blob: cae85b2db87bd3cd1dc1d63446f33d3c7a22e156 [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);
Christoph Lameter81819f02007-05-06 14:49:36 -0700217#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700218static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
219static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
220 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800221static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700222#endif
223
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500224static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225{
226#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700227 /*
228 * The rmw is racy on a preemptible kernel but this is acceptable, so
229 * avoid this_cpu_add()'s irq-disable overhead.
230 */
231 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800232#endif
233}
234
Christoph Lameter81819f02007-05-06 14:49:36 -0700235/********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
238
Christoph Lameter7656c722007-05-09 02:32:40 -0700239static inline void *get_freepointer(struct kmem_cache *s, void *object)
240{
241 return *(void **)(object + s->offset);
242}
243
Eric Dumazet0ad95002011-12-16 16:25:34 +0100244static void prefetch_freepointer(const struct kmem_cache *s, void *object)
245{
246 prefetch(object + s->offset);
247}
248
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500249static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
250{
251 void *p;
252
Joonsoo Kim922d5662016-03-17 14:17:53 -0700253 if (!debug_pagealloc_enabled())
254 return get_freepointer(s, object);
255
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500256 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500257 return p;
258}
259
Christoph Lameter7656c722007-05-09 02:32:40 -0700260static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
261{
262 *(void **)(object + s->offset) = fp;
263}
264
265/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300266#define for_each_object(__p, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700267 for (__p = fixup_red_left(__s, __addr); \
268 __p < (__addr) + (__objects) * (__s)->size; \
269 __p += (__s)->size)
Christoph Lameter7656c722007-05-09 02:32:40 -0700270
Wei Yang54266642014-08-06 16:04:42 -0700271#define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700272 for (__p = fixup_red_left(__s, __addr), __idx = 1; \
273 __idx <= __objects; \
274 __p += (__s)->size, __idx++)
Wei Yang54266642014-08-06 16:04:42 -0700275
Christoph Lameter7656c722007-05-09 02:32:40 -0700276/* Determine object index from a given position */
277static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
278{
279 return (p - addr) / s->size;
280}
281
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800282static inline int order_objects(int order, unsigned long size, int reserved)
283{
284 return ((PAGE_SIZE << order) - reserved) / size;
285}
286
Christoph Lameter834f3d12008-04-14 19:11:31 +0300287static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800288 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300289{
290 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800291 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300292 };
293
294 return x;
295}
296
297static inline int oo_order(struct kmem_cache_order_objects x)
298{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400299 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300300}
301
302static inline int oo_objects(struct kmem_cache_order_objects x)
303{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400304 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300305}
306
Christoph Lameter881db7f2011-06-01 12:25:53 -0500307/*
308 * Per slab locking using the pagelock
309 */
310static __always_inline void slab_lock(struct page *page)
311{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800312 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500313 bit_spin_lock(PG_locked, &page->flags);
314}
315
316static __always_inline void slab_unlock(struct page *page)
317{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800318 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500319 __bit_spin_unlock(PG_locked, &page->flags);
320}
321
Dave Hansena0320862014-01-30 15:46:09 -0800322static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
323{
324 struct page tmp;
325 tmp.counters = counters_new;
326 /*
327 * page->counters can cover frozen/inuse/objects as well
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700328 * as page->_refcount. If we assign to ->counters directly
329 * we run the risk of losing updates to page->_refcount, so
Dave Hansena0320862014-01-30 15:46:09 -0800330 * be careful and only assign to the fields we need.
331 */
332 page->frozen = tmp.frozen;
333 page->inuse = tmp.inuse;
334 page->objects = tmp.objects;
335}
336
Christoph Lameter1d071712011-07-14 12:49:12 -0500337/* Interrupts must be disabled (for the fallback code to work right) */
338static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500339 void *freelist_old, unsigned long counters_old,
340 void *freelist_new, unsigned long counters_new,
341 const char *n)
342{
Christoph Lameter1d071712011-07-14 12:49:12 -0500343 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800344#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
345 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500346 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000347 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700348 freelist_old, counters_old,
349 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700350 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500351 } else
352#endif
353 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500354 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800355 if (page->freelist == freelist_old &&
356 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500357 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800358 set_page_slub_counters(page, counters_new);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500359 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700360 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500361 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500362 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500363 }
364
365 cpu_relax();
366 stat(s, CMPXCHG_DOUBLE_FAIL);
367
368#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700369 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500370#endif
371
Joe Perches6f6528a2015-04-14 15:44:31 -0700372 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373}
374
Christoph Lameter1d071712011-07-14 12:49:12 -0500375static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
376 void *freelist_old, unsigned long counters_old,
377 void *freelist_new, unsigned long counters_new,
378 const char *n)
379{
Heiko Carstens25654092012-01-12 17:17:33 -0800380#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
381 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500382 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000383 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700384 freelist_old, counters_old,
385 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700386 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500387 } else
388#endif
389 {
390 unsigned long flags;
391
392 local_irq_save(flags);
393 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800394 if (page->freelist == freelist_old &&
395 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500396 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800397 set_page_slub_counters(page, counters_new);
Christoph Lameter1d071712011-07-14 12:49:12 -0500398 slab_unlock(page);
399 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700400 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500401 }
402 slab_unlock(page);
403 local_irq_restore(flags);
404 }
405
406 cpu_relax();
407 stat(s, CMPXCHG_DOUBLE_FAIL);
408
409#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700410 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500411#endif
412
Joe Perches6f6528a2015-04-14 15:44:31 -0700413 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500414}
415
Christoph Lameter41ecc552007-05-09 02:32:44 -0700416#ifdef CONFIG_SLUB_DEBUG
417/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500418 * Determine a map of object in use on a page.
419 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500420 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500421 * not vanish from under us.
422 */
423static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
424{
425 void *p;
426 void *addr = page_address(page);
427
428 for (p = page->freelist; p; p = get_freepointer(s, p))
429 set_bit(slab_index(p, s, addr), map);
430}
431
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700432static inline int size_from_object(struct kmem_cache *s)
433{
434 if (s->flags & SLAB_RED_ZONE)
435 return s->size - s->red_left_pad;
436
437 return s->size;
438}
439
440static inline void *restore_red_left(struct kmem_cache *s, void *p)
441{
442 if (s->flags & SLAB_RED_ZONE)
443 p -= s->red_left_pad;
444
445 return p;
446}
447
Christoph Lameter41ecc552007-05-09 02:32:44 -0700448/*
449 * Debug settings:
450 */
Andrey Ryabinin89d3c872015-11-05 18:51:23 -0800451#if defined(CONFIG_SLUB_DEBUG_ON)
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700452static int slub_debug = DEBUG_DEFAULT_FLAGS;
453#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700454static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700455#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700456
457static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700458static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700459
Christoph Lameter7656c722007-05-09 02:32:40 -0700460/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800461 * slub is about to manipulate internal object metadata. This memory lies
462 * outside the range of the allocated object, so accessing it would normally
463 * be reported by kasan as a bounds error. metadata_access_enable() is used
464 * to tell kasan that these accesses are OK.
465 */
466static inline void metadata_access_enable(void)
467{
468 kasan_disable_current();
469}
470
471static inline void metadata_access_disable(void)
472{
473 kasan_enable_current();
474}
475
476/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700477 * Object debugging
478 */
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700479
480/* Verify that a pointer has an address that is valid within a slab page */
481static inline int check_valid_pointer(struct kmem_cache *s,
482 struct page *page, void *object)
483{
484 void *base;
485
486 if (!object)
487 return 1;
488
489 base = page_address(page);
490 object = restore_red_left(s, object);
491 if (object < base || object >= base + page->objects * s->size ||
492 (object - base) % s->size) {
493 return 0;
494 }
495
496 return 1;
497}
498
Daniel Thompson74b41632017-01-24 15:18:02 -0800499static void print_section(char *level, char *text, u8 *addr,
500 unsigned int length)
Christoph Lameter81819f02007-05-06 14:49:36 -0700501{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800502 metadata_access_enable();
Daniel Thompson74b41632017-01-24 15:18:02 -0800503 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200504 length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800505 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700506}
507
Christoph Lameter81819f02007-05-06 14:49:36 -0700508static struct track *get_track(struct kmem_cache *s, void *object,
509 enum track_item alloc)
510{
511 struct track *p;
512
513 if (s->offset)
514 p = object + s->offset + sizeof(void *);
515 else
516 p = object + s->inuse;
517
518 return p + alloc;
519}
520
521static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300522 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700523{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900524 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700525
Christoph Lameter81819f02007-05-06 14:49:36 -0700526 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700527#ifdef CONFIG_STACKTRACE
528 struct stack_trace trace;
529 int i;
530
531 trace.nr_entries = 0;
532 trace.max_entries = TRACK_ADDRS_COUNT;
533 trace.entries = p->addrs;
534 trace.skip = 3;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800535 metadata_access_enable();
Ben Greeard6543e32011-07-07 11:36:36 -0700536 save_stack_trace(&trace);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800537 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700538
539 /* See rant in lockdep.c */
540 if (trace.nr_entries != 0 &&
541 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
542 trace.nr_entries--;
543
544 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
545 p->addrs[i] = 0;
546#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700547 p->addr = addr;
548 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400549 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700550 p->when = jiffies;
551 } else
552 memset(p, 0, sizeof(struct track));
553}
554
Christoph Lameter81819f02007-05-06 14:49:36 -0700555static void init_tracking(struct kmem_cache *s, void *object)
556{
Christoph Lameter24922682007-07-17 04:03:18 -0700557 if (!(s->flags & SLAB_STORE_USER))
558 return;
559
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300560 set_track(s, object, TRACK_FREE, 0UL);
561 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700562}
563
564static void print_track(const char *s, struct track *t)
565{
566 if (!t->addr)
567 return;
568
Fabian Frederickf9f58282014-06-04 16:06:34 -0700569 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
570 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700571#ifdef CONFIG_STACKTRACE
572 {
573 int i;
574 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
575 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700576 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700577 else
578 break;
579 }
580#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700581}
582
Christoph Lameter24922682007-07-17 04:03:18 -0700583static void print_tracking(struct kmem_cache *s, void *object)
584{
585 if (!(s->flags & SLAB_STORE_USER))
586 return;
587
588 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
589 print_track("Freed", get_track(s, object, TRACK_FREE));
590}
591
592static void print_page_info(struct page *page)
593{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700594 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800595 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700596
597}
598
599static void slab_bug(struct kmem_cache *s, char *fmt, ...)
600{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700601 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700602 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700603
604 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700605 vaf.fmt = fmt;
606 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700607 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700608 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700609 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400610
Rusty Russell373d4d02013-01-21 17:17:39 +1030611 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700612 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700613}
614
615static void slab_fix(struct kmem_cache *s, char *fmt, ...)
616{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700617 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700618 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700619
620 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700621 vaf.fmt = fmt;
622 vaf.va = &args;
623 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700624 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700625}
626
627static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700628{
629 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800630 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700631
632 print_tracking(s, p);
633
634 print_page_info(page);
635
Fabian Frederickf9f58282014-06-04 16:06:34 -0700636 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
637 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700638
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700639 if (s->flags & SLAB_RED_ZONE)
Daniel Thompson74b41632017-01-24 15:18:02 -0800640 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
641 s->red_left_pad);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700642 else if (p > addr + 16)
Daniel Thompson74b41632017-01-24 15:18:02 -0800643 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700644
Daniel Thompson74b41632017-01-24 15:18:02 -0800645 print_section(KERN_ERR, "Object ", p,
646 min_t(unsigned long, s->object_size, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700647 if (s->flags & SLAB_RED_ZONE)
Daniel Thompson74b41632017-01-24 15:18:02 -0800648 print_section(KERN_ERR, "Redzone ", p + s->object_size,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500649 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700650
Christoph Lameter81819f02007-05-06 14:49:36 -0700651 if (s->offset)
652 off = s->offset + sizeof(void *);
653 else
654 off = s->inuse;
655
Christoph Lameter24922682007-07-17 04:03:18 -0700656 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700657 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700658
Alexander Potapenko80a92012016-07-28 15:49:07 -0700659 off += kasan_metadata_size(s);
660
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700661 if (off != size_from_object(s))
Christoph Lameter81819f02007-05-06 14:49:36 -0700662 /* Beginning of the filler is the free pointer */
Daniel Thompson74b41632017-01-24 15:18:02 -0800663 print_section(KERN_ERR, "Padding ", p + off,
664 size_from_object(s) - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700665
666 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700667}
668
David Keiteldac9c382014-10-21 16:26:17 -0700669#ifdef CONFIG_SLUB_DEBUG_PANIC_ON
670static void slab_panic(const char *cause)
671{
672 panic("%s\n", cause);
673}
674#else
675static inline void slab_panic(const char *cause) {}
676#endif
677
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800678void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700679 u8 *object, char *reason)
680{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700681 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700682 print_trailer(s, page, object);
David Keiteldac9c382014-10-21 16:26:17 -0700683 slab_panic(reason);
Christoph Lameter81819f02007-05-06 14:49:36 -0700684}
685
Mathieu Malaterrec99dbd92018-06-07 17:05:17 -0700686static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800687 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700688{
689 va_list args;
690 char buf[100];
691
Christoph Lameter24922682007-07-17 04:03:18 -0700692 va_start(args, fmt);
693 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700694 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700695 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700696 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700697 dump_stack();
David Keiteldac9c382014-10-21 16:26:17 -0700698 slab_panic("slab error");
Christoph Lameter81819f02007-05-06 14:49:36 -0700699}
700
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500701static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700702{
703 u8 *p = object;
704
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700705 if (s->flags & SLAB_RED_ZONE)
706 memset(p - s->red_left_pad, val, s->red_left_pad);
707
Christoph Lameter81819f02007-05-06 14:49:36 -0700708 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500709 memset(p, POISON_FREE, s->object_size - 1);
710 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700711 }
712
713 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500714 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700715}
716
Christoph Lameter24922682007-07-17 04:03:18 -0700717static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
718 void *from, void *to)
719{
David Keiteldac9c382014-10-21 16:26:17 -0700720 slab_panic("object poison overwritten");
Christoph Lameter24922682007-07-17 04:03:18 -0700721 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
722 memset(from, data, to - from);
723}
724
725static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
726 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800727 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700728{
729 u8 *fault;
730 u8 *end;
731
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800732 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700733 fault = memchr_inv(start, value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800734 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700735 if (!fault)
736 return 1;
737
738 end = start + bytes;
739 while (end > fault && end[-1] == value)
740 end--;
741
742 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700743 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700744 fault, end - 1, fault[0], value);
745 print_trailer(s, page, object);
746
747 restore_bytes(s, what, value, fault, end);
748 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700749}
750
Christoph Lameter81819f02007-05-06 14:49:36 -0700751/*
752 * Object layout:
753 *
754 * object address
755 * Bytes of the object to be managed.
756 * If the freepointer may overlay the object then the free
757 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700758 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700759 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
760 * 0xa5 (POISON_END)
761 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500762 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700763 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700764 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500765 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700766 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700767 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
768 * 0xcc (RED_ACTIVE) for objects in use.
769 *
770 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700771 * Meta data starts here.
772 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700773 * A. Free pointer (if we cannot overwrite object on free)
774 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700775 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800776 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700777 * before the word boundary.
778 *
779 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700780 *
781 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700782 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700783 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500784 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700785 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700786 * may be used with merged slabcaches.
787 */
788
Christoph Lameter81819f02007-05-06 14:49:36 -0700789static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
790{
791 unsigned long off = s->inuse; /* The end of info */
792
793 if (s->offset)
794 /* Freepointer is placed after the object. */
795 off += sizeof(void *);
796
797 if (s->flags & SLAB_STORE_USER)
798 /* We also have user information there */
799 off += 2 * sizeof(struct track);
800
Alexander Potapenko80a92012016-07-28 15:49:07 -0700801 off += kasan_metadata_size(s);
802
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700803 if (size_from_object(s) == off)
Christoph Lameter81819f02007-05-06 14:49:36 -0700804 return 1;
805
Christoph Lameter24922682007-07-17 04:03:18 -0700806 return check_bytes_and_report(s, page, p, "Object padding",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700807 p + off, POISON_INUSE, size_from_object(s) - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700808}
809
Christoph Lameter39b26462008-04-14 19:11:30 +0300810/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700811static int slab_pad_check(struct kmem_cache *s, struct page *page)
812{
Christoph Lameter24922682007-07-17 04:03:18 -0700813 u8 *start;
814 u8 *fault;
815 u8 *end;
816 int length;
817 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700818
819 if (!(s->flags & SLAB_POISON))
820 return 1;
821
Christoph Lametera973e9d2008-03-01 13:40:44 -0800822 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800823 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300824 end = start + length;
825 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700826 if (!remainder)
827 return 1;
828
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800829 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700830 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800831 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700832 if (!fault)
833 return 1;
834 while (end > fault && end[-1] == POISON_INUSE)
835 end--;
836
837 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Daniel Thompson74b41632017-01-24 15:18:02 -0800838 print_section(KERN_ERR, "Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700839
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200840 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700841 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700842}
843
844static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500845 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700846{
847 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500848 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700849
850 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700851 if (!check_bytes_and_report(s, page, object, "Redzone",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700852 object - s->red_left_pad, val, s->red_left_pad))
853 return 0;
854
855 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500856 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700857 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700858 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500859 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800860 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800861 endobject, POISON_INUSE,
862 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800863 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700864 }
865
866 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500867 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700868 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500869 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700870 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500871 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700872 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700873 /*
874 * check_pad_bytes cleans up on its own.
875 */
876 check_pad_bytes(s, page, p);
877 }
878
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500879 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700880 /*
881 * Object and freepointer overlap. Cannot check
882 * freepointer while object is allocated.
883 */
884 return 1;
885
886 /* Check free pointer validity */
887 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
888 object_err(s, page, p, "Freepointer corrupt");
889 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100890 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700891 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700892 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700893 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800894 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700895 return 0;
896 }
897 return 1;
898}
899
900static int check_slab(struct kmem_cache *s, struct page *page)
901{
Christoph Lameter39b26462008-04-14 19:11:30 +0300902 int maxobj;
903
Christoph Lameter81819f02007-05-06 14:49:36 -0700904 VM_BUG_ON(!irqs_disabled());
905
906 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700907 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700908 return 0;
909 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300910
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800911 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300912 if (page->objects > maxobj) {
913 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800914 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +0300915 return 0;
916 }
917 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700918 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800919 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700920 return 0;
921 }
922 /* Slab_pad_check fixes things up after itself */
923 slab_pad_check(s, page);
924 return 1;
925}
926
927/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700928 * Determine if a certain object on a page is on the freelist. Must hold the
929 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700930 */
931static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
932{
933 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500934 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700935 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800936 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700937
Christoph Lameter881db7f2011-06-01 12:25:53 -0500938 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300939 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700940 if (fp == search)
941 return 1;
942 if (!check_valid_pointer(s, page, fp)) {
943 if (object) {
944 object_err(s, page, object,
945 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800946 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700947 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700948 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800949 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300950 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700951 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700952 return 0;
953 }
954 break;
955 }
956 object = fp;
957 fp = get_freepointer(s, object);
958 nr++;
959 }
960
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800961 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400962 if (max_objects > MAX_OBJS_PER_PAGE)
963 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300964
965 if (page->objects != max_objects) {
Joe Perches756a025f02016-03-17 14:19:47 -0700966 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
967 page->objects, max_objects);
Christoph Lameter224a88b2008-04-14 19:11:31 +0300968 page->objects = max_objects;
969 slab_fix(s, "Number of objects adjusted.");
970 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300971 if (page->inuse != page->objects - nr) {
Joe Perches756a025f02016-03-17 14:19:47 -0700972 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
973 page->inuse, page->objects - nr);
Christoph Lameter39b26462008-04-14 19:11:30 +0300974 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700975 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700976 }
977 return search == NULL;
978}
979
Christoph Lameter0121c6192008-04-29 16:11:12 -0700980static void trace(struct kmem_cache *s, struct page *page, void *object,
981 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700982{
983 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -0700984 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -0700985 s->name,
986 alloc ? "alloc" : "free",
987 object, page->inuse,
988 page->freelist);
989
990 if (!alloc)
Daniel Thompson74b41632017-01-24 15:18:02 -0800991 print_section(KERN_INFO, "Object ", (void *)object,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800992 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700993
994 dump_stack();
995 }
996}
997
Christoph Lameter643b1132007-05-06 14:49:42 -0700998/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700999 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -07001000 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001001static void add_full(struct kmem_cache *s,
1002 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -07001003{
Christoph Lameter643b1132007-05-06 14:49:42 -07001004 if (!(s->flags & SLAB_STORE_USER))
1005 return;
1006
David Rientjes255d0882014-02-10 14:25:39 -08001007 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001008 list_add(&page->lru, &n->full);
1009}
Christoph Lameter643b1132007-05-06 14:49:42 -07001010
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001011static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001012{
1013 if (!(s->flags & SLAB_STORE_USER))
1014 return;
1015
David Rientjes255d0882014-02-10 14:25:39 -08001016 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -07001017 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001018}
1019
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001020/* Tracking of the number of slabs for debugging purposes */
1021static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1022{
1023 struct kmem_cache_node *n = get_node(s, node);
1024
1025 return atomic_long_read(&n->nr_slabs);
1026}
1027
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001028static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1029{
1030 return atomic_long_read(&n->nr_slabs);
1031}
1032
Christoph Lameter205ab992008-04-14 19:11:40 +03001033static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001034{
1035 struct kmem_cache_node *n = get_node(s, node);
1036
1037 /*
1038 * May be called early in order to allocate a slab for the
1039 * kmem_cache_node structure. Solve the chicken-egg
1040 * dilemma by deferring the increment of the count during
1041 * bootstrap (see early_kmem_cache_node_alloc).
1042 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001043 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001044 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001045 atomic_long_add(objects, &n->total_objects);
1046 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001047}
Christoph Lameter205ab992008-04-14 19:11:40 +03001048static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001049{
1050 struct kmem_cache_node *n = get_node(s, node);
1051
1052 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001053 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001054}
1055
1056/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001057static void setup_object_debug(struct kmem_cache *s, struct page *page,
1058 void *object)
1059{
1060 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1061 return;
1062
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001063 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001064 init_tracking(s, object);
1065}
1066
Laura Abbottbecfda62016-03-15 14:55:06 -07001067static inline int alloc_consistency_checks(struct kmem_cache *s,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001068 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001069 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001070{
1071 if (!check_slab(s, page))
Laura Abbottbecfda62016-03-15 14:55:06 -07001072 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001073
Christoph Lameter81819f02007-05-06 14:49:36 -07001074 if (!check_valid_pointer(s, page, object)) {
1075 object_err(s, page, object, "Freelist Pointer check fails");
Laura Abbottbecfda62016-03-15 14:55:06 -07001076 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001077 }
1078
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001079 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Laura Abbottbecfda62016-03-15 14:55:06 -07001080 return 0;
1081
1082 return 1;
1083}
1084
1085static noinline int alloc_debug_processing(struct kmem_cache *s,
1086 struct page *page,
1087 void *object, unsigned long addr)
1088{
1089 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1090 if (!alloc_consistency_checks(s, page, object, addr))
1091 goto bad;
1092 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001093
Christoph Lameter3ec09742007-05-16 22:11:00 -07001094 /* Success perform special debug activities for allocs */
1095 if (s->flags & SLAB_STORE_USER)
1096 set_track(s, object, TRACK_ALLOC, addr);
1097 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001098 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001099 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001100
Christoph Lameter81819f02007-05-06 14:49:36 -07001101bad:
1102 if (PageSlab(page)) {
1103 /*
1104 * If this is a slab page then lets do the best we can
1105 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001106 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001107 */
Christoph Lameter24922682007-07-17 04:03:18 -07001108 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001109 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001110 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001111 }
1112 return 0;
1113}
1114
Laura Abbottbecfda62016-03-15 14:55:06 -07001115static inline int free_consistency_checks(struct kmem_cache *s,
1116 struct page *page, void *object, unsigned long addr)
1117{
1118 if (!check_valid_pointer(s, page, object)) {
1119 slab_err(s, page, "Invalid object pointer 0x%p", object);
1120 return 0;
1121 }
1122
1123 if (on_freelist(s, page, object)) {
1124 object_err(s, page, object, "Object already free");
1125 return 0;
1126 }
1127
1128 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1129 return 0;
1130
1131 if (unlikely(s != page->slab_cache)) {
1132 if (!PageSlab(page)) {
Joe Perches756a025f02016-03-17 14:19:47 -07001133 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1134 object);
Laura Abbottbecfda62016-03-15 14:55:06 -07001135 } else if (!page->slab_cache) {
1136 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1137 object);
1138 dump_stack();
1139 } else
1140 object_err(s, page, object,
1141 "page slab pointer corrupt.");
1142 return 0;
1143 }
1144 return 1;
1145}
1146
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001147/* Supports checking bulk free of a constructed freelist */
Laura Abbott282acb42016-03-15 14:54:59 -07001148static noinline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001149 struct kmem_cache *s, struct page *page,
1150 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001151 unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001152{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001153 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001154 void *object = head;
1155 int cnt = 0;
Laura Abbott282acb42016-03-15 14:54:59 -07001156 unsigned long uninitialized_var(flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001157 int ret = 0;
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001158
Laura Abbott282acb42016-03-15 14:54:59 -07001159 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001160 slab_lock(page);
1161
Laura Abbottbecfda62016-03-15 14:55:06 -07001162 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1163 if (!check_slab(s, page))
1164 goto out;
1165 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001166
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001167next_object:
1168 cnt++;
1169
Laura Abbottbecfda62016-03-15 14:55:06 -07001170 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1171 if (!free_consistency_checks(s, page, object, addr))
1172 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001173 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001174
Christoph Lameter3ec09742007-05-16 22:11:00 -07001175 if (s->flags & SLAB_STORE_USER)
1176 set_track(s, object, TRACK_FREE, addr);
1177 trace(s, page, object, 0);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001178 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001179 init_object(s, object, SLUB_RED_INACTIVE);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001180
1181 /* Reached end of constructed freelist yet? */
1182 if (object != tail) {
1183 object = get_freepointer(s, object);
1184 goto next_object;
1185 }
Laura Abbott804aa132016-03-15 14:55:02 -07001186 ret = 1;
1187
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001188out:
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001189 if (cnt != bulk_cnt)
1190 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1191 bulk_cnt, cnt);
1192
Christoph Lameter881db7f2011-06-01 12:25:53 -05001193 slab_unlock(page);
Laura Abbott282acb42016-03-15 14:54:59 -07001194 spin_unlock_irqrestore(&n->list_lock, flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001195 if (!ret)
1196 slab_fix(s, "Object at 0x%p not freed", object);
1197 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001198}
1199
Christoph Lameter41ecc552007-05-09 02:32:44 -07001200static int __init setup_slub_debug(char *str)
1201{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001202 slub_debug = DEBUG_DEFAULT_FLAGS;
1203 if (*str++ != '=' || !*str)
1204 /*
1205 * No options specified. Switch on full debugging.
1206 */
1207 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001208
1209 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001210 /*
1211 * No options but restriction on slabs. This means full
1212 * debugging for slabs matching a pattern.
1213 */
1214 goto check_slabs;
1215
1216 slub_debug = 0;
1217 if (*str == '-')
1218 /*
1219 * Switch off all debugging measures.
1220 */
1221 goto out;
1222
1223 /*
1224 * Determine which debug features should be switched on
1225 */
Pekka Enberg06428782008-01-07 23:20:27 -08001226 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001227 switch (tolower(*str)) {
1228 case 'f':
Laura Abbottbecfda62016-03-15 14:55:06 -07001229 slub_debug |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001230 break;
1231 case 'z':
1232 slub_debug |= SLAB_RED_ZONE;
1233 break;
1234 case 'p':
1235 slub_debug |= SLAB_POISON;
1236 break;
1237 case 'u':
1238 slub_debug |= SLAB_STORE_USER;
1239 break;
1240 case 't':
1241 slub_debug |= SLAB_TRACE;
1242 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001243 case 'a':
1244 slub_debug |= SLAB_FAILSLAB;
1245 break;
Chris J Arges08303a72015-04-14 15:44:25 -07001246 case 'o':
1247 /*
1248 * Avoid enabling debugging on caches if its minimum
1249 * order would increase as a result.
1250 */
1251 disable_higher_order_debug = 1;
1252 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001253 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001254 pr_err("slub_debug option '%c' unknown. skipped\n",
1255 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001256 }
1257 }
1258
1259check_slabs:
1260 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001261 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001262out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001263 return 1;
1264}
1265
1266__setup("slub_debug", setup_slub_debug);
1267
Joonsoo Kim423c9292014-10-09 15:26:22 -07001268unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001269 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001270 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001271{
1272 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001273 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001274 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001275 if (slub_debug && (!slub_debug_slabs || (name &&
1276 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001277 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001278
1279 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001280}
Jesper Dangaard Brouerb4a64712015-11-20 15:57:41 -08001281#else /* !CONFIG_SLUB_DEBUG */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001282static inline void setup_object_debug(struct kmem_cache *s,
1283 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001284
Christoph Lameter3ec09742007-05-16 22:11:00 -07001285static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001286 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001287
Laura Abbott282acb42016-03-15 14:54:59 -07001288static inline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001289 struct kmem_cache *s, struct page *page,
1290 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001291 unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001292
Christoph Lameter41ecc552007-05-09 02:32:44 -07001293static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1294 { return 1; }
1295static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001296 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001297static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1298 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001299static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1300 struct page *page) {}
Joonsoo Kim423c9292014-10-09 15:26:22 -07001301unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001302 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001303 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001304{
1305 return flags;
1306}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001307#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001308
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001309#define disable_higher_order_debug 0
1310
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001311static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1312 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001313static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1314 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001315static inline void inc_slabs_node(struct kmem_cache *s, int node,
1316 int objects) {}
1317static inline void dec_slabs_node(struct kmem_cache *s, int node,
1318 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001319
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001320#endif /* CONFIG_SLUB_DEBUG */
1321
1322/*
1323 * Hooks for other subsystems that check memory allocations. In a typical
1324 * production configuration these hooks all should produce no code at all.
1325 */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001326static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1327{
1328 kmemleak_alloc(ptr, size, 1, flags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07001329 kasan_kmalloc_large(ptr, size, flags);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001330}
1331
1332static inline void kfree_hook(const void *x)
1333{
1334 kmemleak_free(x);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001335 kasan_kfree_large(x);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001336}
1337
Alexander Potapenko80a92012016-07-28 15:49:07 -07001338static inline void *slab_free_hook(struct kmem_cache *s, void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001339{
Alexander Potapenko80a92012016-07-28 15:49:07 -07001340 void *freeptr;
1341
Roman Bobnievd56791b2013-10-08 15:58:57 -07001342 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001343
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001344 /*
1345 * Trouble is that we may no longer disable interrupts in the fast path
1346 * So in order to make the debug calls that expect irqs to be
1347 * disabled we need to disable interrupts temporarily.
1348 */
1349#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1350 {
1351 unsigned long flags;
1352
1353 local_irq_save(flags);
1354 kmemcheck_slab_free(s, x, s->object_size);
1355 debug_check_no_locks_freed(x, s->object_size);
1356 local_irq_restore(flags);
1357 }
1358#endif
1359 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1360 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001361
Alexander Potapenko80a92012016-07-28 15:49:07 -07001362 freeptr = get_freepointer(s, x);
1363 /*
1364 * kasan_slab_free() may put x into memory quarantine, delaying its
1365 * reuse. In this case the object's freelist pointer is changed.
1366 */
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001367 kasan_slab_free(s, x);
Alexander Potapenko80a92012016-07-28 15:49:07 -07001368 return freeptr;
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001369}
Christoph Lameter205ab992008-04-14 19:11:40 +03001370
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001371static inline void slab_free_freelist_hook(struct kmem_cache *s,
1372 void *head, void *tail)
1373{
1374/*
1375 * Compiler cannot detect this function can be removed if slab_free_hook()
1376 * evaluates to nothing. Thus, catch all relevant config debug options here.
1377 */
1378#if defined(CONFIG_KMEMCHECK) || \
1379 defined(CONFIG_LOCKDEP) || \
1380 defined(CONFIG_DEBUG_KMEMLEAK) || \
1381 defined(CONFIG_DEBUG_OBJECTS_FREE) || \
1382 defined(CONFIG_KASAN)
1383
1384 void *object = head;
1385 void *tail_obj = tail ? : head;
Alexander Potapenko80a92012016-07-28 15:49:07 -07001386 void *freeptr;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001387
1388 do {
Alexander Potapenko80a92012016-07-28 15:49:07 -07001389 freeptr = slab_free_hook(s, object);
1390 } while ((object != tail_obj) && (object = freeptr));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001391#endif
1392}
1393
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001394static void setup_object(struct kmem_cache *s, struct page *page,
1395 void *object)
1396{
1397 setup_object_debug(s, page, object);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07001398 kasan_init_slab_obj(s, object);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001399 if (unlikely(s->ctor)) {
1400 kasan_unpoison_object_data(s, object);
1401 s->ctor(object);
1402 kasan_poison_object_data(s, object);
1403 }
1404}
1405
Christoph Lameter81819f02007-05-06 14:49:36 -07001406/*
1407 * Slab allocation and freeing
1408 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001409static inline struct page *alloc_slab_page(struct kmem_cache *s,
1410 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001411{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001412 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001413 int order = oo_order(oo);
1414
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001415 flags |= __GFP_NOTRACK;
1416
Christoph Lameter2154a332010-07-09 14:07:10 -05001417 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001418 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001419 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001420 page = __alloc_pages_node(node, flags, order);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001421
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001422 if (page && memcg_charge_slab(page, flags, order, s)) {
1423 __free_pages(page, order);
1424 page = NULL;
1425 }
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001426
1427 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001428}
1429
Thomas Garnier210e7a42016-07-26 15:21:59 -07001430#ifdef CONFIG_SLAB_FREELIST_RANDOM
1431/* Pre-initialize the random sequence cache */
1432static int init_cache_random_seq(struct kmem_cache *s)
1433{
1434 int err;
1435 unsigned long i, count = oo_objects(s->oo);
1436
Sean Rees49f68cc2017-02-08 14:30:59 -08001437 /* Bailout if already initialised */
1438 if (s->random_seq)
1439 return 0;
1440
Thomas Garnier210e7a42016-07-26 15:21:59 -07001441 err = cache_random_seq_create(s, count, GFP_KERNEL);
1442 if (err) {
1443 pr_err("SLUB: Unable to initialize free list for %s\n",
1444 s->name);
1445 return err;
1446 }
1447
1448 /* Transform to an offset on the set of pages */
1449 if (s->random_seq) {
1450 for (i = 0; i < count; i++)
1451 s->random_seq[i] *= s->size;
1452 }
1453 return 0;
1454}
1455
1456/* Initialize each random sequence freelist per cache */
1457static void __init init_freelist_randomization(void)
1458{
1459 struct kmem_cache *s;
1460
1461 mutex_lock(&slab_mutex);
1462
1463 list_for_each_entry(s, &slab_caches, list)
1464 init_cache_random_seq(s);
1465
1466 mutex_unlock(&slab_mutex);
1467}
1468
1469/* Get the next entry on the pre-computed freelist randomized */
1470static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1471 unsigned long *pos, void *start,
1472 unsigned long page_limit,
1473 unsigned long freelist_count)
1474{
1475 unsigned int idx;
1476
1477 /*
1478 * If the target page allocation failed, the number of objects on the
1479 * page might be smaller than the usual size defined by the cache.
1480 */
1481 do {
1482 idx = s->random_seq[*pos];
1483 *pos += 1;
1484 if (*pos >= freelist_count)
1485 *pos = 0;
1486 } while (unlikely(idx >= page_limit));
1487
1488 return (char *)start + idx;
1489}
1490
1491/* Shuffle the single linked freelist based on a random pre-computed sequence */
1492static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1493{
1494 void *start;
1495 void *cur;
1496 void *next;
1497 unsigned long idx, pos, page_limit, freelist_count;
1498
1499 if (page->objects < 2 || !s->random_seq)
1500 return false;
1501
1502 freelist_count = oo_objects(s->oo);
1503 pos = get_random_int() % freelist_count;
1504
1505 page_limit = page->objects * s->size;
1506 start = fixup_red_left(s, page_address(page));
1507
1508 /* First entry is used as the base of the freelist */
1509 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1510 freelist_count);
1511 page->freelist = cur;
1512
1513 for (idx = 1; idx < page->objects; idx++) {
1514 setup_object(s, page, cur);
1515 next = next_freelist_entry(s, page, &pos, start, page_limit,
1516 freelist_count);
1517 set_freepointer(s, cur, next);
1518 cur = next;
1519 }
1520 setup_object(s, page, cur);
1521 set_freepointer(s, cur, NULL);
1522
1523 return true;
1524}
1525#else
1526static inline int init_cache_random_seq(struct kmem_cache *s)
1527{
1528 return 0;
1529}
1530static inline void init_freelist_randomization(void) { }
1531static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1532{
1533 return false;
1534}
1535#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1536
Christoph Lameter81819f02007-05-06 14:49:36 -07001537static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1538{
Pekka Enberg06428782008-01-07 23:20:27 -08001539 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001540 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001541 gfp_t alloc_gfp;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001542 void *start, *p;
1543 int idx, order;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001544 bool shuffle;
Christoph Lameter81819f02007-05-06 14:49:36 -07001545
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001546 flags &= gfp_allowed_mask;
1547
Mel Gormand0164ad2015-11-06 16:28:21 -08001548 if (gfpflags_allow_blocking(flags))
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001549 local_irq_enable();
1550
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001551 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001552
Pekka Enbergba522702009-06-24 21:59:51 +03001553 /*
1554 * Let the initial higher-order allocation fail under memory pressure
1555 * so we fall-back to the minimum order allocation.
1556 */
1557 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Mel Gormand0164ad2015-11-06 16:28:21 -08001558 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
Mel Gorman444eb2a42016-03-17 14:19:23 -07001559 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
Pekka Enbergba522702009-06-24 21:59:51 +03001560
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001561 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001562 if (unlikely(!page)) {
1563 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001564 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001565 /*
1566 * Allocation may have failed due to fragmentation.
1567 * Try a lower order alloc if possible
1568 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001569 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001570 if (unlikely(!page))
1571 goto out;
1572 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001573 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001574
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001575 if (kmemcheck_enabled &&
1576 !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001577 int pages = 1 << oo_order(oo);
1578
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001579 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001580
1581 /*
1582 * Objects from caches that have a constructor don't get
1583 * cleared when they're allocated, so we need to do it here.
1584 */
1585 if (s->ctor)
1586 kmemcheck_mark_uninitialized_pages(page, pages);
1587 else
1588 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001589 }
1590
Christoph Lameter834f3d12008-04-14 19:11:31 +03001591 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001592
Glauber Costa1f458cb2012-12-18 14:22:50 -08001593 order = compound_order(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001594 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001595 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001596 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001597 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001598
1599 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001600
1601 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001602 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001603
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001604 kasan_poison_slab(page);
1605
Thomas Garnier210e7a42016-07-26 15:21:59 -07001606 shuffle = shuffle_freelist(s, page);
1607
1608 if (!shuffle) {
1609 for_each_object_idx(p, idx, s, start, page->objects) {
1610 setup_object(s, page, p);
1611 if (likely(idx < page->objects))
1612 set_freepointer(s, p, p + s->size);
1613 else
1614 set_freepointer(s, p, NULL);
1615 }
1616 page->freelist = fixup_red_left(s, start);
Christoph Lameter81819f02007-05-06 14:49:36 -07001617 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001618
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001619 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001620 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001621
Christoph Lameter81819f02007-05-06 14:49:36 -07001622out:
Mel Gormand0164ad2015-11-06 16:28:21 -08001623 if (gfpflags_allow_blocking(flags))
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001624 local_irq_disable();
1625 if (!page)
1626 return NULL;
1627
1628 mod_zone_page_state(page_zone(page),
1629 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1630 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1631 1 << oo_order(oo));
1632
1633 inc_slabs_node(s, page_to_nid(page), page->objects);
1634
Christoph Lameter81819f02007-05-06 14:49:36 -07001635 return page;
1636}
1637
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001638static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1639{
1640 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07001641 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07001642 flags &= ~GFP_SLAB_BUG_MASK;
1643 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1644 invalid_mask, &invalid_mask, flags, &flags);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001645 }
1646
1647 return allocate_slab(s,
1648 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1649}
1650
Christoph Lameter81819f02007-05-06 14:49:36 -07001651static void __free_slab(struct kmem_cache *s, struct page *page)
1652{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001653 int order = compound_order(page);
1654 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001655
Laura Abbottbecfda62016-03-15 14:55:06 -07001656 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001657 void *p;
1658
1659 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001660 for_each_object(p, s, page_address(page),
1661 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001662 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001663 }
1664
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001665 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001666
Christoph Lameter81819f02007-05-06 14:49:36 -07001667 mod_zone_page_state(page_zone(page),
1668 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1669 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001670 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001671
Mel Gorman072bb0a2012-07-31 16:43:58 -07001672 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001673 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001674
Mel Gorman22b751c2013-02-22 16:34:59 -08001675 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001676 if (current->reclaim_state)
1677 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001678 memcg_uncharge_slab(page, order, s);
Se Wang (Patrick) Ohaf343622015-06-25 15:55:04 -07001679 kasan_alloc_pages(page, order);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001680 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001681}
1682
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001683#define need_reserve_slab_rcu \
1684 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1685
Christoph Lameter81819f02007-05-06 14:49:36 -07001686static void rcu_free_slab(struct rcu_head *h)
1687{
1688 struct page *page;
1689
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001690 if (need_reserve_slab_rcu)
1691 page = virt_to_head_page(h);
1692 else
1693 page = container_of((struct list_head *)h, struct page, lru);
1694
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001695 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001696}
1697
1698static void free_slab(struct kmem_cache *s, struct page *page)
1699{
1700 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001701 struct rcu_head *head;
1702
1703 if (need_reserve_slab_rcu) {
1704 int order = compound_order(page);
1705 int offset = (PAGE_SIZE << order) - s->reserved;
1706
1707 VM_BUG_ON(s->reserved != sizeof(*head));
1708 head = page_address(page) + offset;
1709 } else {
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001710 head = &page->rcu_head;
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001711 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001712
1713 call_rcu(head, rcu_free_slab);
1714 } else
1715 __free_slab(s, page);
1716}
1717
1718static void discard_slab(struct kmem_cache *s, struct page *page)
1719{
Christoph Lameter205ab992008-04-14 19:11:40 +03001720 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001721 free_slab(s, page);
1722}
1723
1724/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001725 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001726 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001727static inline void
1728__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001729{
Christoph Lametere95eed52007-05-06 14:49:44 -07001730 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001731 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001732 list_add_tail(&page->lru, &n->partial);
1733 else
1734 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001735}
1736
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001737static inline void add_partial(struct kmem_cache_node *n,
1738 struct page *page, int tail)
1739{
1740 lockdep_assert_held(&n->list_lock);
1741 __add_partial(n, page, tail);
1742}
1743
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001744static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001745 struct page *page)
1746{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001747 lockdep_assert_held(&n->list_lock);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08001748 list_del(&page->lru);
1749 n->nr_partial--;
Christoph Lameter62e346a2010-09-28 08:10:28 -05001750}
1751
Christoph Lameter81819f02007-05-06 14:49:36 -07001752/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001753 * Remove slab from the partial list, freeze it and
1754 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001755 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001756 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001757 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001758static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001759 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001760 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001761{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001762 void *freelist;
1763 unsigned long counters;
1764 struct page new;
1765
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001766 lockdep_assert_held(&n->list_lock);
1767
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001768 /*
1769 * Zap the freelist and set the frozen bit.
1770 * The old freelist is the list of objects for the
1771 * per cpu allocation list.
1772 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001773 freelist = page->freelist;
1774 counters = page->counters;
1775 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001776 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001777 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001778 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001779 new.freelist = NULL;
1780 } else {
1781 new.freelist = freelist;
1782 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001783
Dave Hansena0132ac2014-01-29 14:05:50 -08001784 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001785 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001786
Christoph Lameter7ced3712012-05-09 10:09:53 -05001787 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001788 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001789 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001790 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001791 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001792
1793 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001794 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001795 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001796}
1797
Joonsoo Kim633b0762013-01-21 17:01:25 +09001798static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001799static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001800
Christoph Lameter81819f02007-05-06 14:49:36 -07001801/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001802 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001803 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001804static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1805 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001806{
Christoph Lameter49e22582011-08-09 16:12:27 -05001807 struct page *page, *page2;
1808 void *object = NULL;
Alexey Dobriyanb052d042018-04-05 16:21:10 -07001809 unsigned int available = 0;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001810 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001811
1812 /*
1813 * Racy check. If we mistakenly see no partial slabs then we
1814 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001815 * partial slab and there is none available then get_partials()
1816 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001817 */
1818 if (!n || !n->nr_partial)
1819 return NULL;
1820
1821 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001822 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001823 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001824
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001825 if (!pfmemalloc_match(page, flags))
1826 continue;
1827
Joonsoo Kim633b0762013-01-21 17:01:25 +09001828 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001829 if (!t)
1830 break;
1831
Joonsoo Kim633b0762013-01-21 17:01:25 +09001832 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001833 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001834 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001835 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001836 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001837 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001838 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001839 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001840 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001841 if (!kmem_cache_has_cpu_partial(s)
1842 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001843 break;
1844
Christoph Lameter497b66f2011-08-09 16:12:26 -05001845 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001846 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001847 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001848}
1849
1850/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001851 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001852 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001853static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001854 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001855{
1856#ifdef CONFIG_NUMA
1857 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001858 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001859 struct zone *zone;
1860 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001861 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001862 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001863
1864 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001865 * The defrag ratio allows a configuration of the tradeoffs between
1866 * inter node defragmentation and node local allocations. A lower
1867 * defrag_ratio increases the tendency to do local allocations
1868 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001869 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001870 * If the defrag_ratio is set to 0 then kmalloc() always
1871 * returns node local objects. If the ratio is higher then kmalloc()
1872 * may return off node objects because partial slabs are obtained
1873 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001874 *
Li Peng43efd3e2016-05-19 17:10:43 -07001875 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1876 * (which makes defrag_ratio = 1000) then every (well almost)
1877 * allocation will first attempt to defrag slab caches on other nodes.
1878 * This means scanning over all nodes to look for partial slabs which
1879 * may be expensive if we do it every time we are trying to find a slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001880 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001881 */
Christoph Lameter98246012008-01-07 23:20:26 -08001882 if (!s->remote_node_defrag_ratio ||
1883 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001884 return NULL;
1885
Mel Gormancc9a6c82012-03-21 16:34:11 -07001886 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001887 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001888 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001889 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1890 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001891
Mel Gormancc9a6c82012-03-21 16:34:11 -07001892 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001893
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001894 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001895 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001896 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001897 if (object) {
1898 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001899 * Don't check read_mems_allowed_retry()
1900 * here - if mems_allowed was updated in
1901 * parallel, that was a harmless race
1902 * between allocation and the cpuset
1903 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001904 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001905 return object;
1906 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001907 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001908 }
Mel Gormand26914d2014-04-03 14:47:24 -07001909 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001910#endif
1911 return NULL;
1912}
1913
1914/*
1915 * Get a partial page, lock it and return it.
1916 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001917static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001918 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001919{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001920 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001921 int searchnode = node;
1922
1923 if (node == NUMA_NO_NODE)
1924 searchnode = numa_mem_id();
Christoph Lameter81819f02007-05-06 14:49:36 -07001925
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001926 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001927 if (object || node != NUMA_NO_NODE)
1928 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001929
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001930 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001931}
1932
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001933#ifdef CONFIG_PREEMPT
1934/*
1935 * Calculate the next globally unique transaction for disambiguiation
1936 * during cmpxchg. The transactions start with the cpu number and are then
1937 * incremented by CONFIG_NR_CPUS.
1938 */
1939#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1940#else
1941/*
1942 * No preemption supported therefore also no need to check for
1943 * different cpus.
1944 */
1945#define TID_STEP 1
1946#endif
1947
1948static inline unsigned long next_tid(unsigned long tid)
1949{
1950 return tid + TID_STEP;
1951}
1952
1953static inline unsigned int tid_to_cpu(unsigned long tid)
1954{
1955 return tid % TID_STEP;
1956}
1957
1958static inline unsigned long tid_to_event(unsigned long tid)
1959{
1960 return tid / TID_STEP;
1961}
1962
1963static inline unsigned int init_tid(int cpu)
1964{
1965 return cpu;
1966}
1967
1968static inline void note_cmpxchg_failure(const char *n,
1969 const struct kmem_cache *s, unsigned long tid)
1970{
1971#ifdef SLUB_DEBUG_CMPXCHG
1972 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1973
Fabian Frederickf9f58282014-06-04 16:06:34 -07001974 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001975
1976#ifdef CONFIG_PREEMPT
1977 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001978 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001979 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1980 else
1981#endif
1982 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001983 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001984 tid_to_event(tid), tid_to_event(actual_tid));
1985 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001986 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001987 actual_tid, tid, next_tid(tid));
1988#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001989 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001990}
1991
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001992static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001993{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001994 int cpu;
1995
1996 for_each_possible_cpu(cpu)
1997 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001998}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001999
2000/*
2001 * Remove the cpu slab
2002 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002003static void deactivate_slab(struct kmem_cache *s, struct page *page,
2004 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07002005{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002006 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002007 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2008 int lock = 0;
2009 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002010 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08002011 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002012 struct page new;
2013 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002014
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002015 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002016 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08002017 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07002018 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002019
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002020 /*
2021 * Stage one: Free all available per cpu objects back
2022 * to the page freelist while it is still frozen. Leave the
2023 * last one.
2024 *
2025 * There is no need to take the list->lock because the page
2026 * is still frozen.
2027 */
2028 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2029 void *prior;
2030 unsigned long counters;
2031
2032 do {
2033 prior = page->freelist;
2034 counters = page->counters;
2035 set_freepointer(s, freelist, prior);
2036 new.counters = counters;
2037 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08002038 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002039
Christoph Lameter1d071712011-07-14 12:49:12 -05002040 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002041 prior, counters,
2042 freelist, new.counters,
2043 "drain percpu freelist"));
2044
2045 freelist = nextfree;
2046 }
2047
2048 /*
2049 * Stage two: Ensure that the page is unfrozen while the
2050 * list presence reflects the actual number of objects
2051 * during unfreeze.
2052 *
2053 * We setup the list membership and then perform a cmpxchg
2054 * with the count. If there is a mismatch then the page
2055 * is not unfrozen but the page is on the wrong list.
2056 *
2057 * Then we restart the process which may have to remove
2058 * the page from the list that we just put it on again
2059 * because the number of objects in the slab may have
2060 * changed.
2061 */
2062redo:
2063
2064 old.freelist = page->freelist;
2065 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002066 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002067
2068 /* Determine target state of the slab */
2069 new.counters = old.counters;
2070 if (freelist) {
2071 new.inuse--;
2072 set_freepointer(s, freelist, old.freelist);
2073 new.freelist = freelist;
2074 } else
2075 new.freelist = old.freelist;
2076
2077 new.frozen = 0;
2078
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002079 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002080 m = M_FREE;
2081 else if (new.freelist) {
2082 m = M_PARTIAL;
2083 if (!lock) {
2084 lock = 1;
2085 /*
2086 * Taking the spinlock removes the possiblity
2087 * that acquire_slab() will see a slab page that
2088 * is frozen
2089 */
2090 spin_lock(&n->list_lock);
2091 }
2092 } else {
2093 m = M_FULL;
2094 if (kmem_cache_debug(s) && !lock) {
2095 lock = 1;
2096 /*
2097 * This also ensures that the scanning of full
2098 * slabs from diagnostic functions will not see
2099 * any frozen slabs.
2100 */
2101 spin_lock(&n->list_lock);
2102 }
2103 }
2104
2105 if (l != m) {
2106
2107 if (l == M_PARTIAL)
2108
2109 remove_partial(n, page);
2110
2111 else if (l == M_FULL)
2112
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002113 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002114
2115 if (m == M_PARTIAL) {
2116
2117 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08002118 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002119
2120 } else if (m == M_FULL) {
2121
2122 stat(s, DEACTIVATE_FULL);
2123 add_full(s, n, page);
2124
2125 }
2126 }
2127
2128 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05002129 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002130 old.freelist, old.counters,
2131 new.freelist, new.counters,
2132 "unfreezing slab"))
2133 goto redo;
2134
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002135 if (lock)
2136 spin_unlock(&n->list_lock);
2137
2138 if (m == M_FREE) {
2139 stat(s, DEACTIVATE_EMPTY);
2140 discard_slab(s, page);
2141 stat(s, FREE_SLAB);
2142 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002143}
2144
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002145/*
2146 * Unfreeze all the cpu partial slabs.
2147 *
Christoph Lameter59a09912012-11-28 16:23:00 +00002148 * This function must be called with interrupts disabled
2149 * for the cpu using c (or some other guarantee must be there
2150 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002151 */
Christoph Lameter59a09912012-11-28 16:23:00 +00002152static void unfreeze_partials(struct kmem_cache *s,
2153 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05002154{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002155#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09002156 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08002157 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002158
2159 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002160 struct page new;
2161 struct page old;
2162
2163 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002164
2165 n2 = get_node(s, page_to_nid(page));
2166 if (n != n2) {
2167 if (n)
2168 spin_unlock(&n->list_lock);
2169
2170 n = n2;
2171 spin_lock(&n->list_lock);
2172 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002173
2174 do {
2175
2176 old.freelist = page->freelist;
2177 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002178 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05002179
2180 new.counters = old.counters;
2181 new.freelist = old.freelist;
2182
2183 new.frozen = 0;
2184
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002185 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002186 old.freelist, old.counters,
2187 new.freelist, new.counters,
2188 "unfreezing slab"));
2189
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002190 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002191 page->next = discard_page;
2192 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002193 } else {
2194 add_partial(n, page, DEACTIVATE_TO_TAIL);
2195 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002196 }
2197 }
2198
2199 if (n)
2200 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002201
2202 while (discard_page) {
2203 page = discard_page;
2204 discard_page = discard_page->next;
2205
2206 stat(s, DEACTIVATE_EMPTY);
2207 discard_slab(s, page);
2208 stat(s, FREE_SLAB);
2209 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002210#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002211}
2212
2213/*
2214 * Put a page that was just frozen (in __slab_free) into a partial page
2215 * slot if available. This is done without interrupts disabled and without
2216 * preemption disabled. The cmpxchg is racy and may put the partial page
2217 * onto a random cpus partial slot.
2218 *
2219 * If we did not find a slot then simply move all the partials to the
2220 * per node partial list.
2221 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002222static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002223{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002224#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002225 struct page *oldpage;
2226 int pages;
2227 int pobjects;
2228
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002229 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002230 do {
2231 pages = 0;
2232 pobjects = 0;
2233 oldpage = this_cpu_read(s->cpu_slab->partial);
2234
2235 if (oldpage) {
2236 pobjects = oldpage->pobjects;
2237 pages = oldpage->pages;
2238 if (drain && pobjects > s->cpu_partial) {
2239 unsigned long flags;
2240 /*
2241 * partial array is full. Move the existing
2242 * set to the per node partial list.
2243 */
2244 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002245 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002246 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002247 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002248 pobjects = 0;
2249 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002250 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002251 }
2252 }
2253
2254 pages++;
2255 pobjects += page->objects - page->inuse;
2256
2257 page->pages = pages;
2258 page->pobjects = pobjects;
2259 page->next = oldpage;
2260
Chen Gangd0e0ac92013-07-15 09:05:29 +08002261 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2262 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002263 if (unlikely(!s->cpu_partial)) {
2264 unsigned long flags;
2265
2266 local_irq_save(flags);
2267 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2268 local_irq_restore(flags);
2269 }
2270 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002271#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002272}
2273
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002274static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002275{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002276 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002277 deactivate_slab(s, c->page, c->freelist);
2278
2279 c->tid = next_tid(c->tid);
2280 c->page = NULL;
2281 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002282}
2283
2284/*
2285 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002286 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002287 * Called from IPI handler with interrupts disabled.
2288 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002289static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002290{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002291 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002292
Christoph Lameter49e22582011-08-09 16:12:27 -05002293 if (likely(c)) {
2294 if (c->page)
2295 flush_slab(s, c);
2296
Christoph Lameter59a09912012-11-28 16:23:00 +00002297 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002298 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002299}
2300
2301static void flush_cpu_slab(void *d)
2302{
2303 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002304
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002305 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002306}
2307
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002308static bool has_cpu_slab(int cpu, void *info)
2309{
2310 struct kmem_cache *s = info;
2311 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2312
majianpeng02e1a9c2012-05-17 17:03:26 -07002313 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002314}
2315
Christoph Lameter81819f02007-05-06 14:49:36 -07002316static void flush_all(struct kmem_cache *s)
2317{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002318 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002319}
2320
2321/*
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02002322 * Use the cpu notifier to insure that the cpu slabs are flushed when
2323 * necessary.
2324 */
2325static int slub_cpu_dead(unsigned int cpu)
2326{
2327 struct kmem_cache *s;
2328 unsigned long flags;
2329
2330 mutex_lock(&slab_mutex);
2331 list_for_each_entry(s, &slab_caches, list) {
2332 local_irq_save(flags);
2333 __flush_cpu_slab(s, cpu);
2334 local_irq_restore(flags);
2335 }
2336 mutex_unlock(&slab_mutex);
2337 return 0;
2338}
2339
2340/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002341 * Check if the objects in a per cpu structure fit numa
2342 * locality expectations.
2343 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002344static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002345{
2346#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002347 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002348 return 0;
2349#endif
2350 return 1;
2351}
2352
David Rientjes9a02d692014-06-04 16:06:36 -07002353#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002354static int count_free(struct page *page)
2355{
2356 return page->objects - page->inuse;
2357}
2358
David Rientjes9a02d692014-06-04 16:06:36 -07002359static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2360{
2361 return atomic_long_read(&n->total_objects);
2362}
2363#endif /* CONFIG_SLUB_DEBUG */
2364
2365#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002366static unsigned long count_partial(struct kmem_cache_node *n,
2367 int (*get_count)(struct page *))
2368{
2369 unsigned long flags;
2370 unsigned long x = 0;
2371 struct page *page;
2372
2373 spin_lock_irqsave(&n->list_lock, flags);
2374 list_for_each_entry(page, &n->partial, lru)
2375 x += get_count(page);
2376 spin_unlock_irqrestore(&n->list_lock, flags);
2377 return x;
2378}
David Rientjes9a02d692014-06-04 16:06:36 -07002379#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002380
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002381static noinline void
2382slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2383{
David Rientjes9a02d692014-06-04 16:06:36 -07002384#ifdef CONFIG_SLUB_DEBUG
2385 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2386 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002387 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002388 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002389
David Rientjes9a02d692014-06-04 16:06:36 -07002390 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2391 return;
2392
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07002393 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2394 nid, gfpflags, &gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002395 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2396 s->name, s->object_size, s->size, oo_order(s->oo),
2397 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002398
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002399 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002400 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2401 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002402
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002403 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002404 unsigned long nr_slabs;
2405 unsigned long nr_objs;
2406 unsigned long nr_free;
2407
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002408 nr_free = count_partial(n, count_free);
2409 nr_slabs = node_nr_slabs(n);
2410 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002411
Fabian Frederickf9f58282014-06-04 16:06:34 -07002412 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002413 node, nr_slabs, nr_objs, nr_free);
2414 }
David Rientjes9a02d692014-06-04 16:06:36 -07002415#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002416}
2417
Christoph Lameter497b66f2011-08-09 16:12:26 -05002418static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2419 int node, struct kmem_cache_cpu **pc)
2420{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002421 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002422 struct kmem_cache_cpu *c = *pc;
2423 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002424
Christoph Lameter188fd062012-05-09 10:09:55 -05002425 freelist = get_partial(s, flags, node, c);
2426
2427 if (freelist)
2428 return freelist;
2429
2430 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002431 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002432 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002433 if (c->page)
2434 flush_slab(s, c);
2435
2436 /*
2437 * No other reference to the page yet so we can
2438 * muck around with it freely without cmpxchg
2439 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002440 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002441 page->freelist = NULL;
2442
2443 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002444 c->page = page;
2445 *pc = c;
2446 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002447 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002448
Christoph Lameter6faa6832012-05-09 10:09:51 -05002449 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002450}
2451
Mel Gorman072bb0a2012-07-31 16:43:58 -07002452static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2453{
2454 if (unlikely(PageSlabPfmemalloc(page)))
2455 return gfp_pfmemalloc_allowed(gfpflags);
2456
2457 return true;
2458}
2459
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002460/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002461 * Check the page->freelist of a page and either transfer the freelist to the
2462 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002463 *
2464 * The page is still frozen if the return value is not NULL.
2465 *
2466 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002467 *
2468 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002469 */
2470static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2471{
2472 struct page new;
2473 unsigned long counters;
2474 void *freelist;
2475
2476 do {
2477 freelist = page->freelist;
2478 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002479
Christoph Lameter213eeb92011-11-11 14:07:14 -06002480 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002481 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002482
2483 new.inuse = page->objects;
2484 new.frozen = freelist != NULL;
2485
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002486 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002487 freelist, counters,
2488 NULL, new.counters,
2489 "get_freelist"));
2490
2491 return freelist;
2492}
2493
2494/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002495 * Slow path. The lockless freelist is empty or we need to perform
2496 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002497 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002498 * Processing is still very fast if new objects have been freed to the
2499 * regular freelist. In that case we simply take over the regular freelist
2500 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002501 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002502 * If that is not working then we fall back to the partial lists. We take the
2503 * first element of the freelist as the object to allocate now and move the
2504 * rest of the freelist to the lockless freelist.
2505 *
2506 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002507 * we need to allocate a new slab. This is the slowest path since it involves
2508 * a call to the page allocator and the setup of a new slab.
Christoph Lametera380a3c2015-11-20 15:57:35 -08002509 *
2510 * Version of __slab_alloc to use when we know that interrupts are
2511 * already disabled (which is the case for bulk allocation).
Christoph Lameter81819f02007-05-06 14:49:36 -07002512 */
Christoph Lametera380a3c2015-11-20 15:57:35 -08002513static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002514 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002515{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002516 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002517 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002518
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002519 page = c->page;
Vlastimil Babkaa8a26222020-03-21 18:22:37 -07002520 if (!page) {
2521 /*
2522 * if the node is not online or has no normal memory, just
2523 * ignore the node constraint
2524 */
2525 if (unlikely(node != NUMA_NO_NODE &&
2526 !node_state(node, N_NORMAL_MEMORY)))
2527 node = NUMA_NO_NODE;
Christoph Lameter81819f02007-05-06 14:49:36 -07002528 goto new_slab;
Vlastimil Babkaa8a26222020-03-21 18:22:37 -07002529 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002530redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002531
Christoph Lameter57d437d2012-05-09 10:09:59 -05002532 if (unlikely(!node_match(page, node))) {
Vlastimil Babkaa8a26222020-03-21 18:22:37 -07002533 /*
2534 * same as above but node_match() being false already
2535 * implies node != NUMA_NO_NODE
2536 */
2537 if (!node_state(node, N_NORMAL_MEMORY)) {
2538 node = NUMA_NO_NODE;
2539 goto redo;
2540 } else {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002541 stat(s, ALLOC_NODE_MISMATCH);
2542 deactivate_slab(s, page, c->freelist);
2543 c->page = NULL;
2544 c->freelist = NULL;
2545 goto new_slab;
2546 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002547 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002548
Mel Gorman072bb0a2012-07-31 16:43:58 -07002549 /*
2550 * By rights, we should be searching for a slab page that was
2551 * PFMEMALLOC but right now, we are losing the pfmemalloc
2552 * information when the page leaves the per-cpu allocator
2553 */
2554 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2555 deactivate_slab(s, page, c->freelist);
2556 c->page = NULL;
2557 c->freelist = NULL;
2558 goto new_slab;
2559 }
2560
Eric Dumazet73736e02011-12-13 04:57:06 +01002561 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002562 freelist = c->freelist;
2563 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002564 goto load_freelist;
2565
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002566 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002567
Christoph Lameter6faa6832012-05-09 10:09:51 -05002568 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002569 c->page = NULL;
2570 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002571 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002572 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002573
Christoph Lameter81819f02007-05-06 14:49:36 -07002574 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002575
Christoph Lameter894b8782007-05-10 03:15:16 -07002576load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002577 /*
2578 * freelist is pointing to the list of objects to be used.
2579 * page is pointing to the page from which the objects are obtained.
2580 * That page must be frozen for per cpu allocations to work.
2581 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002582 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002583 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002584 c->tid = next_tid(c->tid);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002585 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002586
Christoph Lameter81819f02007-05-06 14:49:36 -07002587new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002588
Christoph Lameter49e22582011-08-09 16:12:27 -05002589 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002590 page = c->page = c->partial;
2591 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002592 stat(s, CPU_PARTIAL_ALLOC);
2593 c->freelist = NULL;
2594 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002595 }
2596
Christoph Lameter188fd062012-05-09 10:09:55 -05002597 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002598
Christoph Lameterf46974362012-05-09 10:09:54 -05002599 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002600 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002601 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002602 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002603
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002604 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002605 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002606 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002607
Christoph Lameter497b66f2011-08-09 16:12:26 -05002608 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002609 if (kmem_cache_debug(s) &&
2610 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002611 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002612
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002613 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002614 c->page = NULL;
2615 c->freelist = NULL;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002616 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002617}
2618
2619/*
Christoph Lametera380a3c2015-11-20 15:57:35 -08002620 * Another one that disabled interrupt and compensates for possible
2621 * cpu changes by refetching the per cpu area pointer.
2622 */
2623static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2624 unsigned long addr, struct kmem_cache_cpu *c)
2625{
2626 void *p;
2627 unsigned long flags;
2628
2629 local_irq_save(flags);
2630#ifdef CONFIG_PREEMPT
2631 /*
2632 * We may have been preempted and rescheduled on a different
2633 * cpu before disabling interrupts. Need to reload cpu area
2634 * pointer.
2635 */
2636 c = this_cpu_ptr(s->cpu_slab);
2637#endif
2638
2639 p = ___slab_alloc(s, gfpflags, node, addr, c);
2640 local_irq_restore(flags);
2641 return p;
2642}
2643
2644/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002645 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2646 * have the fastpath folded into their functions. So no function call
2647 * overhead for requests that can be satisfied on the fastpath.
2648 *
2649 * The fastpath works by first checking if the lockless freelist can be used.
2650 * If not then __slab_alloc is called for slow processing.
2651 *
2652 * Otherwise we can simply pick the next object from the lockless free list.
2653 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002654static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002655 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002656{
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002657 void *object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002658 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002659 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002660 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002661
Vladimir Davydov8135be52014-12-12 16:56:38 -08002662 s = slab_pre_alloc_hook(s, gfpflags);
2663 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002664 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002665redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002666 /*
2667 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2668 * enabled. We may switch back and forth between cpus while
2669 * reading from one cpu area. That does not matter as long
2670 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002671 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002672 * We should guarantee that tid and kmem_cache are retrieved on
2673 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2674 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002675 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002676 do {
2677 tid = this_cpu_read(s->cpu_slab->tid);
2678 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002679 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2680 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002681
2682 /*
2683 * Irqless object alloc/free algorithm used here depends on sequence
2684 * of fetching cpu_slab's data. tid should be fetched before anything
2685 * on c to guarantee that object and page associated with previous tid
2686 * won't be used with current tid. If we fetch tid first, object and
2687 * page could be one associated with next tid and our alloc/free
2688 * request will be failed. In this case, we will retry. So, no problem.
2689 */
2690 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002691
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002692 /*
2693 * The transaction ids are globally unique per cpu and per operation on
2694 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2695 * occurs on the right processor and that there was no operation on the
2696 * linked list in between.
2697 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002698
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002699 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002700 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002701 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002702 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002703 stat(s, ALLOC_SLOWPATH);
2704 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002705 void *next_object = get_freepointer_safe(s, object);
2706
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002707 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002708 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002709 * operation and if we are on the right processor.
2710 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002711 * The cmpxchg does the following atomically (without lock
2712 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002713 * 1. Relocate first pointer to the current per cpu area.
2714 * 2. Verify that tid and freelist have not been changed
2715 * 3. If they were not changed replace tid and freelist
2716 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002717 * Since this is without lock semantics the protection is only
2718 * against code executing on this cpu *not* from access by
2719 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002720 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002721 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002722 s->cpu_slab->freelist, s->cpu_slab->tid,
2723 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002724 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002725
2726 note_cmpxchg_failure("slab_alloc", s, tid);
2727 goto redo;
2728 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002729 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002730 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002731 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002732
Pekka Enberg74e21342009-11-25 20:14:48 +02002733 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002734 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002735
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002736 slab_post_alloc_hook(s, gfpflags, 1, &object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002737
Christoph Lameter894b8782007-05-10 03:15:16 -07002738 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002739}
2740
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002741static __always_inline void *slab_alloc(struct kmem_cache *s,
2742 gfp_t gfpflags, unsigned long addr)
2743{
2744 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2745}
2746
Christoph Lameter81819f02007-05-06 14:49:36 -07002747void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2748{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002749 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002750
Chen Gangd0e0ac92013-07-15 09:05:29 +08002751 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2752 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002753
2754 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002755}
2756EXPORT_SYMBOL(kmem_cache_alloc);
2757
Li Zefan0f24f122009-12-11 15:45:30 +08002758#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002759void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002760{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002761 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002762 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002763 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002764 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002765}
Richard Kennedy4a923792010-10-21 10:29:19 +01002766EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002767#endif
2768
Christoph Lameter81819f02007-05-06 14:49:36 -07002769#ifdef CONFIG_NUMA
2770void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2771{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002772 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002773
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002774 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002775 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002776
2777 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002778}
2779EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002780
Li Zefan0f24f122009-12-11 15:45:30 +08002781#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002782void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002783 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002784 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002785{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002786 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002787
2788 trace_kmalloc_node(_RET_IP_, ret,
2789 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002790
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002791 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002792 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002793}
Richard Kennedy4a923792010-10-21 10:29:19 +01002794EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002795#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002796#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002797
Christoph Lameter81819f02007-05-06 14:49:36 -07002798/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002799 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002800 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002801 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002802 * So we still attempt to reduce cache line usage. Just take the slab
2803 * lock and free the item. If there is no additional partial page
2804 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002805 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002806static void __slab_free(struct kmem_cache *s, struct page *page,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002807 void *head, void *tail, int cnt,
2808 unsigned long addr)
2809
Christoph Lameter81819f02007-05-06 14:49:36 -07002810{
2811 void *prior;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002812 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002813 struct page new;
2814 unsigned long counters;
2815 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002816 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002817
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002818 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002819
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002820 if (kmem_cache_debug(s) &&
Laura Abbott282acb42016-03-15 14:54:59 -07002821 !free_debug_processing(s, page, head, tail, cnt, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002822 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002823
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002824 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002825 if (unlikely(n)) {
2826 spin_unlock_irqrestore(&n->list_lock, flags);
2827 n = NULL;
2828 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002829 prior = page->freelist;
2830 counters = page->counters;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002831 set_freepointer(s, tail, prior);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002832 new.counters = counters;
2833 was_frozen = new.frozen;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002834 new.inuse -= cnt;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002835 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002836
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002837 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002838
2839 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002840 * Slab was on no list before and will be
2841 * partially empty
2842 * We can defer the list move and instead
2843 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002844 */
2845 new.frozen = 1;
2846
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002847 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002848
LQYMGTb455def2014-12-10 15:42:13 -08002849 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002850 /*
2851 * Speculatively acquire the list_lock.
2852 * If the cmpxchg does not succeed then we may
2853 * drop the list_lock without any processing.
2854 *
2855 * Otherwise the list_lock will synchronize with
2856 * other processors updating the list of slabs.
2857 */
2858 spin_lock_irqsave(&n->list_lock, flags);
2859
2860 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002861 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002862
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002863 } while (!cmpxchg_double_slab(s, page,
2864 prior, counters,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002865 head, new.counters,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002866 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002867
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002868 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002869
2870 /*
2871 * If we just froze the page then put it onto the
2872 * per cpu partial list.
2873 */
Alex Shi8028dce2012-02-03 23:34:56 +08002874 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002875 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002876 stat(s, CPU_PARTIAL_FREE);
2877 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002878 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002879 * The list lock was not taken therefore no list
2880 * activity can be necessary.
2881 */
LQYMGTb455def2014-12-10 15:42:13 -08002882 if (was_frozen)
2883 stat(s, FREE_FROZEN);
2884 return;
2885 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002886
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002887 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002888 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002889
Joonsoo Kim837d6782012-08-16 00:02:40 +09002890 /*
2891 * Objects left in the slab. If it was not on the partial list before
2892 * then add it.
2893 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002894 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2895 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002896 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002897 add_partial(n, page, DEACTIVATE_TO_TAIL);
2898 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002899 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002900 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002901 return;
2902
2903slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002904 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002905 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002906 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002907 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002908 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002909 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002910 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002911 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002912 remove_full(s, n, page);
2913 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002914
Christoph Lameter80f08c12011-06-01 12:25:55 -05002915 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002916 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002917 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002918}
2919
Christoph Lameter894b8782007-05-10 03:15:16 -07002920/*
2921 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2922 * can perform fastpath freeing without additional function calls.
2923 *
2924 * The fastpath is only possible if we are freeing to the current cpu slab
2925 * of this processor. This typically the case if we have just allocated
2926 * the item before.
2927 *
2928 * If fastpath is not possible then fall back to __slab_free where we deal
2929 * with all sorts of special processing.
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002930 *
2931 * Bulk free of a freelist with several objects (all pointing to the
2932 * same page) possible by specifying head and tail ptr, plus objects
2933 * count (cnt). Bulk free indicated by tail pointer being set.
Christoph Lameter894b8782007-05-10 03:15:16 -07002934 */
Alexander Potapenko80a92012016-07-28 15:49:07 -07002935static __always_inline void do_slab_free(struct kmem_cache *s,
2936 struct page *page, void *head, void *tail,
2937 int cnt, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002938{
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002939 void *tail_obj = tail ? : head;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002940 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002941 unsigned long tid;
Christoph Lametera24c5a02011-03-15 12:45:21 -05002942redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002943 /*
2944 * Determine the currently cpus per cpu slab.
2945 * The cpu may change afterward. However that does not matter since
2946 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07002947 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002948 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002949 do {
2950 tid = this_cpu_read(s->cpu_slab->tid);
2951 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002952 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2953 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002954
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002955 /* Same with comment on barrier() in slab_alloc_node() */
2956 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002957
Christoph Lameter442b06b2011-05-17 16:29:31 -05002958 if (likely(page == c->page)) {
Linus Torvaldse6635422020-03-17 11:04:09 -07002959 void **freelist = READ_ONCE(c->freelist);
2960
2961 set_freepointer(s, tail_obj, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002962
Christoph Lameter933393f2011-12-22 11:58:51 -06002963 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002964 s->cpu_slab->freelist, s->cpu_slab->tid,
Linus Torvaldse6635422020-03-17 11:04:09 -07002965 freelist, tid,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002966 head, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002967
2968 note_cmpxchg_failure("slab_free", s, tid);
2969 goto redo;
2970 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002971 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002972 } else
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002973 __slab_free(s, page, head, tail_obj, cnt, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002974
Christoph Lameter894b8782007-05-10 03:15:16 -07002975}
2976
Alexander Potapenko80a92012016-07-28 15:49:07 -07002977static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2978 void *head, void *tail, int cnt,
2979 unsigned long addr)
2980{
2981 slab_free_freelist_hook(s, head, tail);
2982 /*
2983 * slab_free_freelist_hook() could have put the items into quarantine.
2984 * If so, no need to free them.
2985 */
2986 if (s->flags & SLAB_KASAN && !(s->flags & SLAB_DESTROY_BY_RCU))
2987 return;
2988 do_slab_free(s, page, head, tail, cnt, addr);
2989}
2990
2991#ifdef CONFIG_KASAN
2992void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2993{
2994 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2995}
2996#endif
2997
Christoph Lameter81819f02007-05-06 14:49:36 -07002998void kmem_cache_free(struct kmem_cache *s, void *x)
2999{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003000 s = cache_from_obj(s, x);
3001 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00003002 return;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003003 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003004 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07003005}
3006EXPORT_SYMBOL(kmem_cache_free);
3007
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003008struct detached_freelist {
3009 struct page *page;
3010 void *tail;
3011 void *freelist;
3012 int cnt;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003013 struct kmem_cache *s;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003014};
3015
3016/*
3017 * This function progressively scans the array with free objects (with
3018 * a limited look ahead) and extract objects belonging to the same
3019 * page. It builds a detached freelist directly within the given
3020 * page/objects. This can happen without any need for
3021 * synchronization, because the objects are owned by running process.
3022 * The freelist is build up as a single linked list in the objects.
3023 * The idea is, that this detached freelist can then be bulk
3024 * transferred to the real freelist(s), but only requiring a single
3025 * synchronization primitive. Look ahead in the array is limited due
3026 * to performance reasons.
3027 */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003028static inline
3029int build_detached_freelist(struct kmem_cache *s, size_t size,
3030 void **p, struct detached_freelist *df)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003031{
3032 size_t first_skipped_index = 0;
3033 int lookahead = 3;
3034 void *object;
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003035 struct page *page;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003036
3037 /* Always re-init detached_freelist */
3038 df->page = NULL;
3039
3040 do {
3041 object = p[--size];
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003042 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003043 } while (!object && size);
3044
3045 if (!object)
3046 return 0;
3047
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003048 page = virt_to_head_page(object);
3049 if (!s) {
3050 /* Handle kalloc'ed objects */
3051 if (unlikely(!PageSlab(page))) {
3052 BUG_ON(!PageCompound(page));
3053 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003054 __free_pages(page, compound_order(page));
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003055 p[size] = NULL; /* mark object processed */
3056 return size;
3057 }
3058 /* Derive kmem_cache from object */
3059 df->s = page->slab_cache;
3060 } else {
3061 df->s = cache_from_obj(s, object); /* Support for memcg */
3062 }
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003063
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003064 /* Start new detached freelist */
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003065 df->page = page;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003066 set_freepointer(df->s, object, NULL);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003067 df->tail = object;
3068 df->freelist = object;
3069 p[size] = NULL; /* mark object processed */
3070 df->cnt = 1;
3071
3072 while (size) {
3073 object = p[--size];
3074 if (!object)
3075 continue; /* Skip processed objects */
3076
3077 /* df->page is always set at this point */
3078 if (df->page == virt_to_head_page(object)) {
3079 /* Opportunity build freelist */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003080 set_freepointer(df->s, object, df->freelist);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003081 df->freelist = object;
3082 df->cnt++;
3083 p[size] = NULL; /* mark object processed */
3084
3085 continue;
3086 }
3087
3088 /* Limit look ahead search */
3089 if (!--lookahead)
3090 break;
3091
3092 if (!first_skipped_index)
3093 first_skipped_index = size + 1;
3094 }
3095
3096 return first_skipped_index;
3097}
3098
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003099/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003100void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003101{
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003102 if (WARN_ON(!size))
3103 return;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003104
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003105 do {
3106 struct detached_freelist df;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003107
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003108 size = build_detached_freelist(s, size, p, &df);
3109 if (unlikely(!df.page))
3110 continue;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003111
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003112 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003113 } while (likely(size));
Christoph Lameter484748f2015-09-04 15:45:34 -07003114}
3115EXPORT_SYMBOL(kmem_cache_free_bulk);
3116
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003117/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003118int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3119 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003120{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003121 struct kmem_cache_cpu *c;
3122 int i;
3123
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003124 /* memcg and kmem_cache debug support */
3125 s = slab_pre_alloc_hook(s, flags);
3126 if (unlikely(!s))
3127 return false;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003128 /*
3129 * Drain objects in the per cpu slab, while disabling local
3130 * IRQs, which protects against PREEMPT and interrupts
3131 * handlers invoking normal fastpath.
3132 */
3133 local_irq_disable();
3134 c = this_cpu_ptr(s->cpu_slab);
3135
3136 for (i = 0; i < size; i++) {
3137 void *object = c->freelist;
3138
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003139 if (unlikely(!object)) {
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003140 /*
Jann Hornff58bb32020-03-17 01:28:45 +01003141 * We may have removed an object from c->freelist using
3142 * the fastpath in the previous iteration; in that case,
3143 * c->tid has not been bumped yet.
3144 * Since ___slab_alloc() may reenable interrupts while
3145 * allocating memory, we should bump c->tid now.
3146 */
3147 c->tid = next_tid(c->tid);
3148
3149 /*
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003150 * Invoking slow path likely have side-effect
3151 * of re-populating per CPU c->freelist
3152 */
Christoph Lameter87098372015-11-20 15:57:38 -08003153 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003154 _RET_IP_, c);
Christoph Lameter87098372015-11-20 15:57:38 -08003155 if (unlikely(!p[i]))
3156 goto error;
3157
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003158 c = this_cpu_ptr(s->cpu_slab);
3159 continue; /* goto for-loop */
3160 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003161 c->freelist = get_freepointer(s, object);
3162 p[i] = object;
3163 }
3164 c->tid = next_tid(c->tid);
3165 local_irq_enable();
3166
3167 /* Clear memory outside IRQ disabled fastpath loop */
3168 if (unlikely(flags & __GFP_ZERO)) {
3169 int j;
3170
3171 for (j = 0; j < i; j++)
3172 memset(p[j], 0, s->object_size);
3173 }
3174
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003175 /* memcg and kmem_cache debug support */
3176 slab_post_alloc_hook(s, flags, size, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003177 return i;
Christoph Lameter87098372015-11-20 15:57:38 -08003178error:
Christoph Lameter87098372015-11-20 15:57:38 -08003179 local_irq_enable();
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003180 slab_post_alloc_hook(s, flags, i, p);
3181 __kmem_cache_free_bulk(s, i, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003182 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003183}
3184EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3185
3186
Christoph Lameter81819f02007-05-06 14:49:36 -07003187/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003188 * Object placement in a slab is made very easy because we always start at
3189 * offset 0. If we tune the size of the object to the alignment then we can
3190 * get the required alignment by putting one properly sized object after
3191 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07003192 *
3193 * Notice that the allocation order determines the sizes of the per cpu
3194 * caches. Each processor has always one slab available for allocations.
3195 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07003196 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07003197 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07003198 */
3199
3200/*
3201 * Mininum / Maximum order of slab pages. This influences locking overhead
3202 * and slab fragmentation. A higher order reduces the number of partial slabs
3203 * and increases the number of allocations possible without having to
3204 * take the list_lock.
3205 */
3206static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03003207static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003208static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07003209
3210/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003211 * Calculate the order of allocation given an slab object size.
3212 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003213 * The order of allocation has significant impact on performance and other
3214 * system components. Generally order 0 allocations should be preferred since
3215 * order 0 does not cause fragmentation in the page allocator. Larger objects
3216 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003217 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07003218 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07003219 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003220 * In order to reach satisfactory performance we must ensure that a minimum
3221 * number of objects is in one slab. Otherwise we may generate too much
3222 * activity on the partial lists which requires taking the list_lock. This is
3223 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07003224 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003225 * slub_max_order specifies the order where we begin to stop considering the
3226 * number of objects in a slab as critical. If we reach slub_max_order then
3227 * we try to keep the page order as low as possible. So we accept more waste
3228 * of space in favor of a small page order.
3229 *
3230 * Higher order allocations also allow the placement of more objects in a
3231 * slab and thereby reduce object handling overhead. If the user has
3232 * requested a higher mininum order then we start with that one instead of
3233 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07003234 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003235static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003236 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07003237{
3238 int order;
3239 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07003240 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003241
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003242 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04003243 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03003244
Wei Yang9f835702015-11-05 18:45:51 -08003245 for (order = max(min_order, get_order(min_objects * size + reserved));
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003246 order <= max_order; order++) {
3247
Christoph Lameter81819f02007-05-06 14:49:36 -07003248 unsigned long slab_size = PAGE_SIZE << order;
3249
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003250 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003251
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003252 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003253 break;
Christoph Lameter81819f02007-05-06 14:49:36 -07003254 }
Christoph Lameter672bba32007-05-09 02:32:39 -07003255
Christoph Lameter81819f02007-05-06 14:49:36 -07003256 return order;
3257}
3258
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003259static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003260{
3261 int order;
3262 int min_objects;
3263 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003264 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003265
3266 /*
3267 * Attempt to find best configuration for a slab. This
3268 * works by first attempting to generate a layout with
3269 * the best configuration and backing off gradually.
3270 *
Wei Yang422ff4d2015-11-05 18:45:46 -08003271 * First we increase the acceptable waste in a slab. Then
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003272 * we reduce the minimum objects required in a slab.
3273 */
3274 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003275 if (!min_objects)
3276 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003277 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003278 min_objects = min(min_objects, max_objects);
3279
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003280 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003281 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003282 while (fraction >= 4) {
3283 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003284 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003285 if (order <= slub_max_order)
3286 return order;
3287 fraction /= 2;
3288 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03003289 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003290 }
3291
3292 /*
3293 * We were unable to place multiple objects in a slab. Now
3294 * lets see if we can place a single object there.
3295 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003296 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003297 if (order <= slub_max_order)
3298 return order;
3299
3300 /*
3301 * Doh this slab cannot be placed using slub_max_order.
3302 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003303 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03003304 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003305 return order;
3306 return -ENOSYS;
3307}
3308
Pekka Enberg5595cff2008-08-05 09:28:47 +03003309static void
Joonsoo Kim40534972012-05-11 00:50:47 +09003310init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003311{
3312 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003313 spin_lock_init(&n->list_lock);
3314 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003315#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003316 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07003317 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07003318 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003319#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003320}
3321
Christoph Lameter55136592010-08-20 12:37:13 -05003322static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003323{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05003324 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003325 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003326
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003327 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003328 * Must align to double word boundary for the double cmpxchg
3329 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003330 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003331 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3332 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003333
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003334 if (!s->cpu_slab)
3335 return 0;
3336
3337 init_kmem_cache_cpus(s);
3338
3339 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003340}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003341
Christoph Lameter51df1142010-08-20 12:37:15 -05003342static struct kmem_cache *kmem_cache_node;
3343
Christoph Lameter81819f02007-05-06 14:49:36 -07003344/*
3345 * No kmalloc_node yet so do it by hand. We know that this is the first
3346 * slab on the node for this slabcache. There are no concurrent accesses
3347 * possible.
3348 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003349 * Note that this function only works on the kmem_cache_node
3350 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003351 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003352 */
Christoph Lameter55136592010-08-20 12:37:13 -05003353static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003354{
3355 struct page *page;
3356 struct kmem_cache_node *n;
3357
Christoph Lameter51df1142010-08-20 12:37:15 -05003358 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003359
Christoph Lameter51df1142010-08-20 12:37:15 -05003360 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003361
3362 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003363 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003364 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3365 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003366 }
3367
Christoph Lameter81819f02007-05-06 14:49:36 -07003368 n = page->freelist;
3369 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003370 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05003371 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05003372 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003373 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07003374#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003375 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003376 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003377#endif
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003378 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3379 GFP_KERNEL);
Joonsoo Kim40534972012-05-11 00:50:47 +09003380 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003381 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003382
Dave Hansen67b6c902014-01-24 07:20:23 -08003383 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003384 * No locks need to be taken here as it has just been
3385 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003386 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003387 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003388}
3389
3390static void free_kmem_cache_nodes(struct kmem_cache *s)
3391{
3392 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003393 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003394
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003395 for_each_kmem_cache_node(s, node, n) {
3396 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003397 s->node[node] = NULL;
3398 }
3399}
3400
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003401void __kmem_cache_release(struct kmem_cache *s)
3402{
Thomas Garnier210e7a42016-07-26 15:21:59 -07003403 cache_random_seq_destroy(s);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003404 free_percpu(s->cpu_slab);
3405 free_kmem_cache_nodes(s);
3406}
3407
Christoph Lameter55136592010-08-20 12:37:13 -05003408static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003409{
3410 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003411
Christoph Lameterf64dc582007-10-16 01:25:33 -07003412 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003413 struct kmem_cache_node *n;
3414
Alexander Duyck73367bd2010-05-21 14:41:35 -07003415 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003416 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003417 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003418 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003419 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003420 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003421
3422 if (!n) {
3423 free_kmem_cache_nodes(s);
3424 return 0;
3425 }
3426
Christoph Lameter81819f02007-05-06 14:49:36 -07003427 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09003428 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003429 }
3430 return 1;
3431}
Christoph Lameter81819f02007-05-06 14:49:36 -07003432
David Rientjesc0bdb232009-02-25 09:16:35 +02003433static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003434{
3435 if (min < MIN_PARTIAL)
3436 min = MIN_PARTIAL;
3437 else if (min > MAX_PARTIAL)
3438 min = MAX_PARTIAL;
3439 s->min_partial = min;
3440}
3441
Christoph Lameter81819f02007-05-06 14:49:36 -07003442/*
3443 * calculate_sizes() determines the order and the distribution of data within
3444 * a slab object.
3445 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003446static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003447{
3448 unsigned long flags = s->flags;
Alexander Potapenko80a92012016-07-28 15:49:07 -07003449 size_t size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003450 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003451
3452 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003453 * Round up object size to the next word boundary. We can only
3454 * place the free pointer at word boundaries and this determines
3455 * the possible location of the free pointer.
3456 */
3457 size = ALIGN(size, sizeof(void *));
3458
3459#ifdef CONFIG_SLUB_DEBUG
3460 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003461 * Determine if we can poison the object itself. If the user of
3462 * the slab may touch the object after free or before allocation
3463 * then we should never poison the object itself.
3464 */
3465 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003466 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003467 s->flags |= __OBJECT_POISON;
3468 else
3469 s->flags &= ~__OBJECT_POISON;
3470
Christoph Lameter81819f02007-05-06 14:49:36 -07003471
3472 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003473 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003474 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003475 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003476 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003477 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003478 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003479#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003480
3481 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003482 * With that we have determined the number of bytes in actual use
3483 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003484 */
3485 s->inuse = size;
3486
3487 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003488 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003489 /*
3490 * Relocate free pointer after the object if it is not
3491 * permitted to overwrite the first word of the object on
3492 * kmem_cache_free.
3493 *
3494 * This is the case if we do RCU, have a constructor or
3495 * destructor or are poisoning the objects.
3496 */
3497 s->offset = size;
3498 size += sizeof(void *);
3499 }
3500
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003501#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003502 if (flags & SLAB_STORE_USER)
3503 /*
3504 * Need to store information about allocs and frees after
3505 * the object.
3506 */
3507 size += 2 * sizeof(struct track);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003508#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003509
Alexander Potapenko80a92012016-07-28 15:49:07 -07003510 kasan_cache_create(s, &size, &s->flags);
3511#ifdef CONFIG_SLUB_DEBUG
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003512 if (flags & SLAB_RED_ZONE) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003513 /*
3514 * Add some empty padding so that we can catch
3515 * overwrites from earlier objects rather than let
3516 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003517 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003518 * of the object.
3519 */
3520 size += sizeof(void *);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003521
3522 s->red_left_pad = sizeof(void *);
3523 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3524 size += s->red_left_pad;
3525 }
Christoph Lameter41ecc552007-05-09 02:32:44 -07003526#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003527
Christoph Lameter81819f02007-05-06 14:49:36 -07003528 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003529 * SLUB stores one object immediately after another beginning from
3530 * offset 0. In order to align the objects we have to simply size
3531 * each object to conform to the alignment.
3532 */
Christoph Lameter45906852012-11-28 16:23:16 +00003533 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003534 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003535 if (forced_order >= 0)
3536 order = forced_order;
3537 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003538 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003539
Christoph Lameter834f3d12008-04-14 19:11:31 +03003540 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003541 return 0;
3542
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003543 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003544 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003545 s->allocflags |= __GFP_COMP;
3546
3547 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003548 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003549
3550 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3551 s->allocflags |= __GFP_RECLAIMABLE;
3552
Christoph Lameter81819f02007-05-06 14:49:36 -07003553 /*
3554 * Determine the number of objects per slab
3555 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003556 s->oo = oo_make(order, size, s->reserved);
3557 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003558 if (oo_objects(s->oo) > oo_objects(s->max))
3559 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003560
Christoph Lameter834f3d12008-04-14 19:11:31 +03003561 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003562}
3563
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003564static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003565{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003566 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003567 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003568
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003569 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3570 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003571
Christoph Lameter06b285d2008-04-14 19:11:41 +03003572 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003573 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003574 if (disable_higher_order_debug) {
3575 /*
3576 * Disable debugging flags that store metadata if the min slab
3577 * order increased.
3578 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003579 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003580 s->flags &= ~DEBUG_METADATA_FLAGS;
3581 s->offset = 0;
3582 if (!calculate_sizes(s, -1))
3583 goto error;
3584 }
3585 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003586
Heiko Carstens25654092012-01-12 17:17:33 -08003587#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3588 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Laura Abbott149daaf2016-03-15 14:55:09 -07003589 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003590 /* Enable fast mode */
3591 s->flags |= __CMPXCHG_DOUBLE;
3592#endif
3593
David Rientjes3b89d7d2009-02-22 17:40:07 -08003594 /*
3595 * The larger the object size is, the more pages we want on the partial
3596 * list to avoid pounding the page allocator excessively.
3597 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003598 set_min_partial(s, ilog2(s->size) / 2);
3599
3600 /*
3601 * cpu_partial determined the maximum number of objects kept in the
3602 * per cpu partial lists of a processor.
3603 *
3604 * Per cpu partial lists mainly contain slabs that just have one
3605 * object freed. If they are used for allocation then they can be
3606 * filled up again with minimal effort. The slab will never hit the
3607 * per node partial lists and therefore no locking will be required.
3608 *
3609 * This setting also determines
3610 *
3611 * A) The number of objects from per cpu partial slabs dumped to the
3612 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003613 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003614 * per node list when we run out of per cpu objects. We only fetch
3615 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003616 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003617 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003618 s->cpu_partial = 0;
3619 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003620 s->cpu_partial = 2;
3621 else if (s->size >= 1024)
3622 s->cpu_partial = 6;
3623 else if (s->size >= 256)
3624 s->cpu_partial = 13;
3625 else
3626 s->cpu_partial = 30;
3627
Christoph Lameter81819f02007-05-06 14:49:36 -07003628#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003629 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003630#endif
Thomas Garnier210e7a42016-07-26 15:21:59 -07003631
3632 /* Initialize the pre-computed randomized freelist if slab is up */
3633 if (slab_state >= UP) {
3634 if (init_cache_random_seq(s))
3635 goto error;
3636 }
3637
Christoph Lameter55136592010-08-20 12:37:13 -05003638 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003639 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003640
Christoph Lameter55136592010-08-20 12:37:13 -05003641 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003642 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003643
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003644 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003645error:
3646 if (flags & SLAB_PANIC)
Joe Perches756a025f02016-03-17 14:19:47 -07003647 panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n",
3648 s->name, (unsigned long)s->size, s->size,
3649 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003650 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003651}
Christoph Lameter81819f02007-05-06 14:49:36 -07003652
Christoph Lameter33b12c32008-04-25 12:22:43 -07003653static void list_slab_objects(struct kmem_cache *s, struct page *page,
3654 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003655{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003656#ifdef CONFIG_SLUB_DEBUG
3657 void *addr = page_address(page);
3658 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003659 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3660 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003661 if (!map)
3662 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003663 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003664 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003665
Christoph Lameter5f80b132011-04-15 14:48:13 -05003666 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003667 for_each_object(p, s, addr, page->objects) {
3668
3669 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003670 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003671 print_tracking(s, p);
3672 }
3673 }
3674 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003675 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003676#endif
3677}
3678
Christoph Lameter81819f02007-05-06 14:49:36 -07003679/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003680 * Attempt to free all partial slabs on a node.
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003681 * This is called from __kmem_cache_shutdown(). We must take list_lock
3682 * because sysfs file might still access partial list after the shutdowning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003683 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003684static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003685{
Chris Wilson60398922016-08-10 16:27:58 -07003686 LIST_HEAD(discard);
Christoph Lameter81819f02007-05-06 14:49:36 -07003687 struct page *page, *h;
3688
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003689 BUG_ON(irqs_disabled());
3690 spin_lock_irq(&n->list_lock);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003691 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003692 if (!page->inuse) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003693 remove_partial(n, page);
Chris Wilson60398922016-08-10 16:27:58 -07003694 list_add(&page->lru, &discard);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003695 } else {
3696 list_slab_objects(s, page,
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003697 "Objects remaining in %s on __kmem_cache_shutdown()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003698 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003699 }
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003700 spin_unlock_irq(&n->list_lock);
Chris Wilson60398922016-08-10 16:27:58 -07003701
3702 list_for_each_entry_safe(page, h, &discard, lru)
3703 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003704}
3705
3706/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003707 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003708 */
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003709int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003710{
3711 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003712 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003713
3714 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003715 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003716 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003717 free_partial(s, n);
3718 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003719 return 1;
3720 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003721 return 0;
3722}
3723
Christoph Lameter81819f02007-05-06 14:49:36 -07003724/********************************************************************
3725 * Kmalloc subsystem
3726 *******************************************************************/
3727
Christoph Lameter81819f02007-05-06 14:49:36 -07003728static int __init setup_slub_min_order(char *str)
3729{
Pekka Enberg06428782008-01-07 23:20:27 -08003730 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003731
3732 return 1;
3733}
3734
3735__setup("slub_min_order=", setup_slub_min_order);
3736
3737static int __init setup_slub_max_order(char *str)
3738{
Pekka Enberg06428782008-01-07 23:20:27 -08003739 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003740 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003741
3742 return 1;
3743}
3744
3745__setup("slub_max_order=", setup_slub_max_order);
3746
3747static int __init setup_slub_min_objects(char *str)
3748{
Pekka Enberg06428782008-01-07 23:20:27 -08003749 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003750
3751 return 1;
3752}
3753
3754__setup("slub_min_objects=", setup_slub_min_objects);
3755
Christoph Lameter81819f02007-05-06 14:49:36 -07003756void *__kmalloc(size_t size, gfp_t flags)
3757{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003758 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003759 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003760
Christoph Lameter95a05b42013-01-10 19:14:19 +00003761 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003762 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003763
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003764 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003765
3766 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003767 return s;
3768
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003769 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003770
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003771 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003772
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003773 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003774
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003775 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003776}
3777EXPORT_SYMBOL(__kmalloc);
3778
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003779#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003780static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3781{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003782 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003783 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003784
Vladimir Davydov52383432014-06-04 16:06:39 -07003785 flags |= __GFP_COMP | __GFP_NOTRACK;
Vladimir Davydov49491482016-07-26 15:24:24 -07003786 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003787 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003788 ptr = page_address(page);
3789
Roman Bobnievd56791b2013-10-08 15:58:57 -07003790 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003791 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003792}
3793
Christoph Lameter81819f02007-05-06 14:49:36 -07003794void *__kmalloc_node(size_t size, gfp_t flags, int node)
3795{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003796 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003797 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003798
Christoph Lameter95a05b42013-01-10 19:14:19 +00003799 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003800 ret = kmalloc_large_node(size, flags, node);
3801
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003802 trace_kmalloc_node(_RET_IP_, ret,
3803 size, PAGE_SIZE << get_order(size),
3804 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003805
3806 return ret;
3807 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003808
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003809 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003810
3811 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003812 return s;
3813
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003814 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003815
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003816 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003817
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003818 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003819
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003820 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003821}
3822EXPORT_SYMBOL(__kmalloc_node);
3823#endif
3824
Kees Cooked18adc2016-06-23 15:24:05 -07003825#ifdef CONFIG_HARDENED_USERCOPY
3826/*
3827 * Rejects objects that are incorrectly sized.
3828 *
3829 * Returns NULL if check passes, otherwise const char * to name of cache
3830 * to indicate an error.
3831 */
3832const char *__check_heap_object(const void *ptr, unsigned long n,
3833 struct page *page)
3834{
3835 struct kmem_cache *s;
3836 unsigned long offset;
3837 size_t object_size;
3838
3839 /* Find object and usable object size. */
3840 s = page->slab_cache;
3841 object_size = slab_ksize(s);
3842
3843 /* Reject impossible pointers. */
3844 if (ptr < page_address(page))
3845 return s->name;
3846
3847 /* Find offset within object. */
3848 offset = (ptr - page_address(page)) % s->size;
3849
3850 /* Adjust for redzone and reject if within the redzone. */
3851 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3852 if (offset < s->red_left_pad)
3853 return s->name;
3854 offset -= s->red_left_pad;
3855 }
3856
3857 /* Allow address range falling entirely within object size. */
3858 if (offset <= object_size && n <= object_size - offset)
3859 return NULL;
3860
3861 return s->name;
3862}
3863#endif /* CONFIG_HARDENED_USERCOPY */
3864
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003865static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003866{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003867 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003868
Christoph Lameteref8b4522007-10-16 01:24:46 -07003869 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003870 return 0;
3871
Vegard Nossum294a80a2007-12-04 23:45:30 -08003872 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003873
Pekka Enberg76994412008-05-22 19:22:25 +03003874 if (unlikely(!PageSlab(page))) {
3875 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003876 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003877 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003878
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003879 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003880}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003881
3882size_t ksize(const void *object)
3883{
3884 size_t size = __ksize(object);
3885 /* We assume that ksize callers could use whole allocated area,
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07003886 * so we need to unpoison this area.
3887 */
3888 kasan_unpoison_shadow(object, size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003889 return size;
3890}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003891EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003892
3893void kfree(const void *x)
3894{
Christoph Lameter81819f02007-05-06 14:49:36 -07003895 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003896 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003897
Pekka Enberg2121db72009-03-25 11:05:57 +02003898 trace_kfree(_RET_IP_, x);
3899
Satyam Sharma2408c552007-10-16 01:24:44 -07003900 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003901 return;
3902
Christoph Lameterb49af682007-05-06 14:49:41 -07003903 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003904 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003905 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003906 kfree_hook(x);
Se Wang (Patrick) Ohaf343622015-06-25 15:55:04 -07003907 kasan_alloc_pages(page, compound_order(page));
Vladimir Davydov49491482016-07-26 15:24:24 -07003908 __free_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003909 return;
3910 }
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003911 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003912}
3913EXPORT_SYMBOL(kfree);
3914
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003915#define SHRINK_PROMOTE_MAX 32
3916
Christoph Lameter2086d262007-05-06 14:49:46 -07003917/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003918 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3919 * up most to the head of the partial lists. New allocations will then
3920 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003921 *
3922 * The slabs with the least items are placed last. This results in them
3923 * being allocated from last increasing the chance that the last objects
3924 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003925 */
Vladimir Davydovbc01eb92017-03-17 00:48:31 +00003926int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07003927{
3928 int node;
3929 int i;
3930 struct kmem_cache_node *n;
3931 struct page *page;
3932 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003933 struct list_head discard;
3934 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003935 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003936 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003937
Christoph Lameter2086d262007-05-06 14:49:46 -07003938 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003939 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003940 INIT_LIST_HEAD(&discard);
3941 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3942 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003943
3944 spin_lock_irqsave(&n->list_lock, flags);
3945
3946 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003947 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003948 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003949 * Note that concurrent frees may occur while we hold the
3950 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003951 */
3952 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003953 int free = page->objects - page->inuse;
3954
3955 /* Do not reread page->inuse */
3956 barrier();
3957
3958 /* We do not keep full slabs on the list */
3959 BUG_ON(free <= 0);
3960
3961 if (free == page->objects) {
3962 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003963 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003964 } else if (free <= SHRINK_PROMOTE_MAX)
3965 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07003966 }
3967
Christoph Lameter2086d262007-05-06 14:49:46 -07003968 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003969 * Promote the slabs filled up most to the head of the
3970 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07003971 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003972 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3973 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07003974
Christoph Lameter2086d262007-05-06 14:49:46 -07003975 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003976
3977 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003978 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003979 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003980
3981 if (slabs_node(s, node))
3982 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07003983 }
3984
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003985 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07003986}
Christoph Lameter2086d262007-05-06 14:49:46 -07003987
Yasunori Gotob9049e22007-10-21 16:41:37 -07003988static int slab_mem_going_offline_callback(void *arg)
3989{
3990 struct kmem_cache *s;
3991
Christoph Lameter18004c52012-07-06 15:25:12 -05003992 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003993 list_for_each_entry(s, &slab_caches, list)
Vladimir Davydovbc01eb92017-03-17 00:48:31 +00003994 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05003995 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003996
3997 return 0;
3998}
3999
4000static void slab_mem_offline_callback(void *arg)
4001{
4002 struct kmem_cache_node *n;
4003 struct kmem_cache *s;
4004 struct memory_notify *marg = arg;
4005 int offline_node;
4006
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004007 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004008
4009 /*
4010 * If the node still has available memory. we need kmem_cache_node
4011 * for it yet.
4012 */
4013 if (offline_node < 0)
4014 return;
4015
Christoph Lameter18004c52012-07-06 15:25:12 -05004016 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004017 list_for_each_entry(s, &slab_caches, list) {
4018 n = get_node(s, offline_node);
4019 if (n) {
4020 /*
4021 * if n->nr_slabs > 0, slabs still exist on the node
4022 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05004023 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07004024 * callback. So, we must fail.
4025 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03004026 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07004027
4028 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05004029 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004030 }
4031 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004032 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004033}
4034
4035static int slab_mem_going_online_callback(void *arg)
4036{
4037 struct kmem_cache_node *n;
4038 struct kmem_cache *s;
4039 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004040 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004041 int ret = 0;
4042
4043 /*
4044 * If the node's memory is already available, then kmem_cache_node is
4045 * already created. Nothing to do.
4046 */
4047 if (nid < 0)
4048 return 0;
4049
4050 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07004051 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07004052 * allocate a kmem_cache_node structure in order to bring the node
4053 * online.
4054 */
Christoph Lameter18004c52012-07-06 15:25:12 -05004055 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004056 list_for_each_entry(s, &slab_caches, list) {
4057 /*
4058 * XXX: kmem_cache_alloc_node will fallback to other nodes
4059 * since memory is not yet available from the node that
4060 * is brought up.
4061 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05004062 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004063 if (!n) {
4064 ret = -ENOMEM;
4065 goto out;
4066 }
Joonsoo Kim40534972012-05-11 00:50:47 +09004067 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004068 s->node[nid] = n;
4069 }
4070out:
Christoph Lameter18004c52012-07-06 15:25:12 -05004071 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004072 return ret;
4073}
4074
4075static int slab_memory_callback(struct notifier_block *self,
4076 unsigned long action, void *arg)
4077{
4078 int ret = 0;
4079
4080 switch (action) {
4081 case MEM_GOING_ONLINE:
4082 ret = slab_mem_going_online_callback(arg);
4083 break;
4084 case MEM_GOING_OFFLINE:
4085 ret = slab_mem_going_offline_callback(arg);
4086 break;
4087 case MEM_OFFLINE:
4088 case MEM_CANCEL_ONLINE:
4089 slab_mem_offline_callback(arg);
4090 break;
4091 case MEM_ONLINE:
4092 case MEM_CANCEL_OFFLINE:
4093 break;
4094 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08004095 if (ret)
4096 ret = notifier_from_errno(ret);
4097 else
4098 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004099 return ret;
4100}
4101
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004102static struct notifier_block slab_memory_callback_nb = {
4103 .notifier_call = slab_memory_callback,
4104 .priority = SLAB_CALLBACK_PRI,
4105};
Yasunori Gotob9049e22007-10-21 16:41:37 -07004106
Christoph Lameter81819f02007-05-06 14:49:36 -07004107/********************************************************************
4108 * Basic setup of slabs
4109 *******************************************************************/
4110
Christoph Lameter51df1142010-08-20 12:37:15 -05004111/*
4112 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004113 * the page allocator. Allocate them properly then fix up the pointers
4114 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05004115 */
4116
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004117static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05004118{
4119 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004120 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004121 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05004122
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004123 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05004124
Glauber Costa7d557b32013-02-22 20:20:00 +04004125 /*
4126 * This runs very early, and only the boot processor is supposed to be
4127 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4128 * IPIs around.
4129 */
4130 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004131 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05004132 struct page *p;
4133
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004134 list_for_each_entry(p, &n->partial, lru)
4135 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004136
Li Zefan607bf322011-04-12 15:22:26 +08004137#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004138 list_for_each_entry(p, &n->full, lru)
4139 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004140#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05004141 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08004142 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004143 list_add(&s->list, &slab_caches);
4144 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004145}
4146
Christoph Lameter81819f02007-05-06 14:49:36 -07004147void __init kmem_cache_init(void)
4148{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004149 static __initdata struct kmem_cache boot_kmem_cache,
4150 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05004151
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08004152 if (debug_guardpage_minorder())
4153 slub_max_order = 0;
4154
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004155 kmem_cache_node = &boot_kmem_cache_node;
4156 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05004157
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004158 create_boot_cache(kmem_cache_node, "kmem_cache_node",
4159 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004160
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004161 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07004162
4163 /* Able to allocate the per node structures */
4164 slab_state = PARTIAL;
4165
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004166 create_boot_cache(kmem_cache, "kmem_cache",
4167 offsetof(struct kmem_cache, node) +
4168 nr_node_ids * sizeof(struct kmem_cache_node *),
4169 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004170
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004171 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07004172
Christoph Lameter51df1142010-08-20 12:37:15 -05004173 /*
4174 * Allocate kmem_cache_node properly from the kmem_cache slab.
4175 * kmem_cache_node is separately allocated so no need to
4176 * update any list pointers.
4177 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004178 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05004179
4180 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07004181 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004182 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07004183
Thomas Garnier210e7a42016-07-26 15:21:59 -07004184 /* Setup random freelists for each cache */
4185 init_freelist_randomization();
4186
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02004187 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4188 slub_cpu_dead);
Christoph Lameter81819f02007-05-06 14:49:36 -07004189
Fabian Frederickf9f58282014-06-04 16:06:34 -07004190 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004191 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07004192 slub_min_order, slub_max_order, slub_min_objects,
4193 nr_cpu_ids, nr_node_ids);
4194}
4195
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004196void __init kmem_cache_init_late(void)
4197{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004198}
4199
Glauber Costa2633d7a2012-12-18 14:22:34 -08004200struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07004201__kmem_cache_alias(const char *name, size_t size, size_t align,
4202 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07004203{
Vladimir Davydov426589f2015-02-12 14:59:23 -08004204 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07004205
Vladimir Davydova44cb942014-04-07 15:39:23 -07004206 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004207 if (s) {
4208 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004209
Christoph Lameter81819f02007-05-06 14:49:36 -07004210 /*
4211 * Adjust the object sizes so that we clear
4212 * the complete object on kzalloc.
4213 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004214 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004215 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08004216
Vladimir Davydov426589f2015-02-12 14:59:23 -08004217 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004218 c->object_size = s->object_size;
4219 c->inuse = max_t(int, c->inuse,
4220 ALIGN(size, sizeof(void *)));
4221 }
4222
David Rientjes7b8f3b62008-12-17 22:09:46 -08004223 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08004224 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00004225 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08004226 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07004227 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08004228
Christoph Lametercbb79692012-09-05 00:18:32 +00004229 return s;
4230}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004231
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004232int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00004233{
Pekka Enbergaac3a162012-09-05 12:07:44 +03004234 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004235
Pekka Enbergaac3a162012-09-05 12:07:44 +03004236 err = kmem_cache_open(s, flags);
4237 if (err)
4238 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004239
Christoph Lameter45530c42012-11-28 16:23:07 +00004240 /* Mutex is not taken during early boot */
4241 if (slab_state <= UP)
4242 return 0;
4243
Glauber Costa107dab52012-12-18 14:23:05 -08004244 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004245 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004246 if (err)
Dmitry Safonov52b4b952016-02-17 13:11:37 -08004247 __kmem_cache_release(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004248
4249 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004250}
Christoph Lameter81819f02007-05-06 14:49:36 -07004251
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004252void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004253{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004254 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004255 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004256
Christoph Lameter95a05b42013-01-10 19:14:19 +00004257 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004258 return kmalloc_large(size, gfpflags);
4259
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004260 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004261
Satyam Sharma2408c552007-10-16 01:24:44 -07004262 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004263 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004264
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004265 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004266
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004267 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004268 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004269
4270 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004271}
4272
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004273#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004274void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004275 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004276{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004277 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004278 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004279
Christoph Lameter95a05b42013-01-10 19:14:19 +00004280 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004281 ret = kmalloc_large_node(size, gfpflags, node);
4282
4283 trace_kmalloc_node(caller, ret,
4284 size, PAGE_SIZE << get_order(size),
4285 gfpflags, node);
4286
4287 return ret;
4288 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004289
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004290 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004291
Satyam Sharma2408c552007-10-16 01:24:44 -07004292 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004293 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004294
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004295 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004296
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004297 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004298 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004299
4300 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004301}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004302#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004303
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004304#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004305static int count_inuse(struct page *page)
4306{
4307 return page->inuse;
4308}
4309
4310static int count_total(struct page *page)
4311{
4312 return page->objects;
4313}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004314#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004315
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004316#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004317static int validate_slab(struct kmem_cache *s, struct page *page,
4318 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004319{
4320 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004321 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004322
4323 if (!check_slab(s, page) ||
4324 !on_freelist(s, page, NULL))
4325 return 0;
4326
4327 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004328 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004329
Christoph Lameter5f80b132011-04-15 14:48:13 -05004330 get_map(s, page, map);
4331 for_each_object(p, s, addr, page->objects) {
4332 if (test_bit(slab_index(p, s, addr), map))
4333 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4334 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004335 }
4336
Christoph Lameter224a88b2008-04-14 19:11:31 +03004337 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004338 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004339 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004340 return 0;
4341 return 1;
4342}
4343
Christoph Lameter434e2452007-07-17 04:03:30 -07004344static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4345 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004346{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004347 slab_lock(page);
4348 validate_slab(s, page, map);
4349 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004350}
4351
Christoph Lameter434e2452007-07-17 04:03:30 -07004352static int validate_slab_node(struct kmem_cache *s,
4353 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004354{
4355 unsigned long count = 0;
4356 struct page *page;
4357 unsigned long flags;
4358
4359 spin_lock_irqsave(&n->list_lock, flags);
4360
4361 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004362 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004363 count++;
4364 }
4365 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004366 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4367 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004368
4369 if (!(s->flags & SLAB_STORE_USER))
4370 goto out;
4371
4372 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004373 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004374 count++;
4375 }
4376 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004377 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4378 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004379
4380out:
4381 spin_unlock_irqrestore(&n->list_lock, flags);
4382 return count;
4383}
4384
Christoph Lameter434e2452007-07-17 04:03:30 -07004385static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004386{
4387 int node;
4388 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004389 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004390 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004391 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07004392
4393 if (!map)
4394 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004395
4396 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004397 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07004398 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07004399 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004400 return count;
4401}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004402/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004403 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004404 * and freed.
4405 */
4406
4407struct location {
4408 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004409 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004410 long long sum_time;
4411 long min_time;
4412 long max_time;
4413 long min_pid;
4414 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304415 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004416 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004417};
4418
4419struct loc_track {
4420 unsigned long max;
4421 unsigned long count;
4422 struct location *loc;
4423};
4424
4425static void free_loc_track(struct loc_track *t)
4426{
4427 if (t->max)
4428 free_pages((unsigned long)t->loc,
4429 get_order(sizeof(struct location) * t->max));
4430}
4431
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004432static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004433{
4434 struct location *l;
4435 int order;
4436
Christoph Lameter88a420e2007-05-06 14:49:45 -07004437 order = get_order(sizeof(struct location) * max);
4438
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004439 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004440 if (!l)
4441 return 0;
4442
4443 if (t->count) {
4444 memcpy(l, t->loc, sizeof(struct location) * t->count);
4445 free_loc_track(t);
4446 }
4447 t->max = max;
4448 t->loc = l;
4449 return 1;
4450}
4451
4452static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004453 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004454{
4455 long start, end, pos;
4456 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004457 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004458 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004459
4460 start = -1;
4461 end = t->count;
4462
4463 for ( ; ; ) {
4464 pos = start + (end - start + 1) / 2;
4465
4466 /*
4467 * There is nothing at "end". If we end up there
4468 * we need to add something to before end.
4469 */
4470 if (pos == end)
4471 break;
4472
4473 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004474 if (track->addr == caddr) {
4475
4476 l = &t->loc[pos];
4477 l->count++;
4478 if (track->when) {
4479 l->sum_time += age;
4480 if (age < l->min_time)
4481 l->min_time = age;
4482 if (age > l->max_time)
4483 l->max_time = age;
4484
4485 if (track->pid < l->min_pid)
4486 l->min_pid = track->pid;
4487 if (track->pid > l->max_pid)
4488 l->max_pid = track->pid;
4489
Rusty Russell174596a2009-01-01 10:12:29 +10304490 cpumask_set_cpu(track->cpu,
4491 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004492 }
4493 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004494 return 1;
4495 }
4496
Christoph Lameter45edfa52007-05-09 02:32:45 -07004497 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004498 end = pos;
4499 else
4500 start = pos;
4501 }
4502
4503 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004504 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004505 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004506 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004507 return 0;
4508
4509 l = t->loc + pos;
4510 if (pos < t->count)
4511 memmove(l + 1, l,
4512 (t->count - pos) * sizeof(struct location));
4513 t->count++;
4514 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004515 l->addr = track->addr;
4516 l->sum_time = age;
4517 l->min_time = age;
4518 l->max_time = age;
4519 l->min_pid = track->pid;
4520 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304521 cpumask_clear(to_cpumask(l->cpus));
4522 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004523 nodes_clear(l->nodes);
4524 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004525 return 1;
4526}
4527
4528static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004529 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004530 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004531{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004532 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004533 void *p;
4534
Christoph Lameter39b26462008-04-14 19:11:30 +03004535 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004536 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004537
Christoph Lameter224a88b2008-04-14 19:11:31 +03004538 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004539 if (!test_bit(slab_index(p, s, addr), map))
4540 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004541}
4542
4543static int list_locations(struct kmem_cache *s, char *buf,
4544 enum track_item alloc)
4545{
Harvey Harrisone374d482008-01-31 15:20:50 -08004546 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004547 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004548 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004549 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004550 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4551 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004552 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004553
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004554 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4555 GFP_TEMPORARY)) {
4556 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004557 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004558 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004559 /* Push back cpu slabs */
4560 flush_all(s);
4561
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004562 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004563 unsigned long flags;
4564 struct page *page;
4565
Christoph Lameter9e869432007-08-22 14:01:56 -07004566 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004567 continue;
4568
4569 spin_lock_irqsave(&n->list_lock, flags);
4570 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004571 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004572 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004573 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004574 spin_unlock_irqrestore(&n->list_lock, flags);
4575 }
4576
4577 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004578 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004579
Hugh Dickins9c246242008-12-09 13:14:27 -08004580 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004581 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004582 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004583
4584 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004585 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004586 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004587 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004588
4589 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004590 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004591 l->min_time,
4592 (long)div_u64(l->sum_time, l->count),
4593 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004594 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004595 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004596 l->min_time);
4597
4598 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004599 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004600 l->min_pid, l->max_pid);
4601 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004602 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004603 l->min_pid);
4604
Rusty Russell174596a2009-01-01 10:12:29 +10304605 if (num_online_cpus() > 1 &&
4606 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004607 len < PAGE_SIZE - 60)
4608 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4609 " cpus=%*pbl",
4610 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004611
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004612 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004613 len < PAGE_SIZE - 60)
4614 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4615 " nodes=%*pbl",
4616 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004617
Harvey Harrisone374d482008-01-31 15:20:50 -08004618 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004619 }
4620
4621 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004622 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004623 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004624 len += sprintf(buf, "No data\n");
4625 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004626}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004627#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004628
Christoph Lametera5a84752010-10-05 13:57:27 -05004629#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004630static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004631{
4632 u8 *p;
4633
Christoph Lameter95a05b42013-01-10 19:14:19 +00004634 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004635
Fabian Frederickf9f58282014-06-04 16:06:34 -07004636 pr_err("SLUB resiliency testing\n");
4637 pr_err("-----------------------\n");
4638 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004639
4640 p = kzalloc(16, GFP_KERNEL);
4641 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004642 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4643 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004644
4645 validate_slab_cache(kmalloc_caches[4]);
4646
4647 /* Hmmm... The next two are dangerous */
4648 p = kzalloc(32, GFP_KERNEL);
4649 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004650 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4651 p);
4652 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004653
4654 validate_slab_cache(kmalloc_caches[5]);
4655 p = kzalloc(64, GFP_KERNEL);
4656 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4657 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004658 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->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 validate_slab_cache(kmalloc_caches[6]);
4662
Fabian Frederickf9f58282014-06-04 16:06:34 -07004663 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004664 p = kzalloc(128, GFP_KERNEL);
4665 kfree(p);
4666 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004667 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004668 validate_slab_cache(kmalloc_caches[7]);
4669
4670 p = kzalloc(256, GFP_KERNEL);
4671 kfree(p);
4672 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004673 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004674 validate_slab_cache(kmalloc_caches[8]);
4675
4676 p = kzalloc(512, GFP_KERNEL);
4677 kfree(p);
4678 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004679 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004680 validate_slab_cache(kmalloc_caches[9]);
4681}
4682#else
4683#ifdef CONFIG_SYSFS
4684static void resiliency_test(void) {};
4685#endif
4686#endif
4687
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004688#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004689enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004690 SL_ALL, /* All slabs */
4691 SL_PARTIAL, /* Only partially allocated slabs */
4692 SL_CPU, /* Only slabs used for cpu caches */
4693 SL_OBJECTS, /* Determine allocated objects not slabs */
4694 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004695};
4696
Christoph Lameter205ab992008-04-14 19:11:40 +03004697#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004698#define SO_PARTIAL (1 << SL_PARTIAL)
4699#define SO_CPU (1 << SL_CPU)
4700#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004701#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004702
Tejun Heoa4ffb672018-08-24 13:22:21 +09004703#ifdef CONFIG_MEMCG
4704static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4705
4706static int __init setup_slub_memcg_sysfs(char *str)
4707{
4708 int v;
4709
4710 if (get_option(&str, &v) > 0)
4711 memcg_sysfs_enabled = v;
4712
4713 return 1;
4714}
4715
4716__setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4717#endif
4718
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004719static ssize_t show_slab_objects(struct kmem_cache *s,
4720 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004721{
4722 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004723 int node;
4724 int x;
4725 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004726
Chen Gange35e1a92013-07-12 08:23:48 +08004727 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004728 if (!nodes)
4729 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004730
Christoph Lameter205ab992008-04-14 19:11:40 +03004731 if (flags & SO_CPU) {
4732 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004733
Christoph Lameter205ab992008-04-14 19:11:40 +03004734 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004735 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4736 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004737 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004738 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004739
Jason Low4db0c3c2015-04-15 16:14:08 -07004740 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004741 if (!page)
4742 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004743
Christoph Lameterec3ab082012-05-09 10:09:56 -05004744 node = page_to_nid(page);
4745 if (flags & SO_TOTAL)
4746 x = page->objects;
4747 else if (flags & SO_OBJECTS)
4748 x = page->inuse;
4749 else
4750 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004751
Christoph Lameterec3ab082012-05-09 10:09:56 -05004752 total += x;
4753 nodes[node] += x;
4754
Jason Low4db0c3c2015-04-15 16:14:08 -07004755 page = READ_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004756 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004757 node = page_to_nid(page);
4758 if (flags & SO_TOTAL)
4759 WARN_ON_ONCE(1);
4760 else if (flags & SO_OBJECTS)
4761 WARN_ON_ONCE(1);
4762 else
4763 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004764 total += x;
4765 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004766 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004767 }
4768 }
4769
Qian Caif60a6052019-10-14 14:11:51 -07004770 /*
4771 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
4772 * already held which will conflict with an existing lock order:
4773 *
4774 * mem_hotplug_lock->slab_mutex->kernfs_mutex
4775 *
4776 * We don't really need mem_hotplug_lock (to hold off
4777 * slab_mem_going_offline_callback) here because slab's memory hot
4778 * unplug code doesn't destroy the kmem_cache->node[] data.
4779 */
4780
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004781#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004782 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004783 struct kmem_cache_node *n;
4784
4785 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004786
Chen Gangd0e0ac92013-07-15 09:05:29 +08004787 if (flags & SO_TOTAL)
4788 x = atomic_long_read(&n->total_objects);
4789 else if (flags & SO_OBJECTS)
4790 x = atomic_long_read(&n->total_objects) -
4791 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004792 else
4793 x = atomic_long_read(&n->nr_slabs);
4794 total += x;
4795 nodes[node] += x;
4796 }
4797
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004798 } else
4799#endif
4800 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004801 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004802
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004803 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004804 if (flags & SO_TOTAL)
4805 x = count_partial(n, count_total);
4806 else if (flags & SO_OBJECTS)
4807 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004808 else
4809 x = n->nr_partial;
4810 total += x;
4811 nodes[node] += x;
4812 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004813 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004814 x = sprintf(buf, "%lu", total);
4815#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004816 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004817 if (nodes[node])
4818 x += sprintf(buf + x, " N%d=%lu",
4819 node, nodes[node]);
4820#endif
4821 kfree(nodes);
4822 return x + sprintf(buf + x, "\n");
4823}
4824
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004825#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004826static int any_slab_objects(struct kmem_cache *s)
4827{
4828 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004829 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004830
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004831 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004832 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004833 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004834
Christoph Lameter81819f02007-05-06 14:49:36 -07004835 return 0;
4836}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004837#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004838
4839#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004840#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004841
4842struct slab_attribute {
4843 struct attribute attr;
4844 ssize_t (*show)(struct kmem_cache *s, char *buf);
4845 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4846};
4847
4848#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004849 static struct slab_attribute _name##_attr = \
4850 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004851
4852#define SLAB_ATTR(_name) \
4853 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004854 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004855
Christoph Lameter81819f02007-05-06 14:49:36 -07004856static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4857{
4858 return sprintf(buf, "%d\n", s->size);
4859}
4860SLAB_ATTR_RO(slab_size);
4861
4862static ssize_t align_show(struct kmem_cache *s, char *buf)
4863{
4864 return sprintf(buf, "%d\n", s->align);
4865}
4866SLAB_ATTR_RO(align);
4867
4868static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4869{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004870 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004871}
4872SLAB_ATTR_RO(object_size);
4873
4874static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4875{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004876 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004877}
4878SLAB_ATTR_RO(objs_per_slab);
4879
Christoph Lameter06b285d2008-04-14 19:11:41 +03004880static ssize_t order_store(struct kmem_cache *s,
4881 const char *buf, size_t length)
4882{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004883 unsigned long order;
4884 int err;
4885
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004886 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004887 if (err)
4888 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004889
4890 if (order > slub_max_order || order < slub_min_order)
4891 return -EINVAL;
4892
4893 calculate_sizes(s, order);
4894 return length;
4895}
4896
Christoph Lameter81819f02007-05-06 14:49:36 -07004897static ssize_t order_show(struct kmem_cache *s, char *buf)
4898{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004899 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004900}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004901SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004902
David Rientjes73d342b2009-02-22 17:40:09 -08004903static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4904{
4905 return sprintf(buf, "%lu\n", s->min_partial);
4906}
4907
4908static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4909 size_t length)
4910{
4911 unsigned long min;
4912 int err;
4913
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004914 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004915 if (err)
4916 return err;
4917
David Rientjesc0bdb232009-02-25 09:16:35 +02004918 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004919 return length;
4920}
4921SLAB_ATTR(min_partial);
4922
Christoph Lameter49e22582011-08-09 16:12:27 -05004923static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4924{
4925 return sprintf(buf, "%u\n", s->cpu_partial);
4926}
4927
4928static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4929 size_t length)
4930{
Alexey Dobriyanb052d042018-04-05 16:21:10 -07004931 unsigned int objects;
Christoph Lameter49e22582011-08-09 16:12:27 -05004932 int err;
4933
Alexey Dobriyanb052d042018-04-05 16:21:10 -07004934 err = kstrtouint(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004935 if (err)
4936 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004937 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004938 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004939
4940 s->cpu_partial = objects;
4941 flush_all(s);
4942 return length;
4943}
4944SLAB_ATTR(cpu_partial);
4945
Christoph Lameter81819f02007-05-06 14:49:36 -07004946static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4947{
Joe Perches62c70bc2011-01-13 15:45:52 -08004948 if (!s->ctor)
4949 return 0;
4950 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004951}
4952SLAB_ATTR_RO(ctor);
4953
Christoph Lameter81819f02007-05-06 14:49:36 -07004954static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4955{
Gu Zheng4307c142014-08-06 16:04:51 -07004956 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004957}
4958SLAB_ATTR_RO(aliases);
4959
Christoph Lameter81819f02007-05-06 14:49:36 -07004960static ssize_t partial_show(struct kmem_cache *s, char *buf)
4961{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004962 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004963}
4964SLAB_ATTR_RO(partial);
4965
4966static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4967{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004968 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004969}
4970SLAB_ATTR_RO(cpu_slabs);
4971
4972static ssize_t objects_show(struct kmem_cache *s, char *buf)
4973{
Christoph Lameter205ab992008-04-14 19:11:40 +03004974 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004975}
4976SLAB_ATTR_RO(objects);
4977
Christoph Lameter205ab992008-04-14 19:11:40 +03004978static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4979{
4980 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4981}
4982SLAB_ATTR_RO(objects_partial);
4983
Christoph Lameter49e22582011-08-09 16:12:27 -05004984static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4985{
4986 int objects = 0;
4987 int pages = 0;
4988 int cpu;
4989 int len;
4990
4991 for_each_online_cpu(cpu) {
4992 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4993
4994 if (page) {
4995 pages += page->pages;
4996 objects += page->pobjects;
4997 }
4998 }
4999
5000 len = sprintf(buf, "%d(%d)", objects, pages);
5001
5002#ifdef CONFIG_SMP
5003 for_each_online_cpu(cpu) {
5004 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
5005
5006 if (page && len < PAGE_SIZE - 20)
5007 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5008 page->pobjects, page->pages);
5009 }
5010#endif
5011 return len + sprintf(buf + len, "\n");
5012}
5013SLAB_ATTR_RO(slabs_cpu_partial);
5014
Christoph Lameter81819f02007-05-06 14:49:36 -07005015static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5016{
5017 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5018}
5019
5020static ssize_t reclaim_account_store(struct kmem_cache *s,
5021 const char *buf, size_t length)
5022{
5023 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5024 if (buf[0] == '1')
5025 s->flags |= SLAB_RECLAIM_ACCOUNT;
5026 return length;
5027}
5028SLAB_ATTR(reclaim_account);
5029
5030static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5031{
Christoph Lameter5af60832007-05-06 14:49:56 -07005032 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07005033}
5034SLAB_ATTR_RO(hwcache_align);
5035
5036#ifdef CONFIG_ZONE_DMA
5037static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5038{
5039 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5040}
5041SLAB_ATTR_RO(cache_dma);
5042#endif
5043
5044static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5045{
5046 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
5047}
5048SLAB_ATTR_RO(destroy_by_rcu);
5049
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005050static ssize_t reserved_show(struct kmem_cache *s, char *buf)
5051{
5052 return sprintf(buf, "%d\n", s->reserved);
5053}
5054SLAB_ATTR_RO(reserved);
5055
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005056#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005057static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5058{
5059 return show_slab_objects(s, buf, SO_ALL);
5060}
5061SLAB_ATTR_RO(slabs);
5062
5063static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5064{
5065 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5066}
5067SLAB_ATTR_RO(total_objects);
5068
5069static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5070{
Laura Abbottbecfda62016-03-15 14:55:06 -07005071 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
Christoph Lametera5a84752010-10-05 13:57:27 -05005072}
5073
5074static ssize_t sanity_checks_store(struct kmem_cache *s,
5075 const char *buf, size_t length)
5076{
Laura Abbottbecfda62016-03-15 14:55:06 -07005077 s->flags &= ~SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005078 if (buf[0] == '1') {
5079 s->flags &= ~__CMPXCHG_DOUBLE;
Laura Abbottbecfda62016-03-15 14:55:06 -07005080 s->flags |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005081 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005082 return length;
5083}
5084SLAB_ATTR(sanity_checks);
5085
5086static ssize_t trace_show(struct kmem_cache *s, char *buf)
5087{
5088 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5089}
5090
5091static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5092 size_t length)
5093{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005094 /*
5095 * Tracing a merged cache is going to give confusing results
5096 * as well as cause other issues like converting a mergeable
5097 * cache into an umergeable one.
5098 */
5099 if (s->refcount > 1)
5100 return -EINVAL;
5101
Christoph Lametera5a84752010-10-05 13:57:27 -05005102 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005103 if (buf[0] == '1') {
5104 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05005105 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005106 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005107 return length;
5108}
5109SLAB_ATTR(trace);
5110
Christoph Lameter81819f02007-05-06 14:49:36 -07005111static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5112{
5113 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5114}
5115
5116static ssize_t red_zone_store(struct kmem_cache *s,
5117 const char *buf, size_t length)
5118{
5119 if (any_slab_objects(s))
5120 return -EBUSY;
5121
5122 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005123 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005124 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005125 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005126 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005127 return length;
5128}
5129SLAB_ATTR(red_zone);
5130
5131static ssize_t poison_show(struct kmem_cache *s, char *buf)
5132{
5133 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5134}
5135
5136static ssize_t poison_store(struct kmem_cache *s,
5137 const char *buf, size_t length)
5138{
5139 if (any_slab_objects(s))
5140 return -EBUSY;
5141
5142 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005143 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005144 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005145 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005146 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005147 return length;
5148}
5149SLAB_ATTR(poison);
5150
5151static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5152{
5153 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5154}
5155
5156static ssize_t store_user_store(struct kmem_cache *s,
5157 const char *buf, size_t length)
5158{
5159 if (any_slab_objects(s))
5160 return -EBUSY;
5161
5162 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005163 if (buf[0] == '1') {
5164 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07005165 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005166 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005167 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005168 return length;
5169}
5170SLAB_ATTR(store_user);
5171
Christoph Lameter53e15af2007-05-06 14:49:43 -07005172static ssize_t validate_show(struct kmem_cache *s, char *buf)
5173{
5174 return 0;
5175}
5176
5177static ssize_t validate_store(struct kmem_cache *s,
5178 const char *buf, size_t length)
5179{
Christoph Lameter434e2452007-07-17 04:03:30 -07005180 int ret = -EINVAL;
5181
5182 if (buf[0] == '1') {
5183 ret = validate_slab_cache(s);
5184 if (ret >= 0)
5185 ret = length;
5186 }
5187 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07005188}
5189SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05005190
5191static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5192{
5193 if (!(s->flags & SLAB_STORE_USER))
5194 return -ENOSYS;
5195 return list_locations(s, buf, TRACK_ALLOC);
5196}
5197SLAB_ATTR_RO(alloc_calls);
5198
5199static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5200{
5201 if (!(s->flags & SLAB_STORE_USER))
5202 return -ENOSYS;
5203 return list_locations(s, buf, TRACK_FREE);
5204}
5205SLAB_ATTR_RO(free_calls);
5206#endif /* CONFIG_SLUB_DEBUG */
5207
5208#ifdef CONFIG_FAILSLAB
5209static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5210{
5211 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5212}
5213
5214static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5215 size_t length)
5216{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005217 if (s->refcount > 1)
5218 return -EINVAL;
5219
Christoph Lametera5a84752010-10-05 13:57:27 -05005220 s->flags &= ~SLAB_FAILSLAB;
5221 if (buf[0] == '1')
5222 s->flags |= SLAB_FAILSLAB;
5223 return length;
5224}
5225SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005226#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07005227
Christoph Lameter2086d262007-05-06 14:49:46 -07005228static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5229{
5230 return 0;
5231}
5232
5233static ssize_t shrink_store(struct kmem_cache *s,
5234 const char *buf, size_t length)
5235{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005236 if (buf[0] == '1')
5237 kmem_cache_shrink(s);
5238 else
Christoph Lameter2086d262007-05-06 14:49:46 -07005239 return -EINVAL;
5240 return length;
5241}
5242SLAB_ATTR(shrink);
5243
Christoph Lameter81819f02007-05-06 14:49:36 -07005244#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005245static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07005246{
Christoph Lameter98246012008-01-07 23:20:26 -08005247 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07005248}
5249
Christoph Lameter98246012008-01-07 23:20:26 -08005250static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07005251 const char *buf, size_t length)
5252{
Christoph Lameter0121c6192008-04-29 16:11:12 -07005253 unsigned long ratio;
5254 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005255
Jingoo Han3dbb95f2013-09-11 14:20:25 -07005256 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07005257 if (err)
5258 return err;
5259
Christoph Lametere2cb96b2008-08-19 08:51:22 -05005260 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07005261 s->remote_node_defrag_ratio = ratio * 10;
5262
Christoph Lameter81819f02007-05-06 14:49:36 -07005263 return length;
5264}
Christoph Lameter98246012008-01-07 23:20:26 -08005265SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005266#endif
5267
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005268#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005269static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5270{
5271 unsigned long sum = 0;
5272 int cpu;
5273 int len;
5274 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5275
5276 if (!data)
5277 return -ENOMEM;
5278
5279 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005280 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005281
5282 data[cpu] = x;
5283 sum += x;
5284 }
5285
5286 len = sprintf(buf, "%lu", sum);
5287
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005288#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005289 for_each_online_cpu(cpu) {
5290 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005291 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005292 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005293#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005294 kfree(data);
5295 return len + sprintf(buf + len, "\n");
5296}
5297
David Rientjes78eb00c2009-10-15 02:20:22 -07005298static void clear_stat(struct kmem_cache *s, enum stat_item si)
5299{
5300 int cpu;
5301
5302 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005303 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005304}
5305
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005306#define STAT_ATTR(si, text) \
5307static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5308{ \
5309 return show_stat(s, buf, si); \
5310} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005311static ssize_t text##_store(struct kmem_cache *s, \
5312 const char *buf, size_t length) \
5313{ \
5314 if (buf[0] != '0') \
5315 return -EINVAL; \
5316 clear_stat(s, si); \
5317 return length; \
5318} \
5319SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005320
5321STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5322STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5323STAT_ATTR(FREE_FASTPATH, free_fastpath);
5324STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5325STAT_ATTR(FREE_FROZEN, free_frozen);
5326STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5327STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5328STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5329STAT_ATTR(ALLOC_SLAB, alloc_slab);
5330STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005331STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005332STAT_ATTR(FREE_SLAB, free_slab);
5333STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5334STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5335STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5336STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5337STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5338STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005339STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005340STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005341STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5342STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005343STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5344STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005345STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5346STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005347#endif
5348
Pekka Enberg06428782008-01-07 23:20:27 -08005349static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005350 &slab_size_attr.attr,
5351 &object_size_attr.attr,
5352 &objs_per_slab_attr.attr,
5353 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005354 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005355 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005356 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005357 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005358 &partial_attr.attr,
5359 &cpu_slabs_attr.attr,
5360 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005361 &aliases_attr.attr,
5362 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005363 &hwcache_align_attr.attr,
5364 &reclaim_account_attr.attr,
5365 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005366 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005367 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005368 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005369#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005370 &total_objects_attr.attr,
5371 &slabs_attr.attr,
5372 &sanity_checks_attr.attr,
5373 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005374 &red_zone_attr.attr,
5375 &poison_attr.attr,
5376 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005377 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005378 &alloc_calls_attr.attr,
5379 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005380#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005381#ifdef CONFIG_ZONE_DMA
5382 &cache_dma_attr.attr,
5383#endif
5384#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005385 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005386#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005387#ifdef CONFIG_SLUB_STATS
5388 &alloc_fastpath_attr.attr,
5389 &alloc_slowpath_attr.attr,
5390 &free_fastpath_attr.attr,
5391 &free_slowpath_attr.attr,
5392 &free_frozen_attr.attr,
5393 &free_add_partial_attr.attr,
5394 &free_remove_partial_attr.attr,
5395 &alloc_from_partial_attr.attr,
5396 &alloc_slab_attr.attr,
5397 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005398 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005399 &free_slab_attr.attr,
5400 &cpuslab_flush_attr.attr,
5401 &deactivate_full_attr.attr,
5402 &deactivate_empty_attr.attr,
5403 &deactivate_to_head_attr.attr,
5404 &deactivate_to_tail_attr.attr,
5405 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005406 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005407 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005408 &cmpxchg_double_fail_attr.attr,
5409 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005410 &cpu_partial_alloc_attr.attr,
5411 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005412 &cpu_partial_node_attr.attr,
5413 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005414#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005415#ifdef CONFIG_FAILSLAB
5416 &failslab_attr.attr,
5417#endif
5418
Christoph Lameter81819f02007-05-06 14:49:36 -07005419 NULL
5420};
5421
5422static struct attribute_group slab_attr_group = {
5423 .attrs = slab_attrs,
5424};
5425
5426static ssize_t slab_attr_show(struct kobject *kobj,
5427 struct attribute *attr,
5428 char *buf)
5429{
5430 struct slab_attribute *attribute;
5431 struct kmem_cache *s;
5432 int err;
5433
5434 attribute = to_slab_attr(attr);
5435 s = to_slab(kobj);
5436
5437 if (!attribute->show)
5438 return -EIO;
5439
5440 err = attribute->show(s, buf);
5441
5442 return err;
5443}
5444
5445static ssize_t slab_attr_store(struct kobject *kobj,
5446 struct attribute *attr,
5447 const char *buf, size_t len)
5448{
5449 struct slab_attribute *attribute;
5450 struct kmem_cache *s;
5451 int err;
5452
5453 attribute = to_slab_attr(attr);
5454 s = to_slab(kobj);
5455
5456 if (!attribute->store)
5457 return -EIO;
5458
5459 err = attribute->store(s, buf, len);
Johannes Weiner127424c2016-01-20 15:02:32 -08005460#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005461 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08005462 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07005463
Glauber Costa107dab52012-12-18 14:23:05 -08005464 mutex_lock(&slab_mutex);
5465 if (s->max_attr_size < len)
5466 s->max_attr_size = len;
5467
Glauber Costaebe945c2012-12-18 14:23:10 -08005468 /*
5469 * This is a best effort propagation, so this function's return
5470 * value will be determined by the parent cache only. This is
5471 * basically because not all attributes will have a well
5472 * defined semantics for rollbacks - most of the actions will
5473 * have permanent effects.
5474 *
5475 * Returning the error value of any of the children that fail
5476 * is not 100 % defined, in the sense that users seeing the
5477 * error code won't be able to know anything about the state of
5478 * the cache.
5479 *
5480 * Only returning the error code for the parent cache at least
5481 * has well defined semantics. The cache being written to
5482 * directly either failed or succeeded, in which case we loop
5483 * through the descendants with best-effort propagation.
5484 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005485 for_each_memcg_cache(c, s)
5486 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005487 mutex_unlock(&slab_mutex);
5488 }
5489#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005490 return err;
5491}
5492
Glauber Costa107dab52012-12-18 14:23:05 -08005493static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5494{
Johannes Weiner127424c2016-01-20 15:02:32 -08005495#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005496 int i;
5497 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005498 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005499
Vladimir Davydov93030d82014-05-06 12:49:59 -07005500 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005501 return;
5502
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005503 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005504
Glauber Costa107dab52012-12-18 14:23:05 -08005505 /*
5506 * This mean this cache had no attribute written. Therefore, no point
5507 * in copying default values around
5508 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005509 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005510 return;
5511
5512 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5513 char mbuf[64];
5514 char *buf;
5515 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
Thomas Gleixnerc1bb2a82017-06-02 14:46:25 -07005516 ssize_t len;
Glauber Costa107dab52012-12-18 14:23:05 -08005517
5518 if (!attr || !attr->store || !attr->show)
5519 continue;
5520
5521 /*
5522 * It is really bad that we have to allocate here, so we will
5523 * do it only as a fallback. If we actually allocate, though,
5524 * we can just use the allocated buffer until the end.
5525 *
5526 * Most of the slub attributes will tend to be very small in
5527 * size, but sysfs allows buffers up to a page, so they can
5528 * theoretically happen.
5529 */
5530 if (buffer)
5531 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005532 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005533 buf = mbuf;
5534 else {
5535 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5536 if (WARN_ON(!buffer))
5537 continue;
5538 buf = buffer;
5539 }
5540
Thomas Gleixnerc1bb2a82017-06-02 14:46:25 -07005541 len = attr->show(root_cache, buf);
5542 if (len > 0)
5543 attr->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005544 }
5545
5546 if (buffer)
5547 free_page((unsigned long)buffer);
5548#endif
5549}
5550
Christoph Lameter41a21282014-05-06 12:50:08 -07005551static void kmem_cache_release(struct kobject *k)
5552{
5553 slab_kmem_cache_release(to_slab(k));
5554}
5555
Emese Revfy52cf25d2010-01-19 02:58:23 +01005556static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005557 .show = slab_attr_show,
5558 .store = slab_attr_store,
5559};
5560
5561static struct kobj_type slab_ktype = {
5562 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005563 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005564};
5565
5566static int uevent_filter(struct kset *kset, struct kobject *kobj)
5567{
5568 struct kobj_type *ktype = get_ktype(kobj);
5569
5570 if (ktype == &slab_ktype)
5571 return 1;
5572 return 0;
5573}
5574
Emese Revfy9cd43612009-12-31 14:52:51 +01005575static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005576 .filter = uevent_filter,
5577};
5578
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005579static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005580
Vladimir Davydov9a417072014-04-07 15:39:31 -07005581static inline struct kset *cache_kset(struct kmem_cache *s)
5582{
Johannes Weiner127424c2016-01-20 15:02:32 -08005583#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005584 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005585 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005586#endif
5587 return slab_kset;
5588}
5589
Christoph Lameter81819f02007-05-06 14:49:36 -07005590#define ID_STR_LENGTH 64
5591
5592/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005593 *
5594 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005595 */
5596static char *create_unique_id(struct kmem_cache *s)
5597{
5598 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5599 char *p = name;
5600
5601 BUG_ON(!name);
5602
5603 *p++ = ':';
5604 /*
5605 * First flags affecting slabcache operations. We will only
5606 * get here for aliasable slabs so we do not need to support
5607 * too many flags. The flags here must cover all flags that
5608 * are matched during merging to guarantee that the id is
5609 * unique.
5610 */
5611 if (s->flags & SLAB_CACHE_DMA)
5612 *p++ = 'd';
5613 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5614 *p++ = 'a';
Laura Abbottbecfda62016-03-15 14:55:06 -07005615 if (s->flags & SLAB_CONSISTENCY_CHECKS)
Christoph Lameter81819f02007-05-06 14:49:36 -07005616 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005617 if (!(s->flags & SLAB_NOTRACK))
5618 *p++ = 't';
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08005619 if (s->flags & SLAB_ACCOUNT)
5620 *p++ = 'A';
Christoph Lameter81819f02007-05-06 14:49:36 -07005621 if (p != name + 1)
5622 *p++ = '-';
5623 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005624
Christoph Lameter81819f02007-05-06 14:49:36 -07005625 BUG_ON(p > name + ID_STR_LENGTH - 1);
5626 return name;
5627}
5628
5629static int sysfs_slab_add(struct kmem_cache *s)
5630{
5631 int err;
5632 const char *name;
Tejun Heoa4ffb672018-08-24 13:22:21 +09005633 struct kset *kset = cache_kset(s);
Christoph Lameter45530c42012-11-28 16:23:07 +00005634 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005635
Tejun Heoa4ffb672018-08-24 13:22:21 +09005636 if (!kset) {
5637 kobject_init(&s->kobj, &slab_ktype);
5638 return 0;
5639 }
5640
Christoph Lameter81819f02007-05-06 14:49:36 -07005641 if (unmergeable) {
5642 /*
5643 * Slabcache can never be merged so we can use the name proper.
5644 * This is typically the case for debug situations. In that
5645 * case we can catch duplicate names easily.
5646 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005647 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005648 name = s->name;
5649 } else {
5650 /*
5651 * Create a unique name for the slab as a target
5652 * for the symlinks.
5653 */
5654 name = create_unique_id(s);
5655 }
5656
Tejun Heoa4ffb672018-08-24 13:22:21 +09005657 s->kobj.kset = kset;
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005658 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005659 if (err)
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005660 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005661
5662 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005663 if (err)
5664 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005665
Johannes Weiner127424c2016-01-20 15:02:32 -08005666#ifdef CONFIG_MEMCG
Tejun Heoa4ffb672018-08-24 13:22:21 +09005667 if (is_root_cache(s) && memcg_sysfs_enabled) {
Vladimir Davydov9a417072014-04-07 15:39:31 -07005668 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5669 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005670 err = -ENOMEM;
5671 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005672 }
5673 }
5674#endif
5675
Christoph Lameter81819f02007-05-06 14:49:36 -07005676 kobject_uevent(&s->kobj, KOBJ_ADD);
5677 if (!unmergeable) {
5678 /* Setup first alias */
5679 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005680 }
Dave Jones54b6a732014-04-07 15:39:32 -07005681out:
5682 if (!unmergeable)
5683 kfree(name);
5684 return err;
5685out_del_kobj:
5686 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005687 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005688}
5689
Christoph Lameter41a21282014-05-06 12:50:08 -07005690void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005691{
Christoph Lameter97d06602012-07-06 15:25:11 -05005692 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005693 /*
5694 * Sysfs has not been setup yet so no need to remove the
5695 * cache from sysfs.
5696 */
5697 return;
5698
Johannes Weiner127424c2016-01-20 15:02:32 -08005699#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005700 kset_unregister(s->memcg_kset);
5701#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005702 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5703 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005704 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005705}
5706
5707/*
5708 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005709 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005710 */
5711struct saved_alias {
5712 struct kmem_cache *s;
5713 const char *name;
5714 struct saved_alias *next;
5715};
5716
Adrian Bunk5af328a2007-07-17 04:03:27 -07005717static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005718
5719static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5720{
5721 struct saved_alias *al;
5722
Christoph Lameter97d06602012-07-06 15:25:11 -05005723 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005724 /*
5725 * If we have a leftover link then remove it.
5726 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005727 sysfs_remove_link(&slab_kset->kobj, name);
5728 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005729 }
5730
5731 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5732 if (!al)
5733 return -ENOMEM;
5734
5735 al->s = s;
5736 al->name = name;
5737 al->next = alias_list;
5738 alias_list = al;
5739 return 0;
5740}
5741
5742static int __init slab_sysfs_init(void)
5743{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005744 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005745 int err;
5746
Christoph Lameter18004c52012-07-06 15:25:12 -05005747 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005748
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005749 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005750 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005751 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005752 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005753 return -ENOSYS;
5754 }
5755
Christoph Lameter97d06602012-07-06 15:25:11 -05005756 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005757
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005758 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005759 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005760 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005761 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5762 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005763 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005764
5765 while (alias_list) {
5766 struct saved_alias *al = alias_list;
5767
5768 alias_list = alias_list->next;
5769 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005770 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005771 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5772 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005773 kfree(al);
5774 }
5775
Christoph Lameter18004c52012-07-06 15:25:12 -05005776 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005777 resiliency_test();
5778 return 0;
5779}
5780
5781__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005782#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005783
5784/*
5785 * The /proc/slabinfo ABI
5786 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005787#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005788void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005789{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005790 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005791 unsigned long nr_objs = 0;
5792 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005793 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005794 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005795
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005796 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005797 nr_slabs += node_nr_slabs(n);
5798 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005799 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005800 }
5801
Glauber Costa0d7561c2012-10-19 18:20:27 +04005802 sinfo->active_objs = nr_objs - nr_free;
5803 sinfo->num_objs = nr_objs;
5804 sinfo->active_slabs = nr_slabs;
5805 sinfo->num_slabs = nr_slabs;
5806 sinfo->objects_per_slab = oo_objects(s->oo);
5807 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005808}
5809
Glauber Costa0d7561c2012-10-19 18:20:27 +04005810void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005811{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005812}
5813
Glauber Costab7454ad2012-10-19 18:20:25 +04005814ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5815 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005816{
Glauber Costab7454ad2012-10-19 18:20:25 +04005817 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005818}
Linus Torvalds158a9622008-01-02 13:04:48 -08005819#endif /* CONFIG_SLABINFO */