blob: dfc9b4267603eb5e3f40add092cf5e4757b0fc69 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Christoph Lameter81819f02007-05-06 14:49:36 -07002/*
3 * SLUB: A slab allocator that limits cache line use instead of queuing
4 * objects in per cpu and per node lists.
5 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05006 * The allocator synchronizes using per slab locks or atomic operatios
7 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07008 *
Christoph Lametercde53532008-07-04 09:59:22 -07009 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -050010 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070011 */
12
13#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100014#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070015#include <linux/module.h>
16#include <linux/bit_spinlock.h>
17#include <linux/interrupt.h>
18#include <linux/bitops.h>
19#include <linux/slab.h>
Christoph Lameter97d06602012-07-06 15:25:11 -050020#include "slab.h"
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040021#include <linux/proc_fs.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>
Christoph Lameter81819f02007-05-06 14:49:36 -070024#include <linux/cpu.h>
25#include <linux/cpuset.h>
26#include <linux/mempolicy.h>
27#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070028#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070029#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070030#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070031#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090032#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030033#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060034#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080035#include <linux/memcontrol.h>
Kees Cook2482dde2017-09-06 16:19:18 -070036#include <linux/random.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
Matthew Wilcoxb7ccc7f2018-06-07 17:08:46 -070054 * have the ability to do a cmpxchg_double. It only protects:
Christoph Lameter881db7f2011-06-01 12:25:53 -050055 * A. page->freelist -> List of object free in a page
Matthew Wilcoxb7ccc7f2018-06-07 17:08:46 -070056 * B. page->inuse -> Number of objects in use
57 * C. page->objects -> Number of objects in page
58 * D. page->frozen -> frozen state
Christoph Lameter881db7f2011-06-01 12:25:53 -050059 *
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 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800194/* Poison object */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800195#define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800196/* Use cmpxchg_double */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800197#define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
Christoph Lameter81819f02007-05-06 14:49:36 -0700198
Christoph Lameter02cbc872007-05-09 02:32:43 -0700199/*
200 * Tracking user of a slab.
201 */
Ben Greeard6543e32011-07-07 11:36:36 -0700202#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700203struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300204 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700205#ifdef CONFIG_STACKTRACE
206 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
207#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700208 int cpu; /* Was running on cpu */
209 int pid; /* Pid context */
210 unsigned long when; /* When did the operation occur */
211};
212
213enum track_item { TRACK_ALLOC, TRACK_FREE };
214
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500215#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700216static int sysfs_slab_add(struct kmem_cache *);
217static int sysfs_slab_alias(struct kmem_cache *, const char *);
Glauber Costa107dab52012-12-18 14:23:05 -0800218static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Tejun Heobf5eb3d2017-02-22 15:41:11 -0800219static void sysfs_slab_remove(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700220#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700221static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
222static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
223 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800224static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Tejun Heobf5eb3d2017-02-22 15:41:11 -0800225static inline void sysfs_slab_remove(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700226#endif
227
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500228static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800229{
230#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700231 /*
232 * The rmw is racy on a preemptible kernel but this is acceptable, so
233 * avoid this_cpu_add()'s irq-disable overhead.
234 */
235 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800236#endif
237}
238
Christoph Lameter81819f02007-05-06 14:49:36 -0700239/********************************************************************
240 * Core slab cache functions
241 *******************************************************************/
242
Kees Cook2482dde2017-09-06 16:19:18 -0700243/*
244 * Returns freelist pointer (ptr). With hardening, this is obfuscated
245 * with an XOR of the address where the pointer is held and a per-cache
246 * random number.
247 */
248static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
249 unsigned long ptr_addr)
250{
251#ifdef CONFIG_SLAB_FREELIST_HARDENED
Kees Cook9ee0e502020-04-01 21:04:23 -0700252 return (void *)((unsigned long)ptr ^ s->random ^ swab(ptr_addr));
Kees Cook2482dde2017-09-06 16:19:18 -0700253#else
254 return ptr;
255#endif
256}
257
258/* Returns the freelist pointer recorded at location ptr_addr. */
259static inline void *freelist_dereference(const struct kmem_cache *s,
260 void *ptr_addr)
261{
262 return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
263 (unsigned long)ptr_addr);
264}
265
Christoph Lameter7656c722007-05-09 02:32:40 -0700266static inline void *get_freepointer(struct kmem_cache *s, void *object)
267{
Kees Cook2482dde2017-09-06 16:19:18 -0700268 return freelist_dereference(s, object + s->offset);
Christoph Lameter7656c722007-05-09 02:32:40 -0700269}
270
Eric Dumazet0ad95002011-12-16 16:25:34 +0100271static void prefetch_freepointer(const struct kmem_cache *s, void *object)
272{
Vlastimil Babka0882ff92018-08-17 15:44:44 -0700273 prefetch(object + s->offset);
Eric Dumazet0ad95002011-12-16 16:25:34 +0100274}
275
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500276static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
277{
Kees Cook2482dde2017-09-06 16:19:18 -0700278 unsigned long freepointer_addr;
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500279 void *p;
280
Joonsoo Kim922d5662016-03-17 14:17:53 -0700281 if (!debug_pagealloc_enabled())
282 return get_freepointer(s, object);
283
Kees Cook2482dde2017-09-06 16:19:18 -0700284 freepointer_addr = (unsigned long)object + s->offset;
285 probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p));
286 return freelist_ptr(s, p, freepointer_addr);
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500287}
288
Christoph Lameter7656c722007-05-09 02:32:40 -0700289static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
290{
Kees Cook2482dde2017-09-06 16:19:18 -0700291 unsigned long freeptr_addr = (unsigned long)object + s->offset;
292
Alexander Popovce6fa912017-09-06 16:19:22 -0700293#ifdef CONFIG_SLAB_FREELIST_HARDENED
294 BUG_ON(object == fp); /* naive detection of double free or corruption */
295#endif
296
Kees Cook2482dde2017-09-06 16:19:18 -0700297 *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
Christoph Lameter7656c722007-05-09 02:32:40 -0700298}
299
300/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300301#define for_each_object(__p, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700302 for (__p = fixup_red_left(__s, __addr); \
303 __p < (__addr) + (__objects) * (__s)->size; \
304 __p += (__s)->size)
Christoph Lameter7656c722007-05-09 02:32:40 -0700305
Wei Yang54266642014-08-06 16:04:42 -0700306#define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700307 for (__p = fixup_red_left(__s, __addr), __idx = 1; \
308 __idx <= __objects; \
309 __p += (__s)->size, __idx++)
Wei Yang54266642014-08-06 16:04:42 -0700310
Christoph Lameter7656c722007-05-09 02:32:40 -0700311/* Determine object index from a given position */
Alexey Dobriyan284b50d2018-04-05 16:21:35 -0700312static inline unsigned int slab_index(void *p, struct kmem_cache *s, void *addr)
Christoph Lameter7656c722007-05-09 02:32:40 -0700313{
314 return (p - addr) / s->size;
315}
316
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700317static inline unsigned int order_objects(unsigned int order, unsigned int size)
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800318{
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700319 return ((unsigned int)PAGE_SIZE << order) / size;
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800320}
321
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700322static inline struct kmem_cache_order_objects oo_make(unsigned int order,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700323 unsigned int size)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300324{
325 struct kmem_cache_order_objects x = {
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700326 (order << OO_SHIFT) + order_objects(order, size)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300327 };
328
329 return x;
330}
331
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700332static inline unsigned int oo_order(struct kmem_cache_order_objects x)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300333{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400334 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300335}
336
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700337static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300338{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400339 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300340}
341
Christoph Lameter881db7f2011-06-01 12:25:53 -0500342/*
343 * Per slab locking using the pagelock
344 */
345static __always_inline void slab_lock(struct page *page)
346{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800347 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500348 bit_spin_lock(PG_locked, &page->flags);
349}
350
351static __always_inline void slab_unlock(struct page *page)
352{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800353 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500354 __bit_spin_unlock(PG_locked, &page->flags);
355}
356
Christoph Lameter1d071712011-07-14 12:49:12 -0500357/* Interrupts must be disabled (for the fallback code to work right) */
358static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500359 void *freelist_old, unsigned long counters_old,
360 void *freelist_new, unsigned long counters_new,
361 const char *n)
362{
Christoph Lameter1d071712011-07-14 12:49:12 -0500363 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800364#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
365 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500366 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000367 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700368 freelist_old, counters_old,
369 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700370 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500371 } else
372#endif
373 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500374 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800375 if (page->freelist == freelist_old &&
376 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500377 page->freelist = freelist_new;
Matthew Wilcox7d27a042018-06-07 17:08:31 -0700378 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500379 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700380 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500381 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500382 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500383 }
384
385 cpu_relax();
386 stat(s, CMPXCHG_DOUBLE_FAIL);
387
388#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700389 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500390#endif
391
Joe Perches6f6528a2015-04-14 15:44:31 -0700392 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500393}
394
Christoph Lameter1d071712011-07-14 12:49:12 -0500395static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
396 void *freelist_old, unsigned long counters_old,
397 void *freelist_new, unsigned long counters_new,
398 const char *n)
399{
Heiko Carstens25654092012-01-12 17:17:33 -0800400#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
401 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500402 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000403 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700404 freelist_old, counters_old,
405 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700406 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500407 } else
408#endif
409 {
410 unsigned long flags;
411
412 local_irq_save(flags);
413 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800414 if (page->freelist == freelist_old &&
415 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500416 page->freelist = freelist_new;
Matthew Wilcox7d27a042018-06-07 17:08:31 -0700417 page->counters = counters_new;
Christoph Lameter1d071712011-07-14 12:49:12 -0500418 slab_unlock(page);
419 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700420 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500421 }
422 slab_unlock(page);
423 local_irq_restore(flags);
424 }
425
426 cpu_relax();
427 stat(s, CMPXCHG_DOUBLE_FAIL);
428
429#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700430 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500431#endif
432
Joe Perches6f6528a2015-04-14 15:44:31 -0700433 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500434}
435
Christoph Lameter41ecc552007-05-09 02:32:44 -0700436#ifdef CONFIG_SLUB_DEBUG
437/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500438 * Determine a map of object in use on a page.
439 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500440 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500441 * not vanish from under us.
442 */
443static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
444{
445 void *p;
446 void *addr = page_address(page);
447
448 for (p = page->freelist; p; p = get_freepointer(s, p))
449 set_bit(slab_index(p, s, addr), map);
450}
451
Alexey Dobriyan870b1fb2018-04-05 16:21:43 -0700452static inline unsigned int size_from_object(struct kmem_cache *s)
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700453{
454 if (s->flags & SLAB_RED_ZONE)
455 return s->size - s->red_left_pad;
456
457 return s->size;
458}
459
460static inline void *restore_red_left(struct kmem_cache *s, void *p)
461{
462 if (s->flags & SLAB_RED_ZONE)
463 p -= s->red_left_pad;
464
465 return p;
466}
467
Christoph Lameter41ecc552007-05-09 02:32:44 -0700468/*
469 * Debug settings:
470 */
Andrey Ryabinin89d3c872015-11-05 18:51:23 -0800471#if defined(CONFIG_SLUB_DEBUG_ON)
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800472static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700473#else
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800474static slab_flags_t slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700475#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700476
477static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700478static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700479
Christoph Lameter7656c722007-05-09 02:32:40 -0700480/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800481 * slub is about to manipulate internal object metadata. This memory lies
482 * outside the range of the allocated object, so accessing it would normally
483 * be reported by kasan as a bounds error. metadata_access_enable() is used
484 * to tell kasan that these accesses are OK.
485 */
486static inline void metadata_access_enable(void)
487{
488 kasan_disable_current();
489}
490
491static inline void metadata_access_disable(void)
492{
493 kasan_enable_current();
494}
495
496/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700497 * Object debugging
498 */
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700499
500/* Verify that a pointer has an address that is valid within a slab page */
501static inline int check_valid_pointer(struct kmem_cache *s,
502 struct page *page, void *object)
503{
504 void *base;
505
506 if (!object)
507 return 1;
508
509 base = page_address(page);
510 object = restore_red_left(s, object);
511 if (object < base || object >= base + page->objects * s->size ||
512 (object - base) % s->size) {
513 return 0;
514 }
515
516 return 1;
517}
518
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800519static void print_section(char *level, char *text, u8 *addr,
520 unsigned int length)
Christoph Lameter81819f02007-05-06 14:49:36 -0700521{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800522 metadata_access_enable();
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800523 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200524 length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800525 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700526}
527
Christoph Lameter81819f02007-05-06 14:49:36 -0700528static struct track *get_track(struct kmem_cache *s, void *object,
529 enum track_item alloc)
530{
531 struct track *p;
532
533 if (s->offset)
534 p = object + s->offset + sizeof(void *);
535 else
536 p = object + s->inuse;
537
538 return p + alloc;
539}
540
541static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300542 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700543{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900544 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700545
Christoph Lameter81819f02007-05-06 14:49:36 -0700546 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700547#ifdef CONFIG_STACKTRACE
548 struct stack_trace trace;
549 int i;
550
551 trace.nr_entries = 0;
552 trace.max_entries = TRACK_ADDRS_COUNT;
553 trace.entries = p->addrs;
554 trace.skip = 3;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800555 metadata_access_enable();
Ben Greeard6543e32011-07-07 11:36:36 -0700556 save_stack_trace(&trace);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800557 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700558
559 /* See rant in lockdep.c */
560 if (trace.nr_entries != 0 &&
561 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
562 trace.nr_entries--;
563
564 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
565 p->addrs[i] = 0;
566#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700567 p->addr = addr;
568 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400569 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700570 p->when = jiffies;
571 } else
572 memset(p, 0, sizeof(struct track));
573}
574
Christoph Lameter81819f02007-05-06 14:49:36 -0700575static void init_tracking(struct kmem_cache *s, void *object)
576{
Christoph Lameter24922682007-07-17 04:03:18 -0700577 if (!(s->flags & SLAB_STORE_USER))
578 return;
579
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300580 set_track(s, object, TRACK_FREE, 0UL);
581 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700582}
583
Chintan Pandya86609d32018-04-05 16:20:15 -0700584static void print_track(const char *s, struct track *t, unsigned long pr_time)
Christoph Lameter81819f02007-05-06 14:49:36 -0700585{
586 if (!t->addr)
587 return;
588
Fabian Frederickf9f58282014-06-04 16:06:34 -0700589 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Chintan Pandya86609d32018-04-05 16:20:15 -0700590 s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700591#ifdef CONFIG_STACKTRACE
592 {
593 int i;
594 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
595 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700596 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700597 else
598 break;
599 }
600#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700601}
602
Christoph Lameter24922682007-07-17 04:03:18 -0700603static void print_tracking(struct kmem_cache *s, void *object)
604{
Chintan Pandya86609d32018-04-05 16:20:15 -0700605 unsigned long pr_time = jiffies;
Christoph Lameter24922682007-07-17 04:03:18 -0700606 if (!(s->flags & SLAB_STORE_USER))
607 return;
608
Chintan Pandya86609d32018-04-05 16:20:15 -0700609 print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
610 print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
Christoph Lameter24922682007-07-17 04:03:18 -0700611}
612
613static void print_page_info(struct page *page)
614{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700615 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800616 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700617
618}
619
620static void slab_bug(struct kmem_cache *s, char *fmt, ...)
621{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700622 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700623 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700624
625 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700626 vaf.fmt = fmt;
627 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700628 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700629 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700630 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400631
Rusty Russell373d4d02013-01-21 17:17:39 +1030632 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700633 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700634}
635
636static void slab_fix(struct kmem_cache *s, char *fmt, ...)
637{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700638 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700639 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700640
641 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700642 vaf.fmt = fmt;
643 vaf.va = &args;
644 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700645 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700646}
647
Dongli Zhang6c097552020-06-01 21:45:47 -0700648static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
Eugeniu Roscaaf2cf2c2020-09-04 16:35:30 -0700649 void **freelist, void *nextfree)
Dongli Zhang6c097552020-06-01 21:45:47 -0700650{
651 if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
Eugeniu Roscaaf2cf2c2020-09-04 16:35:30 -0700652 !check_valid_pointer(s, page, nextfree) && freelist) {
653 object_err(s, page, *freelist, "Freechain corrupt");
654 *freelist = NULL;
Dongli Zhang6c097552020-06-01 21:45:47 -0700655 slab_fix(s, "Isolate corrupted freechain");
656 return true;
657 }
658
659 return false;
660}
661
Christoph Lameter24922682007-07-17 04:03:18 -0700662static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700663{
664 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800665 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700666
667 print_tracking(s, p);
668
669 print_page_info(page);
670
Fabian Frederickf9f58282014-06-04 16:06:34 -0700671 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
672 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700673
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700674 if (s->flags & SLAB_RED_ZONE)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800675 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
676 s->red_left_pad);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700677 else if (p > addr + 16)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800678 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700679
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800680 print_section(KERN_ERR, "Object ", p,
Alexey Dobriyan1b473f22018-04-05 16:21:17 -0700681 min_t(unsigned int, s->object_size, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700682 if (s->flags & SLAB_RED_ZONE)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800683 print_section(KERN_ERR, "Redzone ", p + s->object_size,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500684 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700685
Christoph Lameter81819f02007-05-06 14:49:36 -0700686 if (s->offset)
687 off = s->offset + sizeof(void *);
688 else
689 off = s->inuse;
690
Christoph Lameter24922682007-07-17 04:03:18 -0700691 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700692 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700693
Alexander Potapenko80a92012016-07-28 15:49:07 -0700694 off += kasan_metadata_size(s);
695
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700696 if (off != size_from_object(s))
Christoph Lameter81819f02007-05-06 14:49:36 -0700697 /* Beginning of the filler is the free pointer */
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800698 print_section(KERN_ERR, "Padding ", p + off,
699 size_from_object(s) - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700700
701 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700702}
703
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800704void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700705 u8 *object, char *reason)
706{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700707 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700708 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700709}
710
Mathieu Malaterrea38965b2018-06-07 17:05:17 -0700711static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800712 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700713{
714 va_list args;
715 char buf[100];
716
Christoph Lameter24922682007-07-17 04:03:18 -0700717 va_start(args, fmt);
718 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700719 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700720 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700721 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700722 dump_stack();
723}
724
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500725static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700726{
727 u8 *p = object;
728
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700729 if (s->flags & SLAB_RED_ZONE)
730 memset(p - s->red_left_pad, val, s->red_left_pad);
731
Christoph Lameter81819f02007-05-06 14:49:36 -0700732 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500733 memset(p, POISON_FREE, s->object_size - 1);
734 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700735 }
736
737 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500738 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700739}
740
Christoph Lameter24922682007-07-17 04:03:18 -0700741static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
742 void *from, void *to)
743{
744 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
745 memset(from, data, to - from);
746}
747
748static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
749 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800750 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700751{
752 u8 *fault;
753 u8 *end;
754
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800755 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700756 fault = memchr_inv(start, value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800757 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700758 if (!fault)
759 return 1;
760
761 end = start + bytes;
762 while (end > fault && end[-1] == value)
763 end--;
764
765 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700766 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700767 fault, end - 1, fault[0], value);
768 print_trailer(s, page, object);
769
770 restore_bytes(s, what, value, fault, end);
771 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700772}
773
Christoph Lameter81819f02007-05-06 14:49:36 -0700774/*
775 * Object layout:
776 *
777 * object address
778 * Bytes of the object to be managed.
779 * If the freepointer may overlay the object then the free
780 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700781 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700782 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
783 * 0xa5 (POISON_END)
784 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500785 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700786 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700787 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500788 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700789 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700790 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
791 * 0xcc (RED_ACTIVE) for objects in use.
792 *
793 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700794 * Meta data starts here.
795 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700796 * A. Free pointer (if we cannot overwrite object on free)
797 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700798 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800799 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700800 * before the word boundary.
801 *
802 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700803 *
804 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700805 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700806 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500807 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700808 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700809 * may be used with merged slabcaches.
810 */
811
Christoph Lameter81819f02007-05-06 14:49:36 -0700812static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
813{
814 unsigned long off = s->inuse; /* The end of info */
815
816 if (s->offset)
817 /* Freepointer is placed after the object. */
818 off += sizeof(void *);
819
820 if (s->flags & SLAB_STORE_USER)
821 /* We also have user information there */
822 off += 2 * sizeof(struct track);
823
Alexander Potapenko80a92012016-07-28 15:49:07 -0700824 off += kasan_metadata_size(s);
825
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700826 if (size_from_object(s) == off)
Christoph Lameter81819f02007-05-06 14:49:36 -0700827 return 1;
828
Christoph Lameter24922682007-07-17 04:03:18 -0700829 return check_bytes_and_report(s, page, p, "Object padding",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700830 p + off, POISON_INUSE, size_from_object(s) - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700831}
832
Christoph Lameter39b26462008-04-14 19:11:30 +0300833/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700834static int slab_pad_check(struct kmem_cache *s, struct page *page)
835{
Christoph Lameter24922682007-07-17 04:03:18 -0700836 u8 *start;
837 u8 *fault;
838 u8 *end;
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800839 u8 *pad;
Christoph Lameter24922682007-07-17 04:03:18 -0700840 int length;
841 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700842
843 if (!(s->flags & SLAB_POISON))
844 return 1;
845
Christoph Lametera973e9d2008-03-01 13:40:44 -0800846 start = page_address(page);
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700847 length = PAGE_SIZE << compound_order(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300848 end = start + length;
849 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700850 if (!remainder)
851 return 1;
852
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800853 pad = end - remainder;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800854 metadata_access_enable();
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800855 fault = memchr_inv(pad, POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800856 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700857 if (!fault)
858 return 1;
859 while (end > fault && end[-1] == POISON_INUSE)
860 end--;
861
862 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800863 print_section(KERN_ERR, "Padding ", pad, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700864
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800865 restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700866 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700867}
868
869static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500870 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700871{
872 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500873 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700874
875 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700876 if (!check_bytes_and_report(s, page, object, "Redzone",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700877 object - s->red_left_pad, val, s->red_left_pad))
878 return 0;
879
880 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500881 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700882 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700883 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500884 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800885 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800886 endobject, POISON_INUSE,
887 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800888 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700889 }
890
891 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500892 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700893 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500894 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700895 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500896 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700897 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700898 /*
899 * check_pad_bytes cleans up on its own.
900 */
901 check_pad_bytes(s, page, p);
902 }
903
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500904 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700905 /*
906 * Object and freepointer overlap. Cannot check
907 * freepointer while object is allocated.
908 */
909 return 1;
910
911 /* Check free pointer validity */
912 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
913 object_err(s, page, p, "Freepointer corrupt");
914 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100915 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700916 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700917 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700918 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800919 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700920 return 0;
921 }
922 return 1;
923}
924
925static int check_slab(struct kmem_cache *s, struct page *page)
926{
Christoph Lameter39b26462008-04-14 19:11:30 +0300927 int maxobj;
928
Christoph Lameter81819f02007-05-06 14:49:36 -0700929 VM_BUG_ON(!irqs_disabled());
930
931 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700932 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700933 return 0;
934 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300935
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700936 maxobj = order_objects(compound_order(page), s->size);
Christoph Lameter39b26462008-04-14 19:11:30 +0300937 if (page->objects > maxobj) {
938 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800939 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +0300940 return 0;
941 }
942 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700943 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800944 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700945 return 0;
946 }
947 /* Slab_pad_check fixes things up after itself */
948 slab_pad_check(s, page);
949 return 1;
950}
951
952/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700953 * Determine if a certain object on a page is on the freelist. Must hold the
954 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700955 */
956static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
957{
958 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500959 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700960 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800961 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700962
Christoph Lameter881db7f2011-06-01 12:25:53 -0500963 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300964 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700965 if (fp == search)
966 return 1;
967 if (!check_valid_pointer(s, page, fp)) {
968 if (object) {
969 object_err(s, page, object,
970 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800971 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700972 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700973 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800974 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300975 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700976 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700977 return 0;
978 }
979 break;
980 }
981 object = fp;
982 fp = get_freepointer(s, object);
983 nr++;
984 }
985
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700986 max_objects = order_objects(compound_order(page), s->size);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400987 if (max_objects > MAX_OBJS_PER_PAGE)
988 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300989
990 if (page->objects != max_objects) {
Joe Perches756a0252016-03-17 14:19:47 -0700991 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
992 page->objects, max_objects);
Christoph Lameter224a88b2008-04-14 19:11:31 +0300993 page->objects = max_objects;
994 slab_fix(s, "Number of objects adjusted.");
995 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300996 if (page->inuse != page->objects - nr) {
Joe Perches756a0252016-03-17 14:19:47 -0700997 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
998 page->inuse, page->objects - nr);
Christoph Lameter39b26462008-04-14 19:11:30 +0300999 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -07001000 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001001 }
1002 return search == NULL;
1003}
1004
Christoph Lameter0121c6192008-04-29 16:11:12 -07001005static void trace(struct kmem_cache *s, struct page *page, void *object,
1006 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -07001007{
1008 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07001009 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -07001010 s->name,
1011 alloc ? "alloc" : "free",
1012 object, page->inuse,
1013 page->freelist);
1014
1015 if (!alloc)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -08001016 print_section(KERN_INFO, "Object ", (void *)object,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001017 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001018
1019 dump_stack();
1020 }
1021}
1022
Christoph Lameter643b1132007-05-06 14:49:42 -07001023/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001024 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -07001025 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001026static void add_full(struct kmem_cache *s,
1027 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -07001028{
Christoph Lameter643b1132007-05-06 14:49:42 -07001029 if (!(s->flags & SLAB_STORE_USER))
1030 return;
1031
David Rientjes255d0882014-02-10 14:25:39 -08001032 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001033 list_add(&page->lru, &n->full);
1034}
Christoph Lameter643b1132007-05-06 14:49:42 -07001035
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001036static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001037{
1038 if (!(s->flags & SLAB_STORE_USER))
1039 return;
1040
David Rientjes255d0882014-02-10 14:25:39 -08001041 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -07001042 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001043}
1044
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001045/* Tracking of the number of slabs for debugging purposes */
1046static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1047{
1048 struct kmem_cache_node *n = get_node(s, node);
1049
1050 return atomic_long_read(&n->nr_slabs);
1051}
1052
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001053static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1054{
1055 return atomic_long_read(&n->nr_slabs);
1056}
1057
Christoph Lameter205ab992008-04-14 19:11:40 +03001058static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001059{
1060 struct kmem_cache_node *n = get_node(s, node);
1061
1062 /*
1063 * May be called early in order to allocate a slab for the
1064 * kmem_cache_node structure. Solve the chicken-egg
1065 * dilemma by deferring the increment of the count during
1066 * bootstrap (see early_kmem_cache_node_alloc).
1067 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001068 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001069 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001070 atomic_long_add(objects, &n->total_objects);
1071 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001072}
Christoph Lameter205ab992008-04-14 19:11:40 +03001073static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001074{
1075 struct kmem_cache_node *n = get_node(s, node);
1076
1077 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001078 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001079}
1080
1081/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001082static void setup_object_debug(struct kmem_cache *s, struct page *page,
1083 void *object)
1084{
1085 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1086 return;
1087
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001088 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001089 init_tracking(s, object);
1090}
1091
Laura Abbottbecfda62016-03-15 14:55:06 -07001092static inline int alloc_consistency_checks(struct kmem_cache *s,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001093 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001094 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001095{
1096 if (!check_slab(s, page))
Laura Abbottbecfda62016-03-15 14:55:06 -07001097 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001098
Christoph Lameter81819f02007-05-06 14:49:36 -07001099 if (!check_valid_pointer(s, page, object)) {
1100 object_err(s, page, object, "Freelist Pointer check fails");
Laura Abbottbecfda62016-03-15 14:55:06 -07001101 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001102 }
1103
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001104 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Laura Abbottbecfda62016-03-15 14:55:06 -07001105 return 0;
1106
1107 return 1;
1108}
1109
1110static noinline int alloc_debug_processing(struct kmem_cache *s,
1111 struct page *page,
1112 void *object, unsigned long addr)
1113{
1114 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1115 if (!alloc_consistency_checks(s, page, object, addr))
1116 goto bad;
1117 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001118
Christoph Lameter3ec09742007-05-16 22:11:00 -07001119 /* Success perform special debug activities for allocs */
1120 if (s->flags & SLAB_STORE_USER)
1121 set_track(s, object, TRACK_ALLOC, addr);
1122 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001123 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001124 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001125
Christoph Lameter81819f02007-05-06 14:49:36 -07001126bad:
1127 if (PageSlab(page)) {
1128 /*
1129 * If this is a slab page then lets do the best we can
1130 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001131 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001132 */
Christoph Lameter24922682007-07-17 04:03:18 -07001133 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001134 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001135 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001136 }
1137 return 0;
1138}
1139
Laura Abbottbecfda62016-03-15 14:55:06 -07001140static inline int free_consistency_checks(struct kmem_cache *s,
1141 struct page *page, void *object, unsigned long addr)
1142{
1143 if (!check_valid_pointer(s, page, object)) {
1144 slab_err(s, page, "Invalid object pointer 0x%p", object);
1145 return 0;
1146 }
1147
1148 if (on_freelist(s, page, object)) {
1149 object_err(s, page, object, "Object already free");
1150 return 0;
1151 }
1152
1153 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1154 return 0;
1155
1156 if (unlikely(s != page->slab_cache)) {
1157 if (!PageSlab(page)) {
Joe Perches756a0252016-03-17 14:19:47 -07001158 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1159 object);
Laura Abbottbecfda62016-03-15 14:55:06 -07001160 } else if (!page->slab_cache) {
1161 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1162 object);
1163 dump_stack();
1164 } else
1165 object_err(s, page, object,
1166 "page slab pointer corrupt.");
1167 return 0;
1168 }
1169 return 1;
1170}
1171
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001172/* Supports checking bulk free of a constructed freelist */
Laura Abbott282acb42016-03-15 14:54:59 -07001173static noinline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001174 struct kmem_cache *s, struct page *page,
1175 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001176 unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001177{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001178 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001179 void *object = head;
1180 int cnt = 0;
Laura Abbott282acb42016-03-15 14:54:59 -07001181 unsigned long uninitialized_var(flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001182 int ret = 0;
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001183
Laura Abbott282acb42016-03-15 14:54:59 -07001184 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001185 slab_lock(page);
1186
Laura Abbottbecfda62016-03-15 14:55:06 -07001187 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1188 if (!check_slab(s, page))
1189 goto out;
1190 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001191
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001192next_object:
1193 cnt++;
1194
Laura Abbottbecfda62016-03-15 14:55:06 -07001195 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1196 if (!free_consistency_checks(s, page, object, addr))
1197 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001198 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001199
Christoph Lameter3ec09742007-05-16 22:11:00 -07001200 if (s->flags & SLAB_STORE_USER)
1201 set_track(s, object, TRACK_FREE, addr);
1202 trace(s, page, object, 0);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001203 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001204 init_object(s, object, SLUB_RED_INACTIVE);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001205
1206 /* Reached end of constructed freelist yet? */
1207 if (object != tail) {
1208 object = get_freepointer(s, object);
1209 goto next_object;
1210 }
Laura Abbott804aa132016-03-15 14:55:02 -07001211 ret = 1;
1212
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001213out:
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001214 if (cnt != bulk_cnt)
1215 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1216 bulk_cnt, cnt);
1217
Christoph Lameter881db7f2011-06-01 12:25:53 -05001218 slab_unlock(page);
Laura Abbott282acb42016-03-15 14:54:59 -07001219 spin_unlock_irqrestore(&n->list_lock, flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001220 if (!ret)
1221 slab_fix(s, "Object at 0x%p not freed", object);
1222 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001223}
1224
Christoph Lameter41ecc552007-05-09 02:32:44 -07001225static int __init setup_slub_debug(char *str)
1226{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001227 slub_debug = DEBUG_DEFAULT_FLAGS;
1228 if (*str++ != '=' || !*str)
1229 /*
1230 * No options specified. Switch on full debugging.
1231 */
1232 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001233
1234 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001235 /*
1236 * No options but restriction on slabs. This means full
1237 * debugging for slabs matching a pattern.
1238 */
1239 goto check_slabs;
1240
1241 slub_debug = 0;
1242 if (*str == '-')
1243 /*
1244 * Switch off all debugging measures.
1245 */
1246 goto out;
1247
1248 /*
1249 * Determine which debug features should be switched on
1250 */
Pekka Enberg06428782008-01-07 23:20:27 -08001251 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001252 switch (tolower(*str)) {
1253 case 'f':
Laura Abbottbecfda62016-03-15 14:55:06 -07001254 slub_debug |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001255 break;
1256 case 'z':
1257 slub_debug |= SLAB_RED_ZONE;
1258 break;
1259 case 'p':
1260 slub_debug |= SLAB_POISON;
1261 break;
1262 case 'u':
1263 slub_debug |= SLAB_STORE_USER;
1264 break;
1265 case 't':
1266 slub_debug |= SLAB_TRACE;
1267 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001268 case 'a':
1269 slub_debug |= SLAB_FAILSLAB;
1270 break;
Chris J Arges08303a72015-04-14 15:44:25 -07001271 case 'o':
1272 /*
1273 * Avoid enabling debugging on caches if its minimum
1274 * order would increase as a result.
1275 */
1276 disable_higher_order_debug = 1;
1277 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001278 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001279 pr_err("slub_debug option '%c' unknown. skipped\n",
1280 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001281 }
1282 }
1283
1284check_slabs:
1285 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001286 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001287out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001288 return 1;
1289}
1290
1291__setup("slub_debug", setup_slub_debug);
1292
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001293slab_flags_t kmem_cache_flags(unsigned int object_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001294 slab_flags_t flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001295 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001296{
1297 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001298 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001299 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001300 if (slub_debug && (!slub_debug_slabs || (name &&
1301 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001302 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001303
1304 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001305}
Jesper Dangaard Brouerb4a64712015-11-20 15:57:41 -08001306#else /* !CONFIG_SLUB_DEBUG */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001307static inline void setup_object_debug(struct kmem_cache *s,
1308 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001309
Christoph Lameter3ec09742007-05-16 22:11:00 -07001310static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001311 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001312
Laura Abbott282acb42016-03-15 14:54:59 -07001313static inline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001314 struct kmem_cache *s, struct page *page,
1315 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001316 unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001317
Christoph Lameter41ecc552007-05-09 02:32:44 -07001318static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1319 { return 1; }
1320static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001321 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001322static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1323 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001324static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1325 struct page *page) {}
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001326slab_flags_t kmem_cache_flags(unsigned int object_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001327 slab_flags_t flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001328 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001329{
1330 return flags;
1331}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001332#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001333
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001334#define disable_higher_order_debug 0
1335
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001336static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1337 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001338static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1339 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001340static inline void inc_slabs_node(struct kmem_cache *s, int node,
1341 int objects) {}
1342static inline void dec_slabs_node(struct kmem_cache *s, int node,
1343 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001344
Dongli Zhang6c097552020-06-01 21:45:47 -07001345static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
Eugeniu Roscaaf2cf2c2020-09-04 16:35:30 -07001346 void **freelist, void *nextfree)
Dongli Zhang6c097552020-06-01 21:45:47 -07001347{
1348 return false;
1349}
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001350#endif /* CONFIG_SLUB_DEBUG */
1351
1352/*
1353 * Hooks for other subsystems that check memory allocations. In a typical
1354 * production configuration these hooks all should produce no code at all.
1355 */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001356static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1357{
1358 kmemleak_alloc(ptr, size, 1, flags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07001359 kasan_kmalloc_large(ptr, size, flags);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001360}
1361
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08001362static __always_inline void kfree_hook(void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001363{
1364 kmemleak_free(x);
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08001365 kasan_kfree_large(x, _RET_IP_);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001366}
1367
Andrey Konovalovc3895392018-04-10 16:30:31 -07001368static __always_inline bool slab_free_hook(struct kmem_cache *s, void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001369{
1370 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001371
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001372 /*
1373 * Trouble is that we may no longer disable interrupts in the fast path
1374 * So in order to make the debug calls that expect irqs to be
1375 * disabled we need to disable interrupts temporarily.
1376 */
Levin, Alexander (Sasha Levin)4675ff02017-11-15 17:36:02 -08001377#ifdef CONFIG_LOCKDEP
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001378 {
1379 unsigned long flags;
1380
1381 local_irq_save(flags);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001382 debug_check_no_locks_freed(x, s->object_size);
1383 local_irq_restore(flags);
1384 }
1385#endif
1386 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1387 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001388
Andrey Konovalovc3895392018-04-10 16:30:31 -07001389 /* KASAN might put x into memory quarantine, delaying its reuse */
1390 return kasan_slab_free(s, x, _RET_IP_);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001391}
Christoph Lameter205ab992008-04-14 19:11:40 +03001392
Andrey Konovalovc3895392018-04-10 16:30:31 -07001393static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1394 void **head, void **tail)
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001395{
1396/*
1397 * Compiler cannot detect this function can be removed if slab_free_hook()
1398 * evaluates to nothing. Thus, catch all relevant config debug options here.
1399 */
Levin, Alexander (Sasha Levin)4675ff02017-11-15 17:36:02 -08001400#if defined(CONFIG_LOCKDEP) || \
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001401 defined(CONFIG_DEBUG_KMEMLEAK) || \
1402 defined(CONFIG_DEBUG_OBJECTS_FREE) || \
1403 defined(CONFIG_KASAN)
1404
Andrey Konovalovc3895392018-04-10 16:30:31 -07001405 void *object;
1406 void *next = *head;
1407 void *old_tail = *tail ? *tail : *head;
1408
1409 /* Head and tail of the reconstructed freelist */
1410 *head = NULL;
1411 *tail = NULL;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001412
1413 do {
Andrey Konovalovc3895392018-04-10 16:30:31 -07001414 object = next;
1415 next = get_freepointer(s, object);
1416 /* If object's reuse doesn't have to be delayed */
1417 if (!slab_free_hook(s, object)) {
1418 /* Move object to the new freelist */
1419 set_freepointer(s, object, *head);
1420 *head = object;
1421 if (!*tail)
1422 *tail = object;
1423 }
1424 } while (object != old_tail);
1425
1426 if (*head == *tail)
1427 *tail = NULL;
1428
1429 return *head != NULL;
1430#else
1431 return true;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001432#endif
1433}
1434
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001435static void setup_object(struct kmem_cache *s, struct page *page,
1436 void *object)
1437{
1438 setup_object_debug(s, page, object);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07001439 kasan_init_slab_obj(s, object);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001440 if (unlikely(s->ctor)) {
1441 kasan_unpoison_object_data(s, object);
1442 s->ctor(object);
1443 kasan_poison_object_data(s, object);
1444 }
1445}
1446
Christoph Lameter81819f02007-05-06 14:49:36 -07001447/*
1448 * Slab allocation and freeing
1449 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001450static inline struct page *alloc_slab_page(struct kmem_cache *s,
1451 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001452{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001453 struct page *page;
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001454 unsigned int order = oo_order(oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001455
Christoph Lameter2154a332010-07-09 14:07:10 -05001456 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001457 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001458 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001459 page = __alloc_pages_node(node, flags, order);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001460
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001461 if (page && memcg_charge_slab(page, flags, order, s)) {
1462 __free_pages(page, order);
1463 page = NULL;
1464 }
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001465
1466 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001467}
1468
Thomas Garnier210e7a42016-07-26 15:21:59 -07001469#ifdef CONFIG_SLAB_FREELIST_RANDOM
1470/* Pre-initialize the random sequence cache */
1471static int init_cache_random_seq(struct kmem_cache *s)
1472{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001473 unsigned int count = oo_objects(s->oo);
Thomas Garnier210e7a42016-07-26 15:21:59 -07001474 int err;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001475
Sean Reesa8100072017-02-08 14:30:59 -08001476 /* Bailout if already initialised */
1477 if (s->random_seq)
1478 return 0;
1479
Thomas Garnier210e7a42016-07-26 15:21:59 -07001480 err = cache_random_seq_create(s, count, GFP_KERNEL);
1481 if (err) {
1482 pr_err("SLUB: Unable to initialize free list for %s\n",
1483 s->name);
1484 return err;
1485 }
1486
1487 /* Transform to an offset on the set of pages */
1488 if (s->random_seq) {
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001489 unsigned int i;
1490
Thomas Garnier210e7a42016-07-26 15:21:59 -07001491 for (i = 0; i < count; i++)
1492 s->random_seq[i] *= s->size;
1493 }
1494 return 0;
1495}
1496
1497/* Initialize each random sequence freelist per cache */
1498static void __init init_freelist_randomization(void)
1499{
1500 struct kmem_cache *s;
1501
1502 mutex_lock(&slab_mutex);
1503
1504 list_for_each_entry(s, &slab_caches, list)
1505 init_cache_random_seq(s);
1506
1507 mutex_unlock(&slab_mutex);
1508}
1509
1510/* Get the next entry on the pre-computed freelist randomized */
1511static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1512 unsigned long *pos, void *start,
1513 unsigned long page_limit,
1514 unsigned long freelist_count)
1515{
1516 unsigned int idx;
1517
1518 /*
1519 * If the target page allocation failed, the number of objects on the
1520 * page might be smaller than the usual size defined by the cache.
1521 */
1522 do {
1523 idx = s->random_seq[*pos];
1524 *pos += 1;
1525 if (*pos >= freelist_count)
1526 *pos = 0;
1527 } while (unlikely(idx >= page_limit));
1528
1529 return (char *)start + idx;
1530}
1531
1532/* Shuffle the single linked freelist based on a random pre-computed sequence */
1533static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1534{
1535 void *start;
1536 void *cur;
1537 void *next;
1538 unsigned long idx, pos, page_limit, freelist_count;
1539
1540 if (page->objects < 2 || !s->random_seq)
1541 return false;
1542
1543 freelist_count = oo_objects(s->oo);
1544 pos = get_random_int() % freelist_count;
1545
1546 page_limit = page->objects * s->size;
1547 start = fixup_red_left(s, page_address(page));
1548
1549 /* First entry is used as the base of the freelist */
1550 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1551 freelist_count);
1552 page->freelist = cur;
1553
1554 for (idx = 1; idx < page->objects; idx++) {
1555 setup_object(s, page, cur);
1556 next = next_freelist_entry(s, page, &pos, start, page_limit,
1557 freelist_count);
1558 set_freepointer(s, cur, next);
1559 cur = next;
1560 }
1561 setup_object(s, page, cur);
1562 set_freepointer(s, cur, NULL);
1563
1564 return true;
1565}
1566#else
1567static inline int init_cache_random_seq(struct kmem_cache *s)
1568{
1569 return 0;
1570}
1571static inline void init_freelist_randomization(void) { }
1572static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1573{
1574 return false;
1575}
1576#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1577
Christoph Lameter81819f02007-05-06 14:49:36 -07001578static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1579{
Pekka Enberg06428782008-01-07 23:20:27 -08001580 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001581 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001582 gfp_t alloc_gfp;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001583 void *start, *p;
1584 int idx, order;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001585 bool shuffle;
Christoph Lameter81819f02007-05-06 14:49:36 -07001586
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001587 flags &= gfp_allowed_mask;
1588
Mel Gormand0164ad2015-11-06 16:28:21 -08001589 if (gfpflags_allow_blocking(flags))
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001590 local_irq_enable();
1591
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001592 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001593
Pekka Enbergba522702009-06-24 21:59:51 +03001594 /*
1595 * Let the initial higher-order allocation fail under memory pressure
1596 * so we fall-back to the minimum order allocation.
1597 */
1598 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Mel Gormand0164ad2015-11-06 16:28:21 -08001599 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
Mel Gorman444eb2a42016-03-17 14:19:23 -07001600 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
Pekka Enbergba522702009-06-24 21:59:51 +03001601
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001602 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001603 if (unlikely(!page)) {
1604 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001605 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001606 /*
1607 * Allocation may have failed due to fragmentation.
1608 * Try a lower order alloc if possible
1609 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001610 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001611 if (unlikely(!page))
1612 goto out;
1613 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001614 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001615
Christoph Lameter834f3d12008-04-14 19:11:31 +03001616 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001617
Glauber Costa1f458cb2012-12-18 14:22:50 -08001618 order = compound_order(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001619 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001620 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001621 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001622 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001623
1624 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001625
1626 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001627 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001628
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001629 kasan_poison_slab(page);
1630
Thomas Garnier210e7a42016-07-26 15:21:59 -07001631 shuffle = shuffle_freelist(s, page);
1632
1633 if (!shuffle) {
1634 for_each_object_idx(p, idx, s, start, page->objects) {
1635 setup_object(s, page, p);
1636 if (likely(idx < page->objects))
1637 set_freepointer(s, p, p + s->size);
1638 else
1639 set_freepointer(s, p, NULL);
1640 }
1641 page->freelist = fixup_red_left(s, start);
Christoph Lameter81819f02007-05-06 14:49:36 -07001642 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001643
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001644 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001645 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001646
Christoph Lameter81819f02007-05-06 14:49:36 -07001647out:
Mel Gormand0164ad2015-11-06 16:28:21 -08001648 if (gfpflags_allow_blocking(flags))
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001649 local_irq_disable();
1650 if (!page)
1651 return NULL;
1652
Johannes Weiner7779f212017-07-06 15:40:55 -07001653 mod_lruvec_page_state(page,
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001654 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1655 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1656 1 << oo_order(oo));
1657
1658 inc_slabs_node(s, page_to_nid(page), page->objects);
1659
Christoph Lameter81819f02007-05-06 14:49:36 -07001660 return page;
1661}
1662
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001663static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1664{
1665 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07001666 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07001667 flags &= ~GFP_SLAB_BUG_MASK;
1668 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1669 invalid_mask, &invalid_mask, flags, &flags);
Borislav Petkov65b9de72017-02-22 15:41:02 -08001670 dump_stack();
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001671 }
1672
1673 return allocate_slab(s,
1674 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1675}
1676
Christoph Lameter81819f02007-05-06 14:49:36 -07001677static void __free_slab(struct kmem_cache *s, struct page *page)
1678{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001679 int order = compound_order(page);
1680 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001681
Laura Abbottbecfda62016-03-15 14:55:06 -07001682 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001683 void *p;
1684
1685 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001686 for_each_object(p, s, page_address(page),
1687 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001688 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001689 }
1690
Johannes Weiner7779f212017-07-06 15:40:55 -07001691 mod_lruvec_page_state(page,
Christoph Lameter81819f02007-05-06 14:49:36 -07001692 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1693 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001694 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001695
Mel Gorman072bb0a2012-07-31 16:43:58 -07001696 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001697 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001698
Matthew Wilcoxd4fc5062018-06-07 17:08:26 -07001699 page->mapping = NULL;
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001700 if (current->reclaim_state)
1701 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001702 memcg_uncharge_slab(page, order, s);
1703 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001704}
1705
1706static void rcu_free_slab(struct rcu_head *h)
1707{
Matthew Wilcoxbf68c212018-06-07 17:09:05 -07001708 struct page *page = container_of(h, struct page, rcu_head);
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001709
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001710 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001711}
1712
1713static void free_slab(struct kmem_cache *s, struct page *page)
1714{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001715 if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
Matthew Wilcoxbf68c212018-06-07 17:09:05 -07001716 call_rcu(&page->rcu_head, rcu_free_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07001717 } else
1718 __free_slab(s, page);
1719}
1720
1721static void discard_slab(struct kmem_cache *s, struct page *page)
1722{
Christoph Lameter205ab992008-04-14 19:11:40 +03001723 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001724 free_slab(s, page);
1725}
1726
1727/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001728 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001729 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001730static inline void
1731__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001732{
Christoph Lametere95eed52007-05-06 14:49:44 -07001733 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001734 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001735 list_add_tail(&page->lru, &n->partial);
1736 else
1737 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001738}
1739
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001740static inline void add_partial(struct kmem_cache_node *n,
1741 struct page *page, int tail)
1742{
1743 lockdep_assert_held(&n->list_lock);
1744 __add_partial(n, page, tail);
1745}
1746
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001747static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001748 struct page *page)
1749{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001750 lockdep_assert_held(&n->list_lock);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08001751 list_del(&page->lru);
1752 n->nr_partial--;
Christoph Lameter62e346a2010-09-28 08:10:28 -05001753}
1754
Christoph Lameter81819f02007-05-06 14:49:36 -07001755/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001756 * Remove slab from the partial list, freeze it and
1757 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001758 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001759 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001760 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001761static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001762 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001763 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001764{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001765 void *freelist;
1766 unsigned long counters;
1767 struct page new;
1768
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001769 lockdep_assert_held(&n->list_lock);
1770
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001771 /*
1772 * Zap the freelist and set the frozen bit.
1773 * The old freelist is the list of objects for the
1774 * per cpu allocation list.
1775 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001776 freelist = page->freelist;
1777 counters = page->counters;
1778 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001779 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001780 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001781 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001782 new.freelist = NULL;
1783 } else {
1784 new.freelist = freelist;
1785 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001786
Dave Hansena0132ac2014-01-29 14:05:50 -08001787 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001788 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001789
Christoph Lameter7ced3712012-05-09 10:09:53 -05001790 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001791 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001792 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001793 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001794 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001795
1796 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001797 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001798 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001799}
1800
Joonsoo Kim633b0762013-01-21 17:01:25 +09001801static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001802static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001803
Christoph Lameter81819f02007-05-06 14:49:36 -07001804/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001805 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001806 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001807static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1808 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001809{
Christoph Lameter49e22582011-08-09 16:12:27 -05001810 struct page *page, *page2;
1811 void *object = NULL;
Alexey Dobriyane5d99982018-04-05 16:21:10 -07001812 unsigned int available = 0;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001813 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001814
1815 /*
1816 * Racy check. If we mistakenly see no partial slabs then we
1817 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001818 * partial slab and there is none available then get_partials()
1819 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001820 */
1821 if (!n || !n->nr_partial)
1822 return NULL;
1823
1824 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001825 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001826 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001827
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001828 if (!pfmemalloc_match(page, flags))
1829 continue;
1830
Joonsoo Kim633b0762013-01-21 17:01:25 +09001831 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001832 if (!t)
1833 break;
1834
Joonsoo Kim633b0762013-01-21 17:01:25 +09001835 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001836 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001837 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001838 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001839 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001840 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001841 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001842 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001843 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001844 if (!kmem_cache_has_cpu_partial(s)
Wei Yange6d0e1d2017-07-06 15:36:34 -07001845 || available > slub_cpu_partial(s) / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001846 break;
1847
Christoph Lameter497b66f2011-08-09 16:12:26 -05001848 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001849 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001850 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001851}
1852
1853/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001854 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001855 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001856static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001857 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001858{
1859#ifdef CONFIG_NUMA
1860 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001861 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001862 struct zone *zone;
1863 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001864 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001865 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001866
1867 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001868 * The defrag ratio allows a configuration of the tradeoffs between
1869 * inter node defragmentation and node local allocations. A lower
1870 * defrag_ratio increases the tendency to do local allocations
1871 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001872 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001873 * If the defrag_ratio is set to 0 then kmalloc() always
1874 * returns node local objects. If the ratio is higher then kmalloc()
1875 * may return off node objects because partial slabs are obtained
1876 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001877 *
Li Peng43efd3e2016-05-19 17:10:43 -07001878 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1879 * (which makes defrag_ratio = 1000) then every (well almost)
1880 * allocation will first attempt to defrag slab caches on other nodes.
1881 * This means scanning over all nodes to look for partial slabs which
1882 * may be expensive if we do it every time we are trying to find a slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001883 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001884 */
Christoph Lameter98246012008-01-07 23:20:26 -08001885 if (!s->remote_node_defrag_ratio ||
1886 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001887 return NULL;
1888
Mel Gormancc9a6c82012-03-21 16:34:11 -07001889 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001890 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001891 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001892 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1893 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001894
Mel Gormancc9a6c82012-03-21 16:34:11 -07001895 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001896
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001897 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001898 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001899 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001900 if (object) {
1901 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001902 * Don't check read_mems_allowed_retry()
1903 * here - if mems_allowed was updated in
1904 * parallel, that was a harmless race
1905 * between allocation and the cpuset
1906 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001907 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001908 return object;
1909 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001910 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001911 }
Mel Gormand26914d2014-04-03 14:47:24 -07001912 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001913#endif
1914 return NULL;
1915}
1916
1917/*
1918 * Get a partial page, lock it and return it.
1919 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001920static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001921 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001922{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001923 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001924 int searchnode = node;
1925
1926 if (node == NUMA_NO_NODE)
1927 searchnode = numa_mem_id();
Christoph Lameter81819f02007-05-06 14:49:36 -07001928
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001929 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001930 if (object || node != NUMA_NO_NODE)
1931 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001932
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001933 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001934}
1935
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001936#ifdef CONFIG_PREEMPT
1937/*
1938 * Calculate the next globally unique transaction for disambiguiation
1939 * during cmpxchg. The transactions start with the cpu number and are then
1940 * incremented by CONFIG_NR_CPUS.
1941 */
1942#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1943#else
1944/*
1945 * No preemption supported therefore also no need to check for
1946 * different cpus.
1947 */
1948#define TID_STEP 1
1949#endif
1950
1951static inline unsigned long next_tid(unsigned long tid)
1952{
1953 return tid + TID_STEP;
1954}
1955
1956static inline unsigned int tid_to_cpu(unsigned long tid)
1957{
1958 return tid % TID_STEP;
1959}
1960
1961static inline unsigned long tid_to_event(unsigned long tid)
1962{
1963 return tid / TID_STEP;
1964}
1965
1966static inline unsigned int init_tid(int cpu)
1967{
1968 return cpu;
1969}
1970
1971static inline void note_cmpxchg_failure(const char *n,
1972 const struct kmem_cache *s, unsigned long tid)
1973{
1974#ifdef SLUB_DEBUG_CMPXCHG
1975 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1976
Fabian Frederickf9f58282014-06-04 16:06:34 -07001977 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001978
1979#ifdef CONFIG_PREEMPT
1980 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001981 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001982 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1983 else
1984#endif
1985 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001986 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001987 tid_to_event(tid), tid_to_event(actual_tid));
1988 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001989 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001990 actual_tid, tid, next_tid(tid));
1991#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001992 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001993}
1994
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001995static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001996{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001997 int cpu;
1998
1999 for_each_possible_cpu(cpu)
2000 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002001}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002002
2003/*
2004 * Remove the cpu slab
2005 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002006static void deactivate_slab(struct kmem_cache *s, struct page *page,
Wei Yangd4ff6d32017-07-06 15:36:25 -07002007 void *freelist, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002008{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002009 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002010 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2011 int lock = 0;
2012 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002013 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08002014 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002015 struct page new;
2016 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002017
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002018 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002019 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08002020 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07002021 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002022
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002023 /*
2024 * Stage one: Free all available per cpu objects back
2025 * to the page freelist while it is still frozen. Leave the
2026 * last one.
2027 *
2028 * There is no need to take the list->lock because the page
2029 * is still frozen.
2030 */
2031 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2032 void *prior;
2033 unsigned long counters;
2034
Dongli Zhang6c097552020-06-01 21:45:47 -07002035 /*
2036 * If 'nextfree' is invalid, it is possible that the object at
2037 * 'freelist' is already corrupted. So isolate all objects
2038 * starting at 'freelist'.
2039 */
Eugeniu Roscaaf2cf2c2020-09-04 16:35:30 -07002040 if (freelist_corrupted(s, page, &freelist, nextfree))
Dongli Zhang6c097552020-06-01 21:45:47 -07002041 break;
2042
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002043 do {
2044 prior = page->freelist;
2045 counters = page->counters;
2046 set_freepointer(s, freelist, prior);
2047 new.counters = counters;
2048 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08002049 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002050
Christoph Lameter1d071712011-07-14 12:49:12 -05002051 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002052 prior, counters,
2053 freelist, new.counters,
2054 "drain percpu freelist"));
2055
2056 freelist = nextfree;
2057 }
2058
2059 /*
2060 * Stage two: Ensure that the page is unfrozen while the
2061 * list presence reflects the actual number of objects
2062 * during unfreeze.
2063 *
2064 * We setup the list membership and then perform a cmpxchg
2065 * with the count. If there is a mismatch then the page
2066 * is not unfrozen but the page is on the wrong list.
2067 *
2068 * Then we restart the process which may have to remove
2069 * the page from the list that we just put it on again
2070 * because the number of objects in the slab may have
2071 * changed.
2072 */
2073redo:
2074
2075 old.freelist = page->freelist;
2076 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002077 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002078
2079 /* Determine target state of the slab */
2080 new.counters = old.counters;
2081 if (freelist) {
2082 new.inuse--;
2083 set_freepointer(s, freelist, old.freelist);
2084 new.freelist = freelist;
2085 } else
2086 new.freelist = old.freelist;
2087
2088 new.frozen = 0;
2089
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002090 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002091 m = M_FREE;
2092 else if (new.freelist) {
2093 m = M_PARTIAL;
2094 if (!lock) {
2095 lock = 1;
2096 /*
2097 * Taking the spinlock removes the possiblity
2098 * that acquire_slab() will see a slab page that
2099 * is frozen
2100 */
2101 spin_lock(&n->list_lock);
2102 }
2103 } else {
2104 m = M_FULL;
2105 if (kmem_cache_debug(s) && !lock) {
2106 lock = 1;
2107 /*
2108 * This also ensures that the scanning of full
2109 * slabs from diagnostic functions will not see
2110 * any frozen slabs.
2111 */
2112 spin_lock(&n->list_lock);
2113 }
2114 }
2115
2116 if (l != m) {
2117
2118 if (l == M_PARTIAL)
2119
2120 remove_partial(n, page);
2121
2122 else if (l == M_FULL)
2123
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002124 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002125
2126 if (m == M_PARTIAL) {
2127
2128 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08002129 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002130
2131 } else if (m == M_FULL) {
2132
2133 stat(s, DEACTIVATE_FULL);
2134 add_full(s, n, page);
2135
2136 }
2137 }
2138
2139 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05002140 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002141 old.freelist, old.counters,
2142 new.freelist, new.counters,
2143 "unfreezing slab"))
2144 goto redo;
2145
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002146 if (lock)
2147 spin_unlock(&n->list_lock);
2148
2149 if (m == M_FREE) {
2150 stat(s, DEACTIVATE_EMPTY);
2151 discard_slab(s, page);
2152 stat(s, FREE_SLAB);
2153 }
Wei Yangd4ff6d32017-07-06 15:36:25 -07002154
2155 c->page = NULL;
2156 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002157}
2158
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002159/*
2160 * Unfreeze all the cpu partial slabs.
2161 *
Christoph Lameter59a09912012-11-28 16:23:00 +00002162 * This function must be called with interrupts disabled
2163 * for the cpu using c (or some other guarantee must be there
2164 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002165 */
Christoph Lameter59a09912012-11-28 16:23:00 +00002166static void unfreeze_partials(struct kmem_cache *s,
2167 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05002168{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002169#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09002170 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08002171 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002172
2173 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002174 struct page new;
2175 struct page old;
2176
2177 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002178
2179 n2 = get_node(s, page_to_nid(page));
2180 if (n != n2) {
2181 if (n)
2182 spin_unlock(&n->list_lock);
2183
2184 n = n2;
2185 spin_lock(&n->list_lock);
2186 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002187
2188 do {
2189
2190 old.freelist = page->freelist;
2191 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002192 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05002193
2194 new.counters = old.counters;
2195 new.freelist = old.freelist;
2196
2197 new.frozen = 0;
2198
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002199 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002200 old.freelist, old.counters,
2201 new.freelist, new.counters,
2202 "unfreezing slab"));
2203
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002204 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002205 page->next = discard_page;
2206 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002207 } else {
2208 add_partial(n, page, DEACTIVATE_TO_TAIL);
2209 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002210 }
2211 }
2212
2213 if (n)
2214 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002215
2216 while (discard_page) {
2217 page = discard_page;
2218 discard_page = discard_page->next;
2219
2220 stat(s, DEACTIVATE_EMPTY);
2221 discard_slab(s, page);
2222 stat(s, FREE_SLAB);
2223 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002224#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002225}
2226
2227/*
2228 * Put a page that was just frozen (in __slab_free) into a partial page
Miles Chen0d2d5d42018-01-31 16:15:47 -08002229 * slot if available.
Christoph Lameter49e22582011-08-09 16:12:27 -05002230 *
2231 * If we did not find a slot then simply move all the partials to the
2232 * per node partial list.
2233 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002234static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002235{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002236#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002237 struct page *oldpage;
2238 int pages;
2239 int pobjects;
2240
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002241 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002242 do {
2243 pages = 0;
2244 pobjects = 0;
2245 oldpage = this_cpu_read(s->cpu_slab->partial);
2246
2247 if (oldpage) {
2248 pobjects = oldpage->pobjects;
2249 pages = oldpage->pages;
2250 if (drain && pobjects > s->cpu_partial) {
2251 unsigned long flags;
2252 /*
2253 * partial array is full. Move the existing
2254 * set to the per node partial list.
2255 */
2256 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002257 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002258 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002259 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002260 pobjects = 0;
2261 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002262 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002263 }
2264 }
2265
2266 pages++;
2267 pobjects += page->objects - page->inuse;
2268
2269 page->pages = pages;
2270 page->pobjects = pobjects;
2271 page->next = oldpage;
2272
Chen Gangd0e0ac92013-07-15 09:05:29 +08002273 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2274 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002275 if (unlikely(!s->cpu_partial)) {
2276 unsigned long flags;
2277
2278 local_irq_save(flags);
2279 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2280 local_irq_restore(flags);
2281 }
2282 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002283#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002284}
2285
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002286static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002287{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002288 stat(s, CPUSLAB_FLUSH);
Wei Yangd4ff6d32017-07-06 15:36:25 -07002289 deactivate_slab(s, c->page, c->freelist, c);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002290
2291 c->tid = next_tid(c->tid);
Christoph Lameter81819f02007-05-06 14:49:36 -07002292}
2293
2294/*
2295 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002296 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002297 * Called from IPI handler with interrupts disabled.
2298 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002299static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002300{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002301 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002302
Christoph Lameter49e22582011-08-09 16:12:27 -05002303 if (likely(c)) {
2304 if (c->page)
2305 flush_slab(s, c);
2306
Christoph Lameter59a09912012-11-28 16:23:00 +00002307 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002308 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002309}
2310
2311static void flush_cpu_slab(void *d)
2312{
2313 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002314
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002315 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002316}
2317
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002318static bool has_cpu_slab(int cpu, void *info)
2319{
2320 struct kmem_cache *s = info;
2321 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2322
Wei Yanga93cf072017-07-06 15:36:31 -07002323 return c->page || slub_percpu_partial(c);
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002324}
2325
Christoph Lameter81819f02007-05-06 14:49:36 -07002326static void flush_all(struct kmem_cache *s)
2327{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002328 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002329}
2330
2331/*
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02002332 * Use the cpu notifier to insure that the cpu slabs are flushed when
2333 * necessary.
2334 */
2335static int slub_cpu_dead(unsigned int cpu)
2336{
2337 struct kmem_cache *s;
2338 unsigned long flags;
2339
2340 mutex_lock(&slab_mutex);
2341 list_for_each_entry(s, &slab_caches, list) {
2342 local_irq_save(flags);
2343 __flush_cpu_slab(s, cpu);
2344 local_irq_restore(flags);
2345 }
2346 mutex_unlock(&slab_mutex);
2347 return 0;
2348}
2349
2350/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002351 * Check if the objects in a per cpu structure fit numa
2352 * locality expectations.
2353 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002354static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002355{
2356#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002357 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002358 return 0;
2359#endif
2360 return 1;
2361}
2362
David Rientjes9a02d692014-06-04 16:06:36 -07002363#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002364static int count_free(struct page *page)
2365{
2366 return page->objects - page->inuse;
2367}
2368
David Rientjes9a02d692014-06-04 16:06:36 -07002369static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2370{
2371 return atomic_long_read(&n->total_objects);
2372}
2373#endif /* CONFIG_SLUB_DEBUG */
2374
2375#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002376static unsigned long count_partial(struct kmem_cache_node *n,
2377 int (*get_count)(struct page *))
2378{
2379 unsigned long flags;
2380 unsigned long x = 0;
2381 struct page *page;
2382
2383 spin_lock_irqsave(&n->list_lock, flags);
2384 list_for_each_entry(page, &n->partial, lru)
2385 x += get_count(page);
2386 spin_unlock_irqrestore(&n->list_lock, flags);
2387 return x;
2388}
David Rientjes9a02d692014-06-04 16:06:36 -07002389#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002390
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002391static noinline void
2392slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2393{
David Rientjes9a02d692014-06-04 16:06:36 -07002394#ifdef CONFIG_SLUB_DEBUG
2395 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2396 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002397 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002398 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002399
David Rientjes9a02d692014-06-04 16:06:36 -07002400 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2401 return;
2402
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07002403 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2404 nid, gfpflags, &gfpflags);
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07002405 pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
Fabian Frederickf9f58282014-06-04 16:06:34 -07002406 s->name, s->object_size, s->size, oo_order(s->oo),
2407 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002408
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002409 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002410 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2411 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002412
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002413 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002414 unsigned long nr_slabs;
2415 unsigned long nr_objs;
2416 unsigned long nr_free;
2417
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002418 nr_free = count_partial(n, count_free);
2419 nr_slabs = node_nr_slabs(n);
2420 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002421
Fabian Frederickf9f58282014-06-04 16:06:34 -07002422 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002423 node, nr_slabs, nr_objs, nr_free);
2424 }
David Rientjes9a02d692014-06-04 16:06:36 -07002425#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002426}
2427
Christoph Lameter497b66f2011-08-09 16:12:26 -05002428static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2429 int node, struct kmem_cache_cpu **pc)
2430{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002431 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002432 struct kmem_cache_cpu *c = *pc;
2433 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002434
Matthew Wilcox128227e2018-06-07 17:05:13 -07002435 WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2436
Christoph Lameter188fd062012-05-09 10:09:55 -05002437 freelist = get_partial(s, flags, node, c);
2438
2439 if (freelist)
2440 return freelist;
2441
2442 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002443 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002444 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002445 if (c->page)
2446 flush_slab(s, c);
2447
2448 /*
2449 * No other reference to the page yet so we can
2450 * muck around with it freely without cmpxchg
2451 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002452 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002453 page->freelist = NULL;
2454
2455 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002456 c->page = page;
2457 *pc = c;
2458 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002459 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002460
Christoph Lameter6faa6832012-05-09 10:09:51 -05002461 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002462}
2463
Mel Gorman072bb0a2012-07-31 16:43:58 -07002464static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2465{
2466 if (unlikely(PageSlabPfmemalloc(page)))
2467 return gfp_pfmemalloc_allowed(gfpflags);
2468
2469 return true;
2470}
2471
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002472/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002473 * Check the page->freelist of a page and either transfer the freelist to the
2474 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002475 *
2476 * The page is still frozen if the return value is not NULL.
2477 *
2478 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002479 *
2480 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002481 */
2482static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2483{
2484 struct page new;
2485 unsigned long counters;
2486 void *freelist;
2487
2488 do {
2489 freelist = page->freelist;
2490 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002491
Christoph Lameter213eeb92011-11-11 14:07:14 -06002492 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002493 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002494
2495 new.inuse = page->objects;
2496 new.frozen = freelist != NULL;
2497
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002498 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002499 freelist, counters,
2500 NULL, new.counters,
2501 "get_freelist"));
2502
2503 return freelist;
2504}
2505
2506/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002507 * Slow path. The lockless freelist is empty or we need to perform
2508 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002509 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002510 * Processing is still very fast if new objects have been freed to the
2511 * regular freelist. In that case we simply take over the regular freelist
2512 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002513 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002514 * If that is not working then we fall back to the partial lists. We take the
2515 * first element of the freelist as the object to allocate now and move the
2516 * rest of the freelist to the lockless freelist.
2517 *
2518 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002519 * we need to allocate a new slab. This is the slowest path since it involves
2520 * a call to the page allocator and the setup of a new slab.
Christoph Lametera380a3c2015-11-20 15:57:35 -08002521 *
2522 * Version of __slab_alloc to use when we know that interrupts are
2523 * already disabled (which is the case for bulk allocation).
Christoph Lameter81819f02007-05-06 14:49:36 -07002524 */
Christoph Lametera380a3c2015-11-20 15:57:35 -08002525static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002526 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002527{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002528 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002529 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002530
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002531 page = c->page;
Vlastimil Babka3e79ba62020-03-21 18:22:37 -07002532 if (!page) {
2533 /*
2534 * if the node is not online or has no normal memory, just
2535 * ignore the node constraint
2536 */
2537 if (unlikely(node != NUMA_NO_NODE &&
2538 !node_state(node, N_NORMAL_MEMORY)))
2539 node = NUMA_NO_NODE;
Christoph Lameter81819f02007-05-06 14:49:36 -07002540 goto new_slab;
Vlastimil Babka3e79ba62020-03-21 18:22:37 -07002541 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002542redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002543
Christoph Lameter57d437d2012-05-09 10:09:59 -05002544 if (unlikely(!node_match(page, node))) {
Vlastimil Babka3e79ba62020-03-21 18:22:37 -07002545 /*
2546 * same as above but node_match() being false already
2547 * implies node != NUMA_NO_NODE
2548 */
2549 if (!node_state(node, N_NORMAL_MEMORY)) {
2550 node = NUMA_NO_NODE;
2551 goto redo;
2552 } else {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002553 stat(s, ALLOC_NODE_MISMATCH);
Wei Yangd4ff6d32017-07-06 15:36:25 -07002554 deactivate_slab(s, page, c->freelist, c);
Joonsoo Kima561ce02014-10-09 15:26:15 -07002555 goto new_slab;
2556 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002557 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002558
Mel Gorman072bb0a2012-07-31 16:43:58 -07002559 /*
2560 * By rights, we should be searching for a slab page that was
2561 * PFMEMALLOC but right now, we are losing the pfmemalloc
2562 * information when the page leaves the per-cpu allocator
2563 */
2564 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
Wei Yangd4ff6d32017-07-06 15:36:25 -07002565 deactivate_slab(s, page, c->freelist, c);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002566 goto new_slab;
2567 }
2568
Eric Dumazet73736e02011-12-13 04:57:06 +01002569 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002570 freelist = c->freelist;
2571 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002572 goto load_freelist;
2573
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002574 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002575
Christoph Lameter6faa6832012-05-09 10:09:51 -05002576 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002577 c->page = NULL;
2578 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002579 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002580 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002581
Christoph Lameter81819f02007-05-06 14:49:36 -07002582 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002583
Christoph Lameter894b8782007-05-10 03:15:16 -07002584load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002585 /*
2586 * freelist is pointing to the list of objects to be used.
2587 * page is pointing to the page from which the objects are obtained.
2588 * That page must be frozen for per cpu allocations to work.
2589 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002590 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002591 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002592 c->tid = next_tid(c->tid);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002593 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002594
Christoph Lameter81819f02007-05-06 14:49:36 -07002595new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002596
Wei Yanga93cf072017-07-06 15:36:31 -07002597 if (slub_percpu_partial(c)) {
2598 page = c->page = slub_percpu_partial(c);
2599 slub_set_percpu_partial(c, page);
Christoph Lameter49e22582011-08-09 16:12:27 -05002600 stat(s, CPU_PARTIAL_ALLOC);
Christoph Lameter49e22582011-08-09 16:12:27 -05002601 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002602 }
2603
Christoph Lameter188fd062012-05-09 10:09:55 -05002604 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002605
Christoph Lameterf46974362012-05-09 10:09:54 -05002606 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002607 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002608 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002609 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002610
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002611 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002612 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002613 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002614
Christoph Lameter497b66f2011-08-09 16:12:26 -05002615 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002616 if (kmem_cache_debug(s) &&
2617 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002618 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002619
Wei Yangd4ff6d32017-07-06 15:36:25 -07002620 deactivate_slab(s, page, get_freepointer(s, freelist), c);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002621 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002622}
2623
2624/*
Christoph Lametera380a3c2015-11-20 15:57:35 -08002625 * Another one that disabled interrupt and compensates for possible
2626 * cpu changes by refetching the per cpu area pointer.
2627 */
2628static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2629 unsigned long addr, struct kmem_cache_cpu *c)
2630{
2631 void *p;
2632 unsigned long flags;
2633
2634 local_irq_save(flags);
2635#ifdef CONFIG_PREEMPT
2636 /*
2637 * We may have been preempted and rescheduled on a different
2638 * cpu before disabling interrupts. Need to reload cpu area
2639 * pointer.
2640 */
2641 c = this_cpu_ptr(s->cpu_slab);
2642#endif
2643
2644 p = ___slab_alloc(s, gfpflags, node, addr, c);
2645 local_irq_restore(flags);
2646 return p;
2647}
2648
2649/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002650 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2651 * have the fastpath folded into their functions. So no function call
2652 * overhead for requests that can be satisfied on the fastpath.
2653 *
2654 * The fastpath works by first checking if the lockless freelist can be used.
2655 * If not then __slab_alloc is called for slow processing.
2656 *
2657 * Otherwise we can simply pick the next object from the lockless free list.
2658 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002659static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002660 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002661{
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002662 void *object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002663 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002664 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002665 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002666
Vladimir Davydov8135be52014-12-12 16:56:38 -08002667 s = slab_pre_alloc_hook(s, gfpflags);
2668 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002669 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002670redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002671 /*
2672 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2673 * enabled. We may switch back and forth between cpus while
2674 * reading from one cpu area. That does not matter as long
2675 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002676 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002677 * We should guarantee that tid and kmem_cache are retrieved on
2678 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2679 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002680 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002681 do {
2682 tid = this_cpu_read(s->cpu_slab->tid);
2683 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002684 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2685 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002686
2687 /*
2688 * Irqless object alloc/free algorithm used here depends on sequence
2689 * of fetching cpu_slab's data. tid should be fetched before anything
2690 * on c to guarantee that object and page associated with previous tid
2691 * won't be used with current tid. If we fetch tid first, object and
2692 * page could be one associated with next tid and our alloc/free
2693 * request will be failed. In this case, we will retry. So, no problem.
2694 */
2695 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002696
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002697 /*
2698 * The transaction ids are globally unique per cpu and per operation on
2699 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2700 * occurs on the right processor and that there was no operation on the
2701 * linked list in between.
2702 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002703
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002704 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002705 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002706 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002707 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002708 stat(s, ALLOC_SLOWPATH);
2709 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002710 void *next_object = get_freepointer_safe(s, object);
2711
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002712 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002713 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002714 * operation and if we are on the right processor.
2715 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002716 * The cmpxchg does the following atomically (without lock
2717 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002718 * 1. Relocate first pointer to the current per cpu area.
2719 * 2. Verify that tid and freelist have not been changed
2720 * 3. If they were not changed replace tid and freelist
2721 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002722 * Since this is without lock semantics the protection is only
2723 * against code executing on this cpu *not* from access by
2724 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002725 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002726 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002727 s->cpu_slab->freelist, s->cpu_slab->tid,
2728 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002729 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002730
2731 note_cmpxchg_failure("slab_alloc", s, tid);
2732 goto redo;
2733 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002734 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002735 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002736 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002737
Pekka Enberg74e21342009-11-25 20:14:48 +02002738 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002739 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002740
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002741 slab_post_alloc_hook(s, gfpflags, 1, &object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002742
Christoph Lameter894b8782007-05-10 03:15:16 -07002743 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002744}
2745
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002746static __always_inline void *slab_alloc(struct kmem_cache *s,
2747 gfp_t gfpflags, unsigned long addr)
2748{
2749 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2750}
2751
Christoph Lameter81819f02007-05-06 14:49:36 -07002752void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2753{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002754 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002755
Chen Gangd0e0ac92013-07-15 09:05:29 +08002756 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2757 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002758
2759 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002760}
2761EXPORT_SYMBOL(kmem_cache_alloc);
2762
Li Zefan0f24f122009-12-11 15:45:30 +08002763#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002764void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002765{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002766 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002767 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002768 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002769 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002770}
Richard Kennedy4a923792010-10-21 10:29:19 +01002771EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002772#endif
2773
Christoph Lameter81819f02007-05-06 14:49:36 -07002774#ifdef CONFIG_NUMA
2775void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2776{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002777 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002778
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002779 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002780 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002781
2782 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002783}
2784EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002785
Li Zefan0f24f122009-12-11 15:45:30 +08002786#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002787void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002788 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002789 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002790{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002791 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002792
2793 trace_kmalloc_node(_RET_IP_, ret,
2794 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002795
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002796 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002797 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002798}
Richard Kennedy4a923792010-10-21 10:29:19 +01002799EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002800#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002801#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002802
Christoph Lameter81819f02007-05-06 14:49:36 -07002803/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002804 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002805 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002806 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002807 * So we still attempt to reduce cache line usage. Just take the slab
2808 * lock and free the item. If there is no additional partial page
2809 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002810 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002811static void __slab_free(struct kmem_cache *s, struct page *page,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002812 void *head, void *tail, int cnt,
2813 unsigned long addr)
2814
Christoph Lameter81819f02007-05-06 14:49:36 -07002815{
2816 void *prior;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002817 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002818 struct page new;
2819 unsigned long counters;
2820 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002821 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002822
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002823 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002824
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002825 if (kmem_cache_debug(s) &&
Laura Abbott282acb42016-03-15 14:54:59 -07002826 !free_debug_processing(s, page, head, tail, cnt, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002827 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002828
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002829 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002830 if (unlikely(n)) {
2831 spin_unlock_irqrestore(&n->list_lock, flags);
2832 n = NULL;
2833 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002834 prior = page->freelist;
2835 counters = page->counters;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002836 set_freepointer(s, tail, prior);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002837 new.counters = counters;
2838 was_frozen = new.frozen;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002839 new.inuse -= cnt;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002840 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002841
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002842 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002843
2844 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002845 * Slab was on no list before and will be
2846 * partially empty
2847 * We can defer the list move and instead
2848 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002849 */
2850 new.frozen = 1;
2851
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002852 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002853
LQYMGTb455def2014-12-10 15:42:13 -08002854 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002855 /*
2856 * Speculatively acquire the list_lock.
2857 * If the cmpxchg does not succeed then we may
2858 * drop the list_lock without any processing.
2859 *
2860 * Otherwise the list_lock will synchronize with
2861 * other processors updating the list of slabs.
2862 */
2863 spin_lock_irqsave(&n->list_lock, flags);
2864
2865 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002866 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002867
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002868 } while (!cmpxchg_double_slab(s, page,
2869 prior, counters,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002870 head, new.counters,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002871 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002872
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002873 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002874
2875 /*
2876 * If we just froze the page then put it onto the
2877 * per cpu partial list.
2878 */
Alex Shi8028dce2012-02-03 23:34:56 +08002879 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002880 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002881 stat(s, CPU_PARTIAL_FREE);
2882 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002883 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002884 * The list lock was not taken therefore no list
2885 * activity can be necessary.
2886 */
LQYMGTb455def2014-12-10 15:42:13 -08002887 if (was_frozen)
2888 stat(s, FREE_FROZEN);
2889 return;
2890 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002891
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002892 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002893 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002894
Joonsoo Kim837d6782012-08-16 00:02:40 +09002895 /*
2896 * Objects left in the slab. If it was not on the partial list before
2897 * then add it.
2898 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002899 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2900 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002901 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002902 add_partial(n, page, DEACTIVATE_TO_TAIL);
2903 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002904 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002905 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002906 return;
2907
2908slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002909 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002910 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002911 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002912 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002913 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002914 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002915 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002916 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002917 remove_full(s, n, page);
2918 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002919
Christoph Lameter80f08c12011-06-01 12:25:55 -05002920 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002921 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002922 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002923}
2924
Christoph Lameter894b8782007-05-10 03:15:16 -07002925/*
2926 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2927 * can perform fastpath freeing without additional function calls.
2928 *
2929 * The fastpath is only possible if we are freeing to the current cpu slab
2930 * of this processor. This typically the case if we have just allocated
2931 * the item before.
2932 *
2933 * If fastpath is not possible then fall back to __slab_free where we deal
2934 * with all sorts of special processing.
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002935 *
2936 * Bulk free of a freelist with several objects (all pointing to the
2937 * same page) possible by specifying head and tail ptr, plus objects
2938 * count (cnt). Bulk free indicated by tail pointer being set.
Christoph Lameter894b8782007-05-10 03:15:16 -07002939 */
Alexander Potapenko80a92012016-07-28 15:49:07 -07002940static __always_inline void do_slab_free(struct kmem_cache *s,
2941 struct page *page, void *head, void *tail,
2942 int cnt, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002943{
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002944 void *tail_obj = tail ? : head;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002945 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002946 unsigned long tid;
Christoph Lametera24c5a02011-03-15 12:45:21 -05002947redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002948 /*
2949 * Determine the currently cpus per cpu slab.
2950 * The cpu may change afterward. However that does not matter since
2951 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07002952 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002953 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002954 do {
2955 tid = this_cpu_read(s->cpu_slab->tid);
2956 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002957 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2958 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002959
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002960 /* Same with comment on barrier() in slab_alloc_node() */
2961 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002962
Christoph Lameter442b06b2011-05-17 16:29:31 -05002963 if (likely(page == c->page)) {
Linus Torvalds451d4a22020-03-17 11:04:09 -07002964 void **freelist = READ_ONCE(c->freelist);
2965
2966 set_freepointer(s, tail_obj, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002967
Christoph Lameter933393f2011-12-22 11:58:51 -06002968 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002969 s->cpu_slab->freelist, s->cpu_slab->tid,
Linus Torvalds451d4a22020-03-17 11:04:09 -07002970 freelist, tid,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002971 head, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002972
2973 note_cmpxchg_failure("slab_free", s, tid);
2974 goto redo;
2975 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002976 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002977 } else
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002978 __slab_free(s, page, head, tail_obj, cnt, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002979
Christoph Lameter894b8782007-05-10 03:15:16 -07002980}
2981
Alexander Potapenko80a92012016-07-28 15:49:07 -07002982static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2983 void *head, void *tail, int cnt,
2984 unsigned long addr)
2985{
Alexander Potapenko80a92012016-07-28 15:49:07 -07002986 /*
Andrey Konovalovc3895392018-04-10 16:30:31 -07002987 * With KASAN enabled slab_free_freelist_hook modifies the freelist
2988 * to remove objects, whose reuse must be delayed.
Alexander Potapenko80a92012016-07-28 15:49:07 -07002989 */
Andrey Konovalovc3895392018-04-10 16:30:31 -07002990 if (slab_free_freelist_hook(s, &head, &tail))
2991 do_slab_free(s, page, head, tail, cnt, addr);
Alexander Potapenko80a92012016-07-28 15:49:07 -07002992}
2993
2994#ifdef CONFIG_KASAN
2995void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2996{
2997 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2998}
2999#endif
3000
Christoph Lameter81819f02007-05-06 14:49:36 -07003001void kmem_cache_free(struct kmem_cache *s, void *x)
3002{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003003 s = cache_from_obj(s, x);
3004 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00003005 return;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003006 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003007 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07003008}
3009EXPORT_SYMBOL(kmem_cache_free);
3010
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003011struct detached_freelist {
3012 struct page *page;
3013 void *tail;
3014 void *freelist;
3015 int cnt;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003016 struct kmem_cache *s;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003017};
3018
3019/*
3020 * This function progressively scans the array with free objects (with
3021 * a limited look ahead) and extract objects belonging to the same
3022 * page. It builds a detached freelist directly within the given
3023 * page/objects. This can happen without any need for
3024 * synchronization, because the objects are owned by running process.
3025 * The freelist is build up as a single linked list in the objects.
3026 * The idea is, that this detached freelist can then be bulk
3027 * transferred to the real freelist(s), but only requiring a single
3028 * synchronization primitive. Look ahead in the array is limited due
3029 * to performance reasons.
3030 */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003031static inline
3032int build_detached_freelist(struct kmem_cache *s, size_t size,
3033 void **p, struct detached_freelist *df)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003034{
3035 size_t first_skipped_index = 0;
3036 int lookahead = 3;
3037 void *object;
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003038 struct page *page;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003039
3040 /* Always re-init detached_freelist */
3041 df->page = NULL;
3042
3043 do {
3044 object = p[--size];
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003045 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003046 } while (!object && size);
3047
3048 if (!object)
3049 return 0;
3050
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003051 page = virt_to_head_page(object);
3052 if (!s) {
3053 /* Handle kalloc'ed objects */
3054 if (unlikely(!PageSlab(page))) {
3055 BUG_ON(!PageCompound(page));
3056 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003057 __free_pages(page, compound_order(page));
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003058 p[size] = NULL; /* mark object processed */
3059 return size;
3060 }
3061 /* Derive kmem_cache from object */
3062 df->s = page->slab_cache;
3063 } else {
3064 df->s = cache_from_obj(s, object); /* Support for memcg */
3065 }
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003066
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003067 /* Start new detached freelist */
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003068 df->page = page;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003069 set_freepointer(df->s, object, NULL);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003070 df->tail = object;
3071 df->freelist = object;
3072 p[size] = NULL; /* mark object processed */
3073 df->cnt = 1;
3074
3075 while (size) {
3076 object = p[--size];
3077 if (!object)
3078 continue; /* Skip processed objects */
3079
3080 /* df->page is always set at this point */
3081 if (df->page == virt_to_head_page(object)) {
3082 /* Opportunity build freelist */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003083 set_freepointer(df->s, object, df->freelist);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003084 df->freelist = object;
3085 df->cnt++;
3086 p[size] = NULL; /* mark object processed */
3087
3088 continue;
3089 }
3090
3091 /* Limit look ahead search */
3092 if (!--lookahead)
3093 break;
3094
3095 if (!first_skipped_index)
3096 first_skipped_index = size + 1;
3097 }
3098
3099 return first_skipped_index;
3100}
3101
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003102/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003103void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003104{
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003105 if (WARN_ON(!size))
3106 return;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003107
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003108 do {
3109 struct detached_freelist df;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003110
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003111 size = build_detached_freelist(s, size, p, &df);
Arnd Bergmann84582c82016-12-12 16:41:35 -08003112 if (!df.page)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003113 continue;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003114
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003115 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003116 } while (likely(size));
Christoph Lameter484748f2015-09-04 15:45:34 -07003117}
3118EXPORT_SYMBOL(kmem_cache_free_bulk);
3119
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003120/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003121int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3122 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003123{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003124 struct kmem_cache_cpu *c;
3125 int i;
3126
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003127 /* memcg and kmem_cache debug support */
3128 s = slab_pre_alloc_hook(s, flags);
3129 if (unlikely(!s))
3130 return false;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003131 /*
3132 * Drain objects in the per cpu slab, while disabling local
3133 * IRQs, which protects against PREEMPT and interrupts
3134 * handlers invoking normal fastpath.
3135 */
3136 local_irq_disable();
3137 c = this_cpu_ptr(s->cpu_slab);
3138
3139 for (i = 0; i < size; i++) {
3140 void *object = c->freelist;
3141
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003142 if (unlikely(!object)) {
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003143 /*
Jann Horn30f6cae2020-03-17 01:28:45 +01003144 * We may have removed an object from c->freelist using
3145 * the fastpath in the previous iteration; in that case,
3146 * c->tid has not been bumped yet.
3147 * Since ___slab_alloc() may reenable interrupts while
3148 * allocating memory, we should bump c->tid now.
3149 */
3150 c->tid = next_tid(c->tid);
3151
3152 /*
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003153 * Invoking slow path likely have side-effect
3154 * of re-populating per CPU c->freelist
3155 */
Christoph Lameter87098372015-11-20 15:57:38 -08003156 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003157 _RET_IP_, c);
Christoph Lameter87098372015-11-20 15:57:38 -08003158 if (unlikely(!p[i]))
3159 goto error;
3160
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003161 c = this_cpu_ptr(s->cpu_slab);
3162 continue; /* goto for-loop */
3163 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003164 c->freelist = get_freepointer(s, object);
3165 p[i] = object;
3166 }
3167 c->tid = next_tid(c->tid);
3168 local_irq_enable();
3169
3170 /* Clear memory outside IRQ disabled fastpath loop */
3171 if (unlikely(flags & __GFP_ZERO)) {
3172 int j;
3173
3174 for (j = 0; j < i; j++)
3175 memset(p[j], 0, s->object_size);
3176 }
3177
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003178 /* memcg and kmem_cache debug support */
3179 slab_post_alloc_hook(s, flags, size, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003180 return i;
Christoph Lameter87098372015-11-20 15:57:38 -08003181error:
Christoph Lameter87098372015-11-20 15:57:38 -08003182 local_irq_enable();
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003183 slab_post_alloc_hook(s, flags, i, p);
3184 __kmem_cache_free_bulk(s, i, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003185 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003186}
3187EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3188
3189
Christoph Lameter81819f02007-05-06 14:49:36 -07003190/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003191 * Object placement in a slab is made very easy because we always start at
3192 * offset 0. If we tune the size of the object to the alignment then we can
3193 * get the required alignment by putting one properly sized object after
3194 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07003195 *
3196 * Notice that the allocation order determines the sizes of the per cpu
3197 * caches. Each processor has always one slab available for allocations.
3198 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07003199 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07003200 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07003201 */
3202
3203/*
3204 * Mininum / Maximum order of slab pages. This influences locking overhead
3205 * and slab fragmentation. A higher order reduces the number of partial slabs
3206 * and increases the number of allocations possible without having to
3207 * take the list_lock.
3208 */
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003209static unsigned int slub_min_order;
3210static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3211static unsigned int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07003212
3213/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003214 * Calculate the order of allocation given an slab object size.
3215 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003216 * The order of allocation has significant impact on performance and other
3217 * system components. Generally order 0 allocations should be preferred since
3218 * order 0 does not cause fragmentation in the page allocator. Larger objects
3219 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003220 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07003221 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07003222 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003223 * In order to reach satisfactory performance we must ensure that a minimum
3224 * number of objects is in one slab. Otherwise we may generate too much
3225 * activity on the partial lists which requires taking the list_lock. This is
3226 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07003227 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003228 * slub_max_order specifies the order where we begin to stop considering the
3229 * number of objects in a slab as critical. If we reach slub_max_order then
3230 * we try to keep the page order as low as possible. So we accept more waste
3231 * of space in favor of a small page order.
3232 *
3233 * Higher order allocations also allow the placement of more objects in a
3234 * slab and thereby reduce object handling overhead. If the user has
3235 * requested a higher mininum order then we start with that one instead of
3236 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07003237 */
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003238static inline unsigned int slab_order(unsigned int size,
3239 unsigned int min_objects, unsigned int max_order,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003240 unsigned int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003241{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003242 unsigned int min_order = slub_min_order;
3243 unsigned int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003244
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003245 if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04003246 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03003247
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003248 for (order = max(min_order, (unsigned int)get_order(min_objects * size));
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003249 order <= max_order; order++) {
3250
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003251 unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3252 unsigned int rem;
Christoph Lameter81819f02007-05-06 14:49:36 -07003253
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003254 rem = slab_size % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003255
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003256 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003257 break;
Christoph Lameter81819f02007-05-06 14:49:36 -07003258 }
Christoph Lameter672bba32007-05-09 02:32:39 -07003259
Christoph Lameter81819f02007-05-06 14:49:36 -07003260 return order;
3261}
3262
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003263static inline int calculate_order(unsigned int size)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003264{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003265 unsigned int order;
3266 unsigned int min_objects;
3267 unsigned int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003268
3269 /*
3270 * Attempt to find best configuration for a slab. This
3271 * works by first attempting to generate a layout with
3272 * the best configuration and backing off gradually.
3273 *
Wei Yang422ff4d2015-11-05 18:45:46 -08003274 * First we increase the acceptable waste in a slab. Then
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003275 * we reduce the minimum objects required in a slab.
3276 */
3277 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003278 if (!min_objects)
3279 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003280 max_objects = order_objects(slub_max_order, size);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003281 min_objects = min(min_objects, max_objects);
3282
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003283 while (min_objects > 1) {
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003284 unsigned int fraction;
3285
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003286 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003287 while (fraction >= 4) {
3288 order = slab_order(size, min_objects,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003289 slub_max_order, fraction);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003290 if (order <= slub_max_order)
3291 return order;
3292 fraction /= 2;
3293 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03003294 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003295 }
3296
3297 /*
3298 * We were unable to place multiple objects in a slab. Now
3299 * lets see if we can place a single object there.
3300 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003301 order = slab_order(size, 1, slub_max_order, 1);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003302 if (order <= slub_max_order)
3303 return order;
3304
3305 /*
3306 * Doh this slab cannot be placed using slub_max_order.
3307 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003308 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03003309 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003310 return order;
3311 return -ENOSYS;
3312}
3313
Pekka Enberg5595cff2008-08-05 09:28:47 +03003314static void
Joonsoo Kim40534972012-05-11 00:50:47 +09003315init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003316{
3317 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003318 spin_lock_init(&n->list_lock);
3319 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003320#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003321 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07003322 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07003323 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003324#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003325}
3326
Christoph Lameter55136592010-08-20 12:37:13 -05003327static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003328{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05003329 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003330 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003331
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003332 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003333 * Must align to double word boundary for the double cmpxchg
3334 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003335 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003336 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3337 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003338
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003339 if (!s->cpu_slab)
3340 return 0;
3341
3342 init_kmem_cache_cpus(s);
3343
3344 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003345}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003346
Christoph Lameter51df1142010-08-20 12:37:15 -05003347static struct kmem_cache *kmem_cache_node;
3348
Christoph Lameter81819f02007-05-06 14:49:36 -07003349/*
3350 * No kmalloc_node yet so do it by hand. We know that this is the first
3351 * slab on the node for this slabcache. There are no concurrent accesses
3352 * possible.
3353 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003354 * Note that this function only works on the kmem_cache_node
3355 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003356 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003357 */
Christoph Lameter55136592010-08-20 12:37:13 -05003358static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003359{
3360 struct page *page;
3361 struct kmem_cache_node *n;
3362
Christoph Lameter51df1142010-08-20 12:37:15 -05003363 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003364
Christoph Lameter51df1142010-08-20 12:37:15 -05003365 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003366
3367 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003368 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003369 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3370 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003371 }
3372
Christoph Lameter81819f02007-05-06 14:49:36 -07003373 n = page->freelist;
3374 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003375 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05003376 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05003377 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003378 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07003379#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003380 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003381 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003382#endif
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003383 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3384 GFP_KERNEL);
Joonsoo Kim40534972012-05-11 00:50:47 +09003385 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003386 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003387
Dave Hansen67b6c902014-01-24 07:20:23 -08003388 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003389 * No locks need to be taken here as it has just been
3390 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003391 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003392 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003393}
3394
3395static void free_kmem_cache_nodes(struct kmem_cache *s)
3396{
3397 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003398 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003399
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003400 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003401 s->node[node] = NULL;
Alexander Potapenkoea37df52017-09-06 16:19:15 -07003402 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003403 }
3404}
3405
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003406void __kmem_cache_release(struct kmem_cache *s)
3407{
Thomas Garnier210e7a42016-07-26 15:21:59 -07003408 cache_random_seq_destroy(s);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003409 free_percpu(s->cpu_slab);
3410 free_kmem_cache_nodes(s);
3411}
3412
Christoph Lameter55136592010-08-20 12:37:13 -05003413static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003414{
3415 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003416
Christoph Lameterf64dc582007-10-16 01:25:33 -07003417 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003418 struct kmem_cache_node *n;
3419
Alexander Duyck73367bd2010-05-21 14:41:35 -07003420 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003421 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003422 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003423 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003424 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003425 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003426
3427 if (!n) {
3428 free_kmem_cache_nodes(s);
3429 return 0;
3430 }
3431
Joonsoo Kim40534972012-05-11 00:50:47 +09003432 init_kmem_cache_node(n);
Alexander Potapenkoea37df52017-09-06 16:19:15 -07003433 s->node[node] = n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003434 }
3435 return 1;
3436}
Christoph Lameter81819f02007-05-06 14:49:36 -07003437
David Rientjesc0bdb232009-02-25 09:16:35 +02003438static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003439{
3440 if (min < MIN_PARTIAL)
3441 min = MIN_PARTIAL;
3442 else if (min > MAX_PARTIAL)
3443 min = MAX_PARTIAL;
3444 s->min_partial = min;
3445}
3446
Wei Yange6d0e1d2017-07-06 15:36:34 -07003447static void set_cpu_partial(struct kmem_cache *s)
3448{
3449#ifdef CONFIG_SLUB_CPU_PARTIAL
3450 /*
3451 * cpu_partial determined the maximum number of objects kept in the
3452 * per cpu partial lists of a processor.
3453 *
3454 * Per cpu partial lists mainly contain slabs that just have one
3455 * object freed. If they are used for allocation then they can be
3456 * filled up again with minimal effort. The slab will never hit the
3457 * per node partial lists and therefore no locking will be required.
3458 *
3459 * This setting also determines
3460 *
3461 * A) The number of objects from per cpu partial slabs dumped to the
3462 * per node list when we reach the limit.
3463 * B) The number of objects in cpu partial slabs to extract from the
3464 * per node list when we run out of per cpu objects. We only fetch
3465 * 50% to keep some capacity around for frees.
3466 */
3467 if (!kmem_cache_has_cpu_partial(s))
3468 s->cpu_partial = 0;
3469 else if (s->size >= PAGE_SIZE)
3470 s->cpu_partial = 2;
3471 else if (s->size >= 1024)
3472 s->cpu_partial = 6;
3473 else if (s->size >= 256)
3474 s->cpu_partial = 13;
3475 else
3476 s->cpu_partial = 30;
3477#endif
3478}
3479
Christoph Lameter81819f02007-05-06 14:49:36 -07003480/*
3481 * calculate_sizes() determines the order and the distribution of data within
3482 * a slab object.
3483 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003484static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003485{
Alexey Dobriyand50112e2017-11-15 17:32:18 -08003486 slab_flags_t flags = s->flags;
Alexey Dobriyanbe4a7982018-04-05 16:21:28 -07003487 unsigned int size = s->object_size;
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003488 unsigned int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003489
3490 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003491 * Round up object size to the next word boundary. We can only
3492 * place the free pointer at word boundaries and this determines
3493 * the possible location of the free pointer.
3494 */
3495 size = ALIGN(size, sizeof(void *));
3496
3497#ifdef CONFIG_SLUB_DEBUG
3498 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003499 * Determine if we can poison the object itself. If the user of
3500 * the slab may touch the object after free or before allocation
3501 * then we should never poison the object itself.
3502 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003503 if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003504 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003505 s->flags |= __OBJECT_POISON;
3506 else
3507 s->flags &= ~__OBJECT_POISON;
3508
Christoph Lameter81819f02007-05-06 14:49:36 -07003509
3510 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003511 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003512 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003513 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003514 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003515 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003516 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003517#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003518
3519 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003520 * With that we have determined the number of bytes in actual use
3521 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003522 */
3523 s->inuse = size;
3524
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003525 if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003526 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003527 /*
3528 * Relocate free pointer after the object if it is not
3529 * permitted to overwrite the first word of the object on
3530 * kmem_cache_free.
3531 *
3532 * This is the case if we do RCU, have a constructor or
3533 * destructor or are poisoning the objects.
3534 */
3535 s->offset = size;
3536 size += sizeof(void *);
3537 }
3538
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003539#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003540 if (flags & SLAB_STORE_USER)
3541 /*
3542 * Need to store information about allocs and frees after
3543 * the object.
3544 */
3545 size += 2 * sizeof(struct track);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003546#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003547
Alexander Potapenko80a92012016-07-28 15:49:07 -07003548 kasan_cache_create(s, &size, &s->flags);
3549#ifdef CONFIG_SLUB_DEBUG
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003550 if (flags & SLAB_RED_ZONE) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003551 /*
3552 * Add some empty padding so that we can catch
3553 * overwrites from earlier objects rather than let
3554 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003555 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003556 * of the object.
3557 */
3558 size += sizeof(void *);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003559
3560 s->red_left_pad = sizeof(void *);
3561 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3562 size += s->red_left_pad;
3563 }
Christoph Lameter41ecc552007-05-09 02:32:44 -07003564#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003565
Christoph Lameter81819f02007-05-06 14:49:36 -07003566 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003567 * SLUB stores one object immediately after another beginning from
3568 * offset 0. In order to align the objects we have to simply size
3569 * each object to conform to the alignment.
3570 */
Christoph Lameter45906852012-11-28 16:23:16 +00003571 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003572 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003573 if (forced_order >= 0)
3574 order = forced_order;
3575 else
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003576 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003577
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003578 if ((int)order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003579 return 0;
3580
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003581 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003582 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003583 s->allocflags |= __GFP_COMP;
3584
3585 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003586 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003587
Nicolas Boichat62d342d2019-03-28 20:43:42 -07003588 if (s->flags & SLAB_CACHE_DMA32)
3589 s->allocflags |= GFP_DMA32;
3590
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003591 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3592 s->allocflags |= __GFP_RECLAIMABLE;
3593
Christoph Lameter81819f02007-05-06 14:49:36 -07003594 /*
3595 * Determine the number of objects per slab
3596 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003597 s->oo = oo_make(order, size);
3598 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03003599 if (oo_objects(s->oo) > oo_objects(s->max))
3600 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003601
Christoph Lameter834f3d12008-04-14 19:11:31 +03003602 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003603}
3604
Alexey Dobriyand50112e2017-11-15 17:32:18 -08003605static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003606{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003607 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Kees Cook2482dde2017-09-06 16:19:18 -07003608#ifdef CONFIG_SLAB_FREELIST_HARDENED
3609 s->random = get_random_long();
3610#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003611
Christoph Lameter06b285d2008-04-14 19:11:41 +03003612 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003613 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003614 if (disable_higher_order_debug) {
3615 /*
3616 * Disable debugging flags that store metadata if the min slab
3617 * order increased.
3618 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003619 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003620 s->flags &= ~DEBUG_METADATA_FLAGS;
3621 s->offset = 0;
3622 if (!calculate_sizes(s, -1))
3623 goto error;
3624 }
3625 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003626
Heiko Carstens25654092012-01-12 17:17:33 -08003627#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3628 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Laura Abbott149daaf2016-03-15 14:55:09 -07003629 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003630 /* Enable fast mode */
3631 s->flags |= __CMPXCHG_DOUBLE;
3632#endif
3633
David Rientjes3b89d7d2009-02-22 17:40:07 -08003634 /*
3635 * The larger the object size is, the more pages we want on the partial
3636 * list to avoid pounding the page allocator excessively.
3637 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003638 set_min_partial(s, ilog2(s->size) / 2);
3639
Wei Yange6d0e1d2017-07-06 15:36:34 -07003640 set_cpu_partial(s);
Christoph Lameter49e22582011-08-09 16:12:27 -05003641
Christoph Lameter81819f02007-05-06 14:49:36 -07003642#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003643 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003644#endif
Thomas Garnier210e7a42016-07-26 15:21:59 -07003645
3646 /* Initialize the pre-computed randomized freelist if slab is up */
3647 if (slab_state >= UP) {
3648 if (init_cache_random_seq(s))
3649 goto error;
3650 }
3651
Christoph Lameter55136592010-08-20 12:37:13 -05003652 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003653 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003654
Christoph Lameter55136592010-08-20 12:37:13 -05003655 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003656 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003657
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003658 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003659error:
3660 if (flags & SLAB_PANIC)
Alexey Dobriyan44065b22018-04-05 16:21:20 -07003661 panic("Cannot create slab %s size=%u realsize=%u order=%u offset=%u flags=%lx\n",
3662 s->name, s->size, s->size,
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -08003663 oo_order(s->oo), s->offset, (unsigned long)flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003664 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003665}
Christoph Lameter81819f02007-05-06 14:49:36 -07003666
Christoph Lameter33b12c32008-04-25 12:22:43 -07003667static void list_slab_objects(struct kmem_cache *s, struct page *page,
3668 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003669{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003670#ifdef CONFIG_SLUB_DEBUG
3671 void *addr = page_address(page);
3672 void *p;
Kees Cook6396bb22018-06-12 14:03:40 -07003673 unsigned long *map = kcalloc(BITS_TO_LONGS(page->objects),
3674 sizeof(long),
3675 GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003676 if (!map)
3677 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003678 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003679 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003680
Christoph Lameter5f80b132011-04-15 14:48:13 -05003681 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003682 for_each_object(p, s, addr, page->objects) {
3683
3684 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003685 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003686 print_tracking(s, p);
3687 }
3688 }
3689 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003690 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003691#endif
3692}
3693
Christoph Lameter81819f02007-05-06 14:49:36 -07003694/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003695 * Attempt to free all partial slabs on a node.
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003696 * This is called from __kmem_cache_shutdown(). We must take list_lock
3697 * because sysfs file might still access partial list after the shutdowning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003698 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003699static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003700{
Chris Wilson60398922016-08-10 16:27:58 -07003701 LIST_HEAD(discard);
Christoph Lameter81819f02007-05-06 14:49:36 -07003702 struct page *page, *h;
3703
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003704 BUG_ON(irqs_disabled());
3705 spin_lock_irq(&n->list_lock);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003706 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003707 if (!page->inuse) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003708 remove_partial(n, page);
Chris Wilson60398922016-08-10 16:27:58 -07003709 list_add(&page->lru, &discard);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003710 } else {
3711 list_slab_objects(s, page,
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003712 "Objects remaining in %s on __kmem_cache_shutdown()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003713 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003714 }
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003715 spin_unlock_irq(&n->list_lock);
Chris Wilson60398922016-08-10 16:27:58 -07003716
3717 list_for_each_entry_safe(page, h, &discard, lru)
3718 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003719}
3720
Shakeel Buttf9e13c02018-04-05 16:21:57 -07003721bool __kmem_cache_empty(struct kmem_cache *s)
3722{
3723 int node;
3724 struct kmem_cache_node *n;
3725
3726 for_each_kmem_cache_node(s, node, n)
3727 if (n->nr_partial || slabs_node(s, node))
3728 return false;
3729 return true;
3730}
3731
Christoph Lameter81819f02007-05-06 14:49:36 -07003732/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003733 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003734 */
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003735int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003736{
3737 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003738 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003739
3740 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003741 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003742 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003743 free_partial(s, n);
3744 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003745 return 1;
3746 }
Tejun Heobf5eb3d2017-02-22 15:41:11 -08003747 sysfs_slab_remove(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003748 return 0;
3749}
3750
Christoph Lameter81819f02007-05-06 14:49:36 -07003751/********************************************************************
3752 * Kmalloc subsystem
3753 *******************************************************************/
3754
Christoph Lameter81819f02007-05-06 14:49:36 -07003755static int __init setup_slub_min_order(char *str)
3756{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003757 get_option(&str, (int *)&slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003758
3759 return 1;
3760}
3761
3762__setup("slub_min_order=", setup_slub_min_order);
3763
3764static int __init setup_slub_max_order(char *str)
3765{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003766 get_option(&str, (int *)&slub_max_order);
3767 slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003768
3769 return 1;
3770}
3771
3772__setup("slub_max_order=", setup_slub_max_order);
3773
3774static int __init setup_slub_min_objects(char *str)
3775{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003776 get_option(&str, (int *)&slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003777
3778 return 1;
3779}
3780
3781__setup("slub_min_objects=", setup_slub_min_objects);
3782
Christoph Lameter81819f02007-05-06 14:49:36 -07003783void *__kmalloc(size_t size, gfp_t flags)
3784{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003785 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003786 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003787
Christoph Lameter95a05b42013-01-10 19:14:19 +00003788 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003789 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003790
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003791 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003792
3793 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003794 return s;
3795
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003796 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003797
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003798 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003799
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003800 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003801
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003802 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003803}
3804EXPORT_SYMBOL(__kmalloc);
3805
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003806#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003807static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3808{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003809 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003810 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003811
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -08003812 flags |= __GFP_COMP;
Vladimir Davydov49491482016-07-26 15:24:24 -07003813 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003814 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003815 ptr = page_address(page);
3816
Roman Bobnievd56791b2013-10-08 15:58:57 -07003817 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003818 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003819}
3820
Christoph Lameter81819f02007-05-06 14:49:36 -07003821void *__kmalloc_node(size_t size, gfp_t flags, int node)
3822{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003823 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003824 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003825
Christoph Lameter95a05b42013-01-10 19:14:19 +00003826 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003827 ret = kmalloc_large_node(size, flags, node);
3828
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003829 trace_kmalloc_node(_RET_IP_, ret,
3830 size, PAGE_SIZE << get_order(size),
3831 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003832
3833 return ret;
3834 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003835
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003836 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003837
3838 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003839 return s;
3840
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003841 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003842
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003843 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003844
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003845 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003846
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003847 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003848}
3849EXPORT_SYMBOL(__kmalloc_node);
3850#endif
3851
Kees Cooked18adc2016-06-23 15:24:05 -07003852#ifdef CONFIG_HARDENED_USERCOPY
3853/*
Kees Cookafcc90f82018-01-10 15:17:01 -08003854 * Rejects incorrectly sized objects and objects that are to be copied
3855 * to/from userspace but do not fall entirely within the containing slab
3856 * cache's usercopy region.
Kees Cooked18adc2016-06-23 15:24:05 -07003857 *
3858 * Returns NULL if check passes, otherwise const char * to name of cache
3859 * to indicate an error.
3860 */
Kees Cookf4e6e282018-01-10 14:48:22 -08003861void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
3862 bool to_user)
Kees Cooked18adc2016-06-23 15:24:05 -07003863{
3864 struct kmem_cache *s;
Alexey Dobriyan44065b22018-04-05 16:21:20 -07003865 unsigned int offset;
Kees Cooked18adc2016-06-23 15:24:05 -07003866 size_t object_size;
3867
3868 /* Find object and usable object size. */
3869 s = page->slab_cache;
Kees Cooked18adc2016-06-23 15:24:05 -07003870
3871 /* Reject impossible pointers. */
3872 if (ptr < page_address(page))
Kees Cookf4e6e282018-01-10 14:48:22 -08003873 usercopy_abort("SLUB object not in SLUB page?!", NULL,
3874 to_user, 0, n);
Kees Cooked18adc2016-06-23 15:24:05 -07003875
3876 /* Find offset within object. */
3877 offset = (ptr - page_address(page)) % s->size;
3878
3879 /* Adjust for redzone and reject if within the redzone. */
3880 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3881 if (offset < s->red_left_pad)
Kees Cookf4e6e282018-01-10 14:48:22 -08003882 usercopy_abort("SLUB object in left red zone",
3883 s->name, to_user, offset, n);
Kees Cooked18adc2016-06-23 15:24:05 -07003884 offset -= s->red_left_pad;
3885 }
3886
Kees Cookafcc90f82018-01-10 15:17:01 -08003887 /* Allow address range falling entirely within usercopy region. */
3888 if (offset >= s->useroffset &&
3889 offset - s->useroffset <= s->usersize &&
3890 n <= s->useroffset - offset + s->usersize)
Kees Cookf4e6e282018-01-10 14:48:22 -08003891 return;
Kees Cooked18adc2016-06-23 15:24:05 -07003892
Kees Cookafcc90f82018-01-10 15:17:01 -08003893 /*
3894 * If the copy is still within the allocated object, produce
3895 * a warning instead of rejecting the copy. This is intended
3896 * to be a temporary method to find any missing usercopy
3897 * whitelists.
3898 */
3899 object_size = slab_ksize(s);
Kees Cook2d891fb2017-11-30 13:04:32 -08003900 if (usercopy_fallback &&
3901 offset <= object_size && n <= object_size - offset) {
Kees Cookafcc90f82018-01-10 15:17:01 -08003902 usercopy_warn("SLUB object", s->name, to_user, offset, n);
3903 return;
3904 }
3905
Kees Cookf4e6e282018-01-10 14:48:22 -08003906 usercopy_abort("SLUB object", s->name, to_user, offset, n);
Kees Cooked18adc2016-06-23 15:24:05 -07003907}
3908#endif /* CONFIG_HARDENED_USERCOPY */
3909
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003910static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003911{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003912 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003913
Christoph Lameteref8b4522007-10-16 01:24:46 -07003914 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003915 return 0;
3916
Vegard Nossum294a80a2007-12-04 23:45:30 -08003917 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003918
Pekka Enberg76994412008-05-22 19:22:25 +03003919 if (unlikely(!PageSlab(page))) {
3920 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003921 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003922 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003923
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003924 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003925}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003926
3927size_t ksize(const void *object)
3928{
3929 size_t size = __ksize(object);
3930 /* We assume that ksize callers could use whole allocated area,
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07003931 * so we need to unpoison this area.
3932 */
3933 kasan_unpoison_shadow(object, size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003934 return size;
3935}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003936EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003937
3938void kfree(const void *x)
3939{
Christoph Lameter81819f02007-05-06 14:49:36 -07003940 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003941 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003942
Pekka Enberg2121db72009-03-25 11:05:57 +02003943 trace_kfree(_RET_IP_, x);
3944
Satyam Sharma2408c552007-10-16 01:24:44 -07003945 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003946 return;
3947
Christoph Lameterb49af682007-05-06 14:49:41 -07003948 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003949 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003950 BUG_ON(!PageCompound(page));
Dmitry Vyukov47adccc2018-02-06 15:36:23 -08003951 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003952 __free_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003953 return;
3954 }
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003955 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003956}
3957EXPORT_SYMBOL(kfree);
3958
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003959#define SHRINK_PROMOTE_MAX 32
3960
Christoph Lameter2086d262007-05-06 14:49:46 -07003961/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003962 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3963 * up most to the head of the partial lists. New allocations will then
3964 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003965 *
3966 * The slabs with the least items are placed last. This results in them
3967 * being allocated from last increasing the chance that the last objects
3968 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003969 */
Tejun Heoc9fc5862017-02-22 15:41:27 -08003970int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07003971{
3972 int node;
3973 int i;
3974 struct kmem_cache_node *n;
3975 struct page *page;
3976 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003977 struct list_head discard;
3978 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003979 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003980 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003981
Christoph Lameter2086d262007-05-06 14:49:46 -07003982 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003983 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003984 INIT_LIST_HEAD(&discard);
3985 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3986 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003987
3988 spin_lock_irqsave(&n->list_lock, flags);
3989
3990 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003991 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003992 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003993 * Note that concurrent frees may occur while we hold the
3994 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003995 */
3996 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003997 int free = page->objects - page->inuse;
3998
3999 /* Do not reread page->inuse */
4000 barrier();
4001
4002 /* We do not keep full slabs on the list */
4003 BUG_ON(free <= 0);
4004
4005 if (free == page->objects) {
4006 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004007 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004008 } else if (free <= SHRINK_PROMOTE_MAX)
4009 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07004010 }
4011
Christoph Lameter2086d262007-05-06 14:49:46 -07004012 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004013 * Promote the slabs filled up most to the head of the
4014 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07004015 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004016 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4017 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07004018
Christoph Lameter2086d262007-05-06 14:49:46 -07004019 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004020
4021 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004022 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004023 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08004024
4025 if (slabs_node(s, node))
4026 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07004027 }
4028
Vladimir Davydovce3712d2015-02-12 14:59:44 -08004029 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07004030}
Christoph Lameter2086d262007-05-06 14:49:46 -07004031
Tejun Heoc9fc5862017-02-22 15:41:27 -08004032#ifdef CONFIG_MEMCG
Tejun Heo01fb58b2017-02-22 15:41:30 -08004033static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s)
4034{
Tejun Heo50862ce72017-02-22 15:41:33 -08004035 /*
4036 * Called with all the locks held after a sched RCU grace period.
4037 * Even if @s becomes empty after shrinking, we can't know that @s
4038 * doesn't have allocations already in-flight and thus can't
4039 * destroy @s until the associated memcg is released.
4040 *
4041 * However, let's remove the sysfs files for empty caches here.
4042 * Each cache has a lot of interface files which aren't
4043 * particularly useful for empty draining caches; otherwise, we can
4044 * easily end up with millions of unnecessary sysfs files on
4045 * systems which have a lot of memory and transient cgroups.
4046 */
4047 if (!__kmem_cache_shrink(s))
4048 sysfs_slab_remove(s);
Tejun Heo01fb58b2017-02-22 15:41:30 -08004049}
4050
Tejun Heoc9fc5862017-02-22 15:41:27 -08004051void __kmemcg_cache_deactivate(struct kmem_cache *s)
4052{
4053 /*
4054 * Disable empty slabs caching. Used to avoid pinning offline
4055 * memory cgroups by kmem pages that can be freed.
4056 */
Wei Yange6d0e1d2017-07-06 15:36:34 -07004057 slub_set_cpu_partial(s, 0);
Tejun Heoc9fc5862017-02-22 15:41:27 -08004058 s->min_partial = 0;
4059
4060 /*
4061 * s->cpu_partial is checked locklessly (see put_cpu_partial), so
Tejun Heo01fb58b2017-02-22 15:41:30 -08004062 * we have to make sure the change is visible before shrinking.
Tejun Heoc9fc5862017-02-22 15:41:27 -08004063 */
Tejun Heo01fb58b2017-02-22 15:41:30 -08004064 slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu);
Tejun Heoc9fc5862017-02-22 15:41:27 -08004065}
4066#endif
4067
Yasunori Gotob9049e22007-10-21 16:41:37 -07004068static int slab_mem_going_offline_callback(void *arg)
4069{
4070 struct kmem_cache *s;
4071
Christoph Lameter18004c52012-07-06 15:25:12 -05004072 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004073 list_for_each_entry(s, &slab_caches, list)
Tejun Heoc9fc5862017-02-22 15:41:27 -08004074 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05004075 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004076
4077 return 0;
4078}
4079
4080static void slab_mem_offline_callback(void *arg)
4081{
4082 struct kmem_cache_node *n;
4083 struct kmem_cache *s;
4084 struct memory_notify *marg = arg;
4085 int offline_node;
4086
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004087 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004088
4089 /*
4090 * If the node still has available memory. we need kmem_cache_node
4091 * for it yet.
4092 */
4093 if (offline_node < 0)
4094 return;
4095
Christoph Lameter18004c52012-07-06 15:25:12 -05004096 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004097 list_for_each_entry(s, &slab_caches, list) {
4098 n = get_node(s, offline_node);
4099 if (n) {
4100 /*
4101 * if n->nr_slabs > 0, slabs still exist on the node
4102 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05004103 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07004104 * callback. So, we must fail.
4105 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03004106 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07004107
4108 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05004109 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004110 }
4111 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004112 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004113}
4114
4115static int slab_mem_going_online_callback(void *arg)
4116{
4117 struct kmem_cache_node *n;
4118 struct kmem_cache *s;
4119 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004120 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004121 int ret = 0;
4122
4123 /*
4124 * If the node's memory is already available, then kmem_cache_node is
4125 * already created. Nothing to do.
4126 */
4127 if (nid < 0)
4128 return 0;
4129
4130 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07004131 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07004132 * allocate a kmem_cache_node structure in order to bring the node
4133 * online.
4134 */
Christoph Lameter18004c52012-07-06 15:25:12 -05004135 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004136 list_for_each_entry(s, &slab_caches, list) {
4137 /*
4138 * XXX: kmem_cache_alloc_node will fallback to other nodes
4139 * since memory is not yet available from the node that
4140 * is brought up.
4141 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05004142 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004143 if (!n) {
4144 ret = -ENOMEM;
4145 goto out;
4146 }
Joonsoo Kim40534972012-05-11 00:50:47 +09004147 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004148 s->node[nid] = n;
4149 }
4150out:
Christoph Lameter18004c52012-07-06 15:25:12 -05004151 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004152 return ret;
4153}
4154
4155static int slab_memory_callback(struct notifier_block *self,
4156 unsigned long action, void *arg)
4157{
4158 int ret = 0;
4159
4160 switch (action) {
4161 case MEM_GOING_ONLINE:
4162 ret = slab_mem_going_online_callback(arg);
4163 break;
4164 case MEM_GOING_OFFLINE:
4165 ret = slab_mem_going_offline_callback(arg);
4166 break;
4167 case MEM_OFFLINE:
4168 case MEM_CANCEL_ONLINE:
4169 slab_mem_offline_callback(arg);
4170 break;
4171 case MEM_ONLINE:
4172 case MEM_CANCEL_OFFLINE:
4173 break;
4174 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08004175 if (ret)
4176 ret = notifier_from_errno(ret);
4177 else
4178 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004179 return ret;
4180}
4181
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004182static struct notifier_block slab_memory_callback_nb = {
4183 .notifier_call = slab_memory_callback,
4184 .priority = SLAB_CALLBACK_PRI,
4185};
Yasunori Gotob9049e22007-10-21 16:41:37 -07004186
Christoph Lameter81819f02007-05-06 14:49:36 -07004187/********************************************************************
4188 * Basic setup of slabs
4189 *******************************************************************/
4190
Christoph Lameter51df1142010-08-20 12:37:15 -05004191/*
4192 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004193 * the page allocator. Allocate them properly then fix up the pointers
4194 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05004195 */
4196
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004197static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05004198{
4199 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004200 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004201 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05004202
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004203 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05004204
Glauber Costa7d557b32013-02-22 20:20:00 +04004205 /*
4206 * This runs very early, and only the boot processor is supposed to be
4207 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4208 * IPIs around.
4209 */
4210 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004211 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05004212 struct page *p;
4213
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004214 list_for_each_entry(p, &n->partial, lru)
4215 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004216
Li Zefan607bf322011-04-12 15:22:26 +08004217#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004218 list_for_each_entry(p, &n->full, lru)
4219 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004220#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05004221 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08004222 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004223 list_add(&s->list, &slab_caches);
Tejun Heo510ded32017-02-22 15:41:24 -08004224 memcg_link_cache(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004225 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004226}
4227
Christoph Lameter81819f02007-05-06 14:49:36 -07004228void __init kmem_cache_init(void)
4229{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004230 static __initdata struct kmem_cache boot_kmem_cache,
4231 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05004232
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08004233 if (debug_guardpage_minorder())
4234 slub_max_order = 0;
4235
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004236 kmem_cache_node = &boot_kmem_cache_node;
4237 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05004238
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004239 create_boot_cache(kmem_cache_node, "kmem_cache_node",
David Windsor8eb82842017-06-10 22:50:28 -04004240 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004241
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004242 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07004243
4244 /* Able to allocate the per node structures */
4245 slab_state = PARTIAL;
4246
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004247 create_boot_cache(kmem_cache, "kmem_cache",
4248 offsetof(struct kmem_cache, node) +
4249 nr_node_ids * sizeof(struct kmem_cache_node *),
David Windsor8eb82842017-06-10 22:50:28 -04004250 SLAB_HWCACHE_ALIGN, 0, 0);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004251
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004252 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004253 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05004254
4255 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07004256 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004257 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07004258
Thomas Garnier210e7a42016-07-26 15:21:59 -07004259 /* Setup random freelists for each cache */
4260 init_freelist_randomization();
4261
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02004262 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4263 slub_cpu_dead);
Christoph Lameter81819f02007-05-06 14:49:36 -07004264
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004265 pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004266 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07004267 slub_min_order, slub_max_order, slub_min_objects,
4268 nr_cpu_ids, nr_node_ids);
4269}
4270
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004271void __init kmem_cache_init_late(void)
4272{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004273}
4274
Glauber Costa2633d7a2012-12-18 14:22:34 -08004275struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -07004276__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08004277 slab_flags_t flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07004278{
Vladimir Davydov426589f2015-02-12 14:59:23 -08004279 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07004280
Vladimir Davydova44cb942014-04-07 15:39:23 -07004281 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004282 if (s) {
4283 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004284
Christoph Lameter81819f02007-05-06 14:49:36 -07004285 /*
4286 * Adjust the object sizes so that we clear
4287 * the complete object on kzalloc.
4288 */
Alexey Dobriyan1b473f22018-04-05 16:21:17 -07004289 s->object_size = max(s->object_size, size);
Alexey Dobriyan52ee6d72018-04-05 16:21:06 -07004290 s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08004291
Vladimir Davydov426589f2015-02-12 14:59:23 -08004292 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004293 c->object_size = s->object_size;
Alexey Dobriyan52ee6d72018-04-05 16:21:06 -07004294 c->inuse = max(c->inuse, ALIGN(size, sizeof(void *)));
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004295 }
4296
David Rientjes7b8f3b62008-12-17 22:09:46 -08004297 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08004298 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00004299 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08004300 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07004301 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08004302
Christoph Lametercbb79692012-09-05 00:18:32 +00004303 return s;
4304}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004305
Alexey Dobriyand50112e2017-11-15 17:32:18 -08004306int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00004307{
Pekka Enbergaac3a162012-09-05 12:07:44 +03004308 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004309
Pekka Enbergaac3a162012-09-05 12:07:44 +03004310 err = kmem_cache_open(s, flags);
4311 if (err)
4312 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004313
Christoph Lameter45530c42012-11-28 16:23:07 +00004314 /* Mutex is not taken during early boot */
4315 if (slab_state <= UP)
4316 return 0;
4317
Glauber Costa107dab52012-12-18 14:23:05 -08004318 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004319 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004320 if (err)
Dmitry Safonov52b4b952016-02-17 13:11:37 -08004321 __kmem_cache_release(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004322
4323 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004324}
Christoph Lameter81819f02007-05-06 14:49:36 -07004325
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004326void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004327{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004328 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004329 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004330
Christoph Lameter95a05b42013-01-10 19:14:19 +00004331 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004332 return kmalloc_large(size, gfpflags);
4333
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004334 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004335
Satyam Sharma2408c552007-10-16 01:24:44 -07004336 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004337 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004338
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004339 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004340
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004341 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004342 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004343
4344 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004345}
4346
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004347#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004348void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004349 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004350{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004351 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004352 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004353
Christoph Lameter95a05b42013-01-10 19:14:19 +00004354 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004355 ret = kmalloc_large_node(size, gfpflags, node);
4356
4357 trace_kmalloc_node(caller, ret,
4358 size, PAGE_SIZE << get_order(size),
4359 gfpflags, node);
4360
4361 return ret;
4362 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004363
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004364 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004365
Satyam Sharma2408c552007-10-16 01:24:44 -07004366 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004367 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004368
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004369 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004370
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004371 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004372 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004373
4374 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004375}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004376#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004377
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004378#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004379static int count_inuse(struct page *page)
4380{
4381 return page->inuse;
4382}
4383
4384static int count_total(struct page *page)
4385{
4386 return page->objects;
4387}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004388#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004389
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004390#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004391static int validate_slab(struct kmem_cache *s, struct page *page,
4392 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004393{
4394 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004395 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004396
4397 if (!check_slab(s, page) ||
4398 !on_freelist(s, page, NULL))
4399 return 0;
4400
4401 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004402 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004403
Christoph Lameter5f80b132011-04-15 14:48:13 -05004404 get_map(s, page, map);
4405 for_each_object(p, s, addr, page->objects) {
4406 if (test_bit(slab_index(p, s, addr), map))
4407 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4408 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004409 }
4410
Christoph Lameter224a88b2008-04-14 19:11:31 +03004411 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004412 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004413 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004414 return 0;
4415 return 1;
4416}
4417
Christoph Lameter434e2452007-07-17 04:03:30 -07004418static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4419 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004420{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004421 slab_lock(page);
4422 validate_slab(s, page, map);
4423 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004424}
4425
Christoph Lameter434e2452007-07-17 04:03:30 -07004426static int validate_slab_node(struct kmem_cache *s,
4427 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004428{
4429 unsigned long count = 0;
4430 struct page *page;
4431 unsigned long flags;
4432
4433 spin_lock_irqsave(&n->list_lock, flags);
4434
4435 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004436 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004437 count++;
4438 }
4439 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004440 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4441 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004442
4443 if (!(s->flags & SLAB_STORE_USER))
4444 goto out;
4445
4446 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004447 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004448 count++;
4449 }
4450 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004451 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4452 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004453
4454out:
4455 spin_unlock_irqrestore(&n->list_lock, flags);
4456 return count;
4457}
4458
Christoph Lameter434e2452007-07-17 04:03:30 -07004459static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004460{
4461 int node;
4462 unsigned long count = 0;
Kees Cook6da2ec52018-06-12 13:55:00 -07004463 unsigned long *map = kmalloc_array(BITS_TO_LONGS(oo_objects(s->max)),
4464 sizeof(unsigned long),
4465 GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004466 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07004467
4468 if (!map)
4469 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004470
4471 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004472 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07004473 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07004474 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004475 return count;
4476}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004477/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004478 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004479 * and freed.
4480 */
4481
4482struct location {
4483 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004484 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004485 long long sum_time;
4486 long min_time;
4487 long max_time;
4488 long min_pid;
4489 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304490 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004491 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004492};
4493
4494struct loc_track {
4495 unsigned long max;
4496 unsigned long count;
4497 struct location *loc;
4498};
4499
4500static void free_loc_track(struct loc_track *t)
4501{
4502 if (t->max)
4503 free_pages((unsigned long)t->loc,
4504 get_order(sizeof(struct location) * t->max));
4505}
4506
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004507static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004508{
4509 struct location *l;
4510 int order;
4511
Christoph Lameter88a420e2007-05-06 14:49:45 -07004512 order = get_order(sizeof(struct location) * max);
4513
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004514 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004515 if (!l)
4516 return 0;
4517
4518 if (t->count) {
4519 memcpy(l, t->loc, sizeof(struct location) * t->count);
4520 free_loc_track(t);
4521 }
4522 t->max = max;
4523 t->loc = l;
4524 return 1;
4525}
4526
4527static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004528 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004529{
4530 long start, end, pos;
4531 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004532 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004533 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004534
4535 start = -1;
4536 end = t->count;
4537
4538 for ( ; ; ) {
4539 pos = start + (end - start + 1) / 2;
4540
4541 /*
4542 * There is nothing at "end". If we end up there
4543 * we need to add something to before end.
4544 */
4545 if (pos == end)
4546 break;
4547
4548 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004549 if (track->addr == caddr) {
4550
4551 l = &t->loc[pos];
4552 l->count++;
4553 if (track->when) {
4554 l->sum_time += age;
4555 if (age < l->min_time)
4556 l->min_time = age;
4557 if (age > l->max_time)
4558 l->max_time = age;
4559
4560 if (track->pid < l->min_pid)
4561 l->min_pid = track->pid;
4562 if (track->pid > l->max_pid)
4563 l->max_pid = track->pid;
4564
Rusty Russell174596a2009-01-01 10:12:29 +10304565 cpumask_set_cpu(track->cpu,
4566 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004567 }
4568 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004569 return 1;
4570 }
4571
Christoph Lameter45edfa52007-05-09 02:32:45 -07004572 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004573 end = pos;
4574 else
4575 start = pos;
4576 }
4577
4578 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004579 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004580 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004581 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004582 return 0;
4583
4584 l = t->loc + pos;
4585 if (pos < t->count)
4586 memmove(l + 1, l,
4587 (t->count - pos) * sizeof(struct location));
4588 t->count++;
4589 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004590 l->addr = track->addr;
4591 l->sum_time = age;
4592 l->min_time = age;
4593 l->max_time = age;
4594 l->min_pid = track->pid;
4595 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304596 cpumask_clear(to_cpumask(l->cpus));
4597 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004598 nodes_clear(l->nodes);
4599 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004600 return 1;
4601}
4602
4603static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004604 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004605 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004606{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004607 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004608 void *p;
4609
Christoph Lameter39b26462008-04-14 19:11:30 +03004610 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004611 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004612
Christoph Lameter224a88b2008-04-14 19:11:31 +03004613 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004614 if (!test_bit(slab_index(p, s, addr), map))
4615 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004616}
4617
4618static int list_locations(struct kmem_cache *s, char *buf,
4619 enum track_item alloc)
4620{
Harvey Harrisone374d482008-01-31 15:20:50 -08004621 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004622 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004623 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004624 int node;
Kees Cook6da2ec52018-06-12 13:55:00 -07004625 unsigned long *map = kmalloc_array(BITS_TO_LONGS(oo_objects(s->max)),
4626 sizeof(unsigned long),
4627 GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004628 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004629
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004630 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
Michal Hocko0ee931c2017-09-13 16:28:29 -07004631 GFP_KERNEL)) {
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004632 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004633 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004634 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004635 /* Push back cpu slabs */
4636 flush_all(s);
4637
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004638 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004639 unsigned long flags;
4640 struct page *page;
4641
Christoph Lameter9e869432007-08-22 14:01:56 -07004642 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004643 continue;
4644
4645 spin_lock_irqsave(&n->list_lock, flags);
4646 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004647 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004648 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004649 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004650 spin_unlock_irqrestore(&n->list_lock, flags);
4651 }
4652
4653 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004654 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004655
Hugh Dickins9c246242008-12-09 13:14:27 -08004656 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004657 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004658 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004659
4660 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004661 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004662 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004663 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004664
4665 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004666 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004667 l->min_time,
4668 (long)div_u64(l->sum_time, l->count),
4669 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004670 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004671 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004672 l->min_time);
4673
4674 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004675 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004676 l->min_pid, l->max_pid);
4677 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004678 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004679 l->min_pid);
4680
Rusty Russell174596a2009-01-01 10:12:29 +10304681 if (num_online_cpus() > 1 &&
4682 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004683 len < PAGE_SIZE - 60)
4684 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4685 " cpus=%*pbl",
4686 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004687
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004688 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004689 len < PAGE_SIZE - 60)
4690 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4691 " nodes=%*pbl",
4692 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004693
Harvey Harrisone374d482008-01-31 15:20:50 -08004694 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004695 }
4696
4697 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004698 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004699 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004700 len += sprintf(buf, "No data\n");
4701 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004702}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004703#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004704
Christoph Lametera5a84752010-10-05 13:57:27 -05004705#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004706static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004707{
4708 u8 *p;
4709
Christoph Lameter95a05b42013-01-10 19:14:19 +00004710 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004711
Fabian Frederickf9f58282014-06-04 16:06:34 -07004712 pr_err("SLUB resiliency testing\n");
4713 pr_err("-----------------------\n");
4714 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004715
4716 p = kzalloc(16, GFP_KERNEL);
4717 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004718 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4719 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004720
4721 validate_slab_cache(kmalloc_caches[4]);
4722
4723 /* Hmmm... The next two are dangerous */
4724 p = kzalloc(32, GFP_KERNEL);
4725 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004726 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4727 p);
4728 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004729
4730 validate_slab_cache(kmalloc_caches[5]);
4731 p = kzalloc(64, GFP_KERNEL);
4732 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4733 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004734 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4735 p);
4736 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004737 validate_slab_cache(kmalloc_caches[6]);
4738
Fabian Frederickf9f58282014-06-04 16:06:34 -07004739 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004740 p = kzalloc(128, GFP_KERNEL);
4741 kfree(p);
4742 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004743 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004744 validate_slab_cache(kmalloc_caches[7]);
4745
4746 p = kzalloc(256, GFP_KERNEL);
4747 kfree(p);
4748 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004749 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004750 validate_slab_cache(kmalloc_caches[8]);
4751
4752 p = kzalloc(512, GFP_KERNEL);
4753 kfree(p);
4754 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004755 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004756 validate_slab_cache(kmalloc_caches[9]);
4757}
4758#else
4759#ifdef CONFIG_SYSFS
4760static void resiliency_test(void) {};
4761#endif
4762#endif
4763
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004764#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004765enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004766 SL_ALL, /* All slabs */
4767 SL_PARTIAL, /* Only partially allocated slabs */
4768 SL_CPU, /* Only slabs used for cpu caches */
4769 SL_OBJECTS, /* Determine allocated objects not slabs */
4770 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004771};
4772
Christoph Lameter205ab992008-04-14 19:11:40 +03004773#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004774#define SO_PARTIAL (1 << SL_PARTIAL)
4775#define SO_CPU (1 << SL_CPU)
4776#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004777#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004778
Tejun Heo1663f262017-02-22 15:41:39 -08004779#ifdef CONFIG_MEMCG
4780static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4781
4782static int __init setup_slub_memcg_sysfs(char *str)
4783{
4784 int v;
4785
4786 if (get_option(&str, &v) > 0)
4787 memcg_sysfs_enabled = v;
4788
4789 return 1;
4790}
4791
4792__setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4793#endif
4794
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004795static ssize_t show_slab_objects(struct kmem_cache *s,
4796 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004797{
4798 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004799 int node;
4800 int x;
4801 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004802
Kees Cook6396bb22018-06-12 14:03:40 -07004803 nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004804 if (!nodes)
4805 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004806
Christoph Lameter205ab992008-04-14 19:11:40 +03004807 if (flags & SO_CPU) {
4808 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004809
Christoph Lameter205ab992008-04-14 19:11:40 +03004810 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004811 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4812 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004813 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004814 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004815
Jason Low4db0c3c2015-04-15 16:14:08 -07004816 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004817 if (!page)
4818 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004819
Christoph Lameterec3ab082012-05-09 10:09:56 -05004820 node = page_to_nid(page);
4821 if (flags & SO_TOTAL)
4822 x = page->objects;
4823 else if (flags & SO_OBJECTS)
4824 x = page->inuse;
4825 else
4826 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004827
Christoph Lameterec3ab082012-05-09 10:09:56 -05004828 total += x;
4829 nodes[node] += x;
4830
Wei Yanga93cf072017-07-06 15:36:31 -07004831 page = slub_percpu_partial_read_once(c);
Christoph Lameter49e22582011-08-09 16:12:27 -05004832 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004833 node = page_to_nid(page);
4834 if (flags & SO_TOTAL)
4835 WARN_ON_ONCE(1);
4836 else if (flags & SO_OBJECTS)
4837 WARN_ON_ONCE(1);
4838 else
4839 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004840 total += x;
4841 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004842 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004843 }
4844 }
4845
Qian Caibb6932c2019-10-14 14:11:51 -07004846 /*
4847 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
4848 * already held which will conflict with an existing lock order:
4849 *
4850 * mem_hotplug_lock->slab_mutex->kernfs_mutex
4851 *
4852 * We don't really need mem_hotplug_lock (to hold off
4853 * slab_mem_going_offline_callback) here because slab's memory hot
4854 * unplug code doesn't destroy the kmem_cache->node[] data.
4855 */
4856
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004857#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004858 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004859 struct kmem_cache_node *n;
4860
4861 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004862
Chen Gangd0e0ac92013-07-15 09:05:29 +08004863 if (flags & SO_TOTAL)
4864 x = atomic_long_read(&n->total_objects);
4865 else if (flags & SO_OBJECTS)
4866 x = atomic_long_read(&n->total_objects) -
4867 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004868 else
4869 x = atomic_long_read(&n->nr_slabs);
4870 total += x;
4871 nodes[node] += x;
4872 }
4873
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004874 } else
4875#endif
4876 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004877 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004878
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004879 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004880 if (flags & SO_TOTAL)
4881 x = count_partial(n, count_total);
4882 else if (flags & SO_OBJECTS)
4883 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004884 else
4885 x = n->nr_partial;
4886 total += x;
4887 nodes[node] += x;
4888 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004889 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004890 x = sprintf(buf, "%lu", total);
4891#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004892 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004893 if (nodes[node])
4894 x += sprintf(buf + x, " N%d=%lu",
4895 node, nodes[node]);
4896#endif
4897 kfree(nodes);
4898 return x + sprintf(buf + x, "\n");
4899}
4900
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004901#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004902static int any_slab_objects(struct kmem_cache *s)
4903{
4904 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004905 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004906
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004907 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004908 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004909 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004910
Christoph Lameter81819f02007-05-06 14:49:36 -07004911 return 0;
4912}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004913#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004914
4915#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004916#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004917
4918struct slab_attribute {
4919 struct attribute attr;
4920 ssize_t (*show)(struct kmem_cache *s, char *buf);
4921 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4922};
4923
4924#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004925 static struct slab_attribute _name##_attr = \
4926 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004927
4928#define SLAB_ATTR(_name) \
4929 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004930 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004931
Christoph Lameter81819f02007-05-06 14:49:36 -07004932static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4933{
Alexey Dobriyan44065b22018-04-05 16:21:20 -07004934 return sprintf(buf, "%u\n", s->size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004935}
4936SLAB_ATTR_RO(slab_size);
4937
4938static ssize_t align_show(struct kmem_cache *s, char *buf)
4939{
Alexey Dobriyan3a3791e2018-04-05 16:21:02 -07004940 return sprintf(buf, "%u\n", s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07004941}
4942SLAB_ATTR_RO(align);
4943
4944static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4945{
Alexey Dobriyan1b473f22018-04-05 16:21:17 -07004946 return sprintf(buf, "%u\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004947}
4948SLAB_ATTR_RO(object_size);
4949
4950static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4951{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004952 return sprintf(buf, "%u\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004953}
4954SLAB_ATTR_RO(objs_per_slab);
4955
Christoph Lameter06b285d2008-04-14 19:11:41 +03004956static ssize_t order_store(struct kmem_cache *s,
4957 const char *buf, size_t length)
4958{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004959 unsigned int order;
Christoph Lameter0121c6192008-04-29 16:11:12 -07004960 int err;
4961
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004962 err = kstrtouint(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004963 if (err)
4964 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004965
4966 if (order > slub_max_order || order < slub_min_order)
4967 return -EINVAL;
4968
4969 calculate_sizes(s, order);
4970 return length;
4971}
4972
Christoph Lameter81819f02007-05-06 14:49:36 -07004973static ssize_t order_show(struct kmem_cache *s, char *buf)
4974{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004975 return sprintf(buf, "%u\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004976}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004977SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004978
David Rientjes73d342b2009-02-22 17:40:09 -08004979static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4980{
4981 return sprintf(buf, "%lu\n", s->min_partial);
4982}
4983
4984static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4985 size_t length)
4986{
4987 unsigned long min;
4988 int err;
4989
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004990 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004991 if (err)
4992 return err;
4993
David Rientjesc0bdb232009-02-25 09:16:35 +02004994 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004995 return length;
4996}
4997SLAB_ATTR(min_partial);
4998
Christoph Lameter49e22582011-08-09 16:12:27 -05004999static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5000{
Wei Yange6d0e1d2017-07-06 15:36:34 -07005001 return sprintf(buf, "%u\n", slub_cpu_partial(s));
Christoph Lameter49e22582011-08-09 16:12:27 -05005002}
5003
5004static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5005 size_t length)
5006{
Alexey Dobriyane5d99982018-04-05 16:21:10 -07005007 unsigned int objects;
Christoph Lameter49e22582011-08-09 16:12:27 -05005008 int err;
5009
Alexey Dobriyane5d99982018-04-05 16:21:10 -07005010 err = kstrtouint(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05005011 if (err)
5012 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09005013 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08005014 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05005015
Wei Yange6d0e1d2017-07-06 15:36:34 -07005016 slub_set_cpu_partial(s, objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05005017 flush_all(s);
5018 return length;
5019}
5020SLAB_ATTR(cpu_partial);
5021
Christoph Lameter81819f02007-05-06 14:49:36 -07005022static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5023{
Joe Perches62c70bc2011-01-13 15:45:52 -08005024 if (!s->ctor)
5025 return 0;
5026 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07005027}
5028SLAB_ATTR_RO(ctor);
5029
Christoph Lameter81819f02007-05-06 14:49:36 -07005030static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5031{
Gu Zheng4307c142014-08-06 16:04:51 -07005032 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005033}
5034SLAB_ATTR_RO(aliases);
5035
Christoph Lameter81819f02007-05-06 14:49:36 -07005036static ssize_t partial_show(struct kmem_cache *s, char *buf)
5037{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08005038 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07005039}
5040SLAB_ATTR_RO(partial);
5041
5042static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5043{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08005044 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07005045}
5046SLAB_ATTR_RO(cpu_slabs);
5047
5048static ssize_t objects_show(struct kmem_cache *s, char *buf)
5049{
Christoph Lameter205ab992008-04-14 19:11:40 +03005050 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07005051}
5052SLAB_ATTR_RO(objects);
5053
Christoph Lameter205ab992008-04-14 19:11:40 +03005054static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5055{
5056 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5057}
5058SLAB_ATTR_RO(objects_partial);
5059
Christoph Lameter49e22582011-08-09 16:12:27 -05005060static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5061{
5062 int objects = 0;
5063 int pages = 0;
5064 int cpu;
5065 int len;
5066
5067 for_each_online_cpu(cpu) {
Wei Yanga93cf072017-07-06 15:36:31 -07005068 struct page *page;
5069
5070 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
Christoph Lameter49e22582011-08-09 16:12:27 -05005071
5072 if (page) {
5073 pages += page->pages;
5074 objects += page->pobjects;
5075 }
5076 }
5077
5078 len = sprintf(buf, "%d(%d)", objects, pages);
5079
5080#ifdef CONFIG_SMP
5081 for_each_online_cpu(cpu) {
Wei Yanga93cf072017-07-06 15:36:31 -07005082 struct page *page;
5083
5084 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
Christoph Lameter49e22582011-08-09 16:12:27 -05005085
5086 if (page && len < PAGE_SIZE - 20)
5087 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5088 page->pobjects, page->pages);
5089 }
5090#endif
5091 return len + sprintf(buf + len, "\n");
5092}
5093SLAB_ATTR_RO(slabs_cpu_partial);
5094
Christoph Lameter81819f02007-05-06 14:49:36 -07005095static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5096{
5097 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5098}
5099
5100static ssize_t reclaim_account_store(struct kmem_cache *s,
5101 const char *buf, size_t length)
5102{
5103 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
5104 if (buf[0] == '1')
5105 s->flags |= SLAB_RECLAIM_ACCOUNT;
5106 return length;
5107}
5108SLAB_ATTR(reclaim_account);
5109
5110static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5111{
Christoph Lameter5af60832007-05-06 14:49:56 -07005112 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07005113}
5114SLAB_ATTR_RO(hwcache_align);
5115
5116#ifdef CONFIG_ZONE_DMA
5117static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5118{
5119 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5120}
5121SLAB_ATTR_RO(cache_dma);
5122#endif
5123
David Windsor8eb82842017-06-10 22:50:28 -04005124static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5125{
Alexey Dobriyan7bbdb812018-04-05 16:21:31 -07005126 return sprintf(buf, "%u\n", s->usersize);
David Windsor8eb82842017-06-10 22:50:28 -04005127}
5128SLAB_ATTR_RO(usersize);
5129
Christoph Lameter81819f02007-05-06 14:49:36 -07005130static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5131{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08005132 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
Christoph Lameter81819f02007-05-06 14:49:36 -07005133}
5134SLAB_ATTR_RO(destroy_by_rcu);
5135
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005136#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005137static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5138{
5139 return show_slab_objects(s, buf, SO_ALL);
5140}
5141SLAB_ATTR_RO(slabs);
5142
5143static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5144{
5145 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5146}
5147SLAB_ATTR_RO(total_objects);
5148
5149static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5150{
Laura Abbottbecfda62016-03-15 14:55:06 -07005151 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
Christoph Lametera5a84752010-10-05 13:57:27 -05005152}
5153
5154static ssize_t sanity_checks_store(struct kmem_cache *s,
5155 const char *buf, size_t length)
5156{
Laura Abbottbecfda62016-03-15 14:55:06 -07005157 s->flags &= ~SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005158 if (buf[0] == '1') {
5159 s->flags &= ~__CMPXCHG_DOUBLE;
Laura Abbottbecfda62016-03-15 14:55:06 -07005160 s->flags |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005161 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005162 return length;
5163}
5164SLAB_ATTR(sanity_checks);
5165
5166static ssize_t trace_show(struct kmem_cache *s, char *buf)
5167{
5168 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5169}
5170
5171static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5172 size_t length)
5173{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005174 /*
5175 * Tracing a merged cache is going to give confusing results
5176 * as well as cause other issues like converting a mergeable
5177 * cache into an umergeable one.
5178 */
5179 if (s->refcount > 1)
5180 return -EINVAL;
5181
Christoph Lametera5a84752010-10-05 13:57:27 -05005182 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005183 if (buf[0] == '1') {
5184 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05005185 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005186 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005187 return length;
5188}
5189SLAB_ATTR(trace);
5190
Christoph Lameter81819f02007-05-06 14:49:36 -07005191static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5192{
5193 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5194}
5195
5196static ssize_t red_zone_store(struct kmem_cache *s,
5197 const char *buf, size_t length)
5198{
5199 if (any_slab_objects(s))
5200 return -EBUSY;
5201
5202 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005203 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005204 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005205 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005206 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005207 return length;
5208}
5209SLAB_ATTR(red_zone);
5210
5211static ssize_t poison_show(struct kmem_cache *s, char *buf)
5212{
5213 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5214}
5215
5216static ssize_t poison_store(struct kmem_cache *s,
5217 const char *buf, size_t length)
5218{
5219 if (any_slab_objects(s))
5220 return -EBUSY;
5221
5222 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005223 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005224 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005225 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005226 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005227 return length;
5228}
5229SLAB_ATTR(poison);
5230
5231static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5232{
5233 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5234}
5235
5236static ssize_t store_user_store(struct kmem_cache *s,
5237 const char *buf, size_t length)
5238{
5239 if (any_slab_objects(s))
5240 return -EBUSY;
5241
5242 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005243 if (buf[0] == '1') {
5244 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07005245 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005246 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005247 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005248 return length;
5249}
5250SLAB_ATTR(store_user);
5251
Christoph Lameter53e15af2007-05-06 14:49:43 -07005252static ssize_t validate_show(struct kmem_cache *s, char *buf)
5253{
5254 return 0;
5255}
5256
5257static ssize_t validate_store(struct kmem_cache *s,
5258 const char *buf, size_t length)
5259{
Christoph Lameter434e2452007-07-17 04:03:30 -07005260 int ret = -EINVAL;
5261
5262 if (buf[0] == '1') {
5263 ret = validate_slab_cache(s);
5264 if (ret >= 0)
5265 ret = length;
5266 }
5267 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07005268}
5269SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05005270
5271static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5272{
5273 if (!(s->flags & SLAB_STORE_USER))
5274 return -ENOSYS;
5275 return list_locations(s, buf, TRACK_ALLOC);
5276}
5277SLAB_ATTR_RO(alloc_calls);
5278
5279static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5280{
5281 if (!(s->flags & SLAB_STORE_USER))
5282 return -ENOSYS;
5283 return list_locations(s, buf, TRACK_FREE);
5284}
5285SLAB_ATTR_RO(free_calls);
5286#endif /* CONFIG_SLUB_DEBUG */
5287
5288#ifdef CONFIG_FAILSLAB
5289static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5290{
5291 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5292}
5293
5294static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5295 size_t length)
5296{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005297 if (s->refcount > 1)
5298 return -EINVAL;
5299
Christoph Lametera5a84752010-10-05 13:57:27 -05005300 s->flags &= ~SLAB_FAILSLAB;
5301 if (buf[0] == '1')
5302 s->flags |= SLAB_FAILSLAB;
5303 return length;
5304}
5305SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005306#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07005307
Christoph Lameter2086d262007-05-06 14:49:46 -07005308static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5309{
5310 return 0;
5311}
5312
5313static ssize_t shrink_store(struct kmem_cache *s,
5314 const char *buf, size_t length)
5315{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005316 if (buf[0] == '1')
5317 kmem_cache_shrink(s);
5318 else
Christoph Lameter2086d262007-05-06 14:49:46 -07005319 return -EINVAL;
5320 return length;
5321}
5322SLAB_ATTR(shrink);
5323
Christoph Lameter81819f02007-05-06 14:49:36 -07005324#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005325static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07005326{
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005327 return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07005328}
5329
Christoph Lameter98246012008-01-07 23:20:26 -08005330static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07005331 const char *buf, size_t length)
5332{
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005333 unsigned int ratio;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005334 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005335
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005336 err = kstrtouint(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07005337 if (err)
5338 return err;
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005339 if (ratio > 100)
5340 return -ERANGE;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005341
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005342 s->remote_node_defrag_ratio = ratio * 10;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005343
Christoph Lameter81819f02007-05-06 14:49:36 -07005344 return length;
5345}
Christoph Lameter98246012008-01-07 23:20:26 -08005346SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005347#endif
5348
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005349#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005350static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5351{
5352 unsigned long sum = 0;
5353 int cpu;
5354 int len;
Kees Cook6da2ec52018-06-12 13:55:00 -07005355 int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005356
5357 if (!data)
5358 return -ENOMEM;
5359
5360 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005361 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005362
5363 data[cpu] = x;
5364 sum += x;
5365 }
5366
5367 len = sprintf(buf, "%lu", sum);
5368
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005369#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005370 for_each_online_cpu(cpu) {
5371 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005372 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005373 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005374#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005375 kfree(data);
5376 return len + sprintf(buf + len, "\n");
5377}
5378
David Rientjes78eb00c2009-10-15 02:20:22 -07005379static void clear_stat(struct kmem_cache *s, enum stat_item si)
5380{
5381 int cpu;
5382
5383 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005384 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005385}
5386
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005387#define STAT_ATTR(si, text) \
5388static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5389{ \
5390 return show_stat(s, buf, si); \
5391} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005392static ssize_t text##_store(struct kmem_cache *s, \
5393 const char *buf, size_t length) \
5394{ \
5395 if (buf[0] != '0') \
5396 return -EINVAL; \
5397 clear_stat(s, si); \
5398 return length; \
5399} \
5400SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005401
5402STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5403STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5404STAT_ATTR(FREE_FASTPATH, free_fastpath);
5405STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5406STAT_ATTR(FREE_FROZEN, free_frozen);
5407STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5408STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5409STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5410STAT_ATTR(ALLOC_SLAB, alloc_slab);
5411STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005412STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005413STAT_ATTR(FREE_SLAB, free_slab);
5414STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5415STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5416STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5417STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5418STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5419STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005420STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005421STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005422STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5423STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005424STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5425STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005426STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5427STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005428#endif
5429
Pekka Enberg06428782008-01-07 23:20:27 -08005430static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005431 &slab_size_attr.attr,
5432 &object_size_attr.attr,
5433 &objs_per_slab_attr.attr,
5434 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005435 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005436 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005437 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005438 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005439 &partial_attr.attr,
5440 &cpu_slabs_attr.attr,
5441 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005442 &aliases_attr.attr,
5443 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005444 &hwcache_align_attr.attr,
5445 &reclaim_account_attr.attr,
5446 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005447 &shrink_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005448 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005449#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005450 &total_objects_attr.attr,
5451 &slabs_attr.attr,
5452 &sanity_checks_attr.attr,
5453 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005454 &red_zone_attr.attr,
5455 &poison_attr.attr,
5456 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005457 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005458 &alloc_calls_attr.attr,
5459 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005460#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005461#ifdef CONFIG_ZONE_DMA
5462 &cache_dma_attr.attr,
5463#endif
5464#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005465 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005466#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005467#ifdef CONFIG_SLUB_STATS
5468 &alloc_fastpath_attr.attr,
5469 &alloc_slowpath_attr.attr,
5470 &free_fastpath_attr.attr,
5471 &free_slowpath_attr.attr,
5472 &free_frozen_attr.attr,
5473 &free_add_partial_attr.attr,
5474 &free_remove_partial_attr.attr,
5475 &alloc_from_partial_attr.attr,
5476 &alloc_slab_attr.attr,
5477 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005478 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005479 &free_slab_attr.attr,
5480 &cpuslab_flush_attr.attr,
5481 &deactivate_full_attr.attr,
5482 &deactivate_empty_attr.attr,
5483 &deactivate_to_head_attr.attr,
5484 &deactivate_to_tail_attr.attr,
5485 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005486 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005487 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005488 &cmpxchg_double_fail_attr.attr,
5489 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005490 &cpu_partial_alloc_attr.attr,
5491 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005492 &cpu_partial_node_attr.attr,
5493 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005494#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005495#ifdef CONFIG_FAILSLAB
5496 &failslab_attr.attr,
5497#endif
David Windsor8eb82842017-06-10 22:50:28 -04005498 &usersize_attr.attr,
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005499
Christoph Lameter81819f02007-05-06 14:49:36 -07005500 NULL
5501};
5502
Arvind Yadav1fdaaa22017-09-06 16:21:56 -07005503static const struct attribute_group slab_attr_group = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005504 .attrs = slab_attrs,
5505};
5506
5507static ssize_t slab_attr_show(struct kobject *kobj,
5508 struct attribute *attr,
5509 char *buf)
5510{
5511 struct slab_attribute *attribute;
5512 struct kmem_cache *s;
5513 int err;
5514
5515 attribute = to_slab_attr(attr);
5516 s = to_slab(kobj);
5517
5518 if (!attribute->show)
5519 return -EIO;
5520
5521 err = attribute->show(s, buf);
5522
5523 return err;
5524}
5525
5526static ssize_t slab_attr_store(struct kobject *kobj,
5527 struct attribute *attr,
5528 const char *buf, size_t len)
5529{
5530 struct slab_attribute *attribute;
5531 struct kmem_cache *s;
5532 int err;
5533
5534 attribute = to_slab_attr(attr);
5535 s = to_slab(kobj);
5536
5537 if (!attribute->store)
5538 return -EIO;
5539
5540 err = attribute->store(s, buf, len);
Johannes Weiner127424c2016-01-20 15:02:32 -08005541#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005542 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08005543 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07005544
Glauber Costa107dab52012-12-18 14:23:05 -08005545 mutex_lock(&slab_mutex);
5546 if (s->max_attr_size < len)
5547 s->max_attr_size = len;
5548
Glauber Costaebe945c2012-12-18 14:23:10 -08005549 /*
5550 * This is a best effort propagation, so this function's return
5551 * value will be determined by the parent cache only. This is
5552 * basically because not all attributes will have a well
5553 * defined semantics for rollbacks - most of the actions will
5554 * have permanent effects.
5555 *
5556 * Returning the error value of any of the children that fail
5557 * is not 100 % defined, in the sense that users seeing the
5558 * error code won't be able to know anything about the state of
5559 * the cache.
5560 *
5561 * Only returning the error code for the parent cache at least
5562 * has well defined semantics. The cache being written to
5563 * directly either failed or succeeded, in which case we loop
5564 * through the descendants with best-effort propagation.
5565 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005566 for_each_memcg_cache(c, s)
5567 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005568 mutex_unlock(&slab_mutex);
5569 }
5570#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005571 return err;
5572}
5573
Glauber Costa107dab52012-12-18 14:23:05 -08005574static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5575{
Johannes Weiner127424c2016-01-20 15:02:32 -08005576#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005577 int i;
5578 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005579 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005580
Vladimir Davydov93030d82014-05-06 12:49:59 -07005581 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005582 return;
5583
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005584 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005585
Glauber Costa107dab52012-12-18 14:23:05 -08005586 /*
5587 * This mean this cache had no attribute written. Therefore, no point
5588 * in copying default values around
5589 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005590 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005591 return;
5592
5593 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5594 char mbuf[64];
5595 char *buf;
5596 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
Thomas Gleixner478fe302017-06-02 14:46:25 -07005597 ssize_t len;
Glauber Costa107dab52012-12-18 14:23:05 -08005598
5599 if (!attr || !attr->store || !attr->show)
5600 continue;
5601
5602 /*
5603 * It is really bad that we have to allocate here, so we will
5604 * do it only as a fallback. If we actually allocate, though,
5605 * we can just use the allocated buffer until the end.
5606 *
5607 * Most of the slub attributes will tend to be very small in
5608 * size, but sysfs allows buffers up to a page, so they can
5609 * theoretically happen.
5610 */
5611 if (buffer)
5612 buf = buffer;
Qian Cai3e632652020-06-01 21:45:57 -07005613 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf) &&
5614 !IS_ENABLED(CONFIG_SLUB_STATS))
Glauber Costa107dab52012-12-18 14:23:05 -08005615 buf = mbuf;
5616 else {
5617 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5618 if (WARN_ON(!buffer))
5619 continue;
5620 buf = buffer;
5621 }
5622
Thomas Gleixner478fe302017-06-02 14:46:25 -07005623 len = attr->show(root_cache, buf);
5624 if (len > 0)
5625 attr->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005626 }
5627
5628 if (buffer)
5629 free_page((unsigned long)buffer);
5630#endif
5631}
5632
Christoph Lameter41a21282014-05-06 12:50:08 -07005633static void kmem_cache_release(struct kobject *k)
5634{
5635 slab_kmem_cache_release(to_slab(k));
5636}
5637
Emese Revfy52cf25d2010-01-19 02:58:23 +01005638static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005639 .show = slab_attr_show,
5640 .store = slab_attr_store,
5641};
5642
5643static struct kobj_type slab_ktype = {
5644 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005645 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005646};
5647
5648static int uevent_filter(struct kset *kset, struct kobject *kobj)
5649{
5650 struct kobj_type *ktype = get_ktype(kobj);
5651
5652 if (ktype == &slab_ktype)
5653 return 1;
5654 return 0;
5655}
5656
Emese Revfy9cd43612009-12-31 14:52:51 +01005657static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005658 .filter = uevent_filter,
5659};
5660
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005661static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005662
Vladimir Davydov9a417072014-04-07 15:39:31 -07005663static inline struct kset *cache_kset(struct kmem_cache *s)
5664{
Johannes Weiner127424c2016-01-20 15:02:32 -08005665#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005666 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005667 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005668#endif
5669 return slab_kset;
5670}
5671
Christoph Lameter81819f02007-05-06 14:49:36 -07005672#define ID_STR_LENGTH 64
5673
5674/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005675 *
5676 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005677 */
5678static char *create_unique_id(struct kmem_cache *s)
5679{
5680 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5681 char *p = name;
5682
5683 BUG_ON(!name);
5684
5685 *p++ = ':';
5686 /*
5687 * First flags affecting slabcache operations. We will only
5688 * get here for aliasable slabs so we do not need to support
5689 * too many flags. The flags here must cover all flags that
5690 * are matched during merging to guarantee that the id is
5691 * unique.
5692 */
5693 if (s->flags & SLAB_CACHE_DMA)
5694 *p++ = 'd';
Nicolas Boichat62d342d2019-03-28 20:43:42 -07005695 if (s->flags & SLAB_CACHE_DMA32)
5696 *p++ = 'D';
Christoph Lameter81819f02007-05-06 14:49:36 -07005697 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5698 *p++ = 'a';
Laura Abbottbecfda62016-03-15 14:55:06 -07005699 if (s->flags & SLAB_CONSISTENCY_CHECKS)
Christoph Lameter81819f02007-05-06 14:49:36 -07005700 *p++ = 'F';
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08005701 if (s->flags & SLAB_ACCOUNT)
5702 *p++ = 'A';
Christoph Lameter81819f02007-05-06 14:49:36 -07005703 if (p != name + 1)
5704 *p++ = '-';
Alexey Dobriyan44065b22018-04-05 16:21:20 -07005705 p += sprintf(p, "%07u", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005706
Christoph Lameter81819f02007-05-06 14:49:36 -07005707 BUG_ON(p > name + ID_STR_LENGTH - 1);
5708 return name;
5709}
5710
Tejun Heo3b7b3142017-06-23 15:08:52 -07005711static void sysfs_slab_remove_workfn(struct work_struct *work)
5712{
5713 struct kmem_cache *s =
5714 container_of(work, struct kmem_cache, kobj_remove_work);
5715
5716 if (!s->kobj.state_in_sysfs)
5717 /*
5718 * For a memcg cache, this may be called during
5719 * deactivation and again on shutdown. Remove only once.
5720 * A cache is never shut down before deactivation is
5721 * complete, so no need to worry about synchronization.
5722 */
Vladimir Davydovf6ba4882017-08-18 15:16:08 -07005723 goto out;
Tejun Heo3b7b3142017-06-23 15:08:52 -07005724
5725#ifdef CONFIG_MEMCG
5726 kset_unregister(s->memcg_kset);
5727#endif
5728 kobject_uevent(&s->kobj, KOBJ_REMOVE);
Vladimir Davydovf6ba4882017-08-18 15:16:08 -07005729out:
Tejun Heo3b7b3142017-06-23 15:08:52 -07005730 kobject_put(&s->kobj);
5731}
5732
Christoph Lameter81819f02007-05-06 14:49:36 -07005733static int sysfs_slab_add(struct kmem_cache *s)
5734{
5735 int err;
5736 const char *name;
Tejun Heo1663f262017-02-22 15:41:39 -08005737 struct kset *kset = cache_kset(s);
Christoph Lameter45530c42012-11-28 16:23:07 +00005738 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005739
Tejun Heo3b7b3142017-06-23 15:08:52 -07005740 INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn);
5741
Tejun Heo1663f262017-02-22 15:41:39 -08005742 if (!kset) {
5743 kobject_init(&s->kobj, &slab_ktype);
5744 return 0;
5745 }
5746
Miles Chen11066382017-11-15 17:32:25 -08005747 if (!unmergeable && disable_higher_order_debug &&
5748 (slub_debug & DEBUG_METADATA_FLAGS))
5749 unmergeable = 1;
5750
Christoph Lameter81819f02007-05-06 14:49:36 -07005751 if (unmergeable) {
5752 /*
5753 * Slabcache can never be merged so we can use the name proper.
5754 * This is typically the case for debug situations. In that
5755 * case we can catch duplicate names easily.
5756 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005757 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005758 name = s->name;
5759 } else {
5760 /*
5761 * Create a unique name for the slab as a target
5762 * for the symlinks.
5763 */
5764 name = create_unique_id(s);
5765 }
5766
Tejun Heo1663f262017-02-22 15:41:39 -08005767 s->kobj.kset = kset;
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005768 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Wang Hai53bb2a62020-06-03 15:56:21 -07005769 if (err) {
5770 kobject_put(&s->kobj);
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005771 goto out;
Wang Hai53bb2a62020-06-03 15:56:21 -07005772 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005773
5774 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005775 if (err)
5776 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005777
Johannes Weiner127424c2016-01-20 15:02:32 -08005778#ifdef CONFIG_MEMCG
Tejun Heo1663f262017-02-22 15:41:39 -08005779 if (is_root_cache(s) && memcg_sysfs_enabled) {
Vladimir Davydov9a417072014-04-07 15:39:31 -07005780 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5781 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005782 err = -ENOMEM;
5783 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005784 }
5785 }
5786#endif
5787
Christoph Lameter81819f02007-05-06 14:49:36 -07005788 kobject_uevent(&s->kobj, KOBJ_ADD);
5789 if (!unmergeable) {
5790 /* Setup first alias */
5791 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005792 }
Dave Jones54b6a732014-04-07 15:39:32 -07005793out:
5794 if (!unmergeable)
5795 kfree(name);
5796 return err;
5797out_del_kobj:
5798 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005799 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005800}
5801
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005802static void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005803{
Christoph Lameter97d06602012-07-06 15:25:11 -05005804 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005805 /*
5806 * Sysfs has not been setup yet so no need to remove the
5807 * cache from sysfs.
5808 */
5809 return;
5810
Tejun Heo3b7b3142017-06-23 15:08:52 -07005811 kobject_get(&s->kobj);
5812 schedule_work(&s->kobj_remove_work);
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005813}
5814
Mikulas Patockad50d82f2018-06-27 23:26:09 -07005815void sysfs_slab_unlink(struct kmem_cache *s)
5816{
5817 if (slab_state >= FULL)
5818 kobject_del(&s->kobj);
5819}
5820
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005821void sysfs_slab_release(struct kmem_cache *s)
5822{
5823 if (slab_state >= FULL)
5824 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005825}
5826
5827/*
5828 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005829 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005830 */
5831struct saved_alias {
5832 struct kmem_cache *s;
5833 const char *name;
5834 struct saved_alias *next;
5835};
5836
Adrian Bunk5af328a2007-07-17 04:03:27 -07005837static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005838
5839static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5840{
5841 struct saved_alias *al;
5842
Christoph Lameter97d06602012-07-06 15:25:11 -05005843 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005844 /*
5845 * If we have a leftover link then remove it.
5846 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005847 sysfs_remove_link(&slab_kset->kobj, name);
5848 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005849 }
5850
5851 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5852 if (!al)
5853 return -ENOMEM;
5854
5855 al->s = s;
5856 al->name = name;
5857 al->next = alias_list;
5858 alias_list = al;
5859 return 0;
5860}
5861
5862static int __init slab_sysfs_init(void)
5863{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005864 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005865 int err;
5866
Christoph Lameter18004c52012-07-06 15:25:12 -05005867 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005868
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005869 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005870 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005871 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005872 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005873 return -ENOSYS;
5874 }
5875
Christoph Lameter97d06602012-07-06 15:25:11 -05005876 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005877
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005878 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005879 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005880 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005881 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5882 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005883 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005884
5885 while (alias_list) {
5886 struct saved_alias *al = alias_list;
5887
5888 alias_list = alias_list->next;
5889 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005890 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005891 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5892 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005893 kfree(al);
5894 }
5895
Christoph Lameter18004c52012-07-06 15:25:12 -05005896 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005897 resiliency_test();
5898 return 0;
5899}
5900
5901__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005902#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005903
5904/*
5905 * The /proc/slabinfo ABI
5906 */
Yang Shi5b365772017-11-15 17:32:03 -08005907#ifdef CONFIG_SLUB_DEBUG
Glauber Costa0d7561c2012-10-19 18:20:27 +04005908void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005909{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005910 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005911 unsigned long nr_objs = 0;
5912 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005913 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005914 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005915
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005916 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005917 nr_slabs += node_nr_slabs(n);
5918 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005919 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005920 }
5921
Glauber Costa0d7561c2012-10-19 18:20:27 +04005922 sinfo->active_objs = nr_objs - nr_free;
5923 sinfo->num_objs = nr_objs;
5924 sinfo->active_slabs = nr_slabs;
5925 sinfo->num_slabs = nr_slabs;
5926 sinfo->objects_per_slab = oo_objects(s->oo);
5927 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005928}
5929
Glauber Costa0d7561c2012-10-19 18:20:27 +04005930void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005931{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005932}
5933
Glauber Costab7454ad2012-10-19 18:20:25 +04005934ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5935 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005936{
Glauber Costab7454ad2012-10-19 18:20:25 +04005937 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005938}
Yang Shi5b365772017-11-15 17:32:03 -08005939#endif /* CONFIG_SLUB_DEBUG */