blob: 8addc535bcdc58794fe40e72a729e4589d44d2b6 [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05005 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07007 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -05009 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070010 */
11
12#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100013#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070014#include <linux/module.h>
15#include <linux/bit_spinlock.h>
16#include <linux/interrupt.h>
17#include <linux/bitops.h>
18#include <linux/slab.h>
Christoph Lameter97d06602012-07-06 15:25:11 -050019#include "slab.h"
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040020#include <linux/proc_fs.h>
Andrew Morton3ac38fa2013-04-29 15:08:06 -070021#include <linux/notifier.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/seq_file.h>
Andrey Ryabinina79316c2015-02-13 14:39:38 -080023#include <linux/kasan.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020024#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070025#include <linux/cpu.h>
26#include <linux/cpuset.h>
27#include <linux/mempolicy.h>
28#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070029#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070031#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070032#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090033#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030034#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060035#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080036#include <linux/memcontrol.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070037
Richard Kennedy4a923792010-10-21 10:29:19 +010038#include <trace/events/kmem.h>
39
Mel Gorman072bb0a2012-07-31 16:43:58 -070040#include "internal.h"
41
Christoph Lameter81819f02007-05-06 14:49:36 -070042/*
43 * Lock order:
Christoph Lameter18004c52012-07-06 15:25:12 -050044 * 1. slab_mutex (Global Mutex)
Christoph Lameter881db7f2011-06-01 12:25:53 -050045 * 2. node->list_lock
46 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070047 *
Christoph Lameter18004c52012-07-06 15:25:12 -050048 * slab_mutex
Christoph Lameter881db7f2011-06-01 12:25:53 -050049 *
Christoph Lameter18004c52012-07-06 15:25:12 -050050 * The role of the slab_mutex is to protect the list of all the slabs
Christoph Lameter881db7f2011-06-01 12:25:53 -050051 * and to synchronize major metadata changes to slab cache structures.
52 *
53 * The slab_lock is only used for debugging and on arches that do not
54 * have the ability to do a cmpxchg_double. It only protects the second
55 * double word in the page struct. Meaning
56 * A. page->freelist -> List of object free in a page
57 * B. page->counters -> Counters of objects
58 * C. page->frozen -> frozen state
59 *
60 * If a slab is frozen then it is exempt from list management. It is not
61 * on any list. The processor that froze the slab is the one who can
62 * perform list operations on the page. Other processors may put objects
63 * onto the freelist but the processor that froze the slab is the only
64 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070065 *
66 * The list_lock protects the partial and full list on each node and
67 * the partial slab counter. If taken then no new slabs may be added or
68 * removed from the lists nor make the number of partial slabs be modified.
69 * (Note that the total number of slabs is an atomic value that may be
70 * modified without taking the list lock).
71 *
72 * The list_lock is a centralized lock and thus we avoid taking it as
73 * much as possible. As long as SLUB does not have to handle partial
74 * slabs, operations can continue without any centralized lock. F.e.
75 * allocating a long series of objects that fill up slabs does not require
76 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070077 * Interrupts are disabled during allocation and deallocation in order to
78 * make the slab allocator safe to use in the context of an irq. In addition
79 * interrupts are disabled to ensure that the processor does not change
80 * while handling per_cpu slabs, due to kernel preemption.
81 *
82 * SLUB assigns one slab for allocation to each processor.
83 * Allocations only occur from these slabs called cpu slabs.
84 *
Christoph Lameter672bba32007-05-09 02:32:39 -070085 * Slabs with free elements are kept on a partial list and during regular
86 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070087 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070088 * We track full slabs for debugging purposes though because otherwise we
89 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070090 *
91 * Slabs are freed when they become empty. Teardown and setup is
92 * minimal so we rely on the page allocators per cpu caches for
93 * fast frees and allocs.
94 *
95 * Overloading of page flags that are otherwise used for LRU management.
96 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070097 * PageActive The slab is frozen and exempt from list processing.
98 * This means that the slab is dedicated to a purpose
99 * such as satisfying allocations for a specific
100 * processor. Objects may be freed in the slab while
101 * it is frozen but slab_free will then skip the usual
102 * list operations. It is up to the processor holding
103 * the slab to integrate the slab into the slab lists
104 * when the slab is no longer needed.
105 *
106 * One use of this flag is to mark slabs that are
107 * used for allocations. Then such a slab becomes a cpu
108 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700109 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700110 * free objects in addition to the regular freelist
111 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700112 *
113 * PageError Slab requires special handling due to debug
114 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700115 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700116 */
117
Christoph Lameteraf537b02010-07-09 14:07:14 -0500118static inline int kmem_cache_debug(struct kmem_cache *s)
119{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500123 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700124#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500125}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700126
Geert Uytterhoeven117d54d2016-08-04 15:31:55 -0700127void *fixup_red_left(struct kmem_cache *s, void *p)
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700128{
129 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
130 p += s->red_left_pad;
131
132 return p;
133}
134
Joonsoo Kim345c9052013-06-19 14:05:52 +0900135static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
136{
137#ifdef CONFIG_SLUB_CPU_PARTIAL
138 return !kmem_cache_debug(s);
139#else
140 return false;
141#endif
142}
143
Christoph Lameter81819f02007-05-06 14:49:36 -0700144/*
145 * Issues still to be resolved:
146 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700147 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
148 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700149 * - Variable sizing of the per node arrays
150 */
151
152/* Enable to test recovery from slab corruption on boot */
153#undef SLUB_RESILIENCY_TEST
154
Christoph Lameterb789ef52011-06-01 12:25:49 -0500155/* Enable to log cmpxchg failures */
156#undef SLUB_DEBUG_CMPXCHG
157
Christoph Lameter81819f02007-05-06 14:49:36 -0700158/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700159 * Mininum number of partial slabs. These will be left on the partial
160 * lists even if they are empty. kmem_cache_shrink may reclaim them.
161 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800162#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700163
Christoph Lameter2086d262007-05-06 14:49:46 -0700164/*
165 * Maximum number of desirable partial slabs.
166 * The existence of more partial slabs makes kmem_cache_shrink
Zhi Yong Wu721ae222013-11-08 20:47:37 +0800167 * sort the partial list by the number of objects in use.
Christoph Lameter2086d262007-05-06 14:49:46 -0700168 */
169#define MAX_PARTIAL 10
170
Laura Abbottbecfda62016-03-15 14:55:06 -0700171#define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
Christoph Lameter81819f02007-05-06 14:49:36 -0700172 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700173
Christoph Lameter81819f02007-05-06 14:49:36 -0700174/*
Laura Abbott149daaf2016-03-15 14:55:09 -0700175 * These debug flags cannot use CMPXCHG because there might be consistency
176 * issues when checking or reading debug information
177 */
178#define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
179 SLAB_TRACE)
180
181
182/*
David Rientjes3de47212009-07-27 18:30:35 -0700183 * Debugging flags that require metadata to be stored in the slab. These get
184 * disabled when slub_debug=O is used and a cache's min order increases with
185 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700186 */
David Rientjes3de47212009-07-27 18:30:35 -0700187#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700188
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400189#define OO_SHIFT 16
190#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500191#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400192
Christoph Lameter81819f02007-05-06 14:49:36 -0700193/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500194#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500195#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700196
Christoph Lameter02cbc872007-05-09 02:32:43 -0700197/*
198 * Tracking user of a slab.
199 */
Ben Greeard6543e32011-07-07 11:36:36 -0700200#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700201struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300202 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700203#ifdef CONFIG_STACKTRACE
204 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
205#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700206 int cpu; /* Was running on cpu */
207 int pid; /* Pid context */
208 unsigned long when; /* When did the operation occur */
209};
210
211enum track_item { TRACK_ALLOC, TRACK_FREE };
212
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500213#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700214static int sysfs_slab_add(struct kmem_cache *);
215static int sysfs_slab_alias(struct kmem_cache *, const char *);
Glauber Costa107dab52012-12-18 14:23:05 -0800216static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Tejun Heobf5eb3d2017-02-22 15:41:11 -0800217static void sysfs_slab_remove(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700218#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700219static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
220static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
221 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800222static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Tejun Heobf5eb3d2017-02-22 15:41:11 -0800223static inline void sysfs_slab_remove(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700224#endif
225
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500226static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800227{
228#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700229 /*
230 * The rmw is racy on a preemptible kernel but this is acceptable, so
231 * avoid this_cpu_add()'s irq-disable overhead.
232 */
233 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800234#endif
235}
236
Christoph Lameter81819f02007-05-06 14:49:36 -0700237/********************************************************************
238 * Core slab cache functions
239 *******************************************************************/
240
Christoph Lameter7656c722007-05-09 02:32:40 -0700241static inline void *get_freepointer(struct kmem_cache *s, void *object)
242{
243 return *(void **)(object + s->offset);
244}
245
Eric Dumazet0ad95002011-12-16 16:25:34 +0100246static void prefetch_freepointer(const struct kmem_cache *s, void *object)
247{
248 prefetch(object + s->offset);
249}
250
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500251static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
252{
253 void *p;
254
Joonsoo Kim922d5662016-03-17 14:17:53 -0700255 if (!debug_pagealloc_enabled())
256 return get_freepointer(s, object);
257
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500258 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500259 return p;
260}
261
Christoph Lameter7656c722007-05-09 02:32:40 -0700262static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
263{
264 *(void **)(object + s->offset) = fp;
265}
266
267/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300268#define for_each_object(__p, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700269 for (__p = fixup_red_left(__s, __addr); \
270 __p < (__addr) + (__objects) * (__s)->size; \
271 __p += (__s)->size)
Christoph Lameter7656c722007-05-09 02:32:40 -0700272
Wei Yang54266642014-08-06 16:04:42 -0700273#define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700274 for (__p = fixup_red_left(__s, __addr), __idx = 1; \
275 __idx <= __objects; \
276 __p += (__s)->size, __idx++)
Wei Yang54266642014-08-06 16:04:42 -0700277
Christoph Lameter7656c722007-05-09 02:32:40 -0700278/* Determine object index from a given position */
279static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
280{
281 return (p - addr) / s->size;
282}
283
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800284static inline int order_objects(int order, unsigned long size, int reserved)
285{
286 return ((PAGE_SIZE << order) - reserved) / size;
287}
288
Christoph Lameter834f3d12008-04-14 19:11:31 +0300289static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800290 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300291{
292 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800293 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300294 };
295
296 return x;
297}
298
299static inline int oo_order(struct kmem_cache_order_objects x)
300{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400301 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300302}
303
304static inline int oo_objects(struct kmem_cache_order_objects x)
305{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400306 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300307}
308
Christoph Lameter881db7f2011-06-01 12:25:53 -0500309/*
310 * Per slab locking using the pagelock
311 */
312static __always_inline void slab_lock(struct page *page)
313{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800314 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500315 bit_spin_lock(PG_locked, &page->flags);
316}
317
318static __always_inline void slab_unlock(struct page *page)
319{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800320 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500321 __bit_spin_unlock(PG_locked, &page->flags);
322}
323
Dave Hansena0320862014-01-30 15:46:09 -0800324static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
325{
326 struct page tmp;
327 tmp.counters = counters_new;
328 /*
329 * page->counters can cover frozen/inuse/objects as well
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700330 * as page->_refcount. If we assign to ->counters directly
331 * we run the risk of losing updates to page->_refcount, so
Dave Hansena0320862014-01-30 15:46:09 -0800332 * be careful and only assign to the fields we need.
333 */
334 page->frozen = tmp.frozen;
335 page->inuse = tmp.inuse;
336 page->objects = tmp.objects;
337}
338
Christoph Lameter1d071712011-07-14 12:49:12 -0500339/* Interrupts must be disabled (for the fallback code to work right) */
340static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500341 void *freelist_old, unsigned long counters_old,
342 void *freelist_new, unsigned long counters_new,
343 const char *n)
344{
Christoph Lameter1d071712011-07-14 12:49:12 -0500345 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800346#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
347 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500348 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000349 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700350 freelist_old, counters_old,
351 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700352 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500353 } else
354#endif
355 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500356 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800357 if (page->freelist == freelist_old &&
358 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500359 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800360 set_page_slub_counters(page, counters_new);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500361 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700362 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500363 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500364 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500365 }
366
367 cpu_relax();
368 stat(s, CMPXCHG_DOUBLE_FAIL);
369
370#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700371 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500372#endif
373
Joe Perches6f6528a2015-04-14 15:44:31 -0700374 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500375}
376
Christoph Lameter1d071712011-07-14 12:49:12 -0500377static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
378 void *freelist_old, unsigned long counters_old,
379 void *freelist_new, unsigned long counters_new,
380 const char *n)
381{
Heiko Carstens25654092012-01-12 17:17:33 -0800382#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
383 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500384 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000385 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700386 freelist_old, counters_old,
387 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700388 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500389 } else
390#endif
391 {
392 unsigned long flags;
393
394 local_irq_save(flags);
395 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800396 if (page->freelist == freelist_old &&
397 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500398 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800399 set_page_slub_counters(page, counters_new);
Christoph Lameter1d071712011-07-14 12:49:12 -0500400 slab_unlock(page);
401 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700402 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500403 }
404 slab_unlock(page);
405 local_irq_restore(flags);
406 }
407
408 cpu_relax();
409 stat(s, CMPXCHG_DOUBLE_FAIL);
410
411#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700412 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500413#endif
414
Joe Perches6f6528a2015-04-14 15:44:31 -0700415 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500416}
417
Christoph Lameter41ecc552007-05-09 02:32:44 -0700418#ifdef CONFIG_SLUB_DEBUG
419/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500420 * Determine a map of object in use on a page.
421 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500422 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500423 * not vanish from under us.
424 */
425static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
426{
427 void *p;
428 void *addr = page_address(page);
429
430 for (p = page->freelist; p; p = get_freepointer(s, p))
431 set_bit(slab_index(p, s, addr), map);
432}
433
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700434static inline int size_from_object(struct kmem_cache *s)
435{
436 if (s->flags & SLAB_RED_ZONE)
437 return s->size - s->red_left_pad;
438
439 return s->size;
440}
441
442static inline void *restore_red_left(struct kmem_cache *s, void *p)
443{
444 if (s->flags & SLAB_RED_ZONE)
445 p -= s->red_left_pad;
446
447 return p;
448}
449
Christoph Lameter41ecc552007-05-09 02:32:44 -0700450/*
451 * Debug settings:
452 */
Andrey Ryabinin89d3c872015-11-05 18:51:23 -0800453#if defined(CONFIG_SLUB_DEBUG_ON)
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700454static int slub_debug = DEBUG_DEFAULT_FLAGS;
455#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700456static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700457#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700458
459static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700460static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700461
Christoph Lameter7656c722007-05-09 02:32:40 -0700462/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800463 * slub is about to manipulate internal object metadata. This memory lies
464 * outside the range of the allocated object, so accessing it would normally
465 * be reported by kasan as a bounds error. metadata_access_enable() is used
466 * to tell kasan that these accesses are OK.
467 */
468static inline void metadata_access_enable(void)
469{
470 kasan_disable_current();
471}
472
473static inline void metadata_access_disable(void)
474{
475 kasan_enable_current();
476}
477
478/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700479 * Object debugging
480 */
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700481
482/* Verify that a pointer has an address that is valid within a slab page */
483static inline int check_valid_pointer(struct kmem_cache *s,
484 struct page *page, void *object)
485{
486 void *base;
487
488 if (!object)
489 return 1;
490
491 base = page_address(page);
492 object = restore_red_left(s, object);
493 if (object < base || object >= base + page->objects * s->size ||
494 (object - base) % s->size) {
495 return 0;
496 }
497
498 return 1;
499}
500
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800501static void print_section(char *level, char *text, u8 *addr,
502 unsigned int length)
Christoph Lameter81819f02007-05-06 14:49:36 -0700503{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800504 metadata_access_enable();
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800505 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200506 length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800507 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700508}
509
Christoph Lameter81819f02007-05-06 14:49:36 -0700510static struct track *get_track(struct kmem_cache *s, void *object,
511 enum track_item alloc)
512{
513 struct track *p;
514
515 if (s->offset)
516 p = object + s->offset + sizeof(void *);
517 else
518 p = object + s->inuse;
519
520 return p + alloc;
521}
522
523static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300524 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700525{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900526 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700527
Christoph Lameter81819f02007-05-06 14:49:36 -0700528 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700529#ifdef CONFIG_STACKTRACE
530 struct stack_trace trace;
531 int i;
532
533 trace.nr_entries = 0;
534 trace.max_entries = TRACK_ADDRS_COUNT;
535 trace.entries = p->addrs;
536 trace.skip = 3;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800537 metadata_access_enable();
Ben Greeard6543e32011-07-07 11:36:36 -0700538 save_stack_trace(&trace);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800539 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700540
541 /* See rant in lockdep.c */
542 if (trace.nr_entries != 0 &&
543 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
544 trace.nr_entries--;
545
546 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
547 p->addrs[i] = 0;
548#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700549 p->addr = addr;
550 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400551 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700552 p->when = jiffies;
553 } else
554 memset(p, 0, sizeof(struct track));
555}
556
Christoph Lameter81819f02007-05-06 14:49:36 -0700557static void init_tracking(struct kmem_cache *s, void *object)
558{
Christoph Lameter24922682007-07-17 04:03:18 -0700559 if (!(s->flags & SLAB_STORE_USER))
560 return;
561
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300562 set_track(s, object, TRACK_FREE, 0UL);
563 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700564}
565
566static void print_track(const char *s, struct track *t)
567{
568 if (!t->addr)
569 return;
570
Fabian Frederickf9f58282014-06-04 16:06:34 -0700571 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
572 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700573#ifdef CONFIG_STACKTRACE
574 {
575 int i;
576 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
577 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700578 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700579 else
580 break;
581 }
582#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700583}
584
Christoph Lameter24922682007-07-17 04:03:18 -0700585static void print_tracking(struct kmem_cache *s, void *object)
586{
587 if (!(s->flags & SLAB_STORE_USER))
588 return;
589
590 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
591 print_track("Freed", get_track(s, object, TRACK_FREE));
592}
593
594static void print_page_info(struct page *page)
595{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700596 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800597 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700598
599}
600
601static void slab_bug(struct kmem_cache *s, char *fmt, ...)
602{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700603 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700604 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700605
606 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700607 vaf.fmt = fmt;
608 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700609 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700610 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700611 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400612
Rusty Russell373d4d02013-01-21 17:17:39 +1030613 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700614 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700615}
616
617static void slab_fix(struct kmem_cache *s, char *fmt, ...)
618{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700619 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700620 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700621
622 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700623 vaf.fmt = fmt;
624 vaf.va = &args;
625 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700626 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700627}
628
629static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700630{
631 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800632 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700633
634 print_tracking(s, p);
635
636 print_page_info(page);
637
Fabian Frederickf9f58282014-06-04 16:06:34 -0700638 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
639 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700640
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700641 if (s->flags & SLAB_RED_ZONE)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800642 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
643 s->red_left_pad);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700644 else if (p > addr + 16)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800645 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700646
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800647 print_section(KERN_ERR, "Object ", p,
648 min_t(unsigned long, s->object_size, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700649 if (s->flags & SLAB_RED_ZONE)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800650 print_section(KERN_ERR, "Redzone ", p + s->object_size,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500651 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700652
Christoph Lameter81819f02007-05-06 14:49:36 -0700653 if (s->offset)
654 off = s->offset + sizeof(void *);
655 else
656 off = s->inuse;
657
Christoph Lameter24922682007-07-17 04:03:18 -0700658 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700659 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700660
Alexander Potapenko80a92012016-07-28 15:49:07 -0700661 off += kasan_metadata_size(s);
662
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700663 if (off != size_from_object(s))
Christoph Lameter81819f02007-05-06 14:49:36 -0700664 /* Beginning of the filler is the free pointer */
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800665 print_section(KERN_ERR, "Padding ", p + off,
666 size_from_object(s) - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700667
668 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700669}
670
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800671void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700672 u8 *object, char *reason)
673{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700674 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700675 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700676}
677
Chen Gangd0e0ac92013-07-15 09:05:29 +0800678static void slab_err(struct kmem_cache *s, struct page *page,
679 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700680{
681 va_list args;
682 char buf[100];
683
Christoph Lameter24922682007-07-17 04:03:18 -0700684 va_start(args, fmt);
685 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700686 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700687 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700688 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700689 dump_stack();
690}
691
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500692static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700693{
694 u8 *p = object;
695
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700696 if (s->flags & SLAB_RED_ZONE)
697 memset(p - s->red_left_pad, val, s->red_left_pad);
698
Christoph Lameter81819f02007-05-06 14:49:36 -0700699 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500700 memset(p, POISON_FREE, s->object_size - 1);
701 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700702 }
703
704 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500705 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700706}
707
Christoph Lameter24922682007-07-17 04:03:18 -0700708static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
709 void *from, void *to)
710{
711 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
712 memset(from, data, to - from);
713}
714
715static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
716 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800717 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700718{
719 u8 *fault;
720 u8 *end;
721
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800722 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700723 fault = memchr_inv(start, value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800724 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700725 if (!fault)
726 return 1;
727
728 end = start + bytes;
729 while (end > fault && end[-1] == value)
730 end--;
731
732 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700733 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700734 fault, end - 1, fault[0], value);
735 print_trailer(s, page, object);
736
737 restore_bytes(s, what, value, fault, end);
738 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700739}
740
Christoph Lameter81819f02007-05-06 14:49:36 -0700741/*
742 * Object layout:
743 *
744 * object address
745 * Bytes of the object to be managed.
746 * If the freepointer may overlay the object then the free
747 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700748 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700749 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
750 * 0xa5 (POISON_END)
751 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500752 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700753 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700754 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500755 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700756 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700757 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
758 * 0xcc (RED_ACTIVE) for objects in use.
759 *
760 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700761 * Meta data starts here.
762 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700763 * A. Free pointer (if we cannot overwrite object on free)
764 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700765 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800766 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700767 * before the word boundary.
768 *
769 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700770 *
771 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700772 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700773 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500774 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700775 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700776 * may be used with merged slabcaches.
777 */
778
Christoph Lameter81819f02007-05-06 14:49:36 -0700779static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
780{
781 unsigned long off = s->inuse; /* The end of info */
782
783 if (s->offset)
784 /* Freepointer is placed after the object. */
785 off += sizeof(void *);
786
787 if (s->flags & SLAB_STORE_USER)
788 /* We also have user information there */
789 off += 2 * sizeof(struct track);
790
Alexander Potapenko80a92012016-07-28 15:49:07 -0700791 off += kasan_metadata_size(s);
792
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700793 if (size_from_object(s) == off)
Christoph Lameter81819f02007-05-06 14:49:36 -0700794 return 1;
795
Christoph Lameter24922682007-07-17 04:03:18 -0700796 return check_bytes_and_report(s, page, p, "Object padding",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700797 p + off, POISON_INUSE, size_from_object(s) - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700798}
799
Christoph Lameter39b26462008-04-14 19:11:30 +0300800/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700801static int slab_pad_check(struct kmem_cache *s, struct page *page)
802{
Christoph Lameter24922682007-07-17 04:03:18 -0700803 u8 *start;
804 u8 *fault;
805 u8 *end;
806 int length;
807 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700808
809 if (!(s->flags & SLAB_POISON))
810 return 1;
811
Christoph Lametera973e9d2008-03-01 13:40:44 -0800812 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800813 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300814 end = start + length;
815 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700816 if (!remainder)
817 return 1;
818
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800819 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700820 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800821 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700822 if (!fault)
823 return 1;
824 while (end > fault && end[-1] == POISON_INUSE)
825 end--;
826
827 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800828 print_section(KERN_ERR, "Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700829
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200830 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700831 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700832}
833
834static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500835 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700836{
837 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500838 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700839
840 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700841 if (!check_bytes_and_report(s, page, object, "Redzone",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700842 object - s->red_left_pad, val, s->red_left_pad))
843 return 0;
844
845 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500846 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700847 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700848 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500849 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800850 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800851 endobject, POISON_INUSE,
852 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800853 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700854 }
855
856 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500857 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700858 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500859 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700860 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500861 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700862 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700863 /*
864 * check_pad_bytes cleans up on its own.
865 */
866 check_pad_bytes(s, page, p);
867 }
868
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500869 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700870 /*
871 * Object and freepointer overlap. Cannot check
872 * freepointer while object is allocated.
873 */
874 return 1;
875
876 /* Check free pointer validity */
877 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
878 object_err(s, page, p, "Freepointer corrupt");
879 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100880 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700881 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700882 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700883 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800884 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700885 return 0;
886 }
887 return 1;
888}
889
890static int check_slab(struct kmem_cache *s, struct page *page)
891{
Christoph Lameter39b26462008-04-14 19:11:30 +0300892 int maxobj;
893
Christoph Lameter81819f02007-05-06 14:49:36 -0700894 VM_BUG_ON(!irqs_disabled());
895
896 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700897 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700898 return 0;
899 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300900
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800901 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300902 if (page->objects > maxobj) {
903 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800904 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +0300905 return 0;
906 }
907 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700908 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800909 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700910 return 0;
911 }
912 /* Slab_pad_check fixes things up after itself */
913 slab_pad_check(s, page);
914 return 1;
915}
916
917/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700918 * Determine if a certain object on a page is on the freelist. Must hold the
919 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700920 */
921static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
922{
923 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500924 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700925 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800926 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700927
Christoph Lameter881db7f2011-06-01 12:25:53 -0500928 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300929 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700930 if (fp == search)
931 return 1;
932 if (!check_valid_pointer(s, page, fp)) {
933 if (object) {
934 object_err(s, page, object,
935 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800936 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700937 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700938 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800939 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300940 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700941 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700942 return 0;
943 }
944 break;
945 }
946 object = fp;
947 fp = get_freepointer(s, object);
948 nr++;
949 }
950
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800951 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400952 if (max_objects > MAX_OBJS_PER_PAGE)
953 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300954
955 if (page->objects != max_objects) {
Joe Perches756a0252016-03-17 14:19:47 -0700956 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
957 page->objects, max_objects);
Christoph Lameter224a88b2008-04-14 19:11:31 +0300958 page->objects = max_objects;
959 slab_fix(s, "Number of objects adjusted.");
960 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300961 if (page->inuse != page->objects - nr) {
Joe Perches756a0252016-03-17 14:19:47 -0700962 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
963 page->inuse, page->objects - nr);
Christoph Lameter39b26462008-04-14 19:11:30 +0300964 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700965 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700966 }
967 return search == NULL;
968}
969
Christoph Lameter0121c6192008-04-29 16:11:12 -0700970static void trace(struct kmem_cache *s, struct page *page, void *object,
971 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700972{
973 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -0700974 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -0700975 s->name,
976 alloc ? "alloc" : "free",
977 object, page->inuse,
978 page->freelist);
979
980 if (!alloc)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800981 print_section(KERN_INFO, "Object ", (void *)object,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800982 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700983
984 dump_stack();
985 }
986}
987
Christoph Lameter643b1132007-05-06 14:49:42 -0700988/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700989 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700990 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500991static void add_full(struct kmem_cache *s,
992 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700993{
Christoph Lameter643b1132007-05-06 14:49:42 -0700994 if (!(s->flags & SLAB_STORE_USER))
995 return;
996
David Rientjes255d0882014-02-10 14:25:39 -0800997 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500998 list_add(&page->lru, &n->full);
999}
Christoph Lameter643b1132007-05-06 14:49:42 -07001000
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001001static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001002{
1003 if (!(s->flags & SLAB_STORE_USER))
1004 return;
1005
David Rientjes255d0882014-02-10 14:25:39 -08001006 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -07001007 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001008}
1009
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001010/* Tracking of the number of slabs for debugging purposes */
1011static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1012{
1013 struct kmem_cache_node *n = get_node(s, node);
1014
1015 return atomic_long_read(&n->nr_slabs);
1016}
1017
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001018static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1019{
1020 return atomic_long_read(&n->nr_slabs);
1021}
1022
Christoph Lameter205ab992008-04-14 19:11:40 +03001023static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001024{
1025 struct kmem_cache_node *n = get_node(s, node);
1026
1027 /*
1028 * May be called early in order to allocate a slab for the
1029 * kmem_cache_node structure. Solve the chicken-egg
1030 * dilemma by deferring the increment of the count during
1031 * bootstrap (see early_kmem_cache_node_alloc).
1032 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001033 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001034 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001035 atomic_long_add(objects, &n->total_objects);
1036 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001037}
Christoph Lameter205ab992008-04-14 19:11:40 +03001038static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001039{
1040 struct kmem_cache_node *n = get_node(s, node);
1041
1042 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001043 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001044}
1045
1046/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001047static void setup_object_debug(struct kmem_cache *s, struct page *page,
1048 void *object)
1049{
1050 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1051 return;
1052
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001053 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001054 init_tracking(s, object);
1055}
1056
Laura Abbottbecfda62016-03-15 14:55:06 -07001057static inline int alloc_consistency_checks(struct kmem_cache *s,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001058 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001059 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001060{
1061 if (!check_slab(s, page))
Laura Abbottbecfda62016-03-15 14:55:06 -07001062 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001063
Christoph Lameter81819f02007-05-06 14:49:36 -07001064 if (!check_valid_pointer(s, page, object)) {
1065 object_err(s, page, object, "Freelist Pointer check fails");
Laura Abbottbecfda62016-03-15 14:55:06 -07001066 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001067 }
1068
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001069 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Laura Abbottbecfda62016-03-15 14:55:06 -07001070 return 0;
1071
1072 return 1;
1073}
1074
1075static noinline int alloc_debug_processing(struct kmem_cache *s,
1076 struct page *page,
1077 void *object, unsigned long addr)
1078{
1079 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1080 if (!alloc_consistency_checks(s, page, object, addr))
1081 goto bad;
1082 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001083
Christoph Lameter3ec09742007-05-16 22:11:00 -07001084 /* Success perform special debug activities for allocs */
1085 if (s->flags & SLAB_STORE_USER)
1086 set_track(s, object, TRACK_ALLOC, addr);
1087 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001088 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001089 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001090
Christoph Lameter81819f02007-05-06 14:49:36 -07001091bad:
1092 if (PageSlab(page)) {
1093 /*
1094 * If this is a slab page then lets do the best we can
1095 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001096 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001097 */
Christoph Lameter24922682007-07-17 04:03:18 -07001098 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001099 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001100 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001101 }
1102 return 0;
1103}
1104
Laura Abbottbecfda62016-03-15 14:55:06 -07001105static inline int free_consistency_checks(struct kmem_cache *s,
1106 struct page *page, void *object, unsigned long addr)
1107{
1108 if (!check_valid_pointer(s, page, object)) {
1109 slab_err(s, page, "Invalid object pointer 0x%p", object);
1110 return 0;
1111 }
1112
1113 if (on_freelist(s, page, object)) {
1114 object_err(s, page, object, "Object already free");
1115 return 0;
1116 }
1117
1118 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1119 return 0;
1120
1121 if (unlikely(s != page->slab_cache)) {
1122 if (!PageSlab(page)) {
Joe Perches756a0252016-03-17 14:19:47 -07001123 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1124 object);
Laura Abbottbecfda62016-03-15 14:55:06 -07001125 } else if (!page->slab_cache) {
1126 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1127 object);
1128 dump_stack();
1129 } else
1130 object_err(s, page, object,
1131 "page slab pointer corrupt.");
1132 return 0;
1133 }
1134 return 1;
1135}
1136
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001137/* Supports checking bulk free of a constructed freelist */
Laura Abbott282acb42016-03-15 14:54:59 -07001138static noinline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001139 struct kmem_cache *s, struct page *page,
1140 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001141 unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001142{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001143 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001144 void *object = head;
1145 int cnt = 0;
Laura Abbott282acb42016-03-15 14:54:59 -07001146 unsigned long uninitialized_var(flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001147 int ret = 0;
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001148
Laura Abbott282acb42016-03-15 14:54:59 -07001149 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001150 slab_lock(page);
1151
Laura Abbottbecfda62016-03-15 14:55:06 -07001152 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1153 if (!check_slab(s, page))
1154 goto out;
1155 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001156
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001157next_object:
1158 cnt++;
1159
Laura Abbottbecfda62016-03-15 14:55:06 -07001160 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1161 if (!free_consistency_checks(s, page, object, addr))
1162 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001163 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001164
Christoph Lameter3ec09742007-05-16 22:11:00 -07001165 if (s->flags & SLAB_STORE_USER)
1166 set_track(s, object, TRACK_FREE, addr);
1167 trace(s, page, object, 0);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001168 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001169 init_object(s, object, SLUB_RED_INACTIVE);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001170
1171 /* Reached end of constructed freelist yet? */
1172 if (object != tail) {
1173 object = get_freepointer(s, object);
1174 goto next_object;
1175 }
Laura Abbott804aa132016-03-15 14:55:02 -07001176 ret = 1;
1177
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001178out:
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001179 if (cnt != bulk_cnt)
1180 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1181 bulk_cnt, cnt);
1182
Christoph Lameter881db7f2011-06-01 12:25:53 -05001183 slab_unlock(page);
Laura Abbott282acb42016-03-15 14:54:59 -07001184 spin_unlock_irqrestore(&n->list_lock, flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001185 if (!ret)
1186 slab_fix(s, "Object at 0x%p not freed", object);
1187 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001188}
1189
Christoph Lameter41ecc552007-05-09 02:32:44 -07001190static int __init setup_slub_debug(char *str)
1191{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001192 slub_debug = DEBUG_DEFAULT_FLAGS;
1193 if (*str++ != '=' || !*str)
1194 /*
1195 * No options specified. Switch on full debugging.
1196 */
1197 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001198
1199 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001200 /*
1201 * No options but restriction on slabs. This means full
1202 * debugging for slabs matching a pattern.
1203 */
1204 goto check_slabs;
1205
1206 slub_debug = 0;
1207 if (*str == '-')
1208 /*
1209 * Switch off all debugging measures.
1210 */
1211 goto out;
1212
1213 /*
1214 * Determine which debug features should be switched on
1215 */
Pekka Enberg06428782008-01-07 23:20:27 -08001216 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001217 switch (tolower(*str)) {
1218 case 'f':
Laura Abbottbecfda62016-03-15 14:55:06 -07001219 slub_debug |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001220 break;
1221 case 'z':
1222 slub_debug |= SLAB_RED_ZONE;
1223 break;
1224 case 'p':
1225 slub_debug |= SLAB_POISON;
1226 break;
1227 case 'u':
1228 slub_debug |= SLAB_STORE_USER;
1229 break;
1230 case 't':
1231 slub_debug |= SLAB_TRACE;
1232 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001233 case 'a':
1234 slub_debug |= SLAB_FAILSLAB;
1235 break;
Chris J Arges08303a72015-04-14 15:44:25 -07001236 case 'o':
1237 /*
1238 * Avoid enabling debugging on caches if its minimum
1239 * order would increase as a result.
1240 */
1241 disable_higher_order_debug = 1;
1242 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001243 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001244 pr_err("slub_debug option '%c' unknown. skipped\n",
1245 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001246 }
1247 }
1248
1249check_slabs:
1250 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001251 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001252out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001253 return 1;
1254}
1255
1256__setup("slub_debug", setup_slub_debug);
1257
Joonsoo Kim423c9292014-10-09 15:26:22 -07001258unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001259 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001260 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001261{
1262 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001263 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001264 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001265 if (slub_debug && (!slub_debug_slabs || (name &&
1266 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001267 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001268
1269 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001270}
Jesper Dangaard Brouerb4a64712015-11-20 15:57:41 -08001271#else /* !CONFIG_SLUB_DEBUG */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001272static inline void setup_object_debug(struct kmem_cache *s,
1273 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001274
Christoph Lameter3ec09742007-05-16 22:11:00 -07001275static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001276 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001277
Laura Abbott282acb42016-03-15 14:54:59 -07001278static inline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001279 struct kmem_cache *s, struct page *page,
1280 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001281 unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001282
Christoph Lameter41ecc552007-05-09 02:32:44 -07001283static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1284 { return 1; }
1285static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001286 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001287static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1288 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001289static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1290 struct page *page) {}
Joonsoo Kim423c9292014-10-09 15:26:22 -07001291unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001292 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001293 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001294{
1295 return flags;
1296}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001297#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001298
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001299#define disable_higher_order_debug 0
1300
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001301static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1302 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001303static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1304 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001305static inline void inc_slabs_node(struct kmem_cache *s, int node,
1306 int objects) {}
1307static inline void dec_slabs_node(struct kmem_cache *s, int node,
1308 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001309
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001310#endif /* CONFIG_SLUB_DEBUG */
1311
1312/*
1313 * Hooks for other subsystems that check memory allocations. In a typical
1314 * production configuration these hooks all should produce no code at all.
1315 */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001316static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1317{
1318 kmemleak_alloc(ptr, size, 1, flags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07001319 kasan_kmalloc_large(ptr, size, flags);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001320}
1321
1322static inline void kfree_hook(const void *x)
1323{
1324 kmemleak_free(x);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001325 kasan_kfree_large(x);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001326}
1327
Alexander Potapenko80a92012016-07-28 15:49:07 -07001328static inline void *slab_free_hook(struct kmem_cache *s, void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001329{
Alexander Potapenko80a92012016-07-28 15:49:07 -07001330 void *freeptr;
1331
Roman Bobnievd56791b2013-10-08 15:58:57 -07001332 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001333
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001334 /*
1335 * Trouble is that we may no longer disable interrupts in the fast path
1336 * So in order to make the debug calls that expect irqs to be
1337 * disabled we need to disable interrupts temporarily.
1338 */
1339#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1340 {
1341 unsigned long flags;
1342
1343 local_irq_save(flags);
1344 kmemcheck_slab_free(s, x, s->object_size);
1345 debug_check_no_locks_freed(x, s->object_size);
1346 local_irq_restore(flags);
1347 }
1348#endif
1349 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1350 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001351
Alexander Potapenko80a92012016-07-28 15:49:07 -07001352 freeptr = get_freepointer(s, x);
1353 /*
1354 * kasan_slab_free() may put x into memory quarantine, delaying its
1355 * reuse. In this case the object's freelist pointer is changed.
1356 */
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001357 kasan_slab_free(s, x);
Alexander Potapenko80a92012016-07-28 15:49:07 -07001358 return freeptr;
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001359}
Christoph Lameter205ab992008-04-14 19:11:40 +03001360
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001361static inline void slab_free_freelist_hook(struct kmem_cache *s,
1362 void *head, void *tail)
1363{
1364/*
1365 * Compiler cannot detect this function can be removed if slab_free_hook()
1366 * evaluates to nothing. Thus, catch all relevant config debug options here.
1367 */
1368#if defined(CONFIG_KMEMCHECK) || \
1369 defined(CONFIG_LOCKDEP) || \
1370 defined(CONFIG_DEBUG_KMEMLEAK) || \
1371 defined(CONFIG_DEBUG_OBJECTS_FREE) || \
1372 defined(CONFIG_KASAN)
1373
1374 void *object = head;
1375 void *tail_obj = tail ? : head;
Alexander Potapenko80a92012016-07-28 15:49:07 -07001376 void *freeptr;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001377
1378 do {
Alexander Potapenko80a92012016-07-28 15:49:07 -07001379 freeptr = slab_free_hook(s, object);
1380 } while ((object != tail_obj) && (object = freeptr));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001381#endif
1382}
1383
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001384static void setup_object(struct kmem_cache *s, struct page *page,
1385 void *object)
1386{
1387 setup_object_debug(s, page, object);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07001388 kasan_init_slab_obj(s, object);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001389 if (unlikely(s->ctor)) {
1390 kasan_unpoison_object_data(s, object);
1391 s->ctor(object);
1392 kasan_poison_object_data(s, object);
1393 }
1394}
1395
Christoph Lameter81819f02007-05-06 14:49:36 -07001396/*
1397 * Slab allocation and freeing
1398 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001399static inline struct page *alloc_slab_page(struct kmem_cache *s,
1400 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001401{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001402 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001403 int order = oo_order(oo);
1404
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001405 flags |= __GFP_NOTRACK;
1406
Christoph Lameter2154a332010-07-09 14:07:10 -05001407 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001408 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001409 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001410 page = __alloc_pages_node(node, flags, order);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001411
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001412 if (page && memcg_charge_slab(page, flags, order, s)) {
1413 __free_pages(page, order);
1414 page = NULL;
1415 }
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001416
1417 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001418}
1419
Thomas Garnier210e7a42016-07-26 15:21:59 -07001420#ifdef CONFIG_SLAB_FREELIST_RANDOM
1421/* Pre-initialize the random sequence cache */
1422static int init_cache_random_seq(struct kmem_cache *s)
1423{
1424 int err;
1425 unsigned long i, count = oo_objects(s->oo);
1426
Sean Reesa8100072017-02-08 14:30:59 -08001427 /* Bailout if already initialised */
1428 if (s->random_seq)
1429 return 0;
1430
Thomas Garnier210e7a42016-07-26 15:21:59 -07001431 err = cache_random_seq_create(s, count, GFP_KERNEL);
1432 if (err) {
1433 pr_err("SLUB: Unable to initialize free list for %s\n",
1434 s->name);
1435 return err;
1436 }
1437
1438 /* Transform to an offset on the set of pages */
1439 if (s->random_seq) {
1440 for (i = 0; i < count; i++)
1441 s->random_seq[i] *= s->size;
1442 }
1443 return 0;
1444}
1445
1446/* Initialize each random sequence freelist per cache */
1447static void __init init_freelist_randomization(void)
1448{
1449 struct kmem_cache *s;
1450
1451 mutex_lock(&slab_mutex);
1452
1453 list_for_each_entry(s, &slab_caches, list)
1454 init_cache_random_seq(s);
1455
1456 mutex_unlock(&slab_mutex);
1457}
1458
1459/* Get the next entry on the pre-computed freelist randomized */
1460static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1461 unsigned long *pos, void *start,
1462 unsigned long page_limit,
1463 unsigned long freelist_count)
1464{
1465 unsigned int idx;
1466
1467 /*
1468 * If the target page allocation failed, the number of objects on the
1469 * page might be smaller than the usual size defined by the cache.
1470 */
1471 do {
1472 idx = s->random_seq[*pos];
1473 *pos += 1;
1474 if (*pos >= freelist_count)
1475 *pos = 0;
1476 } while (unlikely(idx >= page_limit));
1477
1478 return (char *)start + idx;
1479}
1480
1481/* Shuffle the single linked freelist based on a random pre-computed sequence */
1482static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1483{
1484 void *start;
1485 void *cur;
1486 void *next;
1487 unsigned long idx, pos, page_limit, freelist_count;
1488
1489 if (page->objects < 2 || !s->random_seq)
1490 return false;
1491
1492 freelist_count = oo_objects(s->oo);
1493 pos = get_random_int() % freelist_count;
1494
1495 page_limit = page->objects * s->size;
1496 start = fixup_red_left(s, page_address(page));
1497
1498 /* First entry is used as the base of the freelist */
1499 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1500 freelist_count);
1501 page->freelist = cur;
1502
1503 for (idx = 1; idx < page->objects; idx++) {
1504 setup_object(s, page, cur);
1505 next = next_freelist_entry(s, page, &pos, start, page_limit,
1506 freelist_count);
1507 set_freepointer(s, cur, next);
1508 cur = next;
1509 }
1510 setup_object(s, page, cur);
1511 set_freepointer(s, cur, NULL);
1512
1513 return true;
1514}
1515#else
1516static inline int init_cache_random_seq(struct kmem_cache *s)
1517{
1518 return 0;
1519}
1520static inline void init_freelist_randomization(void) { }
1521static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1522{
1523 return false;
1524}
1525#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1526
Christoph Lameter81819f02007-05-06 14:49:36 -07001527static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1528{
Pekka Enberg06428782008-01-07 23:20:27 -08001529 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001530 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001531 gfp_t alloc_gfp;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001532 void *start, *p;
1533 int idx, order;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001534 bool shuffle;
Christoph Lameter81819f02007-05-06 14:49:36 -07001535
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001536 flags &= gfp_allowed_mask;
1537
Mel Gormand0164ad2015-11-06 16:28:21 -08001538 if (gfpflags_allow_blocking(flags))
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001539 local_irq_enable();
1540
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001541 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001542
Pekka Enbergba522702009-06-24 21:59:51 +03001543 /*
1544 * Let the initial higher-order allocation fail under memory pressure
1545 * so we fall-back to the minimum order allocation.
1546 */
1547 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Mel Gormand0164ad2015-11-06 16:28:21 -08001548 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
Mel Gorman444eb2a42016-03-17 14:19:23 -07001549 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
Pekka Enbergba522702009-06-24 21:59:51 +03001550
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001551 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001552 if (unlikely(!page)) {
1553 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001554 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001555 /*
1556 * Allocation may have failed due to fragmentation.
1557 * Try a lower order alloc if possible
1558 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001559 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001560 if (unlikely(!page))
1561 goto out;
1562 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001563 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001564
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001565 if (kmemcheck_enabled &&
1566 !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001567 int pages = 1 << oo_order(oo);
1568
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001569 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001570
1571 /*
1572 * Objects from caches that have a constructor don't get
1573 * cleared when they're allocated, so we need to do it here.
1574 */
1575 if (s->ctor)
1576 kmemcheck_mark_uninitialized_pages(page, pages);
1577 else
1578 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001579 }
1580
Christoph Lameter834f3d12008-04-14 19:11:31 +03001581 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001582
Glauber Costa1f458cb2012-12-18 14:22:50 -08001583 order = compound_order(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001584 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001585 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001586 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001587 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001588
1589 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001590
1591 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001592 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001593
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001594 kasan_poison_slab(page);
1595
Thomas Garnier210e7a42016-07-26 15:21:59 -07001596 shuffle = shuffle_freelist(s, page);
1597
1598 if (!shuffle) {
1599 for_each_object_idx(p, idx, s, start, page->objects) {
1600 setup_object(s, page, p);
1601 if (likely(idx < page->objects))
1602 set_freepointer(s, p, p + s->size);
1603 else
1604 set_freepointer(s, p, NULL);
1605 }
1606 page->freelist = fixup_red_left(s, start);
Christoph Lameter81819f02007-05-06 14:49:36 -07001607 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001608
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001609 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001610 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001611
Christoph Lameter81819f02007-05-06 14:49:36 -07001612out:
Mel Gormand0164ad2015-11-06 16:28:21 -08001613 if (gfpflags_allow_blocking(flags))
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001614 local_irq_disable();
1615 if (!page)
1616 return NULL;
1617
1618 mod_zone_page_state(page_zone(page),
1619 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1620 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1621 1 << oo_order(oo));
1622
1623 inc_slabs_node(s, page_to_nid(page), page->objects);
1624
Christoph Lameter81819f02007-05-06 14:49:36 -07001625 return page;
1626}
1627
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001628static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1629{
1630 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07001631 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07001632 flags &= ~GFP_SLAB_BUG_MASK;
1633 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1634 invalid_mask, &invalid_mask, flags, &flags);
Borislav Petkov65b9de72017-02-22 15:41:02 -08001635 dump_stack();
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001636 }
1637
1638 return allocate_slab(s,
1639 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1640}
1641
Christoph Lameter81819f02007-05-06 14:49:36 -07001642static void __free_slab(struct kmem_cache *s, struct page *page)
1643{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001644 int order = compound_order(page);
1645 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001646
Laura Abbottbecfda62016-03-15 14:55:06 -07001647 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001648 void *p;
1649
1650 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001651 for_each_object(p, s, page_address(page),
1652 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001653 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001654 }
1655
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001656 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001657
Christoph Lameter81819f02007-05-06 14:49:36 -07001658 mod_zone_page_state(page_zone(page),
1659 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1660 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001661 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001662
Mel Gorman072bb0a2012-07-31 16:43:58 -07001663 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001664 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001665
Mel Gorman22b751c2013-02-22 16:34:59 -08001666 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001667 if (current->reclaim_state)
1668 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001669 memcg_uncharge_slab(page, order, s);
1670 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001671}
1672
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001673#define need_reserve_slab_rcu \
1674 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1675
Christoph Lameter81819f02007-05-06 14:49:36 -07001676static void rcu_free_slab(struct rcu_head *h)
1677{
1678 struct page *page;
1679
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001680 if (need_reserve_slab_rcu)
1681 page = virt_to_head_page(h);
1682 else
1683 page = container_of((struct list_head *)h, struct page, lru);
1684
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001685 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001686}
1687
1688static void free_slab(struct kmem_cache *s, struct page *page)
1689{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001690 if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001691 struct rcu_head *head;
1692
1693 if (need_reserve_slab_rcu) {
1694 int order = compound_order(page);
1695 int offset = (PAGE_SIZE << order) - s->reserved;
1696
1697 VM_BUG_ON(s->reserved != sizeof(*head));
1698 head = page_address(page) + offset;
1699 } else {
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001700 head = &page->rcu_head;
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001701 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001702
1703 call_rcu(head, rcu_free_slab);
1704 } else
1705 __free_slab(s, page);
1706}
1707
1708static void discard_slab(struct kmem_cache *s, struct page *page)
1709{
Christoph Lameter205ab992008-04-14 19:11:40 +03001710 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001711 free_slab(s, page);
1712}
1713
1714/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001715 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001716 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001717static inline void
1718__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001719{
Christoph Lametere95eed52007-05-06 14:49:44 -07001720 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001721 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001722 list_add_tail(&page->lru, &n->partial);
1723 else
1724 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001725}
1726
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001727static inline void add_partial(struct kmem_cache_node *n,
1728 struct page *page, int tail)
1729{
1730 lockdep_assert_held(&n->list_lock);
1731 __add_partial(n, page, tail);
1732}
1733
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001734static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001735 struct page *page)
1736{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001737 lockdep_assert_held(&n->list_lock);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08001738 list_del(&page->lru);
1739 n->nr_partial--;
Christoph Lameter62e346a2010-09-28 08:10:28 -05001740}
1741
Christoph Lameter81819f02007-05-06 14:49:36 -07001742/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001743 * Remove slab from the partial list, freeze it and
1744 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001745 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001746 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001747 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001748static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001749 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001750 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001751{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001752 void *freelist;
1753 unsigned long counters;
1754 struct page new;
1755
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001756 lockdep_assert_held(&n->list_lock);
1757
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001758 /*
1759 * Zap the freelist and set the frozen bit.
1760 * The old freelist is the list of objects for the
1761 * per cpu allocation list.
1762 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001763 freelist = page->freelist;
1764 counters = page->counters;
1765 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001766 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001767 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001768 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001769 new.freelist = NULL;
1770 } else {
1771 new.freelist = freelist;
1772 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001773
Dave Hansena0132ac2014-01-29 14:05:50 -08001774 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001775 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001776
Christoph Lameter7ced3712012-05-09 10:09:53 -05001777 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001778 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001779 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001780 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001781 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001782
1783 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001784 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001785 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001786}
1787
Joonsoo Kim633b0762013-01-21 17:01:25 +09001788static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001789static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001790
Christoph Lameter81819f02007-05-06 14:49:36 -07001791/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001792 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001793 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001794static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1795 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001796{
Christoph Lameter49e22582011-08-09 16:12:27 -05001797 struct page *page, *page2;
1798 void *object = NULL;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001799 int available = 0;
1800 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001801
1802 /*
1803 * Racy check. If we mistakenly see no partial slabs then we
1804 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001805 * partial slab and there is none available then get_partials()
1806 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001807 */
1808 if (!n || !n->nr_partial)
1809 return NULL;
1810
1811 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001812 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001813 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001814
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001815 if (!pfmemalloc_match(page, flags))
1816 continue;
1817
Joonsoo Kim633b0762013-01-21 17:01:25 +09001818 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001819 if (!t)
1820 break;
1821
Joonsoo Kim633b0762013-01-21 17:01:25 +09001822 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001823 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001824 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001825 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001826 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001827 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001828 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001829 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001830 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001831 if (!kmem_cache_has_cpu_partial(s)
1832 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001833 break;
1834
Christoph Lameter497b66f2011-08-09 16:12:26 -05001835 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001836 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001837 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001838}
1839
1840/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001841 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001842 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001843static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001844 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001845{
1846#ifdef CONFIG_NUMA
1847 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001848 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001849 struct zone *zone;
1850 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001851 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001852 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001853
1854 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001855 * The defrag ratio allows a configuration of the tradeoffs between
1856 * inter node defragmentation and node local allocations. A lower
1857 * defrag_ratio increases the tendency to do local allocations
1858 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001859 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001860 * If the defrag_ratio is set to 0 then kmalloc() always
1861 * returns node local objects. If the ratio is higher then kmalloc()
1862 * may return off node objects because partial slabs are obtained
1863 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001864 *
Li Peng43efd3e2016-05-19 17:10:43 -07001865 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1866 * (which makes defrag_ratio = 1000) then every (well almost)
1867 * allocation will first attempt to defrag slab caches on other nodes.
1868 * This means scanning over all nodes to look for partial slabs which
1869 * may be expensive if we do it every time we are trying to find a slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001870 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001871 */
Christoph Lameter98246012008-01-07 23:20:26 -08001872 if (!s->remote_node_defrag_ratio ||
1873 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001874 return NULL;
1875
Mel Gormancc9a6c82012-03-21 16:34:11 -07001876 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001877 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001878 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001879 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1880 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001881
Mel Gormancc9a6c82012-03-21 16:34:11 -07001882 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001883
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001884 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001885 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001886 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001887 if (object) {
1888 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001889 * Don't check read_mems_allowed_retry()
1890 * here - if mems_allowed was updated in
1891 * parallel, that was a harmless race
1892 * between allocation and the cpuset
1893 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001894 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001895 return object;
1896 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001897 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001898 }
Mel Gormand26914d2014-04-03 14:47:24 -07001899 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001900#endif
1901 return NULL;
1902}
1903
1904/*
1905 * Get a partial page, lock it and return it.
1906 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001907static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001908 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001909{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001910 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001911 int searchnode = node;
1912
1913 if (node == NUMA_NO_NODE)
1914 searchnode = numa_mem_id();
1915 else if (!node_present_pages(node))
1916 searchnode = node_to_mem_node(node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001917
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001918 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001919 if (object || node != NUMA_NO_NODE)
1920 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001921
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001922 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001923}
1924
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001925#ifdef CONFIG_PREEMPT
1926/*
1927 * Calculate the next globally unique transaction for disambiguiation
1928 * during cmpxchg. The transactions start with the cpu number and are then
1929 * incremented by CONFIG_NR_CPUS.
1930 */
1931#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1932#else
1933/*
1934 * No preemption supported therefore also no need to check for
1935 * different cpus.
1936 */
1937#define TID_STEP 1
1938#endif
1939
1940static inline unsigned long next_tid(unsigned long tid)
1941{
1942 return tid + TID_STEP;
1943}
1944
1945static inline unsigned int tid_to_cpu(unsigned long tid)
1946{
1947 return tid % TID_STEP;
1948}
1949
1950static inline unsigned long tid_to_event(unsigned long tid)
1951{
1952 return tid / TID_STEP;
1953}
1954
1955static inline unsigned int init_tid(int cpu)
1956{
1957 return cpu;
1958}
1959
1960static inline void note_cmpxchg_failure(const char *n,
1961 const struct kmem_cache *s, unsigned long tid)
1962{
1963#ifdef SLUB_DEBUG_CMPXCHG
1964 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1965
Fabian Frederickf9f58282014-06-04 16:06:34 -07001966 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001967
1968#ifdef CONFIG_PREEMPT
1969 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001970 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001971 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1972 else
1973#endif
1974 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001975 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001976 tid_to_event(tid), tid_to_event(actual_tid));
1977 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001978 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001979 actual_tid, tid, next_tid(tid));
1980#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001981 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001982}
1983
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001984static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001985{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001986 int cpu;
1987
1988 for_each_possible_cpu(cpu)
1989 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001990}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001991
1992/*
1993 * Remove the cpu slab
1994 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08001995static void deactivate_slab(struct kmem_cache *s, struct page *page,
1996 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001997{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001998 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001999 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2000 int lock = 0;
2001 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002002 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08002003 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002004 struct page new;
2005 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002006
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002007 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002008 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08002009 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07002010 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002011
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002012 /*
2013 * Stage one: Free all available per cpu objects back
2014 * to the page freelist while it is still frozen. Leave the
2015 * last one.
2016 *
2017 * There is no need to take the list->lock because the page
2018 * is still frozen.
2019 */
2020 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2021 void *prior;
2022 unsigned long counters;
2023
2024 do {
2025 prior = page->freelist;
2026 counters = page->counters;
2027 set_freepointer(s, freelist, prior);
2028 new.counters = counters;
2029 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08002030 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002031
Christoph Lameter1d071712011-07-14 12:49:12 -05002032 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002033 prior, counters,
2034 freelist, new.counters,
2035 "drain percpu freelist"));
2036
2037 freelist = nextfree;
2038 }
2039
2040 /*
2041 * Stage two: Ensure that the page is unfrozen while the
2042 * list presence reflects the actual number of objects
2043 * during unfreeze.
2044 *
2045 * We setup the list membership and then perform a cmpxchg
2046 * with the count. If there is a mismatch then the page
2047 * is not unfrozen but the page is on the wrong list.
2048 *
2049 * Then we restart the process which may have to remove
2050 * the page from the list that we just put it on again
2051 * because the number of objects in the slab may have
2052 * changed.
2053 */
2054redo:
2055
2056 old.freelist = page->freelist;
2057 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002058 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002059
2060 /* Determine target state of the slab */
2061 new.counters = old.counters;
2062 if (freelist) {
2063 new.inuse--;
2064 set_freepointer(s, freelist, old.freelist);
2065 new.freelist = freelist;
2066 } else
2067 new.freelist = old.freelist;
2068
2069 new.frozen = 0;
2070
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002071 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002072 m = M_FREE;
2073 else if (new.freelist) {
2074 m = M_PARTIAL;
2075 if (!lock) {
2076 lock = 1;
2077 /*
2078 * Taking the spinlock removes the possiblity
2079 * that acquire_slab() will see a slab page that
2080 * is frozen
2081 */
2082 spin_lock(&n->list_lock);
2083 }
2084 } else {
2085 m = M_FULL;
2086 if (kmem_cache_debug(s) && !lock) {
2087 lock = 1;
2088 /*
2089 * This also ensures that the scanning of full
2090 * slabs from diagnostic functions will not see
2091 * any frozen slabs.
2092 */
2093 spin_lock(&n->list_lock);
2094 }
2095 }
2096
2097 if (l != m) {
2098
2099 if (l == M_PARTIAL)
2100
2101 remove_partial(n, page);
2102
2103 else if (l == M_FULL)
2104
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002105 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002106
2107 if (m == M_PARTIAL) {
2108
2109 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08002110 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002111
2112 } else if (m == M_FULL) {
2113
2114 stat(s, DEACTIVATE_FULL);
2115 add_full(s, n, page);
2116
2117 }
2118 }
2119
2120 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05002121 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002122 old.freelist, old.counters,
2123 new.freelist, new.counters,
2124 "unfreezing slab"))
2125 goto redo;
2126
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002127 if (lock)
2128 spin_unlock(&n->list_lock);
2129
2130 if (m == M_FREE) {
2131 stat(s, DEACTIVATE_EMPTY);
2132 discard_slab(s, page);
2133 stat(s, FREE_SLAB);
2134 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002135}
2136
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002137/*
2138 * Unfreeze all the cpu partial slabs.
2139 *
Christoph Lameter59a09912012-11-28 16:23:00 +00002140 * This function must be called with interrupts disabled
2141 * for the cpu using c (or some other guarantee must be there
2142 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002143 */
Christoph Lameter59a09912012-11-28 16:23:00 +00002144static void unfreeze_partials(struct kmem_cache *s,
2145 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05002146{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002147#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09002148 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08002149 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002150
2151 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002152 struct page new;
2153 struct page old;
2154
2155 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002156
2157 n2 = get_node(s, page_to_nid(page));
2158 if (n != n2) {
2159 if (n)
2160 spin_unlock(&n->list_lock);
2161
2162 n = n2;
2163 spin_lock(&n->list_lock);
2164 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002165
2166 do {
2167
2168 old.freelist = page->freelist;
2169 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002170 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05002171
2172 new.counters = old.counters;
2173 new.freelist = old.freelist;
2174
2175 new.frozen = 0;
2176
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002177 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002178 old.freelist, old.counters,
2179 new.freelist, new.counters,
2180 "unfreezing slab"));
2181
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002182 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002183 page->next = discard_page;
2184 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002185 } else {
2186 add_partial(n, page, DEACTIVATE_TO_TAIL);
2187 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002188 }
2189 }
2190
2191 if (n)
2192 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002193
2194 while (discard_page) {
2195 page = discard_page;
2196 discard_page = discard_page->next;
2197
2198 stat(s, DEACTIVATE_EMPTY);
2199 discard_slab(s, page);
2200 stat(s, FREE_SLAB);
2201 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002202#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002203}
2204
2205/*
2206 * Put a page that was just frozen (in __slab_free) into a partial page
2207 * slot if available. This is done without interrupts disabled and without
2208 * preemption disabled. The cmpxchg is racy and may put the partial page
2209 * onto a random cpus partial slot.
2210 *
2211 * If we did not find a slot then simply move all the partials to the
2212 * per node partial list.
2213 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002214static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002215{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002216#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002217 struct page *oldpage;
2218 int pages;
2219 int pobjects;
2220
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002221 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002222 do {
2223 pages = 0;
2224 pobjects = 0;
2225 oldpage = this_cpu_read(s->cpu_slab->partial);
2226
2227 if (oldpage) {
2228 pobjects = oldpage->pobjects;
2229 pages = oldpage->pages;
2230 if (drain && pobjects > s->cpu_partial) {
2231 unsigned long flags;
2232 /*
2233 * partial array is full. Move the existing
2234 * set to the per node partial list.
2235 */
2236 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002237 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002238 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002239 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002240 pobjects = 0;
2241 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002242 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002243 }
2244 }
2245
2246 pages++;
2247 pobjects += page->objects - page->inuse;
2248
2249 page->pages = pages;
2250 page->pobjects = pobjects;
2251 page->next = oldpage;
2252
Chen Gangd0e0ac92013-07-15 09:05:29 +08002253 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2254 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002255 if (unlikely(!s->cpu_partial)) {
2256 unsigned long flags;
2257
2258 local_irq_save(flags);
2259 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2260 local_irq_restore(flags);
2261 }
2262 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002263#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002264}
2265
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002266static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002267{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002268 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002269 deactivate_slab(s, c->page, c->freelist);
2270
2271 c->tid = next_tid(c->tid);
2272 c->page = NULL;
2273 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002274}
2275
2276/*
2277 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002278 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002279 * Called from IPI handler with interrupts disabled.
2280 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002281static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002282{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002283 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002284
Christoph Lameter49e22582011-08-09 16:12:27 -05002285 if (likely(c)) {
2286 if (c->page)
2287 flush_slab(s, c);
2288
Christoph Lameter59a09912012-11-28 16:23:00 +00002289 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002290 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002291}
2292
2293static void flush_cpu_slab(void *d)
2294{
2295 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002296
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002297 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002298}
2299
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002300static bool has_cpu_slab(int cpu, void *info)
2301{
2302 struct kmem_cache *s = info;
2303 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2304
majianpeng02e1a9c2012-05-17 17:03:26 -07002305 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002306}
2307
Christoph Lameter81819f02007-05-06 14:49:36 -07002308static void flush_all(struct kmem_cache *s)
2309{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002310 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002311}
2312
2313/*
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02002314 * Use the cpu notifier to insure that the cpu slabs are flushed when
2315 * necessary.
2316 */
2317static int slub_cpu_dead(unsigned int cpu)
2318{
2319 struct kmem_cache *s;
2320 unsigned long flags;
2321
2322 mutex_lock(&slab_mutex);
2323 list_for_each_entry(s, &slab_caches, list) {
2324 local_irq_save(flags);
2325 __flush_cpu_slab(s, cpu);
2326 local_irq_restore(flags);
2327 }
2328 mutex_unlock(&slab_mutex);
2329 return 0;
2330}
2331
2332/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002333 * Check if the objects in a per cpu structure fit numa
2334 * locality expectations.
2335 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002336static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002337{
2338#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002339 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002340 return 0;
2341#endif
2342 return 1;
2343}
2344
David Rientjes9a02d692014-06-04 16:06:36 -07002345#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002346static int count_free(struct page *page)
2347{
2348 return page->objects - page->inuse;
2349}
2350
David Rientjes9a02d692014-06-04 16:06:36 -07002351static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2352{
2353 return atomic_long_read(&n->total_objects);
2354}
2355#endif /* CONFIG_SLUB_DEBUG */
2356
2357#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002358static unsigned long count_partial(struct kmem_cache_node *n,
2359 int (*get_count)(struct page *))
2360{
2361 unsigned long flags;
2362 unsigned long x = 0;
2363 struct page *page;
2364
2365 spin_lock_irqsave(&n->list_lock, flags);
2366 list_for_each_entry(page, &n->partial, lru)
2367 x += get_count(page);
2368 spin_unlock_irqrestore(&n->list_lock, flags);
2369 return x;
2370}
David Rientjes9a02d692014-06-04 16:06:36 -07002371#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002372
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002373static noinline void
2374slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2375{
David Rientjes9a02d692014-06-04 16:06:36 -07002376#ifdef CONFIG_SLUB_DEBUG
2377 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2378 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002379 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002380 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002381
David Rientjes9a02d692014-06-04 16:06:36 -07002382 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2383 return;
2384
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07002385 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2386 nid, gfpflags, &gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002387 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2388 s->name, s->object_size, s->size, oo_order(s->oo),
2389 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002390
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002391 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002392 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2393 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002394
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002395 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002396 unsigned long nr_slabs;
2397 unsigned long nr_objs;
2398 unsigned long nr_free;
2399
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002400 nr_free = count_partial(n, count_free);
2401 nr_slabs = node_nr_slabs(n);
2402 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002403
Fabian Frederickf9f58282014-06-04 16:06:34 -07002404 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002405 node, nr_slabs, nr_objs, nr_free);
2406 }
David Rientjes9a02d692014-06-04 16:06:36 -07002407#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002408}
2409
Christoph Lameter497b66f2011-08-09 16:12:26 -05002410static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2411 int node, struct kmem_cache_cpu **pc)
2412{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002413 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002414 struct kmem_cache_cpu *c = *pc;
2415 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002416
Christoph Lameter188fd062012-05-09 10:09:55 -05002417 freelist = get_partial(s, flags, node, c);
2418
2419 if (freelist)
2420 return freelist;
2421
2422 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002423 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002424 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002425 if (c->page)
2426 flush_slab(s, c);
2427
2428 /*
2429 * No other reference to the page yet so we can
2430 * muck around with it freely without cmpxchg
2431 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002432 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002433 page->freelist = NULL;
2434
2435 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002436 c->page = page;
2437 *pc = c;
2438 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002439 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002440
Christoph Lameter6faa6832012-05-09 10:09:51 -05002441 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002442}
2443
Mel Gorman072bb0a2012-07-31 16:43:58 -07002444static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2445{
2446 if (unlikely(PageSlabPfmemalloc(page)))
2447 return gfp_pfmemalloc_allowed(gfpflags);
2448
2449 return true;
2450}
2451
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002452/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002453 * Check the page->freelist of a page and either transfer the freelist to the
2454 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002455 *
2456 * The page is still frozen if the return value is not NULL.
2457 *
2458 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002459 *
2460 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002461 */
2462static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2463{
2464 struct page new;
2465 unsigned long counters;
2466 void *freelist;
2467
2468 do {
2469 freelist = page->freelist;
2470 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002471
Christoph Lameter213eeb92011-11-11 14:07:14 -06002472 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002473 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002474
2475 new.inuse = page->objects;
2476 new.frozen = freelist != NULL;
2477
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002478 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002479 freelist, counters,
2480 NULL, new.counters,
2481 "get_freelist"));
2482
2483 return freelist;
2484}
2485
2486/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002487 * Slow path. The lockless freelist is empty or we need to perform
2488 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002489 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002490 * Processing is still very fast if new objects have been freed to the
2491 * regular freelist. In that case we simply take over the regular freelist
2492 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002493 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002494 * If that is not working then we fall back to the partial lists. We take the
2495 * first element of the freelist as the object to allocate now and move the
2496 * rest of the freelist to the lockless freelist.
2497 *
2498 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002499 * we need to allocate a new slab. This is the slowest path since it involves
2500 * a call to the page allocator and the setup of a new slab.
Christoph Lametera380a3c2015-11-20 15:57:35 -08002501 *
2502 * Version of __slab_alloc to use when we know that interrupts are
2503 * already disabled (which is the case for bulk allocation).
Christoph Lameter81819f02007-05-06 14:49:36 -07002504 */
Christoph Lametera380a3c2015-11-20 15:57:35 -08002505static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002506 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002507{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002508 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002509 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002510
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002511 page = c->page;
2512 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002513 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002514redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002515
Christoph Lameter57d437d2012-05-09 10:09:59 -05002516 if (unlikely(!node_match(page, node))) {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002517 int searchnode = node;
2518
2519 if (node != NUMA_NO_NODE && !node_present_pages(node))
2520 searchnode = node_to_mem_node(node);
2521
2522 if (unlikely(!node_match(page, searchnode))) {
2523 stat(s, ALLOC_NODE_MISMATCH);
2524 deactivate_slab(s, page, c->freelist);
2525 c->page = NULL;
2526 c->freelist = NULL;
2527 goto new_slab;
2528 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002529 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002530
Mel Gorman072bb0a2012-07-31 16:43:58 -07002531 /*
2532 * By rights, we should be searching for a slab page that was
2533 * PFMEMALLOC but right now, we are losing the pfmemalloc
2534 * information when the page leaves the per-cpu allocator
2535 */
2536 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2537 deactivate_slab(s, page, c->freelist);
2538 c->page = NULL;
2539 c->freelist = NULL;
2540 goto new_slab;
2541 }
2542
Eric Dumazet73736e02011-12-13 04:57:06 +01002543 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002544 freelist = c->freelist;
2545 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002546 goto load_freelist;
2547
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002548 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002549
Christoph Lameter6faa6832012-05-09 10:09:51 -05002550 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002551 c->page = NULL;
2552 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002553 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002554 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002555
Christoph Lameter81819f02007-05-06 14:49:36 -07002556 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002557
Christoph Lameter894b8782007-05-10 03:15:16 -07002558load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002559 /*
2560 * freelist is pointing to the list of objects to be used.
2561 * page is pointing to the page from which the objects are obtained.
2562 * That page must be frozen for per cpu allocations to work.
2563 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002564 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002565 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002566 c->tid = next_tid(c->tid);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002567 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002568
Christoph Lameter81819f02007-05-06 14:49:36 -07002569new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002570
Christoph Lameter49e22582011-08-09 16:12:27 -05002571 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002572 page = c->page = c->partial;
2573 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002574 stat(s, CPU_PARTIAL_ALLOC);
2575 c->freelist = NULL;
2576 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002577 }
2578
Christoph Lameter188fd062012-05-09 10:09:55 -05002579 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002580
Christoph Lameterf46974362012-05-09 10:09:54 -05002581 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002582 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002583 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002584 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002585
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002586 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002587 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002588 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002589
Christoph Lameter497b66f2011-08-09 16:12:26 -05002590 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002591 if (kmem_cache_debug(s) &&
2592 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002593 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002594
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002595 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002596 c->page = NULL;
2597 c->freelist = NULL;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002598 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002599}
2600
2601/*
Christoph Lametera380a3c2015-11-20 15:57:35 -08002602 * Another one that disabled interrupt and compensates for possible
2603 * cpu changes by refetching the per cpu area pointer.
2604 */
2605static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2606 unsigned long addr, struct kmem_cache_cpu *c)
2607{
2608 void *p;
2609 unsigned long flags;
2610
2611 local_irq_save(flags);
2612#ifdef CONFIG_PREEMPT
2613 /*
2614 * We may have been preempted and rescheduled on a different
2615 * cpu before disabling interrupts. Need to reload cpu area
2616 * pointer.
2617 */
2618 c = this_cpu_ptr(s->cpu_slab);
2619#endif
2620
2621 p = ___slab_alloc(s, gfpflags, node, addr, c);
2622 local_irq_restore(flags);
2623 return p;
2624}
2625
2626/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002627 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2628 * have the fastpath folded into their functions. So no function call
2629 * overhead for requests that can be satisfied on the fastpath.
2630 *
2631 * The fastpath works by first checking if the lockless freelist can be used.
2632 * If not then __slab_alloc is called for slow processing.
2633 *
2634 * Otherwise we can simply pick the next object from the lockless free list.
2635 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002636static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002637 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002638{
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002639 void *object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002640 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002641 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002642 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002643
Vladimir Davydov8135be52014-12-12 16:56:38 -08002644 s = slab_pre_alloc_hook(s, gfpflags);
2645 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002646 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002647redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002648 /*
2649 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2650 * enabled. We may switch back and forth between cpus while
2651 * reading from one cpu area. That does not matter as long
2652 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002653 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002654 * We should guarantee that tid and kmem_cache are retrieved on
2655 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2656 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002657 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002658 do {
2659 tid = this_cpu_read(s->cpu_slab->tid);
2660 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002661 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2662 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002663
2664 /*
2665 * Irqless object alloc/free algorithm used here depends on sequence
2666 * of fetching cpu_slab's data. tid should be fetched before anything
2667 * on c to guarantee that object and page associated with previous tid
2668 * won't be used with current tid. If we fetch tid first, object and
2669 * page could be one associated with next tid and our alloc/free
2670 * request will be failed. In this case, we will retry. So, no problem.
2671 */
2672 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002673
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002674 /*
2675 * The transaction ids are globally unique per cpu and per operation on
2676 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2677 * occurs on the right processor and that there was no operation on the
2678 * linked list in between.
2679 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002680
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002681 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002682 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002683 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002684 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002685 stat(s, ALLOC_SLOWPATH);
2686 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002687 void *next_object = get_freepointer_safe(s, object);
2688
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002689 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002690 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002691 * operation and if we are on the right processor.
2692 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002693 * The cmpxchg does the following atomically (without lock
2694 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002695 * 1. Relocate first pointer to the current per cpu area.
2696 * 2. Verify that tid and freelist have not been changed
2697 * 3. If they were not changed replace tid and freelist
2698 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002699 * Since this is without lock semantics the protection is only
2700 * against code executing on this cpu *not* from access by
2701 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002702 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002703 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002704 s->cpu_slab->freelist, s->cpu_slab->tid,
2705 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002706 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002707
2708 note_cmpxchg_failure("slab_alloc", s, tid);
2709 goto redo;
2710 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002711 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002712 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002713 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002714
Pekka Enberg74e21342009-11-25 20:14:48 +02002715 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002716 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002717
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002718 slab_post_alloc_hook(s, gfpflags, 1, &object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002719
Christoph Lameter894b8782007-05-10 03:15:16 -07002720 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002721}
2722
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002723static __always_inline void *slab_alloc(struct kmem_cache *s,
2724 gfp_t gfpflags, unsigned long addr)
2725{
2726 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2727}
2728
Christoph Lameter81819f02007-05-06 14:49:36 -07002729void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2730{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002731 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002732
Chen Gangd0e0ac92013-07-15 09:05:29 +08002733 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2734 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002735
2736 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002737}
2738EXPORT_SYMBOL(kmem_cache_alloc);
2739
Li Zefan0f24f122009-12-11 15:45:30 +08002740#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002741void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002742{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002743 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002744 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002745 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002746 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002747}
Richard Kennedy4a923792010-10-21 10:29:19 +01002748EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002749#endif
2750
Christoph Lameter81819f02007-05-06 14:49:36 -07002751#ifdef CONFIG_NUMA
2752void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2753{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002754 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002755
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002756 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002757 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002758
2759 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002760}
2761EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002762
Li Zefan0f24f122009-12-11 15:45:30 +08002763#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002764void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002765 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002766 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002767{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002768 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002769
2770 trace_kmalloc_node(_RET_IP_, ret,
2771 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002772
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002773 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002774 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002775}
Richard Kennedy4a923792010-10-21 10:29:19 +01002776EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002777#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002778#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002779
Christoph Lameter81819f02007-05-06 14:49:36 -07002780/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002781 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002782 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002783 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002784 * So we still attempt to reduce cache line usage. Just take the slab
2785 * lock and free the item. If there is no additional partial page
2786 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002787 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002788static void __slab_free(struct kmem_cache *s, struct page *page,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002789 void *head, void *tail, int cnt,
2790 unsigned long addr)
2791
Christoph Lameter81819f02007-05-06 14:49:36 -07002792{
2793 void *prior;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002794 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002795 struct page new;
2796 unsigned long counters;
2797 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002798 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002799
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002800 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002801
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002802 if (kmem_cache_debug(s) &&
Laura Abbott282acb42016-03-15 14:54:59 -07002803 !free_debug_processing(s, page, head, tail, cnt, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002804 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002805
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002806 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002807 if (unlikely(n)) {
2808 spin_unlock_irqrestore(&n->list_lock, flags);
2809 n = NULL;
2810 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002811 prior = page->freelist;
2812 counters = page->counters;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002813 set_freepointer(s, tail, prior);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002814 new.counters = counters;
2815 was_frozen = new.frozen;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002816 new.inuse -= cnt;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002817 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002818
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002819 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002820
2821 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002822 * Slab was on no list before and will be
2823 * partially empty
2824 * We can defer the list move and instead
2825 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002826 */
2827 new.frozen = 1;
2828
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002829 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002830
LQYMGTb455def2014-12-10 15:42:13 -08002831 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002832 /*
2833 * Speculatively acquire the list_lock.
2834 * If the cmpxchg does not succeed then we may
2835 * drop the list_lock without any processing.
2836 *
2837 * Otherwise the list_lock will synchronize with
2838 * other processors updating the list of slabs.
2839 */
2840 spin_lock_irqsave(&n->list_lock, flags);
2841
2842 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002843 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002844
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002845 } while (!cmpxchg_double_slab(s, page,
2846 prior, counters,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002847 head, new.counters,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002848 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002849
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002850 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002851
2852 /*
2853 * If we just froze the page then put it onto the
2854 * per cpu partial list.
2855 */
Alex Shi8028dce2012-02-03 23:34:56 +08002856 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002857 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002858 stat(s, CPU_PARTIAL_FREE);
2859 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002860 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002861 * The list lock was not taken therefore no list
2862 * activity can be necessary.
2863 */
LQYMGTb455def2014-12-10 15:42:13 -08002864 if (was_frozen)
2865 stat(s, FREE_FROZEN);
2866 return;
2867 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002868
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002869 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002870 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002871
Joonsoo Kim837d6782012-08-16 00:02:40 +09002872 /*
2873 * Objects left in the slab. If it was not on the partial list before
2874 * then add it.
2875 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002876 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2877 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002878 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002879 add_partial(n, page, DEACTIVATE_TO_TAIL);
2880 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002881 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002882 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002883 return;
2884
2885slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002886 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002887 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002888 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002889 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002890 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002891 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002892 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002893 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002894 remove_full(s, n, page);
2895 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002896
Christoph Lameter80f08c12011-06-01 12:25:55 -05002897 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002898 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002899 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002900}
2901
Christoph Lameter894b8782007-05-10 03:15:16 -07002902/*
2903 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2904 * can perform fastpath freeing without additional function calls.
2905 *
2906 * The fastpath is only possible if we are freeing to the current cpu slab
2907 * of this processor. This typically the case if we have just allocated
2908 * the item before.
2909 *
2910 * If fastpath is not possible then fall back to __slab_free where we deal
2911 * with all sorts of special processing.
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002912 *
2913 * Bulk free of a freelist with several objects (all pointing to the
2914 * same page) possible by specifying head and tail ptr, plus objects
2915 * count (cnt). Bulk free indicated by tail pointer being set.
Christoph Lameter894b8782007-05-10 03:15:16 -07002916 */
Alexander Potapenko80a92012016-07-28 15:49:07 -07002917static __always_inline void do_slab_free(struct kmem_cache *s,
2918 struct page *page, void *head, void *tail,
2919 int cnt, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002920{
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002921 void *tail_obj = tail ? : head;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002922 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002923 unsigned long tid;
Christoph Lametera24c5a02011-03-15 12:45:21 -05002924redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002925 /*
2926 * Determine the currently cpus per cpu slab.
2927 * The cpu may change afterward. However that does not matter since
2928 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07002929 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002930 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002931 do {
2932 tid = this_cpu_read(s->cpu_slab->tid);
2933 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002934 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2935 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002936
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002937 /* Same with comment on barrier() in slab_alloc_node() */
2938 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002939
Christoph Lameter442b06b2011-05-17 16:29:31 -05002940 if (likely(page == c->page)) {
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002941 set_freepointer(s, tail_obj, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002942
Christoph Lameter933393f2011-12-22 11:58:51 -06002943 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002944 s->cpu_slab->freelist, s->cpu_slab->tid,
2945 c->freelist, tid,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002946 head, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002947
2948 note_cmpxchg_failure("slab_free", s, tid);
2949 goto redo;
2950 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002951 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002952 } else
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002953 __slab_free(s, page, head, tail_obj, cnt, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002954
Christoph Lameter894b8782007-05-10 03:15:16 -07002955}
2956
Alexander Potapenko80a92012016-07-28 15:49:07 -07002957static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2958 void *head, void *tail, int cnt,
2959 unsigned long addr)
2960{
2961 slab_free_freelist_hook(s, head, tail);
2962 /*
2963 * slab_free_freelist_hook() could have put the items into quarantine.
2964 * If so, no need to free them.
2965 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08002966 if (s->flags & SLAB_KASAN && !(s->flags & SLAB_TYPESAFE_BY_RCU))
Alexander Potapenko80a92012016-07-28 15:49:07 -07002967 return;
2968 do_slab_free(s, page, head, tail, cnt, addr);
2969}
2970
2971#ifdef CONFIG_KASAN
2972void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2973{
2974 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2975}
2976#endif
2977
Christoph Lameter81819f02007-05-06 14:49:36 -07002978void kmem_cache_free(struct kmem_cache *s, void *x)
2979{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002980 s = cache_from_obj(s, x);
2981 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002982 return;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002983 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002984 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002985}
2986EXPORT_SYMBOL(kmem_cache_free);
2987
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08002988struct detached_freelist {
2989 struct page *page;
2990 void *tail;
2991 void *freelist;
2992 int cnt;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07002993 struct kmem_cache *s;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08002994};
2995
2996/*
2997 * This function progressively scans the array with free objects (with
2998 * a limited look ahead) and extract objects belonging to the same
2999 * page. It builds a detached freelist directly within the given
3000 * page/objects. This can happen without any need for
3001 * synchronization, because the objects are owned by running process.
3002 * The freelist is build up as a single linked list in the objects.
3003 * The idea is, that this detached freelist can then be bulk
3004 * transferred to the real freelist(s), but only requiring a single
3005 * synchronization primitive. Look ahead in the array is limited due
3006 * to performance reasons.
3007 */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003008static inline
3009int build_detached_freelist(struct kmem_cache *s, size_t size,
3010 void **p, struct detached_freelist *df)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003011{
3012 size_t first_skipped_index = 0;
3013 int lookahead = 3;
3014 void *object;
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003015 struct page *page;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003016
3017 /* Always re-init detached_freelist */
3018 df->page = NULL;
3019
3020 do {
3021 object = p[--size];
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003022 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003023 } while (!object && size);
3024
3025 if (!object)
3026 return 0;
3027
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003028 page = virt_to_head_page(object);
3029 if (!s) {
3030 /* Handle kalloc'ed objects */
3031 if (unlikely(!PageSlab(page))) {
3032 BUG_ON(!PageCompound(page));
3033 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003034 __free_pages(page, compound_order(page));
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003035 p[size] = NULL; /* mark object processed */
3036 return size;
3037 }
3038 /* Derive kmem_cache from object */
3039 df->s = page->slab_cache;
3040 } else {
3041 df->s = cache_from_obj(s, object); /* Support for memcg */
3042 }
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003043
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003044 /* Start new detached freelist */
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003045 df->page = page;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003046 set_freepointer(df->s, object, NULL);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003047 df->tail = object;
3048 df->freelist = object;
3049 p[size] = NULL; /* mark object processed */
3050 df->cnt = 1;
3051
3052 while (size) {
3053 object = p[--size];
3054 if (!object)
3055 continue; /* Skip processed objects */
3056
3057 /* df->page is always set at this point */
3058 if (df->page == virt_to_head_page(object)) {
3059 /* Opportunity build freelist */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003060 set_freepointer(df->s, object, df->freelist);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003061 df->freelist = object;
3062 df->cnt++;
3063 p[size] = NULL; /* mark object processed */
3064
3065 continue;
3066 }
3067
3068 /* Limit look ahead search */
3069 if (!--lookahead)
3070 break;
3071
3072 if (!first_skipped_index)
3073 first_skipped_index = size + 1;
3074 }
3075
3076 return first_skipped_index;
3077}
3078
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003079/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003080void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003081{
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003082 if (WARN_ON(!size))
3083 return;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003084
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003085 do {
3086 struct detached_freelist df;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003087
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003088 size = build_detached_freelist(s, size, p, &df);
Arnd Bergmann84582c82016-12-12 16:41:35 -08003089 if (!df.page)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003090 continue;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003091
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003092 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003093 } while (likely(size));
Christoph Lameter484748f2015-09-04 15:45:34 -07003094}
3095EXPORT_SYMBOL(kmem_cache_free_bulk);
3096
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003097/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003098int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3099 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003100{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003101 struct kmem_cache_cpu *c;
3102 int i;
3103
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003104 /* memcg and kmem_cache debug support */
3105 s = slab_pre_alloc_hook(s, flags);
3106 if (unlikely(!s))
3107 return false;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003108 /*
3109 * Drain objects in the per cpu slab, while disabling local
3110 * IRQs, which protects against PREEMPT and interrupts
3111 * handlers invoking normal fastpath.
3112 */
3113 local_irq_disable();
3114 c = this_cpu_ptr(s->cpu_slab);
3115
3116 for (i = 0; i < size; i++) {
3117 void *object = c->freelist;
3118
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003119 if (unlikely(!object)) {
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003120 /*
3121 * Invoking slow path likely have side-effect
3122 * of re-populating per CPU c->freelist
3123 */
Christoph Lameter87098372015-11-20 15:57:38 -08003124 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003125 _RET_IP_, c);
Christoph Lameter87098372015-11-20 15:57:38 -08003126 if (unlikely(!p[i]))
3127 goto error;
3128
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003129 c = this_cpu_ptr(s->cpu_slab);
3130 continue; /* goto for-loop */
3131 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003132 c->freelist = get_freepointer(s, object);
3133 p[i] = object;
3134 }
3135 c->tid = next_tid(c->tid);
3136 local_irq_enable();
3137
3138 /* Clear memory outside IRQ disabled fastpath loop */
3139 if (unlikely(flags & __GFP_ZERO)) {
3140 int j;
3141
3142 for (j = 0; j < i; j++)
3143 memset(p[j], 0, s->object_size);
3144 }
3145
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003146 /* memcg and kmem_cache debug support */
3147 slab_post_alloc_hook(s, flags, size, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003148 return i;
Christoph Lameter87098372015-11-20 15:57:38 -08003149error:
Christoph Lameter87098372015-11-20 15:57:38 -08003150 local_irq_enable();
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003151 slab_post_alloc_hook(s, flags, i, p);
3152 __kmem_cache_free_bulk(s, i, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003153 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003154}
3155EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3156
3157
Christoph Lameter81819f02007-05-06 14:49:36 -07003158/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003159 * Object placement in a slab is made very easy because we always start at
3160 * offset 0. If we tune the size of the object to the alignment then we can
3161 * get the required alignment by putting one properly sized object after
3162 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07003163 *
3164 * Notice that the allocation order determines the sizes of the per cpu
3165 * caches. Each processor has always one slab available for allocations.
3166 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07003167 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07003168 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07003169 */
3170
3171/*
3172 * Mininum / Maximum order of slab pages. This influences locking overhead
3173 * and slab fragmentation. A higher order reduces the number of partial slabs
3174 * and increases the number of allocations possible without having to
3175 * take the list_lock.
3176 */
3177static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03003178static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003179static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07003180
3181/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003182 * Calculate the order of allocation given an slab object size.
3183 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003184 * The order of allocation has significant impact on performance and other
3185 * system components. Generally order 0 allocations should be preferred since
3186 * order 0 does not cause fragmentation in the page allocator. Larger objects
3187 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003188 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07003189 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07003190 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003191 * In order to reach satisfactory performance we must ensure that a minimum
3192 * number of objects is in one slab. Otherwise we may generate too much
3193 * activity on the partial lists which requires taking the list_lock. This is
3194 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07003195 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003196 * slub_max_order specifies the order where we begin to stop considering the
3197 * number of objects in a slab as critical. If we reach slub_max_order then
3198 * we try to keep the page order as low as possible. So we accept more waste
3199 * of space in favor of a small page order.
3200 *
3201 * Higher order allocations also allow the placement of more objects in a
3202 * slab and thereby reduce object handling overhead. If the user has
3203 * requested a higher mininum order then we start with that one instead of
3204 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07003205 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003206static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003207 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07003208{
3209 int order;
3210 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07003211 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003212
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003213 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04003214 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03003215
Wei Yang9f835702015-11-05 18:45:51 -08003216 for (order = max(min_order, get_order(min_objects * size + reserved));
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003217 order <= max_order; order++) {
3218
Christoph Lameter81819f02007-05-06 14:49:36 -07003219 unsigned long slab_size = PAGE_SIZE << order;
3220
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003221 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003222
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003223 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003224 break;
Christoph Lameter81819f02007-05-06 14:49:36 -07003225 }
Christoph Lameter672bba32007-05-09 02:32:39 -07003226
Christoph Lameter81819f02007-05-06 14:49:36 -07003227 return order;
3228}
3229
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003230static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003231{
3232 int order;
3233 int min_objects;
3234 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003235 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003236
3237 /*
3238 * Attempt to find best configuration for a slab. This
3239 * works by first attempting to generate a layout with
3240 * the best configuration and backing off gradually.
3241 *
Wei Yang422ff4d2015-11-05 18:45:46 -08003242 * First we increase the acceptable waste in a slab. Then
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003243 * we reduce the minimum objects required in a slab.
3244 */
3245 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003246 if (!min_objects)
3247 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003248 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003249 min_objects = min(min_objects, max_objects);
3250
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003251 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003252 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003253 while (fraction >= 4) {
3254 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003255 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003256 if (order <= slub_max_order)
3257 return order;
3258 fraction /= 2;
3259 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03003260 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003261 }
3262
3263 /*
3264 * We were unable to place multiple objects in a slab. Now
3265 * lets see if we can place a single object there.
3266 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003267 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003268 if (order <= slub_max_order)
3269 return order;
3270
3271 /*
3272 * Doh this slab cannot be placed using slub_max_order.
3273 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003274 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03003275 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003276 return order;
3277 return -ENOSYS;
3278}
3279
Pekka Enberg5595cff2008-08-05 09:28:47 +03003280static void
Joonsoo Kim40534972012-05-11 00:50:47 +09003281init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003282{
3283 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003284 spin_lock_init(&n->list_lock);
3285 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003286#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003287 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07003288 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07003289 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003290#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003291}
3292
Christoph Lameter55136592010-08-20 12:37:13 -05003293static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003294{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05003295 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003296 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003297
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003298 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003299 * Must align to double word boundary for the double cmpxchg
3300 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003301 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003302 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3303 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003304
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003305 if (!s->cpu_slab)
3306 return 0;
3307
3308 init_kmem_cache_cpus(s);
3309
3310 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003311}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003312
Christoph Lameter51df1142010-08-20 12:37:15 -05003313static struct kmem_cache *kmem_cache_node;
3314
Christoph Lameter81819f02007-05-06 14:49:36 -07003315/*
3316 * No kmalloc_node yet so do it by hand. We know that this is the first
3317 * slab on the node for this slabcache. There are no concurrent accesses
3318 * possible.
3319 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003320 * Note that this function only works on the kmem_cache_node
3321 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003322 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003323 */
Christoph Lameter55136592010-08-20 12:37:13 -05003324static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003325{
3326 struct page *page;
3327 struct kmem_cache_node *n;
3328
Christoph Lameter51df1142010-08-20 12:37:15 -05003329 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003330
Christoph Lameter51df1142010-08-20 12:37:15 -05003331 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003332
3333 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003334 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003335 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3336 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003337 }
3338
Christoph Lameter81819f02007-05-06 14:49:36 -07003339 n = page->freelist;
3340 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003341 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05003342 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05003343 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003344 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07003345#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003346 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003347 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003348#endif
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003349 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3350 GFP_KERNEL);
Joonsoo Kim40534972012-05-11 00:50:47 +09003351 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003352 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003353
Dave Hansen67b6c902014-01-24 07:20:23 -08003354 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003355 * No locks need to be taken here as it has just been
3356 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003357 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003358 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003359}
3360
3361static void free_kmem_cache_nodes(struct kmem_cache *s)
3362{
3363 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003364 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003365
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003366 for_each_kmem_cache_node(s, node, n) {
3367 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003368 s->node[node] = NULL;
3369 }
3370}
3371
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003372void __kmem_cache_release(struct kmem_cache *s)
3373{
Thomas Garnier210e7a42016-07-26 15:21:59 -07003374 cache_random_seq_destroy(s);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003375 free_percpu(s->cpu_slab);
3376 free_kmem_cache_nodes(s);
3377}
3378
Christoph Lameter55136592010-08-20 12:37:13 -05003379static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003380{
3381 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003382
Christoph Lameterf64dc582007-10-16 01:25:33 -07003383 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003384 struct kmem_cache_node *n;
3385
Alexander Duyck73367bd2010-05-21 14:41:35 -07003386 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003387 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003388 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003389 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003390 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003391 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003392
3393 if (!n) {
3394 free_kmem_cache_nodes(s);
3395 return 0;
3396 }
3397
Christoph Lameter81819f02007-05-06 14:49:36 -07003398 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09003399 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003400 }
3401 return 1;
3402}
Christoph Lameter81819f02007-05-06 14:49:36 -07003403
David Rientjesc0bdb232009-02-25 09:16:35 +02003404static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003405{
3406 if (min < MIN_PARTIAL)
3407 min = MIN_PARTIAL;
3408 else if (min > MAX_PARTIAL)
3409 min = MAX_PARTIAL;
3410 s->min_partial = min;
3411}
3412
Christoph Lameter81819f02007-05-06 14:49:36 -07003413/*
3414 * calculate_sizes() determines the order and the distribution of data within
3415 * a slab object.
3416 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003417static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003418{
3419 unsigned long flags = s->flags;
Alexander Potapenko80a92012016-07-28 15:49:07 -07003420 size_t size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003421 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003422
3423 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003424 * Round up object size to the next word boundary. We can only
3425 * place the free pointer at word boundaries and this determines
3426 * the possible location of the free pointer.
3427 */
3428 size = ALIGN(size, sizeof(void *));
3429
3430#ifdef CONFIG_SLUB_DEBUG
3431 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003432 * Determine if we can poison the object itself. If the user of
3433 * the slab may touch the object after free or before allocation
3434 * then we should never poison the object itself.
3435 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003436 if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003437 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003438 s->flags |= __OBJECT_POISON;
3439 else
3440 s->flags &= ~__OBJECT_POISON;
3441
Christoph Lameter81819f02007-05-06 14:49:36 -07003442
3443 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003444 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003445 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003446 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003447 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003448 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003449 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003450#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003451
3452 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003453 * With that we have determined the number of bytes in actual use
3454 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003455 */
3456 s->inuse = size;
3457
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003458 if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003459 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003460 /*
3461 * Relocate free pointer after the object if it is not
3462 * permitted to overwrite the first word of the object on
3463 * kmem_cache_free.
3464 *
3465 * This is the case if we do RCU, have a constructor or
3466 * destructor or are poisoning the objects.
3467 */
3468 s->offset = size;
3469 size += sizeof(void *);
3470 }
3471
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003472#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003473 if (flags & SLAB_STORE_USER)
3474 /*
3475 * Need to store information about allocs and frees after
3476 * the object.
3477 */
3478 size += 2 * sizeof(struct track);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003479#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003480
Alexander Potapenko80a92012016-07-28 15:49:07 -07003481 kasan_cache_create(s, &size, &s->flags);
3482#ifdef CONFIG_SLUB_DEBUG
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003483 if (flags & SLAB_RED_ZONE) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003484 /*
3485 * Add some empty padding so that we can catch
3486 * overwrites from earlier objects rather than let
3487 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003488 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003489 * of the object.
3490 */
3491 size += sizeof(void *);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003492
3493 s->red_left_pad = sizeof(void *);
3494 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3495 size += s->red_left_pad;
3496 }
Christoph Lameter41ecc552007-05-09 02:32:44 -07003497#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003498
Christoph Lameter81819f02007-05-06 14:49:36 -07003499 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003500 * SLUB stores one object immediately after another beginning from
3501 * offset 0. In order to align the objects we have to simply size
3502 * each object to conform to the alignment.
3503 */
Christoph Lameter45906852012-11-28 16:23:16 +00003504 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003505 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003506 if (forced_order >= 0)
3507 order = forced_order;
3508 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003509 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003510
Christoph Lameter834f3d12008-04-14 19:11:31 +03003511 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003512 return 0;
3513
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003514 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003515 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003516 s->allocflags |= __GFP_COMP;
3517
3518 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003519 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003520
3521 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3522 s->allocflags |= __GFP_RECLAIMABLE;
3523
Christoph Lameter81819f02007-05-06 14:49:36 -07003524 /*
3525 * Determine the number of objects per slab
3526 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003527 s->oo = oo_make(order, size, s->reserved);
3528 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003529 if (oo_objects(s->oo) > oo_objects(s->max))
3530 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003531
Christoph Lameter834f3d12008-04-14 19:11:31 +03003532 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003533}
3534
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003535static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003536{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003537 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003538 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003539
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003540 if (need_reserve_slab_rcu && (s->flags & SLAB_TYPESAFE_BY_RCU))
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003541 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003542
Christoph Lameter06b285d2008-04-14 19:11:41 +03003543 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003544 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003545 if (disable_higher_order_debug) {
3546 /*
3547 * Disable debugging flags that store metadata if the min slab
3548 * order increased.
3549 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003550 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003551 s->flags &= ~DEBUG_METADATA_FLAGS;
3552 s->offset = 0;
3553 if (!calculate_sizes(s, -1))
3554 goto error;
3555 }
3556 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003557
Heiko Carstens25654092012-01-12 17:17:33 -08003558#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3559 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Laura Abbott149daaf2016-03-15 14:55:09 -07003560 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003561 /* Enable fast mode */
3562 s->flags |= __CMPXCHG_DOUBLE;
3563#endif
3564
David Rientjes3b89d7d2009-02-22 17:40:07 -08003565 /*
3566 * The larger the object size is, the more pages we want on the partial
3567 * list to avoid pounding the page allocator excessively.
3568 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003569 set_min_partial(s, ilog2(s->size) / 2);
3570
3571 /*
3572 * cpu_partial determined the maximum number of objects kept in the
3573 * per cpu partial lists of a processor.
3574 *
3575 * Per cpu partial lists mainly contain slabs that just have one
3576 * object freed. If they are used for allocation then they can be
3577 * filled up again with minimal effort. The slab will never hit the
3578 * per node partial lists and therefore no locking will be required.
3579 *
3580 * This setting also determines
3581 *
3582 * A) The number of objects from per cpu partial slabs dumped to the
3583 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003584 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003585 * per node list when we run out of per cpu objects. We only fetch
3586 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003587 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003588 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003589 s->cpu_partial = 0;
3590 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003591 s->cpu_partial = 2;
3592 else if (s->size >= 1024)
3593 s->cpu_partial = 6;
3594 else if (s->size >= 256)
3595 s->cpu_partial = 13;
3596 else
3597 s->cpu_partial = 30;
3598
Christoph Lameter81819f02007-05-06 14:49:36 -07003599#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003600 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003601#endif
Thomas Garnier210e7a42016-07-26 15:21:59 -07003602
3603 /* Initialize the pre-computed randomized freelist if slab is up */
3604 if (slab_state >= UP) {
3605 if (init_cache_random_seq(s))
3606 goto error;
3607 }
3608
Christoph Lameter55136592010-08-20 12:37:13 -05003609 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003610 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003611
Christoph Lameter55136592010-08-20 12:37:13 -05003612 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003613 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003614
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003615 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003616error:
3617 if (flags & SLAB_PANIC)
Joe Perches756a0252016-03-17 14:19:47 -07003618 panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n",
3619 s->name, (unsigned long)s->size, s->size,
3620 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003621 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003622}
Christoph Lameter81819f02007-05-06 14:49:36 -07003623
Christoph Lameter33b12c32008-04-25 12:22:43 -07003624static void list_slab_objects(struct kmem_cache *s, struct page *page,
3625 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003626{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003627#ifdef CONFIG_SLUB_DEBUG
3628 void *addr = page_address(page);
3629 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003630 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3631 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003632 if (!map)
3633 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003634 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003635 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003636
Christoph Lameter5f80b132011-04-15 14:48:13 -05003637 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003638 for_each_object(p, s, addr, page->objects) {
3639
3640 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003641 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003642 print_tracking(s, p);
3643 }
3644 }
3645 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003646 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003647#endif
3648}
3649
Christoph Lameter81819f02007-05-06 14:49:36 -07003650/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003651 * Attempt to free all partial slabs on a node.
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003652 * This is called from __kmem_cache_shutdown(). We must take list_lock
3653 * because sysfs file might still access partial list after the shutdowning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003654 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003655static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003656{
Chris Wilson60398922016-08-10 16:27:58 -07003657 LIST_HEAD(discard);
Christoph Lameter81819f02007-05-06 14:49:36 -07003658 struct page *page, *h;
3659
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003660 BUG_ON(irqs_disabled());
3661 spin_lock_irq(&n->list_lock);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003662 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003663 if (!page->inuse) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003664 remove_partial(n, page);
Chris Wilson60398922016-08-10 16:27:58 -07003665 list_add(&page->lru, &discard);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003666 } else {
3667 list_slab_objects(s, page,
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003668 "Objects remaining in %s on __kmem_cache_shutdown()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003669 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003670 }
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003671 spin_unlock_irq(&n->list_lock);
Chris Wilson60398922016-08-10 16:27:58 -07003672
3673 list_for_each_entry_safe(page, h, &discard, lru)
3674 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003675}
3676
3677/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003678 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003679 */
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003680int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003681{
3682 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003683 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003684
3685 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003686 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003687 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003688 free_partial(s, n);
3689 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003690 return 1;
3691 }
Tejun Heobf5eb3d2017-02-22 15:41:11 -08003692 sysfs_slab_remove(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003693 return 0;
3694}
3695
Christoph Lameter81819f02007-05-06 14:49:36 -07003696/********************************************************************
3697 * Kmalloc subsystem
3698 *******************************************************************/
3699
Christoph Lameter81819f02007-05-06 14:49:36 -07003700static int __init setup_slub_min_order(char *str)
3701{
Pekka Enberg06428782008-01-07 23:20:27 -08003702 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003703
3704 return 1;
3705}
3706
3707__setup("slub_min_order=", setup_slub_min_order);
3708
3709static int __init setup_slub_max_order(char *str)
3710{
Pekka Enberg06428782008-01-07 23:20:27 -08003711 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003712 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003713
3714 return 1;
3715}
3716
3717__setup("slub_max_order=", setup_slub_max_order);
3718
3719static int __init setup_slub_min_objects(char *str)
3720{
Pekka Enberg06428782008-01-07 23:20:27 -08003721 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003722
3723 return 1;
3724}
3725
3726__setup("slub_min_objects=", setup_slub_min_objects);
3727
Christoph Lameter81819f02007-05-06 14:49:36 -07003728void *__kmalloc(size_t size, gfp_t flags)
3729{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003730 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003731 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003732
Christoph Lameter95a05b42013-01-10 19:14:19 +00003733 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003734 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003735
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003736 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003737
3738 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003739 return s;
3740
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003741 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003742
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003743 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003744
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003745 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003746
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003747 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003748}
3749EXPORT_SYMBOL(__kmalloc);
3750
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003751#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003752static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3753{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003754 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003755 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003756
Vladimir Davydov52383432014-06-04 16:06:39 -07003757 flags |= __GFP_COMP | __GFP_NOTRACK;
Vladimir Davydov49491482016-07-26 15:24:24 -07003758 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003759 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003760 ptr = page_address(page);
3761
Roman Bobnievd56791b2013-10-08 15:58:57 -07003762 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003763 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003764}
3765
Christoph Lameter81819f02007-05-06 14:49:36 -07003766void *__kmalloc_node(size_t size, gfp_t flags, int node)
3767{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003768 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003769 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003770
Christoph Lameter95a05b42013-01-10 19:14:19 +00003771 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003772 ret = kmalloc_large_node(size, flags, node);
3773
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003774 trace_kmalloc_node(_RET_IP_, ret,
3775 size, PAGE_SIZE << get_order(size),
3776 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003777
3778 return ret;
3779 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003780
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003781 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003782
3783 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003784 return s;
3785
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003786 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003787
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003788 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003789
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003790 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003791
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003792 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003793}
3794EXPORT_SYMBOL(__kmalloc_node);
3795#endif
3796
Kees Cooked18adc2016-06-23 15:24:05 -07003797#ifdef CONFIG_HARDENED_USERCOPY
3798/*
3799 * Rejects objects that are incorrectly sized.
3800 *
3801 * Returns NULL if check passes, otherwise const char * to name of cache
3802 * to indicate an error.
3803 */
3804const char *__check_heap_object(const void *ptr, unsigned long n,
3805 struct page *page)
3806{
3807 struct kmem_cache *s;
3808 unsigned long offset;
3809 size_t object_size;
3810
3811 /* Find object and usable object size. */
3812 s = page->slab_cache;
3813 object_size = slab_ksize(s);
3814
3815 /* Reject impossible pointers. */
3816 if (ptr < page_address(page))
3817 return s->name;
3818
3819 /* Find offset within object. */
3820 offset = (ptr - page_address(page)) % s->size;
3821
3822 /* Adjust for redzone and reject if within the redzone. */
3823 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3824 if (offset < s->red_left_pad)
3825 return s->name;
3826 offset -= s->red_left_pad;
3827 }
3828
3829 /* Allow address range falling entirely within object size. */
3830 if (offset <= object_size && n <= object_size - offset)
3831 return NULL;
3832
3833 return s->name;
3834}
3835#endif /* CONFIG_HARDENED_USERCOPY */
3836
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003837static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003838{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003839 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003840
Christoph Lameteref8b4522007-10-16 01:24:46 -07003841 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003842 return 0;
3843
Vegard Nossum294a80a2007-12-04 23:45:30 -08003844 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003845
Pekka Enberg76994412008-05-22 19:22:25 +03003846 if (unlikely(!PageSlab(page))) {
3847 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003848 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003849 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003850
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003851 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003852}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003853
3854size_t ksize(const void *object)
3855{
3856 size_t size = __ksize(object);
3857 /* We assume that ksize callers could use whole allocated area,
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07003858 * so we need to unpoison this area.
3859 */
3860 kasan_unpoison_shadow(object, size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003861 return size;
3862}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003863EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003864
3865void kfree(const void *x)
3866{
Christoph Lameter81819f02007-05-06 14:49:36 -07003867 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003868 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003869
Pekka Enberg2121db72009-03-25 11:05:57 +02003870 trace_kfree(_RET_IP_, x);
3871
Satyam Sharma2408c552007-10-16 01:24:44 -07003872 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003873 return;
3874
Christoph Lameterb49af682007-05-06 14:49:41 -07003875 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003876 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003877 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003878 kfree_hook(x);
Vladimir Davydov49491482016-07-26 15:24:24 -07003879 __free_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003880 return;
3881 }
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003882 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003883}
3884EXPORT_SYMBOL(kfree);
3885
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003886#define SHRINK_PROMOTE_MAX 32
3887
Christoph Lameter2086d262007-05-06 14:49:46 -07003888/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003889 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3890 * up most to the head of the partial lists. New allocations will then
3891 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003892 *
3893 * The slabs with the least items are placed last. This results in them
3894 * being allocated from last increasing the chance that the last objects
3895 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003896 */
Tejun Heoc9fc5862017-02-22 15:41:27 -08003897int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07003898{
3899 int node;
3900 int i;
3901 struct kmem_cache_node *n;
3902 struct page *page;
3903 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003904 struct list_head discard;
3905 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003906 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003907 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003908
Christoph Lameter2086d262007-05-06 14:49:46 -07003909 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003910 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003911 INIT_LIST_HEAD(&discard);
3912 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3913 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003914
3915 spin_lock_irqsave(&n->list_lock, flags);
3916
3917 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003918 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003919 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003920 * Note that concurrent frees may occur while we hold the
3921 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003922 */
3923 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003924 int free = page->objects - page->inuse;
3925
3926 /* Do not reread page->inuse */
3927 barrier();
3928
3929 /* We do not keep full slabs on the list */
3930 BUG_ON(free <= 0);
3931
3932 if (free == page->objects) {
3933 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003934 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003935 } else if (free <= SHRINK_PROMOTE_MAX)
3936 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07003937 }
3938
Christoph Lameter2086d262007-05-06 14:49:46 -07003939 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003940 * Promote the slabs filled up most to the head of the
3941 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07003942 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003943 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3944 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07003945
Christoph Lameter2086d262007-05-06 14:49:46 -07003946 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003947
3948 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003949 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003950 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003951
3952 if (slabs_node(s, node))
3953 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07003954 }
3955
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003956 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07003957}
Christoph Lameter2086d262007-05-06 14:49:46 -07003958
Tejun Heoc9fc5862017-02-22 15:41:27 -08003959#ifdef CONFIG_MEMCG
Tejun Heo01fb58b2017-02-22 15:41:30 -08003960static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s)
3961{
Tejun Heo50862ce72017-02-22 15:41:33 -08003962 /*
3963 * Called with all the locks held after a sched RCU grace period.
3964 * Even if @s becomes empty after shrinking, we can't know that @s
3965 * doesn't have allocations already in-flight and thus can't
3966 * destroy @s until the associated memcg is released.
3967 *
3968 * However, let's remove the sysfs files for empty caches here.
3969 * Each cache has a lot of interface files which aren't
3970 * particularly useful for empty draining caches; otherwise, we can
3971 * easily end up with millions of unnecessary sysfs files on
3972 * systems which have a lot of memory and transient cgroups.
3973 */
3974 if (!__kmem_cache_shrink(s))
3975 sysfs_slab_remove(s);
Tejun Heo01fb58b2017-02-22 15:41:30 -08003976}
3977
Tejun Heoc9fc5862017-02-22 15:41:27 -08003978void __kmemcg_cache_deactivate(struct kmem_cache *s)
3979{
3980 /*
3981 * Disable empty slabs caching. Used to avoid pinning offline
3982 * memory cgroups by kmem pages that can be freed.
3983 */
3984 s->cpu_partial = 0;
3985 s->min_partial = 0;
3986
3987 /*
3988 * s->cpu_partial is checked locklessly (see put_cpu_partial), so
Tejun Heo01fb58b2017-02-22 15:41:30 -08003989 * we have to make sure the change is visible before shrinking.
Tejun Heoc9fc5862017-02-22 15:41:27 -08003990 */
Tejun Heo01fb58b2017-02-22 15:41:30 -08003991 slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu);
Tejun Heoc9fc5862017-02-22 15:41:27 -08003992}
3993#endif
3994
Yasunori Gotob9049e22007-10-21 16:41:37 -07003995static int slab_mem_going_offline_callback(void *arg)
3996{
3997 struct kmem_cache *s;
3998
Christoph Lameter18004c52012-07-06 15:25:12 -05003999 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004000 list_for_each_entry(s, &slab_caches, list)
Tejun Heoc9fc5862017-02-22 15:41:27 -08004001 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05004002 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004003
4004 return 0;
4005}
4006
4007static void slab_mem_offline_callback(void *arg)
4008{
4009 struct kmem_cache_node *n;
4010 struct kmem_cache *s;
4011 struct memory_notify *marg = arg;
4012 int offline_node;
4013
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004014 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004015
4016 /*
4017 * If the node still has available memory. we need kmem_cache_node
4018 * for it yet.
4019 */
4020 if (offline_node < 0)
4021 return;
4022
Christoph Lameter18004c52012-07-06 15:25:12 -05004023 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004024 list_for_each_entry(s, &slab_caches, list) {
4025 n = get_node(s, offline_node);
4026 if (n) {
4027 /*
4028 * if n->nr_slabs > 0, slabs still exist on the node
4029 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05004030 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07004031 * callback. So, we must fail.
4032 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03004033 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07004034
4035 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05004036 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004037 }
4038 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004039 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004040}
4041
4042static int slab_mem_going_online_callback(void *arg)
4043{
4044 struct kmem_cache_node *n;
4045 struct kmem_cache *s;
4046 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004047 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004048 int ret = 0;
4049
4050 /*
4051 * If the node's memory is already available, then kmem_cache_node is
4052 * already created. Nothing to do.
4053 */
4054 if (nid < 0)
4055 return 0;
4056
4057 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07004058 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07004059 * allocate a kmem_cache_node structure in order to bring the node
4060 * online.
4061 */
Christoph Lameter18004c52012-07-06 15:25:12 -05004062 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004063 list_for_each_entry(s, &slab_caches, list) {
4064 /*
4065 * XXX: kmem_cache_alloc_node will fallback to other nodes
4066 * since memory is not yet available from the node that
4067 * is brought up.
4068 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05004069 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004070 if (!n) {
4071 ret = -ENOMEM;
4072 goto out;
4073 }
Joonsoo Kim40534972012-05-11 00:50:47 +09004074 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004075 s->node[nid] = n;
4076 }
4077out:
Christoph Lameter18004c52012-07-06 15:25:12 -05004078 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004079 return ret;
4080}
4081
4082static int slab_memory_callback(struct notifier_block *self,
4083 unsigned long action, void *arg)
4084{
4085 int ret = 0;
4086
4087 switch (action) {
4088 case MEM_GOING_ONLINE:
4089 ret = slab_mem_going_online_callback(arg);
4090 break;
4091 case MEM_GOING_OFFLINE:
4092 ret = slab_mem_going_offline_callback(arg);
4093 break;
4094 case MEM_OFFLINE:
4095 case MEM_CANCEL_ONLINE:
4096 slab_mem_offline_callback(arg);
4097 break;
4098 case MEM_ONLINE:
4099 case MEM_CANCEL_OFFLINE:
4100 break;
4101 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08004102 if (ret)
4103 ret = notifier_from_errno(ret);
4104 else
4105 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004106 return ret;
4107}
4108
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004109static struct notifier_block slab_memory_callback_nb = {
4110 .notifier_call = slab_memory_callback,
4111 .priority = SLAB_CALLBACK_PRI,
4112};
Yasunori Gotob9049e22007-10-21 16:41:37 -07004113
Christoph Lameter81819f02007-05-06 14:49:36 -07004114/********************************************************************
4115 * Basic setup of slabs
4116 *******************************************************************/
4117
Christoph Lameter51df1142010-08-20 12:37:15 -05004118/*
4119 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004120 * the page allocator. Allocate them properly then fix up the pointers
4121 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05004122 */
4123
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004124static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05004125{
4126 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004127 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004128 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05004129
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004130 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05004131
Glauber Costa7d557b32013-02-22 20:20:00 +04004132 /*
4133 * This runs very early, and only the boot processor is supposed to be
4134 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4135 * IPIs around.
4136 */
4137 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004138 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05004139 struct page *p;
4140
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004141 list_for_each_entry(p, &n->partial, lru)
4142 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004143
Li Zefan607bf322011-04-12 15:22:26 +08004144#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004145 list_for_each_entry(p, &n->full, lru)
4146 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004147#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05004148 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08004149 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004150 list_add(&s->list, &slab_caches);
Tejun Heo510ded32017-02-22 15:41:24 -08004151 memcg_link_cache(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004152 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004153}
4154
Christoph Lameter81819f02007-05-06 14:49:36 -07004155void __init kmem_cache_init(void)
4156{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004157 static __initdata struct kmem_cache boot_kmem_cache,
4158 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05004159
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08004160 if (debug_guardpage_minorder())
4161 slub_max_order = 0;
4162
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004163 kmem_cache_node = &boot_kmem_cache_node;
4164 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05004165
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004166 create_boot_cache(kmem_cache_node, "kmem_cache_node",
4167 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004168
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004169 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07004170
4171 /* Able to allocate the per node structures */
4172 slab_state = PARTIAL;
4173
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004174 create_boot_cache(kmem_cache, "kmem_cache",
4175 offsetof(struct kmem_cache, node) +
4176 nr_node_ids * sizeof(struct kmem_cache_node *),
4177 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004178
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004179 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07004180
Christoph Lameter51df1142010-08-20 12:37:15 -05004181 /*
4182 * Allocate kmem_cache_node properly from the kmem_cache slab.
4183 * kmem_cache_node is separately allocated so no need to
4184 * update any list pointers.
4185 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004186 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05004187
4188 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07004189 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004190 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07004191
Thomas Garnier210e7a42016-07-26 15:21:59 -07004192 /* Setup random freelists for each cache */
4193 init_freelist_randomization();
4194
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02004195 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4196 slub_cpu_dead);
Christoph Lameter81819f02007-05-06 14:49:36 -07004197
Fabian Frederickf9f58282014-06-04 16:06:34 -07004198 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004199 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07004200 slub_min_order, slub_max_order, slub_min_objects,
4201 nr_cpu_ids, nr_node_ids);
4202}
4203
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004204void __init kmem_cache_init_late(void)
4205{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004206}
4207
Glauber Costa2633d7a2012-12-18 14:22:34 -08004208struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07004209__kmem_cache_alias(const char *name, size_t size, size_t align,
4210 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07004211{
Vladimir Davydov426589f2015-02-12 14:59:23 -08004212 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07004213
Vladimir Davydova44cb942014-04-07 15:39:23 -07004214 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004215 if (s) {
4216 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004217
Christoph Lameter81819f02007-05-06 14:49:36 -07004218 /*
4219 * Adjust the object sizes so that we clear
4220 * the complete object on kzalloc.
4221 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004222 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004223 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08004224
Vladimir Davydov426589f2015-02-12 14:59:23 -08004225 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004226 c->object_size = s->object_size;
4227 c->inuse = max_t(int, c->inuse,
4228 ALIGN(size, sizeof(void *)));
4229 }
4230
David Rientjes7b8f3b62008-12-17 22:09:46 -08004231 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08004232 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00004233 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08004234 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07004235 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08004236
Christoph Lametercbb79692012-09-05 00:18:32 +00004237 return s;
4238}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004239
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004240int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00004241{
Pekka Enbergaac3a162012-09-05 12:07:44 +03004242 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004243
Pekka Enbergaac3a162012-09-05 12:07:44 +03004244 err = kmem_cache_open(s, flags);
4245 if (err)
4246 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004247
Christoph Lameter45530c42012-11-28 16:23:07 +00004248 /* Mutex is not taken during early boot */
4249 if (slab_state <= UP)
4250 return 0;
4251
Glauber Costa107dab52012-12-18 14:23:05 -08004252 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004253 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004254 if (err)
Dmitry Safonov52b4b952016-02-17 13:11:37 -08004255 __kmem_cache_release(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004256
4257 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004258}
Christoph Lameter81819f02007-05-06 14:49:36 -07004259
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004260void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004261{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004262 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004263 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004264
Christoph Lameter95a05b42013-01-10 19:14:19 +00004265 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004266 return kmalloc_large(size, gfpflags);
4267
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004268 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004269
Satyam Sharma2408c552007-10-16 01:24:44 -07004270 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004271 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004272
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004273 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004274
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004275 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004276 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004277
4278 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004279}
4280
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004281#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004282void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004283 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004284{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004285 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004286 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004287
Christoph Lameter95a05b42013-01-10 19:14:19 +00004288 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004289 ret = kmalloc_large_node(size, gfpflags, node);
4290
4291 trace_kmalloc_node(caller, ret,
4292 size, PAGE_SIZE << get_order(size),
4293 gfpflags, node);
4294
4295 return ret;
4296 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004297
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004298 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004299
Satyam Sharma2408c552007-10-16 01:24:44 -07004300 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004301 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004302
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004303 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004304
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004305 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004306 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004307
4308 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004309}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004310#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004311
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004312#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004313static int count_inuse(struct page *page)
4314{
4315 return page->inuse;
4316}
4317
4318static int count_total(struct page *page)
4319{
4320 return page->objects;
4321}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004322#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004323
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004324#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004325static int validate_slab(struct kmem_cache *s, struct page *page,
4326 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004327{
4328 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004329 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004330
4331 if (!check_slab(s, page) ||
4332 !on_freelist(s, page, NULL))
4333 return 0;
4334
4335 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004336 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004337
Christoph Lameter5f80b132011-04-15 14:48:13 -05004338 get_map(s, page, map);
4339 for_each_object(p, s, addr, page->objects) {
4340 if (test_bit(slab_index(p, s, addr), map))
4341 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4342 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004343 }
4344
Christoph Lameter224a88b2008-04-14 19:11:31 +03004345 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004346 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004347 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004348 return 0;
4349 return 1;
4350}
4351
Christoph Lameter434e2452007-07-17 04:03:30 -07004352static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4353 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004354{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004355 slab_lock(page);
4356 validate_slab(s, page, map);
4357 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004358}
4359
Christoph Lameter434e2452007-07-17 04:03:30 -07004360static int validate_slab_node(struct kmem_cache *s,
4361 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004362{
4363 unsigned long count = 0;
4364 struct page *page;
4365 unsigned long flags;
4366
4367 spin_lock_irqsave(&n->list_lock, flags);
4368
4369 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004370 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004371 count++;
4372 }
4373 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004374 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4375 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004376
4377 if (!(s->flags & SLAB_STORE_USER))
4378 goto out;
4379
4380 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004381 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004382 count++;
4383 }
4384 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004385 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4386 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004387
4388out:
4389 spin_unlock_irqrestore(&n->list_lock, flags);
4390 return count;
4391}
4392
Christoph Lameter434e2452007-07-17 04:03:30 -07004393static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004394{
4395 int node;
4396 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004397 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004398 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004399 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07004400
4401 if (!map)
4402 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004403
4404 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004405 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07004406 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07004407 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004408 return count;
4409}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004410/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004411 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004412 * and freed.
4413 */
4414
4415struct location {
4416 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004417 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004418 long long sum_time;
4419 long min_time;
4420 long max_time;
4421 long min_pid;
4422 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304423 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004424 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004425};
4426
4427struct loc_track {
4428 unsigned long max;
4429 unsigned long count;
4430 struct location *loc;
4431};
4432
4433static void free_loc_track(struct loc_track *t)
4434{
4435 if (t->max)
4436 free_pages((unsigned long)t->loc,
4437 get_order(sizeof(struct location) * t->max));
4438}
4439
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004440static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004441{
4442 struct location *l;
4443 int order;
4444
Christoph Lameter88a420e2007-05-06 14:49:45 -07004445 order = get_order(sizeof(struct location) * max);
4446
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004447 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004448 if (!l)
4449 return 0;
4450
4451 if (t->count) {
4452 memcpy(l, t->loc, sizeof(struct location) * t->count);
4453 free_loc_track(t);
4454 }
4455 t->max = max;
4456 t->loc = l;
4457 return 1;
4458}
4459
4460static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004461 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004462{
4463 long start, end, pos;
4464 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004465 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004466 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004467
4468 start = -1;
4469 end = t->count;
4470
4471 for ( ; ; ) {
4472 pos = start + (end - start + 1) / 2;
4473
4474 /*
4475 * There is nothing at "end". If we end up there
4476 * we need to add something to before end.
4477 */
4478 if (pos == end)
4479 break;
4480
4481 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004482 if (track->addr == caddr) {
4483
4484 l = &t->loc[pos];
4485 l->count++;
4486 if (track->when) {
4487 l->sum_time += age;
4488 if (age < l->min_time)
4489 l->min_time = age;
4490 if (age > l->max_time)
4491 l->max_time = age;
4492
4493 if (track->pid < l->min_pid)
4494 l->min_pid = track->pid;
4495 if (track->pid > l->max_pid)
4496 l->max_pid = track->pid;
4497
Rusty Russell174596a2009-01-01 10:12:29 +10304498 cpumask_set_cpu(track->cpu,
4499 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004500 }
4501 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004502 return 1;
4503 }
4504
Christoph Lameter45edfa52007-05-09 02:32:45 -07004505 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004506 end = pos;
4507 else
4508 start = pos;
4509 }
4510
4511 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004512 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004513 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004514 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004515 return 0;
4516
4517 l = t->loc + pos;
4518 if (pos < t->count)
4519 memmove(l + 1, l,
4520 (t->count - pos) * sizeof(struct location));
4521 t->count++;
4522 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004523 l->addr = track->addr;
4524 l->sum_time = age;
4525 l->min_time = age;
4526 l->max_time = age;
4527 l->min_pid = track->pid;
4528 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304529 cpumask_clear(to_cpumask(l->cpus));
4530 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004531 nodes_clear(l->nodes);
4532 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004533 return 1;
4534}
4535
4536static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004537 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004538 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004539{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004540 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004541 void *p;
4542
Christoph Lameter39b26462008-04-14 19:11:30 +03004543 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004544 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004545
Christoph Lameter224a88b2008-04-14 19:11:31 +03004546 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004547 if (!test_bit(slab_index(p, s, addr), map))
4548 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004549}
4550
4551static int list_locations(struct kmem_cache *s, char *buf,
4552 enum track_item alloc)
4553{
Harvey Harrisone374d482008-01-31 15:20:50 -08004554 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004555 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004556 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004557 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004558 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4559 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004560 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004561
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004562 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4563 GFP_TEMPORARY)) {
4564 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004565 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004566 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004567 /* Push back cpu slabs */
4568 flush_all(s);
4569
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004570 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004571 unsigned long flags;
4572 struct page *page;
4573
Christoph Lameter9e869432007-08-22 14:01:56 -07004574 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004575 continue;
4576
4577 spin_lock_irqsave(&n->list_lock, flags);
4578 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004579 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004580 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004581 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004582 spin_unlock_irqrestore(&n->list_lock, flags);
4583 }
4584
4585 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004586 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004587
Hugh Dickins9c246242008-12-09 13:14:27 -08004588 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004589 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004590 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004591
4592 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004593 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004594 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004595 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004596
4597 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004598 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004599 l->min_time,
4600 (long)div_u64(l->sum_time, l->count),
4601 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004602 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004603 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004604 l->min_time);
4605
4606 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004607 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004608 l->min_pid, l->max_pid);
4609 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004610 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004611 l->min_pid);
4612
Rusty Russell174596a2009-01-01 10:12:29 +10304613 if (num_online_cpus() > 1 &&
4614 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004615 len < PAGE_SIZE - 60)
4616 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4617 " cpus=%*pbl",
4618 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004619
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004620 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004621 len < PAGE_SIZE - 60)
4622 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4623 " nodes=%*pbl",
4624 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004625
Harvey Harrisone374d482008-01-31 15:20:50 -08004626 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004627 }
4628
4629 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004630 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004631 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004632 len += sprintf(buf, "No data\n");
4633 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004634}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004635#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004636
Christoph Lametera5a84752010-10-05 13:57:27 -05004637#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004638static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004639{
4640 u8 *p;
4641
Christoph Lameter95a05b42013-01-10 19:14:19 +00004642 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004643
Fabian Frederickf9f58282014-06-04 16:06:34 -07004644 pr_err("SLUB resiliency testing\n");
4645 pr_err("-----------------------\n");
4646 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004647
4648 p = kzalloc(16, GFP_KERNEL);
4649 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004650 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4651 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004652
4653 validate_slab_cache(kmalloc_caches[4]);
4654
4655 /* Hmmm... The next two are dangerous */
4656 p = kzalloc(32, GFP_KERNEL);
4657 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004658 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4659 p);
4660 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004661
4662 validate_slab_cache(kmalloc_caches[5]);
4663 p = kzalloc(64, GFP_KERNEL);
4664 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4665 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004666 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4667 p);
4668 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004669 validate_slab_cache(kmalloc_caches[6]);
4670
Fabian Frederickf9f58282014-06-04 16:06:34 -07004671 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004672 p = kzalloc(128, GFP_KERNEL);
4673 kfree(p);
4674 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004675 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004676 validate_slab_cache(kmalloc_caches[7]);
4677
4678 p = kzalloc(256, GFP_KERNEL);
4679 kfree(p);
4680 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004681 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004682 validate_slab_cache(kmalloc_caches[8]);
4683
4684 p = kzalloc(512, GFP_KERNEL);
4685 kfree(p);
4686 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004687 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004688 validate_slab_cache(kmalloc_caches[9]);
4689}
4690#else
4691#ifdef CONFIG_SYSFS
4692static void resiliency_test(void) {};
4693#endif
4694#endif
4695
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004696#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004697enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004698 SL_ALL, /* All slabs */
4699 SL_PARTIAL, /* Only partially allocated slabs */
4700 SL_CPU, /* Only slabs used for cpu caches */
4701 SL_OBJECTS, /* Determine allocated objects not slabs */
4702 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004703};
4704
Christoph Lameter205ab992008-04-14 19:11:40 +03004705#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004706#define SO_PARTIAL (1 << SL_PARTIAL)
4707#define SO_CPU (1 << SL_CPU)
4708#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004709#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004710
Tejun Heo1663f262017-02-22 15:41:39 -08004711#ifdef CONFIG_MEMCG
4712static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4713
4714static int __init setup_slub_memcg_sysfs(char *str)
4715{
4716 int v;
4717
4718 if (get_option(&str, &v) > 0)
4719 memcg_sysfs_enabled = v;
4720
4721 return 1;
4722}
4723
4724__setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4725#endif
4726
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004727static ssize_t show_slab_objects(struct kmem_cache *s,
4728 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004729{
4730 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004731 int node;
4732 int x;
4733 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004734
Chen Gange35e1a92013-07-12 08:23:48 +08004735 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004736 if (!nodes)
4737 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004738
Christoph Lameter205ab992008-04-14 19:11:40 +03004739 if (flags & SO_CPU) {
4740 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004741
Christoph Lameter205ab992008-04-14 19:11:40 +03004742 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004743 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4744 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004745 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004746 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004747
Jason Low4db0c3c2015-04-15 16:14:08 -07004748 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004749 if (!page)
4750 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004751
Christoph Lameterec3ab082012-05-09 10:09:56 -05004752 node = page_to_nid(page);
4753 if (flags & SO_TOTAL)
4754 x = page->objects;
4755 else if (flags & SO_OBJECTS)
4756 x = page->inuse;
4757 else
4758 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004759
Christoph Lameterec3ab082012-05-09 10:09:56 -05004760 total += x;
4761 nodes[node] += x;
4762
Jason Low4db0c3c2015-04-15 16:14:08 -07004763 page = READ_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004764 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004765 node = page_to_nid(page);
4766 if (flags & SO_TOTAL)
4767 WARN_ON_ONCE(1);
4768 else if (flags & SO_OBJECTS)
4769 WARN_ON_ONCE(1);
4770 else
4771 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004772 total += x;
4773 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004774 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004775 }
4776 }
4777
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004778 get_online_mems();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004779#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004780 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004781 struct kmem_cache_node *n;
4782
4783 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004784
Chen Gangd0e0ac92013-07-15 09:05:29 +08004785 if (flags & SO_TOTAL)
4786 x = atomic_long_read(&n->total_objects);
4787 else if (flags & SO_OBJECTS)
4788 x = atomic_long_read(&n->total_objects) -
4789 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004790 else
4791 x = atomic_long_read(&n->nr_slabs);
4792 total += x;
4793 nodes[node] += x;
4794 }
4795
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004796 } else
4797#endif
4798 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004799 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004800
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004801 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004802 if (flags & SO_TOTAL)
4803 x = count_partial(n, count_total);
4804 else if (flags & SO_OBJECTS)
4805 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004806 else
4807 x = n->nr_partial;
4808 total += x;
4809 nodes[node] += x;
4810 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004811 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004812 x = sprintf(buf, "%lu", total);
4813#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004814 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004815 if (nodes[node])
4816 x += sprintf(buf + x, " N%d=%lu",
4817 node, nodes[node]);
4818#endif
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004819 put_online_mems();
Christoph Lameter81819f02007-05-06 14:49:36 -07004820 kfree(nodes);
4821 return x + sprintf(buf + x, "\n");
4822}
4823
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004824#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004825static int any_slab_objects(struct kmem_cache *s)
4826{
4827 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004828 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004829
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004830 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004831 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004832 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004833
Christoph Lameter81819f02007-05-06 14:49:36 -07004834 return 0;
4835}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004836#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004837
4838#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004839#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004840
4841struct slab_attribute {
4842 struct attribute attr;
4843 ssize_t (*show)(struct kmem_cache *s, char *buf);
4844 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4845};
4846
4847#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004848 static struct slab_attribute _name##_attr = \
4849 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004850
4851#define SLAB_ATTR(_name) \
4852 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004853 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004854
Christoph Lameter81819f02007-05-06 14:49:36 -07004855static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4856{
4857 return sprintf(buf, "%d\n", s->size);
4858}
4859SLAB_ATTR_RO(slab_size);
4860
4861static ssize_t align_show(struct kmem_cache *s, char *buf)
4862{
4863 return sprintf(buf, "%d\n", s->align);
4864}
4865SLAB_ATTR_RO(align);
4866
4867static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4868{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004869 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004870}
4871SLAB_ATTR_RO(object_size);
4872
4873static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4874{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004875 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004876}
4877SLAB_ATTR_RO(objs_per_slab);
4878
Christoph Lameter06b285d2008-04-14 19:11:41 +03004879static ssize_t order_store(struct kmem_cache *s,
4880 const char *buf, size_t length)
4881{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004882 unsigned long order;
4883 int err;
4884
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004885 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004886 if (err)
4887 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004888
4889 if (order > slub_max_order || order < slub_min_order)
4890 return -EINVAL;
4891
4892 calculate_sizes(s, order);
4893 return length;
4894}
4895
Christoph Lameter81819f02007-05-06 14:49:36 -07004896static ssize_t order_show(struct kmem_cache *s, char *buf)
4897{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004898 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004899}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004900SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004901
David Rientjes73d342b2009-02-22 17:40:09 -08004902static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4903{
4904 return sprintf(buf, "%lu\n", s->min_partial);
4905}
4906
4907static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4908 size_t length)
4909{
4910 unsigned long min;
4911 int err;
4912
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004913 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004914 if (err)
4915 return err;
4916
David Rientjesc0bdb232009-02-25 09:16:35 +02004917 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004918 return length;
4919}
4920SLAB_ATTR(min_partial);
4921
Christoph Lameter49e22582011-08-09 16:12:27 -05004922static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4923{
4924 return sprintf(buf, "%u\n", s->cpu_partial);
4925}
4926
4927static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4928 size_t length)
4929{
4930 unsigned long objects;
4931 int err;
4932
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004933 err = kstrtoul(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004934 if (err)
4935 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004936 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004937 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004938
4939 s->cpu_partial = objects;
4940 flush_all(s);
4941 return length;
4942}
4943SLAB_ATTR(cpu_partial);
4944
Christoph Lameter81819f02007-05-06 14:49:36 -07004945static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4946{
Joe Perches62c70bc2011-01-13 15:45:52 -08004947 if (!s->ctor)
4948 return 0;
4949 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004950}
4951SLAB_ATTR_RO(ctor);
4952
Christoph Lameter81819f02007-05-06 14:49:36 -07004953static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4954{
Gu Zheng4307c142014-08-06 16:04:51 -07004955 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004956}
4957SLAB_ATTR_RO(aliases);
4958
Christoph Lameter81819f02007-05-06 14:49:36 -07004959static ssize_t partial_show(struct kmem_cache *s, char *buf)
4960{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004961 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004962}
4963SLAB_ATTR_RO(partial);
4964
4965static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4966{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004967 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004968}
4969SLAB_ATTR_RO(cpu_slabs);
4970
4971static ssize_t objects_show(struct kmem_cache *s, char *buf)
4972{
Christoph Lameter205ab992008-04-14 19:11:40 +03004973 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004974}
4975SLAB_ATTR_RO(objects);
4976
Christoph Lameter205ab992008-04-14 19:11:40 +03004977static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4978{
4979 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4980}
4981SLAB_ATTR_RO(objects_partial);
4982
Christoph Lameter49e22582011-08-09 16:12:27 -05004983static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4984{
4985 int objects = 0;
4986 int pages = 0;
4987 int cpu;
4988 int len;
4989
4990 for_each_online_cpu(cpu) {
4991 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4992
4993 if (page) {
4994 pages += page->pages;
4995 objects += page->pobjects;
4996 }
4997 }
4998
4999 len = sprintf(buf, "%d(%d)", objects, pages);
5000
5001#ifdef CONFIG_SMP
5002 for_each_online_cpu(cpu) {
5003 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
5004
5005 if (page && len < PAGE_SIZE - 20)
5006 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5007 page->pobjects, page->pages);
5008 }
5009#endif
5010 return len + sprintf(buf + len, "\n");
5011}
5012SLAB_ATTR_RO(slabs_cpu_partial);
5013
Christoph Lameter81819f02007-05-06 14:49:36 -07005014static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5015{
5016 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5017}
5018
5019static ssize_t reclaim_account_store(struct kmem_cache *s,
5020 const char *buf, size_t length)
5021{
5022 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5023 if (buf[0] == '1')
5024 s->flags |= SLAB_RECLAIM_ACCOUNT;
5025 return length;
5026}
5027SLAB_ATTR(reclaim_account);
5028
5029static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5030{
Christoph Lameter5af60832007-05-06 14:49:56 -07005031 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07005032}
5033SLAB_ATTR_RO(hwcache_align);
5034
5035#ifdef CONFIG_ZONE_DMA
5036static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5037{
5038 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5039}
5040SLAB_ATTR_RO(cache_dma);
5041#endif
5042
5043static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5044{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08005045 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
Christoph Lameter81819f02007-05-06 14:49:36 -07005046}
5047SLAB_ATTR_RO(destroy_by_rcu);
5048
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005049static ssize_t reserved_show(struct kmem_cache *s, char *buf)
5050{
5051 return sprintf(buf, "%d\n", s->reserved);
5052}
5053SLAB_ATTR_RO(reserved);
5054
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005055#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005056static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5057{
5058 return show_slab_objects(s, buf, SO_ALL);
5059}
5060SLAB_ATTR_RO(slabs);
5061
5062static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5063{
5064 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5065}
5066SLAB_ATTR_RO(total_objects);
5067
5068static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5069{
Laura Abbottbecfda62016-03-15 14:55:06 -07005070 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
Christoph Lametera5a84752010-10-05 13:57:27 -05005071}
5072
5073static ssize_t sanity_checks_store(struct kmem_cache *s,
5074 const char *buf, size_t length)
5075{
Laura Abbottbecfda62016-03-15 14:55:06 -07005076 s->flags &= ~SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005077 if (buf[0] == '1') {
5078 s->flags &= ~__CMPXCHG_DOUBLE;
Laura Abbottbecfda62016-03-15 14:55:06 -07005079 s->flags |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005080 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005081 return length;
5082}
5083SLAB_ATTR(sanity_checks);
5084
5085static ssize_t trace_show(struct kmem_cache *s, char *buf)
5086{
5087 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5088}
5089
5090static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5091 size_t length)
5092{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005093 /*
5094 * Tracing a merged cache is going to give confusing results
5095 * as well as cause other issues like converting a mergeable
5096 * cache into an umergeable one.
5097 */
5098 if (s->refcount > 1)
5099 return -EINVAL;
5100
Christoph Lametera5a84752010-10-05 13:57:27 -05005101 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005102 if (buf[0] == '1') {
5103 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05005104 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005105 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005106 return length;
5107}
5108SLAB_ATTR(trace);
5109
Christoph Lameter81819f02007-05-06 14:49:36 -07005110static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5111{
5112 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5113}
5114
5115static ssize_t red_zone_store(struct kmem_cache *s,
5116 const char *buf, size_t length)
5117{
5118 if (any_slab_objects(s))
5119 return -EBUSY;
5120
5121 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005122 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005123 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005124 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005125 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005126 return length;
5127}
5128SLAB_ATTR(red_zone);
5129
5130static ssize_t poison_show(struct kmem_cache *s, char *buf)
5131{
5132 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5133}
5134
5135static ssize_t poison_store(struct kmem_cache *s,
5136 const char *buf, size_t length)
5137{
5138 if (any_slab_objects(s))
5139 return -EBUSY;
5140
5141 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005142 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005143 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005144 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005145 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005146 return length;
5147}
5148SLAB_ATTR(poison);
5149
5150static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5151{
5152 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5153}
5154
5155static ssize_t store_user_store(struct kmem_cache *s,
5156 const char *buf, size_t length)
5157{
5158 if (any_slab_objects(s))
5159 return -EBUSY;
5160
5161 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005162 if (buf[0] == '1') {
5163 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07005164 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005165 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005166 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005167 return length;
5168}
5169SLAB_ATTR(store_user);
5170
Christoph Lameter53e15af2007-05-06 14:49:43 -07005171static ssize_t validate_show(struct kmem_cache *s, char *buf)
5172{
5173 return 0;
5174}
5175
5176static ssize_t validate_store(struct kmem_cache *s,
5177 const char *buf, size_t length)
5178{
Christoph Lameter434e2452007-07-17 04:03:30 -07005179 int ret = -EINVAL;
5180
5181 if (buf[0] == '1') {
5182 ret = validate_slab_cache(s);
5183 if (ret >= 0)
5184 ret = length;
5185 }
5186 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07005187}
5188SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05005189
5190static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5191{
5192 if (!(s->flags & SLAB_STORE_USER))
5193 return -ENOSYS;
5194 return list_locations(s, buf, TRACK_ALLOC);
5195}
5196SLAB_ATTR_RO(alloc_calls);
5197
5198static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5199{
5200 if (!(s->flags & SLAB_STORE_USER))
5201 return -ENOSYS;
5202 return list_locations(s, buf, TRACK_FREE);
5203}
5204SLAB_ATTR_RO(free_calls);
5205#endif /* CONFIG_SLUB_DEBUG */
5206
5207#ifdef CONFIG_FAILSLAB
5208static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5209{
5210 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5211}
5212
5213static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5214 size_t length)
5215{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005216 if (s->refcount > 1)
5217 return -EINVAL;
5218
Christoph Lametera5a84752010-10-05 13:57:27 -05005219 s->flags &= ~SLAB_FAILSLAB;
5220 if (buf[0] == '1')
5221 s->flags |= SLAB_FAILSLAB;
5222 return length;
5223}
5224SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005225#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07005226
Christoph Lameter2086d262007-05-06 14:49:46 -07005227static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5228{
5229 return 0;
5230}
5231
5232static ssize_t shrink_store(struct kmem_cache *s,
5233 const char *buf, size_t length)
5234{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005235 if (buf[0] == '1')
5236 kmem_cache_shrink(s);
5237 else
Christoph Lameter2086d262007-05-06 14:49:46 -07005238 return -EINVAL;
5239 return length;
5240}
5241SLAB_ATTR(shrink);
5242
Christoph Lameter81819f02007-05-06 14:49:36 -07005243#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005244static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07005245{
Christoph Lameter98246012008-01-07 23:20:26 -08005246 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07005247}
5248
Christoph Lameter98246012008-01-07 23:20:26 -08005249static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07005250 const char *buf, size_t length)
5251{
Christoph Lameter0121c6192008-04-29 16:11:12 -07005252 unsigned long ratio;
5253 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005254
Jingoo Han3dbb95f2013-09-11 14:20:25 -07005255 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07005256 if (err)
5257 return err;
5258
Christoph Lametere2cb96b2008-08-19 08:51:22 -05005259 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07005260 s->remote_node_defrag_ratio = ratio * 10;
5261
Christoph Lameter81819f02007-05-06 14:49:36 -07005262 return length;
5263}
Christoph Lameter98246012008-01-07 23:20:26 -08005264SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005265#endif
5266
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005267#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005268static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5269{
5270 unsigned long sum = 0;
5271 int cpu;
5272 int len;
5273 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5274
5275 if (!data)
5276 return -ENOMEM;
5277
5278 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005279 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005280
5281 data[cpu] = x;
5282 sum += x;
5283 }
5284
5285 len = sprintf(buf, "%lu", sum);
5286
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005287#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005288 for_each_online_cpu(cpu) {
5289 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005290 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005291 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005292#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005293 kfree(data);
5294 return len + sprintf(buf + len, "\n");
5295}
5296
David Rientjes78eb00c2009-10-15 02:20:22 -07005297static void clear_stat(struct kmem_cache *s, enum stat_item si)
5298{
5299 int cpu;
5300
5301 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005302 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005303}
5304
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005305#define STAT_ATTR(si, text) \
5306static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5307{ \
5308 return show_stat(s, buf, si); \
5309} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005310static ssize_t text##_store(struct kmem_cache *s, \
5311 const char *buf, size_t length) \
5312{ \
5313 if (buf[0] != '0') \
5314 return -EINVAL; \
5315 clear_stat(s, si); \
5316 return length; \
5317} \
5318SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005319
5320STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5321STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5322STAT_ATTR(FREE_FASTPATH, free_fastpath);
5323STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5324STAT_ATTR(FREE_FROZEN, free_frozen);
5325STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5326STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5327STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5328STAT_ATTR(ALLOC_SLAB, alloc_slab);
5329STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005330STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005331STAT_ATTR(FREE_SLAB, free_slab);
5332STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5333STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5334STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5335STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5336STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5337STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005338STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005339STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005340STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5341STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005342STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5343STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005344STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5345STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005346#endif
5347
Pekka Enberg06428782008-01-07 23:20:27 -08005348static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005349 &slab_size_attr.attr,
5350 &object_size_attr.attr,
5351 &objs_per_slab_attr.attr,
5352 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005353 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005354 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005355 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005356 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005357 &partial_attr.attr,
5358 &cpu_slabs_attr.attr,
5359 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005360 &aliases_attr.attr,
5361 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005362 &hwcache_align_attr.attr,
5363 &reclaim_account_attr.attr,
5364 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005365 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005366 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005367 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005368#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005369 &total_objects_attr.attr,
5370 &slabs_attr.attr,
5371 &sanity_checks_attr.attr,
5372 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005373 &red_zone_attr.attr,
5374 &poison_attr.attr,
5375 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005376 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005377 &alloc_calls_attr.attr,
5378 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005379#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005380#ifdef CONFIG_ZONE_DMA
5381 &cache_dma_attr.attr,
5382#endif
5383#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005384 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005385#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005386#ifdef CONFIG_SLUB_STATS
5387 &alloc_fastpath_attr.attr,
5388 &alloc_slowpath_attr.attr,
5389 &free_fastpath_attr.attr,
5390 &free_slowpath_attr.attr,
5391 &free_frozen_attr.attr,
5392 &free_add_partial_attr.attr,
5393 &free_remove_partial_attr.attr,
5394 &alloc_from_partial_attr.attr,
5395 &alloc_slab_attr.attr,
5396 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005397 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005398 &free_slab_attr.attr,
5399 &cpuslab_flush_attr.attr,
5400 &deactivate_full_attr.attr,
5401 &deactivate_empty_attr.attr,
5402 &deactivate_to_head_attr.attr,
5403 &deactivate_to_tail_attr.attr,
5404 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005405 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005406 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005407 &cmpxchg_double_fail_attr.attr,
5408 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005409 &cpu_partial_alloc_attr.attr,
5410 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005411 &cpu_partial_node_attr.attr,
5412 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005413#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005414#ifdef CONFIG_FAILSLAB
5415 &failslab_attr.attr,
5416#endif
5417
Christoph Lameter81819f02007-05-06 14:49:36 -07005418 NULL
5419};
5420
5421static struct attribute_group slab_attr_group = {
5422 .attrs = slab_attrs,
5423};
5424
5425static ssize_t slab_attr_show(struct kobject *kobj,
5426 struct attribute *attr,
5427 char *buf)
5428{
5429 struct slab_attribute *attribute;
5430 struct kmem_cache *s;
5431 int err;
5432
5433 attribute = to_slab_attr(attr);
5434 s = to_slab(kobj);
5435
5436 if (!attribute->show)
5437 return -EIO;
5438
5439 err = attribute->show(s, buf);
5440
5441 return err;
5442}
5443
5444static ssize_t slab_attr_store(struct kobject *kobj,
5445 struct attribute *attr,
5446 const char *buf, size_t len)
5447{
5448 struct slab_attribute *attribute;
5449 struct kmem_cache *s;
5450 int err;
5451
5452 attribute = to_slab_attr(attr);
5453 s = to_slab(kobj);
5454
5455 if (!attribute->store)
5456 return -EIO;
5457
5458 err = attribute->store(s, buf, len);
Johannes Weiner127424c2016-01-20 15:02:32 -08005459#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005460 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08005461 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07005462
Glauber Costa107dab52012-12-18 14:23:05 -08005463 mutex_lock(&slab_mutex);
5464 if (s->max_attr_size < len)
5465 s->max_attr_size = len;
5466
Glauber Costaebe945c2012-12-18 14:23:10 -08005467 /*
5468 * This is a best effort propagation, so this function's return
5469 * value will be determined by the parent cache only. This is
5470 * basically because not all attributes will have a well
5471 * defined semantics for rollbacks - most of the actions will
5472 * have permanent effects.
5473 *
5474 * Returning the error value of any of the children that fail
5475 * is not 100 % defined, in the sense that users seeing the
5476 * error code won't be able to know anything about the state of
5477 * the cache.
5478 *
5479 * Only returning the error code for the parent cache at least
5480 * has well defined semantics. The cache being written to
5481 * directly either failed or succeeded, in which case we loop
5482 * through the descendants with best-effort propagation.
5483 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005484 for_each_memcg_cache(c, s)
5485 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005486 mutex_unlock(&slab_mutex);
5487 }
5488#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005489 return err;
5490}
5491
Glauber Costa107dab52012-12-18 14:23:05 -08005492static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5493{
Johannes Weiner127424c2016-01-20 15:02:32 -08005494#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005495 int i;
5496 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005497 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005498
Vladimir Davydov93030d82014-05-06 12:49:59 -07005499 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005500 return;
5501
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005502 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005503
Glauber Costa107dab52012-12-18 14:23:05 -08005504 /*
5505 * This mean this cache had no attribute written. Therefore, no point
5506 * in copying default values around
5507 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005508 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005509 return;
5510
5511 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5512 char mbuf[64];
5513 char *buf;
5514 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
Thomas Gleixner478fe302017-06-02 14:46:25 -07005515 ssize_t len;
Glauber Costa107dab52012-12-18 14:23:05 -08005516
5517 if (!attr || !attr->store || !attr->show)
5518 continue;
5519
5520 /*
5521 * It is really bad that we have to allocate here, so we will
5522 * do it only as a fallback. If we actually allocate, though,
5523 * we can just use the allocated buffer until the end.
5524 *
5525 * Most of the slub attributes will tend to be very small in
5526 * size, but sysfs allows buffers up to a page, so they can
5527 * theoretically happen.
5528 */
5529 if (buffer)
5530 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005531 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005532 buf = mbuf;
5533 else {
5534 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5535 if (WARN_ON(!buffer))
5536 continue;
5537 buf = buffer;
5538 }
5539
Thomas Gleixner478fe302017-06-02 14:46:25 -07005540 len = attr->show(root_cache, buf);
5541 if (len > 0)
5542 attr->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005543 }
5544
5545 if (buffer)
5546 free_page((unsigned long)buffer);
5547#endif
5548}
5549
Christoph Lameter41a21282014-05-06 12:50:08 -07005550static void kmem_cache_release(struct kobject *k)
5551{
5552 slab_kmem_cache_release(to_slab(k));
5553}
5554
Emese Revfy52cf25d2010-01-19 02:58:23 +01005555static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005556 .show = slab_attr_show,
5557 .store = slab_attr_store,
5558};
5559
5560static struct kobj_type slab_ktype = {
5561 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005562 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005563};
5564
5565static int uevent_filter(struct kset *kset, struct kobject *kobj)
5566{
5567 struct kobj_type *ktype = get_ktype(kobj);
5568
5569 if (ktype == &slab_ktype)
5570 return 1;
5571 return 0;
5572}
5573
Emese Revfy9cd43612009-12-31 14:52:51 +01005574static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005575 .filter = uevent_filter,
5576};
5577
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005578static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005579
Vladimir Davydov9a417072014-04-07 15:39:31 -07005580static inline struct kset *cache_kset(struct kmem_cache *s)
5581{
Johannes Weiner127424c2016-01-20 15:02:32 -08005582#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005583 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005584 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005585#endif
5586 return slab_kset;
5587}
5588
Christoph Lameter81819f02007-05-06 14:49:36 -07005589#define ID_STR_LENGTH 64
5590
5591/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005592 *
5593 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005594 */
5595static char *create_unique_id(struct kmem_cache *s)
5596{
5597 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5598 char *p = name;
5599
5600 BUG_ON(!name);
5601
5602 *p++ = ':';
5603 /*
5604 * First flags affecting slabcache operations. We will only
5605 * get here for aliasable slabs so we do not need to support
5606 * too many flags. The flags here must cover all flags that
5607 * are matched during merging to guarantee that the id is
5608 * unique.
5609 */
5610 if (s->flags & SLAB_CACHE_DMA)
5611 *p++ = 'd';
5612 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5613 *p++ = 'a';
Laura Abbottbecfda62016-03-15 14:55:06 -07005614 if (s->flags & SLAB_CONSISTENCY_CHECKS)
Christoph Lameter81819f02007-05-06 14:49:36 -07005615 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005616 if (!(s->flags & SLAB_NOTRACK))
5617 *p++ = 't';
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08005618 if (s->flags & SLAB_ACCOUNT)
5619 *p++ = 'A';
Christoph Lameter81819f02007-05-06 14:49:36 -07005620 if (p != name + 1)
5621 *p++ = '-';
5622 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005623
Christoph Lameter81819f02007-05-06 14:49:36 -07005624 BUG_ON(p > name + ID_STR_LENGTH - 1);
5625 return name;
5626}
5627
Tejun Heo3b7b3142017-06-23 15:08:52 -07005628static void sysfs_slab_remove_workfn(struct work_struct *work)
5629{
5630 struct kmem_cache *s =
5631 container_of(work, struct kmem_cache, kobj_remove_work);
5632
5633 if (!s->kobj.state_in_sysfs)
5634 /*
5635 * For a memcg cache, this may be called during
5636 * deactivation and again on shutdown. Remove only once.
5637 * A cache is never shut down before deactivation is
5638 * complete, so no need to worry about synchronization.
5639 */
5640 return;
5641
5642#ifdef CONFIG_MEMCG
5643 kset_unregister(s->memcg_kset);
5644#endif
5645 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5646 kobject_del(&s->kobj);
5647 kobject_put(&s->kobj);
5648}
5649
Christoph Lameter81819f02007-05-06 14:49:36 -07005650static int sysfs_slab_add(struct kmem_cache *s)
5651{
5652 int err;
5653 const char *name;
Tejun Heo1663f262017-02-22 15:41:39 -08005654 struct kset *kset = cache_kset(s);
Christoph Lameter45530c42012-11-28 16:23:07 +00005655 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005656
Tejun Heo3b7b3142017-06-23 15:08:52 -07005657 INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5658
Tejun Heo1663f262017-02-22 15:41:39 -08005659 if (!kset) {
5660 kobject_init(&s->kobj, &slab_ktype);
5661 return 0;
5662 }
5663
Christoph Lameter81819f02007-05-06 14:49:36 -07005664 if (unmergeable) {
5665 /*
5666 * Slabcache can never be merged so we can use the name proper.
5667 * This is typically the case for debug situations. In that
5668 * case we can catch duplicate names easily.
5669 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005670 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005671 name = s->name;
5672 } else {
5673 /*
5674 * Create a unique name for the slab as a target
5675 * for the symlinks.
5676 */
5677 name = create_unique_id(s);
5678 }
5679
Tejun Heo1663f262017-02-22 15:41:39 -08005680 s->kobj.kset = kset;
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005681 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005682 if (err)
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005683 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005684
5685 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005686 if (err)
5687 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005688
Johannes Weiner127424c2016-01-20 15:02:32 -08005689#ifdef CONFIG_MEMCG
Tejun Heo1663f262017-02-22 15:41:39 -08005690 if (is_root_cache(s) && memcg_sysfs_enabled) {
Vladimir Davydov9a417072014-04-07 15:39:31 -07005691 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5692 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005693 err = -ENOMEM;
5694 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005695 }
5696 }
5697#endif
5698
Christoph Lameter81819f02007-05-06 14:49:36 -07005699 kobject_uevent(&s->kobj, KOBJ_ADD);
5700 if (!unmergeable) {
5701 /* Setup first alias */
5702 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005703 }
Dave Jones54b6a732014-04-07 15:39:32 -07005704out:
5705 if (!unmergeable)
5706 kfree(name);
5707 return err;
5708out_del_kobj:
5709 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005710 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005711}
5712
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005713static void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005714{
Christoph Lameter97d06602012-07-06 15:25:11 -05005715 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005716 /*
5717 * Sysfs has not been setup yet so no need to remove the
5718 * cache from sysfs.
5719 */
5720 return;
5721
Tejun Heo3b7b3142017-06-23 15:08:52 -07005722 kobject_get(&s->kobj);
5723 schedule_work(&s->kobj_remove_work);
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005724}
5725
5726void sysfs_slab_release(struct kmem_cache *s)
5727{
5728 if (slab_state >= FULL)
5729 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005730}
5731
5732/*
5733 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005734 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005735 */
5736struct saved_alias {
5737 struct kmem_cache *s;
5738 const char *name;
5739 struct saved_alias *next;
5740};
5741
Adrian Bunk5af328a2007-07-17 04:03:27 -07005742static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005743
5744static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5745{
5746 struct saved_alias *al;
5747
Christoph Lameter97d06602012-07-06 15:25:11 -05005748 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005749 /*
5750 * If we have a leftover link then remove it.
5751 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005752 sysfs_remove_link(&slab_kset->kobj, name);
5753 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005754 }
5755
5756 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5757 if (!al)
5758 return -ENOMEM;
5759
5760 al->s = s;
5761 al->name = name;
5762 al->next = alias_list;
5763 alias_list = al;
5764 return 0;
5765}
5766
5767static int __init slab_sysfs_init(void)
5768{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005769 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005770 int err;
5771
Christoph Lameter18004c52012-07-06 15:25:12 -05005772 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005773
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005774 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005775 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005776 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005777 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005778 return -ENOSYS;
5779 }
5780
Christoph Lameter97d06602012-07-06 15:25:11 -05005781 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005782
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005783 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005784 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005785 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005786 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5787 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005788 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005789
5790 while (alias_list) {
5791 struct saved_alias *al = alias_list;
5792
5793 alias_list = alias_list->next;
5794 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005795 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005796 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5797 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005798 kfree(al);
5799 }
5800
Christoph Lameter18004c52012-07-06 15:25:12 -05005801 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005802 resiliency_test();
5803 return 0;
5804}
5805
5806__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005807#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005808
5809/*
5810 * The /proc/slabinfo ABI
5811 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005812#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005813void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005814{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005815 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005816 unsigned long nr_objs = 0;
5817 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005818 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005819 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005820
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005821 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005822 nr_slabs += node_nr_slabs(n);
5823 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005824 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005825 }
5826
Glauber Costa0d7561c2012-10-19 18:20:27 +04005827 sinfo->active_objs = nr_objs - nr_free;
5828 sinfo->num_objs = nr_objs;
5829 sinfo->active_slabs = nr_slabs;
5830 sinfo->num_slabs = nr_slabs;
5831 sinfo->objects_per_slab = oo_objects(s->oo);
5832 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005833}
5834
Glauber Costa0d7561c2012-10-19 18:20:27 +04005835void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005836{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005837}
5838
Glauber Costab7454ad2012-10-19 18:20:25 +04005839ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5840 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005841{
Glauber Costab7454ad2012-10-19 18:20:25 +04005842 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005843}
Linus Torvalds158a9622008-01-02 13:04:48 -08005844#endif /* CONFIG_SLABINFO */