slab_list) {
3728                 if (!page->inuse) {
3729                         remove_partial(n, page);
3730                         list_add(&page->slab_list, &discard);
3731                 } else {
3732                         list_slab_objects(s, page,
3733                         "Objects remaining in %s on __kmem_cache_shutdown()");
3734                 }
3735         }
3736         spin_unlock_irq(&n->list_lock);
3737 
3738         list_for_each_entry_safe(page, h, &discard, slab_list)
3739                 discard_slab(s, page);
3740 }
3741 
3742 bool __kmem_cache_empty(struct kmem_cache *s)
3743 {
3744         int node;
3745         struct kmem_cache_node *n;
3746 
3747         for_each_kmem_cache_node(s, node, n)
3748                 if (n->nr_partial || slabs_node(s, node))
3749                         return false;
3750         return true;
3751 }
3752 
3753 /*
3754  * Release all resources used by a slab cache.
3755  */
3756 int __kmem_cache_shutdown(struct kmem_cache *s)
3757 {
3758         int node;
3759         struct kmem_cache_node *n;
3760 
3761         flush_all(s);
3762         /* Attempt to free all objects */
3763         for_each_kmem_cache_node(s, node, n) {
3764                 free_partial(s, n);
3765                 if (n->nr_partial || slabs_node(s, node))
3766                         return 1;
3767         }
3768         sysfs_slab_remove(s);
3769         return 0;
3770 }
3771 
3772 /********************************************************************
3773  *              Kmalloc subsystem
3774  *******************************************************************/
3775 
3776 static int __init setup_slub_min_order(char *str)
3777 {
3778         get_option(&str, (int *)&slub_min_order);
3779 
3780         return 1;
3781 }
3782 
3783 __setup("slub_min_order=", setup_slub_min_order);
3784 
3785 static int __init setup_slub_max_order(char *str)
3786 {
3787         get_option(&str, (int *)&slub_max_order);
3788         slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
3789 
3790         return 1;
3791 }
3792 
3793 __setup("slub_max_order=", setup_slub_max_order);
3794 
3795 static int __init setup_slub_min_objects(char *str)
3796 {
3797         get_option(&str, (int *)&slub_min_objects);
3798 
3799         return 1;
3800 }
3801 
3802 __setup("slub_min_objects=", setup_slub_min_objects);
3803 
3804 void *__kmalloc(size_t size, gfp_t flags)
3805 {
3806         struct kmem_cache *s;
3807         void *ret;
3808 
3809         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3810                 return kmalloc_large(size, flags);
3811 
3812         s = kmalloc_slab(size, flags);
3813 
3814         if (unlikely(ZERO_OR_NULL_PTR(s)))
3815                 return s;
3816 
3817         ret = slab_alloc(s, flags, _RET_IP_);
3818 
3819         trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
3820 
3821         ret = kasan_kmalloc(s, ret, size, flags);
3822 
3823         return ret;
3824 }
3825 EXPORT_SYMBOL(__kmalloc);
3826 
3827 #ifdef CONFIG_NUMA
3828 static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3829 {
3830         struct page *page;
3831         void *ptr = NULL;
3832         unsigned int order = get_order(size);
3833 
3834         flags |= __GFP_COMP;
3835         page = alloc_pages_node(node, flags, order);
3836         if (page) {
3837                 ptr = page_address(page);
3838                 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE,
3839                                     1 << order);
3840         }
3841 
3842         return kmalloc_large_node_hook(ptr, size, flags);
3843 }
3844 
3845 void *__kmalloc_node(size_t size, gfp_t flags, int node)
3846 {
3847         struct kmem_cache *s;
3848         void *ret;
3849 
3850         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
3851                 ret = kmalloc_large_node(size, flags, node);
3852 
3853                 trace_kmalloc_node(_RET_IP_, ret,
3854                                    size, PAGE_SIZE << get_order(size),
3855                                    flags, node);
3856 
3857                 return ret;
3858         }
3859 
3860         s = kmalloc_slab(size, flags);
3861 
3862         if (unlikely(ZERO_OR_NULL_PTR(s)))
3863                 return s;
3864 
3865         ret = slab_alloc_node(s, flags, node, _RET_IP_);
3866 
3867         trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
3868 
3869         ret = kasan_kmalloc(s, ret, size, flags);
3870 
3871         return ret;
3872 }
3873 EXPORT_SYMBOL(__kmalloc_node);
3874 #endif  /* CONFIG_NUMA */
3875 
3876 #ifdef CONFIG_HARDENED_USERCOPY
3877 /*
3878  * Rejects incorrectly sized objects and objects that are to be copied
3879  * to/from userspace but do not fall entirely within the containing slab
3880  * cache's usercopy region.
3881  *
3882  * Returns NULL if check passes, otherwise const char * to name of cache
3883  * to indicate an error.
3884  */
3885 void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3886                          bool to_user)
3887 {
3888         struct kmem_cache *s;
3889         unsigned int offset;
3890         size_t object_size;
3891 
3892         ptr = kasan_reset_tag(ptr);
3893 
3894         /* Find object and usable object size. */
3895         s = page->slab_cache;
3896 
3897         /* Reject impossible pointers. */
3898         if (ptr < page_address(page))
3899                 usercopy_abort("SLUB object not in SLUB page?!", NULL,
3900                                to_user, 0, n);
3901 
3902         /* Find offset within object. */
3903         offset = (ptr - page_address(page)) % s->size;
3904 
3905         /* Adjust for redzone and reject if within the redzone. */
3906         if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3907                 if (offset < s->red_left_pad)
3908                         usercopy_abort("SLUB object in left red zone",
3909                                        s->name, to_user, offset, n);
3910                 offset -= s->red_left_pad;
3911         }
3912 
3913         /* Allow address range falling entirely within usercopy region. */
3914         if (offset >= s->useroffset &&
3915             offset - s->useroffset <= s->usersize &&
3916             n <= s->useroffset - offset + s->usersize)
3917                 return;
3918 
3919         /*
3920          * If the copy is still within the allocated object, produce
3921          * a warning instead of rejecting the copy. This is intended
3922          * to be a temporary method to find any missing usercopy
3923          * whitelists.
3924          */
3925         object_size = slab_ksize(s);
3926         if (usercopy_fallback &&
3927             offset <= object_size && n <= object_size - offset) {
3928                 usercopy_warn("SLUB object", s->name, to_user, offset, n);
3929                 return;
3930         }
3931 
3932         usercopy_abort("SLUB object", s->name, to_user, offset, n);
3933 }
3934 #endif /* CONFIG_HARDENED_USERCOPY */
3935 
3936 size_t __ksize(const void *object)
3937 {
3938         struct page *page;
3939 
3940         if (unlikely(object == ZERO_SIZE_PTR))
3941                 return 0;
3942 
3943         page = virt_to_head_page(object);
3944 
3945         if (unlikely(!PageSlab(page))) {
3946                 WARN_ON(!PageCompound(page));
3947                 return page_size(page);
3948         }
3949 
3950         return slab_ksize(page->slab_cache);
3951 }
3952 EXPORT_SYMBOL(__ksize);
3953 
3954 void kfree(const void *x)
3955 {
3956         struct page *page;
3957         void *object = (void *)x;
3958 
3959         trace_kfree(_RET_IP_, x);
3960 
3961         if (unlikely(ZERO_OR_NULL_PTR(x)))
3962                 return;
3963 
3964         page = virt_to_head_page(x);
3965         if (unlikely(!PageSlab(page))) {
3966                 unsigned int order = compound_order(page);
3967 
3968                 BUG_ON(!PageCompound(page));
3969                 kfree_hook(object);
3970                 mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE,
3971                                     -(1 << order));
3972                 __free_pages(page, order);
3973                 return;
3974         }
3975         slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
3976 }
3977 EXPORT_SYMBOL(kfree);
3978 
3979 #define SHRINK_PROMOTE_MAX 32
3980 
3981 /*
3982  * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3983  * up most to the head of the partial lists. New allocations will then
3984  * fill those up and thus they can be removed from the partial lists.
3985  *
3986  * The slabs with the least items are placed last. This results in them
3987  * being allocated from last increasing the chance that the last objects
3988  * are freed in them.
3989  */
3990 int __kmem_cache_shrink(struct kmem_cache *s)
3991 {
3992         int node;
3993         int i;
3994         struct kmem_cache_node *n;
3995         struct page *page;
3996         struct page *t;
3997         struct list_head discard;
3998         struct list_head promote[SHRINK_PROMOTE_MAX];
3999         unsigned long flags;
4000         int ret = 0;
4001 
4002         flush_all(s);
4003         for_each_kmem_cache_node(s, node, n) {
4004                 INIT_LIST_HEAD(&discard);
4005                 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4006                         INIT_LIST_HEAD(promote + i);
4007 
4008                 spin_lock_irqsave(&n->list_lock, flags);
4009 
4010                 /*
4011                  * Build lists of slabs to discard or promote.
4012                  *
4013                  * Note that concurrent frees may occur while we hold the
4014                  * list_lock. page->inuse here is the upper limit.
4015                  */
4016                 list_for_each_entry_safe(page, t, &n->partial, slab_list) {
4017                         int free = page->objects - page->inuse;
4018 
4019                         /* Do not reread page->inuse */
4020                         barrier();
4021 
4022                         /* We do not keep full slabs on the list */
4023                         BUG_ON(free <= 0);
4024 
4025                         if (free == page->objects) {
4026                                 list_move(&page->slab_list, &discard);
4027                                 n->nr_partial--;
4028                         } else if (free <= SHRINK_PROMOTE_MAX)
4029                                 list_move(&page->slab_list, promote + free - 1);
4030                 }
4031 
4032                 /*
4033                  * Promote the slabs filled up most to the head of the
4034                  * partial list.
4035                  */
4036                 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4037                         list_splice(promote + i, &n->partial);
4038 
4039                 spin_unlock_irqrestore(&n->list_lock, flags);
4040 
4041                 /* Release empty slabs */
4042                 list_for_each_entry_safe(page, t, &discard, slab_list)
4043                         discard_slab(s, page);
4044 
4045                 if (slabs_node(s, node))
4046                         ret = 1;
4047         }
4048 
4049         return ret;
4050 }
4051 
4052 #ifdef CONFIG_MEMCG
4053 void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s)
4054 {
4055         /*
4056          * Called with all the locks held after a sched RCU grace period.
4057          * Even if @s becomes empty after shrinking, we can't know that @s
4058          * doesn't have allocations already in-flight and thus can't
4059          * destroy @s until the associated memcg is released.
4060          *
4061          * However, let's remove the sysfs files for empty caches here.
4062          * Each cache has a lot of interface files which aren't
4063          * particularly useful for empty draining caches; otherwise, we can
4064          * easily end up with millions of unnecessary sysfs files on
4065          * systems which have a lot of memory and transient cgroups.
4066          */
4067         if (!__kmem_cache_shrink(s))
4068                 sysfs_slab_remove(s);
4069 }
4070 
4071 void __kmemcg_cache_deactivate(struct kmem_cache *s)
4072 {
4073         /*
4074          * Disable empty slabs caching. Used to avoid pinning offline
4075          * memory cgroups by kmem pages that can be freed.
4076          */
4077         slub_set_cpu_partial(s, 0);
4078         s->min_partial = 0;
4079 }
4080 #endif  /* CONFIG_MEMCG */
4081 
4082 static int slab_mem_going_offline_callback(void *arg)
4083 {
4084         struct kmem_cache *s;
4085 
4086         mutex_lock(&slab_mutex);
4087         list_for_each_entry(s, &slab_caches, list)
4088                 __kmem_cache_shrink(s);
4089         mutex_unlock(&slab_mutex);
4090 
4091         return 0;
4092 }
4093 
4094 static void slab_mem_offline_callback(void *arg)
4095 {
4096         struct kmem_cache_node *n;
4097         struct kmem_cache *s;
4098         struct memory_notify *marg = arg;
4099         int offline_node;
4100 
4101         offline_node = marg->status_change_nid_normal;
4102 
4103         /*
4104          * If the node still has available memory. we need kmem_cache_node
4105          * for it yet.
4106          */
4107         if (offline_node < 0)
4108                 return;
4109 
4110         mutex_lock(&slab_mutex);
4111         list_for_each_entry(s, &slab_caches, list) {
4112                 n = get_node(s, offline_node);
4113                 if (n) {
4114                         /*
4115                          * if n->nr_slabs > 0, slabs still exist on the node
4116                          * that is going down. We were unable to free them,
4117                          * and offline_pages() function shouldn't call this
4118                          * callback. So, we must fail.
4119                          */
4120                         BUG_ON(slabs_node(s, offline_node));
4121 
4122                         s->node[offline_node] = NULL;
4123                         kmem_cache_free(kmem_cache_node, n);
4124                 }
4125         }
4126         mutex_unlock(&slab_mutex);
4127 }
4128 
4129 static int slab_mem_going_online_callback(void *arg)
4130 {
4131         struct kmem_cache_node *n;
4132         struct kmem_cache *s;
4133         struct memory_notify *marg = arg;
4134         int nid = marg->status_change_nid_normal;
4135         int ret = 0;
4136 
4137         /*
4138          * If the node's memory is already available, then kmem_cache_node is
4139          * already created. Nothing to do.
4140          */
4141         if (nid < 0)
4142                 return 0;
4143 
4144         /*
4145          * We are bringing a node online. No memory is available yet. We must
4146          * allocate a kmem_cache_node structure in order to bring the node
4147          * online.
4148          */
4149         mutex_lock(&slab_mutex);
4150         list_for_each_entry(s, &slab_caches, list) {
4151                 /*
4152                  * XXX: kmem_cache_alloc_node will fallback to other nodes
4153                  *      since memory is not yet available from the node that
4154                  *      is brought up.
4155                  */
4156                 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
4157                 if (!n) {
4158                         ret = -ENOMEM;
4159                         goto out;
4160                 }
4161                 init_kmem_cache_node(n);
4162                 s->node[nid] = n;
4163         }
4164 out:
4165         mutex_unlock(&slab_mutex);
4166         return ret;
4167 }
4168 
4169 static int slab_memory_callback(struct notifier_block *self,
4170                                 unsigned long action, void *arg)
4171 {
4172         int ret = 0;
4173 
4174         switch (action) {
4175         case MEM_GOING_ONLINE:
4176                 ret = slab_mem_going_online_callback(arg);
4177                 break;
4178         case MEM_GOING_OFFLINE:
4179                 ret = slab_mem_going_offline_callback(arg);
4180                 break;
4181         case MEM_OFFLINE:
4182         case MEM_CANCEL_ONLINE:
4183                 slab_mem_offline_callback(arg);
4184                 break;
4185         case MEM_ONLINE:
4186         case MEM_CANCEL_OFFLINE:
4187                 break;
4188         }
4189         if (ret)
4190                 ret = notifier_from_errno(ret);
4191         else
4192                 ret = NOTIFY_OK;
4193         return ret;
4194 }
4195 
4196 static struct notifier_block slab_memory_callback_nb = {
4197         .notifier_call = slab_memory_callback,
4198         .priority = SLAB_CALLBACK_PRI,
4199 };
4200 
4201 /********************************************************************
4202  *                      Basic setup of slabs
4203  *******************************************************************/
4204 
4205 /*
4206  * Used for early kmem_cache structures that were allocated using
4207  * the page allocator. Allocate them properly then fix up the pointers
4208  * that may be pointing to the wrong kmem_cache structure.
4209  */
4210 
4211 static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
4212 {
4213         int node;
4214         struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
4215         struct kmem_cache_node *n;
4216 
4217         memcpy(s, static_cache, kmem_cache->object_size);
4218 
4219         /*
4220          * This runs very early, and only the boot processor is supposed to be
4221          * up.  Even if it weren't true, IRQs are not up so we couldn't fire
4222          * IPIs around.
4223          */
4224         __flush_cpu_slab(s, smp_processor_id());
4225         for_each_kmem_cache_node(s, node, n) {
4226                 struct page *p;
4227 
4228                 list_for_each_entry(p, &n->partial, slab_list)
4229                         p->slab_cache = s;
4230 
4231 #ifdef CONFIG_SLUB_DEBUG
4232                 list_for_each_entry(p, &n->full, slab_list)
4233                         p->slab_cache = s;
4234 #endif
4235         }
4236         slab_init_memcg_params(s);
4237         list_add(&s->list, &slab_caches);
4238         memcg_link_cache(s, NULL);
4239         return s;
4240 }
4241 
4242 void __init kmem_cache_init(void)
4243 {
4244         static __initdata struct kmem_cache boot_kmem_cache,
4245                 boot_kmem_cache_node;
4246 
4247         if (debug_guardpage_minorder())
4248                 slub_max_order = 0;
4249 
4250         kmem_cache_node = &boot_kmem_cache_node;
4251         kmem_cache = &boot_kmem_cache;
4252 
4253         create_boot_cache(kmem_cache_node, "kmem_cache_node",
4254                 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
4255 
4256         register_hotmemory_notifier(&slab_memory_callback_nb);
4257 
4258         /* Able to allocate the per node structures */
4259         slab_state = PARTIAL;
4260 
4261         create_boot_cache(kmem_cache, "kmem_cache",
4262                         offsetof(struct kmem_cache, node) +
4263                                 nr_node_ids * sizeof(struct kmem_cache_node *),
4264                        SLAB_HWCACHE_ALIGN, 0, 0);
4265 
4266         kmem_cache = bootstrap(&boot_kmem_cache);
4267         kmem_cache_node = bootstrap(&boot_kmem_cache_node);
4268 
4269         /* Now we can use the kmem_cache to allocate kmalloc slabs */
4270         setup_kmalloc_cache_index_table();
4271         create_kmalloc_caches(0);
4272 
4273         /* Setup random freelists for each cache */
4274         init_freelist_randomization();
4275 
4276         cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4277                                   slub_cpu_dead);
4278 
4279         pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
4280                 cache_line_size(),
4281                 slub_min_order, slub_max_order, slub_min_objects,
4282                 nr_cpu_ids, nr_node_ids);
4283 }
4284 
4285 void __init kmem_cache_init_late(void)
4286 {
4287 }
4288 
4289 struct kmem_cache *
4290 __kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
4291                    slab_flags_t flags, void (*ctor)(void *))
4292 {
4293         struct kmem_cache *s, *c;
4294 
4295         s = find_mergeable(size, align, flags, name, ctor);
4296         if (s) {
4297                 s->refcount++;
4298 
4299                 /*
4300                  * Adjust the object sizes so that we clear
4301                  * the complete object on kzalloc.
4302                  */
4303                 s->object_size = max(s->object_size, size);
4304                 s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
4305 
4306                 for_each_memcg_cache(c, s) {
4307                         c->object_size = s->object_size;
4308                         c->inuse = max(c->inuse, ALIGN(size, sizeof(void *)));
4309                 }
4310 
4311                 if (sysfs_slab_alias(s, name)) {
4312                         s->refcount--;
4313                         s = NULL;
4314                 }
4315         }
4316 
4317         return s;
4318 }
4319 
4320 int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
4321 {
4322         int err;
4323 
4324         err = kmem_cache_open(s, flags);
4325         if (err)
4326                 return err;
4327 
4328         /* Mutex is not taken during early boot */
4329         if (slab_state <= UP)
4330                 return 0;
4331 
4332         memcg_propagate_slab_attrs(s);
4333         err = sysfs_slab_add(s);
4334         if (err)
4335                 __kmem_cache_release(s);
4336 
4337         return err;
4338 }
4339 
4340 void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
4341 {
4342         struct kmem_cache *s;
4343         void *ret;
4344 
4345         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
4346                 return kmalloc_large(size, gfpflags);
4347 
4348         s = kmalloc_slab(size, gfpflags);
4349 
4350         if (unlikely(ZERO_OR_NULL_PTR(s)))
4351                 return s;
4352 
4353         ret = slab_alloc(s, gfpflags, caller);
4354 
4355         /* Honor the call site pointer we received. */
4356         trace_kmalloc(caller, ret, size, s->size, gfpflags);
4357 
4358         return ret;
4359 }
4360 
4361 #ifdef CONFIG_NUMA
4362 void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
4363                                         int node, unsigned long caller)
4364 {
4365         struct kmem_cache *s;
4366         void *ret;
4367 
4368         if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
4369                 ret = kmalloc_large_node(size, gfpflags, node);
4370 
4371                 trace_kmalloc_node(caller, ret,
4372                                    size, PAGE_SIZE << get_order(size),
4373                                    gfpflags, node);
4374 
4375                 return ret;
4376         }
4377 
4378         s = kmalloc_slab(size, gfpflags);
4379 
4380         if (unlikely(ZERO_OR_NULL_PTR(s)))
4381                 return s;
4382 
4383         ret = slab_alloc_node(s, gfpflags, node, caller);
4384 
4385         /* Honor the call site pointer we received. */
4386         trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
4387 
4388         return ret;
4389 }
4390 #endif
4391 
4392 #ifdef CONFIG_SYSFS
4393 static int count_inuse(struct page *page)
4394 {
4395         return page->inuse;
4396 }
4397 
4398 static int count_total(struct page *page)
4399 {
4400         return page->objects;
4401 }
4402 #endif
4403 
4404 #ifdef CONFIG_SLUB_DEBUG
4405 static int validate_slab(struct kmem_cache *s, struct page *page,
4406                                                 unsigned long *map)
4407 {
4408         void *p;
4409         void *addr = page_address(page);
4410 
4411         if (!check_slab(s, page) ||
4412                         !on_freelist(s, page, NULL))
4413                 return 0;
4414 
4415         /* Now we know that a valid freelist exists */
4416         bitmap_zero(map, page->objects);
4417 
4418         get_map(s, page, map);
4419         for_each_object(p, s, addr, page->objects) {
4420                 if (test_bit(slab_index(p, s, addr), map))
4421                         if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4422                                 return 0;
4423         }
4424 
4425         for_each_object(p, s, addr, page->objects)
4426                 if (!test_bit(slab_index(p, s, addr), map))
4427                         if (!check_object(s, page, p, SLUB_RED_ACTIVE))
4428                                 return 0;
4429         return 1;
4430 }
4431 
4432 static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4433                                                 unsigned long *map)
4434 {
4435         slab_lock(page);
4436         validate_slab(s, page, map);
4437         slab_unlock(page);
4438 }
4439 
4440 static int validate_slab_node(struct kmem_cache *s,
4441                 struct kmem_cache_node *n, unsigned long *map)
4442 {
4443         unsigned long count = 0;
4444         struct page *page;
4445         unsigned long flags;
4446 
4447         spin_lock_irqsave(&n->list_lock, flags);
4448 
4449         list_for_each_entry(page, &n->partial, slab_list) {
4450                 validate_slab_slab(s, page, map);
4451                 count++;
4452         }
4453         if (count != n->nr_partial)
4454                 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4455                        s->name, count, n->nr_partial);
4456 
4457         if (!(s->flags & SLAB_STORE_USER))
4458                 goto out;
4459 
4460         list_for_each_entry(page, &n->full, slab_list) {
4461                 validate_slab_slab(s, page, map);
4462                 count++;
4463         }
4464         if (count != atomic_long_read(&n->nr_slabs))
4465                 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4466                        s->name, count, atomic_long_read(&n->nr_slabs));
4467 
4468 out:
4469         spin_unlock_irqrestore(&n->list_lock, flags);
4470         return count;
4471 }
4472 
4473 static long validate_slab_cache(struct kmem_cache *s)
4474 {
4475         int node;
4476         unsigned long count = 0;
4477         struct kmem_cache_node *n;
4478         unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4479 
4480         if (!map)
4481                 return -ENOMEM;
4482 
4483         flush_all(s);
4484         for_each_kmem_cache_node(s, node, n)
4485                 count += validate_slab_node(s, n, map);
4486         bitmap_free(map);
4487         return count;
4488 }
4489 /*
4490  * Generate lists of code addresses where slabcache objects are allocated
4491  * and freed.
4492  */
4493 
4494 struct location {
4495         unsigned long count;
4496         unsigned long addr;
4497         long long sum_time;
4498         long min_time;
4499         long max_time;
4500         long min_pid;
4501         long max_pid;
4502         DECLARE_BITMAP(cpus, NR_CPUS);
4503         nodemask_t nodes;
4504 };
4505 
4506 struct loc_track {
4507         unsigned long max;
4508         unsigned long count;
4509         struct location *loc;
4510 };
4511 
4512 static void free_loc_track(struct loc_track *t)
4513 {
4514         if (t->max)
4515                 free_pages((unsigned long)t->loc,
4516                         get_order(sizeof(struct location) * t->max));
4517 }
4518 
4519 static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
4520 {
4521         struct location *l;
4522         int order;
4523 
4524         order = get_order(sizeof(struct location) * max);
4525 
4526         l = (void *)__get_free_pages(flags, order);
4527         if (!l)
4528                 return 0;
4529 
4530         if (t->count) {
4531                 memcpy(l, t->loc, sizeof(struct location) * t->count);
4532                 free_loc_track(t);
4533         }
4534         t->max = max;
4535         t->loc = l;
4536         return 1;
4537 }
4538 
4539 static int add_location(struct loc_track *t, struct kmem_cache *s,
4540                                 const struct track *track)
4541 {
4542         long start, end, pos;
4543         struct location *l;
4544         unsigned long caddr;
4545         unsigned long age = jiffies - track->when;
4546 
4547         start = -1;
4548         end = t->count;
4549 
4550         for ( ; ; ) {
4551                 pos = start + (end - start + 1) / 2;
4552 
4553                 /*
4554                  * There is nothing at "end". If we end up there
4555                  * we need to add something to before end.
4556                  */
4557                 if (pos == end)
4558                         break;
4559 
4560                 caddr = t->loc[pos].addr;
4561                 if (track->addr == caddr) {
4562 
4563                         l = &t->loc[pos];
4564                         l->count++;
4565                         if (track->when) {
4566                                 l->sum_time += age;
4567                                 if (age < l->min_time)
4568                                         l->min_time = age;
4569                                 if (age > l->max_time)
4570                                         l->max_time = age;
4571 
4572                                 if (track->pid < l->min_pid)
4573                                         l->min_pid = track->pid;
4574                                 if (track->pid > l->max_pid)
4575                                         l->max_pid = track->pid;
4576 
4577                                 cpumask_set_cpu(track->cpu,
4578                                                 to_cpumask(l->cpus));
4579                         }
4580                         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4581                         return 1;
4582                 }
4583 
4584                 if (track->addr < caddr)
4585                         end = pos;
4586                 else
4587                         start = pos;
4588         }
4589 
4590         /*
4591          * Not found. Insert new tracking element.
4592          */
4593         if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
4594                 return 0;
4595 
4596         l = t->loc + pos;
4597         if (pos < t->count)
4598                 memmove(l + 1, l,
4599                         (t->count - pos) * sizeof(struct location));
4600         t->count++;
4601         l->count = 1;
4602         l->addr = track->addr;
4603         l->sum_time = age;
4604         l->min_time = age;
4605         l->max_time = age;
4606         l->min_pid = track->pid;
4607         l->max_pid = track->pid;
4608         cpumask_clear(to_cpumask(l->cpus));
4609         cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
4610         nodes_clear(l->nodes);
4611         node_set(page_to_nid(virt_to_page(track)), l->nodes);
4612         return 1;
4613 }
4614 
4615 static void process_slab(struct loc_track *t, struct kmem_cache *s,
4616                 struct page *page, enum track_item alloc,
4617                 unsigned long *map)
4618 {
4619         void *addr = page_address(page);
4620         void *p;
4621 
4622         bitmap_zero(map, page->objects);
4623         get_map(s, page, map);
4624 
4625         for_each_object(p, s, addr, page->objects)
4626                 if (!test_bit(slab_index(p, s, addr), map))
4627                         add_location(t, s, get_track(s, p, alloc));
4628 }
4629 
4630 static int list_locations(struct kmem_cache *s, char *buf,
4631                                         enum track_item alloc)
4632 {
4633         int len = 0;
4634         unsigned long i;
4635         struct loc_track t = { 0, 0, NULL };
4636         int node;
4637         struct kmem_cache_node *n;
4638         unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
4639 
4640         if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4641                                      GFP_KERNEL)) {
4642                 bitmap_free(map);
4643                 return sprintf(buf, "Out of memory\n");
4644         }
4645         /* Push back cpu slabs */
4646         flush_all(s);
4647 
4648         for_each_kmem_cache_node(s, node, n) {
4649                 unsigned long flags;
4650                 struct page *page;
4651 
4652                 if (!atomic_long_read(&n->nr_slabs))
4653                         continue;
4654 
4655                 spin_lock_irqsave(&n->list_lock, flags);
4656                 list_for_each_entry(page, &n->partial, slab_list)
4657                         process_slab(&t, s, page, alloc, map);
4658                 list_for_each_entry(page, &n->full, slab_list)
4659                         process_slab(&t, s, page, alloc, map);
4660                 spin_unlock_irqrestore(&n->list_lock, flags);
4661         }
4662 
4663         for (i = 0; i < t.count; i++) {
4664                 struct location *l = &t.loc[i];
4665 
4666                 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
4667                         break;
4668                 len += sprintf(buf + len, "%7ld ", l->count);
4669 
4670                 if (l->addr)
4671                         len += sprintf(buf + len, "%pS", (void *)l->addr);
4672                 else
4673                         len += sprintf(buf + len, "<not-available>");
4674 
4675                 if (l->sum_time != l->min_time) {
4676                         len += sprintf(buf + len, " age=%ld/%ld/%ld",
4677                                 l->min_time,
4678                                 (long)div_u64(l->sum_time, l->count),
4679                                 l->max_time);
4680                 } else
4681                         len += sprintf(buf + len, " age=%ld",
4682                                 l->min_time);
4683 
4684                 if (l->min_pid != l->max_pid)
4685                         len += sprintf(buf + len, " pid=%ld-%ld",
4686                                 l->min_pid, l->max_pid);
4687                 else
4688                         len += sprintf(buf + len, " pid=%ld",
4689                                 l->min_pid);
4690 
4691                 if (num_online_cpus() > 1 &&
4692                                 !cpumask_empty(to_cpumask(l->cpus)) &&
4693                                 len < PAGE_SIZE - 60)
4694                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4695                                          " cpus=%*pbl",
4696                                          cpumask_pr_args(to_cpumask(l->cpus)));
4697 
4698                 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
4699                                 len < PAGE_SIZE - 60)
4700                         len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4701                                          " nodes=%*pbl",
4702                                          nodemask_pr_args(&l->nodes));
4703 
4704                 len += sprintf(buf + len, "\n");
4705         }
4706 
4707         free_loc_track(&t);
4708         bitmap_free(map);
4709         if (!t.count)
4710                 len += sprintf(buf, "No data\n");
4711         return len;
4712 }
4713 #endif  /* CONFIG_SLUB_DEBUG */
4714 
4715 #ifdef SLUB_RESILIENCY_TEST
4716 static void __init resiliency_test(void)
4717 {
4718         u8 *p;
4719         int type = KMALLOC_NORMAL;
4720 
4721         BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
4722 
4723         pr_err("SLUB resiliency testing\n");
4724         pr_err("-----------------------\n");
4725         pr_err("A. Corruption after allocation\n");
4726 
4727         p = kzalloc(16, GFP_KERNEL);
4728         p[16] = 0x12;
4729         pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4730                p + 16);
4731 
4732         validate_slab_cache(kmalloc_caches[type][4]);
4733 
4734         /* Hmmm... The next two are dangerous */
4735         p = kzalloc(32, GFP_KERNEL);
4736         p[32 + sizeof(void *)] = 0x34;
4737         pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4738                p);
4739         pr_err("If allocated object is overwritten then not detectable\n\n");
4740 
4741         validate_slab_cache(kmalloc_caches[type][5]);
4742         p = kzalloc(64, GFP_KERNEL);
4743         p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4744         *p = 0x56;
4745         pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4746                p);
4747         pr_err("If allocated object is overwritten then not detectable\n\n");
4748         validate_slab_cache(kmalloc_caches[type][6]);
4749 
4750         pr_err("\nB. Corruption after free\n");
4751         p = kzalloc(128, GFP_KERNEL);
4752         kfree(p);
4753         *p = 0x78;
4754         pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
4755         validate_slab_cache(kmalloc_caches[type][7]);
4756 
4757         p = kzalloc(256, GFP_KERNEL);
4758         kfree(p);
4759         p[50] = 0x9a;
4760         pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
4761         validate_slab_cache(kmalloc_caches[type][8]);
4762 
4763         p = kzalloc(512, GFP_KERNEL);
4764         kfree(p);
4765         p[512] = 0xab;
4766         pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
4767         validate_slab_cache(kmalloc_caches[type][9]);
4768 }
4769 #else
4770 #ifdef CONFIG_SYSFS
4771 static void resiliency_test(void) {};
4772 #endif
4773 #endif  /* SLUB_RESILIENCY_TEST */
4774 
4775 #ifdef CONFIG_SYSFS
4776 enum slab_stat_type {
4777         SL_ALL,                 /* All slabs */
4778         SL_PARTIAL,             /* Only partially allocated slabs */
4779         SL_CPU,                 /* Only slabs used for cpu caches */
4780         SL_OBJECTS,             /* Determine allocated objects not slabs */
4781         SL_TOTAL                /* Determine object capacity not slabs */
4782 };
4783 
4784 #define SO_ALL          (1 << SL_ALL)
4785 #define SO_PARTIAL      (1 << SL_PARTIAL)
4786 #define SO_CPU          (1 << SL_CPU)
4787 #define SO_OBJECTS      (1 << SL_OBJECTS)
4788 #define SO_TOTAL        (1 << SL_TOTAL)
4789 
4790 #ifdef CONFIG_MEMCG
4791 static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4792 
4793 static int __init setup_slub_memcg_sysfs(char *str)
4794 {
4795         int v;
4796 
4797         if (get_option(&str, &v) > 0)
4798                 memcg_sysfs_enabled = v;
4799 
4800         return 1;
4801 }
4802 
4803 __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4804 #endif
4805 
4806 static ssize_t show_slab_objects(struct kmem_cache *s,
4807                             char *buf, unsigned long flags)
4808 {
4809         unsigned long total = 0;
4810         int node;
4811         int x;
4812         unsigned long *nodes;
4813 
4814         nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
4815         if (!nodes)
4816                 return -ENOMEM;
4817 
4818         if (flags & SO_CPU) {
4819                 int cpu;
4820 
4821                 for_each_possible_cpu(cpu) {
4822                         struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4823                                                                cpu);
4824                         int node;
4825                         struct page *page;
4826 
4827                         page = READ_ONCE(c->page);
4828                         if (!page)
4829                                 continue;
4830 
4831                         node = page_to_nid(page);
4832                         if (flags & SO_TOTAL)
4833                                 x = page->objects;
4834                         else if (flags & SO_OBJECTS)
4835                                 x = page->inuse;
4836                         else
4837                                 x = 1;
4838 
4839                         total += x;
4840                         nodes[node] += x;
4841 
4842                         page = slub_percpu_partial_read_once(c);
4843                         if (page) {
4844                                 node = page_to_nid(page);
4845                                 if (flags & SO_TOTAL)
4846                                         WARN_ON_ONCE(1);
4847                                 else if (flags & SO_OBJECTS)
4848                                         WARN_ON_ONCE(1);
4849                                 else
4850                                         x = page->pages;
4851                                 total += x;
4852                                 nodes[node] += x;
4853                         }
4854                 }
4855         }
4856 
4857         /*
4858          * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
4859          * already held which will conflict with an existing lock order:
4860          *
4861          * mem_hotplug_lock->slab_mutex->kernfs_mutex
4862          *
4863          * We don't really need mem_hotplug_lock (to hold off
4864          * slab_mem_going_offline_callback) here because slab's memory hot
4865          * unplug code doesn't destroy the kmem_cache->node[] data.
4866          */
4867 
4868 #ifdef CONFIG_SLUB_DEBUG
4869         if (flags & SO_ALL) {
4870                 struct kmem_cache_node *n;
4871 
4872                 for_each_kmem_cache_node(s, node, n) {
4873 
4874                         if (flags & SO_TOTAL)
4875                                 x = atomic_long_read(&n->total_objects);
4876                         else if (flags & SO_OBJECTS)
4877                                 x = atomic_long_read(&n->total_objects) -
4878                                         count_partial(n, count_free);
4879                         else
4880                                 x = atomic_long_read(&n->nr_slabs);
4881                         total += x;
4882                         nodes[node] += x;
4883                 }
4884 
4885         } else
4886 #endif
4887         if (flags & SO_PARTIAL) {
4888                 struct kmem_cache_node *n;
4889 
4890                 for_each_kmem_cache_node(s, node, n) {
4891                         if (flags & SO_TOTAL)
4892                                 x = count_partial(n, count_total);
4893                         else if (flags & SO_OBJECTS)
4894                                 x = count_partial(n, count_inuse);
4895                         else
4896                                 x = n->nr_partial;
4897                         total += x;
4898                         nodes[node] += x;
4899                 }
4900         }
4901         x = sprintf(buf, "%lu", total);
4902 #ifdef CONFIG_NUMA
4903         for (node = 0; node < nr_node_ids; node++)
4904                 if (nodes[node])
4905                         x += sprintf(buf + x, " N%d=%lu",
4906                                         node, nodes[node]);
4907 #endif
4908         kfree(nodes);
4909         return x + sprintf(buf + x, "\n");
4910 }
4911 
4912 #ifdef CONFIG_SLUB_DEBUG
4913 static int any_slab_objects(struct kmem_cache *s)
4914 {
4915         int node;
4916         struct kmem_cache_node *n;
4917 
4918         for_each_kmem_cache_node(s, node, n)
4919                 if (atomic_long_read(&n->total_objects))
4920                         return 1;
4921 
4922         return 0;
4923 }
4924 #endif
4925 
4926 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4927 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4928 
4929 struct slab_attribute {
4930         struct attribute attr;
4931         ssize_t (*show)(struct kmem_cache *s, char *buf);
4932         ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4933 };
4934 
4935 #define SLAB_ATTR_RO(_name) \
4936         static struct slab_attribute _name##_attr = \
4937         __ATTR(_name, 0400, _name##_show, NULL)
4938 
4939 #define SLAB_ATTR(_name) \
4940         static struct slab_attribute _name##_attr =  \
4941         __ATTR(_name, 0600, _name##_show, _name##_store)
4942 
4943 static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4944 {
4945         return sprintf(buf, "%u\n", s->size);
4946 }
4947 SLAB_ATTR_RO(slab_size);
4948 
4949 static ssize_t align_show(struct kmem_cache *s, char *buf)
4950 {
4951         return sprintf(buf, "%u\n", s->align);
4952 }
4953 SLAB_ATTR_RO(align);
4954 
4955 static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4956 {
4957         return sprintf(buf, "%u\n", s->object_size);
4958 }
4959 SLAB_ATTR_RO(object_size);
4960 
4961 static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4962 {
4963         return sprintf(buf, "%u\n", oo_objects(s->oo));
4964 }
4965 SLAB_ATTR_RO(objs_per_slab);
4966 
4967 static ssize_t order_store(struct kmem_cache *s,
4968                                 const char *buf, size_t length)
4969 {
4970         unsigned int order;
4971         int err;
4972 
4973         err = kstrtouint(buf, 10, &order);
4974         if (err)
4975                 return err;
4976 
4977         if (order > slub_max_order || order < slub_min_order)
4978                 return -EINVAL;
4979 
4980         calculate_sizes(s, order);
4981         return length;
4982 }
4983 
4984 static ssize_t order_show(struct kmem_cache *s, char *buf)
4985 {
4986         return sprintf(buf, "%u\n", oo_order(s->oo));
4987 }
4988 SLAB_ATTR(order);
4989 
4990 static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4991 {
4992         return sprintf(buf, "%lu\n", s->min_partial);
4993 }
4994 
4995 static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4996                                  size_t length)
4997 {
4998         unsigned long min;
4999         int err;
5000 
5001         err = kstrtoul(buf, 10, &min);
5002         if (err)
5003                 return err;
5004 
5005         set_min_partial(s, min);
5006         return length;
5007 }
5008 SLAB_ATTR(min_partial);
5009 
5010 static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5011 {
5012         return sprintf(buf, "%u\n", slub_cpu_partial(s));
5013 }
5014 
5015 static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5016                                  size_t length)
5017 {
5018         unsigned int objects;
5019         int err;
5020 
5021         err = kstrtouint(buf, 10, &objects);
5022         if (err)
5023                 return err;
5024         if (objects && !kmem_cache_has_cpu_partial(s))
5025                 return -EINVAL;
5026 
5027         slub_set_cpu_partial(s, objects);
5028         flush_all(s);
5029         return length;
5030 }
5031 SLAB_ATTR(cpu_partial);
5032 
5033 static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5034 {
5035         if (!s->ctor)
5036                 return 0;
5037         return sprintf(buf, "%pS\n", s->ctor);
5038 }
5039 SLAB_ATTR_RO(ctor);
5040 
5041 static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5042 {
5043         return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
5044 }
5045 SLAB_ATTR_RO(aliases);
5046 
5047 static ssize_t partial_show(struct kmem_cache *s, char *buf)
5048 {
5049         return show_slab_objects(s, buf, SO_PARTIAL);
5050 }
5051 SLAB_ATTR_RO(partial);
5052 
5053 static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5054 {
5055         return show_slab_objects(s, buf, SO_CPU);
5056 }
5057 SLAB_ATTR_RO(cpu_slabs);
5058 
5059 static ssize_t objects_show(struct kmem_cache *s, char *buf)
5060 {
5061         return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
5062 }
5063 SLAB_ATTR_RO(objects);
5064 
5065 static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5066 {
5067         return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5068 }
5069 SLAB_ATTR_RO(objects_partial);
5070 
5071 static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5072 {
5073         int objects = 0;
5074         int pages = 0;
5075         int cpu;
5076         int len;
5077 
5078         for_each_online_cpu(cpu) {
5079                 struct page *page;
5080 
5081                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5082 
5083                 if (page) {
5084                         pages += page->pages;
5085                         objects += page->pobjects;
5086                 }
5087         }
5088 
5089         len = sprintf(buf, "%d(%d)", objects, pages);
5090 
5091 #ifdef CONFIG_SMP
5092         for_each_online_cpu(cpu) {
5093                 struct page *page;
5094 
5095                 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
5096 
5097                 if (page && len < PAGE_SIZE - 20)
5098                         len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5099                                 page->pobjects, page->pages);
5100         }
5101 #endif
5102         return len + sprintf(buf + len, "\n");
5103 }
5104 SLAB_ATTR_RO(slabs_cpu_partial);
5105 
5106 static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5107 {
5108         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5109 }
5110 
5111 static ssize_t reclaim_account_store(struct kmem_cache *s,
5112                                 const char *buf, size_t length)
5113 {
5114         s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5115         if (buf[0] == '1')
5116                 s->flags |= SLAB_RECLAIM_ACCOUNT;
5117         return length;
5118 }
5119 SLAB_ATTR(reclaim_account);
5120 
5121 static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5122 {
5123         return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5124 }
5125 SLAB_ATTR_RO(hwcache_align);
5126 
5127 #ifdef CONFIG_ZONE_DMA
5128 static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5129 {
5130         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5131 }
5132 SLAB_ATTR_RO(cache_dma);
5133 #endif
5134 
5135 static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5136 {
5137         return sprintf(buf, "%u\n", s->usersize);
5138 }
5139 SLAB_ATTR_RO(usersize);
5140 
5141 static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5142 {
5143         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
5144 }
5145 SLAB_ATTR_RO(destroy_by_rcu);
5146 
5147 #ifdef CONFIG_SLUB_DEBUG
5148 static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5149 {
5150         return show_slab_objects(s, buf, SO_ALL);
5151 }
5152 SLAB_ATTR_RO(slabs);
5153 
5154 static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5155 {
5156         return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5157 }
5158 SLAB_ATTR_RO(total_objects);
5159 
5160 static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5161 {
5162         return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
5163 }
5164 
5165 static ssize_t sanity_checks_store(struct kmem_cache *s,
5166                                 const char *buf, size_t length)
5167 {
5168         s->flags &= ~SLAB_CONSISTENCY_CHECKS;
5169         if (buf[0] == '1') {
5170                 s->flags &= ~__CMPXCHG_DOUBLE;
5171                 s->flags |= SLAB_CONSISTENCY_CHECKS;
5172         }
5173         return length;
5174 }
5175 SLAB_ATTR(sanity_checks);
5176 
5177 static ssize_t trace_show(struct kmem_cache *s, char *buf)
5178 {
5179         return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5180 }
5181 
5182 static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5183                                                         size_t length)
5184 {
5185         /*
5186          * Tracing a merged cache is going to give confusing results
5187          * as well as cause other issues like converting a mergeable
5188          * cache into an umergeable one.
5189          */
5190         if (s->refcount > 1)
5191                 return -EINVAL;
5192 
5193         s->flags &= ~SLAB_TRACE;
5194         if (buf[0] == '1') {
5195                 s->flags &= ~__CMPXCHG_DOUBLE;
5196                 s->flags |= SLAB_TRACE;
5197         }
5198         return length;
5199 }
5200 SLAB_ATTR(trace);
5201 
5202 static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5203 {
5204         return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5205 }
5206 
5207 static ssize_t red_zone_store(struct kmem_cache *s,
5208                                 const char *buf, size_t length)
5209 {
5210         if (any_slab_objects(s))
5211                 return -EBUSY;
5212 
5213         s->flags &= ~SLAB_RED_ZONE;
5214         if (buf[0] == '1') {
5215                 s->flags |= SLAB_RED_ZONE;
5216         }
5217         calculate_sizes(s, -1);
5218         return length;
5219 }
5220 SLAB_ATTR(red_zone);
5221 
5222 static ssize_t poison_show(struct kmem_cache *s, char *buf)
5223 {
5224         return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5225 }
5226 
5227 static ssize_t poison_store(struct kmem_cache *s,
5228                                 const char *buf, size_t length)
5229 {
5230         if (any_slab_objects(s))
5231                 return -EBUSY;
5232 
5233         s->flags &= ~SLAB_POISON;
5234         if (buf[0] == '1') {
5235                 s->flags |= SLAB_POISON;
5236         }
5237         calculate_sizes(s, -1);
5238         return length;
5239 }
5240 SLAB_ATTR(poison);
5241 
5242 static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5243 {
5244         return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5245 }
5246 
5247 static ssize_t store_user_store(struct kmem_cache *s,
5248                                 const char *buf, size_t length)
5249 {
5250         if (any_slab_objects(s))
5251                 return -EBUSY;
5252 
5253         s->flags &= ~SLAB_STORE_USER;
5254         if (buf[0] == '1') {
5255                 s->flags &= ~__CMPXCHG_DOUBLE;
5256                 s->flags |= SLAB_STORE_USER;
5257         }
5258         calculate_sizes(s, -1);
5259         return length;
5260 }
5261 SLAB_ATTR(store_user);
5262 
5263 static ssize_t validate_show(struct kmem_cache *s, char *buf)
5264 {
5265         return 0;
5266 }
5267 
5268 static ssize_t validate_store(struct kmem_cache *s,
5269                         const char *buf, size_t length)
5270 {
5271         int ret = -EINVAL;
5272 
5273         if (buf[0] == '1') {
5274                 ret = validate_slab_cache(s);
5275                 if (ret >= 0)
5276                         ret = length;
5277         }
5278         return ret;
5279 }
5280 SLAB_ATTR(validate);
5281 
5282 static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5283 {
5284         if (!(s->flags & SLAB_STORE_USER))
5285                 return -ENOSYS;
5286         return list_locations(s, buf, TRACK_ALLOC);
5287 }
5288 SLAB_ATTR_RO(alloc_calls);
5289 
5290 static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5291 {
5292         if (!(s->flags & SLAB_STORE_USER))
5293                 return -ENOSYS;
5294         return list_locations(s, buf, TRACK_FREE);
5295 }
5296 SLAB_ATTR_RO(free_calls);
5297 #endif /* CONFIG_SLUB_DEBUG */
5298 
5299 #ifdef CONFIG_FAILSLAB
5300 static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5301 {
5302         return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5303 }
5304 
5305 static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5306                                                         size_t length)
5307 {
5308         if (s->refcount > 1)
5309                 return -EINVAL;
5310 
5311         s->flags &= ~SLAB_FAILSLAB;
5312         if (buf[0] == '1')
5313                 s->flags |= SLAB_FAILSLAB;
5314         return length;
5315 }
5316 SLAB_ATTR(failslab);
5317 #endif
5318 
5319 static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5320 {
5321         return 0;
5322 }
5323 
5324 static ssize_t shrink_store(struct kmem_cache *s,
5325                         const char *buf, size_t length)
5326 {
5327         if (buf[0] == '1')
5328                 kmem_cache_shrink_all(s);
5329         else
5330                 return -EINVAL;
5331         return length;
5332 }
5333 SLAB_ATTR(shrink);
5334 
5335 #ifdef CONFIG_NUMA
5336 static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
5337 {
5338         return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
5339 }
5340 
5341 static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
5342                                 const char *buf, size_t length)
5343 {
5344         unsigned int ratio;
5345         int err;
5346 
5347         err = kstrtouint(buf, 10, &ratio);
5348         if (err)
5349                 return err;
5350         if (ratio > 100)
5351                 return -ERANGE;
5352 
5353         s->remote_node_defrag_ratio = ratio * 10;
5354 
5355         return length;
5356 }
5357 SLAB_ATTR(remote_node_defrag_ratio);
5358 #endif
5359 
5360 #ifdef CONFIG_SLUB_STATS
5361 static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5362 {
5363         unsigned long sum  = 0;
5364         int cpu;
5365         int len;
5366         int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
5367 
5368         if (!data)
5369                 return -ENOMEM;
5370 
5371         for_each_online_cpu(cpu) {
5372                 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
5373 
5374                 data[cpu] = x;
5375                 sum += x;
5376         }
5377 
5378         len = sprintf(buf, "%lu", sum);
5379 
5380 #ifdef CONFIG_SMP
5381         for_each_online_cpu(cpu) {
5382                 if (data[cpu] && len < PAGE_SIZE - 20)
5383                         len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
5384         }
5385 #endif
5386         kfree(data);
5387         return len + sprintf(buf + len, "\n");
5388 }
5389 
5390 static void clear_stat(struct kmem_cache *s, enum stat_item si)
5391 {
5392         int cpu;
5393 
5394         for_each_online_cpu(cpu)
5395                 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
5396 }
5397 
5398 #define STAT_ATTR(si, text)                                     \
5399 static ssize_t text##_show(struct kmem_cache *s, char *buf)     \
5400 {                                                               \
5401         return show_stat(s, buf, si);                           \
5402 }                                                               \
5403 static ssize_t text##_store(struct kmem_cache *s,               \
5404                                 const char *buf, size_t length) \
5405 {                                                               \
5406         if (buf[0] != '0')                                      \
5407                 return -EINVAL;                                 \
5408         clear_stat(s, si);                                      \
5409         return length;                                          \
5410 }                                                               \
5411 SLAB_ATTR(text);                                                \
5412 
5413 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5414 STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5415 STAT_ATTR(FREE_FASTPATH, free_fastpath);
5416 STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5417 STAT_ATTR(FREE_FROZEN, free_frozen);
5418 STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5419 STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5420 STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5421 STAT_ATTR(ALLOC_SLAB, alloc_slab);
5422 STAT_ATTR(ALLOC_REFILL, alloc_refill);
5423 STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
5424 STAT_ATTR(FREE_SLAB, free_slab);
5425 STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5426 STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5427 STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5428 STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5429 STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5430 STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
5431 STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
5432 STAT_ATTR(ORDER_FALLBACK, order_fallback);
5433 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5434 STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
5435 STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5436 STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
5437 STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5438 STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
5439 #endif  /* CONFIG_SLUB_STATS */
5440 
5441 static struct attribute *slab_attrs[] = {
5442         &slab_size_attr.attr,
5443         &object_size_attr.attr,
5444         &objs_per_slab_attr.attr,
5445         &order_attr.attr,
5446         &min_partial_attr.attr,
5447         &cpu_partial_attr.attr,
5448         &objects_attr.attr,
5449         &objects_partial_attr.attr,
5450         &partial_attr.attr,
5451         &cpu_slabs_attr.attr,
5452         &ctor_attr.attr,
5453         &aliases_attr.attr,
5454         &align_attr.attr,
5455         &hwcache_align_attr.attr,
5456         &reclaim_account_attr.attr,
5457         &destroy_by_rcu_attr.attr,
5458         &shrink_attr.attr,
5459         &slabs_cpu_partial_attr.attr,
5460 #ifdef CONFIG_SLUB_DEBUG
5461         &total_objects_attr.attr,
5462         &slabs_attr.attr,
5463         &sanity_checks_attr.attr,
5464         &trace_attr.attr,
5465         &red_zone_attr.attr,
5466         &poison_attr.attr,
5467         &store_user_attr.attr,
5468         &validate_attr.attr,
5469         &alloc_calls_attr.attr,
5470         &free_calls_attr.attr,
5471 #endif
5472 #ifdef CONFIG_ZONE_DMA
5473         &cache_dma_attr.attr,
5474 #endif
5475 #ifdef CONFIG_NUMA
5476         &remote_node_defrag_ratio_attr.attr,
5477 #endif
5478 #ifdef CONFIG_SLUB_STATS
5479         &alloc_fastpath_attr.attr,
5480         &alloc_slowpath_attr.attr,
5481         &free_fastpath_attr.attr,
5482         &free_slowpath_attr.attr,
5483         &free_frozen_attr.attr,
5484         &free_add_partial_attr.attr,
5485         &free_remove_partial_attr.attr,
5486         &alloc_from_partial_attr.attr,
5487         &alloc_slab_attr.attr,
5488         &alloc_refill_attr.attr,
5489         &alloc_node_mismatch_attr.attr,
5490         &free_slab_attr.attr,
5491         &cpuslab_flush_attr.attr,
5492         &deactivate_full_attr.attr,
5493         &deactivate_empty_attr.attr,
5494         &deactivate_to_head_attr.attr,
5495         &deactivate_to_tail_attr.attr,
5496         &deactivate_remote_frees_attr.attr,
5497         &deactivate_bypass_attr.attr,
5498         &order_fallback_attr.attr,
5499         &cmpxchg_double_fail_attr.attr,
5500         &cmpxchg_double_cpu_fail_attr.attr,
5501         &cpu_partial_alloc_attr.attr,
5502         &cpu_partial_free_attr.attr,
5503         &cpu_partial_node_attr.attr,
5504         &cpu_partial_drain_attr.attr,
5505 #endif
5506 #ifdef CONFIG_FAILSLAB
5507         &failslab_attr.attr,
5508 #endif
5509         &usersize_attr.attr,
5510 
5511         NULL
5512 };
5513 
5514 static const struct attribute_group slab_attr_group = {
5515         .attrs = slab_attrs,
5516 };
5517 
5518 static ssize_t slab_attr_show(struct kobject *kobj,
5519                                 struct attribute *attr,
5520                                 char *buf)
5521 {
5522         struct slab_attribute *attribute;
5523         struct kmem_cache *s;
5524         int err;
5525 
5526         attribute = to_slab_attr(attr);
5527         s = to_slab(kobj);
5528 
5529         if (!attribute->show)
5530                 return -EIO;
5531 
5532         err = attribute->show(s, buf);
5533 
5534         return err;
5535 }
5536 
5537 static ssize_t slab_attr_store(struct kobject *kobj,
5538                                 struct attribute *attr,
5539                                 const char *buf, size_t len)
5540 {
5541         struct slab_attribute *attribute;
5542         struct kmem_cache *s;
5543         int err;
5544 
5545         attribute = to_slab_attr(attr);
5546         s = to_slab(kobj);
5547 
5548         if (!attribute->store)
5549                 return -EIO;
5550 
5551         err = attribute->store(s, buf, len);
5552 #ifdef CONFIG_MEMCG
5553         if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5554                 struct kmem_cache *c;
5555 
5556                 mutex_lock(&slab_mutex);
5557                 if (s->max_attr_size < len)
5558                         s->max_attr_size = len;
5559 
5560                 /*
5561                  * This is a best effort propagation, so this function's return
5562                  * value will be determined by the parent cache only. This is
5563                  * basically because not all attributes will have a well
5564                  * defined semantics for rollbacks - most of the actions will
5565                  * have permanent effects.
5566                  *
5567                  * Returning the error value of any of the children that fail
5568                  * is not 100 % defined, in the sense that users seeing the
5569                  * error code won't be able to know anything about the state of
5570                  * the cache.
5571                  *
5572                  * Only returning the error code for the parent cache at least
5573                  * has well defined semantics. The cache being written to
5574                  * directly either failed or succeeded, in which case we loop
5575                  * through the descendants with best-effort propagation.
5576                  */
5577                 for_each_memcg_cache(c, s)
5578                         attribute->store(c, buf, len);
5579                 mutex_unlock(&slab_mutex);
5580         }
5581 #endif
5582         return err;
5583 }
5584 
5585 static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5586 {
5587 #ifdef CONFIG_MEMCG
5588         int i;
5589         char *buffer = NULL;
5590         struct kmem_cache *root_cache;
5591 
5592         if (is_root_cache(s))
5593                 return;
5594 
5595         root_cache = s->memcg_params.root_cache;
5596 
5597         /*
5598          * This mean this cache had no attribute written. Therefore, no point
5599          * in copying default values around
5600          */
5601         if (!root_cache->max_attr_size)
5602                 return;
5603 
5604         for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5605                 char mbuf[64];
5606                 char *buf;
5607                 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5608                 ssize_t len;
5609 
5610                 if (!attr || !attr->store || !attr->show)
5611                         continue;
5612 
5613                 /*
5614                  * It is really bad that we have to allocate here, so we will
5615                  * do it only as a fallback. If we actually allocate, though,
5616                  * we can just use the allocated buffer until the end.
5617                  *
5618                  * Most of the slub attributes will tend to be very small in
5619                  * size, but sysfs allows buffers up to a page, so they can
5620                  * theoretically happen.
5621                  */
5622                 if (buffer)
5623                         buf = buffer;
5624                 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
5625                         buf = mbuf;
5626                 else {
5627                         buffer = (char *) get_zeroed_page(GFP_KERNEL);
5628                         if (WARN_ON(!buffer))
5629                                 continue;
5630                         buf = buffer;
5631                 }
5632 
5633                 len = attr->show(root_cache, buf);
5634                 if (len > 0)
5635                         attr->store(s, buf, len);
5636         }
5637 
5638         if (buffer)
5639                 free_page((unsigned long)buffer);
5640 #endif  /* CONFIG_MEMCG */
5641 }
5642 
5643 static void kmem_cache_release(struct kobject *k)
5644 {
5645         slab_kmem_cache_release(to_slab(k));
5646 }
5647 
5648 static const struct sysfs_ops slab_sysfs_ops = {
5649         .show = slab_attr_show,
5650         .store = slab_attr_store,
5651 };
5652 
5653 static struct kobj_type slab_ktype = {
5654         .sysfs_ops = &slab_sysfs_ops,
5655         .release = kmem_cache_release,
5656 };
5657 
5658 static int uevent_filter(struct kset *kset, struct kobject *kobj)
5659 {
5660         struct kobj_type *ktype = get_ktype(kobj);
5661 
5662         if (ktype == &slab_ktype)
5663                 return 1;
5664         return 0;
5665 }
5666 
5667 static const struct kset_uevent_ops slab_uevent_ops = {
5668         .filter = uevent_filter,
5669 };
5670 
5671 static struct kset *slab_kset;
5672 
5673 static inline struct kset *cache_kset(struct kmem_cache *s)
5674 {
5675 #ifdef CONFIG_MEMCG
5676         if (!is_root_cache(s))
5677                 return s->memcg_params.root_cache->memcg_kset;
5678 #endif
5679         return slab_kset;
5680 }
5681 
5682 #define ID_STR_LENGTH 64
5683 
5684 /* Create a unique string id for a slab cache:
5685  *
5686  * Format       :[flags-]size
5687  */
5688 static char *create_unique_id(struct kmem_cache *s)
5689 {
5690         char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5691         char *p = name;
5692 
5693         BUG_ON(!name);
5694 
5695         *p++ = ':';
5696         /*
5697          * First flags affecting slabcache operations. We will only
5698          * get here for aliasable slabs so we do not need to support
5699          * too many flags. The flags here must cover all flags that
5700          * are matched during merging to guarantee that the id is
5701          * unique.
5702          */
5703         if (s->flags & SLAB_CACHE_DMA)
5704                 *p++ = 'd';
5705         if (s->flags & SLAB_CACHE_DMA32)
5706                 *p++ = 'D';
5707         if (s->flags & SLAB_RECLAIM_ACCOUNT)
5708                 *p++ = 'a';
5709         if (s->flags & SLAB_CONSISTENCY_CHECKS)
5710                 *p++ = 'F';
5711         if (s->flags & SLAB_ACCOUNT)
5712                 *p++ = 'A';
5713         if (p != name + 1)
5714                 *p++ = '-';
5715         p += sprintf(p, "%07u", s->size);
5716 
5717         BUG_ON(p > name + ID_STR_LENGTH - 1);
5718         return name;
5719 }
5720 
5721 static void sysfs_slab_remove_workfn(struct work_struct *work)
5722 {
5723         struct kmem_cache *s =
5724                 container_of(work, struct kmem_cache, kobj_remove_work);
5725 
5726         if (!s->kobj.state_in_sysfs)
5727                 /*
5728                  * For a memcg cache, this may be called during
5729                  * deactivation and again on shutdown.  Remove only once.
5730                  * A cache is never shut down before deactivation is
5731                  * complete, so no need to worry about synchronization.
5732                  */
5733                 goto out;
5734 
5735 #ifdef CONFIG_MEMCG
5736         kset_unregister(s->memcg_kset);
5737 #endif
5738         kobject_uevent(&s->kobj, KOBJ_REMOVE);
5739 out:
5740         kobject_put(&s->kobj);
5741 }
5742 
5743 static int sysfs_slab_add(struct kmem_cache *s)
5744 {
5745         int err;
5746         const char *name;
5747         struct kset *kset = cache_kset(s);
5748         int unmergeable = slab_unmergeable(s);
5749 
5750         INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5751 
5752         if (!kset) {
5753                 kobject_init(&s->kobj, &slab_ktype);
5754                 return 0;
5755         }
5756 
5757         if (!unmergeable && disable_higher_order_debug &&
5758                         (slub_debug & DEBUG_METADATA_FLAGS))
5759                 unmergeable = 1;
5760 
5761         if (unmergeable) {
5762                 /*
5763                  * Slabcache can never be merged so we can use the name proper.
5764                  * This is typically the case for debug situations. In that
5765                  * case we can catch duplicate names easily.
5766                  */
5767                 sysfs_remove_link(&slab_kset->kobj, s->name);
5768                 name = s->name;
5769         } else {
5770                 /*
5771                  * Create a unique name for the slab as a target
5772                  * for the symlinks.
5773                  */
5774                 name = create_unique_id(s);
5775         }
5776 
5777         s->kobj.kset = kset;
5778         err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
5779         if (err) {
5780                 kobject_put(&s->kobj);
5781                 goto out;
5782         }
5783 
5784         err = sysfs_create_group(&s->kobj, &slab_attr_group);
5785         if (err)
5786                 goto out_del_kobj;
5787 
5788 #ifdef CONFIG_MEMCG
5789         if (is_root_cache(s) && memcg_sysfs_enabled) {
5790                 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5791                 if (!s->memcg_kset) {
5792                         err = -ENOMEM;
5793                         goto out_del_kobj;
5794                 }
5795         }
5796 #endif
5797 
5798         kobject_uevent(&s->kobj, KOBJ_ADD);
5799         if (!unmergeable) {
5800                 /* Setup first alias */
5801                 sysfs_slab_alias(s, s->name);
5802         }
5803 out:
5804         if (!unmergeable)
5805                 kfree(name);
5806         return err;
5807 out_del_kobj:
5808         kobject_del(&s->kobj);
5809         goto out;
5810 }
5811 
5812 static void sysfs_slab_remove(struct kmem_cache *s)
5813 {
5814         if (slab_state < FULL)
5815                 /*
5816                  * Sysfs has not been setup yet so no need to remove the
5817                  * cache from sysfs.
5818                  */
5819                 return;
5820 
5821         kobject_get(&s->kobj);
5822         schedule_work(&s->kobj_remove_work);
5823 }
5824 
5825 void sysfs_slab_unlink(struct kmem_cache *s)
5826 {
5827         if (slab_state >= FULL)
5828                 kobject_del(&s->kobj);
5829 }
5830 
5831 void sysfs_slab_release(struct kmem_cache *s)
5832 {
5833         if (slab_state >= FULL)
5834                 kobject_put(&s->kobj);
5835 }
5836 
5837 /*
5838  * Need to buffer aliases during bootup until sysfs becomes
5839  * available lest we lose that information.
5840  */
5841 struct saved_alias {
5842         struct kmem_cache *s;
5843         const char *name;
5844         struct saved_alias *next;
5845 };
5846 
5847 static struct saved_alias *alias_list;
5848 
5849 static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5850 {
5851         struct saved_alias *al;
5852 
5853         if (slab_state == FULL) {
5854                 /*
5855                  * If we have a leftover link then remove it.
5856                  */
5857                 sysfs_remove_link(&slab_kset->kobj, name);
5858                 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
5859         }
5860 
5861         al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5862         if (!al)
5863                 return -ENOMEM;
5864 
5865         al->s = s;
5866         al->name = name;
5867         al->next = alias_list;
5868         alias_list = al;
5869         return 0;
5870 }
5871 
5872 static int __init slab_sysfs_init(void)
5873 {
5874         struct kmem_cache *s;
5875         int err;
5876 
5877         mutex_lock(&slab_mutex);
5878 
5879         slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
5880         if (!slab_kset) {
5881                 mutex_unlock(&slab_mutex);
5882                 pr_err("Cannot register slab subsystem.\n");
5883                 return -ENOSYS;
5884         }
5885 
5886         slab_state = FULL;
5887 
5888         list_for_each_entry(s, &slab_caches, list) {
5889                 err = sysfs_slab_add(s);
5890                 if (err)
5891                         pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5892                                s->name);
5893         }
5894 
5895         while (alias_list) {
5896                 struct saved_alias *al = alias_list;
5897 
5898                 alias_list = alias_list->next;
5899                 err = sysfs_slab_alias(al->s, al->name);
5900                 if (err)
5901                         pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5902                                al->name);
5903                 kfree(al);
5904         }
5905 
5906         mutex_unlock(&slab_mutex);
5907         resiliency_test();
5908         return 0;
5909 }
5910 
5911 __initcall(slab_sysfs_init);
5912 #endif /* CONFIG_SYSFS */
5913 
5914 /*
5915  * The /proc/slabinfo ABI
5916  */
5917 #ifdef CONFIG_SLUB_DEBUG
5918 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
5919 {
5920         unsigned long nr_slabs = 0;
5921         unsigned long nr_objs = 0;
5922         unsigned long nr_free = 0;
5923         int node;
5924         struct kmem_cache_node *n;
5925 
5926         for_each_kmem_cache_node(s, node, n) {
5927                 nr_slabs += node_nr_slabs(n);
5928                 nr_objs += node_nr_objs(n);
5929                 nr_free += count_partial(n, count_free);
5930         }
5931 
5932         sinfo->active_objs = nr_objs - nr_free;
5933         sinfo->num_objs = nr_objs;
5934         sinfo->active_slabs = nr_slabs;
5935         sinfo->num_slabs = nr_slabs;
5936         sinfo->objects_per_slab = oo_objects(s->oo);
5937         sinfo->cache_order = oo_order(s->oo);
5938 }
5939 
5940 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
5941 {
5942 }
5943 
5944 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5945                        size_t count, loff_t *ppos)
5946 {
5947         return -EIO;
5948 }
5949 #endif /* CONFIG_SLUB_DEBUG */

/* [<][>][^][v][top][bottom][index][help] */