Merge suspend-to-idle rework material for v5.4.
[linux-2.6-block.git] / mm / slub.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
81819f0f
CL
2/*
3 * SLUB: A slab allocator that limits cache line use instead of queuing
4 * objects in per cpu and per node lists.
5 *
881db7fb
CL
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.
81819f0f 8 *
cde53535 9 * (C) 2007 SGI, Christoph Lameter
881db7fb 10 * (C) 2011 Linux Foundation, Christoph Lameter
81819f0f
CL
11 */
12
13#include <linux/mm.h>
1eb5ac64 14#include <linux/swap.h> /* struct reclaim_state */
81819f0f
CL
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>
97d06609 20#include "slab.h"
7b3c3a50 21#include <linux/proc_fs.h>
81819f0f 22#include <linux/seq_file.h>
a79316c6 23#include <linux/kasan.h>
81819f0f
CL
24#include <linux/cpu.h>
25#include <linux/cpuset.h>
26#include <linux/mempolicy.h>
27#include <linux/ctype.h>
3ac7fe5a 28#include <linux/debugobjects.h>
81819f0f 29#include <linux/kallsyms.h>
b9049e23 30#include <linux/memory.h>
f8bd2258 31#include <linux/math64.h>
773ff60e 32#include <linux/fault-inject.h>
bfa71457 33#include <linux/stacktrace.h>
4de900b4 34#include <linux/prefetch.h>
2633d7a0 35#include <linux/memcontrol.h>
2482ddec 36#include <linux/random.h>
81819f0f 37
4a92379b
RK
38#include <trace/events/kmem.h>
39
072bb0aa
MG
40#include "internal.h"
41
81819f0f
CL
42/*
43 * Lock order:
18004c5d 44 * 1. slab_mutex (Global Mutex)
881db7fb
CL
45 * 2. node->list_lock
46 * 3. slab_lock(page) (Only on some arches and for debugging)
81819f0f 47 *
18004c5d 48 * slab_mutex
881db7fb 49 *
18004c5d 50 * The role of the slab_mutex is to protect the list of all the slabs
881db7fb
CL
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
b7ccc7f8 54 * have the ability to do a cmpxchg_double. It only protects:
881db7fb 55 * A. page->freelist -> List of object free in a page
b7ccc7f8
MW
56 * B. page->inuse -> Number of objects in use
57 * C. page->objects -> Number of objects in page
58 * D. page->frozen -> frozen state
881db7fb
CL
59 *
60 * If a slab is frozen then it is exempt from list management. It is not
632b2ef0
LX
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.
81819f0f
CL
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.
81819f0f
CL
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 *
672bba3a
CL
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
81819f0f 88 * freed then the slab will show up again on the partial lists.
672bba3a
CL
89 * We track full slabs for debugging purposes though because otherwise we
90 * cannot scan all objects.
81819f0f
CL
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 *
4b6f0750
CL
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
dfb4f096 110 * freelist that allows lockless access to
894b8788
CL
111 * free objects in addition to the regular freelist
112 * that requires the slab lock.
81819f0f
CL
113 *
114 * PageError Slab requires special handling due to debug
115 * options set. This moves slab handling out of
894b8788 116 * the fast path and disables lockless freelists.
81819f0f
CL
117 */
118
af537b0a
CL
119static inline int kmem_cache_debug(struct kmem_cache *s)
120{
5577bd8a 121#ifdef CONFIG_SLUB_DEBUG
af537b0a 122 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
5577bd8a 123#else
af537b0a 124 return 0;
5577bd8a 125#endif
af537b0a 126}
5577bd8a 127
117d54df 128void *fixup_red_left(struct kmem_cache *s, void *p)
d86bd1be
JK
129{
130 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
131 p += s->red_left_pad;
132
133 return p;
134}
135
345c905d
JK
136static 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
81819f0f
CL
145/*
146 * Issues still to be resolved:
147 *
81819f0f
CL
148 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
149 *
81819f0f
CL
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
b789ef51
CL
156/* Enable to log cmpxchg failures */
157#undef SLUB_DEBUG_CMPXCHG
158
2086d26a
CL
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 */
76be8950 163#define MIN_PARTIAL 5
e95eed57 164
2086d26a
CL
165/*
166 * Maximum number of desirable partial slabs.
167 * The existence of more partial slabs makes kmem_cache_shrink
721ae22a 168 * sort the partial list by the number of objects in use.
2086d26a
CL
169 */
170#define MAX_PARTIAL 10
171
becfda68 172#define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
81819f0f 173 SLAB_POISON | SLAB_STORE_USER)
672bba3a 174
149daaf3
LA
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
fa5ec8a1 183/*
3de47213
DR
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.
fa5ec8a1 187 */
3de47213 188#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
fa5ec8a1 189
210b5c06
CG
190#define OO_SHIFT 16
191#define OO_MASK ((1 << OO_SHIFT) - 1)
50d5c41c 192#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
210b5c06 193
81819f0f 194/* Internal SLUB flags */
d50112ed 195/* Poison object */
4fd0b46e 196#define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
d50112ed 197/* Use cmpxchg_double */
4fd0b46e 198#define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
81819f0f 199
02cbc874
CL
200/*
201 * Tracking user of a slab.
202 */
d6543e39 203#define TRACK_ADDRS_COUNT 16
02cbc874 204struct track {
ce71e27c 205 unsigned long addr; /* Called from address */
d6543e39
BG
206#ifdef CONFIG_STACKTRACE
207 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
208#endif
02cbc874
CL
209 int cpu; /* Was running on cpu */
210 int pid; /* Pid context */
211 unsigned long when; /* When did the operation occur */
212};
213
214enum track_item { TRACK_ALLOC, TRACK_FREE };
215
ab4d5ed5 216#ifdef CONFIG_SYSFS
81819f0f
CL
217static int sysfs_slab_add(struct kmem_cache *);
218static int sysfs_slab_alias(struct kmem_cache *, const char *);
107dab5c 219static void memcg_propagate_slab_attrs(struct kmem_cache *s);
bf5eb3de 220static void sysfs_slab_remove(struct kmem_cache *s);
81819f0f 221#else
0c710013
CL
222static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
223static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
224 { return 0; }
107dab5c 225static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
bf5eb3de 226static inline void sysfs_slab_remove(struct kmem_cache *s) { }
81819f0f
CL
227#endif
228
4fdccdfb 229static inline void stat(const struct kmem_cache *s, enum stat_item si)
8ff12cfc
CL
230{
231#ifdef CONFIG_SLUB_STATS
88da03a6
CL
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]);
8ff12cfc
CL
237#endif
238}
239
81819f0f
CL
240/********************************************************************
241 * Core slab cache functions
242 *******************************************************************/
243
2482ddec
KC
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 */
249static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
250 unsigned long ptr_addr)
251{
252#ifdef CONFIG_SLAB_FREELIST_HARDENED
d36a63a9
AK
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 (unsigned long)kasan_reset_tag((void *)ptr_addr));
2482ddec
KC
265#else
266 return ptr;
267#endif
268}
269
270/* Returns the freelist pointer recorded at location ptr_addr. */
271static 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
7656c72b
CL
278static inline void *get_freepointer(struct kmem_cache *s, void *object)
279{
2482ddec 280 return freelist_dereference(s, object + s->offset);
7656c72b
CL
281}
282
0ad9500e
ED
283static void prefetch_freepointer(const struct kmem_cache *s, void *object)
284{
0882ff91 285 prefetch(object + s->offset);
0ad9500e
ED
286}
287
1393d9a1
CL
288static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
289{
2482ddec 290 unsigned long freepointer_addr;
1393d9a1
CL
291 void *p;
292
922d566c
JK
293 if (!debug_pagealloc_enabled())
294 return get_freepointer(s, object);
295
2482ddec
KC
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);
1393d9a1
CL
299}
300
7656c72b
CL
301static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
302{
2482ddec
KC
303 unsigned long freeptr_addr = (unsigned long)object + s->offset;
304
ce6fa91b
AP
305#ifdef CONFIG_SLAB_FREELIST_HARDENED
306 BUG_ON(object == fp); /* naive detection of double free or corruption */
307#endif
308
2482ddec 309 *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
7656c72b
CL
310}
311
312/* Loop over all objects in a slab */
224a88be 313#define for_each_object(__p, __s, __addr, __objects) \
d86bd1be
JK
314 for (__p = fixup_red_left(__s, __addr); \
315 __p < (__addr) + (__objects) * (__s)->size; \
316 __p += (__s)->size)
7656c72b 317
7656c72b 318/* Determine object index from a given position */
284b50dd 319static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr)
7656c72b 320{
6373dca1 321 return (kasan_reset_tag(p) - addr) / s->size;
7656c72b
CL
322}
323
9736d2a9 324static inline unsigned int order_objects(unsigned int order, unsigned int size)
ab9a0f19 325{
9736d2a9 326 return ((unsigned int)PAGE_SIZE << order) / size;
ab9a0f19
LJ
327}
328
19af27af 329static inline struct kmem_cache_order_objects oo_make(unsigned int order,
9736d2a9 330 unsigned int size)
834f3d11
CL
331{
332 struct kmem_cache_order_objects x = {
9736d2a9 333 (order << OO_SHIFT) + order_objects(order, size)
834f3d11
CL
334 };
335
336 return x;
337}
338
19af27af 339static inline unsigned int oo_order(struct kmem_cache_order_objects x)
834f3d11 340{
210b5c06 341 return x.x >> OO_SHIFT;
834f3d11
CL
342}
343
19af27af 344static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
834f3d11 345{
210b5c06 346 return x.x & OO_MASK;
834f3d11
CL
347}
348
881db7fb
CL
349/*
350 * Per slab locking using the pagelock
351 */
352static __always_inline void slab_lock(struct page *page)
353{
48c935ad 354 VM_BUG_ON_PAGE(PageTail(page), page);
881db7fb
CL
355 bit_spin_lock(PG_locked, &page->flags);
356}
357
358static __always_inline void slab_unlock(struct page *page)
359{
48c935ad 360 VM_BUG_ON_PAGE(PageTail(page), page);
881db7fb
CL
361 __bit_spin_unlock(PG_locked, &page->flags);
362}
363
1d07171c
CL
364/* Interrupts must be disabled (for the fallback code to work right) */
365static 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());
2565409f
HC
371#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
372 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
1d07171c 373 if (s->flags & __CMPXCHG_DOUBLE) {
cdcd6298 374 if (cmpxchg_double(&page->freelist, &page->counters,
0aa9a13d
DC
375 freelist_old, counters_old,
376 freelist_new, counters_new))
6f6528a1 377 return true;
1d07171c
CL
378 } else
379#endif
380 {
381 slab_lock(page);
d0e0ac97
CG
382 if (page->freelist == freelist_old &&
383 page->counters == counters_old) {
1d07171c 384 page->freelist = freelist_new;
7d27a04b 385 page->counters = counters_new;
1d07171c 386 slab_unlock(page);
6f6528a1 387 return true;
1d07171c
CL
388 }
389 slab_unlock(page);
390 }
391
392 cpu_relax();
393 stat(s, CMPXCHG_DOUBLE_FAIL);
394
395#ifdef SLUB_DEBUG_CMPXCHG
f9f58285 396 pr_info("%s %s: cmpxchg double redo ", n, s->name);
1d07171c
CL
397#endif
398
6f6528a1 399 return false;
1d07171c
CL
400}
401
b789ef51
CL
402static 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{
2565409f
HC
407#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
408 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
b789ef51 409 if (s->flags & __CMPXCHG_DOUBLE) {
cdcd6298 410 if (cmpxchg_double(&page->freelist, &page->counters,
0aa9a13d
DC
411 freelist_old, counters_old,
412 freelist_new, counters_new))
6f6528a1 413 return true;
b789ef51
CL
414 } else
415#endif
416 {
1d07171c
CL
417 unsigned long flags;
418
419 local_irq_save(flags);
881db7fb 420 slab_lock(page);
d0e0ac97
CG
421 if (page->freelist == freelist_old &&
422 page->counters == counters_old) {
b789ef51 423 page->freelist = freelist_new;
7d27a04b 424 page->counters = counters_new;
881db7fb 425 slab_unlock(page);
1d07171c 426 local_irq_restore(flags);
6f6528a1 427 return true;
b789ef51 428 }
881db7fb 429 slab_unlock(page);
1d07171c 430 local_irq_restore(flags);
b789ef51
CL
431 }
432
433 cpu_relax();
434 stat(s, CMPXCHG_DOUBLE_FAIL);
435
436#ifdef SLUB_DEBUG_CMPXCHG
f9f58285 437 pr_info("%s %s: cmpxchg double redo ", n, s->name);
b789ef51
CL
438#endif
439
6f6528a1 440 return false;
b789ef51
CL
441}
442
41ecc55b 443#ifdef CONFIG_SLUB_DEBUG
5f80b13a
CL
444/*
445 * Determine a map of object in use on a page.
446 *
881db7fb 447 * Node listlock must be held to guarantee that the page does
5f80b13a
CL
448 * not vanish from under us.
449 */
450static 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
870b1fbb 459static inline unsigned int size_from_object(struct kmem_cache *s)
d86bd1be
JK
460{
461 if (s->flags & SLAB_RED_ZONE)
462 return s->size - s->red_left_pad;
463
464 return s->size;
465}
466
467static 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
41ecc55b
CL
475/*
476 * Debug settings:
477 */
89d3c87e 478#if defined(CONFIG_SLUB_DEBUG_ON)
d50112ed 479static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
f0630fff 480#else
d50112ed 481static slab_flags_t slub_debug;
f0630fff 482#endif
41ecc55b
CL
483
484static char *slub_debug_slabs;
fa5ec8a1 485static int disable_higher_order_debug;
41ecc55b 486
a79316c6
AR
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 */
493static inline void metadata_access_enable(void)
494{
495 kasan_disable_current();
496}
497
498static inline void metadata_access_disable(void)
499{
500 kasan_enable_current();
501}
502
81819f0f
CL
503/*
504 * Object debugging
505 */
d86bd1be
JK
506
507/* Verify that a pointer has an address that is valid within a slab page */
508static 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);
338cfaad 517 object = kasan_reset_tag(object);
d86bd1be
JK
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
aa2efd5e
DT
527static void print_section(char *level, char *text, u8 *addr,
528 unsigned int length)
81819f0f 529{
a79316c6 530 metadata_access_enable();
aa2efd5e 531 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
ffc79d28 532 length, 1);
a79316c6 533 metadata_access_disable();
81819f0f
CL
534}
535
81819f0f
CL
536static 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
549static void set_track(struct kmem_cache *s, void *object,
ce71e27c 550 enum track_item alloc, unsigned long addr)
81819f0f 551{
1a00df4a 552 struct track *p = get_track(s, object, alloc);
81819f0f 553
81819f0f 554 if (addr) {
d6543e39 555#ifdef CONFIG_STACKTRACE
79716799 556 unsigned int nr_entries;
d6543e39 557
a79316c6 558 metadata_access_enable();
79716799 559 nr_entries = stack_trace_save(p->addrs, TRACK_ADDRS_COUNT, 3);
a79316c6 560 metadata_access_disable();
d6543e39 561
79716799
TG
562 if (nr_entries < TRACK_ADDRS_COUNT)
563 p->addrs[nr_entries] = 0;
d6543e39 564#endif
81819f0f
CL
565 p->addr = addr;
566 p->cpu = smp_processor_id();
88e4ccf2 567 p->pid = current->pid;
81819f0f 568 p->when = jiffies;
b8ca7ff7 569 } else {
81819f0f 570 memset(p, 0, sizeof(struct track));
b8ca7ff7 571 }
81819f0f
CL
572}
573
81819f0f
CL
574static void init_tracking(struct kmem_cache *s, void *object)
575{
24922684
CL
576 if (!(s->flags & SLAB_STORE_USER))
577 return;
578
ce71e27c
EGM
579 set_track(s, object, TRACK_FREE, 0UL);
580 set_track(s, object, TRACK_ALLOC, 0UL);
81819f0f
CL
581}
582
86609d33 583static void print_track(const char *s, struct track *t, unsigned long pr_time)
81819f0f
CL
584{
585 if (!t->addr)
586 return;
587
f9f58285 588 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
86609d33 589 s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
d6543e39
BG
590#ifdef CONFIG_STACKTRACE
591 {
592 int i;
593 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
594 if (t->addrs[i])
f9f58285 595 pr_err("\t%pS\n", (void *)t->addrs[i]);
d6543e39
BG
596 else
597 break;
598 }
599#endif
24922684
CL
600}
601
602static void print_tracking(struct kmem_cache *s, void *object)
603{
86609d33 604 unsigned long pr_time = jiffies;
24922684
CL
605 if (!(s->flags & SLAB_STORE_USER))
606 return;
607
86609d33
CP
608 print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
609 print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
24922684
CL
610}
611
612static void print_page_info(struct page *page)
613{
f9f58285 614 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
d0e0ac97 615 page, page->objects, page->inuse, page->freelist, page->flags);
24922684
CL
616
617}
618
619static void slab_bug(struct kmem_cache *s, char *fmt, ...)
620{
ecc42fbe 621 struct va_format vaf;
24922684 622 va_list args;
24922684
CL
623
624 va_start(args, fmt);
ecc42fbe
FF
625 vaf.fmt = fmt;
626 vaf.va = &args;
f9f58285 627 pr_err("=============================================================================\n");
ecc42fbe 628 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
f9f58285 629 pr_err("-----------------------------------------------------------------------------\n\n");
645df230 630
373d4d09 631 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
ecc42fbe 632 va_end(args);
81819f0f
CL
633}
634
24922684
CL
635static void slab_fix(struct kmem_cache *s, char *fmt, ...)
636{
ecc42fbe 637 struct va_format vaf;
24922684 638 va_list args;
24922684
CL
639
640 va_start(args, fmt);
ecc42fbe
FF
641 vaf.fmt = fmt;
642 vaf.va = &args;
643 pr_err("FIX %s: %pV\n", s->name, &vaf);
24922684 644 va_end(args);
24922684
CL
645}
646
647static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
81819f0f
CL
648{
649 unsigned int off; /* Offset of last byte */
a973e9dd 650 u8 *addr = page_address(page);
24922684
CL
651
652 print_tracking(s, p);
653
654 print_page_info(page);
655
f9f58285
FF
656 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
657 p, p - addr, get_freepointer(s, p));
24922684 658
d86bd1be 659 if (s->flags & SLAB_RED_ZONE)
aa2efd5e
DT
660 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
661 s->red_left_pad);
d86bd1be 662 else if (p > addr + 16)
aa2efd5e 663 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
81819f0f 664
aa2efd5e 665 print_section(KERN_ERR, "Object ", p,
1b473f29 666 min_t(unsigned int, s->object_size, PAGE_SIZE));
81819f0f 667 if (s->flags & SLAB_RED_ZONE)
aa2efd5e 668 print_section(KERN_ERR, "Redzone ", p + s->object_size,
3b0efdfa 669 s->inuse - s->object_size);
81819f0f 670
81819f0f
CL
671 if (s->offset)
672 off = s->offset + sizeof(void *);
673 else
674 off = s->inuse;
675
24922684 676 if (s->flags & SLAB_STORE_USER)
81819f0f 677 off += 2 * sizeof(struct track);
81819f0f 678
80a9201a
AP
679 off += kasan_metadata_size(s);
680
d86bd1be 681 if (off != size_from_object(s))
81819f0f 682 /* Beginning of the filler is the free pointer */
aa2efd5e
DT
683 print_section(KERN_ERR, "Padding ", p + off,
684 size_from_object(s) - off);
24922684
CL
685
686 dump_stack();
81819f0f
CL
687}
688
75c66def 689void object_err(struct kmem_cache *s, struct page *page,
81819f0f
CL
690 u8 *object, char *reason)
691{
3dc50637 692 slab_bug(s, "%s", reason);
24922684 693 print_trailer(s, page, object);
81819f0f
CL
694}
695
a38965bf 696static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
d0e0ac97 697 const char *fmt, ...)
81819f0f
CL
698{
699 va_list args;
700 char buf[100];
701
24922684
CL
702 va_start(args, fmt);
703 vsnprintf(buf, sizeof(buf), fmt, args);
81819f0f 704 va_end(args);
3dc50637 705 slab_bug(s, "%s", buf);
24922684 706 print_page_info(page);
81819f0f
CL
707 dump_stack();
708}
709
f7cb1933 710static void init_object(struct kmem_cache *s, void *object, u8 val)
81819f0f
CL
711{
712 u8 *p = object;
713
d86bd1be
JK
714 if (s->flags & SLAB_RED_ZONE)
715 memset(p - s->red_left_pad, val, s->red_left_pad);
716
81819f0f 717 if (s->flags & __OBJECT_POISON) {
3b0efdfa
CL
718 memset(p, POISON_FREE, s->object_size - 1);
719 p[s->object_size - 1] = POISON_END;
81819f0f
CL
720 }
721
722 if (s->flags & SLAB_RED_ZONE)
3b0efdfa 723 memset(p + s->object_size, val, s->inuse - s->object_size);
81819f0f
CL
724}
725
24922684
CL
726static 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
733static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
734 u8 *object, char *what,
06428780 735 u8 *start, unsigned int value, unsigned int bytes)
24922684
CL
736{
737 u8 *fault;
738 u8 *end;
739
a79316c6 740 metadata_access_enable();
79824820 741 fault = memchr_inv(start, value, bytes);
a79316c6 742 metadata_access_disable();
24922684
CL
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);
f9f58285 751 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
24922684
CL
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;
81819f0f
CL
757}
758
81819f0f
CL
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.
672bba3a 766 *
81819f0f
CL
767 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
768 * 0xa5 (POISON_END)
769 *
3b0efdfa 770 * object + s->object_size
81819f0f 771 * Padding to reach word boundary. This is also used for Redzoning.
672bba3a 772 * Padding is extended by another word if Redzoning is enabled and
3b0efdfa 773 * object_size == inuse.
672bba3a 774 *
81819f0f
CL
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
672bba3a
CL
779 * Meta data starts here.
780 *
81819f0f
CL
781 * A. Free pointer (if we cannot overwrite object on free)
782 * B. Tracking data for SLAB_STORE_USER
672bba3a 783 * C. Padding to reach required alignment boundary or at mininum
6446faa2 784 * one word if debugging is on to be able to detect writes
672bba3a
CL
785 * before the word boundary.
786 *
787 * Padding is done using 0x5a (POISON_INUSE)
81819f0f
CL
788 *
789 * object + s->size
672bba3a 790 * Nothing is used beyond s->size.
81819f0f 791 *
3b0efdfa 792 * If slabcaches are merged then the object_size and inuse boundaries are mostly
672bba3a 793 * ignored. And therefore no slab options that rely on these boundaries
81819f0f
CL
794 * may be used with merged slabcaches.
795 */
796
81819f0f
CL
797static 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
80a9201a
AP
809 off += kasan_metadata_size(s);
810
d86bd1be 811 if (size_from_object(s) == off)
81819f0f
CL
812 return 1;
813
24922684 814 return check_bytes_and_report(s, page, p, "Object padding",
d86bd1be 815 p + off, POISON_INUSE, size_from_object(s) - off);
81819f0f
CL
816}
817
39b26464 818/* Check the pad bytes at the end of a slab page */
81819f0f
CL
819static int slab_pad_check(struct kmem_cache *s, struct page *page)
820{
24922684
CL
821 u8 *start;
822 u8 *fault;
823 u8 *end;
5d682681 824 u8 *pad;
24922684
CL
825 int length;
826 int remainder;
81819f0f
CL
827
828 if (!(s->flags & SLAB_POISON))
829 return 1;
830
a973e9dd 831 start = page_address(page);
9736d2a9 832 length = PAGE_SIZE << compound_order(page);
39b26464
CL
833 end = start + length;
834 remainder = length % s->size;
81819f0f
CL
835 if (!remainder)
836 return 1;
837
5d682681 838 pad = end - remainder;
a79316c6 839 metadata_access_enable();
5d682681 840 fault = memchr_inv(pad, POISON_INUSE, remainder);
a79316c6 841 metadata_access_disable();
24922684
CL
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);
5d682681 848 print_section(KERN_ERR, "Padding ", pad, remainder);
24922684 849
5d682681 850 restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
24922684 851 return 0;
81819f0f
CL
852}
853
854static int check_object(struct kmem_cache *s, struct page *page,
f7cb1933 855 void *object, u8 val)
81819f0f
CL
856{
857 u8 *p = object;
3b0efdfa 858 u8 *endobject = object + s->object_size;
81819f0f
CL
859
860 if (s->flags & SLAB_RED_ZONE) {
d86bd1be
JK
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
24922684 865 if (!check_bytes_and_report(s, page, object, "Redzone",
3b0efdfa 866 endobject, val, s->inuse - s->object_size))
81819f0f 867 return 0;
81819f0f 868 } else {
3b0efdfa 869 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
3adbefee 870 check_bytes_and_report(s, page, p, "Alignment padding",
d0e0ac97
CG
871 endobject, POISON_INUSE,
872 s->inuse - s->object_size);
3adbefee 873 }
81819f0f
CL
874 }
875
876 if (s->flags & SLAB_POISON) {
f7cb1933 877 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
24922684 878 (!check_bytes_and_report(s, page, p, "Poison", p,
3b0efdfa 879 POISON_FREE, s->object_size - 1) ||
24922684 880 !check_bytes_and_report(s, page, p, "Poison",
3b0efdfa 881 p + s->object_size - 1, POISON_END, 1)))
81819f0f 882 return 0;
81819f0f
CL
883 /*
884 * check_pad_bytes cleans up on its own.
885 */
886 check_pad_bytes(s, page, p);
887 }
888
f7cb1933 889 if (!s->offset && val == SLUB_RED_ACTIVE)
81819f0f
CL
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 /*
9f6c708e 900 * No choice but to zap it and thus lose the remainder
81819f0f 901 * of the free objects in this slab. May cause
672bba3a 902 * another error because the object count is now wrong.
81819f0f 903 */
a973e9dd 904 set_freepointer(s, p, NULL);
81819f0f
CL
905 return 0;
906 }
907 return 1;
908}
909
910static int check_slab(struct kmem_cache *s, struct page *page)
911{
39b26464
CL
912 int maxobj;
913
81819f0f
CL
914 VM_BUG_ON(!irqs_disabled());
915
916 if (!PageSlab(page)) {
24922684 917 slab_err(s, page, "Not a valid slab page");
81819f0f
CL
918 return 0;
919 }
39b26464 920
9736d2a9 921 maxobj = order_objects(compound_order(page), s->size);
39b26464
CL
922 if (page->objects > maxobj) {
923 slab_err(s, page, "objects %u > max %u",
f6edde9c 924 page->objects, maxobj);
39b26464
CL
925 return 0;
926 }
927 if (page->inuse > page->objects) {
24922684 928 slab_err(s, page, "inuse %u > max %u",
f6edde9c 929 page->inuse, page->objects);
81819f0f
CL
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/*
672bba3a
CL
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.
81819f0f
CL
940 */
941static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
942{
943 int nr = 0;
881db7fb 944 void *fp;
81819f0f 945 void *object = NULL;
f6edde9c 946 int max_objects;
81819f0f 947
881db7fb 948 fp = page->freelist;
39b26464 949 while (fp && nr <= page->objects) {
81819f0f
CL
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");
a973e9dd 956 set_freepointer(s, object, NULL);
81819f0f 957 } else {
24922684 958 slab_err(s, page, "Freepointer corrupt");
a973e9dd 959 page->freelist = NULL;
39b26464 960 page->inuse = page->objects;
24922684 961 slab_fix(s, "Freelist cleared");
81819f0f
CL
962 return 0;
963 }
964 break;
965 }
966 object = fp;
967 fp = get_freepointer(s, object);
968 nr++;
969 }
970
9736d2a9 971 max_objects = order_objects(compound_order(page), s->size);
210b5c06
CG
972 if (max_objects > MAX_OBJS_PER_PAGE)
973 max_objects = MAX_OBJS_PER_PAGE;
224a88be
CL
974
975 if (page->objects != max_objects) {
756a025f
JP
976 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
977 page->objects, max_objects);
224a88be
CL
978 page->objects = max_objects;
979 slab_fix(s, "Number of objects adjusted.");
980 }
39b26464 981 if (page->inuse != page->objects - nr) {
756a025f
JP
982 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
983 page->inuse, page->objects - nr);
39b26464 984 page->inuse = page->objects - nr;
24922684 985 slab_fix(s, "Object count adjusted.");
81819f0f
CL
986 }
987 return search == NULL;
988}
989
0121c619
CL
990static void trace(struct kmem_cache *s, struct page *page, void *object,
991 int alloc)
3ec09742
CL
992{
993 if (s->flags & SLAB_TRACE) {
f9f58285 994 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
3ec09742
CL
995 s->name,
996 alloc ? "alloc" : "free",
997 object, page->inuse,
998 page->freelist);
999
1000 if (!alloc)
aa2efd5e 1001 print_section(KERN_INFO, "Object ", (void *)object,
d0e0ac97 1002 s->object_size);
3ec09742
CL
1003
1004 dump_stack();
1005 }
1006}
1007
643b1138 1008/*
672bba3a 1009 * Tracking of fully allocated slabs for debugging purposes.
643b1138 1010 */
5cc6eee8
CL
1011static void add_full(struct kmem_cache *s,
1012 struct kmem_cache_node *n, struct page *page)
643b1138 1013{
5cc6eee8
CL
1014 if (!(s->flags & SLAB_STORE_USER))
1015 return;
1016
255d0884 1017 lockdep_assert_held(&n->list_lock);
916ac052 1018 list_add(&page->slab_list, &n->full);
643b1138
CL
1019}
1020
c65c1877 1021static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
643b1138 1022{
643b1138
CL
1023 if (!(s->flags & SLAB_STORE_USER))
1024 return;
1025
255d0884 1026 lockdep_assert_held(&n->list_lock);
916ac052 1027 list_del(&page->slab_list);
643b1138
CL
1028}
1029
0f389ec6
CL
1030/* Tracking of the number of slabs for debugging purposes */
1031static 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
26c02cf0
AB
1038static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1039{
1040 return atomic_long_read(&n->nr_slabs);
1041}
1042
205ab99d 1043static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
0f389ec6
CL
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 */
338b2642 1053 if (likely(n)) {
0f389ec6 1054 atomic_long_inc(&n->nr_slabs);
205ab99d
CL
1055 atomic_long_add(objects, &n->total_objects);
1056 }
0f389ec6 1057}
205ab99d 1058static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
0f389ec6
CL
1059{
1060 struct kmem_cache_node *n = get_node(s, node);
1061
1062 atomic_long_dec(&n->nr_slabs);
205ab99d 1063 atomic_long_sub(objects, &n->total_objects);
0f389ec6
CL
1064}
1065
1066/* Object debug checks for alloc/free paths */
3ec09742
CL
1067static 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
f7cb1933 1073 init_object(s, object, SLUB_RED_INACTIVE);
3ec09742
CL
1074 init_tracking(s, object);
1075}
1076
a7101224
AK
1077static void setup_page_debug(struct kmem_cache *s, void *addr, int order)
1078{
1079 if (!(s->flags & SLAB_POISON))
1080 return;
1081
1082 metadata_access_enable();
1083 memset(addr, POISON_INUSE, PAGE_SIZE << order);
1084 metadata_access_disable();
1085}
1086
becfda68 1087static inline int alloc_consistency_checks(struct kmem_cache *s,
278d7756 1088 struct page *page, void *object)
81819f0f
CL
1089{
1090 if (!check_slab(s, page))
becfda68 1091 return 0;
81819f0f 1092
81819f0f
CL
1093 if (!check_valid_pointer(s, page, object)) {
1094 object_err(s, page, object, "Freelist Pointer check fails");
becfda68 1095 return 0;
81819f0f
CL
1096 }
1097
f7cb1933 1098 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
becfda68
LA
1099 return 0;
1100
1101 return 1;
1102}
1103
1104static noinline int alloc_debug_processing(struct kmem_cache *s,
1105 struct page *page,
1106 void *object, unsigned long addr)
1107{
1108 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
278d7756 1109 if (!alloc_consistency_checks(s, page, object))
becfda68
LA
1110 goto bad;
1111 }
81819f0f 1112
3ec09742
CL
1113 /* Success perform special debug activities for allocs */
1114 if (s->flags & SLAB_STORE_USER)
1115 set_track(s, object, TRACK_ALLOC, addr);
1116 trace(s, page, object, 1);
f7cb1933 1117 init_object(s, object, SLUB_RED_ACTIVE);
81819f0f 1118 return 1;
3ec09742 1119
81819f0f
CL
1120bad:
1121 if (PageSlab(page)) {
1122 /*
1123 * If this is a slab page then lets do the best we can
1124 * to avoid issues in the future. Marking all objects
672bba3a 1125 * as used avoids touching the remaining objects.
81819f0f 1126 */
24922684 1127 slab_fix(s, "Marking all objects used");
39b26464 1128 page->inuse = page->objects;
a973e9dd 1129 page->freelist = NULL;
81819f0f
CL
1130 }
1131 return 0;
1132}
1133
becfda68
LA
1134static inline int free_consistency_checks(struct kmem_cache *s,
1135 struct page *page, void *object, unsigned long addr)
81819f0f 1136{
81819f0f 1137 if (!check_valid_pointer(s, page, object)) {
70d71228 1138 slab_err(s, page, "Invalid object pointer 0x%p", object);
becfda68 1139 return 0;
81819f0f
CL
1140 }
1141
1142 if (on_freelist(s, page, object)) {
24922684 1143 object_err(s, page, object, "Object already free");
becfda68 1144 return 0;
81819f0f
CL
1145 }
1146
f7cb1933 1147 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
becfda68 1148 return 0;
81819f0f 1149
1b4f59e3 1150 if (unlikely(s != page->slab_cache)) {
3adbefee 1151 if (!PageSlab(page)) {
756a025f
JP
1152 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1153 object);
1b4f59e3 1154 } else if (!page->slab_cache) {
f9f58285
FF
1155 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1156 object);
70d71228 1157 dump_stack();
06428780 1158 } else
24922684
CL
1159 object_err(s, page, object,
1160 "page slab pointer corrupt.");
becfda68
LA
1161 return 0;
1162 }
1163 return 1;
1164}
1165
1166/* Supports checking bulk free of a constructed freelist */
1167static noinline int free_debug_processing(
1168 struct kmem_cache *s, struct page *page,
1169 void *head, void *tail, int bulk_cnt,
1170 unsigned long addr)
1171{
1172 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1173 void *object = head;
1174 int cnt = 0;
1175 unsigned long uninitialized_var(flags);
1176 int ret = 0;
1177
1178 spin_lock_irqsave(&n->list_lock, flags);
1179 slab_lock(page);
1180
1181 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1182 if (!check_slab(s, page))
1183 goto out;
1184 }
1185
1186next_object:
1187 cnt++;
1188
1189 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1190 if (!free_consistency_checks(s, page, object, addr))
1191 goto out;
81819f0f 1192 }
3ec09742 1193
3ec09742
CL
1194 if (s->flags & SLAB_STORE_USER)
1195 set_track(s, object, TRACK_FREE, addr);
1196 trace(s, page, object, 0);
81084651 1197 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
f7cb1933 1198 init_object(s, object, SLUB_RED_INACTIVE);
81084651
JDB
1199
1200 /* Reached end of constructed freelist yet? */
1201 if (object != tail) {
1202 object = get_freepointer(s, object);
1203 goto next_object;
1204 }
804aa132
LA
1205 ret = 1;
1206
5c2e4bbb 1207out:
81084651
JDB
1208 if (cnt != bulk_cnt)
1209 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1210 bulk_cnt, cnt);
1211
881db7fb 1212 slab_unlock(page);
282acb43 1213 spin_unlock_irqrestore(&n->list_lock, flags);
804aa132
LA
1214 if (!ret)
1215 slab_fix(s, "Object at 0x%p not freed", object);
1216 return ret;
81819f0f
CL
1217}
1218
41ecc55b
CL
1219static int __init setup_slub_debug(char *str)
1220{
f0630fff
CL
1221 slub_debug = DEBUG_DEFAULT_FLAGS;
1222 if (*str++ != '=' || !*str)
1223 /*
1224 * No options specified. Switch on full debugging.
1225 */
1226 goto out;
1227
1228 if (*str == ',')
1229 /*
1230 * No options but restriction on slabs. This means full
1231 * debugging for slabs matching a pattern.
1232 */
1233 goto check_slabs;
1234
1235 slub_debug = 0;
1236 if (*str == '-')
1237 /*
1238 * Switch off all debugging measures.
1239 */
1240 goto out;
1241
1242 /*
1243 * Determine which debug features should be switched on
1244 */
06428780 1245 for (; *str && *str != ','; str++) {
f0630fff
CL
1246 switch (tolower(*str)) {
1247 case 'f':
becfda68 1248 slub_debug |= SLAB_CONSISTENCY_CHECKS;
f0630fff
CL
1249 break;
1250 case 'z':
1251 slub_debug |= SLAB_RED_ZONE;
1252 break;
1253 case 'p':
1254 slub_debug |= SLAB_POISON;
1255 break;
1256 case 'u':
1257 slub_debug |= SLAB_STORE_USER;
1258 break;
1259 case 't':
1260 slub_debug |= SLAB_TRACE;
1261 break;
4c13dd3b
DM
1262 case 'a':
1263 slub_debug |= SLAB_FAILSLAB;
1264 break;
08303a73
CA
1265 case 'o':
1266 /*
1267 * Avoid enabling debugging on caches if its minimum
1268 * order would increase as a result.
1269 */
1270 disable_higher_order_debug = 1;
1271 break;
f0630fff 1272 default:
f9f58285
FF
1273 pr_err("slub_debug option '%c' unknown. skipped\n",
1274 *str);
f0630fff 1275 }
41ecc55b
CL
1276 }
1277
f0630fff 1278check_slabs:
41ecc55b
CL
1279 if (*str == ',')
1280 slub_debug_slabs = str + 1;
f0630fff 1281out:
6471384a
AP
1282 if ((static_branch_unlikely(&init_on_alloc) ||
1283 static_branch_unlikely(&init_on_free)) &&
1284 (slub_debug & SLAB_POISON))
1285 pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
41ecc55b
CL
1286 return 1;
1287}
1288
1289__setup("slub_debug", setup_slub_debug);
1290
c5fd3ca0
AT
1291/*
1292 * kmem_cache_flags - apply debugging options to the cache
1293 * @object_size: the size of an object without meta data
1294 * @flags: flags to set
1295 * @name: name of the cache
1296 * @ctor: constructor function
1297 *
1298 * Debug option(s) are applied to @flags. In addition to the debug
1299 * option(s), if a slab name (or multiple) is specified i.e.
1300 * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1301 * then only the select slabs will receive the debug option(s).
1302 */
0293d1fd 1303slab_flags_t kmem_cache_flags(unsigned int object_size,
d50112ed 1304 slab_flags_t flags, const char *name,
51cc5068 1305 void (*ctor)(void *))
41ecc55b 1306{
c5fd3ca0
AT
1307 char *iter;
1308 size_t len;
1309
1310 /* If slub_debug = 0, it folds into the if conditional. */
1311 if (!slub_debug_slabs)
1312 return flags | slub_debug;
1313
1314 len = strlen(name);
1315 iter = slub_debug_slabs;
1316 while (*iter) {
1317 char *end, *glob;
1318 size_t cmplen;
1319
9cf3a8d8 1320 end = strchrnul(iter, ',');
c5fd3ca0
AT
1321
1322 glob = strnchr(iter, end - iter, '*');
1323 if (glob)
1324 cmplen = glob - iter;
1325 else
1326 cmplen = max_t(size_t, len, (end - iter));
1327
1328 if (!strncmp(name, iter, cmplen)) {
1329 flags |= slub_debug;
1330 break;
1331 }
1332
1333 if (!*end)
1334 break;
1335 iter = end + 1;
1336 }
ba0268a8
CL
1337
1338 return flags;
41ecc55b 1339}
b4a64718 1340#else /* !CONFIG_SLUB_DEBUG */
3ec09742
CL
1341static inline void setup_object_debug(struct kmem_cache *s,
1342 struct page *page, void *object) {}
a7101224
AK
1343static inline void setup_page_debug(struct kmem_cache *s,
1344 void *addr, int order) {}
41ecc55b 1345
3ec09742 1346static inline int alloc_debug_processing(struct kmem_cache *s,
ce71e27c 1347 struct page *page, void *object, unsigned long addr) { return 0; }
41ecc55b 1348
282acb43 1349static inline int free_debug_processing(
81084651
JDB
1350 struct kmem_cache *s, struct page *page,
1351 void *head, void *tail, int bulk_cnt,
282acb43 1352 unsigned long addr) { return 0; }
41ecc55b 1353
41ecc55b
CL
1354static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1355 { return 1; }
1356static inline int check_object(struct kmem_cache *s, struct page *page,
f7cb1933 1357 void *object, u8 val) { return 1; }
5cc6eee8
CL
1358static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1359 struct page *page) {}
c65c1877
PZ
1360static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1361 struct page *page) {}
0293d1fd 1362slab_flags_t kmem_cache_flags(unsigned int object_size,
d50112ed 1363 slab_flags_t flags, const char *name,
51cc5068 1364 void (*ctor)(void *))
ba0268a8
CL
1365{
1366 return flags;
1367}
41ecc55b 1368#define slub_debug 0
0f389ec6 1369
fdaa45e9
IM
1370#define disable_higher_order_debug 0
1371
0f389ec6
CL
1372static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1373 { return 0; }
26c02cf0
AB
1374static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1375 { return 0; }
205ab99d
CL
1376static inline void inc_slabs_node(struct kmem_cache *s, int node,
1377 int objects) {}
1378static inline void dec_slabs_node(struct kmem_cache *s, int node,
1379 int objects) {}
7d550c56 1380
02e72cc6
AR
1381#endif /* CONFIG_SLUB_DEBUG */
1382
1383/*
1384 * Hooks for other subsystems that check memory allocations. In a typical
1385 * production configuration these hooks all should produce no code at all.
1386 */
0116523c 1387static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
d56791b3 1388{
53128245 1389 ptr = kasan_kmalloc_large(ptr, size, flags);
a2f77575 1390 /* As ptr might get tagged, call kmemleak hook after KASAN. */
d56791b3 1391 kmemleak_alloc(ptr, size, 1, flags);
53128245 1392 return ptr;
d56791b3
RB
1393}
1394
ee3ce779 1395static __always_inline void kfree_hook(void *x)
d56791b3
RB
1396{
1397 kmemleak_free(x);
ee3ce779 1398 kasan_kfree_large(x, _RET_IP_);
d56791b3
RB
1399}
1400
c3895391 1401static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
d56791b3
RB
1402{
1403 kmemleak_free_recursive(x, s->flags);
7d550c56 1404
02e72cc6
AR
1405 /*
1406 * Trouble is that we may no longer disable interrupts in the fast path
1407 * So in order to make the debug calls that expect irqs to be
1408 * disabled we need to disable interrupts temporarily.
1409 */
4675ff05 1410#ifdef CONFIG_LOCKDEP
02e72cc6
AR
1411 {
1412 unsigned long flags;
1413
1414 local_irq_save(flags);
02e72cc6
AR
1415 debug_check_no_locks_freed(x, s->object_size);
1416 local_irq_restore(flags);
1417 }
1418#endif
1419 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1420 debug_check_no_obj_freed(x, s->object_size);
0316bec2 1421
c3895391
AK
1422 /* KASAN might put x into memory quarantine, delaying its reuse */
1423 return kasan_slab_free(s, x, _RET_IP_);
02e72cc6 1424}
205ab99d 1425
c3895391
AK
1426static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1427 void **head, void **tail)
81084651 1428{
6471384a
AP
1429
1430 void *object;
1431 void *next = *head;
1432 void *old_tail = *tail ? *tail : *head;
1433 int rsize;
1434
1b7e816f
LA
1435 if (slab_want_init_on_free(s)) {
1436 void *p = NULL;
1437
6471384a
AP
1438 do {
1439 object = next;
1440 next = get_freepointer(s, object);
1441 /*
1442 * Clear the object and the metadata, but don't touch
1443 * the redzone.
1444 */
1445 memset(object, 0, s->object_size);
1446 rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad
1447 : 0;
1448 memset((char *)object + s->inuse, 0,
1449 s->size - s->inuse - rsize);
1b7e816f
LA
1450 set_freepointer(s, object, p);
1451 p = object;
6471384a 1452 } while (object != old_tail);
1b7e816f 1453 }
6471384a 1454
81084651
JDB
1455/*
1456 * Compiler cannot detect this function can be removed if slab_free_hook()
1457 * evaluates to nothing. Thus, catch all relevant config debug options here.
1458 */
4675ff05 1459#if defined(CONFIG_LOCKDEP) || \
81084651
JDB
1460 defined(CONFIG_DEBUG_KMEMLEAK) || \
1461 defined(CONFIG_DEBUG_OBJECTS_FREE) || \
1462 defined(CONFIG_KASAN)
1463
6471384a 1464 next = *head;
c3895391
AK
1465
1466 /* Head and tail of the reconstructed freelist */
1467 *head = NULL;
1468 *tail = NULL;
81084651
JDB
1469
1470 do {
c3895391
AK
1471 object = next;
1472 next = get_freepointer(s, object);
1473 /* If object's reuse doesn't have to be delayed */
1474 if (!slab_free_hook(s, object)) {
1475 /* Move object to the new freelist */
1476 set_freepointer(s, object, *head);
1477 *head = object;
1478 if (!*tail)
1479 *tail = object;
1480 }
1481 } while (object != old_tail);
1482
1483 if (*head == *tail)
1484 *tail = NULL;
1485
1486 return *head != NULL;
1487#else
1488 return true;
81084651
JDB
1489#endif
1490}
1491
4d176711 1492static void *setup_object(struct kmem_cache *s, struct page *page,
588f8ba9
TG
1493 void *object)
1494{
1495 setup_object_debug(s, page, object);
4d176711 1496 object = kasan_init_slab_obj(s, object);
588f8ba9
TG
1497 if (unlikely(s->ctor)) {
1498 kasan_unpoison_object_data(s, object);
1499 s->ctor(object);
1500 kasan_poison_object_data(s, object);
1501 }
4d176711 1502 return object;
588f8ba9
TG
1503}
1504
81819f0f
CL
1505/*
1506 * Slab allocation and freeing
1507 */
5dfb4175
VD
1508static inline struct page *alloc_slab_page(struct kmem_cache *s,
1509 gfp_t flags, int node, struct kmem_cache_order_objects oo)
65c3376a 1510{
5dfb4175 1511 struct page *page;
19af27af 1512 unsigned int order = oo_order(oo);
65c3376a 1513
2154a336 1514 if (node == NUMA_NO_NODE)
5dfb4175 1515 page = alloc_pages(flags, order);
65c3376a 1516 else
96db800f 1517 page = __alloc_pages_node(node, flags, order);
5dfb4175 1518
6cea1d56 1519 if (page && charge_slab_page(page, flags, order, s)) {
f3ccb2c4
VD
1520 __free_pages(page, order);
1521 page = NULL;
1522 }
5dfb4175
VD
1523
1524 return page;
65c3376a
CL
1525}
1526
210e7a43
TG
1527#ifdef CONFIG_SLAB_FREELIST_RANDOM
1528/* Pre-initialize the random sequence cache */
1529static int init_cache_random_seq(struct kmem_cache *s)
1530{
19af27af 1531 unsigned int count = oo_objects(s->oo);
210e7a43 1532 int err;
210e7a43 1533
a810007a
SR
1534 /* Bailout if already initialised */
1535 if (s->random_seq)
1536 return 0;
1537
210e7a43
TG
1538 err = cache_random_seq_create(s, count, GFP_KERNEL);
1539 if (err) {
1540 pr_err("SLUB: Unable to initialize free list for %s\n",
1541 s->name);
1542 return err;
1543 }
1544
1545 /* Transform to an offset on the set of pages */
1546 if (s->random_seq) {
19af27af
AD
1547 unsigned int i;
1548
210e7a43
TG
1549 for (i = 0; i < count; i++)
1550 s->random_seq[i] *= s->size;
1551 }
1552 return 0;
1553}
1554
1555/* Initialize each random sequence freelist per cache */
1556static void __init init_freelist_randomization(void)
1557{
1558 struct kmem_cache *s;
1559
1560 mutex_lock(&slab_mutex);
1561
1562 list_for_each_entry(s, &slab_caches, list)
1563 init_cache_random_seq(s);
1564
1565 mutex_unlock(&slab_mutex);
1566}
1567
1568/* Get the next entry on the pre-computed freelist randomized */
1569static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1570 unsigned long *pos, void *start,
1571 unsigned long page_limit,
1572 unsigned long freelist_count)
1573{
1574 unsigned int idx;
1575
1576 /*
1577 * If the target page allocation failed, the number of objects on the
1578 * page might be smaller than the usual size defined by the cache.
1579 */
1580 do {
1581 idx = s->random_seq[*pos];
1582 *pos += 1;
1583 if (*pos >= freelist_count)
1584 *pos = 0;
1585 } while (unlikely(idx >= page_limit));
1586
1587 return (char *)start + idx;
1588}
1589
1590/* Shuffle the single linked freelist based on a random pre-computed sequence */
1591static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1592{
1593 void *start;
1594 void *cur;
1595 void *next;
1596 unsigned long idx, pos, page_limit, freelist_count;
1597
1598 if (page->objects < 2 || !s->random_seq)
1599 return false;
1600
1601 freelist_count = oo_objects(s->oo);
1602 pos = get_random_int() % freelist_count;
1603
1604 page_limit = page->objects * s->size;
1605 start = fixup_red_left(s, page_address(page));
1606
1607 /* First entry is used as the base of the freelist */
1608 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1609 freelist_count);
4d176711 1610 cur = setup_object(s, page, cur);
210e7a43
TG
1611 page->freelist = cur;
1612
1613 for (idx = 1; idx < page->objects; idx++) {
210e7a43
TG
1614 next = next_freelist_entry(s, page, &pos, start, page_limit,
1615 freelist_count);
4d176711 1616 next = setup_object(s, page, next);
210e7a43
TG
1617 set_freepointer(s, cur, next);
1618 cur = next;
1619 }
210e7a43
TG
1620 set_freepointer(s, cur, NULL);
1621
1622 return true;
1623}
1624#else
1625static inline int init_cache_random_seq(struct kmem_cache *s)
1626{
1627 return 0;
1628}
1629static inline void init_freelist_randomization(void) { }
1630static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1631{
1632 return false;
1633}
1634#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1635
81819f0f
CL
1636static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1637{
06428780 1638 struct page *page;
834f3d11 1639 struct kmem_cache_order_objects oo = s->oo;
ba52270d 1640 gfp_t alloc_gfp;
4d176711 1641 void *start, *p, *next;
588f8ba9 1642 int idx, order;
210e7a43 1643 bool shuffle;
81819f0f 1644
7e0528da
CL
1645 flags &= gfp_allowed_mask;
1646
d0164adc 1647 if (gfpflags_allow_blocking(flags))
7e0528da
CL
1648 local_irq_enable();
1649
b7a49f0d 1650 flags |= s->allocflags;
e12ba74d 1651
ba52270d
PE
1652 /*
1653 * Let the initial higher-order allocation fail under memory pressure
1654 * so we fall-back to the minimum order allocation.
1655 */
1656 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
d0164adc 1657 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
444eb2a4 1658 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
ba52270d 1659
5dfb4175 1660 page = alloc_slab_page(s, alloc_gfp, node, oo);
65c3376a
CL
1661 if (unlikely(!page)) {
1662 oo = s->min;
80c3a998 1663 alloc_gfp = flags;
65c3376a
CL
1664 /*
1665 * Allocation may have failed due to fragmentation.
1666 * Try a lower order alloc if possible
1667 */
5dfb4175 1668 page = alloc_slab_page(s, alloc_gfp, node, oo);
588f8ba9
TG
1669 if (unlikely(!page))
1670 goto out;
1671 stat(s, ORDER_FALLBACK);
65c3376a 1672 }
5a896d9e 1673
834f3d11 1674 page->objects = oo_objects(oo);
81819f0f 1675
1f458cbf 1676 order = compound_order(page);
1b4f59e3 1677 page->slab_cache = s;
c03f94cc 1678 __SetPageSlab(page);
2f064f34 1679 if (page_is_pfmemalloc(page))
072bb0aa 1680 SetPageSlabPfmemalloc(page);
81819f0f 1681
a7101224 1682 kasan_poison_slab(page);
81819f0f 1683
a7101224 1684 start = page_address(page);
81819f0f 1685
a7101224 1686 setup_page_debug(s, start, order);
0316bec2 1687
210e7a43
TG
1688 shuffle = shuffle_freelist(s, page);
1689
1690 if (!shuffle) {
4d176711
AK
1691 start = fixup_red_left(s, start);
1692 start = setup_object(s, page, start);
1693 page->freelist = start;
18e50661
AK
1694 for (idx = 0, p = start; idx < page->objects - 1; idx++) {
1695 next = p + s->size;
1696 next = setup_object(s, page, next);
1697 set_freepointer(s, p, next);
1698 p = next;
1699 }
1700 set_freepointer(s, p, NULL);
81819f0f 1701 }
81819f0f 1702
e6e82ea1 1703 page->inuse = page->objects;
8cb0a506 1704 page->frozen = 1;
588f8ba9 1705
81819f0f 1706out:
d0164adc 1707 if (gfpflags_allow_blocking(flags))
588f8ba9
TG
1708 local_irq_disable();
1709 if (!page)
1710 return NULL;
1711
588f8ba9
TG
1712 inc_slabs_node(s, page_to_nid(page), page->objects);
1713
81819f0f
CL
1714 return page;
1715}
1716
588f8ba9
TG
1717static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1718{
1719 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
bacdcb34 1720 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
72baeef0
MH
1721 flags &= ~GFP_SLAB_BUG_MASK;
1722 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1723 invalid_mask, &invalid_mask, flags, &flags);
65b9de75 1724 dump_stack();
588f8ba9
TG
1725 }
1726
1727 return allocate_slab(s,
1728 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1729}
1730
81819f0f
CL
1731static void __free_slab(struct kmem_cache *s, struct page *page)
1732{
834f3d11
CL
1733 int order = compound_order(page);
1734 int pages = 1 << order;
81819f0f 1735
becfda68 1736 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
81819f0f
CL
1737 void *p;
1738
1739 slab_pad_check(s, page);
224a88be
CL
1740 for_each_object(p, s, page_address(page),
1741 page->objects)
f7cb1933 1742 check_object(s, page, p, SLUB_RED_INACTIVE);
81819f0f
CL
1743 }
1744
072bb0aa 1745 __ClearPageSlabPfmemalloc(page);
49bd5221 1746 __ClearPageSlab(page);
1f458cbf 1747
d4fc5069 1748 page->mapping = NULL;
1eb5ac64
NP
1749 if (current->reclaim_state)
1750 current->reclaim_state->reclaimed_slab += pages;
6cea1d56 1751 uncharge_slab_page(page, order, s);
27ee57c9 1752 __free_pages(page, order);
81819f0f
CL
1753}
1754
1755static void rcu_free_slab(struct rcu_head *h)
1756{
bf68c214 1757 struct page *page = container_of(h, struct page, rcu_head);
da9a638c 1758
1b4f59e3 1759 __free_slab(page->slab_cache, page);
81819f0f
CL
1760}
1761
1762static void free_slab(struct kmem_cache *s, struct page *page)
1763{
5f0d5a3a 1764 if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
bf68c214 1765 call_rcu(&page->rcu_head, rcu_free_slab);
81819f0f
CL
1766 } else
1767 __free_slab(s, page);
1768}
1769
1770static void discard_slab(struct kmem_cache *s, struct page *page)
1771{
205ab99d 1772 dec_slabs_node(s, page_to_nid(page), page->objects);
81819f0f
CL
1773 free_slab(s, page);
1774}
1775
1776/*
5cc6eee8 1777 * Management of partially allocated slabs.
81819f0f 1778 */
1e4dd946
SR
1779static inline void
1780__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
81819f0f 1781{
e95eed57 1782 n->nr_partial++;
136333d1 1783 if (tail == DEACTIVATE_TO_TAIL)
916ac052 1784 list_add_tail(&page->slab_list, &n->partial);
7c2e132c 1785 else
916ac052 1786 list_add(&page->slab_list, &n->partial);
81819f0f
CL
1787}
1788
1e4dd946
SR
1789static inline void add_partial(struct kmem_cache_node *n,
1790 struct page *page, int tail)
62e346a8 1791{
c65c1877 1792 lockdep_assert_held(&n->list_lock);
1e4dd946
SR
1793 __add_partial(n, page, tail);
1794}
c65c1877 1795
1e4dd946
SR
1796static inline void remove_partial(struct kmem_cache_node *n,
1797 struct page *page)
1798{
1799 lockdep_assert_held(&n->list_lock);
916ac052 1800 list_del(&page->slab_list);
52b4b950 1801 n->nr_partial--;
1e4dd946
SR
1802}
1803
81819f0f 1804/*
7ced3719
CL
1805 * Remove slab from the partial list, freeze it and
1806 * return the pointer to the freelist.
81819f0f 1807 *
497b66f2 1808 * Returns a list of objects or NULL if it fails.
81819f0f 1809 */
497b66f2 1810static inline void *acquire_slab(struct kmem_cache *s,
acd19fd1 1811 struct kmem_cache_node *n, struct page *page,
633b0764 1812 int mode, int *objects)
81819f0f 1813{
2cfb7455
CL
1814 void *freelist;
1815 unsigned long counters;
1816 struct page new;
1817
c65c1877
PZ
1818 lockdep_assert_held(&n->list_lock);
1819
2cfb7455
CL
1820 /*
1821 * Zap the freelist and set the frozen bit.
1822 * The old freelist is the list of objects for the
1823 * per cpu allocation list.
1824 */
7ced3719
CL
1825 freelist = page->freelist;
1826 counters = page->counters;
1827 new.counters = counters;
633b0764 1828 *objects = new.objects - new.inuse;
23910c50 1829 if (mode) {
7ced3719 1830 new.inuse = page->objects;
23910c50
PE
1831 new.freelist = NULL;
1832 } else {
1833 new.freelist = freelist;
1834 }
2cfb7455 1835
a0132ac0 1836 VM_BUG_ON(new.frozen);
7ced3719 1837 new.frozen = 1;
2cfb7455 1838
7ced3719 1839 if (!__cmpxchg_double_slab(s, page,
2cfb7455 1840 freelist, counters,
02d7633f 1841 new.freelist, new.counters,
7ced3719 1842 "acquire_slab"))
7ced3719 1843 return NULL;
2cfb7455
CL
1844
1845 remove_partial(n, page);
7ced3719 1846 WARN_ON(!freelist);
49e22585 1847 return freelist;
81819f0f
CL
1848}
1849
633b0764 1850static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
8ba00bb6 1851static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
49e22585 1852
81819f0f 1853/*
672bba3a 1854 * Try to allocate a partial slab from a specific node.
81819f0f 1855 */
8ba00bb6
JK
1856static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1857 struct kmem_cache_cpu *c, gfp_t flags)
81819f0f 1858{
49e22585
CL
1859 struct page *page, *page2;
1860 void *object = NULL;
e5d9998f 1861 unsigned int available = 0;
633b0764 1862 int objects;
81819f0f
CL
1863
1864 /*
1865 * Racy check. If we mistakenly see no partial slabs then we
1866 * just allocate an empty slab. If we mistakenly try to get a
672bba3a
CL
1867 * partial slab and there is none available then get_partials()
1868 * will return NULL.
81819f0f
CL
1869 */
1870 if (!n || !n->nr_partial)
1871 return NULL;
1872
1873 spin_lock(&n->list_lock);
916ac052 1874 list_for_each_entry_safe(page, page2, &n->partial, slab_list) {
8ba00bb6 1875 void *t;
49e22585 1876
8ba00bb6
JK
1877 if (!pfmemalloc_match(page, flags))
1878 continue;
1879
633b0764 1880 t = acquire_slab(s, n, page, object == NULL, &objects);
49e22585
CL
1881 if (!t)
1882 break;
1883
633b0764 1884 available += objects;
12d79634 1885 if (!object) {
49e22585 1886 c->page = page;
49e22585 1887 stat(s, ALLOC_FROM_PARTIAL);
49e22585 1888 object = t;
49e22585 1889 } else {
633b0764 1890 put_cpu_partial(s, page, 0);
8028dcea 1891 stat(s, CPU_PARTIAL_NODE);
49e22585 1892 }
345c905d 1893 if (!kmem_cache_has_cpu_partial(s)
e6d0e1dc 1894 || available > slub_cpu_partial(s) / 2)
49e22585
CL
1895 break;
1896
497b66f2 1897 }
81819f0f 1898 spin_unlock(&n->list_lock);
497b66f2 1899 return object;
81819f0f
CL
1900}
1901
1902/*
672bba3a 1903 * Get a page from somewhere. Search in increasing NUMA distances.
81819f0f 1904 */
de3ec035 1905static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
acd19fd1 1906 struct kmem_cache_cpu *c)
81819f0f
CL
1907{
1908#ifdef CONFIG_NUMA
1909 struct zonelist *zonelist;
dd1a239f 1910 struct zoneref *z;
54a6eb5c
MG
1911 struct zone *zone;
1912 enum zone_type high_zoneidx = gfp_zone(flags);
497b66f2 1913 void *object;
cc9a6c87 1914 unsigned int cpuset_mems_cookie;
81819f0f
CL
1915
1916 /*
672bba3a
CL
1917 * The defrag ratio allows a configuration of the tradeoffs between
1918 * inter node defragmentation and node local allocations. A lower
1919 * defrag_ratio increases the tendency to do local allocations
1920 * instead of attempting to obtain partial slabs from other nodes.
81819f0f 1921 *
672bba3a
CL
1922 * If the defrag_ratio is set to 0 then kmalloc() always
1923 * returns node local objects. If the ratio is higher then kmalloc()
1924 * may return off node objects because partial slabs are obtained
1925 * from other nodes and filled up.
81819f0f 1926 *
43efd3ea
LP
1927 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1928 * (which makes defrag_ratio = 1000) then every (well almost)
1929 * allocation will first attempt to defrag slab caches on other nodes.
1930 * This means scanning over all nodes to look for partial slabs which
1931 * may be expensive if we do it every time we are trying to find a slab
672bba3a 1932 * with available objects.
81819f0f 1933 */
9824601e
CL
1934 if (!s->remote_node_defrag_ratio ||
1935 get_cycles() % 1024 > s->remote_node_defrag_ratio)
81819f0f
CL
1936 return NULL;
1937
cc9a6c87 1938 do {
d26914d1 1939 cpuset_mems_cookie = read_mems_allowed_begin();
2a389610 1940 zonelist = node_zonelist(mempolicy_slab_node(), flags);
cc9a6c87
MG
1941 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1942 struct kmem_cache_node *n;
1943
1944 n = get_node(s, zone_to_nid(zone));
1945
dee2f8aa 1946 if (n && cpuset_zone_allowed(zone, flags) &&
cc9a6c87 1947 n->nr_partial > s->min_partial) {
8ba00bb6 1948 object = get_partial_node(s, n, c, flags);
cc9a6c87
MG
1949 if (object) {
1950 /*
d26914d1
MG
1951 * Don't check read_mems_allowed_retry()
1952 * here - if mems_allowed was updated in
1953 * parallel, that was a harmless race
1954 * between allocation and the cpuset
1955 * update
cc9a6c87 1956 */
cc9a6c87
MG
1957 return object;
1958 }
c0ff7453 1959 }
81819f0f 1960 }
d26914d1 1961 } while (read_mems_allowed_retry(cpuset_mems_cookie));
6dfd1b65 1962#endif /* CONFIG_NUMA */
81819f0f
CL
1963 return NULL;
1964}
1965
1966/*
1967 * Get a partial page, lock it and return it.
1968 */
497b66f2 1969static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
acd19fd1 1970 struct kmem_cache_cpu *c)
81819f0f 1971{
497b66f2 1972 void *object;
a561ce00
JK
1973 int searchnode = node;
1974
1975 if (node == NUMA_NO_NODE)
1976 searchnode = numa_mem_id();
1977 else if (!node_present_pages(node))
1978 searchnode = node_to_mem_node(node);
81819f0f 1979
8ba00bb6 1980 object = get_partial_node(s, get_node(s, searchnode), c, flags);
497b66f2
CL
1981 if (object || node != NUMA_NO_NODE)
1982 return object;
81819f0f 1983
acd19fd1 1984 return get_any_partial(s, flags, c);
81819f0f
CL
1985}
1986
8a5ec0ba
CL
1987#ifdef CONFIG_PREEMPT
1988/*
1989 * Calculate the next globally unique transaction for disambiguiation
1990 * during cmpxchg. The transactions start with the cpu number and are then
1991 * incremented by CONFIG_NR_CPUS.
1992 */
1993#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1994#else
1995/*
1996 * No preemption supported therefore also no need to check for
1997 * different cpus.
1998 */
1999#define TID_STEP 1
2000#endif
2001
2002static inline unsigned long next_tid(unsigned long tid)
2003{
2004 return tid + TID_STEP;
2005}
2006
2007static inline unsigned int tid_to_cpu(unsigned long tid)
2008{
2009 return tid % TID_STEP;
2010}
2011
2012static inline unsigned long tid_to_event(unsigned long tid)
2013{
2014 return tid / TID_STEP;
2015}
2016
2017static inline unsigned int init_tid(int cpu)
2018{
2019 return cpu;
2020}
2021
2022static inline void note_cmpxchg_failure(const char *n,
2023 const struct kmem_cache *s, unsigned long tid)
2024{
2025#ifdef SLUB_DEBUG_CMPXCHG
2026 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2027
f9f58285 2028 pr_info("%s %s: cmpxchg redo ", n, s->name);
8a5ec0ba
CL
2029
2030#ifdef CONFIG_PREEMPT
2031 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
f9f58285 2032 pr_warn("due to cpu change %d -> %d\n",
8a5ec0ba
CL
2033 tid_to_cpu(tid), tid_to_cpu(actual_tid));
2034 else
2035#endif
2036 if (tid_to_event(tid) != tid_to_event(actual_tid))
f9f58285 2037 pr_warn("due to cpu running other code. Event %ld->%ld\n",
8a5ec0ba
CL
2038 tid_to_event(tid), tid_to_event(actual_tid));
2039 else
f9f58285 2040 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
8a5ec0ba
CL
2041 actual_tid, tid, next_tid(tid));
2042#endif
4fdccdfb 2043 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
8a5ec0ba
CL
2044}
2045
788e1aad 2046static void init_kmem_cache_cpus(struct kmem_cache *s)
8a5ec0ba 2047{
8a5ec0ba
CL
2048 int cpu;
2049
2050 for_each_possible_cpu(cpu)
2051 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
8a5ec0ba 2052}
2cfb7455 2053
81819f0f
CL
2054/*
2055 * Remove the cpu slab
2056 */
d0e0ac97 2057static void deactivate_slab(struct kmem_cache *s, struct page *page,
d4ff6d35 2058 void *freelist, struct kmem_cache_cpu *c)
81819f0f 2059{
2cfb7455 2060 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
2cfb7455
CL
2061 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2062 int lock = 0;
2063 enum slab_modes l = M_NONE, m = M_NONE;
2cfb7455 2064 void *nextfree;
136333d1 2065 int tail = DEACTIVATE_TO_HEAD;
2cfb7455
CL
2066 struct page new;
2067 struct page old;
2068
2069 if (page->freelist) {
84e554e6 2070 stat(s, DEACTIVATE_REMOTE_FREES);
136333d1 2071 tail = DEACTIVATE_TO_TAIL;
2cfb7455
CL
2072 }
2073
894b8788 2074 /*
2cfb7455
CL
2075 * Stage one: Free all available per cpu objects back
2076 * to the page freelist while it is still frozen. Leave the
2077 * last one.
2078 *
2079 * There is no need to take the list->lock because the page
2080 * is still frozen.
2081 */
2082 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2083 void *prior;
2084 unsigned long counters;
2085
2086 do {
2087 prior = page->freelist;
2088 counters = page->counters;
2089 set_freepointer(s, freelist, prior);
2090 new.counters = counters;
2091 new.inuse--;
a0132ac0 2092 VM_BUG_ON(!new.frozen);
2cfb7455 2093
1d07171c 2094 } while (!__cmpxchg_double_slab(s, page,
2cfb7455
CL
2095 prior, counters,
2096 freelist, new.counters,
2097 "drain percpu freelist"));
2098
2099 freelist = nextfree;
2100 }
2101
894b8788 2102 /*
2cfb7455
CL
2103 * Stage two: Ensure that the page is unfrozen while the
2104 * list presence reflects the actual number of objects
2105 * during unfreeze.
2106 *
2107 * We setup the list membership and then perform a cmpxchg
2108 * with the count. If there is a mismatch then the page
2109 * is not unfrozen but the page is on the wrong list.
2110 *
2111 * Then we restart the process which may have to remove
2112 * the page from the list that we just put it on again
2113 * because the number of objects in the slab may have
2114 * changed.
894b8788 2115 */
2cfb7455 2116redo:
894b8788 2117
2cfb7455
CL
2118 old.freelist = page->freelist;
2119 old.counters = page->counters;
a0132ac0 2120 VM_BUG_ON(!old.frozen);
7c2e132c 2121
2cfb7455
CL
2122 /* Determine target state of the slab */
2123 new.counters = old.counters;
2124 if (freelist) {
2125 new.inuse--;
2126 set_freepointer(s, freelist, old.freelist);
2127 new.freelist = freelist;
2128 } else
2129 new.freelist = old.freelist;
2130
2131 new.frozen = 0;
2132
8a5b20ae 2133 if (!new.inuse && n->nr_partial >= s->min_partial)
2cfb7455
CL
2134 m = M_FREE;
2135 else if (new.freelist) {
2136 m = M_PARTIAL;
2137 if (!lock) {
2138 lock = 1;
2139 /*
8bb4e7a2 2140 * Taking the spinlock removes the possibility
2cfb7455
CL
2141 * that acquire_slab() will see a slab page that
2142 * is frozen
2143 */
2144 spin_lock(&n->list_lock);
2145 }
2146 } else {
2147 m = M_FULL;
2148 if (kmem_cache_debug(s) && !lock) {
2149 lock = 1;
2150 /*
2151 * This also ensures that the scanning of full
2152 * slabs from diagnostic functions will not see
2153 * any frozen slabs.
2154 */
2155 spin_lock(&n->list_lock);
2156 }
2157 }
2158
2159 if (l != m) {
2cfb7455 2160 if (l == M_PARTIAL)
2cfb7455 2161 remove_partial(n, page);
2cfb7455 2162 else if (l == M_FULL)
c65c1877 2163 remove_full(s, n, page);
2cfb7455 2164
88349a28 2165 if (m == M_PARTIAL)
2cfb7455 2166 add_partial(n, page, tail);
88349a28 2167 else if (m == M_FULL)
2cfb7455 2168 add_full(s, n, page);
2cfb7455
CL
2169 }
2170
2171 l = m;
1d07171c 2172 if (!__cmpxchg_double_slab(s, page,
2cfb7455
CL
2173 old.freelist, old.counters,
2174 new.freelist, new.counters,
2175 "unfreezing slab"))
2176 goto redo;
2177
2cfb7455
CL
2178 if (lock)
2179 spin_unlock(&n->list_lock);
2180
88349a28
WY
2181 if (m == M_PARTIAL)
2182 stat(s, tail);
2183 else if (m == M_FULL)
2184 stat(s, DEACTIVATE_FULL);
2185 else if (m == M_FREE) {
2cfb7455
CL
2186 stat(s, DEACTIVATE_EMPTY);
2187 discard_slab(s, page);
2188 stat(s, FREE_SLAB);
894b8788 2189 }
d4ff6d35
WY
2190
2191 c->page = NULL;
2192 c->freelist = NULL;
81819f0f
CL
2193}
2194
d24ac77f
JK
2195/*
2196 * Unfreeze all the cpu partial slabs.
2197 *
59a09917
CL
2198 * This function must be called with interrupts disabled
2199 * for the cpu using c (or some other guarantee must be there
2200 * to guarantee no concurrent accesses).
d24ac77f 2201 */
59a09917
CL
2202static void unfreeze_partials(struct kmem_cache *s,
2203 struct kmem_cache_cpu *c)
49e22585 2204{
345c905d 2205#ifdef CONFIG_SLUB_CPU_PARTIAL
43d77867 2206 struct kmem_cache_node *n = NULL, *n2 = NULL;
9ada1934 2207 struct page *page, *discard_page = NULL;
49e22585
CL
2208
2209 while ((page = c->partial)) {
49e22585
CL
2210 struct page new;
2211 struct page old;
2212
2213 c->partial = page->next;
43d77867
JK
2214
2215 n2 = get_node(s, page_to_nid(page));
2216 if (n != n2) {
2217 if (n)
2218 spin_unlock(&n->list_lock);
2219
2220 n = n2;
2221 spin_lock(&n->list_lock);
2222 }
49e22585
CL
2223
2224 do {
2225
2226 old.freelist = page->freelist;
2227 old.counters = page->counters;
a0132ac0 2228 VM_BUG_ON(!old.frozen);
49e22585
CL
2229
2230 new.counters = old.counters;
2231 new.freelist = old.freelist;
2232
2233 new.frozen = 0;
2234
d24ac77f 2235 } while (!__cmpxchg_double_slab(s, page,
49e22585
CL
2236 old.freelist, old.counters,
2237 new.freelist, new.counters,
2238 "unfreezing slab"));
2239
8a5b20ae 2240 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
9ada1934
SL
2241 page->next = discard_page;
2242 discard_page = page;
43d77867
JK
2243 } else {
2244 add_partial(n, page, DEACTIVATE_TO_TAIL);
2245 stat(s, FREE_ADD_PARTIAL);
49e22585
CL
2246 }
2247 }
2248
2249 if (n)
2250 spin_unlock(&n->list_lock);
9ada1934
SL
2251
2252 while (discard_page) {
2253 page = discard_page;
2254 discard_page = discard_page->next;
2255
2256 stat(s, DEACTIVATE_EMPTY);
2257 discard_slab(s, page);
2258 stat(s, FREE_SLAB);
2259 }
6dfd1b65 2260#endif /* CONFIG_SLUB_CPU_PARTIAL */
49e22585
CL
2261}
2262
2263/*
9234bae9
WY
2264 * Put a page that was just frozen (in __slab_free|get_partial_node) into a
2265 * partial page slot if available.
49e22585
CL
2266 *
2267 * If we did not find a slot then simply move all the partials to the
2268 * per node partial list.
2269 */
633b0764 2270static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
49e22585 2271{
345c905d 2272#ifdef CONFIG_SLUB_CPU_PARTIAL
49e22585
CL
2273 struct page *oldpage;
2274 int pages;
2275 int pobjects;
2276
d6e0b7fa 2277 preempt_disable();
49e22585
CL
2278 do {
2279 pages = 0;
2280 pobjects = 0;
2281 oldpage = this_cpu_read(s->cpu_slab->partial);
2282
2283 if (oldpage) {
2284 pobjects = oldpage->pobjects;
2285 pages = oldpage->pages;
2286 if (drain && pobjects > s->cpu_partial) {
2287 unsigned long flags;
2288 /*
2289 * partial array is full. Move the existing
2290 * set to the per node partial list.
2291 */
2292 local_irq_save(flags);
59a09917 2293 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
49e22585 2294 local_irq_restore(flags);
e24fc410 2295 oldpage = NULL;
49e22585
CL
2296 pobjects = 0;
2297 pages = 0;
8028dcea 2298 stat(s, CPU_PARTIAL_DRAIN);
49e22585
CL
2299 }
2300 }
2301
2302 pages++;
2303 pobjects += page->objects - page->inuse;
2304
2305 page->pages = pages;
2306 page->pobjects = pobjects;
2307 page->next = oldpage;
2308
d0e0ac97
CG
2309 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2310 != oldpage);
d6e0b7fa
VD
2311 if (unlikely(!s->cpu_partial)) {
2312 unsigned long flags;
2313
2314 local_irq_save(flags);
2315 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2316 local_irq_restore(flags);
2317 }
2318 preempt_enable();
6dfd1b65 2319#endif /* CONFIG_SLUB_CPU_PARTIAL */
49e22585
CL
2320}
2321
dfb4f096 2322static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
81819f0f 2323{
84e554e6 2324 stat(s, CPUSLAB_FLUSH);
d4ff6d35 2325 deactivate_slab(s, c->page, c->freelist, c);
c17dda40
CL
2326
2327 c->tid = next_tid(c->tid);
81819f0f
CL
2328}
2329
2330/*
2331 * Flush cpu slab.
6446faa2 2332 *
81819f0f
CL
2333 * Called from IPI handler with interrupts disabled.
2334 */
0c710013 2335static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
81819f0f 2336{
9dfc6e68 2337 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
81819f0f 2338
1265ef2d
WY
2339 if (c->page)
2340 flush_slab(s, c);
49e22585 2341
1265ef2d 2342 unfreeze_partials(s, c);
81819f0f
CL
2343}
2344
2345static void flush_cpu_slab(void *d)
2346{
2347 struct kmem_cache *s = d;
81819f0f 2348
dfb4f096 2349 __flush_cpu_slab(s, smp_processor_id());
81819f0f
CL
2350}
2351
a8364d55
GBY
2352static bool has_cpu_slab(int cpu, void *info)
2353{
2354 struct kmem_cache *s = info;
2355 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2356
a93cf07b 2357 return c->page || slub_percpu_partial(c);
a8364d55
GBY
2358}
2359
81819f0f
CL
2360static void flush_all(struct kmem_cache *s)
2361{
a8364d55 2362 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
81819f0f
CL
2363}
2364
a96a87bf
SAS
2365/*
2366 * Use the cpu notifier to insure that the cpu slabs are flushed when
2367 * necessary.
2368 */
2369static int slub_cpu_dead(unsigned int cpu)
2370{
2371 struct kmem_cache *s;
2372 unsigned long flags;
2373
2374 mutex_lock(&slab_mutex);
2375 list_for_each_entry(s, &slab_caches, list) {
2376 local_irq_save(flags);
2377 __flush_cpu_slab(s, cpu);
2378 local_irq_restore(flags);
2379 }
2380 mutex_unlock(&slab_mutex);
2381 return 0;
2382}
2383
dfb4f096
CL
2384/*
2385 * Check if the objects in a per cpu structure fit numa
2386 * locality expectations.
2387 */
57d437d2 2388static inline int node_match(struct page *page, int node)
dfb4f096
CL
2389{
2390#ifdef CONFIG_NUMA
6159d0f5 2391 if (node != NUMA_NO_NODE && page_to_nid(page) != node)
dfb4f096
CL
2392 return 0;
2393#endif
2394 return 1;
2395}
2396
9a02d699 2397#ifdef CONFIG_SLUB_DEBUG
781b2ba6
PE
2398static int count_free(struct page *page)
2399{
2400 return page->objects - page->inuse;
2401}
2402
9a02d699
DR
2403static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2404{
2405 return atomic_long_read(&n->total_objects);
2406}
2407#endif /* CONFIG_SLUB_DEBUG */
2408
2409#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
781b2ba6
PE
2410static unsigned long count_partial(struct kmem_cache_node *n,
2411 int (*get_count)(struct page *))
2412{
2413 unsigned long flags;
2414 unsigned long x = 0;
2415 struct page *page;
2416
2417 spin_lock_irqsave(&n->list_lock, flags);
916ac052 2418 list_for_each_entry(page, &n->partial, slab_list)
781b2ba6
PE
2419 x += get_count(page);
2420 spin_unlock_irqrestore(&n->list_lock, flags);
2421 return x;
2422}
9a02d699 2423#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
26c02cf0 2424
781b2ba6
PE
2425static noinline void
2426slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2427{
9a02d699
DR
2428#ifdef CONFIG_SLUB_DEBUG
2429 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2430 DEFAULT_RATELIMIT_BURST);
781b2ba6 2431 int node;
fa45dc25 2432 struct kmem_cache_node *n;
781b2ba6 2433
9a02d699
DR
2434 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2435 return;
2436
5b3810e5
VB
2437 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2438 nid, gfpflags, &gfpflags);
19af27af 2439 pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
f9f58285
FF
2440 s->name, s->object_size, s->size, oo_order(s->oo),
2441 oo_order(s->min));
781b2ba6 2442
3b0efdfa 2443 if (oo_order(s->min) > get_order(s->object_size))
f9f58285
FF
2444 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2445 s->name);
fa5ec8a1 2446
fa45dc25 2447 for_each_kmem_cache_node(s, node, n) {
781b2ba6
PE
2448 unsigned long nr_slabs;
2449 unsigned long nr_objs;
2450 unsigned long nr_free;
2451
26c02cf0
AB
2452 nr_free = count_partial(n, count_free);
2453 nr_slabs = node_nr_slabs(n);
2454 nr_objs = node_nr_objs(n);
781b2ba6 2455
f9f58285 2456 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
781b2ba6
PE
2457 node, nr_slabs, nr_objs, nr_free);
2458 }
9a02d699 2459#endif
781b2ba6
PE
2460}
2461
497b66f2
CL
2462static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2463 int node, struct kmem_cache_cpu **pc)
2464{
6faa6833 2465 void *freelist;
188fd063
CL
2466 struct kmem_cache_cpu *c = *pc;
2467 struct page *page;
497b66f2 2468
128227e7
MW
2469 WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2470
188fd063 2471 freelist = get_partial(s, flags, node, c);
497b66f2 2472
188fd063
CL
2473 if (freelist)
2474 return freelist;
2475
2476 page = new_slab(s, flags, node);
497b66f2 2477 if (page) {
7c8e0181 2478 c = raw_cpu_ptr(s->cpu_slab);
497b66f2
CL
2479 if (c->page)
2480 flush_slab(s, c);
2481
2482 /*
2483 * No other reference to the page yet so we can
2484 * muck around with it freely without cmpxchg
2485 */
6faa6833 2486 freelist = page->freelist;
497b66f2
CL
2487 page->freelist = NULL;
2488
2489 stat(s, ALLOC_SLAB);
497b66f2
CL
2490 c->page = page;
2491 *pc = c;
edde82b6 2492 }
497b66f2 2493
6faa6833 2494 return freelist;
497b66f2
CL
2495}
2496
072bb0aa
MG
2497static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2498{
2499 if (unlikely(PageSlabPfmemalloc(page)))
2500 return gfp_pfmemalloc_allowed(gfpflags);
2501
2502 return true;
2503}
2504
213eeb9f 2505/*
d0e0ac97
CG
2506 * Check the page->freelist of a page and either transfer the freelist to the
2507 * per cpu freelist or deactivate the page.
213eeb9f
CL
2508 *
2509 * The page is still frozen if the return value is not NULL.
2510 *
2511 * If this function returns NULL then the page has been unfrozen.
d24ac77f
JK
2512 *
2513 * This function must be called with interrupt disabled.
213eeb9f
CL
2514 */
2515static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2516{
2517 struct page new;
2518 unsigned long counters;
2519 void *freelist;
2520
2521 do {
2522 freelist = page->freelist;
2523 counters = page->counters;
6faa6833 2524
213eeb9f 2525 new.counters = counters;
a0132ac0 2526 VM_BUG_ON(!new.frozen);
213eeb9f
CL
2527
2528 new.inuse = page->objects;
2529 new.frozen = freelist != NULL;
2530
d24ac77f 2531 } while (!__cmpxchg_double_slab(s, page,
213eeb9f
CL
2532 freelist, counters,
2533 NULL, new.counters,
2534 "get_freelist"));
2535
2536 return freelist;
2537}
2538
81819f0f 2539/*
894b8788
CL
2540 * Slow path. The lockless freelist is empty or we need to perform
2541 * debugging duties.
2542 *
894b8788
CL
2543 * Processing is still very fast if new objects have been freed to the
2544 * regular freelist. In that case we simply take over the regular freelist
2545 * as the lockless freelist and zap the regular freelist.
81819f0f 2546 *
894b8788
CL
2547 * If that is not working then we fall back to the partial lists. We take the
2548 * first element of the freelist as the object to allocate now and move the
2549 * rest of the freelist to the lockless freelist.
81819f0f 2550 *
894b8788 2551 * And if we were unable to get a new slab from the partial slab lists then
6446faa2
CL
2552 * we need to allocate a new slab. This is the slowest path since it involves
2553 * a call to the page allocator and the setup of a new slab.
a380a3c7
CL
2554 *
2555 * Version of __slab_alloc to use when we know that interrupts are
2556 * already disabled (which is the case for bulk allocation).
81819f0f 2557 */
a380a3c7 2558static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
ce71e27c 2559 unsigned long addr, struct kmem_cache_cpu *c)
81819f0f 2560{
6faa6833 2561 void *freelist;
f6e7def7 2562 struct page *page;
81819f0f 2563
f6e7def7
CL
2564 page = c->page;
2565 if (!page)
81819f0f 2566 goto new_slab;
49e22585 2567redo:
6faa6833 2568
57d437d2 2569 if (unlikely(!node_match(page, node))) {
a561ce00
JK
2570 int searchnode = node;
2571
2572 if (node != NUMA_NO_NODE && !node_present_pages(node))
2573 searchnode = node_to_mem_node(node);
2574
2575 if (unlikely(!node_match(page, searchnode))) {
2576 stat(s, ALLOC_NODE_MISMATCH);
d4ff6d35 2577 deactivate_slab(s, page, c->freelist, c);
a561ce00
JK
2578 goto new_slab;
2579 }
fc59c053 2580 }
6446faa2 2581
072bb0aa
MG
2582 /*
2583 * By rights, we should be searching for a slab page that was
2584 * PFMEMALLOC but right now, we are losing the pfmemalloc
2585 * information when the page leaves the per-cpu allocator
2586 */
2587 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
d4ff6d35 2588 deactivate_slab(s, page, c->freelist, c);
072bb0aa
MG
2589 goto new_slab;
2590 }
2591
73736e03 2592 /* must check again c->freelist in case of cpu migration or IRQ */
6faa6833
CL
2593 freelist = c->freelist;
2594 if (freelist)
73736e03 2595 goto load_freelist;
03e404af 2596
f6e7def7 2597 freelist = get_freelist(s, page);
6446faa2 2598
6faa6833 2599 if (!freelist) {
03e404af
CL
2600 c->page = NULL;
2601 stat(s, DEACTIVATE_BYPASS);
fc59c053 2602 goto new_slab;
03e404af 2603 }
6446faa2 2604
84e554e6 2605 stat(s, ALLOC_REFILL);
6446faa2 2606
894b8788 2607load_freelist:
507effea
CL
2608 /*
2609 * freelist is pointing to the list of objects to be used.
2610 * page is pointing to the page from which the objects are obtained.
2611 * That page must be frozen for per cpu allocations to work.
2612 */
a0132ac0 2613 VM_BUG_ON(!c->page->frozen);
6faa6833 2614 c->freelist = get_freepointer(s, freelist);
8a5ec0ba 2615 c->tid = next_tid(c->tid);
6faa6833 2616 return freelist;
81819f0f 2617
81819f0f 2618new_slab:
2cfb7455 2619
a93cf07b
WY
2620 if (slub_percpu_partial(c)) {
2621 page = c->page = slub_percpu_partial(c);
2622 slub_set_percpu_partial(c, page);
49e22585 2623 stat(s, CPU_PARTIAL_ALLOC);
49e22585 2624 goto redo;
81819f0f
CL
2625 }
2626
188fd063 2627 freelist = new_slab_objects(s, gfpflags, node, &c);
01ad8a7b 2628
f4697436 2629 if (unlikely(!freelist)) {
9a02d699 2630 slab_out_of_memory(s, gfpflags, node);
f4697436 2631 return NULL;
81819f0f 2632 }
2cfb7455 2633
f6e7def7 2634 page = c->page;
5091b74a 2635 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
4b6f0750 2636 goto load_freelist;
2cfb7455 2637
497b66f2 2638 /* Only entered in the debug case */
d0e0ac97
CG
2639 if (kmem_cache_debug(s) &&
2640 !alloc_debug_processing(s, page, freelist, addr))
497b66f2 2641 goto new_slab; /* Slab failed checks. Next slab needed */
894b8788 2642
d4ff6d35 2643 deactivate_slab(s, page, get_freepointer(s, freelist), c);
6faa6833 2644 return freelist;
894b8788
CL
2645}
2646
a380a3c7
CL
2647/*
2648 * Another one that disabled interrupt and compensates for possible
2649 * cpu changes by refetching the per cpu area pointer.
2650 */
2651static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2652 unsigned long addr, struct kmem_cache_cpu *c)
2653{
2654 void *p;
2655 unsigned long flags;
2656
2657 local_irq_save(flags);
2658#ifdef CONFIG_PREEMPT
2659 /*
2660 * We may have been preempted and rescheduled on a different
2661 * cpu before disabling interrupts. Need to reload cpu area
2662 * pointer.
2663 */
2664 c = this_cpu_ptr(s->cpu_slab);
2665#endif
2666
2667 p = ___slab_alloc(s, gfpflags, node, addr, c);
2668 local_irq_restore(flags);
2669 return p;
2670}
2671
894b8788
CL
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 */
2b847c3c 2682static __always_inline void *slab_alloc_node(struct kmem_cache *s,
ce71e27c 2683 gfp_t gfpflags, int node, unsigned long addr)
894b8788 2684{
03ec0ed5 2685 void *object;
dfb4f096 2686 struct kmem_cache_cpu *c;
57d437d2 2687 struct page *page;
8a5ec0ba 2688 unsigned long tid;
1f84260c 2689
8135be5a
VD
2690 s = slab_pre_alloc_hook(s, gfpflags);
2691 if (!s)
773ff60e 2692 return NULL;
8a5ec0ba 2693redo:
8a5ec0ba
CL
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.
7cccd80b 2699 *
9aabf810
JK
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.
8a5ec0ba 2703 */
9aabf810
JK
2704 do {
2705 tid = this_cpu_read(s->cpu_slab->tid);
2706 c = raw_cpu_ptr(s->cpu_slab);
859b7a0e
MR
2707 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2708 unlikely(tid != READ_ONCE(c->tid)));
9aabf810
JK
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();
8a5ec0ba 2719
8a5ec0ba
CL
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 */
8a5ec0ba 2726
9dfc6e68 2727 object = c->freelist;
57d437d2 2728 page = c->page;
8eae1492 2729 if (unlikely(!object || !node_match(page, node))) {
dfb4f096 2730 object = __slab_alloc(s, gfpflags, node, addr, c);
8eae1492
DH
2731 stat(s, ALLOC_SLOWPATH);
2732 } else {
0ad9500e
ED
2733 void *next_object = get_freepointer_safe(s, object);
2734
8a5ec0ba 2735 /*
25985edc 2736 * The cmpxchg will only match if there was no additional
8a5ec0ba
CL
2737 * operation and if we are on the right processor.
2738 *
d0e0ac97
CG
2739 * The cmpxchg does the following atomically (without lock
2740 * semantics!)
8a5ec0ba
CL
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 *
d0e0ac97
CG
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.
8a5ec0ba 2748 */
933393f5 2749 if (unlikely(!this_cpu_cmpxchg_double(
8a5ec0ba
CL
2750 s->cpu_slab->freelist, s->cpu_slab->tid,
2751 object, tid,
0ad9500e 2752 next_object, next_tid(tid)))) {
8a5ec0ba
CL
2753
2754 note_cmpxchg_failure("slab_alloc", s, tid);
2755 goto redo;
2756 }
0ad9500e 2757 prefetch_freepointer(s, next_object);
84e554e6 2758 stat(s, ALLOC_FASTPATH);
894b8788 2759 }
6471384a
AP
2760 /*
2761 * If the object has been wiped upon free, make sure it's fully
2762 * initialized by zeroing out freelist pointer.
2763 */
2764 if (unlikely(slab_want_init_on_free(s)) && object)
2765 memset(object + s->offset, 0, sizeof(void *));
8a5ec0ba 2766
6471384a 2767 if (unlikely(slab_want_init_on_alloc(gfpflags, s)) && object)
3b0efdfa 2768 memset(object, 0, s->object_size);
d07dbea4 2769
03ec0ed5 2770 slab_post_alloc_hook(s, gfpflags, 1, &object);
5a896d9e 2771
894b8788 2772 return object;
81819f0f
CL
2773}
2774
2b847c3c
EG
2775static __always_inline void *slab_alloc(struct kmem_cache *s,
2776 gfp_t gfpflags, unsigned long addr)
2777{
2778 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2779}
2780
81819f0f
CL
2781void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2782{
2b847c3c 2783 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
5b882be4 2784
d0e0ac97
CG
2785 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2786 s->size, gfpflags);
5b882be4
EGM
2787
2788 return ret;
81819f0f
CL
2789}
2790EXPORT_SYMBOL(kmem_cache_alloc);
2791
0f24f128 2792#ifdef CONFIG_TRACING
4a92379b
RK
2793void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
2794{
2b847c3c 2795 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
4a92379b 2796 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
0116523c 2797 ret = kasan_kmalloc(s, ret, size, gfpflags);
4a92379b
RK
2798 return ret;
2799}
2800EXPORT_SYMBOL(kmem_cache_alloc_trace);
5b882be4
EGM
2801#endif
2802
81819f0f
CL
2803#ifdef CONFIG_NUMA
2804void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2805{
2b847c3c 2806 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
5b882be4 2807
ca2b84cb 2808 trace_kmem_cache_alloc_node(_RET_IP_, ret,
3b0efdfa 2809 s->object_size, s->size, gfpflags, node);
5b882be4
EGM
2810
2811 return ret;
81819f0f
CL
2812}
2813EXPORT_SYMBOL(kmem_cache_alloc_node);
81819f0f 2814
0f24f128 2815#ifdef CONFIG_TRACING
4a92379b 2816void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
5b882be4 2817 gfp_t gfpflags,
4a92379b 2818 int node, size_t size)
5b882be4 2819{
2b847c3c 2820 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
4a92379b
RK
2821
2822 trace_kmalloc_node(_RET_IP_, ret,
2823 size, s->size, gfpflags, node);
0316bec2 2824
0116523c 2825 ret = kasan_kmalloc(s, ret, size, gfpflags);
4a92379b 2826 return ret;
5b882be4 2827}
4a92379b 2828EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
5b882be4 2829#endif
6dfd1b65 2830#endif /* CONFIG_NUMA */
5b882be4 2831
81819f0f 2832/*
94e4d712 2833 * Slow path handling. This may still be called frequently since objects
894b8788 2834 * have a longer lifetime than the cpu slabs in most processing loads.
81819f0f 2835 *
894b8788
CL
2836 * So we still attempt to reduce cache line usage. Just take the slab
2837 * lock and free the item. If there is no additional partial page
2838 * handling required then we can return immediately.
81819f0f 2839 */
894b8788 2840static void __slab_free(struct kmem_cache *s, struct page *page,
81084651
JDB
2841 void *head, void *tail, int cnt,
2842 unsigned long addr)
2843
81819f0f
CL
2844{
2845 void *prior;
2cfb7455 2846 int was_frozen;
2cfb7455
CL
2847 struct page new;
2848 unsigned long counters;
2849 struct kmem_cache_node *n = NULL;
61728d1e 2850 unsigned long uninitialized_var(flags);
81819f0f 2851
8a5ec0ba 2852 stat(s, FREE_SLOWPATH);
81819f0f 2853
19c7ff9e 2854 if (kmem_cache_debug(s) &&
282acb43 2855 !free_debug_processing(s, page, head, tail, cnt, addr))
80f08c19 2856 return;
6446faa2 2857
2cfb7455 2858 do {
837d678d
JK
2859 if (unlikely(n)) {
2860 spin_unlock_irqrestore(&n->list_lock, flags);
2861 n = NULL;
2862 }
2cfb7455
CL
2863 prior = page->freelist;
2864 counters = page->counters;
81084651 2865 set_freepointer(s, tail, prior);
2cfb7455
CL
2866 new.counters = counters;
2867 was_frozen = new.frozen;
81084651 2868 new.inuse -= cnt;
837d678d 2869 if ((!new.inuse || !prior) && !was_frozen) {
49e22585 2870
c65c1877 2871 if (kmem_cache_has_cpu_partial(s) && !prior) {
49e22585
CL
2872
2873 /*
d0e0ac97
CG
2874 * Slab was on no list before and will be
2875 * partially empty
2876 * We can defer the list move and instead
2877 * freeze it.
49e22585
CL
2878 */
2879 new.frozen = 1;
2880
c65c1877 2881 } else { /* Needs to be taken off a list */
49e22585 2882
b455def2 2883 n = get_node(s, page_to_nid(page));
49e22585
CL
2884 /*
2885 * Speculatively acquire the list_lock.
2886 * If the cmpxchg does not succeed then we may
2887 * drop the list_lock without any processing.
2888 *
2889 * Otherwise the list_lock will synchronize with
2890 * other processors updating the list of slabs.
2891 */
2892 spin_lock_irqsave(&n->list_lock, flags);
2893
2894 }
2cfb7455 2895 }
81819f0f 2896
2cfb7455
CL
2897 } while (!cmpxchg_double_slab(s, page,
2898 prior, counters,
81084651 2899 head, new.counters,
2cfb7455 2900 "__slab_free"));
81819f0f 2901
2cfb7455 2902 if (likely(!n)) {
49e22585
CL
2903
2904 /*
2905 * If we just froze the page then put it onto the
2906 * per cpu partial list.
2907 */
8028dcea 2908 if (new.frozen && !was_frozen) {
49e22585 2909 put_cpu_partial(s, page, 1);
8028dcea
AS
2910 stat(s, CPU_PARTIAL_FREE);
2911 }
49e22585 2912 /*
2cfb7455
CL
2913 * The list lock was not taken therefore no list
2914 * activity can be necessary.
2915 */
b455def2
L
2916 if (was_frozen)
2917 stat(s, FREE_FROZEN);
2918 return;
2919 }
81819f0f 2920
8a5b20ae 2921 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
837d678d
JK
2922 goto slab_empty;
2923
81819f0f 2924 /*
837d678d
JK
2925 * Objects left in the slab. If it was not on the partial list before
2926 * then add it.
81819f0f 2927 */
345c905d 2928 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
a4d3f891 2929 remove_full(s, n, page);
837d678d
JK
2930 add_partial(n, page, DEACTIVATE_TO_TAIL);
2931 stat(s, FREE_ADD_PARTIAL);
8ff12cfc 2932 }
80f08c19 2933 spin_unlock_irqrestore(&n->list_lock, flags);
81819f0f
CL
2934 return;
2935
2936slab_empty:
a973e9dd 2937 if (prior) {
81819f0f 2938 /*
6fbabb20 2939 * Slab on the partial list.
81819f0f 2940 */
5cc6eee8 2941 remove_partial(n, page);
84e554e6 2942 stat(s, FREE_REMOVE_PARTIAL);
c65c1877 2943 } else {
6fbabb20 2944 /* Slab must be on the full list */
c65c1877
PZ
2945 remove_full(s, n, page);
2946 }
2cfb7455 2947
80f08c19 2948 spin_unlock_irqrestore(&n->list_lock, flags);
84e554e6 2949 stat(s, FREE_SLAB);
81819f0f 2950 discard_slab(s, page);
81819f0f
CL
2951}
2952
894b8788
CL
2953/*
2954 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2955 * can perform fastpath freeing without additional function calls.
2956 *
2957 * The fastpath is only possible if we are freeing to the current cpu slab
2958 * of this processor. This typically the case if we have just allocated
2959 * the item before.
2960 *
2961 * If fastpath is not possible then fall back to __slab_free where we deal
2962 * with all sorts of special processing.
81084651
JDB
2963 *
2964 * Bulk free of a freelist with several objects (all pointing to the
2965 * same page) possible by specifying head and tail ptr, plus objects
2966 * count (cnt). Bulk free indicated by tail pointer being set.
894b8788 2967 */
80a9201a
AP
2968static __always_inline void do_slab_free(struct kmem_cache *s,
2969 struct page *page, void *head, void *tail,
2970 int cnt, unsigned long addr)
894b8788 2971{
81084651 2972 void *tail_obj = tail ? : head;
dfb4f096 2973 struct kmem_cache_cpu *c;
8a5ec0ba 2974 unsigned long tid;
8a5ec0ba
CL
2975redo:
2976 /*
2977 * Determine the currently cpus per cpu slab.
2978 * The cpu may change afterward. However that does not matter since
2979 * data is retrieved via this pointer. If we are on the same cpu
2ae44005 2980 * during the cmpxchg then the free will succeed.
8a5ec0ba 2981 */
9aabf810
JK
2982 do {
2983 tid = this_cpu_read(s->cpu_slab->tid);
2984 c = raw_cpu_ptr(s->cpu_slab);
859b7a0e
MR
2985 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2986 unlikely(tid != READ_ONCE(c->tid)));
c016b0bd 2987
9aabf810
JK
2988 /* Same with comment on barrier() in slab_alloc_node() */
2989 barrier();
c016b0bd 2990
442b06bc 2991 if (likely(page == c->page)) {
81084651 2992 set_freepointer(s, tail_obj, c->freelist);
8a5ec0ba 2993
933393f5 2994 if (unlikely(!this_cpu_cmpxchg_double(
8a5ec0ba
CL
2995 s->cpu_slab->freelist, s->cpu_slab->tid,
2996 c->freelist, tid,
81084651 2997 head, next_tid(tid)))) {
8a5ec0ba
CL
2998
2999 note_cmpxchg_failure("slab_free", s, tid);
3000 goto redo;
3001 }
84e554e6 3002 stat(s, FREE_FASTPATH);
894b8788 3003 } else
81084651 3004 __slab_free(s, page, head, tail_obj, cnt, addr);
894b8788 3005
894b8788
CL
3006}
3007
80a9201a
AP
3008static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
3009 void *head, void *tail, int cnt,
3010 unsigned long addr)
3011{
80a9201a 3012 /*
c3895391
AK
3013 * With KASAN enabled slab_free_freelist_hook modifies the freelist
3014 * to remove objects, whose reuse must be delayed.
80a9201a 3015 */
c3895391
AK
3016 if (slab_free_freelist_hook(s, &head, &tail))
3017 do_slab_free(s, page, head, tail, cnt, addr);
80a9201a
AP
3018}
3019
2bd926b4 3020#ifdef CONFIG_KASAN_GENERIC
80a9201a
AP
3021void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3022{
3023 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
3024}
3025#endif
3026
81819f0f
CL
3027void kmem_cache_free(struct kmem_cache *s, void *x)
3028{
b9ce5ef4
GC
3029 s = cache_from_obj(s, x);
3030 if (!s)
79576102 3031 return;
81084651 3032 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
ca2b84cb 3033 trace_kmem_cache_free(_RET_IP_, x);
81819f0f
CL
3034}
3035EXPORT_SYMBOL(kmem_cache_free);
3036
d0ecd894 3037struct detached_freelist {
fbd02630 3038 struct page *page;
d0ecd894
JDB
3039 void *tail;
3040 void *freelist;
3041 int cnt;
376bf125 3042 struct kmem_cache *s;
d0ecd894 3043};
fbd02630 3044
d0ecd894
JDB
3045/*
3046 * This function progressively scans the array with free objects (with
3047 * a limited look ahead) and extract objects belonging to the same
3048 * page. It builds a detached freelist directly within the given
3049 * page/objects. This can happen without any need for
3050 * synchronization, because the objects are owned by running process.
3051 * The freelist is build up as a single linked list in the objects.
3052 * The idea is, that this detached freelist can then be bulk
3053 * transferred to the real freelist(s), but only requiring a single
3054 * synchronization primitive. Look ahead in the array is limited due
3055 * to performance reasons.
3056 */
376bf125
JDB
3057static inline
3058int build_detached_freelist(struct kmem_cache *s, size_t size,
3059 void **p, struct detached_freelist *df)
d0ecd894
JDB
3060{
3061 size_t first_skipped_index = 0;
3062 int lookahead = 3;
3063 void *object;
ca257195 3064 struct page *page;
fbd02630 3065
d0ecd894
JDB
3066 /* Always re-init detached_freelist */
3067 df->page = NULL;
fbd02630 3068
d0ecd894
JDB
3069 do {
3070 object = p[--size];
ca257195 3071 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
d0ecd894 3072 } while (!object && size);
3eed034d 3073
d0ecd894
JDB
3074 if (!object)
3075 return 0;
fbd02630 3076
ca257195
JDB
3077 page = virt_to_head_page(object);
3078 if (!s) {
3079 /* Handle kalloc'ed objects */
3080 if (unlikely(!PageSlab(page))) {
3081 BUG_ON(!PageCompound(page));
3082 kfree_hook(object);
4949148a 3083 __free_pages(page, compound_order(page));
ca257195
JDB
3084 p[size] = NULL; /* mark object processed */
3085 return size;
3086 }
3087 /* Derive kmem_cache from object */
3088 df->s = page->slab_cache;
3089 } else {
3090 df->s = cache_from_obj(s, object); /* Support for memcg */
3091 }
376bf125 3092
d0ecd894 3093 /* Start new detached freelist */
ca257195 3094 df->page = page;
376bf125 3095 set_freepointer(df->s, object, NULL);
d0ecd894
JDB
3096 df->tail = object;
3097 df->freelist = object;
3098 p[size] = NULL; /* mark object processed */
3099 df->cnt = 1;
3100
3101 while (size) {
3102 object = p[--size];
3103 if (!object)
3104 continue; /* Skip processed objects */
3105
3106 /* df->page is always set at this point */
3107 if (df->page == virt_to_head_page(object)) {
3108 /* Opportunity build freelist */
376bf125 3109 set_freepointer(df->s, object, df->freelist);
d0ecd894
JDB
3110 df->freelist = object;
3111 df->cnt++;
3112 p[size] = NULL; /* mark object processed */
3113
3114 continue;
fbd02630 3115 }
d0ecd894
JDB
3116
3117 /* Limit look ahead search */
3118 if (!--lookahead)
3119 break;
3120
3121 if (!first_skipped_index)
3122 first_skipped_index = size + 1;
fbd02630 3123 }
d0ecd894
JDB
3124
3125 return first_skipped_index;
3126}
3127
d0ecd894 3128/* Note that interrupts must be enabled when calling this function. */
376bf125 3129void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
d0ecd894
JDB
3130{
3131 if (WARN_ON(!size))
3132 return;
3133
3134 do {
3135 struct detached_freelist df;
3136
3137 size = build_detached_freelist(s, size, p, &df);
84582c8a 3138 if (!df.page)
d0ecd894
JDB
3139 continue;
3140
376bf125 3141 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
d0ecd894 3142 } while (likely(size));
484748f0
CL
3143}
3144EXPORT_SYMBOL(kmem_cache_free_bulk);
3145
994eb764 3146/* Note that interrupts must be enabled when calling this function. */
865762a8
JDB
3147int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3148 void **p)
484748f0 3149{
994eb764
JDB
3150 struct kmem_cache_cpu *c;
3151 int i;
3152
03ec0ed5
JDB
3153 /* memcg and kmem_cache debug support */
3154 s = slab_pre_alloc_hook(s, flags);
3155 if (unlikely(!s))
3156 return false;
994eb764
JDB
3157 /*
3158 * Drain objects in the per cpu slab, while disabling local
3159 * IRQs, which protects against PREEMPT and interrupts
3160 * handlers invoking normal fastpath.
3161 */
3162 local_irq_disable();
3163 c = this_cpu_ptr(s->cpu_slab);
3164
3165 for (i = 0; i < size; i++) {
3166 void *object = c->freelist;
3167
ebe909e0 3168 if (unlikely(!object)) {
ebe909e0
JDB
3169 /*
3170 * Invoking slow path likely have side-effect
3171 * of re-populating per CPU c->freelist
3172 */
87098373 3173 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
ebe909e0 3174 _RET_IP_, c);
87098373
CL
3175 if (unlikely(!p[i]))
3176 goto error;
3177
ebe909e0
JDB
3178 c = this_cpu_ptr(s->cpu_slab);
3179 continue; /* goto for-loop */
3180 }
994eb764
JDB
3181 c->freelist = get_freepointer(s, object);
3182 p[i] = object;
3183 }
3184 c->tid = next_tid(c->tid);
3185 local_irq_enable();
3186
3187 /* Clear memory outside IRQ disabled fastpath loop */
6471384a 3188 if (unlikely(slab_want_init_on_alloc(flags, s))) {
994eb764
JDB
3189 int j;
3190
3191 for (j = 0; j < i; j++)
3192 memset(p[j], 0, s->object_size);
3193 }
3194
03ec0ed5
JDB
3195 /* memcg and kmem_cache debug support */
3196 slab_post_alloc_hook(s, flags, size, p);
865762a8 3197 return i;
87098373 3198error:
87098373 3199 local_irq_enable();
03ec0ed5
JDB
3200 slab_post_alloc_hook(s, flags, i, p);
3201 __kmem_cache_free_bulk(s, i, p);
865762a8 3202 return 0;
484748f0
CL
3203}
3204EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3205
3206
81819f0f 3207/*
672bba3a
CL
3208 * Object placement in a slab is made very easy because we always start at
3209 * offset 0. If we tune the size of the object to the alignment then we can
3210 * get the required alignment by putting one properly sized object after
3211 * another.
81819f0f
CL
3212 *
3213 * Notice that the allocation order determines the sizes of the per cpu
3214 * caches. Each processor has always one slab available for allocations.
3215 * Increasing the allocation order reduces the number of times that slabs
672bba3a 3216 * must be moved on and off the partial lists and is therefore a factor in
81819f0f 3217 * locking overhead.
81819f0f
CL
3218 */
3219
3220/*
3221 * Mininum / Maximum order of slab pages. This influences locking overhead
3222 * and slab fragmentation. A higher order reduces the number of partial slabs
3223 * and increases the number of allocations possible without having to
3224 * take the list_lock.
3225 */
19af27af
AD
3226static unsigned int slub_min_order;
3227static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3228static unsigned int slub_min_objects;
81819f0f 3229
81819f0f
CL
3230/*
3231 * Calculate the order of allocation given an slab object size.
3232 *
672bba3a
CL
3233 * The order of allocation has significant impact on performance and other
3234 * system components. Generally order 0 allocations should be preferred since
3235 * order 0 does not cause fragmentation in the page allocator. Larger objects
3236 * be problematic to put into order 0 slabs because there may be too much
c124f5b5 3237 * unused space left. We go to a higher order if more than 1/16th of the slab
672bba3a
CL
3238 * would be wasted.
3239 *
3240 * In order to reach satisfactory performance we must ensure that a minimum
3241 * number of objects is in one slab. Otherwise we may generate too much
3242 * activity on the partial lists which requires taking the list_lock. This is
3243 * less a concern for large slabs though which are rarely used.
81819f0f 3244 *
672bba3a
CL
3245 * slub_max_order specifies the order where we begin to stop considering the
3246 * number of objects in a slab as critical. If we reach slub_max_order then
3247 * we try to keep the page order as low as possible. So we accept more waste
3248 * of space in favor of a small page order.
81819f0f 3249 *
672bba3a
CL
3250 * Higher order allocations also allow the placement of more objects in a
3251 * slab and thereby reduce object handling overhead. If the user has
3252 * requested a higher mininum order then we start with that one instead of
3253 * the smallest order which will fit the object.
81819f0f 3254 */
19af27af
AD
3255static inline unsigned int slab_order(unsigned int size,
3256 unsigned int min_objects, unsigned int max_order,
9736d2a9 3257 unsigned int fract_leftover)
81819f0f 3258{
19af27af
AD
3259 unsigned int min_order = slub_min_order;
3260 unsigned int order;
81819f0f 3261
9736d2a9 3262 if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
210b5c06 3263 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
39b26464 3264
9736d2a9 3265 for (order = max(min_order, (unsigned int)get_order(min_objects * size));
5e6d444e 3266 order <= max_order; order++) {
81819f0f 3267
19af27af
AD
3268 unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3269 unsigned int rem;
81819f0f 3270
9736d2a9 3271 rem = slab_size % size;
81819f0f 3272
5e6d444e 3273 if (rem <= slab_size / fract_leftover)
81819f0f 3274 break;
81819f0f 3275 }
672bba3a 3276
81819f0f
CL
3277 return order;
3278}
3279
9736d2a9 3280static inline int calculate_order(unsigned int size)
5e6d444e 3281{
19af27af
AD
3282 unsigned int order;
3283 unsigned int min_objects;
3284 unsigned int max_objects;
5e6d444e
CL
3285
3286 /*
3287 * Attempt to find best configuration for a slab. This
3288 * works by first attempting to generate a layout with
3289 * the best configuration and backing off gradually.
3290 *
422ff4d7 3291 * First we increase the acceptable waste in a slab. Then
5e6d444e
CL
3292 * we reduce the minimum objects required in a slab.
3293 */
3294 min_objects = slub_min_objects;
9b2cd506
CL
3295 if (!min_objects)
3296 min_objects = 4 * (fls(nr_cpu_ids) + 1);
9736d2a9 3297 max_objects = order_objects(slub_max_order, size);
e8120ff1
ZY
3298 min_objects = min(min_objects, max_objects);
3299
5e6d444e 3300 while (min_objects > 1) {
19af27af
AD
3301 unsigned int fraction;
3302
c124f5b5 3303 fraction = 16;
5e6d444e
CL
3304 while (fraction >= 4) {
3305 order = slab_order(size, min_objects,
9736d2a9 3306 slub_max_order, fraction);
5e6d444e
CL
3307 if (order <= slub_max_order)
3308 return order;
3309 fraction /= 2;
3310 }
5086c389 3311 min_objects--;
5e6d444e
CL
3312 }
3313
3314 /*
3315 * We were unable to place multiple objects in a slab. Now
3316 * lets see if we can place a single object there.
3317 */
9736d2a9 3318 order = slab_order(size, 1, slub_max_order, 1);
5e6d444e
CL
3319 if (order <= slub_max_order)
3320 return order;
3321
3322 /*
3323 * Doh this slab cannot be placed using slub_max_order.
3324 */
9736d2a9 3325 order = slab_order(size, 1, MAX_ORDER, 1);
818cf590 3326 if (order < MAX_ORDER)
5e6d444e
CL
3327 return order;
3328 return -ENOSYS;
3329}
3330
5595cffc 3331static void
4053497d 3332init_kmem_cache_node(struct kmem_cache_node *n)
81819f0f
CL
3333{
3334 n->nr_partial = 0;
81819f0f
CL
3335 spin_lock_init(&n->list_lock);
3336 INIT_LIST_HEAD(&n->partial);
8ab1372f 3337#ifdef CONFIG_SLUB_DEBUG
0f389ec6 3338 atomic_long_set(&n->nr_slabs, 0);
02b71b70 3339 atomic_long_set(&n->total_objects, 0);
643b1138 3340 INIT_LIST_HEAD(&n->full);
8ab1372f 3341#endif
81819f0f
CL
3342}
3343
55136592 3344static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
4c93c355 3345{
6c182dc0 3346 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
95a05b42 3347 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
4c93c355 3348
8a5ec0ba 3349 /*
d4d84fef
CM
3350 * Must align to double word boundary for the double cmpxchg
3351 * instructions to work; see __pcpu_double_call_return_bool().
8a5ec0ba 3352 */
d4d84fef
CM
3353 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3354 2 * sizeof(void *));
8a5ec0ba
CL
3355
3356 if (!s->cpu_slab)
3357 return 0;
3358
3359 init_kmem_cache_cpus(s);
4c93c355 3360
8a5ec0ba 3361 return 1;
4c93c355 3362}
4c93c355 3363
51df1142
CL
3364static struct kmem_cache *kmem_cache_node;
3365
81819f0f
CL
3366/*
3367 * No kmalloc_node yet so do it by hand. We know that this is the first
3368 * slab on the node for this slabcache. There are no concurrent accesses
3369 * possible.
3370 *
721ae22a
ZYW
3371 * Note that this function only works on the kmem_cache_node
3372 * when allocating for the kmem_cache_node. This is used for bootstrapping
4c93c355 3373 * memory on a fresh node that has no slab structures yet.
81819f0f 3374 */
55136592 3375static void early_kmem_cache_node_alloc(int node)
81819f0f
CL
3376{
3377 struct page *page;
3378 struct kmem_cache_node *n;
3379
51df1142 3380 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
81819f0f 3381
51df1142 3382 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
81819f0f
CL
3383
3384 BUG_ON(!page);
a2f92ee7 3385 if (page_to_nid(page) != node) {
f9f58285
FF
3386 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3387 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
a2f92ee7
CL
3388 }
3389
81819f0f
CL
3390 n = page->freelist;
3391 BUG_ON(!n);
8ab1372f 3392#ifdef CONFIG_SLUB_DEBUG
f7cb1933 3393 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
51df1142 3394 init_tracking(kmem_cache_node, n);
8ab1372f 3395#endif
12b22386 3396 n = kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
505f5dcb 3397 GFP_KERNEL);
12b22386
AK
3398 page->freelist = get_freepointer(kmem_cache_node, n);
3399 page->inuse = 1;
3400 page->frozen = 0;
3401 kmem_cache_node->node[node] = n;
4053497d 3402 init_kmem_cache_node(n);
51df1142 3403 inc_slabs_node(kmem_cache_node, node, page->objects);
6446faa2 3404
67b6c900 3405 /*
1e4dd946
SR
3406 * No locks need to be taken here as it has just been
3407 * initialized and there is no concurrent access.
67b6c900 3408 */
1e4dd946 3409 __add_partial(n, page, DEACTIVATE_TO_HEAD);
81819f0f
CL
3410}
3411
3412static void free_kmem_cache_nodes(struct kmem_cache *s)
3413{
3414 int node;
fa45dc25 3415 struct kmem_cache_node *n;
81819f0f 3416
fa45dc25 3417 for_each_kmem_cache_node(s, node, n) {
81819f0f 3418 s->node[node] = NULL;
ea37df54 3419 kmem_cache_free(kmem_cache_node, n);
81819f0f
CL
3420 }
3421}
3422
52b4b950
DS
3423void __kmem_cache_release(struct kmem_cache *s)
3424{
210e7a43 3425 cache_random_seq_destroy(s);
52b4b950
DS
3426 free_percpu(s->cpu_slab);
3427 free_kmem_cache_nodes(s);
3428}
3429
55136592 3430static int init_kmem_cache_nodes(struct kmem_cache *s)
81819f0f
CL
3431{
3432 int node;
81819f0f 3433
f64dc58c 3434 for_each_node_state(node, N_NORMAL_MEMORY) {
81819f0f
CL
3435 struct kmem_cache_node *n;
3436
73367bd8 3437 if (slab_state == DOWN) {
55136592 3438 early_kmem_cache_node_alloc(node);
73367bd8
AD
3439 continue;
3440 }
51df1142 3441 n = kmem_cache_alloc_node(kmem_cache_node,
55136592 3442 GFP_KERNEL, node);
81819f0f 3443
73367bd8
AD
3444 if (!n) {
3445 free_kmem_cache_nodes(s);
3446 return 0;
81819f0f 3447 }
73367bd8 3448
4053497d 3449 init_kmem_cache_node(n);
ea37df54 3450 s->node[node] = n;
81819f0f
CL
3451 }
3452 return 1;
3453}
81819f0f 3454
c0bdb232 3455static void set_min_partial(struct kmem_cache *s, unsigned long min)
3b89d7d8
DR
3456{
3457 if (min < MIN_PARTIAL)
3458 min = MIN_PARTIAL;
3459 else if (min > MAX_PARTIAL)
3460 min = MAX_PARTIAL;
3461 s->min_partial = min;
3462}
3463
e6d0e1dc
WY
3464static void set_cpu_partial(struct kmem_cache *s)
3465{
3466#ifdef CONFIG_SLUB_CPU_PARTIAL
3467 /*
3468 * cpu_partial determined the maximum number of objects kept in the
3469 * per cpu partial lists of a processor.
3470 *
3471 * Per cpu partial lists mainly contain slabs that just have one
3472 * object freed. If they are used for allocation then they can be
3473 * filled up again with minimal effort. The slab will never hit the
3474 * per node partial lists and therefore no locking will be required.
3475 *
3476 * This setting also determines
3477 *
3478 * A) The number of objects from per cpu partial slabs dumped to the
3479 * per node list when we reach the limit.
3480 * B) The number of objects in cpu partial slabs to extract from the
3481 * per node list when we run out of per cpu objects. We only fetch
3482 * 50% to keep some capacity around for frees.
3483 */
3484 if (!kmem_cache_has_cpu_partial(s))
3485 s->cpu_partial = 0;
3486 else if (s->size >= PAGE_SIZE)
3487 s->cpu_partial = 2;
3488 else if (s->size >= 1024)
3489 s->cpu_partial = 6;
3490 else if (s->size >= 256)
3491 s->cpu_partial = 13;
3492 else
3493 s->cpu_partial = 30;
3494#endif
3495}
3496
81819f0f
CL
3497/*
3498 * calculate_sizes() determines the order and the distribution of data within
3499 * a slab object.
3500 */
06b285dc 3501static int calculate_sizes(struct kmem_cache *s, int forced_order)
81819f0f 3502{
d50112ed 3503 slab_flags_t flags = s->flags;
be4a7988 3504 unsigned int size = s->object_size;
19af27af 3505 unsigned int order;
81819f0f 3506
d8b42bf5
CL
3507 /*
3508 * Round up object size to the next word boundary. We can only
3509 * place the free pointer at word boundaries and this determines
3510 * the possible location of the free pointer.
3511 */
3512 size = ALIGN(size, sizeof(void *));
3513
3514#ifdef CONFIG_SLUB_DEBUG
81819f0f
CL
3515 /*
3516 * Determine if we can poison the object itself. If the user of
3517 * the slab may touch the object after free or before allocation
3518 * then we should never poison the object itself.
3519 */
5f0d5a3a 3520 if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
c59def9f 3521 !s->ctor)
81819f0f
CL
3522 s->flags |= __OBJECT_POISON;
3523 else
3524 s->flags &= ~__OBJECT_POISON;
3525
81819f0f
CL
3526
3527 /*
672bba3a 3528 * If we are Redzoning then check if there is some space between the
81819f0f 3529 * end of the object and the free pointer. If not then add an
672bba3a 3530 * additional word to have some bytes to store Redzone information.
81819f0f 3531 */
3b0efdfa 3532 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
81819f0f 3533 size += sizeof(void *);
41ecc55b 3534#endif
81819f0f
CL
3535
3536 /*
672bba3a
CL
3537 * With that we have determined the number of bytes in actual use
3538 * by the object. This is the potential offset to the free pointer.
81819f0f
CL
3539 */
3540 s->inuse = size;
3541
5f0d5a3a 3542 if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
c59def9f 3543 s->ctor)) {
81819f0f
CL
3544 /*
3545 * Relocate free pointer after the object if it is not
3546 * permitted to overwrite the first word of the object on
3547 * kmem_cache_free.
3548 *
3549 * This is the case if we do RCU, have a constructor or
3550 * destructor or are poisoning the objects.
3551 */
3552 s->offset = size;
3553 size += sizeof(void *);
3554 }
3555
c12b3c62 3556#ifdef CONFIG_SLUB_DEBUG
81819f0f
CL
3557 if (flags & SLAB_STORE_USER)
3558 /*
3559 * Need to store information about allocs and frees after
3560 * the object.
3561 */
3562 size += 2 * sizeof(struct track);
80a9201a 3563#endif
81819f0f 3564
80a9201a
AP
3565 kasan_cache_create(s, &size, &s->flags);
3566#ifdef CONFIG_SLUB_DEBUG
d86bd1be 3567 if (flags & SLAB_RED_ZONE) {
81819f0f
CL
3568 /*
3569 * Add some empty padding so that we can catch
3570 * overwrites from earlier objects rather than let
3571 * tracking information or the free pointer be
0211a9c8 3572 * corrupted if a user writes before the start
81819f0f
CL
3573 * of the object.
3574 */
3575 size += sizeof(void *);
d86bd1be
JK
3576
3577 s->red_left_pad = sizeof(void *);
3578 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3579 size += s->red_left_pad;
3580 }
41ecc55b 3581#endif
672bba3a 3582
81819f0f
CL
3583 /*
3584 * SLUB stores one object immediately after another beginning from
3585 * offset 0. In order to align the objects we have to simply size
3586 * each object to conform to the alignment.
3587 */
45906855 3588 size = ALIGN(size, s->align);
81819f0f 3589 s->size = size;
06b285dc
CL
3590 if (forced_order >= 0)
3591 order = forced_order;
3592 else
9736d2a9 3593 order = calculate_order(size);
81819f0f 3594
19af27af 3595 if ((int)order < 0)
81819f0f
CL
3596 return 0;
3597
b7a49f0d 3598 s->allocflags = 0;
834f3d11 3599 if (order)
b7a49f0d
CL
3600 s->allocflags |= __GFP_COMP;
3601
3602 if (s->flags & SLAB_CACHE_DMA)
2c59dd65 3603 s->allocflags |= GFP_DMA;
b7a49f0d 3604
6d6ea1e9
NB
3605 if (s->flags & SLAB_CACHE_DMA32)
3606 s->allocflags |= GFP_DMA32;
3607
b7a49f0d
CL
3608 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3609 s->allocflags |= __GFP_RECLAIMABLE;
3610
81819f0f
CL
3611 /*
3612 * Determine the number of objects per slab
3613 */
9736d2a9
MW
3614 s->oo = oo_make(order, size);
3615 s->min = oo_make(get_order(size), size);
205ab99d
CL
3616 if (oo_objects(s->oo) > oo_objects(s->max))
3617 s->max = s->oo;
81819f0f 3618
834f3d11 3619 return !!oo_objects(s->oo);
81819f0f
CL
3620}
3621
d50112ed 3622static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
81819f0f 3623{
8a13a4cc 3624 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
2482ddec
KC
3625#ifdef CONFIG_SLAB_FREELIST_HARDENED
3626 s->random = get_random_long();
3627#endif
81819f0f 3628
06b285dc 3629 if (!calculate_sizes(s, -1))
81819f0f 3630 goto error;
3de47213
DR
3631 if (disable_higher_order_debug) {
3632 /*
3633 * Disable debugging flags that store metadata if the min slab
3634 * order increased.
3635 */
3b0efdfa 3636 if (get_order(s->size) > get_order(s->object_size)) {
3de47213
DR
3637 s->flags &= ~DEBUG_METADATA_FLAGS;
3638 s->offset = 0;
3639 if (!calculate_sizes(s, -1))
3640 goto error;
3641 }
3642 }
81819f0f 3643
2565409f
HC
3644#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3645 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
149daaf3 3646 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
b789ef51
CL
3647 /* Enable fast mode */
3648 s->flags |= __CMPXCHG_DOUBLE;
3649#endif
3650
3b89d7d8
DR
3651 /*
3652 * The larger the object size is, the more pages we want on the partial
3653 * list to avoid pounding the page allocator excessively.
3654 */
49e22585
CL
3655 set_min_partial(s, ilog2(s->size) / 2);
3656
e6d0e1dc 3657 set_cpu_partial(s);
49e22585 3658
81819f0f 3659#ifdef CONFIG_NUMA
e2cb96b7 3660 s->remote_node_defrag_ratio = 1000;
81819f0f 3661#endif
210e7a43
TG
3662
3663 /* Initialize the pre-computed randomized freelist if slab is up */
3664 if (slab_state >= UP) {
3665 if (init_cache_random_seq(s))
3666 goto error;
3667 }
3668
55136592 3669 if (!init_kmem_cache_nodes(s))
dfb4f096 3670 goto error;
81819f0f 3671
55136592 3672 if (alloc_kmem_cache_cpus(s))
278b1bb1 3673 return 0;
ff12059e 3674
4c93c355 3675 free_kmem_cache_nodes(s);
81819f0f 3676error:
278b1bb1 3677 return -EINVAL;
81819f0f 3678}
81819f0f 3679
33b12c38
CL
3680static void list_slab_objects(struct kmem_cache *s, struct page *page,
3681 const char *text)
3682{
3683#ifdef CONFIG_SLUB_DEBUG
3684 void *addr = page_address(page);
3685 void *p;
0684e652 3686 unsigned long *map = bitmap_zalloc(page->objects, GFP_ATOMIC);
bbd7d57b
ED
3687 if (!map)
3688 return;
945cf2b6 3689 slab_err(s, page, text, s->name);
33b12c38 3690 slab_lock(page);
33b12c38 3691
5f80b13a 3692 get_map(s, page, map);
33b12c38
CL
3693 for_each_object(p, s, addr, page->objects) {
3694
3695 if (!test_bit(slab_index(p, s, addr), map)) {
f9f58285 3696 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
33b12c38
CL
3697 print_tracking(s, p);
3698 }
3699 }
3700 slab_unlock(page);
0684e652 3701 bitmap_free(map);
33b12c38
CL
3702#endif
3703}
3704
81819f0f 3705/*
599870b1 3706 * Attempt to free all partial slabs on a node.
52b4b950
DS
3707 * This is called from __kmem_cache_shutdown(). We must take list_lock
3708 * because sysfs file might still access partial list after the shutdowning.
81819f0f 3709 */
599870b1 3710static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
81819f0f 3711{
60398923 3712 LIST_HEAD(discard);
81819f0f
CL
3713 struct page *page, *h;
3714
52b4b950
DS
3715 BUG_ON(irqs_disabled());
3716 spin_lock_irq(&n->list_lock);
916ac052 3717 list_for_each_entry_safe(page, h, &n->partial, slab_list) {
81819f0f 3718 if (!page->inuse) {
52b4b950 3719 remove_partial(n, page);
916ac052 3720 list_add(&page->slab_list, &discard);
33b12c38
CL
3721 } else {
3722 list_slab_objects(s, page,
52b4b950 3723 "Objects remaining in %s on __kmem_cache_shutdown()");
599870b1 3724 }
33b12c38 3725 }
52b4b950 3726 spin_unlock_irq(&n->list_lock);
60398923 3727
916ac052 3728 list_for_each_entry_safe(page, h, &discard, slab_list)
60398923 3729 discard_slab(s, page);
81819f0f
CL
3730}
3731
f9e13c0a
SB
3732bool __kmem_cache_empty(struct kmem_cache *s)
3733{
3734 int node;
3735 struct kmem_cache_node *n;
3736
3737 for_each_kmem_cache_node(s, node, n)
3738 if (n->nr_partial || slabs_node(s, node))
3739 return false;
3740 return true;
3741}
3742
81819f0f 3743/*
672bba3a 3744 * Release all resources used by a slab cache.
81819f0f 3745 */
52b4b950 3746int __kmem_cache_shutdown(struct kmem_cache *s)
81819f0f
CL
3747{
3748 int node;
fa45dc25 3749 struct kmem_cache_node *n;
81819f0f
CL
3750
3751 flush_all(s);
81819f0f 3752 /* Attempt to free all objects */
fa45dc25 3753 for_each_kmem_cache_node(s, node, n) {
599870b1
CL
3754 free_partial(s, n);
3755 if (n->nr_partial || slabs_node(s, node))
81819f0f
CL
3756 return 1;
3757 }
bf5eb3de 3758 sysfs_slab_remove(s);
81819f0f
CL
3759 return 0;
3760}
3761
81819f0f
CL
3762/********************************************************************
3763 * Kmalloc subsystem
3764 *******************************************************************/
3765
81819f0f
CL
3766static int __init setup_slub_min_order(char *str)
3767{
19af27af 3768 get_option(&str, (int *)&slub_min_order);
81819f0f
CL
3769
3770 return 1;
3771}
3772
3773__setup("slub_min_order=", setup_slub_min_order);
3774
3775static int __init setup_slub_max_order(char *str)
3776{
19af27af
AD
3777 get_option(&str, (int *)&slub_max_order);
3778 slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
81819f0f
CL
3779
3780 return 1;
3781}
3782
3783__setup("slub_max_order=", setup_slub_max_order);
3784
3785static int __init setup_slub_min_objects(char *str)
3786{
19af27af 3787 get_option(&str, (int *)&slub_min_objects);
81819f0f
CL
3788
3789 return 1;
3790}
3791
3792__setup("slub_min_objects=", setup_slub_min_objects);
3793
81819f0f
CL
3794void *__kmalloc(size_t size, gfp_t flags)
3795{
aadb4bc4 3796 struct kmem_cache *s;
5b882be4 3797 void *ret;
81819f0f 3798
95a05b42 3799 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
eada35ef 3800 return kmalloc_large(size, flags);
aadb4bc4 3801
2c59dd65 3802 s = kmalloc_slab(size, flags);
aadb4bc4
CL
3803
3804 if (unlikely(ZERO_OR_NULL_PTR(s)))
6cb8f913
CL
3805 return s;
3806
2b847c3c 3807 ret = slab_alloc(s, flags, _RET_IP_);
5b882be4 3808
ca2b84cb 3809 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
5b882be4 3810
0116523c 3811 ret = kasan_kmalloc(s, ret, size, flags);
0316bec2 3812
5b882be4 3813 return ret;
81819f0f
CL
3814}
3815EXPORT_SYMBOL(__kmalloc);
3816
5d1f57e4 3817#ifdef CONFIG_NUMA
f619cfe1
CL
3818static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3819{
b1eeab67 3820 struct page *page;
e4f7c0b4 3821 void *ptr = NULL;
f619cfe1 3822
75f296d9 3823 flags |= __GFP_COMP;
4949148a 3824 page = alloc_pages_node(node, flags, get_order(size));
f619cfe1 3825 if (page)
e4f7c0b4
CM
3826 ptr = page_address(page);
3827
0116523c 3828 return kmalloc_large_node_hook(ptr, size, flags);
f619cfe1
CL
3829}
3830
81819f0f
CL
3831void *__kmalloc_node(size_t size, gfp_t flags, int node)
3832{
aadb4bc4 3833 struct kmem_cache *s;
5b882be4 3834 void *ret;
81819f0f 3835
95a05b42 3836 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
5b882be4
EGM
3837 ret = kmalloc_large_node(size, flags, node);
3838
ca2b84cb
EGM
3839 trace_kmalloc_node(_RET_IP_, ret,
3840 size, PAGE_SIZE << get_order(size),
3841 flags, node);
5b882be4
EGM
3842
3843 return ret;
3844 }
aadb4bc4 3845
2c59dd65 3846 s = kmalloc_slab(size, flags);
aadb4bc4
CL
3847
3848 if (unlikely(ZERO_OR_NULL_PTR(s)))
6cb8f913
CL
3849 return s;
3850
2b847c3c 3851 ret = slab_alloc_node(s, flags, node, _RET_IP_);
5b882be4 3852
ca2b84cb 3853 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
5b882be4 3854
0116523c 3855 ret = kasan_kmalloc(s, ret, size, flags);
0316bec2 3856
5b882be4 3857 return ret;
81819f0f
CL
3858}
3859EXPORT_SYMBOL(__kmalloc_node);
6dfd1b65 3860#endif /* CONFIG_NUMA */
81819f0f 3861
ed18adc1
KC
3862#ifdef CONFIG_HARDENED_USERCOPY
3863/*
afcc90f8
KC
3864 * Rejects incorrectly sized objects and objects that are to be copied
3865 * to/from userspace but do not fall entirely within the containing slab
3866 * cache's usercopy region.
ed18adc1
KC
3867 *
3868 * Returns NULL if check passes, otherwise const char * to name of cache
3869 * to indicate an error.
3870 */
f4e6e289
KC
3871void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3872 bool to_user)
ed18adc1
KC
3873{
3874 struct kmem_cache *s;
44065b2e 3875 unsigned int offset;
ed18adc1
KC
3876 size_t object_size;
3877
96fedce2
AK
3878 ptr = kasan_reset_tag(ptr);
3879
ed18adc1
KC
3880 /* Find object and usable object size. */
3881 s = page->slab_cache;
ed18adc1
KC
3882
3883 /* Reject impossible pointers. */
3884 if (ptr < page_address(page))
f4e6e289
KC
3885 usercopy_abort("SLUB object not in SLUB page?!", NULL,
3886 to_user, 0, n);
ed18adc1
KC
3887
3888 /* Find offset within object. */
3889 offset = (ptr - page_address(page)) % s->size;
3890
3891 /* Adjust for redzone and reject if within the redzone. */
3892 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3893 if (offset < s->red_left_pad)
f4e6e289
KC
3894 usercopy_abort("SLUB object in left red zone",
3895 s->name, to_user, offset, n);
ed18adc1
KC
3896 offset -= s->red_left_pad;
3897 }
3898
afcc90f8
KC
3899 /* Allow address range falling entirely within usercopy region. */
3900 if (offset >= s->useroffset &&
3901 offset - s->useroffset <= s->usersize &&
3902 n <= s->useroffset - offset + s->usersize)
f4e6e289 3903 return;
ed18adc1 3904
afcc90f8
KC
3905 /*
3906 * If the copy is still within the allocated object, produce
3907 * a warning instead of rejecting the copy. This is intended
3908 * to be a temporary method to find any missing usercopy
3909 * whitelists.
3910 */
3911 object_size = slab_ksize(s);
2d891fbc
KC
3912 if (usercopy_fallback &&
3913 offset <= object_size && n <= object_size - offset) {
afcc90f8
KC
3914 usercopy_warn("SLUB object", s->name, to_user, offset, n);
3915 return;
3916 }
ed18adc1 3917
f4e6e289 3918 usercopy_abort("SLUB object", s->name, to_user, offset, n);
ed18adc1
KC
3919}
3920#endif /* CONFIG_HARDENED_USERCOPY */
3921
10d1f8cb 3922size_t __ksize(const void *object)
81819f0f 3923{
272c1d21 3924 struct page *page;
81819f0f 3925
ef8b4520 3926 if (unlikely(object == ZERO_SIZE_PTR))
272c1d21
CL
3927 return 0;
3928
294a80a8 3929 page = virt_to_head_page(object);
294a80a8 3930
76994412
PE
3931 if (unlikely(!PageSlab(page))) {
3932 WARN_ON(!PageCompound(page));
294a80a8 3933 return PAGE_SIZE << compound_order(page);
76994412 3934 }
81819f0f 3935
1b4f59e3 3936 return slab_ksize(page->slab_cache);
81819f0f 3937}
10d1f8cb 3938EXPORT_SYMBOL(__ksize);
81819f0f
CL
3939
3940void kfree(const void *x)
3941{
81819f0f 3942 struct page *page;
5bb983b0 3943 void *object = (void *)x;
81819f0f 3944
2121db74
PE
3945 trace_kfree(_RET_IP_, x);
3946
2408c550 3947 if (unlikely(ZERO_OR_NULL_PTR(x)))
81819f0f
CL
3948 return;
3949
b49af68f 3950 page = virt_to_head_page(x);
aadb4bc4 3951 if (unlikely(!PageSlab(page))) {
0937502a 3952 BUG_ON(!PageCompound(page));
47adccce 3953 kfree_hook(object);
4949148a 3954 __free_pages(page, compound_order(page));
aadb4bc4
CL
3955 return;
3956 }
81084651 3957 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
81819f0f
CL
3958}
3959EXPORT_SYMBOL(kfree);
3960
832f37f5
VD
3961#define SHRINK_PROMOTE_MAX 32
3962
2086d26a 3963/*
832f37f5
VD
3964 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3965 * up most to the head of the partial lists. New allocations will then
3966 * fill those up and thus they can be removed from the partial lists.
672bba3a
CL
3967 *
3968 * The slabs with the least items are placed last. This results in them
3969 * being allocated from last increasing the chance that the last objects
3970 * are freed in them.
2086d26a 3971 */
c9fc5864 3972int __kmem_cache_shrink(struct kmem_cache *s)
2086d26a
CL
3973{
3974 int node;
3975 int i;
3976 struct kmem_cache_node *n;
3977 struct page *page;
3978 struct page *t;
832f37f5
VD
3979 struct list_head discard;
3980 struct list_head promote[SHRINK_PROMOTE_MAX];
2086d26a 3981 unsigned long flags;
ce3712d7 3982 int ret = 0;
2086d26a 3983
2086d26a 3984 flush_all(s);
fa45dc25 3985 for_each_kmem_cache_node(s, node, n) {
832f37f5
VD
3986 INIT_LIST_HEAD(&discard);
3987 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3988 INIT_LIST_HEAD(promote + i);
2086d26a
CL
3989
3990 spin_lock_irqsave(&n->list_lock, flags);
3991
3992 /*
832f37f5 3993 * Build lists of slabs to discard or promote.
2086d26a 3994 *
672bba3a
CL
3995 * Note that concurrent frees may occur while we hold the
3996 * list_lock. page->inuse here is the upper limit.
2086d26a 3997 */
916ac052 3998 list_for_each_entry_safe(page, t, &n->partial, slab_list) {
832f37f5
VD
3999 int free = page->objects - page->inuse;
4000
4001 /* Do not reread page->inuse */
4002 barrier();
4003
4004 /* We do not keep full slabs on the list */
4005 BUG_ON(free <= 0);
4006
4007 if (free == page->objects) {
916ac052 4008 list_move(&page->slab_list, &discard);
69cb8e6b 4009 n->nr_partial--;
832f37f5 4010 } else if (free <= SHRINK_PROMOTE_MAX)
916ac052 4011 list_move(&page->slab_list, promote + free - 1);
2086d26a
CL
4012 }
4013
2086d26a 4014 /*
832f37f5
VD
4015 * Promote the slabs filled up most to the head of the
4016 * partial list.
2086d26a 4017 */
832f37f5
VD
4018 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4019 list_splice(promote + i, &n->partial);
2086d26a 4020
2086d26a 4021 spin_unlock_irqrestore(&n->list_lock, flags);
69cb8e6b
CL
4022
4023 /* Release empty slabs */
916ac052 4024 list_for_each_entry_safe(page, t, &discard, slab_list)
69cb8e6b 4025 discard_slab(s, page);
ce3712d7
VD
4026
4027 if (slabs_node(s, node))
4028 ret = 1;
2086d26a
CL
4029 }
4030
ce3712d7 4031 return ret;
2086d26a 4032}
2086d26a 4033
c9fc5864 4034#ifdef CONFIG_MEMCG
43486694 4035void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s)
01fb58bc 4036{
50862ce7
TH
4037 /*
4038 * Called with all the locks held after a sched RCU grace period.
4039 * Even if @s becomes empty after shrinking, we can't know that @s
4040 * doesn't have allocations already in-flight and thus can't
4041 * destroy @s until the associated memcg is released.
4042 *
4043 * However, let's remove the sysfs files for empty caches here.
4044 * Each cache has a lot of interface files which aren't
4045 * particularly useful for empty draining caches; otherwise, we can
4046 * easily end up with millions of unnecessary sysfs files on
4047 * systems which have a lot of memory and transient cgroups.
4048 */
4049 if (!__kmem_cache_shrink(s))
4050 sysfs_slab_remove(s);
01fb58bc
TH
4051}
4052
c9fc5864
TH
4053void __kmemcg_cache_deactivate(struct kmem_cache *s)
4054{
4055 /*
4056 * Disable empty slabs caching. Used to avoid pinning offline
4057 * memory cgroups by kmem pages that can be freed.
4058 */
e6d0e1dc 4059 slub_set_cpu_partial(s, 0);
c9fc5864 4060 s->min_partial = 0;
c9fc5864 4061}
6dfd1b65 4062#endif /* CONFIG_MEMCG */
c9fc5864 4063
b9049e23
YG
4064static int slab_mem_going_offline_callback(void *arg)
4065{
4066 struct kmem_cache *s;
4067
18004c5d 4068 mutex_lock(&slab_mutex);
b9049e23 4069 list_for_each_entry(s, &slab_caches, list)
c9fc5864 4070 __kmem_cache_shrink(s);
18004c5d 4071 mutex_unlock(&slab_mutex);
b9049e23
YG
4072
4073 return 0;
4074}
4075
4076static void slab_mem_offline_callback(void *arg)
4077{
4078 struct kmem_cache_node *n;
4079 struct kmem_cache *s;
4080 struct memory_notify *marg = arg;
4081 int offline_node;
4082
b9d5ab25 4083 offline_node = marg->status_change_nid_normal;
b9049e23
YG
4084
4085 /*
4086 * If the node still has available memory. we need kmem_cache_node
4087 * for it yet.
4088 */
4089 if (offline_node < 0)
4090 return;
4091
18004c5d 4092 mutex_lock(&slab_mutex);
b9049e23
YG
4093 list_for_each_entry(s, &slab_caches, list) {
4094 n = get_node(s, offline_node);
4095 if (n) {
4096 /*
4097 * if n->nr_slabs > 0, slabs still exist on the node
4098 * that is going down. We were unable to free them,
c9404c9c 4099 * and offline_pages() function shouldn't call this
b9049e23
YG
4100 * callback. So, we must fail.
4101 */
0f389ec6 4102 BUG_ON(slabs_node(s, offline_node));
b9049e23
YG
4103
4104 s->node[offline_node] = NULL;
8de66a0c 4105 kmem_cache_free(kmem_cache_node, n);
b9049e23
YG
4106 }
4107 }
18004c5d 4108 mutex_unlock(&slab_mutex);
b9049e23
YG
4109}
4110
4111static int slab_mem_going_online_callback(void *arg)
4112{
4113 struct kmem_cache_node *n;
4114 struct kmem_cache *s;
4115 struct memory_notify *marg = arg;
b9d5ab25 4116 int nid = marg->status_change_nid_normal;
b9049e23
YG
4117 int ret = 0;
4118
4119 /*
4120 * If the node's memory is already available, then kmem_cache_node is
4121 * already created. Nothing to do.
4122 */
4123 if (nid < 0)
4124 return 0;
4125
4126 /*
0121c619 4127 * We are bringing a node online. No memory is available yet. We must
b9049e23
YG
4128 * allocate a kmem_cache_node structure in order to bring the node
4129 * online.
4130 */
18004c5d 4131 mutex_lock(&slab_mutex);
b9049e23
YG
4132 list_for_each_entry(s, &slab_caches, list) {
4133 /*
4134 * XXX: kmem_cache_alloc_node will fallback to other nodes
4135 * since memory is not yet available from the node that
4136 * is brought up.
4137 */
8de66a0c 4138 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
b9049e23
YG
4139 if (!n) {
4140 ret = -ENOMEM;
4141 goto out;
4142 }
4053497d 4143 init_kmem_cache_node(n);
b9049e23
YG
4144 s->node[nid] = n;
4145 }
4146out:
18004c5d 4147 mutex_unlock(&slab_mutex);
b9049e23
YG
4148 return ret;
4149}
4150
4151static int slab_memory_callback(struct notifier_block *self,
4152 unsigned long action, void *arg)
4153{
4154 int ret = 0;
4155
4156 switch (action) {
4157 case MEM_GOING_ONLINE:
4158 ret = slab_mem_going_online_callback(arg);
4159 break;
4160 case MEM_GOING_OFFLINE:
4161 ret = slab_mem_going_offline_callback(arg);
4162 break;
4163 case MEM_OFFLINE:
4164 case MEM_CANCEL_ONLINE:
4165 slab_mem_offline_callback(arg);
4166 break;
4167 case MEM_ONLINE:
4168 case MEM_CANCEL_OFFLINE:
4169 break;
4170 }
dc19f9db
KH
4171 if (ret)
4172 ret = notifier_from_errno(ret);
4173 else
4174 ret = NOTIFY_OK;
b9049e23
YG
4175 return ret;
4176}
4177
3ac38faa
AM
4178static struct notifier_block slab_memory_callback_nb = {
4179 .notifier_call = slab_memory_callback,
4180 .priority = SLAB_CALLBACK_PRI,
4181};
b9049e23 4182
81819f0f
CL
4183/********************************************************************
4184 * Basic setup of slabs
4185 *******************************************************************/
4186
51df1142
CL
4187/*
4188 * Used for early kmem_cache structures that were allocated using
dffb4d60
CL
4189 * the page allocator. Allocate them properly then fix up the pointers
4190 * that may be pointing to the wrong kmem_cache structure.
51df1142
CL
4191 */
4192
dffb4d60 4193static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
51df1142
CL
4194{
4195 int node;
dffb4d60 4196 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
fa45dc25 4197 struct kmem_cache_node *n;
51df1142 4198
dffb4d60 4199 memcpy(s, static_cache, kmem_cache->object_size);
51df1142 4200
7d557b3c
GC
4201 /*
4202 * This runs very early, and only the boot processor is supposed to be
4203 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4204 * IPIs around.
4205 */
4206 __flush_cpu_slab(s, smp_processor_id());
fa45dc25 4207 for_each_kmem_cache_node(s, node, n) {
51df1142
CL
4208 struct page *p;
4209
916ac052 4210 list_for_each_entry(p, &n->partial, slab_list)
fa45dc25 4211 p->slab_cache = s;
51df1142 4212
607bf324 4213#ifdef CONFIG_SLUB_DEBUG
916ac052 4214 list_for_each_entry(p, &n->full, slab_list)
fa45dc25 4215 p->slab_cache = s;
51df1142 4216#endif
51df1142 4217 }
f7ce3190 4218 slab_init_memcg_params(s);
dffb4d60 4219 list_add(&s->list, &slab_caches);
c03914b7 4220 memcg_link_cache(s, NULL);
dffb4d60 4221 return s;
51df1142
CL
4222}
4223
81819f0f
CL
4224void __init kmem_cache_init(void)
4225{
dffb4d60
CL
4226 static __initdata struct kmem_cache boot_kmem_cache,
4227 boot_kmem_cache_node;
51df1142 4228
fc8d8620
SG
4229 if (debug_guardpage_minorder())
4230 slub_max_order = 0;
4231
dffb4d60
CL
4232 kmem_cache_node = &boot_kmem_cache_node;
4233 kmem_cache = &boot_kmem_cache;
51df1142 4234
dffb4d60 4235 create_boot_cache(kmem_cache_node, "kmem_cache_node",
8eb8284b 4236 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
b9049e23 4237
3ac38faa 4238 register_hotmemory_notifier(&slab_memory_callback_nb);
81819f0f
CL
4239
4240 /* Able to allocate the per node structures */
4241 slab_state = PARTIAL;
4242
dffb4d60
CL
4243 create_boot_cache(kmem_cache, "kmem_cache",
4244 offsetof(struct kmem_cache, node) +
4245 nr_node_ids * sizeof(struct kmem_cache_node *),
8eb8284b 4246 SLAB_HWCACHE_ALIGN, 0, 0);
8a13a4cc 4247
dffb4d60 4248 kmem_cache = bootstrap(&boot_kmem_cache);
dffb4d60 4249 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
51df1142
CL
4250
4251 /* Now we can use the kmem_cache to allocate kmalloc slabs */
34cc6990 4252 setup_kmalloc_cache_index_table();
f97d5f63 4253 create_kmalloc_caches(0);
81819f0f 4254
210e7a43
TG
4255 /* Setup random freelists for each cache */
4256 init_freelist_randomization();
4257
a96a87bf
SAS
4258 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4259 slub_cpu_dead);
81819f0f 4260
b9726c26 4261 pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
f97d5f63 4262 cache_line_size(),
81819f0f
CL
4263 slub_min_order, slub_max_order, slub_min_objects,
4264 nr_cpu_ids, nr_node_ids);
4265}
4266
7e85ee0c
PE
4267void __init kmem_cache_init_late(void)
4268{
7e85ee0c
PE
4269}
4270
2633d7a0 4271struct kmem_cache *
f4957d5b 4272__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
d50112ed 4273 slab_flags_t flags, void (*ctor)(void *))
81819f0f 4274{
426589f5 4275 struct kmem_cache *s, *c;
81819f0f 4276
a44cb944 4277 s = find_mergeable(size, align, flags, name, ctor);
81819f0f
CL
4278 if (s) {
4279 s->refcount++;
84d0ddd6 4280
81819f0f
CL
4281 /*
4282 * Adjust the object sizes so that we clear
4283 * the complete object on kzalloc.
4284 */
1b473f29 4285 s->object_size = max(s->object_size, size);
52ee6d74 4286 s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
6446faa2 4287
426589f5 4288 for_each_memcg_cache(c, s) {
84d0ddd6 4289 c->object_size = s->object_size;
52ee6d74 4290 c->inuse = max(c->inuse, ALIGN(size, sizeof(void *)));
84d0ddd6
VD
4291 }
4292
7b8f3b66 4293 if (sysfs_slab_alias(s, name)) {
7b8f3b66 4294 s->refcount--;
cbb79694 4295 s = NULL;
7b8f3b66 4296 }
a0e1d1be 4297 }
6446faa2 4298
cbb79694
CL
4299 return s;
4300}
84c1cf62 4301
d50112ed 4302int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
cbb79694 4303{
aac3a166
PE
4304 int err;
4305
4306 err = kmem_cache_open(s, flags);
4307 if (err)
4308 return err;
20cea968 4309
45530c44
CL
4310 /* Mutex is not taken during early boot */
4311 if (slab_state <= UP)
4312 return 0;
4313
107dab5c 4314 memcg_propagate_slab_attrs(s);
aac3a166 4315 err = sysfs_slab_add(s);
aac3a166 4316 if (err)
52b4b950 4317 __kmem_cache_release(s);
20cea968 4318
aac3a166 4319 return err;
81819f0f 4320}
81819f0f 4321
ce71e27c 4322void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
81819f0f 4323{
aadb4bc4 4324 struct kmem_cache *s;
94b528d0 4325 void *ret;
aadb4bc4 4326
95a05b42 4327 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
eada35ef
PE
4328 return kmalloc_large(size, gfpflags);
4329
2c59dd65 4330 s = kmalloc_slab(size, gfpflags);
81819f0f 4331
2408c550 4332 if (unlikely(ZERO_OR_NULL_PTR(s)))
6cb8f913 4333 return s;
81819f0f 4334
2b847c3c 4335 ret = slab_alloc(s, gfpflags, caller);
94b528d0 4336
25985edc 4337 /* Honor the call site pointer we received. */
ca2b84cb 4338 trace_kmalloc(caller, ret, size, s->size, gfpflags);
94b528d0
EGM
4339
4340 return ret;
81819f0f
CL
4341}
4342
5d1f57e4 4343#ifdef CONFIG_NUMA
81819f0f 4344void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
ce71e27c 4345 int node, unsigned long caller)
81819f0f 4346{
aadb4bc4 4347 struct kmem_cache *s;
94b528d0 4348 void *ret;
aadb4bc4 4349
95a05b42 4350 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
d3e14aa3
XF
4351 ret = kmalloc_large_node(size, gfpflags, node);
4352
4353 trace_kmalloc_node(caller, ret,
4354 size, PAGE_SIZE << get_order(size),
4355 gfpflags, node);
4356
4357 return ret;
4358 }
eada35ef 4359
2c59dd65 4360 s = kmalloc_slab(size, gfpflags);
81819f0f 4361
2408c550 4362 if (unlikely(ZERO_OR_NULL_PTR(s)))
6cb8f913 4363 return s;
81819f0f 4364
2b847c3c 4365 ret = slab_alloc_node(s, gfpflags, node, caller);
94b528d0 4366
25985edc 4367 /* Honor the call site pointer we received. */
ca2b84cb 4368 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
94b528d0
EGM
4369
4370 return ret;
81819f0f 4371}
5d1f57e4 4372#endif
81819f0f 4373
ab4d5ed5 4374#ifdef CONFIG_SYSFS
205ab99d
CL
4375static int count_inuse(struct page *page)
4376{
4377 return page->inuse;
4378}
4379
4380static int count_total(struct page *page)
4381{
4382 return page->objects;
4383}
ab4d5ed5 4384#endif
205ab99d 4385
ab4d5ed5 4386#ifdef CONFIG_SLUB_DEBUG
434e245d
CL
4387static int validate_slab(struct kmem_cache *s, struct page *page,
4388 unsigned long *map)
53e15af0
CL
4389{
4390 void *p;
a973e9dd 4391 void *addr = page_address(page);
53e15af0
CL
4392
4393 if (!check_slab(s, page) ||
4394 !on_freelist(s, page, NULL))
4395 return 0;
4396
4397 /* Now we know that a valid freelist exists */
39b26464 4398 bitmap_zero(map, page->objects);
53e15af0 4399
5f80b13a
CL
4400 get_map(s, page, map);
4401 for_each_object(p, s, addr, page->objects) {
4402 if (test_bit(slab_index(p, s, addr), map))
4403 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4404 return 0;
53e15af0
CL
4405 }
4406
224a88be 4407 for_each_object(p, s, addr, page->objects)
7656c72b 4408 if (!test_bit(slab_index(p, s, addr), map))
37d57443 4409 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
53e15af0
CL
4410 return 0;
4411 return 1;
4412}
4413
434e245d
CL
4414static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4415 unsigned long *map)
53e15af0 4416{
881db7fb
CL
4417 slab_lock(page);
4418 validate_slab(s, page, map);
4419 slab_unlock(page);
53e15af0
CL
4420}
4421
434e245d
CL
4422static int validate_slab_node(struct kmem_cache *s,
4423 struct kmem_cache_node *n, unsigned long *map)
53e15af0
CL
4424{
4425 unsigned long count = 0;
4426 struct page *page;
4427 unsigned long flags;
4428
4429 spin_lock_irqsave(&n->list_lock, flags);
4430
916ac052 4431 list_for_each_entry(page, &n->partial, slab_list) {
434e245d 4432 validate_slab_slab(s, page, map);
53e15af0
CL
4433 count++;
4434 }
4435 if (count != n->nr_partial)
f9f58285
FF
4436 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4437 s->name, count, n->nr_partial);
53e15af0
CL
4438
4439 if (!(s->flags & SLAB_STORE_USER))
4440 goto out;
4441
916ac052 4442 list_for_each_entry(page, &n->full, slab_list) {
434e245d 4443 validate_slab_slab(s, page, map);
53e15af0
CL
4444 count++;
4445 }
4446 if (count != atomic_long_read(&n->nr_slabs))
f9f58285
FF
4447 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4448 s->name, count, atomic_long_read(&n->nr_slabs));
53e15af0
CL
4449
4450out:
4451 spin_unlock_irqrestore(&n->list_lock, flags);
4452 return count;
4453}
4454
434e245d 4455static long validate_slab_cache(struct kmem_cache *s)
53e15af0
CL
4456{
4457 int node;
4458 unsigned long count = 0;
fa45dc25 4459 struct kmem_cache_node *n;
0684e652 4460 unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
434e245d
CL
4461
4462 if (!map)
4463 return -ENOMEM;
53e15af0
CL
4464
4465 flush_all(s);
fa45dc25 4466 for_each_kmem_cache_node(s, node, n)
434e245d 4467 count += validate_slab_node(s, n, map);
0684e652 4468 bitmap_free(map);
53e15af0
CL
4469 return count;
4470}
88a420e4 4471/*
672bba3a 4472 * Generate lists of code addresses where slabcache objects are allocated
88a420e4
CL
4473 * and freed.
4474 */
4475
4476struct location {
4477 unsigned long count;
ce71e27c 4478 unsigned long addr;
45edfa58
CL
4479 long long sum_time;
4480 long min_time;
4481 long max_time;
4482 long min_pid;
4483 long max_pid;
174596a0 4484 DECLARE_BITMAP(cpus, NR_CPUS);
45edfa58 4485 nodemask_t nodes;
88a420e4
CL
4486};
4487
4488struct loc_track {
4489 unsigned long max;
4490 unsigned long count;
4491 struct location *loc;
4492};
4493
4494static void free_loc_track(struct loc_track *t)
4495{
4496 if (t->max)
4497 free_pages((unsigned long)t->loc,
4498 get_order(sizeof(struct location) * t->max));
4499}
4500
68dff6a9 4501static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
88a420e4
CL
4502{
4503 struct location *l;
4504 int order;
4505
88a420e4
CL
4506 order = get_order(sizeof(struct location) * max);
4507
68dff6a9 4508 l = (void *)__get_free_pages(flags, order);
88a420e4
CL
4509 if (!l)
4510 return 0;
4511
4512 if (t->count) {
4513 memcpy(l, t->loc, sizeof(struct location) * t->count);
4514 free_loc_track(t);
4515 }
4516 t->max = max;
4517 t->loc = l;
4518 return 1;
4519}
4520
4521static int add_location(struct loc_track *t, struct kmem_cache *s,
45edfa58 4522 const struct track *track)
88a420e4
CL
4523{
4524 long start, end, pos;
4525 struct location *l;
ce71e27c 4526 unsigned long caddr;
45edfa58 4527 unsigned long age = jiffies - track->when;
88a420e4
CL
4528
4529 start = -1;
4530 end = t->count;
4531
4532 for ( ; ; ) {
4533 pos = start + (end - start + 1) / 2;
4534
4535 /*
4536 * There is nothing at "end". If we end up there
4537 * we need to add something to before end.
4538 */
4539 if (pos == end)
4540 break;
4541
4542 caddr = t->loc[pos].addr;
45edfa58
CL
4543 if (track->addr == caddr) {
4544
4545 l = &t->loc[pos];
4546 l->count++;
4547 if (track->when) {
4548 l->sum_time += age;
4549 if (age < l->min_time)
4550 l->min_time = age;
4551 if (age > l->max_time)
4552 l->max_time = age;
4553
4554 if (track->pid < l->min_pid)
4555 l->min_pid = track->pid;
4556 if (track->pid > l->max_pid)
4557 l->max_pid = track->pid;
4558
174596a0
RR
4559 cpumask_set_cpu(track->cpu,
4560 to_cpumask(l->cpus));
45edfa58
CL
4561 }
4562 node_set(page_to_nid(virt_to_page(track)), l->nodes);
88a420e4
CL
4563 return 1;
4564 }
4565
45edfa58 4566 if (track->addr < caddr)
88a420e4
CL
4567 end = pos;
4568 else
4569 start = pos;
4570 }
4571
4572 /*
672bba3a 4573 * Not found. Insert new tracking element.
88a420e4 4574 */
68dff6a9 4575 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
88a420e4
CL
4576 return 0;
4577
4578 l = t->loc + pos;
4579 if (pos < t->count)
4580 memmove(l + 1, l,
4581 (t->count - pos) * sizeof(struct location));
4582 t->count++;
4583 l->count = 1;
45edfa58
CL
4584 l->addr = track->addr;
4585 l->sum_time = age;
4586 l->min_time = age;
4587 l->max_time = age;
4588 l->min_pid = track->pid;
4589 l->max_pid = track->pid;
174596a0
RR
4590 cpumask_clear(to_cpumask(l->cpus));
4591 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
45edfa58
CL
4592 nodes_clear(l->nodes);
4593 node_set(page_to_nid(virt_to_page(track)), l->nodes);
88a420e4
CL
4594 return 1;
4595}
4596
4597static void process_slab(struct loc_track *t, struct kmem_cache *s,
bbd7d57b 4598 struct page *page, enum track_item alloc,
a5dd5c11 4599 unsigned long *map)
88a420e4 4600{
a973e9dd 4601 void *addr = page_address(page);
88a420e4
CL
4602 void *p;
4603
39b26464 4604 bitmap_zero(map, page->objects);
5f80b13a 4605 get_map(s, page, map);
88a420e4 4606
224a88be 4607 for_each_object(p, s, addr, page->objects)
45edfa58
CL
4608 if (!test_bit(slab_index(p, s, addr), map))
4609 add_location(t, s, get_track(s, p, alloc));
88a420e4
CL
4610}
4611
4612static int list_locations(struct kmem_cache *s, char *buf,
4613 enum track_item alloc)
4614{
e374d483 4615 int len = 0;
88a420e4 4616 unsigned long i;
68dff6a9 4617 struct loc_track t = { 0, 0, NULL };
88a420e4 4618 int node;
fa45dc25 4619 struct kmem_cache_node *n;
0684e652 4620 unsigned long *map = bitmap_alloc(oo_objects(s->max), GFP_KERNEL);
88a420e4 4621
bbd7d57b 4622 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
0ee931c4 4623 GFP_KERNEL)) {
0684e652 4624 bitmap_free(map);
68dff6a9 4625 return sprintf(buf, "Out of memory\n");
bbd7d57b 4626 }
88a420e4
CL
4627 /* Push back cpu slabs */
4628 flush_all(s);
4629
fa45dc25 4630 for_each_kmem_cache_node(s, node, n) {
88a420e4
CL
4631 unsigned long flags;
4632 struct page *page;
4633
9e86943b 4634 if (!atomic_long_read(&n->nr_slabs))
88a420e4
CL
4635 continue;
4636
4637 spin_lock_irqsave(&n->list_lock, flags);
916ac052 4638 list_for_each_entry(page, &n->partial, slab_list)
bbd7d57b 4639 process_slab(&t, s, page, alloc, map);
916ac052 4640 list_for_each_entry(page, &n->full, slab_list)
bbd7d57b 4641 process_slab(&t, s, page, alloc, map);
88a420e4
CL
4642 spin_unlock_irqrestore(&n->list_lock, flags);
4643 }
4644
4645 for (i = 0; i < t.count; i++) {
45edfa58 4646 struct location *l = &t.loc[i];
88a420e4 4647
9c246247 4648 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
88a420e4 4649 break;
e374d483 4650 len += sprintf(buf + len, "%7ld ", l->count);
45edfa58
CL
4651
4652 if (l->addr)
62c70bce 4653 len += sprintf(buf + len, "%pS", (void *)l->addr);
88a420e4 4654 else
e374d483 4655 len += sprintf(buf + len, "<not-available>");
45edfa58
CL
4656
4657 if (l->sum_time != l->min_time) {
e374d483 4658 len += sprintf(buf + len, " age=%ld/%ld/%ld",
f8bd2258
RZ
4659 l->min_time,
4660 (long)div_u64(l->sum_time, l->count),
4661 l->max_time);
45edfa58 4662 } else
e374d483 4663 len += sprintf(buf + len, " age=%ld",
45edfa58
CL
4664 l->min_time);
4665
4666 if (l->min_pid != l->max_pid)
e374d483 4667 len += sprintf(buf + len, " pid=%ld-%ld",
45edfa58
CL
4668 l->min_pid, l->max_pid);
4669 else
e374d483 4670 len += sprintf(buf + len, " pid=%ld",
45edfa58
CL
4671 l->min_pid);
4672
174596a0
RR
4673 if (num_online_cpus() > 1 &&
4674 !cpumask_empty(to_cpumask(l->cpus)) &&
5024c1d7
TH
4675 len < PAGE_SIZE - 60)
4676 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4677 " cpus=%*pbl",
4678 cpumask_pr_args(to_cpumask(l->cpus)));
45edfa58 4679
62bc62a8 4680 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
5024c1d7
TH
4681 len < PAGE_SIZE - 60)
4682 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4683 " nodes=%*pbl",
4684 nodemask_pr_args(&l->nodes));
45edfa58 4685
e374d483 4686 len += sprintf(buf + len, "\n");
88a420e4
CL
4687 }
4688
4689 free_loc_track(&t);
0684e652 4690 bitmap_free(map);
88a420e4 4691 if (!t.count)
e374d483
HH
4692 len += sprintf(buf, "No data\n");
4693 return len;
88a420e4 4694}
6dfd1b65 4695#endif /* CONFIG_SLUB_DEBUG */
88a420e4 4696
a5a84755 4697#ifdef SLUB_RESILIENCY_TEST
c07b8183 4698static void __init resiliency_test(void)
a5a84755
CL
4699{
4700 u8 *p;
cc252eae 4701 int type = KMALLOC_NORMAL;
a5a84755 4702
95a05b42 4703 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
a5a84755 4704
f9f58285
FF
4705 pr_err("SLUB resiliency testing\n");
4706 pr_err("-----------------------\n");
4707 pr_err("A. Corruption after allocation\n");
a5a84755
CL
4708
4709 p = kzalloc(16, GFP_KERNEL);
4710 p[16] = 0x12;
f9f58285
FF
4711 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4712 p + 16);
a5a84755 4713
cc252eae 4714 validate_slab_cache(kmalloc_caches[type][4]);
a5a84755
CL
4715
4716 /* Hmmm... The next two are dangerous */
4717 p = kzalloc(32, GFP_KERNEL);
4718 p[32 + sizeof(void *)] = 0x34;
f9f58285
FF
4719 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4720 p);
4721 pr_err("If allocated object is overwritten then not detectable\n\n");
a5a84755 4722
cc252eae 4723 validate_slab_cache(kmalloc_caches[type][5]);
a5a84755
CL
4724 p = kzalloc(64, GFP_KERNEL);
4725 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4726 *p = 0x56;
f9f58285
FF
4727 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4728 p);
4729 pr_err("If allocated object is overwritten then not detectable\n\n");
cc252eae 4730 validate_slab_cache(kmalloc_caches[type][6]);
a5a84755 4731
f9f58285 4732 pr_err("\nB. Corruption after free\n");
a5a84755
CL
4733 p = kzalloc(128, GFP_KERNEL);
4734 kfree(p);
4735 *p = 0x78;
f9f58285 4736 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
cc252eae 4737 validate_slab_cache(kmalloc_caches[type][7]);
a5a84755
CL
4738
4739 p = kzalloc(256, GFP_KERNEL);
4740 kfree(p);
4741 p[50] = 0x9a;
f9f58285 4742 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
cc252eae 4743 validate_slab_cache(kmalloc_caches[type][8]);
a5a84755
CL
4744
4745 p = kzalloc(512, GFP_KERNEL);
4746 kfree(p);
4747 p[512] = 0xab;
f9f58285 4748 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
cc252eae 4749 validate_slab_cache(kmalloc_caches[type][9]);
a5a84755
CL
4750}
4751#else
4752#ifdef CONFIG_SYSFS
4753static void resiliency_test(void) {};
4754#endif
6dfd1b65 4755#endif /* SLUB_RESILIENCY_TEST */
a5a84755 4756
ab4d5ed5 4757#ifdef CONFIG_SYSFS
81819f0f 4758enum slab_stat_type {
205ab99d
CL
4759 SL_ALL, /* All slabs */
4760 SL_PARTIAL, /* Only partially allocated slabs */
4761 SL_CPU, /* Only slabs used for cpu caches */
4762 SL_OBJECTS, /* Determine allocated objects not slabs */
4763 SL_TOTAL /* Determine object capacity not slabs */
81819f0f
CL
4764};
4765
205ab99d 4766#define SO_ALL (1 << SL_ALL)
81819f0f
CL
4767#define SO_PARTIAL (1 << SL_PARTIAL)
4768#define SO_CPU (1 << SL_CPU)
4769#define SO_OBJECTS (1 << SL_OBJECTS)
205ab99d 4770#define SO_TOTAL (1 << SL_TOTAL)
81819f0f 4771
1663f26d
TH
4772#ifdef CONFIG_MEMCG
4773static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4774
4775static int __init setup_slub_memcg_sysfs(char *str)
4776{
4777 int v;
4778
4779 if (get_option(&str, &v) > 0)
4780 memcg_sysfs_enabled = v;
4781
4782 return 1;
4783}
4784
4785__setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4786#endif
4787
62e5c4b4
CG
4788static ssize_t show_slab_objects(struct kmem_cache *s,
4789 char *buf, unsigned long flags)
81819f0f
CL
4790{
4791 unsigned long total = 0;
81819f0f
CL
4792 int node;
4793 int x;
4794 unsigned long *nodes;
81819f0f 4795
6396bb22 4796 nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
62e5c4b4
CG
4797 if (!nodes)
4798 return -ENOMEM;
81819f0f 4799
205ab99d
CL
4800 if (flags & SO_CPU) {
4801 int cpu;
81819f0f 4802
205ab99d 4803 for_each_possible_cpu(cpu) {
d0e0ac97
CG
4804 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4805 cpu);
ec3ab083 4806 int node;
49e22585 4807 struct page *page;
dfb4f096 4808
4db0c3c2 4809 page = READ_ONCE(c->page);
ec3ab083
CL
4810 if (!page)
4811 continue;
205ab99d 4812
ec3ab083
CL
4813 node = page_to_nid(page);
4814 if (flags & SO_TOTAL)
4815 x = page->objects;
4816 else if (flags & SO_OBJECTS)
4817 x = page->inuse;
4818 else
4819 x = 1;
49e22585 4820
ec3ab083
CL
4821 total += x;
4822 nodes[node] += x;
4823
a93cf07b 4824 page = slub_percpu_partial_read_once(c);
49e22585 4825 if (page) {
8afb1474
LZ
4826 node = page_to_nid(page);
4827 if (flags & SO_TOTAL)
4828 WARN_ON_ONCE(1);
4829 else if (flags & SO_OBJECTS)
4830 WARN_ON_ONCE(1);
4831 else
4832 x = page->pages;
bc6697d8
ED
4833 total += x;
4834 nodes[node] += x;
49e22585 4835 }
81819f0f
CL
4836 }
4837 }
4838
bfc8c901 4839 get_online_mems();
ab4d5ed5 4840#ifdef CONFIG_SLUB_DEBUG
205ab99d 4841 if (flags & SO_ALL) {
fa45dc25
CL
4842 struct kmem_cache_node *n;
4843
4844 for_each_kmem_cache_node(s, node, n) {
205ab99d 4845
d0e0ac97
CG
4846 if (flags & SO_TOTAL)
4847 x = atomic_long_read(&n->total_objects);
4848 else if (flags & SO_OBJECTS)
4849 x = atomic_long_read(&n->total_objects) -
4850 count_partial(n, count_free);
81819f0f 4851 else
205ab99d 4852 x = atomic_long_read(&n->nr_slabs);
81819f0f
CL
4853 total += x;
4854 nodes[node] += x;
4855 }
4856
ab4d5ed5
CL
4857 } else
4858#endif
4859 if (flags & SO_PARTIAL) {
fa45dc25 4860 struct kmem_cache_node *n;
81819f0f 4861
fa45dc25 4862 for_each_kmem_cache_node(s, node, n) {
205ab99d
CL
4863 if (flags & SO_TOTAL)
4864 x = count_partial(n, count_total);
4865 else if (flags & SO_OBJECTS)
4866 x = count_partial(n, count_inuse);
81819f0f 4867 else
205ab99d 4868 x = n->nr_partial;
81819f0f
CL
4869 total += x;
4870 nodes[node] += x;
4871 }
4872 }
81819f0f
CL
4873 x = sprintf(buf, "%lu", total);
4874#ifdef CONFIG_NUMA
fa45dc25 4875 for (node = 0; node < nr_node_ids; node++)
81819f0f
CL
4876 if (nodes[node])
4877 x += sprintf(buf + x, " N%d=%lu",
4878 node, nodes[node]);
4879#endif
bfc8c901 4880 put_online_mems();
81819f0f
CL
4881 kfree(nodes);
4882 return x + sprintf(buf + x, "\n");
4883}
4884
ab4d5ed5 4885#ifdef CONFIG_SLUB_DEBUG
81819f0f
CL
4886static int any_slab_objects(struct kmem_cache *s)
4887{
4888 int node;
fa45dc25 4889 struct kmem_cache_node *n;
81819f0f 4890
fa45dc25 4891 for_each_kmem_cache_node(s, node, n)
4ea33e2d 4892 if (atomic_long_read(&n->total_objects))
81819f0f 4893 return 1;
fa45dc25 4894
81819f0f
CL
4895 return 0;
4896}
ab4d5ed5 4897#endif
81819f0f
CL
4898
4899#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
497888cf 4900#define to_slab(n) container_of(n, struct kmem_cache, kobj)
81819f0f
CL
4901
4902struct slab_attribute {
4903 struct attribute attr;
4904 ssize_t (*show)(struct kmem_cache *s, char *buf);
4905 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4906};
4907
4908#define SLAB_ATTR_RO(_name) \
ab067e99
VK
4909 static struct slab_attribute _name##_attr = \
4910 __ATTR(_name, 0400, _name##_show, NULL)
81819f0f
CL
4911
4912#define SLAB_ATTR(_name) \
4913 static struct slab_attribute _name##_attr = \
ab067e99 4914 __ATTR(_name, 0600, _name##_show, _name##_store)
81819f0f 4915
81819f0f
CL
4916static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4917{
44065b2e 4918 return sprintf(buf, "%u\n", s->size);
81819f0f
CL
4919}
4920SLAB_ATTR_RO(slab_size);
4921
4922static ssize_t align_show(struct kmem_cache *s, char *buf)
4923{
3a3791ec 4924 return sprintf(buf, "%u\n", s->align);
81819f0f
CL
4925}
4926SLAB_ATTR_RO(align);
4927
4928static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4929{
1b473f29 4930 return sprintf(buf, "%u\n", s->object_size);
81819f0f
CL
4931}
4932SLAB_ATTR_RO(object_size);
4933
4934static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4935{
19af27af 4936 return sprintf(buf, "%u\n", oo_objects(s->oo));
81819f0f
CL
4937}
4938SLAB_ATTR_RO(objs_per_slab);
4939
06b285dc
CL
4940static ssize_t order_store(struct kmem_cache *s,
4941 const char *buf, size_t length)
4942{
19af27af 4943 unsigned int order;
0121c619
CL
4944 int err;
4945
19af27af 4946 err = kstrtouint(buf, 10, &order);
0121c619
CL
4947 if (err)
4948 return err;
06b285dc
CL
4949
4950 if (order > slub_max_order || order < slub_min_order)
4951 return -EINVAL;
4952
4953 calculate_sizes(s, order);
4954 return length;
4955}
4956
81819f0f
CL
4957static ssize_t order_show(struct kmem_cache *s, char *buf)
4958{
19af27af 4959 return sprintf(buf, "%u\n", oo_order(s->oo));
81819f0f 4960}
06b285dc 4961SLAB_ATTR(order);
81819f0f 4962
73d342b1
DR
4963static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4964{
4965 return sprintf(buf, "%lu\n", s->min_partial);
4966}
4967
4968static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4969 size_t length)
4970{
4971 unsigned long min;
4972 int err;
4973
3dbb95f7 4974 err = kstrtoul(buf, 10, &min);
73d342b1
DR
4975 if (err)
4976 return err;
4977
c0bdb232 4978 set_min_partial(s, min);
73d342b1
DR
4979 return length;
4980}
4981SLAB_ATTR(min_partial);
4982
49e22585
CL
4983static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4984{
e6d0e1dc 4985 return sprintf(buf, "%u\n", slub_cpu_partial(s));
49e22585
CL
4986}
4987
4988static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4989 size_t length)
4990{
e5d9998f 4991 unsigned int objects;
49e22585
CL
4992 int err;
4993
e5d9998f 4994 err = kstrtouint(buf, 10, &objects);
49e22585
CL
4995 if (err)
4996 return err;
345c905d 4997 if (objects && !kmem_cache_has_cpu_partial(s))
74ee4ef1 4998 return -EINVAL;
49e22585 4999
e6d0e1dc 5000 slub_set_cpu_partial(s, objects);
49e22585
CL
5001 flush_all(s);
5002 return length;
5003}
5004SLAB_ATTR(cpu_partial);
5005
81819f0f
CL
5006static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5007{
62c70bce
JP
5008 if (!s->ctor)
5009 return 0;
5010 return sprintf(buf, "%pS\n", s->ctor);
81819f0f
CL
5011}
5012SLAB_ATTR_RO(ctor);
5013
81819f0f
CL
5014static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5015{
4307c14f 5016 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
81819f0f
CL
5017}
5018SLAB_ATTR_RO(aliases);
5019
81819f0f
CL
5020static ssize_t partial_show(struct kmem_cache *s, char *buf)
5021{
d9acf4b7 5022 return show_slab_objects(s, buf, SO_PARTIAL);
81819f0f
CL
5023}
5024SLAB_ATTR_RO(partial);
5025
5026static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5027{
d9acf4b7 5028 return show_slab_objects(s, buf, SO_CPU);
81819f0f
CL
5029}
5030SLAB_ATTR_RO(cpu_slabs);
5031
5032static ssize_t objects_show(struct kmem_cache *s, char *buf)
5033{
205ab99d 5034 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
81819f0f
CL
5035}
5036SLAB_ATTR_RO(objects);
5037
205ab99d
CL
5038static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5039{
5040 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5041}
5042SLAB_ATTR_RO(objects_partial);
5043
49e22585
CL
5044static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5045{
5046 int objects = 0;
5047 int pages = 0;
5048 int cpu;
5049 int len;
5050
5051 for_each_online_cpu(cpu) {
a93cf07b
WY
5052 struct page *page;
5053
5054 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
49e22585
CL
5055
5056 if (page) {
5057 pages += page->pages;
5058 objects += page->pobjects;
5059 }
5060 }
5061
5062 len = sprintf(buf, "%d(%d)", objects, pages);
5063
5064#ifdef CONFIG_SMP
5065 for_each_online_cpu(cpu) {
a93cf07b
WY
5066 struct page *page;
5067
5068 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
49e22585
CL
5069
5070 if (page && len < PAGE_SIZE - 20)
5071 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5072 page->pobjects, page->pages);
5073 }
5074#endif
5075 return len + sprintf(buf + len, "\n");
5076}
5077SLAB_ATTR_RO(slabs_cpu_partial);
5078
a5a84755
CL
5079static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5080{
5081 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5082}
5083
5084static ssize_t reclaim_account_store(struct kmem_cache *s,
5085 const char *buf, size_t length)
5086{
5087 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5088 if (buf[0] == '1')
5089 s->flags |= SLAB_RECLAIM_ACCOUNT;
5090 return length;
5091}
5092SLAB_ATTR(reclaim_account);
5093
5094static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5095{
5096 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
5097}
5098SLAB_ATTR_RO(hwcache_align);
5099
5100#ifdef CONFIG_ZONE_DMA
5101static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5102{
5103 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5104}
5105SLAB_ATTR_RO(cache_dma);
5106#endif
5107
8eb8284b
DW
5108static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5109{
7bbdb81e 5110 return sprintf(buf, "%u\n", s->usersize);
8eb8284b
DW
5111}
5112SLAB_ATTR_RO(usersize);
5113
a5a84755
CL
5114static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5115{
5f0d5a3a 5116 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
a5a84755
CL
5117}
5118SLAB_ATTR_RO(destroy_by_rcu);
5119
ab4d5ed5 5120#ifdef CONFIG_SLUB_DEBUG
a5a84755
CL
5121static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5122{
5123 return show_slab_objects(s, buf, SO_ALL);
5124}
5125SLAB_ATTR_RO(slabs);
5126
205ab99d
CL
5127static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5128{
5129 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5130}
5131SLAB_ATTR_RO(total_objects);
5132
81819f0f
CL
5133static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5134{
becfda68 5135 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
81819f0f
CL
5136}
5137
5138static ssize_t sanity_checks_store(struct kmem_cache *s,
5139 const char *buf, size_t length)
5140{
becfda68 5141 s->flags &= ~SLAB_CONSISTENCY_CHECKS;
b789ef51
CL
5142 if (buf[0] == '1') {
5143 s->flags &= ~__CMPXCHG_DOUBLE;
becfda68 5144 s->flags |= SLAB_CONSISTENCY_CHECKS;
b789ef51 5145 }
81819f0f
CL
5146 return length;
5147}
5148SLAB_ATTR(sanity_checks);
5149
5150static ssize_t trace_show(struct kmem_cache *s, char *buf)
5151{
5152 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5153}
5154
5155static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5156 size_t length)
5157{
c9e16131
CL
5158 /*
5159 * Tracing a merged cache is going to give confusing results
5160 * as well as cause other issues like converting a mergeable
5161 * cache into an umergeable one.
5162 */
5163 if (s->refcount > 1)
5164 return -EINVAL;
5165
81819f0f 5166 s->flags &= ~SLAB_TRACE;
b789ef51
CL
5167 if (buf[0] == '1') {
5168 s->flags &= ~__CMPXCHG_DOUBLE;
81819f0f 5169 s->flags |= SLAB_TRACE;
b789ef51 5170 }
81819f0f
CL
5171 return length;
5172}
5173SLAB_ATTR(trace);
5174
81819f0f
CL
5175static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5176{
5177 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5178}
5179
5180static ssize_t red_zone_store(struct kmem_cache *s,
5181 const char *buf, size_t length)
5182{
5183 if (any_slab_objects(s))
5184 return -EBUSY;
5185
5186 s->flags &= ~SLAB_RED_ZONE;
b789ef51 5187 if (buf[0] == '1') {
81819f0f 5188 s->flags |= SLAB_RED_ZONE;
b789ef51 5189 }
06b285dc 5190 calculate_sizes(s, -1);
81819f0f
CL
5191 return length;
5192}
5193SLAB_ATTR(red_zone);
5194
5195static ssize_t poison_show(struct kmem_cache *s, char *buf)
5196{
5197 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5198}
5199
5200static ssize_t poison_store(struct kmem_cache *s,
5201 const char *buf, size_t length)
5202{
5203 if (any_slab_objects(s))
5204 return -EBUSY;
5205
5206 s->flags &= ~SLAB_POISON;
b789ef51 5207 if (buf[0] == '1') {
81819f0f 5208 s->flags |= SLAB_POISON;
b789ef51 5209 }
06b285dc 5210 calculate_sizes(s, -1);
81819f0f
CL
5211 return length;
5212}
5213SLAB_ATTR(poison);
5214
5215static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5216{
5217 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5218}
5219
5220static ssize_t store_user_store(struct kmem_cache *s,
5221 const char *buf, size_t length)
5222{
5223 if (any_slab_objects(s))
5224 return -EBUSY;
5225
5226 s->flags &= ~SLAB_STORE_USER;
b789ef51
CL
5227 if (buf[0] == '1') {
5228 s->flags &= ~__CMPXCHG_DOUBLE;
81819f0f 5229 s->flags |= SLAB_STORE_USER;
b789ef51 5230 }
06b285dc 5231 calculate_sizes(s, -1);
81819f0f
CL
5232 return length;
5233}
5234SLAB_ATTR(store_user);
5235
53e15af0
CL
5236static ssize_t validate_show(struct kmem_cache *s, char *buf)
5237{
5238 return 0;
5239}
5240
5241static ssize_t validate_store(struct kmem_cache *s,
5242 const char *buf, size_t length)
5243{
434e245d
CL
5244 int ret = -EINVAL;
5245
5246 if (buf[0] == '1') {
5247 ret = validate_slab_cache(s);
5248 if (ret >= 0)
5249 ret = length;
5250 }
5251 return ret;
53e15af0
CL
5252}
5253SLAB_ATTR(validate);
a5a84755
CL
5254
5255static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5256{
5257 if (!(s->flags & SLAB_STORE_USER))
5258 return -ENOSYS;
5259 return list_locations(s, buf, TRACK_ALLOC);
5260}
5261SLAB_ATTR_RO(alloc_calls);
5262
5263static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5264{
5265 if (!(s->flags & SLAB_STORE_USER))
5266 return -ENOSYS;
5267 return list_locations(s, buf, TRACK_FREE);
5268}
5269SLAB_ATTR_RO(free_calls);
5270#endif /* CONFIG_SLUB_DEBUG */
5271
5272#ifdef CONFIG_FAILSLAB
5273static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5274{
5275 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5276}
5277
5278static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5279 size_t length)
5280{
c9e16131
CL
5281 if (s->refcount > 1)
5282 return -EINVAL;
5283
a5a84755
CL
5284 s->flags &= ~SLAB_FAILSLAB;
5285 if (buf[0] == '1')
5286 s->flags |= SLAB_FAILSLAB;
5287 return length;
5288}
5289SLAB_ATTR(failslab);
ab4d5ed5 5290#endif
53e15af0 5291
2086d26a
CL
5292static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5293{
5294 return 0;
5295}
5296
5297static ssize_t shrink_store(struct kmem_cache *s,
5298 const char *buf, size_t length)
5299{
832f37f5
VD
5300 if (buf[0] == '1')
5301 kmem_cache_shrink(s);
5302 else
2086d26a
CL
5303 return -EINVAL;
5304 return length;
5305}
5306SLAB_ATTR(shrink);
5307
81819f0f 5308#ifdef CONFIG_NUMA
9824601e 5309static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
81819f0f 5310{
eb7235eb 5311 return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
81819f0f
CL
5312}
5313
9824601e 5314static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
81819f0f
CL
5315 const char *buf, size_t length)
5316{
eb7235eb 5317 unsigned int ratio;
0121c619
CL
5318 int err;
5319
eb7235eb 5320 err = kstrtouint(buf, 10, &ratio);
0121c619
CL
5321 if (err)
5322 return err;
eb7235eb
AD
5323 if (ratio > 100)
5324 return -ERANGE;
0121c619 5325
eb7235eb 5326 s->remote_node_defrag_ratio = ratio * 10;
81819f0f 5327
81819f0f
CL
5328 return length;
5329}
9824601e 5330SLAB_ATTR(remote_node_defrag_ratio);
81819f0f
CL
5331#endif
5332
8ff12cfc 5333#ifdef CONFIG_SLUB_STATS
8ff12cfc
CL
5334static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5335{
5336 unsigned long sum = 0;
5337 int cpu;
5338 int len;
6da2ec56 5339 int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
8ff12cfc
CL
5340
5341 if (!data)
5342 return -ENOMEM;
5343
5344 for_each_online_cpu(cpu) {
9dfc6e68 5345 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
8ff12cfc
CL
5346
5347 data[cpu] = x;
5348 sum += x;
5349 }
5350
5351 len = sprintf(buf, "%lu", sum);
5352
50ef37b9 5353#ifdef CONFIG_SMP
8ff12cfc
CL
5354 for_each_online_cpu(cpu) {
5355 if (data[cpu] && len < PAGE_SIZE - 20)
50ef37b9 5356 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
8ff12cfc 5357 }
50ef37b9 5358#endif
8ff12cfc
CL
5359 kfree(data);
5360 return len + sprintf(buf + len, "\n");
5361}
5362
78eb00cc
DR
5363static void clear_stat(struct kmem_cache *s, enum stat_item si)
5364{
5365 int cpu;
5366
5367 for_each_online_cpu(cpu)
9dfc6e68 5368 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
78eb00cc
DR
5369}
5370
8ff12cfc
CL
5371#define STAT_ATTR(si, text) \
5372static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5373{ \
5374 return show_stat(s, buf, si); \
5375} \
78eb00cc
DR
5376static ssize_t text##_store(struct kmem_cache *s, \
5377 const char *buf, size_t length) \
5378{ \
5379 if (buf[0] != '0') \
5380 return -EINVAL; \
5381 clear_stat(s, si); \
5382 return length; \
5383} \
5384SLAB_ATTR(text); \
8ff12cfc
CL
5385
5386STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5387STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5388STAT_ATTR(FREE_FASTPATH, free_fastpath);
5389STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5390STAT_ATTR(FREE_FROZEN, free_frozen);
5391STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5392STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5393STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5394STAT_ATTR(ALLOC_SLAB, alloc_slab);
5395STAT_ATTR(ALLOC_REFILL, alloc_refill);
e36a2652 5396STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
8ff12cfc
CL
5397STAT_ATTR(FREE_SLAB, free_slab);
5398STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5399STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5400STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5401STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5402STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5403STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
03e404af 5404STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
65c3376a 5405STAT_ATTR(ORDER_FALLBACK, order_fallback);
b789ef51
CL
5406STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5407STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
49e22585
CL
5408STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5409STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
8028dcea
AS
5410STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5411STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
6dfd1b65 5412#endif /* CONFIG_SLUB_STATS */
8ff12cfc 5413
06428780 5414static struct attribute *slab_attrs[] = {
81819f0f
CL
5415 &slab_size_attr.attr,
5416 &object_size_attr.attr,
5417 &objs_per_slab_attr.attr,
5418 &order_attr.attr,
73d342b1 5419 &min_partial_attr.attr,
49e22585 5420 &cpu_partial_attr.attr,
81819f0f 5421 &objects_attr.attr,
205ab99d 5422 &objects_partial_attr.attr,
81819f0f
CL
5423 &partial_attr.attr,
5424 &cpu_slabs_attr.attr,
5425 &ctor_attr.attr,
81819f0f
CL
5426 &aliases_attr.attr,
5427 &align_attr.attr,
81819f0f
CL
5428 &hwcache_align_attr.attr,
5429 &reclaim_account_attr.attr,
5430 &destroy_by_rcu_attr.attr,
a5a84755 5431 &shrink_attr.attr,
49e22585 5432 &slabs_cpu_partial_attr.attr,
ab4d5ed5 5433#ifdef CONFIG_SLUB_DEBUG
a5a84755
CL
5434 &total_objects_attr.attr,
5435 &slabs_attr.attr,
5436 &sanity_checks_attr.attr,
5437 &trace_attr.attr,
81819f0f
CL
5438 &red_zone_attr.attr,
5439 &poison_attr.attr,
5440 &store_user_attr.attr,
53e15af0 5441 &validate_attr.attr,
88a420e4
CL
5442 &alloc_calls_attr.attr,
5443 &free_calls_attr.attr,
ab4d5ed5 5444#endif
81819f0f
CL
5445#ifdef CONFIG_ZONE_DMA
5446 &cache_dma_attr.attr,
5447#endif
5448#ifdef CONFIG_NUMA
9824601e 5449 &remote_node_defrag_ratio_attr.attr,
8ff12cfc
CL
5450#endif
5451#ifdef CONFIG_SLUB_STATS
5452 &alloc_fastpath_attr.attr,
5453 &alloc_slowpath_attr.attr,
5454 &free_fastpath_attr.attr,
5455 &free_slowpath_attr.attr,
5456 &free_frozen_attr.attr,
5457 &free_add_partial_attr.attr,
5458 &free_remove_partial_attr.attr,
5459 &alloc_from_partial_attr.attr,
5460 &alloc_slab_attr.attr,
5461 &alloc_refill_attr.attr,
e36a2652 5462 &alloc_node_mismatch_attr.attr,
8ff12cfc
CL
5463 &free_slab_attr.attr,
5464 &cpuslab_flush_attr.attr,
5465 &deactivate_full_attr.attr,
5466 &deactivate_empty_attr.attr,
5467 &deactivate_to_head_attr.attr,
5468 &deactivate_to_tail_attr.attr,
5469 &deactivate_remote_frees_attr.attr,
03e404af 5470 &deactivate_bypass_attr.attr,
65c3376a 5471 &order_fallback_attr.attr,
b789ef51
CL
5472 &cmpxchg_double_fail_attr.attr,
5473 &cmpxchg_double_cpu_fail_attr.attr,
49e22585
CL
5474 &cpu_partial_alloc_attr.attr,
5475 &cpu_partial_free_attr.attr,
8028dcea
AS
5476 &cpu_partial_node_attr.attr,
5477 &cpu_partial_drain_attr.attr,
81819f0f 5478#endif
4c13dd3b
DM
5479#ifdef CONFIG_FAILSLAB
5480 &failslab_attr.attr,
5481#endif
8eb8284b 5482 &usersize_attr.attr,
4c13dd3b 5483
81819f0f
CL
5484 NULL
5485};
5486
1fdaaa23 5487static const struct attribute_group slab_attr_group = {
81819f0f
CL
5488 .attrs = slab_attrs,
5489};
5490
5491static ssize_t slab_attr_show(struct kobject *kobj,
5492 struct attribute *attr,
5493 char *buf)
5494{
5495 struct slab_attribute *attribute;
5496 struct kmem_cache *s;
5497 int err;
5498
5499 attribute = to_slab_attr(attr);
5500 s = to_slab(kobj);
5501
5502 if (!attribute->show)
5503 return -EIO;
5504
5505 err = attribute->show(s, buf);
5506
5507 return err;
5508}
5509
5510static ssize_t slab_attr_store(struct kobject *kobj,
5511 struct attribute *attr,
5512 const char *buf, size_t len)
5513{
5514 struct slab_attribute *attribute;
5515 struct kmem_cache *s;
5516 int err;
5517
5518 attribute = to_slab_attr(attr);
5519 s = to_slab(kobj);
5520
5521 if (!attribute->store)
5522 return -EIO;
5523
5524 err = attribute->store(s, buf, len);
127424c8 5525#ifdef CONFIG_MEMCG
107dab5c 5526 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
426589f5 5527 struct kmem_cache *c;
81819f0f 5528
107dab5c
GC
5529 mutex_lock(&slab_mutex);
5530 if (s->max_attr_size < len)
5531 s->max_attr_size = len;
5532
ebe945c2
GC
5533 /*
5534 * This is a best effort propagation, so this function's return
5535 * value will be determined by the parent cache only. This is
5536 * basically because not all attributes will have a well
5537 * defined semantics for rollbacks - most of the actions will
5538 * have permanent effects.
5539 *
5540 * Returning the error value of any of the children that fail
5541 * is not 100 % defined, in the sense that users seeing the
5542 * error code won't be able to know anything about the state of
5543 * the cache.
5544 *
5545 * Only returning the error code for the parent cache at least
5546 * has well defined semantics. The cache being written to
5547 * directly either failed or succeeded, in which case we loop
5548 * through the descendants with best-effort propagation.
5549 */
426589f5
VD
5550 for_each_memcg_cache(c, s)
5551 attribute->store(c, buf, len);
107dab5c
GC
5552 mutex_unlock(&slab_mutex);
5553 }
5554#endif
81819f0f
CL
5555 return err;
5556}
5557
107dab5c
GC
5558static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5559{
127424c8 5560#ifdef CONFIG_MEMCG
107dab5c
GC
5561 int i;
5562 char *buffer = NULL;
93030d83 5563 struct kmem_cache *root_cache;
107dab5c 5564
93030d83 5565 if (is_root_cache(s))
107dab5c
GC
5566 return;
5567
f7ce3190 5568 root_cache = s->memcg_params.root_cache;
93030d83 5569
107dab5c
GC
5570 /*
5571 * This mean this cache had no attribute written. Therefore, no point
5572 * in copying default values around
5573 */
93030d83 5574 if (!root_cache->max_attr_size)
107dab5c
GC
5575 return;
5576
5577 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5578 char mbuf[64];
5579 char *buf;
5580 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
478fe303 5581 ssize_t len;
107dab5c
GC
5582
5583 if (!attr || !attr->store || !attr->show)
5584 continue;
5585
5586 /*
5587 * It is really bad that we have to allocate here, so we will
5588 * do it only as a fallback. If we actually allocate, though,
5589 * we can just use the allocated buffer until the end.
5590 *
5591 * Most of the slub attributes will tend to be very small in
5592 * size, but sysfs allows buffers up to a page, so they can
5593 * theoretically happen.
5594 */
5595 if (buffer)
5596 buf = buffer;
93030d83 5597 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
107dab5c
GC
5598 buf = mbuf;
5599 else {
5600 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5601 if (WARN_ON(!buffer))
5602 continue;
5603 buf = buffer;
5604 }
5605
478fe303
TG
5606 len = attr->show(root_cache, buf);
5607 if (len > 0)
5608 attr->store(s, buf, len);
107dab5c
GC
5609 }
5610
5611 if (buffer)
5612 free_page((unsigned long)buffer);
6dfd1b65 5613#endif /* CONFIG_MEMCG */
107dab5c
GC
5614}
5615
41a21285
CL
5616static void kmem_cache_release(struct kobject *k)
5617{
5618 slab_kmem_cache_release(to_slab(k));
5619}
5620
52cf25d0 5621static const struct sysfs_ops slab_sysfs_ops = {
81819f0f
CL
5622 .show = slab_attr_show,
5623 .store = slab_attr_store,
5624};
5625
5626static struct kobj_type slab_ktype = {
5627 .sysfs_ops = &slab_sysfs_ops,
41a21285 5628 .release = kmem_cache_release,
81819f0f
CL
5629};
5630
5631static int uevent_filter(struct kset *kset, struct kobject *kobj)
5632{
5633 struct kobj_type *ktype = get_ktype(kobj);
5634
5635 if (ktype == &slab_ktype)
5636 return 1;
5637 return 0;
5638}
5639
9cd43611 5640static const struct kset_uevent_ops slab_uevent_ops = {
81819f0f
CL
5641 .filter = uevent_filter,
5642};
5643
27c3a314 5644static struct kset *slab_kset;
81819f0f 5645
9a41707b
VD
5646static inline struct kset *cache_kset(struct kmem_cache *s)
5647{
127424c8 5648#ifdef CONFIG_MEMCG
9a41707b 5649 if (!is_root_cache(s))
f7ce3190 5650 return s->memcg_params.root_cache->memcg_kset;
9a41707b
VD
5651#endif
5652 return slab_kset;
5653}
5654
81819f0f
CL
5655#define ID_STR_LENGTH 64
5656
5657/* Create a unique string id for a slab cache:
6446faa2
CL
5658 *
5659 * Format :[flags-]size
81819f0f
CL
5660 */
5661static char *create_unique_id(struct kmem_cache *s)
5662{
5663 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5664 char *p = name;
5665
5666 BUG_ON(!name);
5667
5668 *p++ = ':';
5669 /*
5670 * First flags affecting slabcache operations. We will only
5671 * get here for aliasable slabs so we do not need to support
5672 * too many flags. The flags here must cover all flags that
5673 * are matched during merging to guarantee that the id is
5674 * unique.
5675 */
5676 if (s->flags & SLAB_CACHE_DMA)
5677 *p++ = 'd';
6d6ea1e9
NB
5678 if (s->flags & SLAB_CACHE_DMA32)
5679 *p++ = 'D';
81819f0f
CL
5680 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5681 *p++ = 'a';
becfda68 5682 if (s->flags & SLAB_CONSISTENCY_CHECKS)
81819f0f 5683 *p++ = 'F';
230e9fc2
VD
5684 if (s->flags & SLAB_ACCOUNT)
5685 *p++ = 'A';
81819f0f
CL
5686 if (p != name + 1)
5687 *p++ = '-';
44065b2e 5688 p += sprintf(p, "%07u", s->size);
2633d7a0 5689
81819f0f
CL
5690 BUG_ON(p > name + ID_STR_LENGTH - 1);
5691 return name;
5692}
5693
3b7b3140
TH
5694static void sysfs_slab_remove_workfn(struct work_struct *work)
5695{
5696 struct kmem_cache *s =
5697 container_of(work, struct kmem_cache, kobj_remove_work);
5698
5699 if (!s->kobj.state_in_sysfs)
5700 /*
5701 * For a memcg cache, this may be called during
5702 * deactivation and again on shutdown. Remove only once.
5703 * A cache is never shut down before deactivation is
5704 * complete, so no need to worry about synchronization.
5705 */
f6ba4880 5706 goto out;
3b7b3140
TH
5707
5708#ifdef CONFIG_MEMCG
5709 kset_unregister(s->memcg_kset);
5710#endif
5711 kobject_uevent(&s->kobj, KOBJ_REMOVE);
f6ba4880 5712out:
3b7b3140
TH
5713 kobject_put(&s->kobj);
5714}
5715
81819f0f
CL
5716static int sysfs_slab_add(struct kmem_cache *s)
5717{
5718 int err;
5719 const char *name;
1663f26d 5720 struct kset *kset = cache_kset(s);
45530c44 5721 int unmergeable = slab_unmergeable(s);
81819f0f 5722
3b7b3140
TH
5723 INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5724
1663f26d
TH
5725 if (!kset) {
5726 kobject_init(&s->kobj, &slab_ktype);
5727 return 0;
5728 }
5729
11066386
MC
5730 if (!unmergeable && disable_higher_order_debug &&
5731 (slub_debug & DEBUG_METADATA_FLAGS))
5732 unmergeable = 1;
5733
81819f0f
CL
5734 if (unmergeable) {
5735 /*
5736 * Slabcache can never be merged so we can use the name proper.
5737 * This is typically the case for debug situations. In that
5738 * case we can catch duplicate names easily.
5739 */
27c3a314 5740 sysfs_remove_link(&slab_kset->kobj, s->name);
81819f0f
CL
5741 name = s->name;
5742 } else {
5743 /*
5744 * Create a unique name for the slab as a target
5745 * for the symlinks.
5746 */
5747 name = create_unique_id(s);
5748 }
5749
1663f26d 5750 s->kobj.kset = kset;
26e4f205 5751 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
54b6a731 5752 if (err)
80da026a 5753 goto out;
81819f0f
CL
5754
5755 err = sysfs_create_group(&s->kobj, &slab_attr_group);
54b6a731
DJ
5756 if (err)
5757 goto out_del_kobj;
9a41707b 5758
127424c8 5759#ifdef CONFIG_MEMCG
1663f26d 5760 if (is_root_cache(s) && memcg_sysfs_enabled) {
9a41707b
VD
5761 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5762 if (!s->memcg_kset) {
54b6a731
DJ
5763 err = -ENOMEM;
5764 goto out_del_kobj;
9a41707b
VD
5765 }
5766 }
5767#endif
5768
81819f0f
CL
5769 kobject_uevent(&s->kobj, KOBJ_ADD);
5770 if (!unmergeable) {
5771 /* Setup first alias */
5772 sysfs_slab_alias(s, s->name);
81819f0f 5773 }
54b6a731
DJ
5774out:
5775 if (!unmergeable)
5776 kfree(name);
5777 return err;
5778out_del_kobj:
5779 kobject_del(&s->kobj);
54b6a731 5780 goto out;
81819f0f
CL
5781}
5782
bf5eb3de 5783static void sysfs_slab_remove(struct kmem_cache *s)
81819f0f 5784{
97d06609 5785 if (slab_state < FULL)
2bce6485
CL
5786 /*
5787 * Sysfs has not been setup yet so no need to remove the
5788 * cache from sysfs.
5789 */
5790 return;
5791
3b7b3140
TH
5792 kobject_get(&s->kobj);
5793 schedule_work(&s->kobj_remove_work);
bf5eb3de
TH
5794}
5795
d50d82fa
MP
5796void sysfs_slab_unlink(struct kmem_cache *s)
5797{
5798 if (slab_state >= FULL)
5799 kobject_del(&s->kobj);
5800}
5801
bf5eb3de
TH
5802void sysfs_slab_release(struct kmem_cache *s)
5803{
5804 if (slab_state >= FULL)
5805 kobject_put(&s->kobj);
81819f0f
CL
5806}
5807
5808/*
5809 * Need to buffer aliases during bootup until sysfs becomes
9f6c708e 5810 * available lest we lose that information.
81819f0f
CL
5811 */
5812struct saved_alias {
5813 struct kmem_cache *s;
5814 const char *name;
5815 struct saved_alias *next;
5816};
5817
5af328a5 5818static struct saved_alias *alias_list;
81819f0f
CL
5819
5820static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5821{
5822 struct saved_alias *al;
5823
97d06609 5824 if (slab_state == FULL) {
81819f0f
CL
5825 /*
5826 * If we have a leftover link then remove it.
5827 */
27c3a314
GKH
5828 sysfs_remove_link(&slab_kset->kobj, name);
5829 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
81819f0f
CL
5830 }
5831
5832 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5833 if (!al)
5834 return -ENOMEM;
5835
5836 al->s = s;
5837 al->name = name;
5838 al->next = alias_list;
5839 alias_list = al;
5840 return 0;
5841}
5842
5843static int __init slab_sysfs_init(void)
5844{
5b95a4ac 5845 struct kmem_cache *s;
81819f0f
CL
5846 int err;
5847
18004c5d 5848 mutex_lock(&slab_mutex);
2bce6485 5849
0ff21e46 5850 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
27c3a314 5851 if (!slab_kset) {
18004c5d 5852 mutex_unlock(&slab_mutex);
f9f58285 5853 pr_err("Cannot register slab subsystem.\n");
81819f0f
CL
5854 return -ENOSYS;
5855 }
5856
97d06609 5857 slab_state = FULL;
26a7bd03 5858
5b95a4ac 5859 list_for_each_entry(s, &slab_caches, list) {
26a7bd03 5860 err = sysfs_slab_add(s);
5d540fb7 5861 if (err)
f9f58285
FF
5862 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5863 s->name);
26a7bd03 5864 }
81819f0f
CL
5865
5866 while (alias_list) {
5867 struct saved_alias *al = alias_list;
5868
5869 alias_list = alias_list->next;
5870 err = sysfs_slab_alias(al->s, al->name);
5d540fb7 5871 if (err)
f9f58285
FF
5872 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5873 al->name);
81819f0f
CL
5874 kfree(al);
5875 }
5876
18004c5d 5877 mutex_unlock(&slab_mutex);
81819f0f
CL
5878 resiliency_test();
5879 return 0;
5880}
5881
5882__initcall(slab_sysfs_init);
ab4d5ed5 5883#endif /* CONFIG_SYSFS */
57ed3eda
PE
5884
5885/*
5886 * The /proc/slabinfo ABI
5887 */
5b365771 5888#ifdef CONFIG_SLUB_DEBUG
0d7561c6 5889void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
57ed3eda 5890{
57ed3eda 5891 unsigned long nr_slabs = 0;
205ab99d
CL
5892 unsigned long nr_objs = 0;
5893 unsigned long nr_free = 0;
57ed3eda 5894 int node;
fa45dc25 5895 struct kmem_cache_node *n;
57ed3eda 5896
fa45dc25 5897 for_each_kmem_cache_node(s, node, n) {
c17fd13e
WL
5898 nr_slabs += node_nr_slabs(n);
5899 nr_objs += node_nr_objs(n);
205ab99d 5900 nr_free += count_partial(n, count_free);
57ed3eda
PE
5901 }
5902
0d7561c6
GC
5903 sinfo->active_objs = nr_objs - nr_free;
5904 sinfo->num_objs = nr_objs;
5905 sinfo->active_slabs = nr_slabs;
5906 sinfo->num_slabs = nr_slabs;
5907 sinfo->objects_per_slab = oo_objects(s->oo);
5908 sinfo->cache_order = oo_order(s->oo);
57ed3eda
PE
5909}
5910
0d7561c6 5911void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
7b3c3a50 5912{
7b3c3a50
AD
5913}
5914
b7454ad3
GC
5915ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5916 size_t count, loff_t *ppos)
7b3c3a50 5917{
b7454ad3 5918 return -EIO;
7b3c3a50 5919}
5b365771 5920#endif /* CONFIG_SLUB_DEBUG */