blob: 236465d3d7f58cf4e049ff231ded3d110ee197f4 [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>
Andrew Mortonf71ca812021-06-15 18:23:39 -070018#include <linux/swab.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070019#include <linux/bitops.h>
20#include <linux/slab.h>
Christoph Lameter97d06602012-07-06 15:25:11 -050021#include "slab.h"
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040022#include <linux/proc_fs.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070023#include <linux/seq_file.h>
Andrey Ryabinina79316c2015-02-13 14:39:38 -080024#include <linux/kasan.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070025#include <linux/cpu.h>
26#include <linux/cpuset.h>
27#include <linux/mempolicy.h>
28#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070029#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030#include <linux/kallsyms.h>
Alexander Potapenko2019e66b2021-01-22 09:18:39 +000031#include <linux/kfence.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070032#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070033#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090034#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030035#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060036#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080037#include <linux/memcontrol.h>
Kees Cook2482ddec2017-09-06 16:19:18 -070038#include <linux/random.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070039
Faiyaz Mohammed54e74122021-06-15 09:19:37 +053040#include <linux/debugfs.h>
Richard Kennedy4a923792010-10-21 10:29:19 +010041#include <trace/events/kmem.h>
Liujie Xie8bc63372021-05-11 19:05:32 +080042#include <trace/hooks/mm.h>
Richard Kennedy4a923792010-10-21 10:29:19 +010043
Mel Gorman072bb0a2012-07-31 16:43:58 -070044#include "internal.h"
45
Christoph Lameter81819f02007-05-06 14:49:36 -070046/*
47 * Lock order:
Christoph Lameter18004c52012-07-06 15:25:12 -050048 * 1. slab_mutex (Global Mutex)
Christoph Lameter881db7f2011-06-01 12:25:53 -050049 * 2. node->list_lock
50 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070051 *
Christoph Lameter18004c52012-07-06 15:25:12 -050052 * slab_mutex
Christoph Lameter881db7f2011-06-01 12:25:53 -050053 *
Christoph Lameter18004c52012-07-06 15:25:12 -050054 * The role of the slab_mutex is to protect the list of all the slabs
Christoph Lameter881db7f2011-06-01 12:25:53 -050055 * and to synchronize major metadata changes to slab cache structures.
56 *
57 * The slab_lock is only used for debugging and on arches that do not
Matthew Wilcoxb7ccc7f2018-06-07 17:08:46 -070058 * have the ability to do a cmpxchg_double. It only protects:
Christoph Lameter881db7f2011-06-01 12:25:53 -050059 * A. page->freelist -> List of object free in a page
Matthew Wilcoxb7ccc7f2018-06-07 17:08:46 -070060 * B. page->inuse -> Number of objects in use
61 * C. page->objects -> Number of objects in page
62 * D. page->frozen -> frozen state
Christoph Lameter881db7f2011-06-01 12:25:53 -050063 *
64 * If a slab is frozen then it is exempt from list management. It is not
Liu Xiang632b2ef2019-05-13 17:16:28 -070065 * on any list except per cpu partial list. The processor that froze the
66 * slab is the one who can perform list operations on the page. Other
67 * processors may put objects onto the freelist but the processor that
68 * froze the slab is the only one that can retrieve the objects from the
69 * page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070070 *
71 * The list_lock protects the partial and full list on each node and
72 * the partial slab counter. If taken then no new slabs may be added or
73 * removed from the lists nor make the number of partial slabs be modified.
74 * (Note that the total number of slabs is an atomic value that may be
75 * modified without taking the list lock).
76 *
77 * The list_lock is a centralized lock and thus we avoid taking it as
78 * much as possible. As long as SLUB does not have to handle partial
79 * slabs, operations can continue without any centralized lock. F.e.
80 * allocating a long series of objects that fill up slabs does not require
81 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070082 * Interrupts are disabled during allocation and deallocation in order to
83 * make the slab allocator safe to use in the context of an irq. In addition
84 * interrupts are disabled to ensure that the processor does not change
85 * while handling per_cpu slabs, due to kernel preemption.
86 *
87 * SLUB assigns one slab for allocation to each processor.
88 * Allocations only occur from these slabs called cpu slabs.
89 *
Christoph Lameter672bba32007-05-09 02:32:39 -070090 * Slabs with free elements are kept on a partial list and during regular
91 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070092 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070093 * We track full slabs for debugging purposes though because otherwise we
94 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070095 *
96 * Slabs are freed when they become empty. Teardown and setup is
97 * minimal so we rely on the page allocators per cpu caches for
98 * fast frees and allocs.
99 *
Yu Zhaoaed68142019-11-30 17:49:34 -0800100 * page->frozen The slab is frozen and exempt from list processing.
Christoph Lameter4b6f0752007-05-16 22:10:53 -0700101 * This means that the slab is dedicated to a purpose
102 * such as satisfying allocations for a specific
103 * processor. Objects may be freed in the slab while
104 * it is frozen but slab_free will then skip the usual
105 * list operations. It is up to the processor holding
106 * the slab to integrate the slab into the slab lists
107 * when the slab is no longer needed.
108 *
109 * One use of this flag is to mark slabs that are
110 * used for allocations. Then such a slab becomes a cpu
111 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700112 * freelist that allows lockless access to
Christoph Lameter894b87882007-05-10 03:15:16 -0700113 * free objects in addition to the regular freelist
114 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700115 *
Yu Zhaoaed68142019-11-30 17:49:34 -0800116 * SLAB_DEBUG_FLAGS Slab requires special handling due to debug
Christoph Lameter81819f02007-05-06 14:49:36 -0700117 * options set. This moves slab handling out of
Christoph Lameter894b87882007-05-10 03:15:16 -0700118 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700119 */
120
Vlastimil Babkaca0cab62020-08-06 23:18:51 -0700121#ifdef CONFIG_SLUB_DEBUG
122#ifdef CONFIG_SLUB_DEBUG_ON
123DEFINE_STATIC_KEY_TRUE(slub_debug_enabled);
124#else
125DEFINE_STATIC_KEY_FALSE(slub_debug_enabled);
126#endif
127#endif
128
Vlastimil Babka59052e82020-08-06 23:18:55 -0700129static inline bool kmem_cache_debug(struct kmem_cache *s)
130{
131 return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS);
Christoph Lameteraf537b02010-07-09 14:07:14 -0500132}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700133
Geert Uytterhoeven117d54d2016-08-04 15:31:55 -0700134void *fixup_red_left(struct kmem_cache *s, void *p)
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700135{
Vlastimil Babka59052e82020-08-06 23:18:55 -0700136 if (kmem_cache_debug_flags(s, SLAB_RED_ZONE))
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700137 p += s->red_left_pad;
138
139 return p;
140}
141
Joonsoo Kim345c9052013-06-19 14:05:52 +0900142static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
143{
144#ifdef CONFIG_SLUB_CPU_PARTIAL
145 return !kmem_cache_debug(s);
146#else
147 return false;
148#endif
149}
150
Christoph Lameter81819f02007-05-06 14:49:36 -0700151/*
152 * Issues still to be resolved:
153 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700154 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
155 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700156 * - Variable sizing of the per node arrays
157 */
158
159/* Enable to test recovery from slab corruption on boot */
160#undef SLUB_RESILIENCY_TEST
161
Christoph Lameterb789ef52011-06-01 12:25:49 -0500162/* Enable to log cmpxchg failures */
163#undef SLUB_DEBUG_CMPXCHG
164
Christoph Lameter81819f02007-05-06 14:49:36 -0700165/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700166 * Mininum number of partial slabs. These will be left on the partial
167 * lists even if they are empty. kmem_cache_shrink may reclaim them.
168 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800169#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700170
Christoph Lameter2086d262007-05-06 14:49:46 -0700171/*
172 * Maximum number of desirable partial slabs.
173 * The existence of more partial slabs makes kmem_cache_shrink
Zhi Yong Wu721ae222013-11-08 20:47:37 +0800174 * sort the partial list by the number of objects in use.
Christoph Lameter2086d262007-05-06 14:49:46 -0700175 */
176#define MAX_PARTIAL 10
177
Laura Abbottbecfda62016-03-15 14:55:06 -0700178#define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
Christoph Lameter81819f02007-05-06 14:49:36 -0700179 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700180
Christoph Lameter81819f02007-05-06 14:49:36 -0700181/*
Laura Abbott149daaf2016-03-15 14:55:09 -0700182 * These debug flags cannot use CMPXCHG because there might be consistency
183 * issues when checking or reading debug information
184 */
185#define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
186 SLAB_TRACE)
187
188
189/*
David Rientjes3de472132009-07-27 18:30:35 -0700190 * Debugging flags that require metadata to be stored in the slab. These get
191 * disabled when slub_debug=O is used and a cache's min order increases with
192 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700193 */
David Rientjes3de472132009-07-27 18:30:35 -0700194#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700195
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400196#define OO_SHIFT 16
197#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500198#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400199
Christoph Lameter81819f02007-05-06 14:49:36 -0700200/* Internal SLUB flags */
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800201/* Poison object */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800202#define __OBJECT_POISON ((slab_flags_t __force)0x80000000U)
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800203/* Use cmpxchg_double */
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800204#define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U)
Christoph Lameter81819f02007-05-06 14:49:36 -0700205
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500206#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700207static int sysfs_slab_add(struct kmem_cache *);
208static int sysfs_slab_alias(struct kmem_cache *, const char *);
Christoph Lameter81819f02007-05-06 14:49:36 -0700209#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700210static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
211static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
212 { return 0; }
Christoph Lameter81819f02007-05-06 14:49:36 -0700213#endif
214
Faiyaz Mohammed54e74122021-06-15 09:19:37 +0530215#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
216static void debugfs_slab_add(struct kmem_cache *);
217#else
218static inline void debugfs_slab_add(struct kmem_cache *s) { }
219#endif
220
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500221static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800222{
223#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700224 /*
225 * The rmw is racy on a preemptible kernel but this is acceptable, so
226 * avoid this_cpu_add()'s irq-disable overhead.
227 */
228 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800229#endif
230}
231
Christoph Lameter81819f02007-05-06 14:49:36 -0700232/********************************************************************
233 * Core slab cache functions
234 *******************************************************************/
235
Kees Cook2482ddec2017-09-06 16:19:18 -0700236/*
237 * Returns freelist pointer (ptr). With hardening, this is obfuscated
238 * with an XOR of the address where the pointer is held and a per-cache
239 * random number.
240 */
241static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr,
242 unsigned long ptr_addr)
243{
244#ifdef CONFIG_SLAB_FREELIST_HARDENED
Andrey Konovalovd36a63a2019-02-20 22:19:32 -0800245 /*
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800246 * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged.
Andrey Konovalovd36a63a2019-02-20 22:19:32 -0800247 * Normally, this doesn't cause any issues, as both set_freepointer()
248 * and get_freepointer() are called with a pointer with the same tag.
249 * However, there are some issues with CONFIG_SLUB_DEBUG code. For
250 * example, when __free_slub() iterates over objects in a cache, it
251 * passes untagged pointers to check_object(). check_object() in turns
252 * calls get_freepointer() with an untagged pointer, which causes the
253 * freepointer to be restored incorrectly.
254 */
255 return (void *)((unsigned long)ptr ^ s->random ^
Kees Cook1ad53d92020-04-01 21:04:23 -0700256 swab((unsigned long)kasan_reset_tag((void *)ptr_addr)));
Kees Cook2482ddec2017-09-06 16:19:18 -0700257#else
258 return ptr;
259#endif
260}
261
262/* Returns the freelist pointer recorded at location ptr_addr. */
263static inline void *freelist_dereference(const struct kmem_cache *s,
264 void *ptr_addr)
265{
266 return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr),
267 (unsigned long)ptr_addr);
268}
269
Christoph Lameter7656c722007-05-09 02:32:40 -0700270static inline void *get_freepointer(struct kmem_cache *s, void *object)
271{
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800272 object = kasan_reset_tag(object);
Kees Cook2482ddec2017-09-06 16:19:18 -0700273 return freelist_dereference(s, object + s->offset);
Christoph Lameter7656c722007-05-09 02:32:40 -0700274}
275
Eric Dumazet0ad95002011-12-16 16:25:34 +0100276static void prefetch_freepointer(const struct kmem_cache *s, void *object)
277{
Vlastimil Babka0882ff92018-08-17 15:44:44 -0700278 prefetch(object + s->offset);
Eric Dumazet0ad95002011-12-16 16:25:34 +0100279}
280
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500281static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
282{
Kees Cook2482ddec2017-09-06 16:19:18 -0700283 unsigned long freepointer_addr;
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500284 void *p;
285
Vlastimil Babka8e57f8a2020-01-13 16:29:20 -0800286 if (!debug_pagealloc_enabled_static())
Joonsoo Kim922d5662016-03-17 14:17:53 -0700287 return get_freepointer(s, object);
288
Alexander Potapenko5a18a6d2021-05-22 17:41:56 -0700289 object = kasan_reset_tag(object);
Kees Cook2482ddec2017-09-06 16:19:18 -0700290 freepointer_addr = (unsigned long)object + s->offset;
Christoph Hellwigfe557312020-06-17 09:37:53 +0200291 copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p));
Kees Cook2482ddec2017-09-06 16:19:18 -0700292 return freelist_ptr(s, p, freepointer_addr);
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500293}
294
Christoph Lameter7656c722007-05-09 02:32:40 -0700295static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
296{
Kees Cook2482ddec2017-09-06 16:19:18 -0700297 unsigned long freeptr_addr = (unsigned long)object + s->offset;
298
Alexander Popovce6fa912017-09-06 16:19:22 -0700299#ifdef CONFIG_SLAB_FREELIST_HARDENED
300 BUG_ON(object == fp); /* naive detection of double free or corruption */
301#endif
302
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800303 freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr);
Kees Cook2482ddec2017-09-06 16:19:18 -0700304 *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr);
Christoph Lameter7656c722007-05-09 02:32:40 -0700305}
306
307/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300308#define for_each_object(__p, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700309 for (__p = fixup_red_left(__s, __addr); \
310 __p < (__addr) + (__objects) * (__s)->size; \
311 __p += (__s)->size)
Christoph Lameter7656c722007-05-09 02:32:40 -0700312
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700313static inline unsigned int order_objects(unsigned int order, unsigned int size)
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800314{
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700315 return ((unsigned int)PAGE_SIZE << order) / size;
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800316}
317
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700318static inline struct kmem_cache_order_objects oo_make(unsigned int order,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700319 unsigned int size)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300320{
321 struct kmem_cache_order_objects x = {
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700322 (order << OO_SHIFT) + order_objects(order, size)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300323 };
324
325 return x;
326}
327
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700328static inline unsigned int oo_order(struct kmem_cache_order_objects x)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300329{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400330 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300331}
332
Alexey Dobriyan19af27a2018-04-05 16:21:39 -0700333static inline unsigned int oo_objects(struct kmem_cache_order_objects x)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300334{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400335 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300336}
337
Christoph Lameter881db7f2011-06-01 12:25:53 -0500338/*
339 * Per slab locking using the pagelock
340 */
341static __always_inline void slab_lock(struct page *page)
342{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800343 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500344 bit_spin_lock(PG_locked, &page->flags);
345}
346
347static __always_inline void slab_unlock(struct page *page)
348{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800349 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500350 __bit_spin_unlock(PG_locked, &page->flags);
351}
352
Christoph Lameter1d071712011-07-14 12:49:12 -0500353/* Interrupts must be disabled (for the fallback code to work right) */
354static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500355 void *freelist_old, unsigned long counters_old,
356 void *freelist_new, unsigned long counters_new,
357 const char *n)
358{
Christoph Lameter1d071712011-07-14 12:49:12 -0500359 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800360#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
361 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500362 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000363 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700364 freelist_old, counters_old,
365 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700366 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500367 } else
368#endif
369 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500370 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800371 if (page->freelist == freelist_old &&
372 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373 page->freelist = freelist_new;
Matthew Wilcox7d27a042018-06-07 17:08:31 -0700374 page->counters = counters_new;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500375 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700376 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500377 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500378 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500379 }
380
381 cpu_relax();
382 stat(s, CMPXCHG_DOUBLE_FAIL);
383
384#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700385 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500386#endif
387
Joe Perches6f6528a2015-04-14 15:44:31 -0700388 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500389}
390
Christoph Lameter1d071712011-07-14 12:49:12 -0500391static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
392 void *freelist_old, unsigned long counters_old,
393 void *freelist_new, unsigned long counters_new,
394 const char *n)
395{
Heiko Carstens25654092012-01-12 17:17:33 -0800396#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
397 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500398 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000399 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700400 freelist_old, counters_old,
401 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700402 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500403 } else
404#endif
405 {
406 unsigned long flags;
407
408 local_irq_save(flags);
409 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800410 if (page->freelist == freelist_old &&
411 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500412 page->freelist = freelist_new;
Matthew Wilcox7d27a042018-06-07 17:08:31 -0700413 page->counters = counters_new;
Christoph Lameter1d071712011-07-14 12:49:12 -0500414 slab_unlock(page);
415 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700416 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500417 }
418 slab_unlock(page);
419 local_irq_restore(flags);
420 }
421
422 cpu_relax();
423 stat(s, CMPXCHG_DOUBLE_FAIL);
424
425#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700426 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500427#endif
428
Joe Perches6f6528a2015-04-14 15:44:31 -0700429 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500430}
431
Christoph Lameter41ecc552007-05-09 02:32:44 -0700432#ifdef CONFIG_SLUB_DEBUG
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800433static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)];
434static DEFINE_SPINLOCK(object_map_lock);
435
Vlastimil Babka791f85d2021-05-23 01:28:37 +0200436static void __fill_map(unsigned long *obj_map, struct kmem_cache *s,
437 struct page *page)
438{
439 void *addr = page_address(page);
440 void *p;
441
442 bitmap_zero(obj_map, page->objects);
443
444 for (p = page->freelist; p; p = get_freepointer(s, p))
445 set_bit(__obj_to_index(s, addr, p), obj_map);
446}
447
Christoph Lameter41ecc552007-05-09 02:32:44 -0700448/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500449 * Determine a map of object in use on a page.
450 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500451 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500452 * not vanish from under us.
453 */
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800454static unsigned long *get_map(struct kmem_cache *s, struct page *page)
Jules Irenge31364c22020-04-06 20:08:15 -0700455 __acquires(&object_map_lock)
Christoph Lameter5f80b132011-04-15 14:48:13 -0500456{
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800457 VM_BUG_ON(!irqs_disabled());
458
459 spin_lock(&object_map_lock);
460
Vlastimil Babka791f85d2021-05-23 01:28:37 +0200461 __fill_map(object_map, s, page);
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800462
463 return object_map;
464}
465
Jules Irenge81aba9e2020-04-06 20:08:18 -0700466static void put_map(unsigned long *map) __releases(&object_map_lock)
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800467{
468 VM_BUG_ON(map != object_map);
Yu Zhao90e9f6a2020-01-30 22:11:57 -0800469 spin_unlock(&object_map_lock);
Christoph Lameter5f80b132011-04-15 14:48:13 -0500470}
471
Alexey Dobriyan870b1fb2018-04-05 16:21:43 -0700472static inline unsigned int size_from_object(struct kmem_cache *s)
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700473{
474 if (s->flags & SLAB_RED_ZONE)
475 return s->size - s->red_left_pad;
476
477 return s->size;
478}
479
480static inline void *restore_red_left(struct kmem_cache *s, void *p)
481{
482 if (s->flags & SLAB_RED_ZONE)
483 p -= s->red_left_pad;
484
485 return p;
486}
487
Christoph Lameter41ecc552007-05-09 02:32:44 -0700488/*
489 * Debug settings:
490 */
Andrey Ryabinin89d3c872015-11-05 18:51:23 -0800491#if defined(CONFIG_SLUB_DEBUG_ON)
Vijayanand Jitta14309282021-03-23 08:57:33 +0530492slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700493#else
Vijayanand Jitta14309282021-03-23 08:57:33 +0530494slab_flags_t slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700495#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700496
Vlastimil Babkae17f1df2020-08-06 23:18:35 -0700497static char *slub_debug_string;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700498static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700499
Christoph Lameter7656c722007-05-09 02:32:40 -0700500/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800501 * slub is about to manipulate internal object metadata. This memory lies
502 * outside the range of the allocated object, so accessing it would normally
503 * be reported by kasan as a bounds error. metadata_access_enable() is used
504 * to tell kasan that these accesses are OK.
505 */
506static inline void metadata_access_enable(void)
507{
508 kasan_disable_current();
509}
510
511static inline void metadata_access_disable(void)
512{
513 kasan_enable_current();
514}
515
516/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700517 * Object debugging
518 */
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700519
520/* Verify that a pointer has an address that is valid within a slab page */
521static inline int check_valid_pointer(struct kmem_cache *s,
522 struct page *page, void *object)
523{
524 void *base;
525
526 if (!object)
527 return 1;
528
529 base = page_address(page);
Qian Cai338cfaa2019-02-20 22:19:36 -0800530 object = kasan_reset_tag(object);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700531 object = restore_red_left(s, object);
532 if (object < base || object >= base + page->objects * s->size ||
533 (object - base) % s->size) {
534 return 0;
535 }
536
537 return 1;
538}
539
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800540static void print_section(char *level, char *text, u8 *addr,
541 unsigned int length)
Christoph Lameter81819f02007-05-06 14:49:36 -0700542{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800543 metadata_access_enable();
Kuan-Ying Leec0cce1d2021-08-13 16:54:27 -0700544 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS,
545 16, 1, kasan_reset_tag((void *)addr), length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800546 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700547}
548
Waiman Longcbfc35a2020-05-07 18:36:06 -0700549/*
550 * See comment in calculate_sizes().
551 */
552static inline bool freeptr_outside_object(struct kmem_cache *s)
553{
554 return s->offset >= s->inuse;
555}
556
557/*
558 * Return offset of the end of info block which is inuse + free pointer if
559 * not overlapping with object.
560 */
561static inline unsigned int get_info_end(struct kmem_cache *s)
562{
563 if (freeptr_outside_object(s))
564 return s->inuse + sizeof(void *);
565 else
566 return s->inuse;
567}
568
Christoph Lameter81819f02007-05-06 14:49:36 -0700569static struct track *get_track(struct kmem_cache *s, void *object,
570 enum track_item alloc)
571{
572 struct track *p;
573
Waiman Longcbfc35a2020-05-07 18:36:06 -0700574 p = object + get_info_end(s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700575
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800576 return kasan_reset_tag(p + alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700577}
578
Vijayanand Jittaee8d2c72021-01-12 00:42:08 +0530579/*
580 * This function will be used to loop through all the slab objects in
581 * a page to give track structure for each object, the function fn will
582 * be using this track structure and extract required info into its private
583 * data, the return value will be the number of track structures that are
584 * processed.
585 */
586unsigned long get_each_object_track(struct kmem_cache *s,
587 struct page *page, enum track_item alloc,
588 int (*fn)(const struct kmem_cache *, const void *,
589 const struct track *, void *), void *private)
590{
591 void *p;
592 struct track *t;
593 int ret;
594 unsigned long num_track = 0;
595
596 if (!slub_debug || !(s->flags & SLAB_STORE_USER))
597 return 0;
598
599 slab_lock(page);
600 for_each_object(p, s, page_address(page), page->objects) {
601 t = get_track(s, p, alloc);
602 ret = fn(s, p, t, private);
603 if (ret < 0)
604 break;
605 num_track += 1;
606 }
607 slab_unlock(page);
608 return num_track;
609}
610EXPORT_SYMBOL_GPL(get_each_object_track);
611
Christoph Lameter81819f02007-05-06 14:49:36 -0700612static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300613 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700614{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900615 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700616
Christoph Lameter81819f02007-05-06 14:49:36 -0700617 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700618#ifdef CONFIG_STACKTRACE
Thomas Gleixner79716792019-04-25 11:45:00 +0200619 unsigned int nr_entries;
Ben Greeard6543e32011-07-07 11:36:36 -0700620
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800621 metadata_access_enable();
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800622 nr_entries = stack_trace_save(kasan_reset_tag(p->addrs),
623 TRACK_ADDRS_COUNT, 3);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800624 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700625
Thomas Gleixner79716792019-04-25 11:45:00 +0200626 if (nr_entries < TRACK_ADDRS_COUNT)
627 p->addrs[nr_entries] = 0;
Liujie Xie7d216e22021-06-24 10:10:14 +0800628 trace_android_vh_save_track_hash(alloc == TRACK_ALLOC,
629 (unsigned long)p);
Ben Greeard6543e32011-07-07 11:36:36 -0700630#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700631 p->addr = addr;
632 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400633 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700634 p->when = jiffies;
Thomas Gleixnerb8ca7ff2019-04-10 12:28:05 +0200635 } else {
Christoph Lameter81819f02007-05-06 14:49:36 -0700636 memset(p, 0, sizeof(struct track));
Thomas Gleixnerb8ca7ff2019-04-10 12:28:05 +0200637 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700638}
639
Christoph Lameter81819f02007-05-06 14:49:36 -0700640static void init_tracking(struct kmem_cache *s, void *object)
641{
Christoph Lameter24922682007-07-17 04:03:18 -0700642 if (!(s->flags & SLAB_STORE_USER))
643 return;
644
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300645 set_track(s, object, TRACK_FREE, 0UL);
646 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700647}
648
Chintan Pandya86609d32018-04-05 16:20:15 -0700649static void print_track(const char *s, struct track *t, unsigned long pr_time)
Christoph Lameter81819f02007-05-06 14:49:36 -0700650{
651 if (!t->addr)
652 return;
653
Fabian Frederickf9f58282014-06-04 16:06:34 -0700654 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
Chintan Pandya86609d32018-04-05 16:20:15 -0700655 s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700656#ifdef CONFIG_STACKTRACE
657 {
658 int i;
659 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
660 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700661 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700662 else
663 break;
664 }
665#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700666}
667
Vlastimil Babkae42f1742020-08-06 23:19:05 -0700668void print_tracking(struct kmem_cache *s, void *object)
Christoph Lameter24922682007-07-17 04:03:18 -0700669{
Chintan Pandya86609d32018-04-05 16:20:15 -0700670 unsigned long pr_time = jiffies;
Christoph Lameter24922682007-07-17 04:03:18 -0700671 if (!(s->flags & SLAB_STORE_USER))
672 return;
673
Chintan Pandya86609d32018-04-05 16:20:15 -0700674 print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time);
675 print_track("Freed", get_track(s, object, TRACK_FREE), pr_time);
Christoph Lameter24922682007-07-17 04:03:18 -0700676}
677
678static void print_page_info(struct page *page)
679{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700680 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800681 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700682
683}
684
685static void slab_bug(struct kmem_cache *s, char *fmt, ...)
686{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700687 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700688 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700689
690 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700691 vaf.fmt = fmt;
692 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700693 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700694 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700695 pr_err("-----------------------------------------------------------------------------\n\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700696 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700697}
698
699static void slab_fix(struct kmem_cache *s, char *fmt, ...)
700{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700701 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700702 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700703
704 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700705 vaf.fmt = fmt;
706 vaf.va = &args;
707 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700708 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700709}
710
Dongli Zhang52f23472020-06-01 21:45:47 -0700711static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
Eugeniu Roscadc07a722020-09-04 16:35:30 -0700712 void **freelist, void *nextfree)
Dongli Zhang52f23472020-06-01 21:45:47 -0700713{
714 if ((s->flags & SLAB_CONSISTENCY_CHECKS) &&
Eugeniu Roscadc07a722020-09-04 16:35:30 -0700715 !check_valid_pointer(s, page, nextfree) && freelist) {
716 object_err(s, page, *freelist, "Freechain corrupt");
717 *freelist = NULL;
Dongli Zhang52f23472020-06-01 21:45:47 -0700718 slab_fix(s, "Isolate corrupted freechain");
719 return true;
720 }
721
722 return false;
723}
724
Christoph Lameter24922682007-07-17 04:03:18 -0700725static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700726{
727 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800728 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700729
730 print_tracking(s, p);
731
732 print_page_info(page);
733
Fabian Frederickf9f58282014-06-04 16:06:34 -0700734 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
735 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700736
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700737 if (s->flags & SLAB_RED_ZONE)
Kees Cook4a36fda2021-06-15 18:23:19 -0700738 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800739 s->red_left_pad);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700740 else if (p > addr + 16)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800741 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700742
Kees Cook4a36fda2021-06-15 18:23:19 -0700743 print_section(KERN_ERR, "Object ", p,
Alexey Dobriyan1b473f22018-04-05 16:21:17 -0700744 min_t(unsigned int, s->object_size, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700745 if (s->flags & SLAB_RED_ZONE)
Kees Cook4a36fda2021-06-15 18:23:19 -0700746 print_section(KERN_ERR, "Redzone ", p + s->object_size,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500747 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700748
Waiman Longcbfc35a2020-05-07 18:36:06 -0700749 off = get_info_end(s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700750
Christoph Lameter24922682007-07-17 04:03:18 -0700751 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700752 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700753
Alexander Potapenko80a92012016-07-28 15:49:07 -0700754 off += kasan_metadata_size(s);
755
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700756 if (off != size_from_object(s))
Christoph Lameter81819f02007-05-06 14:49:36 -0700757 /* Beginning of the filler is the free pointer */
Kees Cook4a36fda2021-06-15 18:23:19 -0700758 print_section(KERN_ERR, "Padding ", p + off,
Daniel Thompsonaa2efd52017-01-24 15:18:02 -0800759 size_from_object(s) - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700760
761 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700762}
763
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800764void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700765 u8 *object, char *reason)
766{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700767 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700768 print_trailer(s, page, object);
Georgi Djakova2e00b42021-06-17 02:01:49 -0400769 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700770}
771
Mathieu Malaterrea38965b2018-06-07 17:05:17 -0700772static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800773 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700774{
775 va_list args;
776 char buf[100];
777
Christoph Lameter24922682007-07-17 04:03:18 -0700778 va_start(args, fmt);
779 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700780 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700781 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700782 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700783 dump_stack();
Georgi Djakova2e00b42021-06-17 02:01:49 -0400784 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Christoph Lameter81819f02007-05-06 14:49:36 -0700785}
786
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500787static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700788{
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800789 u8 *p = kasan_reset_tag(object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700790
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700791 if (s->flags & SLAB_RED_ZONE)
792 memset(p - s->red_left_pad, val, s->red_left_pad);
793
Christoph Lameter81819f02007-05-06 14:49:36 -0700794 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500795 memset(p, POISON_FREE, s->object_size - 1);
796 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700797 }
798
799 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500800 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700801}
802
Christoph Lameter24922682007-07-17 04:03:18 -0700803static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
804 void *from, void *to)
805{
806 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
807 memset(from, data, to - from);
808}
809
810static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
811 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800812 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700813{
814 u8 *fault;
815 u8 *end;
Miles Chene1b70dd2019-11-30 17:49:31 -0800816 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700817
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800818 metadata_access_enable();
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800819 fault = memchr_inv(kasan_reset_tag(start), value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800820 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700821 if (!fault)
822 return 1;
823
824 end = start + bytes;
825 while (end > fault && end[-1] == value)
826 end--;
827
828 slab_bug(s, "%s overwritten", what);
Miles Chene1b70dd2019-11-30 17:49:31 -0800829 pr_err("INFO: 0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n",
830 fault, end - 1, fault - addr,
831 fault[0], value);
Christoph Lameter24922682007-07-17 04:03:18 -0700832 print_trailer(s, page, object);
Georgi Djakova2e00b42021-06-17 02:01:49 -0400833 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Christoph Lameter24922682007-07-17 04:03:18 -0700834
835 restore_bytes(s, what, value, fault, end);
836 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700837}
838
Christoph Lameter81819f02007-05-06 14:49:36 -0700839/*
840 * Object layout:
841 *
842 * object address
843 * Bytes of the object to be managed.
844 * If the freepointer may overlay the object then the free
Waiman Longcbfc35a2020-05-07 18:36:06 -0700845 * pointer is at the middle of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700846 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700847 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
848 * 0xa5 (POISON_END)
849 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500850 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700851 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700852 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500853 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700854 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700855 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
856 * 0xcc (RED_ACTIVE) for objects in use.
857 *
858 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700859 * Meta data starts here.
860 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700861 * A. Free pointer (if we cannot overwrite object on free)
862 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700863 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800864 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700865 * before the word boundary.
866 *
867 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700868 *
869 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700870 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700871 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500872 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700873 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700874 * may be used with merged slabcaches.
875 */
876
Christoph Lameter81819f02007-05-06 14:49:36 -0700877static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
878{
Waiman Longcbfc35a2020-05-07 18:36:06 -0700879 unsigned long off = get_info_end(s); /* The end of info */
Christoph Lameter81819f02007-05-06 14:49:36 -0700880
881 if (s->flags & SLAB_STORE_USER)
882 /* We also have user information there */
883 off += 2 * sizeof(struct track);
884
Alexander Potapenko80a92012016-07-28 15:49:07 -0700885 off += kasan_metadata_size(s);
886
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700887 if (size_from_object(s) == off)
Christoph Lameter81819f02007-05-06 14:49:36 -0700888 return 1;
889
Christoph Lameter24922682007-07-17 04:03:18 -0700890 return check_bytes_and_report(s, page, p, "Object padding",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700891 p + off, POISON_INUSE, size_from_object(s) - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700892}
893
Christoph Lameter39b26462008-04-14 19:11:30 +0300894/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700895static int slab_pad_check(struct kmem_cache *s, struct page *page)
896{
Christoph Lameter24922682007-07-17 04:03:18 -0700897 u8 *start;
898 u8 *fault;
899 u8 *end;
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800900 u8 *pad;
Christoph Lameter24922682007-07-17 04:03:18 -0700901 int length;
902 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700903
904 if (!(s->flags & SLAB_POISON))
905 return 1;
906
Christoph Lametera973e9d2008-03-01 13:40:44 -0800907 start = page_address(page);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -0700908 length = page_size(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300909 end = start + length;
910 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700911 if (!remainder)
912 return 1;
913
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800914 pad = end - remainder;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800915 metadata_access_enable();
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -0800916 fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800917 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700918 if (!fault)
919 return 1;
920 while (end > fault && end[-1] == POISON_INUSE)
921 end--;
922
Miles Chene1b70dd2019-11-30 17:49:31 -0800923 slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu",
924 fault, end - 1, fault - start);
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800925 print_section(KERN_ERR, "Padding ", pad, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700926
Balasubramani Vivekanandan5d682682018-01-31 16:15:43 -0800927 restore_bytes(s, "slab padding", POISON_INUSE, fault, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700928 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700929}
930
931static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500932 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700933{
934 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500935 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700936
937 if (s->flags & SLAB_RED_ZONE) {
Kees Cook4a36fda2021-06-15 18:23:19 -0700938 if (!check_bytes_and_report(s, page, object, "Left Redzone",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700939 object - s->red_left_pad, val, s->red_left_pad))
940 return 0;
941
Kees Cook4a36fda2021-06-15 18:23:19 -0700942 if (!check_bytes_and_report(s, page, object, "Right Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500943 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700944 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700945 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500946 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800947 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800948 endobject, POISON_INUSE,
949 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800950 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700951 }
952
953 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500954 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700955 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500956 POISON_FREE, s->object_size - 1) ||
Kees Cook4a36fda2021-06-15 18:23:19 -0700957 !check_bytes_and_report(s, page, p, "End Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500958 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700959 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700960 /*
961 * check_pad_bytes cleans up on its own.
962 */
963 check_pad_bytes(s, page, p);
964 }
965
Waiman Longcbfc35a2020-05-07 18:36:06 -0700966 if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700967 /*
968 * Object and freepointer overlap. Cannot check
969 * freepointer while object is allocated.
970 */
971 return 1;
972
973 /* Check free pointer validity */
974 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
975 object_err(s, page, p, "Freepointer corrupt");
976 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100977 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700978 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700979 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700980 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800981 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700982 return 0;
983 }
984 return 1;
985}
986
987static int check_slab(struct kmem_cache *s, struct page *page)
988{
Christoph Lameter39b26462008-04-14 19:11:30 +0300989 int maxobj;
990
Christoph Lameter81819f02007-05-06 14:49:36 -0700991 VM_BUG_ON(!irqs_disabled());
992
993 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700994 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700995 return 0;
996 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300997
Matthew Wilcox9736d2a2018-06-07 17:09:10 -0700998 maxobj = order_objects(compound_order(page), s->size);
Christoph Lameter39b26462008-04-14 19:11:30 +0300999 if (page->objects > maxobj) {
1000 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -08001001 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +03001002 return 0;
1003 }
1004 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -07001005 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -08001006 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001007 return 0;
1008 }
1009 /* Slab_pad_check fixes things up after itself */
1010 slab_pad_check(s, page);
1011 return 1;
1012}
1013
1014/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001015 * Determine if a certain object on a page is on the freelist. Must hold the
1016 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -07001017 */
1018static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
1019{
1020 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -05001021 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001022 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -08001023 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001024
Christoph Lameter881db7f2011-06-01 12:25:53 -05001025 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +03001026 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001027 if (fp == search)
1028 return 1;
1029 if (!check_valid_pointer(s, page, fp)) {
1030 if (object) {
1031 object_err(s, page, object,
1032 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -08001033 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001034 } else {
Christoph Lameter24922682007-07-17 04:03:18 -07001035 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -08001036 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +03001037 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -07001038 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -07001039 return 0;
1040 }
1041 break;
1042 }
1043 object = fp;
1044 fp = get_freepointer(s, object);
1045 nr++;
1046 }
1047
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07001048 max_objects = order_objects(compound_order(page), s->size);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04001049 if (max_objects > MAX_OBJS_PER_PAGE)
1050 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001051
1052 if (page->objects != max_objects) {
Joe Perches756a0252016-03-17 14:19:47 -07001053 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
1054 page->objects, max_objects);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001055 page->objects = max_objects;
1056 slab_fix(s, "Number of objects adjusted.");
1057 }
Christoph Lameter39b26462008-04-14 19:11:30 +03001058 if (page->inuse != page->objects - nr) {
Joe Perches756a0252016-03-17 14:19:47 -07001059 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
1060 page->inuse, page->objects - nr);
Christoph Lameter39b26462008-04-14 19:11:30 +03001061 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -07001062 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001063 }
1064 return search == NULL;
1065}
1066
Christoph Lameter0121c6192008-04-29 16:11:12 -07001067static void trace(struct kmem_cache *s, struct page *page, void *object,
1068 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -07001069{
1070 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07001071 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -07001072 s->name,
1073 alloc ? "alloc" : "free",
1074 object, page->inuse,
1075 page->freelist);
1076
1077 if (!alloc)
Daniel Thompsonaa2efd52017-01-24 15:18:02 -08001078 print_section(KERN_INFO, "Object ", (void *)object,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001079 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001080
1081 dump_stack();
1082 }
1083}
1084
Christoph Lameter643b1132007-05-06 14:49:42 -07001085/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001086 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -07001087 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001088static void add_full(struct kmem_cache *s,
1089 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -07001090{
Christoph Lameter643b1132007-05-06 14:49:42 -07001091 if (!(s->flags & SLAB_STORE_USER))
1092 return;
1093
David Rientjes255d0882014-02-10 14:25:39 -08001094 lockdep_assert_held(&n->list_lock);
Tobin C. Harding916ac052019-05-13 17:16:12 -07001095 list_add(&page->slab_list, &n->full);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001096}
Christoph Lameter643b1132007-05-06 14:49:42 -07001097
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001098static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001099{
1100 if (!(s->flags & SLAB_STORE_USER))
1101 return;
1102
David Rientjes255d0882014-02-10 14:25:39 -08001103 lockdep_assert_held(&n->list_lock);
Tobin C. Harding916ac052019-05-13 17:16:12 -07001104 list_del(&page->slab_list);
Christoph Lameter643b1132007-05-06 14:49:42 -07001105}
1106
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001107/* Tracking of the number of slabs for debugging purposes */
1108static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1109{
1110 struct kmem_cache_node *n = get_node(s, node);
1111
1112 return atomic_long_read(&n->nr_slabs);
1113}
1114
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001115static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1116{
1117 return atomic_long_read(&n->nr_slabs);
1118}
1119
Christoph Lameter205ab992008-04-14 19:11:40 +03001120static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001121{
1122 struct kmem_cache_node *n = get_node(s, node);
1123
1124 /*
1125 * May be called early in order to allocate a slab for the
1126 * kmem_cache_node structure. Solve the chicken-egg
1127 * dilemma by deferring the increment of the count during
1128 * bootstrap (see early_kmem_cache_node_alloc).
1129 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001130 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001131 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001132 atomic_long_add(objects, &n->total_objects);
1133 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001134}
Christoph Lameter205ab992008-04-14 19:11:40 +03001135static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001136{
1137 struct kmem_cache_node *n = get_node(s, node);
1138
1139 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001140 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001141}
1142
1143/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001144static void setup_object_debug(struct kmem_cache *s, struct page *page,
1145 void *object)
1146{
Vlastimil Babka8fc8d662020-08-06 23:18:58 -07001147 if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))
Christoph Lameter3ec09742007-05-16 22:11:00 -07001148 return;
1149
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001150 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001151 init_tracking(s, object);
1152}
1153
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07001154static
1155void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr)
Andrey Konovalova7101222019-02-20 22:19:23 -08001156{
Vlastimil Babka8fc8d662020-08-06 23:18:58 -07001157 if (!kmem_cache_debug_flags(s, SLAB_POISON))
Andrey Konovalova7101222019-02-20 22:19:23 -08001158 return;
1159
1160 metadata_access_enable();
Andrey Konovalov8d4b6fc2020-12-22 12:02:17 -08001161 memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page));
Andrey Konovalova7101222019-02-20 22:19:23 -08001162 metadata_access_disable();
1163}
1164
Laura Abbottbecfda62016-03-15 14:55:06 -07001165static inline int alloc_consistency_checks(struct kmem_cache *s,
Qian Cai278d7752019-03-05 15:42:10 -08001166 struct page *page, void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07001167{
1168 if (!check_slab(s, page))
Laura Abbottbecfda62016-03-15 14:55:06 -07001169 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001170
Christoph Lameter81819f02007-05-06 14:49:36 -07001171 if (!check_valid_pointer(s, page, object)) {
1172 object_err(s, page, object, "Freelist Pointer check fails");
Laura Abbottbecfda62016-03-15 14:55:06 -07001173 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001174 }
1175
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001176 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Laura Abbottbecfda62016-03-15 14:55:06 -07001177 return 0;
1178
1179 return 1;
1180}
1181
1182static noinline int alloc_debug_processing(struct kmem_cache *s,
1183 struct page *page,
1184 void *object, unsigned long addr)
1185{
1186 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
Qian Cai278d7752019-03-05 15:42:10 -08001187 if (!alloc_consistency_checks(s, page, object))
Laura Abbottbecfda62016-03-15 14:55:06 -07001188 goto bad;
1189 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001190
Christoph Lameter3ec09742007-05-16 22:11:00 -07001191 /* Success perform special debug activities for allocs */
1192 if (s->flags & SLAB_STORE_USER)
1193 set_track(s, object, TRACK_ALLOC, addr);
1194 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001195 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001196 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001197
Christoph Lameter81819f02007-05-06 14:49:36 -07001198bad:
1199 if (PageSlab(page)) {
1200 /*
1201 * If this is a slab page then lets do the best we can
1202 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001203 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001204 */
Christoph Lameter24922682007-07-17 04:03:18 -07001205 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001206 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001207 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001208 }
1209 return 0;
1210}
1211
Laura Abbottbecfda62016-03-15 14:55:06 -07001212static inline int free_consistency_checks(struct kmem_cache *s,
1213 struct page *page, void *object, unsigned long addr)
1214{
1215 if (!check_valid_pointer(s, page, object)) {
1216 slab_err(s, page, "Invalid object pointer 0x%p", object);
1217 return 0;
1218 }
1219
1220 if (on_freelist(s, page, object)) {
1221 object_err(s, page, object, "Object already free");
1222 return 0;
1223 }
1224
1225 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1226 return 0;
1227
1228 if (unlikely(s != page->slab_cache)) {
1229 if (!PageSlab(page)) {
Joe Perches756a0252016-03-17 14:19:47 -07001230 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1231 object);
Laura Abbottbecfda62016-03-15 14:55:06 -07001232 } else if (!page->slab_cache) {
1233 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1234 object);
1235 dump_stack();
1236 } else
1237 object_err(s, page, object,
1238 "page slab pointer corrupt.");
1239 return 0;
1240 }
1241 return 1;
1242}
1243
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001244/* Supports checking bulk free of a constructed freelist */
Laura Abbott282acb42016-03-15 14:54:59 -07001245static noinline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001246 struct kmem_cache *s, struct page *page,
1247 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001248 unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001249{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001250 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001251 void *object = head;
1252 int cnt = 0;
Kees Cook3f649ab2020-06-03 13:09:38 -07001253 unsigned long flags;
Laura Abbott804aa132016-03-15 14:55:02 -07001254 int ret = 0;
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001255
Laura Abbott282acb42016-03-15 14:54:59 -07001256 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001257 slab_lock(page);
1258
Laura Abbottbecfda62016-03-15 14:55:06 -07001259 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1260 if (!check_slab(s, page))
1261 goto out;
1262 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001263
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001264next_object:
1265 cnt++;
1266
Laura Abbottbecfda62016-03-15 14:55:06 -07001267 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1268 if (!free_consistency_checks(s, page, object, addr))
1269 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001270 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001271
Christoph Lameter3ec09742007-05-16 22:11:00 -07001272 if (s->flags & SLAB_STORE_USER)
1273 set_track(s, object, TRACK_FREE, addr);
1274 trace(s, page, object, 0);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001275 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001276 init_object(s, object, SLUB_RED_INACTIVE);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001277
1278 /* Reached end of constructed freelist yet? */
1279 if (object != tail) {
1280 object = get_freepointer(s, object);
1281 goto next_object;
1282 }
Laura Abbott804aa132016-03-15 14:55:02 -07001283 ret = 1;
1284
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001285out:
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001286 if (cnt != bulk_cnt)
1287 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1288 bulk_cnt, cnt);
1289
Christoph Lameter881db7f2011-06-01 12:25:53 -05001290 slab_unlock(page);
Laura Abbott282acb42016-03-15 14:54:59 -07001291 spin_unlock_irqrestore(&n->list_lock, flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001292 if (!ret)
1293 slab_fix(s, "Object at 0x%p not freed", object);
1294 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001295}
1296
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001297/*
1298 * Parse a block of slub_debug options. Blocks are delimited by ';'
1299 *
1300 * @str: start of block
1301 * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified
1302 * @slabs: return start of list of slabs, or NULL when there's no list
1303 * @init: assume this is initial parsing and not per-kmem-create parsing
1304 *
1305 * returns the start of next block if there's any, or NULL
1306 */
1307static char *
1308parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init)
1309{
1310 bool higher_order_disable = false;
1311
1312 /* Skip any completely empty blocks */
1313 while (*str && *str == ';')
1314 str++;
1315
1316 if (*str == ',') {
1317 /*
1318 * No options but restriction on slabs. This means full
1319 * debugging for slabs matching a pattern.
1320 */
1321 *flags = DEBUG_DEFAULT_FLAGS;
1322 goto check_slabs;
1323 }
1324 *flags = 0;
1325
1326 /* Determine which debug features should be switched on */
1327 for (; *str && *str != ',' && *str != ';'; str++) {
1328 switch (tolower(*str)) {
1329 case '-':
1330 *flags = 0;
1331 break;
1332 case 'f':
1333 *flags |= SLAB_CONSISTENCY_CHECKS;
1334 break;
1335 case 'z':
1336 *flags |= SLAB_RED_ZONE;
1337 break;
1338 case 'p':
1339 *flags |= SLAB_POISON;
1340 break;
1341 case 'u':
1342 *flags |= SLAB_STORE_USER;
1343 break;
1344 case 't':
1345 *flags |= SLAB_TRACE;
1346 break;
1347 case 'a':
1348 *flags |= SLAB_FAILSLAB;
1349 break;
1350 case 'o':
1351 /*
1352 * Avoid enabling debugging on caches if its minimum
1353 * order would increase as a result.
1354 */
1355 higher_order_disable = true;
1356 break;
1357 default:
1358 if (init)
1359 pr_err("slub_debug option '%c' unknown. skipped\n", *str);
1360 }
1361 }
1362check_slabs:
1363 if (*str == ',')
1364 *slabs = ++str;
1365 else
1366 *slabs = NULL;
1367
1368 /* Skip over the slab list */
1369 while (*str && *str != ';')
1370 str++;
1371
1372 /* Skip any completely empty blocks */
1373 while (*str && *str == ';')
1374 str++;
1375
1376 if (init && higher_order_disable)
1377 disable_higher_order_debug = 1;
1378
1379 if (*str)
1380 return str;
1381 else
1382 return NULL;
1383}
1384
Christoph Lameter41ecc552007-05-09 02:32:44 -07001385static int __init setup_slub_debug(char *str)
1386{
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001387 slab_flags_t flags;
Vlastimil Babka3425d612021-08-06 17:14:36 +01001388 slab_flags_t global_flags;
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001389 char *saved_str;
1390 char *slab_list;
1391 bool global_slub_debug_changed = false;
1392 bool slab_list_specified = false;
1393
Vlastimil Babka3425d612021-08-06 17:14:36 +01001394 global_flags = DEBUG_DEFAULT_FLAGS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001395 if (*str++ != '=' || !*str)
1396 /*
1397 * No options specified. Switch on full debugging.
1398 */
1399 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001400
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001401 saved_str = str;
1402 while (str) {
1403 str = parse_slub_debug_flags(str, &flags, &slab_list, true);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001404
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001405 if (!slab_list) {
Vlastimil Babka3425d612021-08-06 17:14:36 +01001406 global_flags = flags;
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001407 global_slub_debug_changed = true;
1408 } else {
1409 slab_list_specified = true;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001410 }
1411 }
1412
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001413 /*
1414 * For backwards compatibility, a single list of flags with list of
Vlastimil Babka3425d612021-08-06 17:14:36 +01001415 * slabs means debugging is only changed for those slabs, so the global
1416 * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending
1417 * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001418 * long as there is no option specifying flags without a slab list.
1419 */
1420 if (slab_list_specified) {
1421 if (!global_slub_debug_changed)
Vlastimil Babka3425d612021-08-06 17:14:36 +01001422 global_flags = slub_debug;
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001423 slub_debug_string = saved_str;
1424 }
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001425out:
Vlastimil Babka3425d612021-08-06 17:14:36 +01001426 slub_debug = global_flags;
Vlastimil Babkaca0cab62020-08-06 23:18:51 -07001427 if (slub_debug != 0 || slub_debug_string)
1428 static_branch_enable(&slub_debug_enabled);
Alexander Potapenko64713842019-07-11 20:59:19 -07001429 if ((static_branch_unlikely(&init_on_alloc) ||
1430 static_branch_unlikely(&init_on_free)) &&
1431 (slub_debug & SLAB_POISON))
1432 pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n");
Christoph Lameter41ecc552007-05-09 02:32:44 -07001433 return 1;
1434}
1435
1436__setup("slub_debug", setup_slub_debug);
1437
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001438/*
1439 * kmem_cache_flags - apply debugging options to the cache
1440 * @object_size: the size of an object without meta data
1441 * @flags: flags to set
1442 * @name: name of the cache
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001443 *
1444 * Debug option(s) are applied to @flags. In addition to the debug
1445 * option(s), if a slab name (or multiple) is specified i.e.
1446 * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ...
1447 * then only the select slabs will receive the debug option(s).
1448 */
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001449slab_flags_t kmem_cache_flags(unsigned int object_size,
Nikolay Borisov2e95bc62021-02-24 12:00:58 -08001450 slab_flags_t flags, const char *name)
Christoph Lameter41ecc552007-05-09 02:32:44 -07001451{
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001452 char *iter;
1453 size_t len;
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001454 char *next_block;
1455 slab_flags_t block_flags;
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001456
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001457 len = strlen(name);
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001458 next_block = slub_debug_string;
1459 /* Go through all blocks of debug options, see if any matches our slab's name */
1460 while (next_block) {
1461 next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false);
1462 if (!iter)
1463 continue;
1464 /* Found a block that has a slab list, search it */
1465 while (*iter) {
1466 char *end, *glob;
1467 size_t cmplen;
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001468
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001469 end = strchrnul(iter, ',');
1470 if (next_block && next_block < end)
1471 end = next_block - 1;
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001472
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001473 glob = strnchr(iter, end - iter, '*');
1474 if (glob)
1475 cmplen = glob - iter;
1476 else
1477 cmplen = max_t(size_t, len, (end - iter));
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001478
Vlastimil Babkae17f1df2020-08-06 23:18:35 -07001479 if (!strncmp(name, iter, cmplen)) {
1480 flags |= block_flags;
1481 return flags;
1482 }
1483
1484 if (!*end || *end == ';')
1485 break;
1486 iter = end + 1;
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001487 }
Aaron Tomlinc5fd3ca2018-10-26 15:03:15 -07001488 }
Christoph Lameterba0268a2007-09-11 15:24:11 -07001489
Eric Farman484cfac2020-10-02 22:21:41 -07001490 return flags | slub_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001491}
Jesper Dangaard Brouerb4a64712015-11-20 15:57:41 -08001492#else /* !CONFIG_SLUB_DEBUG */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001493static inline void setup_object_debug(struct kmem_cache *s,
1494 struct page *page, void *object) {}
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07001495static inline
1496void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001497
Christoph Lameter3ec09742007-05-16 22:11:00 -07001498static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001499 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001500
Laura Abbott282acb42016-03-15 14:54:59 -07001501static inline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001502 struct kmem_cache *s, struct page *page,
1503 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001504 unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001505
Christoph Lameter41ecc552007-05-09 02:32:44 -07001506static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1507 { return 1; }
1508static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001509 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001510static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1511 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001512static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1513 struct page *page) {}
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001514slab_flags_t kmem_cache_flags(unsigned int object_size,
Nikolay Borisov2e95bc62021-02-24 12:00:58 -08001515 slab_flags_t flags, const char *name)
Christoph Lameterba0268a2007-09-11 15:24:11 -07001516{
1517 return flags;
1518}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001519#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001520
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001521#define disable_higher_order_debug 0
1522
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001523static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1524 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001525static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1526 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001527static inline void inc_slabs_node(struct kmem_cache *s, int node,
1528 int objects) {}
1529static inline void dec_slabs_node(struct kmem_cache *s, int node,
1530 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001531
Dongli Zhang52f23472020-06-01 21:45:47 -07001532static bool freelist_corrupted(struct kmem_cache *s, struct page *page,
Eugeniu Roscadc07a722020-09-04 16:35:30 -07001533 void **freelist, void *nextfree)
Dongli Zhang52f23472020-06-01 21:45:47 -07001534{
1535 return false;
1536}
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001537#endif /* CONFIG_SLUB_DEBUG */
1538
1539/*
1540 * Hooks for other subsystems that check memory allocations. In a typical
1541 * production configuration these hooks all should produce no code at all.
1542 */
Andrey Konovalov01165232018-12-28 00:29:37 -08001543static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001544{
Andrey Konovalov53128242019-02-20 22:19:11 -08001545 ptr = kasan_kmalloc_large(ptr, size, flags);
Andrey Konovalova2f77572019-02-20 22:19:16 -08001546 /* As ptr might get tagged, call kmemleak hook after KASAN. */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001547 kmemleak_alloc(ptr, size, 1, flags);
Andrey Konovalov53128242019-02-20 22:19:11 -08001548 return ptr;
Roman Bobnievd56791b2013-10-08 15:58:57 -07001549}
1550
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08001551static __always_inline void kfree_hook(void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001552{
1553 kmemleak_free(x);
Andrey Konovalov3cd65f52021-02-03 15:35:05 +11001554 kasan_kfree_large(x);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001555}
1556
Andrey Konovalov24690d72021-03-18 17:01:41 +11001557static __always_inline bool slab_free_hook(struct kmem_cache *s,
1558 void *x, bool init)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001559{
1560 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001561
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001562 /*
1563 * Trouble is that we may no longer disable interrupts in the fast path
1564 * So in order to make the debug calls that expect irqs to be
1565 * disabled we need to disable interrupts temporarily.
1566 */
Levin, Alexander (Sasha Levin)4675ff02017-11-15 17:36:02 -08001567#ifdef CONFIG_LOCKDEP
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001568 {
1569 unsigned long flags;
1570
1571 local_irq_save(flags);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001572 debug_check_no_locks_freed(x, s->object_size);
1573 local_irq_restore(flags);
1574 }
1575#endif
1576 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1577 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001578
Marco Elvercfbe1632020-08-06 23:19:12 -07001579 /* Use KCSAN to help debug racy use-after-free. */
1580 if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
1581 __kcsan_check_access(x, s->object_size,
1582 KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
1583
Andrey Konovalov24690d72021-03-18 17:01:41 +11001584 /*
1585 * As memory initialization might be integrated into KASAN,
1586 * kasan_slab_free and initialization memset's must be
1587 * kept together to avoid discrepancies in behavior.
1588 *
1589 * The initialization memset's clear the object and the metadata,
1590 * but don't touch the SLAB redzone.
1591 */
1592 if (init) {
1593 int rsize;
1594
1595 if (!kasan_has_integrated_init())
1596 memset(kasan_reset_tag(x), 0, s->object_size);
1597 rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
1598 memset((char *)kasan_reset_tag(x) + s->inuse, 0,
1599 s->size - s->inuse - rsize);
1600 }
1601 /* KASAN might put x into memory quarantine, delaying its reuse. */
1602 return kasan_slab_free(s, x, init);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001603}
Christoph Lameter205ab992008-04-14 19:11:40 +03001604
Andrey Konovalovc3895392018-04-10 16:30:31 -07001605static inline bool slab_free_freelist_hook(struct kmem_cache *s,
1606 void **head, void **tail)
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001607{
Alexander Potapenko64713842019-07-11 20:59:19 -07001608
1609 void *object;
1610 void *next = *head;
1611 void *old_tail = *tail ? *tail : *head;
Alexander Potapenko64713842019-07-11 20:59:19 -07001612
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00001613 if (is_kfence_address(next)) {
Andrey Konovalov24690d72021-03-18 17:01:41 +11001614 slab_free_hook(s, next, false);
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00001615 return true;
1616 }
1617
Laura Abbottaea4df42019-11-15 17:34:50 -08001618 /* Head and tail of the reconstructed freelist */
1619 *head = NULL;
1620 *tail = NULL;
Laura Abbott1b7e8162019-07-31 15:32:40 -04001621
Laura Abbottaea4df42019-11-15 17:34:50 -08001622 do {
1623 object = next;
1624 next = get_freepointer(s, object);
1625
Andrey Konovalovc3895392018-04-10 16:30:31 -07001626 /* If object's reuse doesn't have to be delayed */
Andrey Konovalov24690d72021-03-18 17:01:41 +11001627 if (!slab_free_hook(s, object, slab_want_init_on_free(s))) {
Andrey Konovalovc3895392018-04-10 16:30:31 -07001628 /* Move object to the new freelist */
1629 set_freepointer(s, object, *head);
1630 *head = object;
1631 if (!*tail)
1632 *tail = object;
1633 }
1634 } while (object != old_tail);
1635
1636 if (*head == *tail)
1637 *tail = NULL;
1638
1639 return *head != NULL;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001640}
1641
Andrey Konovalov4d176712018-12-28 00:30:23 -08001642static void *setup_object(struct kmem_cache *s, struct page *page,
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001643 void *object)
1644{
1645 setup_object_debug(s, page, object);
Andrey Konovalov4d176712018-12-28 00:30:23 -08001646 object = kasan_init_slab_obj(s, object);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001647 if (unlikely(s->ctor)) {
1648 kasan_unpoison_object_data(s, object);
1649 s->ctor(object);
1650 kasan_poison_object_data(s, object);
1651 }
Andrey Konovalov4d176712018-12-28 00:30:23 -08001652 return object;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001653}
1654
Christoph Lameter81819f02007-05-06 14:49:36 -07001655/*
1656 * Slab allocation and freeing
1657 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001658static inline struct page *alloc_slab_page(struct kmem_cache *s,
1659 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001660{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001661 struct page *page;
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001662 unsigned int order = oo_order(oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001663
Christoph Lameter2154a332010-07-09 14:07:10 -05001664 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001665 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001666 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001667 page = __alloc_pages_node(node, flags, order);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001668
Roman Gushchin10befea2020-08-06 23:21:27 -07001669 if (page)
Roman Gushchin74d555b2020-08-06 23:21:44 -07001670 account_slab_page(page, order, s);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001671
1672 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001673}
1674
Thomas Garnier210e7a42016-07-26 15:21:59 -07001675#ifdef CONFIG_SLAB_FREELIST_RANDOM
1676/* Pre-initialize the random sequence cache */
1677static int init_cache_random_seq(struct kmem_cache *s)
1678{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001679 unsigned int count = oo_objects(s->oo);
Thomas Garnier210e7a42016-07-26 15:21:59 -07001680 int err;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001681
Sean Reesa8100072017-02-08 14:30:59 -08001682 /* Bailout if already initialised */
1683 if (s->random_seq)
1684 return 0;
1685
Thomas Garnier210e7a42016-07-26 15:21:59 -07001686 err = cache_random_seq_create(s, count, GFP_KERNEL);
1687 if (err) {
1688 pr_err("SLUB: Unable to initialize free list for %s\n",
1689 s->name);
1690 return err;
1691 }
1692
1693 /* Transform to an offset on the set of pages */
1694 if (s->random_seq) {
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07001695 unsigned int i;
1696
Thomas Garnier210e7a42016-07-26 15:21:59 -07001697 for (i = 0; i < count; i++)
1698 s->random_seq[i] *= s->size;
1699 }
1700 return 0;
1701}
1702
1703/* Initialize each random sequence freelist per cache */
1704static void __init init_freelist_randomization(void)
1705{
1706 struct kmem_cache *s;
1707
1708 mutex_lock(&slab_mutex);
1709
1710 list_for_each_entry(s, &slab_caches, list)
1711 init_cache_random_seq(s);
1712
1713 mutex_unlock(&slab_mutex);
1714}
1715
1716/* Get the next entry on the pre-computed freelist randomized */
1717static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1718 unsigned long *pos, void *start,
1719 unsigned long page_limit,
1720 unsigned long freelist_count)
1721{
1722 unsigned int idx;
1723
1724 /*
1725 * If the target page allocation failed, the number of objects on the
1726 * page might be smaller than the usual size defined by the cache.
1727 */
1728 do {
1729 idx = s->random_seq[*pos];
1730 *pos += 1;
1731 if (*pos >= freelist_count)
1732 *pos = 0;
1733 } while (unlikely(idx >= page_limit));
1734
1735 return (char *)start + idx;
1736}
1737
1738/* Shuffle the single linked freelist based on a random pre-computed sequence */
1739static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1740{
1741 void *start;
1742 void *cur;
1743 void *next;
1744 unsigned long idx, pos, page_limit, freelist_count;
1745
1746 if (page->objects < 2 || !s->random_seq)
1747 return false;
1748
1749 freelist_count = oo_objects(s->oo);
1750 pos = get_random_int() % freelist_count;
1751
1752 page_limit = page->objects * s->size;
1753 start = fixup_red_left(s, page_address(page));
1754
1755 /* First entry is used as the base of the freelist */
1756 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1757 freelist_count);
Andrey Konovalov4d176712018-12-28 00:30:23 -08001758 cur = setup_object(s, page, cur);
Thomas Garnier210e7a42016-07-26 15:21:59 -07001759 page->freelist = cur;
1760
1761 for (idx = 1; idx < page->objects; idx++) {
Thomas Garnier210e7a42016-07-26 15:21:59 -07001762 next = next_freelist_entry(s, page, &pos, start, page_limit,
1763 freelist_count);
Andrey Konovalov4d176712018-12-28 00:30:23 -08001764 next = setup_object(s, page, next);
Thomas Garnier210e7a42016-07-26 15:21:59 -07001765 set_freepointer(s, cur, next);
1766 cur = next;
1767 }
Thomas Garnier210e7a42016-07-26 15:21:59 -07001768 set_freepointer(s, cur, NULL);
1769
1770 return true;
1771}
1772#else
1773static inline int init_cache_random_seq(struct kmem_cache *s)
1774{
1775 return 0;
1776}
1777static inline void init_freelist_randomization(void) { }
1778static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1779{
1780 return false;
1781}
1782#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1783
Christoph Lameter81819f02007-05-06 14:49:36 -07001784static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1785{
Pekka Enberg06428782008-01-07 23:20:27 -08001786 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001787 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001788 gfp_t alloc_gfp;
Andrey Konovalov4d176712018-12-28 00:30:23 -08001789 void *start, *p, *next;
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07001790 int idx;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001791 bool shuffle;
Christoph Lameter81819f02007-05-06 14:49:36 -07001792
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001793 flags &= gfp_allowed_mask;
1794
Mel Gormand0164ad2015-11-06 16:28:21 -08001795 if (gfpflags_allow_blocking(flags))
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001796 local_irq_enable();
1797
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001798 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001799
Pekka Enbergba522702009-06-24 21:59:51 +03001800 /*
1801 * Let the initial higher-order allocation fail under memory pressure
1802 * so we fall-back to the minimum order allocation.
1803 */
1804 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Mel Gormand0164ad2015-11-06 16:28:21 -08001805 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
Mel Gorman444eb2a42016-03-17 14:19:23 -07001806 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
Pekka Enbergba522702009-06-24 21:59:51 +03001807
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001808 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001809 if (unlikely(!page)) {
1810 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001811 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001812 /*
1813 * Allocation may have failed due to fragmentation.
1814 * Try a lower order alloc if possible
1815 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001816 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001817 if (unlikely(!page))
1818 goto out;
1819 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001820 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001821
Christoph Lameter834f3d12008-04-14 19:11:31 +03001822 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001823
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001824 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001825 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001826 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001827 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001828
Andrey Konovalova7101222019-02-20 22:19:23 -08001829 kasan_poison_slab(page);
1830
Christoph Lameter81819f02007-05-06 14:49:36 -07001831 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001832
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07001833 setup_page_debug(s, page, start);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001834
Thomas Garnier210e7a42016-07-26 15:21:59 -07001835 shuffle = shuffle_freelist(s, page);
1836
1837 if (!shuffle) {
Andrey Konovalov4d176712018-12-28 00:30:23 -08001838 start = fixup_red_left(s, start);
1839 start = setup_object(s, page, start);
1840 page->freelist = start;
Andrey Konovalov18e50662019-02-20 22:19:28 -08001841 for (idx = 0, p = start; idx < page->objects - 1; idx++) {
1842 next = p + s->size;
1843 next = setup_object(s, page, next);
1844 set_freepointer(s, p, next);
1845 p = next;
1846 }
1847 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001848 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001849
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001850 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001851 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001852
Christoph Lameter81819f02007-05-06 14:49:36 -07001853out:
Mel Gormand0164ad2015-11-06 16:28:21 -08001854 if (gfpflags_allow_blocking(flags))
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001855 local_irq_disable();
1856 if (!page)
1857 return NULL;
1858
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001859 inc_slabs_node(s, page_to_nid(page), page->objects);
1860
Christoph Lameter81819f02007-05-06 14:49:36 -07001861 return page;
1862}
1863
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001864static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1865{
Long Li44405092020-08-06 23:18:28 -07001866 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1867 flags = kmalloc_fix_flags(flags);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001868
1869 return allocate_slab(s,
1870 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1871}
1872
Christoph Lameter81819f02007-05-06 14:49:36 -07001873static void __free_slab(struct kmem_cache *s, struct page *page)
1874{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001875 int order = compound_order(page);
1876 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001877
Vlastimil Babka8fc8d662020-08-06 23:18:58 -07001878 if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001879 void *p;
1880
1881 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001882 for_each_object(p, s, page_address(page),
1883 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001884 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001885 }
1886
Mel Gorman072bb0a2012-07-31 16:43:58 -07001887 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001888 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001889
Matthew Wilcoxd4fc5062018-06-07 17:08:26 -07001890 page->mapping = NULL;
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001891 if (current->reclaim_state)
1892 current->reclaim_state->reclaimed_slab += pages;
Roman Gushchin74d555b2020-08-06 23:21:44 -07001893 unaccount_slab_page(page, order, s);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001894 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001895}
1896
1897static void rcu_free_slab(struct rcu_head *h)
1898{
Matthew Wilcoxbf68c212018-06-07 17:09:05 -07001899 struct page *page = container_of(h, struct page, rcu_head);
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001900
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001901 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001902}
1903
1904static void free_slab(struct kmem_cache *s, struct page *page)
1905{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001906 if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) {
Matthew Wilcoxbf68c212018-06-07 17:09:05 -07001907 call_rcu(&page->rcu_head, rcu_free_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07001908 } else
1909 __free_slab(s, page);
1910}
1911
1912static void discard_slab(struct kmem_cache *s, struct page *page)
1913{
Christoph Lameter205ab992008-04-14 19:11:40 +03001914 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001915 free_slab(s, page);
1916}
1917
1918/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001919 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001920 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001921static inline void
1922__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001923{
Christoph Lametere95eed52007-05-06 14:49:44 -07001924 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001925 if (tail == DEACTIVATE_TO_TAIL)
Tobin C. Harding916ac052019-05-13 17:16:12 -07001926 list_add_tail(&page->slab_list, &n->partial);
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001927 else
Tobin C. Harding916ac052019-05-13 17:16:12 -07001928 list_add(&page->slab_list, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001929}
1930
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001931static inline void add_partial(struct kmem_cache_node *n,
1932 struct page *page, int tail)
1933{
1934 lockdep_assert_held(&n->list_lock);
1935 __add_partial(n, page, tail);
1936}
1937
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001938static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001939 struct page *page)
1940{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001941 lockdep_assert_held(&n->list_lock);
Tobin C. Harding916ac052019-05-13 17:16:12 -07001942 list_del(&page->slab_list);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08001943 n->nr_partial--;
Christoph Lameter62e346a2010-09-28 08:10:28 -05001944}
1945
Christoph Lameter81819f02007-05-06 14:49:36 -07001946/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001947 * Remove slab from the partial list, freeze it and
1948 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001949 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001950 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001951 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001952static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001953 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001954 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001955{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001956 void *freelist;
1957 unsigned long counters;
1958 struct page new;
1959
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001960 lockdep_assert_held(&n->list_lock);
1961
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001962 /*
1963 * Zap the freelist and set the frozen bit.
1964 * The old freelist is the list of objects for the
1965 * per cpu allocation list.
1966 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001967 freelist = page->freelist;
1968 counters = page->counters;
1969 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001970 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001971 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001972 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001973 new.freelist = NULL;
1974 } else {
1975 new.freelist = freelist;
1976 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001977
Dave Hansena0132ac2014-01-29 14:05:50 -08001978 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001979 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001980
Christoph Lameter7ced3712012-05-09 10:09:53 -05001981 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001982 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001983 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001984 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001985 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001986
1987 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001988 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001989 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001990}
1991
Joonsoo Kim633b0762013-01-21 17:01:25 +09001992static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001993static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001994
Christoph Lameter81819f02007-05-06 14:49:36 -07001995/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001996 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001997 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001998static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1999 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07002000{
Christoph Lameter49e22582011-08-09 16:12:27 -05002001 struct page *page, *page2;
2002 void *object = NULL;
Alexey Dobriyane5d99982018-04-05 16:21:10 -07002003 unsigned int available = 0;
Joonsoo Kim633b0762013-01-21 17:01:25 +09002004 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002005
2006 /*
2007 * Racy check. If we mistakenly see no partial slabs then we
2008 * just allocate an empty slab. If we mistakenly try to get a
Chen Tao70b6d252020-10-15 20:10:01 -07002009 * partial slab and there is none available then get_partial()
Christoph Lameter672bba32007-05-09 02:32:39 -07002010 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07002011 */
2012 if (!n || !n->nr_partial)
2013 return NULL;
2014
2015 spin_lock(&n->list_lock);
Tobin C. Harding916ac052019-05-13 17:16:12 -07002016 list_for_each_entry_safe(page, page2, &n->partial, slab_list) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07002017 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05002018
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07002019 if (!pfmemalloc_match(page, flags))
2020 continue;
2021
Joonsoo Kim633b0762013-01-21 17:01:25 +09002022 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05002023 if (!t)
Linus Torvaldse1759162021-03-10 10:18:04 -08002024 break;
Christoph Lameter49e22582011-08-09 16:12:27 -05002025
Joonsoo Kim633b0762013-01-21 17:01:25 +09002026 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08002027 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002028 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05002029 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002030 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05002031 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09002032 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08002033 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05002034 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002035 if (!kmem_cache_has_cpu_partial(s)
Wei Yange6d0e1d2017-07-06 15:36:34 -07002036 || available > slub_cpu_partial(s) / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05002037 break;
2038
Christoph Lameter497b66f2011-08-09 16:12:26 -05002039 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002040 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002041 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002042}
2043
2044/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002045 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07002046 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08002047static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05002048 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002049{
2050#ifdef CONFIG_NUMA
2051 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07002052 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07002053 struct zone *zone;
Joonsoo Kim97a225e2020-06-03 15:59:01 -07002054 enum zone_type highest_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002055 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07002056 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07002057
2058 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07002059 * The defrag ratio allows a configuration of the tradeoffs between
2060 * inter node defragmentation and node local allocations. A lower
2061 * defrag_ratio increases the tendency to do local allocations
2062 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07002063 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002064 * If the defrag_ratio is set to 0 then kmalloc() always
2065 * returns node local objects. If the ratio is higher then kmalloc()
2066 * may return off node objects because partial slabs are obtained
2067 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07002068 *
Li Peng43efd3e2016-05-19 17:10:43 -07002069 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
2070 * (which makes defrag_ratio = 1000) then every (well almost)
2071 * allocation will first attempt to defrag slab caches on other nodes.
2072 * This means scanning over all nodes to look for partial slabs which
2073 * may be expensive if we do it every time we are trying to find a slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002074 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07002075 */
Christoph Lameter98246012008-01-07 23:20:26 -08002076 if (!s->remote_node_defrag_ratio ||
2077 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07002078 return NULL;
2079
Mel Gormancc9a6c82012-03-21 16:34:11 -07002080 do {
Mel Gormand26914d2014-04-03 14:47:24 -07002081 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07002082 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Joonsoo Kim97a225e2020-06-03 15:59:01 -07002083 for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
Mel Gormancc9a6c82012-03-21 16:34:11 -07002084 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07002085
Mel Gormancc9a6c82012-03-21 16:34:11 -07002086 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07002087
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08002088 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07002089 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07002090 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07002091 if (object) {
2092 /*
Mel Gormand26914d2014-04-03 14:47:24 -07002093 * Don't check read_mems_allowed_retry()
2094 * here - if mems_allowed was updated in
2095 * parallel, that was a harmless race
2096 * between allocation and the cpuset
2097 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07002098 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07002099 return object;
2100 }
Miao Xiec0ff7452010-05-24 14:32:08 -07002101 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002102 }
Mel Gormand26914d2014-04-03 14:47:24 -07002103 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07002104#endif /* CONFIG_NUMA */
Christoph Lameter81819f02007-05-06 14:49:36 -07002105 return NULL;
2106}
2107
2108/*
2109 * Get a partial page, lock it and return it.
2110 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05002111static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05002112 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002113{
Christoph Lameter497b66f2011-08-09 16:12:26 -05002114 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07002115 int searchnode = node;
2116
2117 if (node == NUMA_NO_NODE)
2118 searchnode = numa_mem_id();
Christoph Lameter81819f02007-05-06 14:49:36 -07002119
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07002120 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002121 if (object || node != NUMA_NO_NODE)
2122 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002123
Christoph Lameteracd19fd2011-08-09 16:12:25 -05002124 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07002125}
2126
Thomas Gleixner923717c2019-10-15 21:18:12 +02002127#ifdef CONFIG_PREEMPTION
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002128/*
Ethon Paul0d645ed2020-06-04 16:49:34 -07002129 * Calculate the next globally unique transaction for disambiguation
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002130 * during cmpxchg. The transactions start with the cpu number and are then
2131 * incremented by CONFIG_NR_CPUS.
2132 */
2133#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
2134#else
2135/*
2136 * No preemption supported therefore also no need to check for
2137 * different cpus.
2138 */
2139#define TID_STEP 1
2140#endif
2141
2142static inline unsigned long next_tid(unsigned long tid)
2143{
2144 return tid + TID_STEP;
2145}
2146
Qian Cai9d5f0be2019-09-23 15:33:52 -07002147#ifdef SLUB_DEBUG_CMPXCHG
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002148static inline unsigned int tid_to_cpu(unsigned long tid)
2149{
2150 return tid % TID_STEP;
2151}
2152
2153static inline unsigned long tid_to_event(unsigned long tid)
2154{
2155 return tid / TID_STEP;
2156}
Qian Cai9d5f0be2019-09-23 15:33:52 -07002157#endif
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002158
2159static inline unsigned int init_tid(int cpu)
2160{
2161 return cpu;
2162}
2163
2164static inline void note_cmpxchg_failure(const char *n,
2165 const struct kmem_cache *s, unsigned long tid)
2166{
2167#ifdef SLUB_DEBUG_CMPXCHG
2168 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
2169
Fabian Frederickf9f58282014-06-04 16:06:34 -07002170 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002171
Thomas Gleixner923717c2019-10-15 21:18:12 +02002172#ifdef CONFIG_PREEMPTION
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002173 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002174 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002175 tid_to_cpu(tid), tid_to_cpu(actual_tid));
2176 else
2177#endif
2178 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002179 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002180 tid_to_event(tid), tid_to_event(actual_tid));
2181 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07002182 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002183 actual_tid, tid, next_tid(tid));
2184#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05002185 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002186}
2187
Fengguang Wu788e1aa2012-09-28 16:34:05 +08002188static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002189{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002190 int cpu;
2191
2192 for_each_possible_cpu(cpu)
2193 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002194}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002195
2196/*
2197 * Remove the cpu slab
2198 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002199static void deactivate_slab(struct kmem_cache *s, struct page *page,
Wei Yangd4ff6d32017-07-06 15:36:25 -07002200 void *freelist, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002201{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002202 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002203 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
2204 int lock = 0;
2205 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002206 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08002207 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002208 struct page new;
2209 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002210
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002211 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002212 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08002213 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b87882007-05-10 03:15:16 -07002214 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002215
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002216 /*
2217 * Stage one: Free all available per cpu objects back
2218 * to the page freelist while it is still frozen. Leave the
2219 * last one.
2220 *
2221 * There is no need to take the list->lock because the page
2222 * is still frozen.
2223 */
2224 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2225 void *prior;
2226 unsigned long counters;
2227
Dongli Zhang52f23472020-06-01 21:45:47 -07002228 /*
2229 * If 'nextfree' is invalid, it is possible that the object at
2230 * 'freelist' is already corrupted. So isolate all objects
2231 * starting at 'freelist'.
2232 */
Eugeniu Roscadc07a722020-09-04 16:35:30 -07002233 if (freelist_corrupted(s, page, &freelist, nextfree))
Dongli Zhang52f23472020-06-01 21:45:47 -07002234 break;
2235
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002236 do {
2237 prior = page->freelist;
2238 counters = page->counters;
2239 set_freepointer(s, freelist, prior);
2240 new.counters = counters;
2241 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08002242 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002243
Christoph Lameter1d071712011-07-14 12:49:12 -05002244 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002245 prior, counters,
2246 freelist, new.counters,
2247 "drain percpu freelist"));
2248
2249 freelist = nextfree;
2250 }
2251
2252 /*
2253 * Stage two: Ensure that the page is unfrozen while the
2254 * list presence reflects the actual number of objects
2255 * during unfreeze.
2256 *
2257 * We setup the list membership and then perform a cmpxchg
2258 * with the count. If there is a mismatch then the page
2259 * is not unfrozen but the page is on the wrong list.
2260 *
2261 * Then we restart the process which may have to remove
2262 * the page from the list that we just put it on again
2263 * because the number of objects in the slab may have
2264 * changed.
2265 */
2266redo:
2267
2268 old.freelist = page->freelist;
2269 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002270 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002271
2272 /* Determine target state of the slab */
2273 new.counters = old.counters;
2274 if (freelist) {
2275 new.inuse--;
2276 set_freepointer(s, freelist, old.freelist);
2277 new.freelist = freelist;
2278 } else
2279 new.freelist = old.freelist;
2280
2281 new.frozen = 0;
2282
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002283 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002284 m = M_FREE;
2285 else if (new.freelist) {
2286 m = M_PARTIAL;
2287 if (!lock) {
2288 lock = 1;
2289 /*
Wei Yang8bb4e7a2019-03-05 15:46:22 -08002290 * Taking the spinlock removes the possibility
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002291 * that acquire_slab() will see a slab page that
2292 * is frozen
2293 */
2294 spin_lock(&n->list_lock);
2295 }
2296 } else {
2297 m = M_FULL;
Abel Wu9cf7a112020-10-13 16:48:47 -07002298#ifdef CONFIG_SLUB_DEBUG
2299 if ((s->flags & SLAB_STORE_USER) && !lock) {
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002300 lock = 1;
2301 /*
2302 * This also ensures that the scanning of full
2303 * slabs from diagnostic functions will not see
2304 * any frozen slabs.
2305 */
2306 spin_lock(&n->list_lock);
2307 }
Abel Wu9cf7a112020-10-13 16:48:47 -07002308#endif
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002309 }
2310
2311 if (l != m) {
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002312 if (l == M_PARTIAL)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002313 remove_partial(n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002314 else if (l == M_FULL)
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002315 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002316
Wei Yang88349a22018-12-28 00:33:13 -08002317 if (m == M_PARTIAL)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002318 add_partial(n, page, tail);
Wei Yang88349a22018-12-28 00:33:13 -08002319 else if (m == M_FULL)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002320 add_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002321 }
2322
2323 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05002324 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002325 old.freelist, old.counters,
2326 new.freelist, new.counters,
2327 "unfreezing slab"))
2328 goto redo;
2329
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002330 if (lock)
2331 spin_unlock(&n->list_lock);
2332
Wei Yang88349a22018-12-28 00:33:13 -08002333 if (m == M_PARTIAL)
2334 stat(s, tail);
2335 else if (m == M_FULL)
2336 stat(s, DEACTIVATE_FULL);
2337 else if (m == M_FREE) {
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002338 stat(s, DEACTIVATE_EMPTY);
2339 discard_slab(s, page);
2340 stat(s, FREE_SLAB);
2341 }
Wei Yangd4ff6d32017-07-06 15:36:25 -07002342
2343 c->page = NULL;
2344 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002345}
2346
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002347/*
2348 * Unfreeze all the cpu partial slabs.
2349 *
Christoph Lameter59a09912012-11-28 16:23:00 +00002350 * This function must be called with interrupts disabled
2351 * for the cpu using c (or some other guarantee must be there
2352 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002353 */
Christoph Lameter59a09912012-11-28 16:23:00 +00002354static void unfreeze_partials(struct kmem_cache *s,
2355 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05002356{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002357#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09002358 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08002359 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002360
chenqiwu4c7ba222020-04-01 21:04:16 -07002361 while ((page = slub_percpu_partial(c))) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002362 struct page new;
2363 struct page old;
2364
chenqiwu4c7ba222020-04-01 21:04:16 -07002365 slub_set_percpu_partial(c, page);
Joonsoo Kim43d77862012-06-09 02:23:16 +09002366
2367 n2 = get_node(s, page_to_nid(page));
2368 if (n != n2) {
2369 if (n)
2370 spin_unlock(&n->list_lock);
2371
2372 n = n2;
2373 spin_lock(&n->list_lock);
2374 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002375
2376 do {
2377
2378 old.freelist = page->freelist;
2379 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002380 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05002381
2382 new.counters = old.counters;
2383 new.freelist = old.freelist;
2384
2385 new.frozen = 0;
2386
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002387 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002388 old.freelist, old.counters,
2389 new.freelist, new.counters,
2390 "unfreezing slab"));
2391
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002392 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002393 page->next = discard_page;
2394 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002395 } else {
2396 add_partial(n, page, DEACTIVATE_TO_TAIL);
2397 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002398 }
2399 }
2400
2401 if (n)
2402 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002403
2404 while (discard_page) {
2405 page = discard_page;
2406 discard_page = discard_page->next;
2407
2408 stat(s, DEACTIVATE_EMPTY);
2409 discard_slab(s, page);
2410 stat(s, FREE_SLAB);
2411 }
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07002412#endif /* CONFIG_SLUB_CPU_PARTIAL */
Christoph Lameter49e22582011-08-09 16:12:27 -05002413}
2414
2415/*
Wei Yang9234bae2019-03-05 15:43:10 -08002416 * Put a page that was just frozen (in __slab_free|get_partial_node) into a
2417 * partial page slot if available.
Christoph Lameter49e22582011-08-09 16:12:27 -05002418 *
2419 * If we did not find a slot then simply move all the partials to the
2420 * per node partial list.
2421 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002422static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002423{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002424#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002425 struct page *oldpage;
2426 int pages;
2427 int pobjects;
2428
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002429 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002430 do {
2431 pages = 0;
2432 pobjects = 0;
2433 oldpage = this_cpu_read(s->cpu_slab->partial);
2434
2435 if (oldpage) {
2436 pobjects = oldpage->pobjects;
2437 pages = oldpage->pages;
chenqiwubbd4e302020-04-01 21:04:19 -07002438 if (drain && pobjects > slub_cpu_partial(s)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002439 unsigned long flags;
2440 /*
2441 * partial array is full. Move the existing
2442 * set to the per node partial list.
2443 */
2444 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002445 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002446 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002447 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002448 pobjects = 0;
2449 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002450 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002451 }
2452 }
2453
2454 pages++;
2455 pobjects += page->objects - page->inuse;
2456
2457 page->pages = pages;
2458 page->pobjects = pobjects;
2459 page->next = oldpage;
2460
Chen Gangd0e0ac92013-07-15 09:05:29 +08002461 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2462 != oldpage);
chenqiwubbd4e302020-04-01 21:04:19 -07002463 if (unlikely(!slub_cpu_partial(s))) {
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002464 unsigned long flags;
2465
2466 local_irq_save(flags);
2467 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2468 local_irq_restore(flags);
2469 }
2470 preempt_enable();
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07002471#endif /* CONFIG_SLUB_CPU_PARTIAL */
Christoph Lameter49e22582011-08-09 16:12:27 -05002472}
2473
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002474static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002475{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002476 stat(s, CPUSLAB_FLUSH);
Wei Yangd4ff6d32017-07-06 15:36:25 -07002477 deactivate_slab(s, c->page, c->freelist, c);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002478
2479 c->tid = next_tid(c->tid);
Christoph Lameter81819f02007-05-06 14:49:36 -07002480}
2481
2482/*
2483 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002484 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002485 * Called from IPI handler with interrupts disabled.
2486 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002487static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002488{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002489 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002490
Wei Yang1265ef22018-12-28 00:33:06 -08002491 if (c->page)
2492 flush_slab(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002493
Wei Yang1265ef22018-12-28 00:33:06 -08002494 unfreeze_partials(s, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07002495}
2496
2497static void flush_cpu_slab(void *d)
2498{
2499 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002500
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002501 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002502}
2503
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002504static bool has_cpu_slab(int cpu, void *info)
2505{
2506 struct kmem_cache *s = info;
2507 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2508
Wei Yanga93cf072017-07-06 15:36:31 -07002509 return c->page || slub_percpu_partial(c);
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002510}
2511
Christoph Lameter81819f02007-05-06 14:49:36 -07002512static void flush_all(struct kmem_cache *s)
2513{
Sebastian Andrzej Siewiorcb923152020-01-17 10:01:37 +01002514 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07002515}
2516
2517/*
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02002518 * Use the cpu notifier to insure that the cpu slabs are flushed when
2519 * necessary.
2520 */
2521static int slub_cpu_dead(unsigned int cpu)
2522{
2523 struct kmem_cache *s;
2524 unsigned long flags;
2525
2526 mutex_lock(&slab_mutex);
2527 list_for_each_entry(s, &slab_caches, list) {
2528 local_irq_save(flags);
2529 __flush_cpu_slab(s, cpu);
2530 local_irq_restore(flags);
2531 }
2532 mutex_unlock(&slab_mutex);
2533 return 0;
2534}
2535
2536/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002537 * Check if the objects in a per cpu structure fit numa
2538 * locality expectations.
2539 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002540static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002541{
2542#ifdef CONFIG_NUMA
Wei Yang6159d0f2018-12-28 00:33:09 -08002543 if (node != NUMA_NO_NODE && page_to_nid(page) != node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002544 return 0;
2545#endif
2546 return 1;
2547}
2548
David Rientjes9a02d692014-06-04 16:06:36 -07002549#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002550static int count_free(struct page *page)
2551{
2552 return page->objects - page->inuse;
2553}
2554
David Rientjes9a02d692014-06-04 16:06:36 -07002555static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2556{
2557 return atomic_long_read(&n->total_objects);
2558}
2559#endif /* CONFIG_SLUB_DEBUG */
2560
2561#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002562static unsigned long count_partial(struct kmem_cache_node *n,
2563 int (*get_count)(struct page *))
2564{
2565 unsigned long flags;
2566 unsigned long x = 0;
2567 struct page *page;
2568
2569 spin_lock_irqsave(&n->list_lock, flags);
Tobin C. Harding916ac052019-05-13 17:16:12 -07002570 list_for_each_entry(page, &n->partial, slab_list)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002571 x += get_count(page);
2572 spin_unlock_irqrestore(&n->list_lock, flags);
2573 return x;
2574}
David Rientjes9a02d692014-06-04 16:06:36 -07002575#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002576
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002577static noinline void
2578slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2579{
David Rientjes9a02d692014-06-04 16:06:36 -07002580#ifdef CONFIG_SLUB_DEBUG
2581 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2582 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002583 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002584 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002585
David Rientjes9a02d692014-06-04 16:06:36 -07002586 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2587 return;
2588
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07002589 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2590 nid, gfpflags, &gfpflags);
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07002591 pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n",
Fabian Frederickf9f58282014-06-04 16:06:34 -07002592 s->name, s->object_size, s->size, oo_order(s->oo),
2593 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002594
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002595 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002596 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2597 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002598
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002599 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002600 unsigned long nr_slabs;
2601 unsigned long nr_objs;
2602 unsigned long nr_free;
2603
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002604 nr_free = count_partial(n, count_free);
2605 nr_slabs = node_nr_slabs(n);
2606 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002607
Fabian Frederickf9f58282014-06-04 16:06:34 -07002608 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002609 node, nr_slabs, nr_objs, nr_free);
2610 }
David Rientjes9a02d692014-06-04 16:06:36 -07002611#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002612}
2613
Christoph Lameter497b66f2011-08-09 16:12:26 -05002614static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2615 int node, struct kmem_cache_cpu **pc)
2616{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002617 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002618 struct kmem_cache_cpu *c = *pc;
2619 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002620
Matthew Wilcox128227e2018-06-07 17:05:13 -07002621 WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO));
2622
Christoph Lameter188fd062012-05-09 10:09:55 -05002623 freelist = get_partial(s, flags, node, c);
2624
2625 if (freelist)
2626 return freelist;
2627
2628 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002629 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002630 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002631 if (c->page)
2632 flush_slab(s, c);
2633
2634 /*
2635 * No other reference to the page yet so we can
2636 * muck around with it freely without cmpxchg
2637 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002638 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002639 page->freelist = NULL;
2640
2641 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002642 c->page = page;
2643 *pc = c;
Peng Wangedde82b2019-03-05 15:42:00 -08002644 }
Christoph Lameter497b66f2011-08-09 16:12:26 -05002645
Christoph Lameter6faa6832012-05-09 10:09:51 -05002646 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002647}
2648
Mel Gorman072bb0a2012-07-31 16:43:58 -07002649static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2650{
2651 if (unlikely(PageSlabPfmemalloc(page)))
2652 return gfp_pfmemalloc_allowed(gfpflags);
2653
2654 return true;
2655}
2656
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002657/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002658 * Check the page->freelist of a page and either transfer the freelist to the
2659 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002660 *
2661 * The page is still frozen if the return value is not NULL.
2662 *
2663 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002664 *
2665 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002666 */
2667static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2668{
2669 struct page new;
2670 unsigned long counters;
2671 void *freelist;
2672
2673 do {
2674 freelist = page->freelist;
2675 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002676
Christoph Lameter213eeb92011-11-11 14:07:14 -06002677 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002678 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002679
2680 new.inuse = page->objects;
2681 new.frozen = freelist != NULL;
2682
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002683 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002684 freelist, counters,
2685 NULL, new.counters,
2686 "get_freelist"));
2687
2688 return freelist;
2689}
2690
2691/*
Christoph Lameter894b87882007-05-10 03:15:16 -07002692 * Slow path. The lockless freelist is empty or we need to perform
2693 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002694 *
Christoph Lameter894b87882007-05-10 03:15:16 -07002695 * Processing is still very fast if new objects have been freed to the
2696 * regular freelist. In that case we simply take over the regular freelist
2697 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002698 *
Christoph Lameter894b87882007-05-10 03:15:16 -07002699 * If that is not working then we fall back to the partial lists. We take the
2700 * first element of the freelist as the object to allocate now and move the
2701 * rest of the freelist to the lockless freelist.
2702 *
2703 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002704 * we need to allocate a new slab. This is the slowest path since it involves
2705 * a call to the page allocator and the setup of a new slab.
Christoph Lametera380a3c2015-11-20 15:57:35 -08002706 *
2707 * Version of __slab_alloc to use when we know that interrupts are
2708 * already disabled (which is the case for bulk allocation).
Christoph Lameter81819f02007-05-06 14:49:36 -07002709 */
Christoph Lametera380a3c2015-11-20 15:57:35 -08002710static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002711 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002712{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002713 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002714 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002715
Abel Wu9f986d92020-10-13 16:48:43 -07002716 stat(s, ALLOC_SLOWPATH);
2717
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002718 page = c->page;
Vlastimil Babka0715e6c2020-03-21 18:22:37 -07002719 if (!page) {
2720 /*
2721 * if the node is not online or has no normal memory, just
2722 * ignore the node constraint
2723 */
2724 if (unlikely(node != NUMA_NO_NODE &&
2725 !node_state(node, N_NORMAL_MEMORY)))
2726 node = NUMA_NO_NODE;
Christoph Lameter81819f02007-05-06 14:49:36 -07002727 goto new_slab;
Vlastimil Babka0715e6c2020-03-21 18:22:37 -07002728 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002729redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002730
Christoph Lameter57d437d2012-05-09 10:09:59 -05002731 if (unlikely(!node_match(page, node))) {
Vlastimil Babka0715e6c2020-03-21 18:22:37 -07002732 /*
2733 * same as above but node_match() being false already
2734 * implies node != NUMA_NO_NODE
2735 */
2736 if (!node_state(node, N_NORMAL_MEMORY)) {
2737 node = NUMA_NO_NODE;
2738 goto redo;
2739 } else {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002740 stat(s, ALLOC_NODE_MISMATCH);
Wei Yangd4ff6d32017-07-06 15:36:25 -07002741 deactivate_slab(s, page, c->freelist, c);
Joonsoo Kima561ce02014-10-09 15:26:15 -07002742 goto new_slab;
2743 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002744 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002745
Mel Gorman072bb0a2012-07-31 16:43:58 -07002746 /*
2747 * By rights, we should be searching for a slab page that was
2748 * PFMEMALLOC but right now, we are losing the pfmemalloc
2749 * information when the page leaves the per-cpu allocator
2750 */
2751 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
Wei Yangd4ff6d32017-07-06 15:36:25 -07002752 deactivate_slab(s, page, c->freelist, c);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002753 goto new_slab;
2754 }
2755
Eric Dumazet73736e02011-12-13 04:57:06 +01002756 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002757 freelist = c->freelist;
2758 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002759 goto load_freelist;
2760
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002761 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002762
Christoph Lameter6faa6832012-05-09 10:09:51 -05002763 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002764 c->page = NULL;
2765 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002766 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002767 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002768
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002770
Christoph Lameter894b87882007-05-10 03:15:16 -07002771load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002772 /*
2773 * freelist is pointing to the list of objects to be used.
2774 * page is pointing to the page from which the objects are obtained.
2775 * That page must be frozen for per cpu allocations to work.
2776 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002777 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002778 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002779 c->tid = next_tid(c->tid);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002780 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002781
Christoph Lameter81819f02007-05-06 14:49:36 -07002782new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002783
Wei Yanga93cf072017-07-06 15:36:31 -07002784 if (slub_percpu_partial(c)) {
2785 page = c->page = slub_percpu_partial(c);
2786 slub_set_percpu_partial(c, page);
Christoph Lameter49e22582011-08-09 16:12:27 -05002787 stat(s, CPU_PARTIAL_ALLOC);
Christoph Lameter49e22582011-08-09 16:12:27 -05002788 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002789 }
2790
Christoph Lameter188fd062012-05-09 10:09:55 -05002791 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002792
Christoph Lameterf46974362012-05-09 10:09:54 -05002793 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002794 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002795 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002796 }
Christoph Lameter894b87882007-05-10 03:15:16 -07002797
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002798 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002799 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002800 goto load_freelist;
Christoph Lameter894b87882007-05-10 03:15:16 -07002801
Christoph Lameter497b66f2011-08-09 16:12:26 -05002802 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002803 if (kmem_cache_debug(s) &&
2804 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002805 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002806
Wei Yangd4ff6d32017-07-06 15:36:25 -07002807 deactivate_slab(s, page, get_freepointer(s, freelist), c);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002808 return freelist;
Christoph Lameter894b87882007-05-10 03:15:16 -07002809}
2810
2811/*
Christoph Lametera380a3c2015-11-20 15:57:35 -08002812 * Another one that disabled interrupt and compensates for possible
2813 * cpu changes by refetching the per cpu area pointer.
2814 */
2815static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2816 unsigned long addr, struct kmem_cache_cpu *c)
2817{
2818 void *p;
2819 unsigned long flags;
2820
2821 local_irq_save(flags);
Thomas Gleixner923717c2019-10-15 21:18:12 +02002822#ifdef CONFIG_PREEMPTION
Christoph Lametera380a3c2015-11-20 15:57:35 -08002823 /*
2824 * We may have been preempted and rescheduled on a different
2825 * cpu before disabling interrupts. Need to reload cpu area
2826 * pointer.
2827 */
2828 c = this_cpu_ptr(s->cpu_slab);
2829#endif
2830
2831 p = ___slab_alloc(s, gfpflags, node, addr, c);
2832 local_irq_restore(flags);
2833 return p;
2834}
2835
2836/*
Alexander Potapenko0f181f92019-10-14 14:11:57 -07002837 * If the object has been wiped upon free, make sure it's fully initialized by
2838 * zeroing out freelist pointer.
2839 */
2840static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s,
2841 void *obj)
2842{
2843 if (unlikely(slab_want_init_on_free(s)) && obj)
Andrey Konovalov5c96cfe2021-01-23 21:01:38 -08002844 memset((void *)((char *)kasan_reset_tag(obj) + s->offset),
2845 0, sizeof(void *));
Alexander Potapenko0f181f92019-10-14 14:11:57 -07002846}
2847
2848/*
Christoph Lameter894b87882007-05-10 03:15:16 -07002849 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2850 * have the fastpath folded into their functions. So no function call
2851 * overhead for requests that can be satisfied on the fastpath.
2852 *
2853 * The fastpath works by first checking if the lockless freelist can be used.
2854 * If not then __slab_alloc is called for slow processing.
2855 *
2856 * Otherwise we can simply pick the next object from the lockless free list.
2857 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002858static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002859 gfp_t gfpflags, int node, unsigned long addr, size_t orig_size)
Christoph Lameter894b87882007-05-10 03:15:16 -07002860{
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002861 void *object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002862 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002863 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002864 unsigned long tid;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07002865 struct obj_cgroup *objcg = NULL;
Andrey Konovalov5a7af112021-03-18 17:01:41 +11002866 bool init = false;
Christoph Lameter1f842602008-01-07 23:20:30 -08002867
Roman Gushchin964d4bd2020-08-06 23:20:56 -07002868 s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags);
Vladimir Davydov8135be52014-12-12 16:56:38 -08002869 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002870 return NULL;
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002871
2872 object = kfence_alloc(s, orig_size, gfpflags);
2873 if (unlikely(object))
2874 goto out;
2875
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002876redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002877 /*
2878 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2879 * enabled. We may switch back and forth between cpus while
2880 * reading from one cpu area. That does not matter as long
2881 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002882 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002883 * We should guarantee that tid and kmem_cache are retrieved on
Thomas Gleixner923717c2019-10-15 21:18:12 +02002884 * the same cpu. It could be different if CONFIG_PREEMPTION so we need
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002885 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002886 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002887 do {
2888 tid = this_cpu_read(s->cpu_slab->tid);
2889 c = raw_cpu_ptr(s->cpu_slab);
Thomas Gleixner923717c2019-10-15 21:18:12 +02002890 } while (IS_ENABLED(CONFIG_PREEMPTION) &&
Mark Rutland859b7a02015-03-25 15:55:23 -07002891 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002892
2893 /*
2894 * Irqless object alloc/free algorithm used here depends on sequence
2895 * of fetching cpu_slab's data. tid should be fetched before anything
2896 * on c to guarantee that object and page associated with previous tid
2897 * won't be used with current tid. If we fetch tid first, object and
2898 * page could be one associated with next tid and our alloc/free
2899 * request will be failed. In this case, we will retry. So, no problem.
2900 */
2901 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002902
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002903 /*
2904 * The transaction ids are globally unique per cpu and per operation on
2905 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2906 * occurs on the right processor and that there was no operation on the
2907 * linked list in between.
2908 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002909
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002910 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002911 page = c->page;
Laurent Dufour22e46632020-11-13 22:51:53 -08002912 if (unlikely(!object || !page || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002913 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002914 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002915 void *next_object = get_freepointer_safe(s, object);
2916
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002917 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002918 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002919 * operation and if we are on the right processor.
2920 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002921 * The cmpxchg does the following atomically (without lock
2922 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002923 * 1. Relocate first pointer to the current per cpu area.
2924 * 2. Verify that tid and freelist have not been changed
2925 * 3. If they were not changed replace tid and freelist
2926 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002927 * Since this is without lock semantics the protection is only
2928 * against code executing on this cpu *not* from access by
2929 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002930 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002931 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002932 s->cpu_slab->freelist, s->cpu_slab->tid,
2933 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002934 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002935
2936 note_cmpxchg_failure("slab_alloc", s, tid);
2937 goto redo;
2938 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002939 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002940 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b87882007-05-10 03:15:16 -07002941 }
Alexander Potapenko0f181f92019-10-14 14:11:57 -07002942
Andrey Konovalov5c96cfe2021-01-23 21:01:38 -08002943 maybe_wipe_obj_freeptr(s, object);
Andrey Konovalov5a7af112021-03-18 17:01:41 +11002944 init = slab_want_init_on_alloc(gfpflags, s);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002945
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002946out:
Andrey Konovalov5a7af112021-03-18 17:01:41 +11002947 slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002948
Christoph Lameter894b87882007-05-10 03:15:16 -07002949 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002950}
2951
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002952static __always_inline void *slab_alloc(struct kmem_cache *s,
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002953 gfp_t gfpflags, unsigned long addr, size_t orig_size)
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002954{
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002955 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size);
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002956}
2957
Christoph Lameter81819f02007-05-06 14:49:36 -07002958void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2959{
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002960 void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002961
Chen Gangd0e0ac92013-07-15 09:05:29 +08002962 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2963 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002964
2965 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002966}
2967EXPORT_SYMBOL(kmem_cache_alloc);
2968
Li Zefan0f24f122009-12-11 15:45:30 +08002969#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002970void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002971{
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002972 void *ret = slab_alloc(s, gfpflags, _RET_IP_, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002973 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Andrey Konovalov01165232018-12-28 00:29:37 -08002974 ret = kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002975 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002976}
Richard Kennedy4a923792010-10-21 10:29:19 +01002977EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002978#endif
2979
Christoph Lameter81819f02007-05-06 14:49:36 -07002980#ifdef CONFIG_NUMA
2981void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2982{
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002983 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002984
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002985 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002986 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002987
2988 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002989}
2990EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002991
Li Zefan0f24f122009-12-11 15:45:30 +08002992#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002993void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002994 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002995 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002996{
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00002997 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002998
2999 trace_kmalloc_node(_RET_IP_, ret,
3000 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003001
Andrey Konovalov01165232018-12-28 00:29:37 -08003002 ret = kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01003003 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003004}
Richard Kennedy4a923792010-10-21 10:29:19 +01003005EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003006#endif
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07003007#endif /* CONFIG_NUMA */
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003008
Christoph Lameter81819f02007-05-06 14:49:36 -07003009/*
Kim Phillips94e4d712015-02-10 14:09:37 -08003010 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b87882007-05-10 03:15:16 -07003011 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07003012 *
Christoph Lameter894b87882007-05-10 03:15:16 -07003013 * So we still attempt to reduce cache line usage. Just take the slab
3014 * lock and free the item. If there is no additional partial page
3015 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07003016 */
Christoph Lameter894b87882007-05-10 03:15:16 -07003017static void __slab_free(struct kmem_cache *s, struct page *page,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003018 void *head, void *tail, int cnt,
3019 unsigned long addr)
3020
Christoph Lameter81819f02007-05-06 14:49:36 -07003021{
3022 void *prior;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003023 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003024 struct page new;
3025 unsigned long counters;
3026 struct kmem_cache_node *n = NULL;
Kees Cook3f649ab2020-06-03 13:09:38 -07003027 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003028
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003029 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07003030
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00003031 if (kfence_free(head))
3032 return;
3033
Christoph Lameter19c7ff92012-05-30 12:54:46 -05003034 if (kmem_cache_debug(s) &&
Laura Abbott282acb42016-03-15 14:54:59 -07003035 !free_debug_processing(s, page, head, tail, cnt, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05003036 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08003037
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003038 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09003039 if (unlikely(n)) {
3040 spin_unlock_irqrestore(&n->list_lock, flags);
3041 n = NULL;
3042 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003043 prior = page->freelist;
3044 counters = page->counters;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003045 set_freepointer(s, tail, prior);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003046 new.counters = counters;
3047 was_frozen = new.frozen;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003048 new.inuse -= cnt;
Joonsoo Kim837d6782012-08-16 00:02:40 +09003049 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05003050
Peter Zijlstrac65c1872014-01-10 13:23:49 +01003051 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05003052
3053 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08003054 * Slab was on no list before and will be
3055 * partially empty
3056 * We can defer the list move and instead
3057 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05003058 */
3059 new.frozen = 1;
3060
Peter Zijlstrac65c1872014-01-10 13:23:49 +01003061 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05003062
LQYMGTb455def2014-12-10 15:42:13 -08003063 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05003064 /*
3065 * Speculatively acquire the list_lock.
3066 * If the cmpxchg does not succeed then we may
3067 * drop the list_lock without any processing.
3068 *
3069 * Otherwise the list_lock will synchronize with
3070 * other processors updating the list of slabs.
3071 */
3072 spin_lock_irqsave(&n->list_lock, flags);
3073
3074 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003075 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003076
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003077 } while (!cmpxchg_double_slab(s, page,
3078 prior, counters,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003079 head, new.counters,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003080 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07003081
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003082 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05003083
Abel Wuc270cf32020-10-13 16:48:40 -07003084 if (likely(was_frozen)) {
3085 /*
3086 * The list lock was not taken therefore no list
3087 * activity can be necessary.
3088 */
3089 stat(s, FREE_FROZEN);
3090 } else if (new.frozen) {
3091 /*
3092 * If we just froze the page then put it onto the
3093 * per cpu partial list.
3094 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003095 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08003096 stat(s, CPU_PARTIAL_FREE);
3097 }
Abel Wuc270cf32020-10-13 16:48:40 -07003098
LQYMGTb455def2014-12-10 15:42:13 -08003099 return;
3100 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003101
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07003102 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09003103 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07003104
Joonsoo Kim837d6782012-08-16 00:02:40 +09003105 /*
3106 * Objects left in the slab. If it was not on the partial list before
3107 * then add it.
3108 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003109 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
Liu Xianga4d3f892019-05-13 17:16:22 -07003110 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09003111 add_partial(n, page, DEACTIVATE_TO_TAIL);
3112 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003113 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05003114 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003115 return;
3116
3117slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08003118 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003119 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05003120 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07003121 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05003122 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06003123 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01003124 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05003125 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01003126 remove_full(s, n, page);
3127 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05003128
Christoph Lameter80f08c12011-06-01 12:25:55 -05003129 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06003130 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07003131 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003132}
3133
Christoph Lameter894b87882007-05-10 03:15:16 -07003134/*
3135 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
3136 * can perform fastpath freeing without additional function calls.
3137 *
3138 * The fastpath is only possible if we are freeing to the current cpu slab
3139 * of this processor. This typically the case if we have just allocated
3140 * the item before.
3141 *
3142 * If fastpath is not possible then fall back to __slab_free where we deal
3143 * with all sorts of special processing.
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003144 *
3145 * Bulk free of a freelist with several objects (all pointing to the
3146 * same page) possible by specifying head and tail ptr, plus objects
3147 * count (cnt). Bulk free indicated by tail pointer being set.
Christoph Lameter894b87882007-05-10 03:15:16 -07003148 */
Alexander Potapenko80a92012016-07-28 15:49:07 -07003149static __always_inline void do_slab_free(struct kmem_cache *s,
3150 struct page *page, void *head, void *tail,
3151 int cnt, unsigned long addr)
Christoph Lameter894b87882007-05-10 03:15:16 -07003152{
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003153 void *tail_obj = tail ? : head;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003154 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003155 unsigned long tid;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003156
Bharata B Raod1b2cf62020-10-13 16:53:09 -07003157 memcg_slab_free_hook(s, &head, 1);
Christoph Lametera24c5a02011-03-15 12:45:21 -05003158redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003159 /*
3160 * Determine the currently cpus per cpu slab.
3161 * The cpu may change afterward. However that does not matter since
3162 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07003163 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003164 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08003165 do {
3166 tid = this_cpu_read(s->cpu_slab->tid);
3167 c = raw_cpu_ptr(s->cpu_slab);
Thomas Gleixner923717c2019-10-15 21:18:12 +02003168 } while (IS_ENABLED(CONFIG_PREEMPTION) &&
Mark Rutland859b7a02015-03-25 15:55:23 -07003169 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05003170
Joonsoo Kim9aabf812015-02-10 14:09:32 -08003171 /* Same with comment on barrier() in slab_alloc_node() */
3172 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05003173
Christoph Lameter442b06b2011-05-17 16:29:31 -05003174 if (likely(page == c->page)) {
Linus Torvalds50761902020-03-17 11:04:09 -07003175 void **freelist = READ_ONCE(c->freelist);
3176
3177 set_freepointer(s, tail_obj, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003178
Christoph Lameter933393f2011-12-22 11:58:51 -06003179 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003180 s->cpu_slab->freelist, s->cpu_slab->tid,
Linus Torvalds50761902020-03-17 11:04:09 -07003181 freelist, tid,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003182 head, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003183
3184 note_cmpxchg_failure("slab_free", s, tid);
3185 goto redo;
3186 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06003187 stat(s, FREE_FASTPATH);
Christoph Lameter894b87882007-05-10 03:15:16 -07003188 } else
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003189 __slab_free(s, page, head, tail_obj, cnt, addr);
Christoph Lameter894b87882007-05-10 03:15:16 -07003190
Christoph Lameter894b87882007-05-10 03:15:16 -07003191}
3192
Alexander Potapenko80a92012016-07-28 15:49:07 -07003193static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
3194 void *head, void *tail, int cnt,
3195 unsigned long addr)
3196{
Alexander Potapenko80a92012016-07-28 15:49:07 -07003197 /*
Andrey Konovalovc3895392018-04-10 16:30:31 -07003198 * With KASAN enabled slab_free_freelist_hook modifies the freelist
3199 * to remove objects, whose reuse must be delayed.
Alexander Potapenko80a92012016-07-28 15:49:07 -07003200 */
Andrey Konovalovc3895392018-04-10 16:30:31 -07003201 if (slab_free_freelist_hook(s, &head, &tail))
3202 do_slab_free(s, page, head, tail, cnt, addr);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003203}
3204
Andrey Konovalov2bd926b2018-12-28 00:29:53 -08003205#ifdef CONFIG_KASAN_GENERIC
Alexander Potapenko80a92012016-07-28 15:49:07 -07003206void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
3207{
3208 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
3209}
3210#endif
3211
Christoph Lameter81819f02007-05-06 14:49:36 -07003212void kmem_cache_free(struct kmem_cache *s, void *x)
3213{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003214 s = cache_from_obj(s, x);
3215 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00003216 return;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003217 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003218 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07003219}
3220EXPORT_SYMBOL(kmem_cache_free);
3221
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003222struct detached_freelist {
3223 struct page *page;
3224 void *tail;
3225 void *freelist;
3226 int cnt;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003227 struct kmem_cache *s;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003228};
3229
3230/*
3231 * This function progressively scans the array with free objects (with
3232 * a limited look ahead) and extract objects belonging to the same
3233 * page. It builds a detached freelist directly within the given
3234 * page/objects. This can happen without any need for
3235 * synchronization, because the objects are owned by running process.
3236 * The freelist is build up as a single linked list in the objects.
3237 * The idea is, that this detached freelist can then be bulk
3238 * transferred to the real freelist(s), but only requiring a single
3239 * synchronization primitive. Look ahead in the array is limited due
3240 * to performance reasons.
3241 */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003242static inline
3243int build_detached_freelist(struct kmem_cache *s, size_t size,
3244 void **p, struct detached_freelist *df)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003245{
3246 size_t first_skipped_index = 0;
3247 int lookahead = 3;
3248 void *object;
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003249 struct page *page;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003250
3251 /* Always re-init detached_freelist */
3252 df->page = NULL;
3253
3254 do {
3255 object = p[--size];
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003256 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003257 } while (!object && size);
3258
3259 if (!object)
3260 return 0;
3261
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003262 page = virt_to_head_page(object);
3263 if (!s) {
3264 /* Handle kalloc'ed objects */
3265 if (unlikely(!PageSlab(page))) {
3266 BUG_ON(!PageCompound(page));
3267 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003268 __free_pages(page, compound_order(page));
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003269 p[size] = NULL; /* mark object processed */
3270 return size;
3271 }
3272 /* Derive kmem_cache from object */
3273 df->s = page->slab_cache;
3274 } else {
3275 df->s = cache_from_obj(s, object); /* Support for memcg */
3276 }
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003277
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00003278 if (is_kfence_address(object)) {
Andrey Konovalov24690d72021-03-18 17:01:41 +11003279 slab_free_hook(df->s, object, false);
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00003280 __kfence_free(object);
3281 p[size] = NULL; /* mark object processed */
3282 return size;
3283 }
3284
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003285 /* Start new detached freelist */
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003286 df->page = page;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003287 set_freepointer(df->s, object, NULL);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003288 df->tail = object;
3289 df->freelist = object;
3290 p[size] = NULL; /* mark object processed */
3291 df->cnt = 1;
3292
3293 while (size) {
3294 object = p[--size];
3295 if (!object)
3296 continue; /* Skip processed objects */
3297
3298 /* df->page is always set at this point */
3299 if (df->page == virt_to_head_page(object)) {
3300 /* Opportunity build freelist */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003301 set_freepointer(df->s, object, df->freelist);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003302 df->freelist = object;
3303 df->cnt++;
3304 p[size] = NULL; /* mark object processed */
3305
3306 continue;
3307 }
3308
3309 /* Limit look ahead search */
3310 if (!--lookahead)
3311 break;
3312
3313 if (!first_skipped_index)
3314 first_skipped_index = size + 1;
3315 }
3316
3317 return first_skipped_index;
3318}
3319
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003320/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003321void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003322{
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003323 if (WARN_ON(!size))
3324 return;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003325
Bharata B Raod1b2cf62020-10-13 16:53:09 -07003326 memcg_slab_free_hook(s, p, size);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003327 do {
3328 struct detached_freelist df;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003329
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003330 size = build_detached_freelist(s, size, p, &df);
Arnd Bergmann84582c82016-12-12 16:41:35 -08003331 if (!df.page)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003332 continue;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003333
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003334 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003335 } while (likely(size));
Christoph Lameter484748f2015-09-04 15:45:34 -07003336}
3337EXPORT_SYMBOL(kmem_cache_free_bulk);
3338
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003339/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003340int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3341 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003342{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003343 struct kmem_cache_cpu *c;
3344 int i;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003345 struct obj_cgroup *objcg = NULL;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003346
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003347 /* memcg and kmem_cache debug support */
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003348 s = slab_pre_alloc_hook(s, &objcg, size, flags);
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003349 if (unlikely(!s))
3350 return false;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003351 /*
3352 * Drain objects in the per cpu slab, while disabling local
3353 * IRQs, which protects against PREEMPT and interrupts
3354 * handlers invoking normal fastpath.
3355 */
3356 local_irq_disable();
3357 c = this_cpu_ptr(s->cpu_slab);
3358
3359 for (i = 0; i < size; i++) {
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00003360 void *object = kfence_alloc(s, s->object_size, flags);
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003361
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00003362 if (unlikely(object)) {
3363 p[i] = object;
3364 continue;
3365 }
3366
3367 object = c->freelist;
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003368 if (unlikely(!object)) {
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003369 /*
Jann Hornfd4d9c72020-03-17 01:28:45 +01003370 * We may have removed an object from c->freelist using
3371 * the fastpath in the previous iteration; in that case,
3372 * c->tid has not been bumped yet.
3373 * Since ___slab_alloc() may reenable interrupts while
3374 * allocating memory, we should bump c->tid now.
3375 */
3376 c->tid = next_tid(c->tid);
3377
3378 /*
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003379 * Invoking slow path likely have side-effect
3380 * of re-populating per CPU c->freelist
3381 */
Christoph Lameter87098372015-11-20 15:57:38 -08003382 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003383 _RET_IP_, c);
Christoph Lameter87098372015-11-20 15:57:38 -08003384 if (unlikely(!p[i]))
3385 goto error;
3386
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003387 c = this_cpu_ptr(s->cpu_slab);
Alexander Potapenko0f181f92019-10-14 14:11:57 -07003388 maybe_wipe_obj_freeptr(s, p[i]);
3389
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003390 continue; /* goto for-loop */
3391 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003392 c->freelist = get_freepointer(s, object);
3393 p[i] = object;
Alexander Potapenko0f181f92019-10-14 14:11:57 -07003394 maybe_wipe_obj_freeptr(s, p[i]);
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003395 }
3396 c->tid = next_tid(c->tid);
3397 local_irq_enable();
3398
Andrey Konovalov5a7af112021-03-18 17:01:41 +11003399 /*
3400 * memcg and kmem_cache debug support and memory initialization.
3401 * Done outside of the IRQ disabled fastpath loop.
3402 */
3403 slab_post_alloc_hook(s, objcg, flags, size, p,
3404 slab_want_init_on_alloc(flags, s));
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003405 return i;
Christoph Lameter87098372015-11-20 15:57:38 -08003406error:
Christoph Lameter87098372015-11-20 15:57:38 -08003407 local_irq_enable();
Andrey Konovalov5a7af112021-03-18 17:01:41 +11003408 slab_post_alloc_hook(s, objcg, flags, i, p, false);
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003409 __kmem_cache_free_bulk(s, i, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003410 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003411}
3412EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3413
3414
Christoph Lameter81819f02007-05-06 14:49:36 -07003415/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003416 * Object placement in a slab is made very easy because we always start at
3417 * offset 0. If we tune the size of the object to the alignment then we can
3418 * get the required alignment by putting one properly sized object after
3419 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07003420 *
3421 * Notice that the allocation order determines the sizes of the per cpu
3422 * caches. Each processor has always one slab available for allocations.
3423 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07003424 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07003425 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07003426 */
3427
3428/*
3429 * Mininum / Maximum order of slab pages. This influences locking overhead
3430 * and slab fragmentation. A higher order reduces the number of partial slabs
3431 * and increases the number of allocations possible without having to
3432 * take the list_lock.
3433 */
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003434static unsigned int slub_min_order;
3435static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
3436static unsigned int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07003437
3438/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003439 * Calculate the order of allocation given an slab object size.
3440 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003441 * The order of allocation has significant impact on performance and other
3442 * system components. Generally order 0 allocations should be preferred since
3443 * order 0 does not cause fragmentation in the page allocator. Larger objects
3444 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003445 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07003446 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07003447 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003448 * In order to reach satisfactory performance we must ensure that a minimum
3449 * number of objects is in one slab. Otherwise we may generate too much
3450 * activity on the partial lists which requires taking the list_lock. This is
3451 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07003452 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003453 * slub_max_order specifies the order where we begin to stop considering the
3454 * number of objects in a slab as critical. If we reach slub_max_order then
3455 * we try to keep the page order as low as possible. So we accept more waste
3456 * of space in favor of a small page order.
3457 *
3458 * Higher order allocations also allow the placement of more objects in a
3459 * slab and thereby reduce object handling overhead. If the user has
3460 * requested a higher mininum order then we start with that one instead of
3461 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07003462 */
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003463static inline unsigned int slab_order(unsigned int size,
3464 unsigned int min_objects, unsigned int max_order,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003465 unsigned int fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003466{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003467 unsigned int min_order = slub_min_order;
3468 unsigned int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003469
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003470 if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04003471 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03003472
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003473 for (order = max(min_order, (unsigned int)get_order(min_objects * size));
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003474 order <= max_order; order++) {
3475
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003476 unsigned int slab_size = (unsigned int)PAGE_SIZE << order;
3477 unsigned int rem;
Christoph Lameter81819f02007-05-06 14:49:36 -07003478
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003479 rem = slab_size % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003480
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003481 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003482 break;
Christoph Lameter81819f02007-05-06 14:49:36 -07003483 }
Christoph Lameter672bba32007-05-09 02:32:39 -07003484
Christoph Lameter81819f02007-05-06 14:49:36 -07003485 return order;
3486}
3487
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003488static inline int calculate_order(unsigned int size)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003489{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003490 unsigned int order;
3491 unsigned int min_objects;
3492 unsigned int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003493
3494 /*
3495 * Attempt to find best configuration for a slab. This
3496 * works by first attempting to generate a layout with
3497 * the best configuration and backing off gradually.
3498 *
Wei Yang422ff4d2015-11-05 18:45:46 -08003499 * First we increase the acceptable waste in a slab. Then
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003500 * we reduce the minimum objects required in a slab.
3501 */
3502 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003503 if (!min_objects)
3504 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003505 max_objects = order_objects(slub_max_order, size);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003506 min_objects = min(min_objects, max_objects);
3507
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003508 while (min_objects > 1) {
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003509 unsigned int fraction;
3510
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003511 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003512 while (fraction >= 4) {
3513 order = slab_order(size, min_objects,
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003514 slub_max_order, fraction);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003515 if (order <= slub_max_order)
3516 return order;
3517 fraction /= 2;
3518 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03003519 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003520 }
3521
3522 /*
3523 * We were unable to place multiple objects in a slab. Now
3524 * lets see if we can place a single object there.
3525 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003526 order = slab_order(size, 1, slub_max_order, 1);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003527 if (order <= slub_max_order)
3528 return order;
3529
3530 /*
3531 * Doh this slab cannot be placed using slub_max_order.
3532 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003533 order = slab_order(size, 1, MAX_ORDER, 1);
David Rientjes818cf592009-04-23 09:58:22 +03003534 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003535 return order;
3536 return -ENOSYS;
3537}
3538
Pekka Enberg5595cff2008-08-05 09:28:47 +03003539static void
Joonsoo Kim40534972012-05-11 00:50:47 +09003540init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003541{
3542 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003543 spin_lock_init(&n->list_lock);
3544 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003545#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003546 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07003547 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07003548 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003549#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003550}
3551
Christoph Lameter55136592010-08-20 12:37:13 -05003552static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003553{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05003554 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003555 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003556
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003557 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003558 * Must align to double word boundary for the double cmpxchg
3559 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003560 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003561 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3562 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003563
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003564 if (!s->cpu_slab)
3565 return 0;
3566
3567 init_kmem_cache_cpus(s);
3568
3569 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003570}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003571
Christoph Lameter51df1142010-08-20 12:37:15 -05003572static struct kmem_cache *kmem_cache_node;
3573
Christoph Lameter81819f02007-05-06 14:49:36 -07003574/*
3575 * No kmalloc_node yet so do it by hand. We know that this is the first
3576 * slab on the node for this slabcache. There are no concurrent accesses
3577 * possible.
3578 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003579 * Note that this function only works on the kmem_cache_node
3580 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003581 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003582 */
Christoph Lameter55136592010-08-20 12:37:13 -05003583static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003584{
3585 struct page *page;
3586 struct kmem_cache_node *n;
3587
Christoph Lameter51df1142010-08-20 12:37:15 -05003588 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003589
Christoph Lameter51df1142010-08-20 12:37:15 -05003590 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003591
3592 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003593 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003594 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3595 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003596 }
3597
Christoph Lameter81819f02007-05-06 14:49:36 -07003598 n = page->freelist;
3599 BUG_ON(!n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003600#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003601 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003602 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003603#endif
Andrey Konovalov5a7af112021-03-18 17:01:41 +11003604 n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false);
Andrey Konovalov12b22382018-12-28 00:29:41 -08003605 page->freelist = get_freepointer(kmem_cache_node, n);
3606 page->inuse = 1;
3607 page->frozen = 0;
3608 kmem_cache_node->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09003609 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003610 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003611
Dave Hansen67b6c902014-01-24 07:20:23 -08003612 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003613 * No locks need to be taken here as it has just been
3614 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003615 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003616 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003617}
3618
3619static void free_kmem_cache_nodes(struct kmem_cache *s)
3620{
3621 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003622 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003623
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003624 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003625 s->node[node] = NULL;
Alexander Potapenkoea37df52017-09-06 16:19:15 -07003626 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003627 }
3628}
3629
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003630void __kmem_cache_release(struct kmem_cache *s)
3631{
Thomas Garnier210e7a42016-07-26 15:21:59 -07003632 cache_random_seq_destroy(s);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003633 free_percpu(s->cpu_slab);
3634 free_kmem_cache_nodes(s);
3635}
3636
Christoph Lameter55136592010-08-20 12:37:13 -05003637static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003638{
3639 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003640
Christoph Lameterf64dc582007-10-16 01:25:33 -07003641 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003642 struct kmem_cache_node *n;
3643
Alexander Duyck73367bd2010-05-21 14:41:35 -07003644 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003645 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003646 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003647 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003648 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003649 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003650
3651 if (!n) {
3652 free_kmem_cache_nodes(s);
3653 return 0;
3654 }
3655
Joonsoo Kim40534972012-05-11 00:50:47 +09003656 init_kmem_cache_node(n);
Alexander Potapenkoea37df52017-09-06 16:19:15 -07003657 s->node[node] = n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003658 }
3659 return 1;
3660}
Christoph Lameter81819f02007-05-06 14:49:36 -07003661
David Rientjesc0bdb232009-02-25 09:16:35 +02003662static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003663{
3664 if (min < MIN_PARTIAL)
3665 min = MIN_PARTIAL;
3666 else if (min > MAX_PARTIAL)
3667 min = MAX_PARTIAL;
3668 s->min_partial = min;
3669}
3670
Wei Yange6d0e1d2017-07-06 15:36:34 -07003671static void set_cpu_partial(struct kmem_cache *s)
3672{
3673#ifdef CONFIG_SLUB_CPU_PARTIAL
3674 /*
3675 * cpu_partial determined the maximum number of objects kept in the
3676 * per cpu partial lists of a processor.
3677 *
3678 * Per cpu partial lists mainly contain slabs that just have one
3679 * object freed. If they are used for allocation then they can be
3680 * filled up again with minimal effort. The slab will never hit the
3681 * per node partial lists and therefore no locking will be required.
3682 *
3683 * This setting also determines
3684 *
3685 * A) The number of objects from per cpu partial slabs dumped to the
3686 * per node list when we reach the limit.
3687 * B) The number of objects in cpu partial slabs to extract from the
3688 * per node list when we run out of per cpu objects. We only fetch
3689 * 50% to keep some capacity around for frees.
3690 */
3691 if (!kmem_cache_has_cpu_partial(s))
chenqiwubbd4e302020-04-01 21:04:19 -07003692 slub_set_cpu_partial(s, 0);
Wei Yange6d0e1d2017-07-06 15:36:34 -07003693 else if (s->size >= PAGE_SIZE)
chenqiwubbd4e302020-04-01 21:04:19 -07003694 slub_set_cpu_partial(s, 2);
Wei Yange6d0e1d2017-07-06 15:36:34 -07003695 else if (s->size >= 1024)
chenqiwubbd4e302020-04-01 21:04:19 -07003696 slub_set_cpu_partial(s, 6);
Wei Yange6d0e1d2017-07-06 15:36:34 -07003697 else if (s->size >= 256)
chenqiwubbd4e302020-04-01 21:04:19 -07003698 slub_set_cpu_partial(s, 13);
Wei Yange6d0e1d2017-07-06 15:36:34 -07003699 else
chenqiwubbd4e302020-04-01 21:04:19 -07003700 slub_set_cpu_partial(s, 30);
Wei Yange6d0e1d2017-07-06 15:36:34 -07003701#endif
3702}
3703
Christoph Lameter81819f02007-05-06 14:49:36 -07003704/*
3705 * calculate_sizes() determines the order and the distribution of data within
3706 * a slab object.
3707 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003708static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003709{
Alexey Dobriyand50112e2017-11-15 17:32:18 -08003710 slab_flags_t flags = s->flags;
Alexey Dobriyanbe4a7982018-04-05 16:21:28 -07003711 unsigned int size = s->object_size;
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003712 unsigned int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003713
3714 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003715 * Round up object size to the next word boundary. We can only
3716 * place the free pointer at word boundaries and this determines
3717 * the possible location of the free pointer.
3718 */
3719 size = ALIGN(size, sizeof(void *));
3720
3721#ifdef CONFIG_SLUB_DEBUG
3722 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003723 * Determine if we can poison the object itself. If the user of
3724 * the slab may touch the object after free or before allocation
3725 * then we should never poison the object itself.
3726 */
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08003727 if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003728 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003729 s->flags |= __OBJECT_POISON;
3730 else
3731 s->flags &= ~__OBJECT_POISON;
3732
Christoph Lameter81819f02007-05-06 14:49:36 -07003733
3734 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003735 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003736 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003737 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003738 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003739 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003740 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003741#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003742
3743 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003744 * With that we have determined the number of bytes in actual use
Kees Cookf6ed2352021-06-15 18:23:26 -07003745 * by the object and redzoning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003746 */
3747 s->inuse = size;
3748
Kees Cook4314c8c2021-06-15 18:23:22 -07003749 if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) ||
3750 ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) ||
3751 s->ctor) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003752 /*
3753 * Relocate free pointer after the object if it is not
3754 * permitted to overwrite the first word of the object on
3755 * kmem_cache_free.
3756 *
3757 * This is the case if we do RCU, have a constructor or
Kees Cook4314c8c2021-06-15 18:23:22 -07003758 * destructor, are poisoning the objects, or are
3759 * redzoning an object smaller than sizeof(void *).
Waiman Longcbfc35a2020-05-07 18:36:06 -07003760 *
3761 * The assumption that s->offset >= s->inuse means free
3762 * pointer is outside of the object is used in the
3763 * freeptr_outside_object() function. If that is no
3764 * longer true, the function needs to be modified.
Christoph Lameter81819f02007-05-06 14:49:36 -07003765 */
3766 s->offset = size;
3767 size += sizeof(void *);
Kees Cookf6ed2352021-06-15 18:23:26 -07003768 } else {
Kees Cook3202fa62020-04-01 21:04:27 -07003769 /*
3770 * Store freelist pointer near middle of object to keep
3771 * it away from the edges of the object to avoid small
3772 * sized over/underflows from neighboring allocations.
3773 */
Kees Cookf6ed2352021-06-15 18:23:26 -07003774 s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
Christoph Lameter81819f02007-05-06 14:49:36 -07003775 }
3776
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003777#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003778 if (flags & SLAB_STORE_USER)
3779 /*
3780 * Need to store information about allocs and frees after
3781 * the object.
3782 */
3783 size += 2 * sizeof(struct track);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003784#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003785
Alexander Potapenko80a92012016-07-28 15:49:07 -07003786 kasan_cache_create(s, &size, &s->flags);
3787#ifdef CONFIG_SLUB_DEBUG
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003788 if (flags & SLAB_RED_ZONE) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003789 /*
3790 * Add some empty padding so that we can catch
3791 * overwrites from earlier objects rather than let
3792 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003793 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003794 * of the object.
3795 */
3796 size += sizeof(void *);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003797
3798 s->red_left_pad = sizeof(void *);
3799 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3800 size += s->red_left_pad;
3801 }
Christoph Lameter41ecc552007-05-09 02:32:44 -07003802#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003803
Christoph Lameter81819f02007-05-06 14:49:36 -07003804 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003805 * SLUB stores one object immediately after another beginning from
3806 * offset 0. In order to align the objects we have to simply size
3807 * each object to conform to the alignment.
3808 */
Christoph Lameter45906852012-11-28 16:23:16 +00003809 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003810 s->size = size;
Roman Gushchin4138fdf2020-08-06 23:20:42 -07003811 s->reciprocal_size = reciprocal_value(size);
Christoph Lameter06b285d2008-04-14 19:11:41 +03003812 if (forced_order >= 0)
3813 order = forced_order;
3814 else
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003815 order = calculate_order(size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003816
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003817 if ((int)order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003818 return 0;
3819
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003820 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003821 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003822 s->allocflags |= __GFP_COMP;
3823
3824 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003825 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003826
Nicolas Boichat6d6ea1e2019-03-28 20:43:42 -07003827 if (s->flags & SLAB_CACHE_DMA32)
3828 s->allocflags |= GFP_DMA32;
3829
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003830 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3831 s->allocflags |= __GFP_RECLAIMABLE;
3832
Christoph Lameter81819f02007-05-06 14:49:36 -07003833 /*
3834 * Determine the number of objects per slab
3835 */
Matthew Wilcox9736d2a2018-06-07 17:09:10 -07003836 s->oo = oo_make(order, size);
3837 s->min = oo_make(get_order(size), size);
Christoph Lameter205ab992008-04-14 19:11:40 +03003838 if (oo_objects(s->oo) > oo_objects(s->max))
3839 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003840
Christoph Lameter834f3d12008-04-14 19:11:31 +03003841 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003842}
3843
Alexey Dobriyand50112e2017-11-15 17:32:18 -08003844static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003845{
Nikolay Borisov2e95bc62021-02-24 12:00:58 -08003846 s->flags = kmem_cache_flags(s->size, flags, s->name);
Kees Cook2482ddec2017-09-06 16:19:18 -07003847#ifdef CONFIG_SLAB_FREELIST_HARDENED
3848 s->random = get_random_long();
3849#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003850
Christoph Lameter06b285d2008-04-14 19:11:41 +03003851 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003852 goto error;
David Rientjes3de472132009-07-27 18:30:35 -07003853 if (disable_higher_order_debug) {
3854 /*
3855 * Disable debugging flags that store metadata if the min slab
3856 * order increased.
3857 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003858 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de472132009-07-27 18:30:35 -07003859 s->flags &= ~DEBUG_METADATA_FLAGS;
3860 s->offset = 0;
3861 if (!calculate_sizes(s, -1))
3862 goto error;
3863 }
3864 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003865
Heiko Carstens25654092012-01-12 17:17:33 -08003866#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3867 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Laura Abbott149daaf2016-03-15 14:55:09 -07003868 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003869 /* Enable fast mode */
3870 s->flags |= __CMPXCHG_DOUBLE;
3871#endif
3872
David Rientjes3b89d7d2009-02-22 17:40:07 -08003873 /*
3874 * The larger the object size is, the more pages we want on the partial
3875 * list to avoid pounding the page allocator excessively.
3876 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003877 set_min_partial(s, ilog2(s->size) / 2);
3878
Wei Yange6d0e1d2017-07-06 15:36:34 -07003879 set_cpu_partial(s);
Christoph Lameter49e22582011-08-09 16:12:27 -05003880
Christoph Lameter81819f02007-05-06 14:49:36 -07003881#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003882 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003883#endif
Thomas Garnier210e7a42016-07-26 15:21:59 -07003884
3885 /* Initialize the pre-computed randomized freelist if slab is up */
3886 if (slab_state >= UP) {
3887 if (init_cache_random_seq(s))
3888 goto error;
3889 }
3890
Christoph Lameter55136592010-08-20 12:37:13 -05003891 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003892 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003893
Christoph Lameter55136592010-08-20 12:37:13 -05003894 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003895 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003896
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003897 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003898error:
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003899 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003900}
Christoph Lameter81819f02007-05-06 14:49:36 -07003901
Christoph Lameter33b12c32008-04-25 12:22:43 -07003902static void list_slab_objects(struct kmem_cache *s, struct page *page,
Sebastian Andrzej Siewior55860d92020-06-25 20:29:55 -07003903 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003904{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003905#ifdef CONFIG_SLUB_DEBUG
3906 void *addr = page_address(page);
Sebastian Andrzej Siewior55860d92020-06-25 20:29:55 -07003907 unsigned long *map;
Christoph Lameter33b12c32008-04-25 12:22:43 -07003908 void *p;
Christopher Lameteraa456c72020-06-01 21:45:53 -07003909
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003910 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003911 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003912
Yu Zhao90e9f6a2020-01-30 22:11:57 -08003913 map = get_map(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003914 for_each_object(p, s, addr, page->objects) {
3915
Roman Gushchin4138fdf2020-08-06 23:20:42 -07003916 if (!test_bit(__obj_to_index(s, addr, p), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003917 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003918 print_tracking(s, p);
3919 }
3920 }
Sebastian Andrzej Siewior55860d92020-06-25 20:29:55 -07003921 put_map(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003922 slab_unlock(page);
3923#endif
3924}
3925
Christoph Lameter81819f02007-05-06 14:49:36 -07003926/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003927 * Attempt to free all partial slabs on a node.
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003928 * This is called from __kmem_cache_shutdown(). We must take list_lock
3929 * because sysfs file might still access partial list after the shutdowning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003930 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003931static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003932{
Chris Wilson60398922016-08-10 16:27:58 -07003933 LIST_HEAD(discard);
Christoph Lameter81819f02007-05-06 14:49:36 -07003934 struct page *page, *h;
3935
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003936 BUG_ON(irqs_disabled());
3937 spin_lock_irq(&n->list_lock);
Tobin C. Harding916ac052019-05-13 17:16:12 -07003938 list_for_each_entry_safe(page, h, &n->partial, slab_list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003939 if (!page->inuse) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003940 remove_partial(n, page);
Tobin C. Harding916ac052019-05-13 17:16:12 -07003941 list_add(&page->slab_list, &discard);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003942 } else {
3943 list_slab_objects(s, page,
Sebastian Andrzej Siewior55860d92020-06-25 20:29:55 -07003944 "Objects remaining in %s on __kmem_cache_shutdown()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003945 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003946 }
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003947 spin_unlock_irq(&n->list_lock);
Chris Wilson60398922016-08-10 16:27:58 -07003948
Tobin C. Harding916ac052019-05-13 17:16:12 -07003949 list_for_each_entry_safe(page, h, &discard, slab_list)
Chris Wilson60398922016-08-10 16:27:58 -07003950 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003951}
3952
Shakeel Buttf9e13c02018-04-05 16:21:57 -07003953bool __kmem_cache_empty(struct kmem_cache *s)
3954{
3955 int node;
3956 struct kmem_cache_node *n;
3957
3958 for_each_kmem_cache_node(s, node, n)
3959 if (n->nr_partial || slabs_node(s, node))
3960 return false;
3961 return true;
3962}
3963
Christoph Lameter81819f02007-05-06 14:49:36 -07003964/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003965 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003966 */
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003967int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003968{
3969 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003970 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003971
3972 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003973 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003974 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003975 free_partial(s, n);
3976 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003977 return 1;
3978 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003979 return 0;
3980}
3981
Christoph Lameter81819f02007-05-06 14:49:36 -07003982/********************************************************************
3983 * Kmalloc subsystem
3984 *******************************************************************/
3985
Christoph Lameter81819f02007-05-06 14:49:36 -07003986static int __init setup_slub_min_order(char *str)
3987{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003988 get_option(&str, (int *)&slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003989
3990 return 1;
3991}
3992
3993__setup("slub_min_order=", setup_slub_min_order);
3994
3995static int __init setup_slub_max_order(char *str)
3996{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07003997 get_option(&str, (int *)&slub_max_order);
3998 slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003999
4000 return 1;
4001}
4002
4003__setup("slub_max_order=", setup_slub_max_order);
4004
4005static int __init setup_slub_min_objects(char *str)
4006{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07004007 get_option(&str, (int *)&slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07004008
4009 return 1;
4010}
4011
4012__setup("slub_min_objects=", setup_slub_min_objects);
4013
Christoph Lameter81819f02007-05-06 14:49:36 -07004014void *__kmalloc(size_t size, gfp_t flags)
4015{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004016 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004017 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004018
Christoph Lameter95a05b42013-01-10 19:14:19 +00004019 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004020 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004021
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004022 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004023
4024 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004025 return s;
4026
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004027 ret = slab_alloc(s, flags, _RET_IP_, size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004028
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004029 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004030
Andrey Konovalov01165232018-12-28 00:29:37 -08004031 ret = kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08004032
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004033 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004034}
4035EXPORT_SYMBOL(__kmalloc);
4036
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004037#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08004038static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
4039{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01004040 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01004041 void *ptr = NULL;
Vlastimil Babka6a486c02019-10-06 17:58:42 -07004042 unsigned int order = get_order(size);
Christoph Lameterf619cfe2008-03-01 13:56:40 -08004043
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -08004044 flags |= __GFP_COMP;
Vlastimil Babka6a486c02019-10-06 17:58:42 -07004045 page = alloc_pages_node(node, flags, order);
4046 if (page) {
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01004047 ptr = page_address(page);
Muchun Songcbb86d62021-02-24 12:04:26 -08004048 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
4049 PAGE_SIZE << order);
Vlastimil Babka6a486c02019-10-06 17:58:42 -07004050 }
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01004051
Andrey Konovalov01165232018-12-28 00:29:37 -08004052 return kmalloc_large_node_hook(ptr, size, flags);
Christoph Lameterf619cfe2008-03-01 13:56:40 -08004053}
4054
Christoph Lameter81819f02007-05-06 14:49:36 -07004055void *__kmalloc_node(size_t size, gfp_t flags, int node)
4056{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004057 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004058 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004059
Christoph Lameter95a05b42013-01-10 19:14:19 +00004060 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004061 ret = kmalloc_large_node(size, flags, node);
4062
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004063 trace_kmalloc_node(_RET_IP_, ret,
4064 size, PAGE_SIZE << get_order(size),
4065 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004066
4067 return ret;
4068 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004069
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004070 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004071
4072 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004073 return s;
4074
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004075 ret = slab_alloc_node(s, flags, node, _RET_IP_, size);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004076
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004077 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004078
Andrey Konovalov01165232018-12-28 00:29:37 -08004079 ret = kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08004080
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03004081 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004082}
4083EXPORT_SYMBOL(__kmalloc_node);
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07004084#endif /* CONFIG_NUMA */
Christoph Lameter81819f02007-05-06 14:49:36 -07004085
Kees Cooked18adc2016-06-23 15:24:05 -07004086#ifdef CONFIG_HARDENED_USERCOPY
4087/*
Kees Cookafcc90f82018-01-10 15:17:01 -08004088 * Rejects incorrectly sized objects and objects that are to be copied
4089 * to/from userspace but do not fall entirely within the containing slab
4090 * cache's usercopy region.
Kees Cooked18adc2016-06-23 15:24:05 -07004091 *
4092 * Returns NULL if check passes, otherwise const char * to name of cache
4093 * to indicate an error.
4094 */
Kees Cookf4e6e282018-01-10 14:48:22 -08004095void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
4096 bool to_user)
Kees Cooked18adc2016-06-23 15:24:05 -07004097{
4098 struct kmem_cache *s;
Alexey Dobriyan44065b22018-04-05 16:21:20 -07004099 unsigned int offset;
Kees Cooked18adc2016-06-23 15:24:05 -07004100 size_t object_size;
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004101 bool is_kfence = is_kfence_address(ptr);
Kees Cooked18adc2016-06-23 15:24:05 -07004102
Andrey Konovalov96fedce2019-01-08 15:23:15 -08004103 ptr = kasan_reset_tag(ptr);
4104
Kees Cooked18adc2016-06-23 15:24:05 -07004105 /* Find object and usable object size. */
4106 s = page->slab_cache;
Kees Cooked18adc2016-06-23 15:24:05 -07004107
4108 /* Reject impossible pointers. */
4109 if (ptr < page_address(page))
Kees Cookf4e6e282018-01-10 14:48:22 -08004110 usercopy_abort("SLUB object not in SLUB page?!", NULL,
4111 to_user, 0, n);
Kees Cooked18adc2016-06-23 15:24:05 -07004112
4113 /* Find offset within object. */
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004114 if (is_kfence)
4115 offset = ptr - kfence_object_start(ptr);
4116 else
4117 offset = (ptr - page_address(page)) % s->size;
Kees Cooked18adc2016-06-23 15:24:05 -07004118
4119 /* Adjust for redzone and reject if within the redzone. */
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004120 if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) {
Kees Cooked18adc2016-06-23 15:24:05 -07004121 if (offset < s->red_left_pad)
Kees Cookf4e6e282018-01-10 14:48:22 -08004122 usercopy_abort("SLUB object in left red zone",
4123 s->name, to_user, offset, n);
Kees Cooked18adc2016-06-23 15:24:05 -07004124 offset -= s->red_left_pad;
4125 }
4126
Kees Cookafcc90f82018-01-10 15:17:01 -08004127 /* Allow address range falling entirely within usercopy region. */
4128 if (offset >= s->useroffset &&
4129 offset - s->useroffset <= s->usersize &&
4130 n <= s->useroffset - offset + s->usersize)
Kees Cookf4e6e282018-01-10 14:48:22 -08004131 return;
Kees Cooked18adc2016-06-23 15:24:05 -07004132
Kees Cookafcc90f82018-01-10 15:17:01 -08004133 /*
4134 * If the copy is still within the allocated object, produce
4135 * a warning instead of rejecting the copy. This is intended
4136 * to be a temporary method to find any missing usercopy
4137 * whitelists.
4138 */
4139 object_size = slab_ksize(s);
Kees Cook2d891fb2017-11-30 13:04:32 -08004140 if (usercopy_fallback &&
4141 offset <= object_size && n <= object_size - offset) {
Kees Cookafcc90f82018-01-10 15:17:01 -08004142 usercopy_warn("SLUB object", s->name, to_user, offset, n);
4143 return;
4144 }
4145
Kees Cookf4e6e282018-01-10 14:48:22 -08004146 usercopy_abort("SLUB object", s->name, to_user, offset, n);
Kees Cooked18adc2016-06-23 15:24:05 -07004147}
4148#endif /* CONFIG_HARDENED_USERCOPY */
4149
Marco Elver10d1f8c2019-07-11 20:54:14 -07004150size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07004151{
Christoph Lameter272c1d22007-06-08 13:46:49 -07004152 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07004153
Christoph Lameteref8b4522007-10-16 01:24:46 -07004154 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07004155 return 0;
4156
Vegard Nossum294a80a2007-12-04 23:45:30 -08004157 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08004158
Pekka Enberg76994412008-05-22 19:22:25 +03004159 if (unlikely(!PageSlab(page))) {
4160 WARN_ON(!PageCompound(page));
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07004161 return page_size(page);
Pekka Enberg76994412008-05-22 19:22:25 +03004162 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004163
Glauber Costa1b4f59e32012-10-22 18:05:36 +04004164 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07004165}
Marco Elver10d1f8c2019-07-11 20:54:14 -07004166EXPORT_SYMBOL(__ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07004167
4168void kfree(const void *x)
4169{
Christoph Lameter81819f02007-05-06 14:49:36 -07004170 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08004171 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07004172
Pekka Enberg2121db72009-03-25 11:05:57 +02004173 trace_kfree(_RET_IP_, x);
4174
Satyam Sharma2408c552007-10-16 01:24:44 -07004175 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07004176 return;
4177
Christoph Lameterb49af682007-05-06 14:49:41 -07004178 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004179 if (unlikely(!PageSlab(page))) {
Vlastimil Babka6a486c02019-10-06 17:58:42 -07004180 unsigned int order = compound_order(page);
4181
Christoph Lameter09375022008-05-28 10:32:22 -07004182 BUG_ON(!PageCompound(page));
Dmitry Vyukov47adccc2018-02-06 15:36:23 -08004183 kfree_hook(object);
Muchun Songcbb86d62021-02-24 12:04:26 -08004184 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
4185 -(PAGE_SIZE << order));
Vlastimil Babka6a486c02019-10-06 17:58:42 -07004186 __free_pages(page, order);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004187 return;
4188 }
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08004189 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07004190}
4191EXPORT_SYMBOL(kfree);
4192
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004193#define SHRINK_PROMOTE_MAX 32
4194
Christoph Lameter2086d262007-05-06 14:49:46 -07004195/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004196 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
4197 * up most to the head of the partial lists. New allocations will then
4198 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07004199 *
4200 * The slabs with the least items are placed last. This results in them
4201 * being allocated from last increasing the chance that the last objects
4202 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07004203 */
Tejun Heoc9fc5862017-02-22 15:41:27 -08004204int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07004205{
4206 int node;
4207 int i;
4208 struct kmem_cache_node *n;
4209 struct page *page;
4210 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004211 struct list_head discard;
4212 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07004213 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08004214 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07004215
Christoph Lameter2086d262007-05-06 14:49:46 -07004216 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004217 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004218 INIT_LIST_HEAD(&discard);
4219 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
4220 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07004221
4222 spin_lock_irqsave(&n->list_lock, flags);
4223
4224 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004225 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07004226 *
Christoph Lameter672bba32007-05-09 02:32:39 -07004227 * Note that concurrent frees may occur while we hold the
4228 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07004229 */
Tobin C. Harding916ac052019-05-13 17:16:12 -07004230 list_for_each_entry_safe(page, t, &n->partial, slab_list) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004231 int free = page->objects - page->inuse;
4232
4233 /* Do not reread page->inuse */
4234 barrier();
4235
4236 /* We do not keep full slabs on the list */
4237 BUG_ON(free <= 0);
4238
4239 if (free == page->objects) {
Tobin C. Harding916ac052019-05-13 17:16:12 -07004240 list_move(&page->slab_list, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004241 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004242 } else if (free <= SHRINK_PROMOTE_MAX)
Tobin C. Harding916ac052019-05-13 17:16:12 -07004243 list_move(&page->slab_list, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07004244 }
4245
Christoph Lameter2086d262007-05-06 14:49:46 -07004246 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004247 * Promote the slabs filled up most to the head of the
4248 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07004249 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004250 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
4251 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07004252
Christoph Lameter2086d262007-05-06 14:49:46 -07004253 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004254
4255 /* Release empty slabs */
Tobin C. Harding916ac052019-05-13 17:16:12 -07004256 list_for_each_entry_safe(page, t, &discard, slab_list)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05004257 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08004258
4259 if (slabs_node(s, node))
4260 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07004261 }
4262
Vladimir Davydovce3712d2015-02-12 14:59:44 -08004263 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07004264}
Christoph Lameter2086d262007-05-06 14:49:46 -07004265
Yasunori Gotob9049e22007-10-21 16:41:37 -07004266static int slab_mem_going_offline_callback(void *arg)
4267{
4268 struct kmem_cache *s;
4269
Christoph Lameter18004c52012-07-06 15:25:12 -05004270 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004271 list_for_each_entry(s, &slab_caches, list)
Tejun Heoc9fc5862017-02-22 15:41:27 -08004272 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05004273 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004274
4275 return 0;
4276}
4277
4278static void slab_mem_offline_callback(void *arg)
4279{
4280 struct kmem_cache_node *n;
4281 struct kmem_cache *s;
4282 struct memory_notify *marg = arg;
4283 int offline_node;
4284
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004285 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004286
4287 /*
4288 * If the node still has available memory. we need kmem_cache_node
4289 * for it yet.
4290 */
4291 if (offline_node < 0)
4292 return;
4293
Christoph Lameter18004c52012-07-06 15:25:12 -05004294 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004295 list_for_each_entry(s, &slab_caches, list) {
4296 n = get_node(s, offline_node);
4297 if (n) {
4298 /*
4299 * if n->nr_slabs > 0, slabs still exist on the node
4300 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05004301 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07004302 * callback. So, we must fail.
4303 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03004304 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07004305
4306 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05004307 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004308 }
4309 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004310 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004311}
4312
4313static int slab_mem_going_online_callback(void *arg)
4314{
4315 struct kmem_cache_node *n;
4316 struct kmem_cache *s;
4317 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004318 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004319 int ret = 0;
4320
4321 /*
4322 * If the node's memory is already available, then kmem_cache_node is
4323 * already created. Nothing to do.
4324 */
4325 if (nid < 0)
4326 return 0;
4327
4328 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07004329 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07004330 * allocate a kmem_cache_node structure in order to bring the node
4331 * online.
4332 */
Christoph Lameter18004c52012-07-06 15:25:12 -05004333 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004334 list_for_each_entry(s, &slab_caches, list) {
4335 /*
4336 * XXX: kmem_cache_alloc_node will fallback to other nodes
4337 * since memory is not yet available from the node that
4338 * is brought up.
4339 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05004340 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004341 if (!n) {
4342 ret = -ENOMEM;
4343 goto out;
4344 }
Joonsoo Kim40534972012-05-11 00:50:47 +09004345 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004346 s->node[nid] = n;
4347 }
4348out:
Christoph Lameter18004c52012-07-06 15:25:12 -05004349 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004350 return ret;
4351}
4352
4353static int slab_memory_callback(struct notifier_block *self,
4354 unsigned long action, void *arg)
4355{
4356 int ret = 0;
4357
4358 switch (action) {
4359 case MEM_GOING_ONLINE:
4360 ret = slab_mem_going_online_callback(arg);
4361 break;
4362 case MEM_GOING_OFFLINE:
4363 ret = slab_mem_going_offline_callback(arg);
4364 break;
4365 case MEM_OFFLINE:
4366 case MEM_CANCEL_ONLINE:
4367 slab_mem_offline_callback(arg);
4368 break;
4369 case MEM_ONLINE:
4370 case MEM_CANCEL_OFFLINE:
4371 break;
4372 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08004373 if (ret)
4374 ret = notifier_from_errno(ret);
4375 else
4376 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004377 return ret;
4378}
4379
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004380static struct notifier_block slab_memory_callback_nb = {
4381 .notifier_call = slab_memory_callback,
4382 .priority = SLAB_CALLBACK_PRI,
4383};
Yasunori Gotob9049e22007-10-21 16:41:37 -07004384
Christoph Lameter81819f02007-05-06 14:49:36 -07004385/********************************************************************
4386 * Basic setup of slabs
4387 *******************************************************************/
4388
Christoph Lameter51df1142010-08-20 12:37:15 -05004389/*
4390 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004391 * the page allocator. Allocate them properly then fix up the pointers
4392 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05004393 */
4394
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004395static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05004396{
4397 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004398 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004399 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05004400
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004401 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05004402
Glauber Costa7d557b32013-02-22 20:20:00 +04004403 /*
4404 * This runs very early, and only the boot processor is supposed to be
4405 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4406 * IPIs around.
4407 */
4408 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004409 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05004410 struct page *p;
4411
Tobin C. Harding916ac052019-05-13 17:16:12 -07004412 list_for_each_entry(p, &n->partial, slab_list)
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004413 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004414
Li Zefan607bf322011-04-12 15:22:26 +08004415#ifdef CONFIG_SLUB_DEBUG
Tobin C. Harding916ac052019-05-13 17:16:12 -07004416 list_for_each_entry(p, &n->full, slab_list)
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004417 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004418#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05004419 }
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004420 list_add(&s->list, &slab_caches);
4421 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004422}
4423
Christoph Lameter81819f02007-05-06 14:49:36 -07004424void __init kmem_cache_init(void)
4425{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004426 static __initdata struct kmem_cache boot_kmem_cache,
4427 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05004428
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08004429 if (debug_guardpage_minorder())
4430 slub_max_order = 0;
4431
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004432 kmem_cache_node = &boot_kmem_cache_node;
4433 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05004434
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004435 create_boot_cache(kmem_cache_node, "kmem_cache_node",
David Windsor8eb82842017-06-10 22:50:28 -04004436 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004437
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004438 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07004439
4440 /* Able to allocate the per node structures */
4441 slab_state = PARTIAL;
4442
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004443 create_boot_cache(kmem_cache, "kmem_cache",
4444 offsetof(struct kmem_cache, node) +
4445 nr_node_ids * sizeof(struct kmem_cache_node *),
David Windsor8eb82842017-06-10 22:50:28 -04004446 SLAB_HWCACHE_ALIGN, 0, 0);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004447
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004448 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004449 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05004450
4451 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07004452 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004453 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07004454
Thomas Garnier210e7a42016-07-26 15:21:59 -07004455 /* Setup random freelists for each cache */
4456 init_freelist_randomization();
4457
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02004458 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4459 slub_cpu_dead);
Christoph Lameter81819f02007-05-06 14:49:36 -07004460
Alexey Dobriyanb9726c22019-03-05 15:48:26 -08004461 pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004462 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07004463 slub_min_order, slub_max_order, slub_min_objects,
4464 nr_cpu_ids, nr_node_ids);
4465}
4466
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004467void __init kmem_cache_init_late(void)
4468{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004469}
4470
Glauber Costa2633d7a2012-12-18 14:22:34 -08004471struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -07004472__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08004473 slab_flags_t flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07004474{
Roman Gushchin10befea2020-08-06 23:21:27 -07004475 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004476
Vladimir Davydova44cb942014-04-07 15:39:23 -07004477 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004478 if (s) {
4479 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004480
Christoph Lameter81819f02007-05-06 14:49:36 -07004481 /*
4482 * Adjust the object sizes so that we clear
4483 * the complete object on kzalloc.
4484 */
Alexey Dobriyan1b473f22018-04-05 16:21:17 -07004485 s->object_size = max(s->object_size, size);
Alexey Dobriyan52ee6d72018-04-05 16:21:06 -07004486 s->inuse = max(s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08004487
David Rientjes7b8f3b62008-12-17 22:09:46 -08004488 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08004489 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00004490 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08004491 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07004492 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08004493
Christoph Lametercbb79692012-09-05 00:18:32 +00004494 return s;
4495}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004496
Alexey Dobriyand50112e2017-11-15 17:32:18 -08004497int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00004498{
Pekka Enbergaac3a162012-09-05 12:07:44 +03004499 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004500
Pekka Enbergaac3a162012-09-05 12:07:44 +03004501 err = kmem_cache_open(s, flags);
4502 if (err)
4503 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004504
Christoph Lameter45530c42012-11-28 16:23:07 +00004505 /* Mutex is not taken during early boot */
4506 if (slab_state <= UP)
4507 return 0;
4508
Pekka Enbergaac3a162012-09-05 12:07:44 +03004509 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004510 if (err)
Dmitry Safonov52b4b952016-02-17 13:11:37 -08004511 __kmem_cache_release(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004512
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05304513 if (s->flags & SLAB_STORE_USER)
4514 debugfs_slab_add(s);
4515
Pekka Enbergaac3a162012-09-05 12:07:44 +03004516 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004517}
Christoph Lameter81819f02007-05-06 14:49:36 -07004518
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004519void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004520{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004521 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004522 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004523
Christoph Lameter95a05b42013-01-10 19:14:19 +00004524 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004525 return kmalloc_large(size, gfpflags);
4526
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004527 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004528
Satyam Sharma2408c552007-10-16 01:24:44 -07004529 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004530 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004531
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004532 ret = slab_alloc(s, gfpflags, caller, size);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004533
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004534 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004535 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004536
4537 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004538}
Daniel Vetterfd7cb572020-03-23 15:49:00 +01004539EXPORT_SYMBOL(__kmalloc_track_caller);
Christoph Lameter81819f02007-05-06 14:49:36 -07004540
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004541#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004542void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004543 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004544{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004545 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004546 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004547
Christoph Lameter95a05b42013-01-10 19:14:19 +00004548 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004549 ret = kmalloc_large_node(size, gfpflags, node);
4550
4551 trace_kmalloc_node(caller, ret,
4552 size, PAGE_SIZE << get_order(size),
4553 gfpflags, node);
4554
4555 return ret;
4556 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004557
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004558 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004559
Satyam Sharma2408c552007-10-16 01:24:44 -07004560 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004561 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004562
Alexander Potapenko2019e66b2021-01-22 09:18:39 +00004563 ret = slab_alloc_node(s, gfpflags, node, caller, size);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004564
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004565 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004566 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004567
4568 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004569}
Daniel Vetterfd7cb572020-03-23 15:49:00 +01004570EXPORT_SYMBOL(__kmalloc_node_track_caller);
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004571#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004572
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004573#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004574static int count_inuse(struct page *page)
4575{
4576 return page->inuse;
4577}
4578
4579static int count_total(struct page *page)
4580{
4581 return page->objects;
4582}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004583#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004584
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004585#ifdef CONFIG_SLUB_DEBUG
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004586static void validate_slab(struct kmem_cache *s, struct page *page)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004587{
4588 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004589 void *addr = page_address(page);
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004590 unsigned long *map;
4591
4592 slab_lock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004593
Yu Zhaodd98afd2019-11-30 17:49:37 -08004594 if (!check_slab(s, page) || !on_freelist(s, page, NULL))
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004595 goto unlock;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004596
4597 /* Now we know that a valid freelist exists */
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004598 map = get_map(s, page);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004599 for_each_object(p, s, addr, page->objects) {
Roman Gushchin4138fdf2020-08-06 23:20:42 -07004600 u8 val = test_bit(__obj_to_index(s, addr, p), map) ?
Yu Zhaodd98afd2019-11-30 17:49:37 -08004601 SLUB_RED_INACTIVE : SLUB_RED_ACTIVE;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004602
Yu Zhaodd98afd2019-11-30 17:49:37 -08004603 if (!check_object(s, page, p, val))
4604 break;
4605 }
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004606 put_map(map);
4607unlock:
Christoph Lameter881db7f2011-06-01 12:25:53 -05004608 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004609}
4610
Christoph Lameter434e2452007-07-17 04:03:30 -07004611static int validate_slab_node(struct kmem_cache *s,
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004612 struct kmem_cache_node *n)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004613{
4614 unsigned long count = 0;
4615 struct page *page;
4616 unsigned long flags;
4617
4618 spin_lock_irqsave(&n->list_lock, flags);
4619
Tobin C. Harding916ac052019-05-13 17:16:12 -07004620 list_for_each_entry(page, &n->partial, slab_list) {
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004621 validate_slab(s, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004622 count++;
4623 }
4624 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004625 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4626 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004627
4628 if (!(s->flags & SLAB_STORE_USER))
4629 goto out;
4630
Tobin C. Harding916ac052019-05-13 17:16:12 -07004631 list_for_each_entry(page, &n->full, slab_list) {
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004632 validate_slab(s, page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004633 count++;
4634 }
4635 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004636 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4637 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004638
4639out:
4640 spin_unlock_irqrestore(&n->list_lock, flags);
4641 return count;
4642}
4643
Christoph Lameter434e2452007-07-17 04:03:30 -07004644static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004645{
4646 int node;
4647 unsigned long count = 0;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004648 struct kmem_cache_node *n;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004649
4650 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004651 for_each_kmem_cache_node(s, node, n)
Yu Zhao90e9f6a2020-01-30 22:11:57 -08004652 count += validate_slab_node(s, n);
4653
Christoph Lameter53e15af2007-05-06 14:49:43 -07004654 return count;
4655}
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05304656
4657#ifdef CONFIG_DEBUG_FS
Christoph Lameter88a420e2007-05-06 14:49:45 -07004658/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004659 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004660 * and freed.
4661 */
4662
4663struct location {
4664 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004665 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004666 long long sum_time;
4667 long min_time;
4668 long max_time;
4669 long min_pid;
4670 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304671 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004672 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004673};
4674
4675struct loc_track {
4676 unsigned long max;
4677 unsigned long count;
4678 struct location *loc;
4679};
4680
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05304681static struct dentry *slab_debugfs_root;
4682
Christoph Lameter88a420e2007-05-06 14:49:45 -07004683static void free_loc_track(struct loc_track *t)
4684{
4685 if (t->max)
4686 free_pages((unsigned long)t->loc,
4687 get_order(sizeof(struct location) * t->max));
4688}
4689
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004690static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004691{
4692 struct location *l;
4693 int order;
4694
Christoph Lameter88a420e2007-05-06 14:49:45 -07004695 order = get_order(sizeof(struct location) * max);
4696
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004697 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004698 if (!l)
4699 return 0;
4700
4701 if (t->count) {
4702 memcpy(l, t->loc, sizeof(struct location) * t->count);
4703 free_loc_track(t);
4704 }
4705 t->max = max;
4706 t->loc = l;
4707 return 1;
4708}
4709
4710static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004711 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004712{
4713 long start, end, pos;
4714 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004715 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004716 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004717
4718 start = -1;
4719 end = t->count;
4720
4721 for ( ; ; ) {
4722 pos = start + (end - start + 1) / 2;
4723
4724 /*
4725 * There is nothing at "end". If we end up there
4726 * we need to add something to before end.
4727 */
4728 if (pos == end)
4729 break;
4730
4731 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004732 if (track->addr == caddr) {
4733
4734 l = &t->loc[pos];
4735 l->count++;
4736 if (track->when) {
4737 l->sum_time += age;
4738 if (age < l->min_time)
4739 l->min_time = age;
4740 if (age > l->max_time)
4741 l->max_time = age;
4742
4743 if (track->pid < l->min_pid)
4744 l->min_pid = track->pid;
4745 if (track->pid > l->max_pid)
4746 l->max_pid = track->pid;
4747
Rusty Russell174596a2009-01-01 10:12:29 +10304748 cpumask_set_cpu(track->cpu,
4749 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004750 }
4751 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004752 return 1;
4753 }
4754
Christoph Lameter45edfa52007-05-09 02:32:45 -07004755 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004756 end = pos;
4757 else
4758 start = pos;
4759 }
4760
4761 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004762 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004763 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004764 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004765 return 0;
4766
4767 l = t->loc + pos;
4768 if (pos < t->count)
4769 memmove(l + 1, l,
4770 (t->count - pos) * sizeof(struct location));
4771 t->count++;
4772 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004773 l->addr = track->addr;
4774 l->sum_time = age;
4775 l->min_time = age;
4776 l->max_time = age;
4777 l->min_pid = track->pid;
4778 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304779 cpumask_clear(to_cpumask(l->cpus));
4780 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004781 nodes_clear(l->nodes);
4782 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004783 return 1;
4784}
4785
4786static void process_slab(struct loc_track *t, struct kmem_cache *s,
Vlastimil Babka791f85d2021-05-23 01:28:37 +02004787 struct page *page, enum track_item alloc,
4788 unsigned long *obj_map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004789{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004790 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004791 void *p;
4792
Vlastimil Babka791f85d2021-05-23 01:28:37 +02004793 __fill_map(obj_map, s, page);
4794
Christoph Lameter224a88b2008-04-14 19:11:31 +03004795 for_each_object(p, s, addr, page->objects)
Vlastimil Babka791f85d2021-05-23 01:28:37 +02004796 if (!test_bit(__obj_to_index(s, addr, p), obj_map))
Christoph Lameter45edfa52007-05-09 02:32:45 -07004797 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004798}
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05304799#endif /* CONFIG_DEBUG_FS */
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07004800#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter88a420e2007-05-06 14:49:45 -07004801
Christoph Lametera5a84752010-10-05 13:57:27 -05004802#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004803static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004804{
4805 u8 *p;
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004806 int type = KMALLOC_NORMAL;
Christoph Lametera5a84752010-10-05 13:57:27 -05004807
Christoph Lameter95a05b42013-01-10 19:14:19 +00004808 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004809
Fabian Frederickf9f58282014-06-04 16:06:34 -07004810 pr_err("SLUB resiliency testing\n");
4811 pr_err("-----------------------\n");
4812 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004813
4814 p = kzalloc(16, GFP_KERNEL);
4815 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004816 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4817 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004818
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004819 validate_slab_cache(kmalloc_caches[type][4]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004820
4821 /* Hmmm... The next two are dangerous */
4822 p = kzalloc(32, GFP_KERNEL);
4823 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004824 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4825 p);
4826 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004827
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004828 validate_slab_cache(kmalloc_caches[type][5]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004829 p = kzalloc(64, GFP_KERNEL);
4830 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4831 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004832 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4833 p);
4834 pr_err("If allocated object is overwritten then not detectable\n\n");
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004835 validate_slab_cache(kmalloc_caches[type][6]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004836
Fabian Frederickf9f58282014-06-04 16:06:34 -07004837 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004838 p = kzalloc(128, GFP_KERNEL);
4839 kfree(p);
4840 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004841 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004842 validate_slab_cache(kmalloc_caches[type][7]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004843
4844 p = kzalloc(256, GFP_KERNEL);
4845 kfree(p);
4846 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004847 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004848 validate_slab_cache(kmalloc_caches[type][8]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004849
4850 p = kzalloc(512, GFP_KERNEL);
4851 kfree(p);
4852 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004853 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07004854 validate_slab_cache(kmalloc_caches[type][9]);
Christoph Lametera5a84752010-10-05 13:57:27 -05004855}
4856#else
4857#ifdef CONFIG_SYSFS
4858static void resiliency_test(void) {};
4859#endif
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07004860#endif /* SLUB_RESILIENCY_TEST */
Christoph Lametera5a84752010-10-05 13:57:27 -05004861
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004862#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004863enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004864 SL_ALL, /* All slabs */
4865 SL_PARTIAL, /* Only partially allocated slabs */
4866 SL_CPU, /* Only slabs used for cpu caches */
4867 SL_OBJECTS, /* Determine allocated objects not slabs */
4868 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004869};
4870
Christoph Lameter205ab992008-04-14 19:11:40 +03004871#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004872#define SO_PARTIAL (1 << SL_PARTIAL)
4873#define SO_CPU (1 << SL_CPU)
4874#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004875#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004876
Tejun Heo1663f262017-02-22 15:41:39 -08004877#ifdef CONFIG_MEMCG
4878static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON);
4879
4880static int __init setup_slub_memcg_sysfs(char *str)
4881{
4882 int v;
4883
4884 if (get_option(&str, &v) > 0)
4885 memcg_sysfs_enabled = v;
4886
4887 return 1;
4888}
4889
4890__setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs);
4891#endif
4892
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004893static ssize_t show_slab_objects(struct kmem_cache *s,
4894 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004895{
4896 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004897 int node;
4898 int x;
4899 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004900
Kees Cook6396bb22018-06-12 14:03:40 -07004901 nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004902 if (!nodes)
4903 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004904
Christoph Lameter205ab992008-04-14 19:11:40 +03004905 if (flags & SO_CPU) {
4906 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004907
Christoph Lameter205ab992008-04-14 19:11:40 +03004908 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004909 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4910 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004911 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004912 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004913
Jason Low4db0c3c2015-04-15 16:14:08 -07004914 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004915 if (!page)
4916 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004917
Christoph Lameterec3ab082012-05-09 10:09:56 -05004918 node = page_to_nid(page);
4919 if (flags & SO_TOTAL)
4920 x = page->objects;
4921 else if (flags & SO_OBJECTS)
4922 x = page->inuse;
4923 else
4924 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004925
Christoph Lameterec3ab082012-05-09 10:09:56 -05004926 total += x;
4927 nodes[node] += x;
4928
Wei Yanga93cf072017-07-06 15:36:31 -07004929 page = slub_percpu_partial_read_once(c);
Christoph Lameter49e22582011-08-09 16:12:27 -05004930 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004931 node = page_to_nid(page);
4932 if (flags & SO_TOTAL)
4933 WARN_ON_ONCE(1);
4934 else if (flags & SO_OBJECTS)
4935 WARN_ON_ONCE(1);
4936 else
4937 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004938 total += x;
4939 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004940 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004941 }
4942 }
4943
Qian Caie4f8e512019-10-14 14:11:51 -07004944 /*
4945 * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex"
4946 * already held which will conflict with an existing lock order:
4947 *
4948 * mem_hotplug_lock->slab_mutex->kernfs_mutex
4949 *
4950 * We don't really need mem_hotplug_lock (to hold off
4951 * slab_mem_going_offline_callback) here because slab's memory hot
4952 * unplug code doesn't destroy the kmem_cache->node[] data.
4953 */
4954
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004955#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004956 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004957 struct kmem_cache_node *n;
4958
4959 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004960
Chen Gangd0e0ac92013-07-15 09:05:29 +08004961 if (flags & SO_TOTAL)
4962 x = atomic_long_read(&n->total_objects);
4963 else if (flags & SO_OBJECTS)
4964 x = atomic_long_read(&n->total_objects) -
4965 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004966 else
4967 x = atomic_long_read(&n->nr_slabs);
4968 total += x;
4969 nodes[node] += x;
4970 }
4971
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004972 } else
4973#endif
4974 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004975 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004976
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004977 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004978 if (flags & SO_TOTAL)
4979 x = count_partial(n, count_total);
4980 else if (flags & SO_OBJECTS)
4981 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004982 else
4983 x = n->nr_partial;
4984 total += x;
4985 nodes[node] += x;
4986 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004987 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004988 x = sprintf(buf, "%lu", total);
4989#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004990 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004991 if (nodes[node])
4992 x += sprintf(buf + x, " N%d=%lu",
4993 node, nodes[node]);
4994#endif
4995 kfree(nodes);
4996 return x + sprintf(buf + x, "\n");
4997}
4998
Christoph Lameter81819f02007-05-06 14:49:36 -07004999#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03005000#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07005001
5002struct slab_attribute {
5003 struct attribute attr;
5004 ssize_t (*show)(struct kmem_cache *s, char *buf);
5005 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
5006};
5007
5008#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04005009 static struct slab_attribute _name##_attr = \
5010 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07005011
5012#define SLAB_ATTR(_name) \
5013 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04005014 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07005015
Christoph Lameter81819f02007-05-06 14:49:36 -07005016static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
5017{
Alexey Dobriyan44065b22018-04-05 16:21:20 -07005018 return sprintf(buf, "%u\n", s->size);
Christoph Lameter81819f02007-05-06 14:49:36 -07005019}
5020SLAB_ATTR_RO(slab_size);
5021
5022static ssize_t align_show(struct kmem_cache *s, char *buf)
5023{
Alexey Dobriyan3a3791e2018-04-05 16:21:02 -07005024 return sprintf(buf, "%u\n", s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07005025}
5026SLAB_ATTR_RO(align);
5027
5028static ssize_t object_size_show(struct kmem_cache *s, char *buf)
5029{
Alexey Dobriyan1b473f22018-04-05 16:21:17 -07005030 return sprintf(buf, "%u\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07005031}
5032SLAB_ATTR_RO(object_size);
5033
5034static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
5035{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07005036 return sprintf(buf, "%u\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07005037}
5038SLAB_ATTR_RO(objs_per_slab);
5039
5040static ssize_t order_show(struct kmem_cache *s, char *buf)
5041{
Alexey Dobriyan19af27a2018-04-05 16:21:39 -07005042 return sprintf(buf, "%u\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07005043}
Vlastimil Babka32a6f402020-08-06 23:18:41 -07005044SLAB_ATTR_RO(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07005045
David Rientjes73d342b2009-02-22 17:40:09 -08005046static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
5047{
5048 return sprintf(buf, "%lu\n", s->min_partial);
5049}
5050
5051static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
5052 size_t length)
5053{
5054 unsigned long min;
5055 int err;
5056
Jingoo Han3dbb95f2013-09-11 14:20:25 -07005057 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08005058 if (err)
5059 return err;
5060
David Rientjesc0bdb232009-02-25 09:16:35 +02005061 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08005062 return length;
5063}
5064SLAB_ATTR(min_partial);
5065
Christoph Lameter49e22582011-08-09 16:12:27 -05005066static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
5067{
Wei Yange6d0e1d2017-07-06 15:36:34 -07005068 return sprintf(buf, "%u\n", slub_cpu_partial(s));
Christoph Lameter49e22582011-08-09 16:12:27 -05005069}
5070
5071static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
5072 size_t length)
5073{
Alexey Dobriyane5d99982018-04-05 16:21:10 -07005074 unsigned int objects;
Christoph Lameter49e22582011-08-09 16:12:27 -05005075 int err;
5076
Alexey Dobriyane5d99982018-04-05 16:21:10 -07005077 err = kstrtouint(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05005078 if (err)
5079 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09005080 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08005081 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05005082
Wei Yange6d0e1d2017-07-06 15:36:34 -07005083 slub_set_cpu_partial(s, objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05005084 flush_all(s);
5085 return length;
5086}
5087SLAB_ATTR(cpu_partial);
5088
Christoph Lameter81819f02007-05-06 14:49:36 -07005089static ssize_t ctor_show(struct kmem_cache *s, char *buf)
5090{
Joe Perches62c70bc2011-01-13 15:45:52 -08005091 if (!s->ctor)
5092 return 0;
5093 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07005094}
5095SLAB_ATTR_RO(ctor);
5096
Christoph Lameter81819f02007-05-06 14:49:36 -07005097static ssize_t aliases_show(struct kmem_cache *s, char *buf)
5098{
Gu Zheng4307c142014-08-06 16:04:51 -07005099 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005100}
5101SLAB_ATTR_RO(aliases);
5102
Christoph Lameter81819f02007-05-06 14:49:36 -07005103static ssize_t partial_show(struct kmem_cache *s, char *buf)
5104{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08005105 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07005106}
5107SLAB_ATTR_RO(partial);
5108
5109static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
5110{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08005111 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07005112}
5113SLAB_ATTR_RO(cpu_slabs);
5114
5115static ssize_t objects_show(struct kmem_cache *s, char *buf)
5116{
Christoph Lameter205ab992008-04-14 19:11:40 +03005117 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07005118}
5119SLAB_ATTR_RO(objects);
5120
Christoph Lameter205ab992008-04-14 19:11:40 +03005121static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
5122{
5123 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
5124}
5125SLAB_ATTR_RO(objects_partial);
5126
Christoph Lameter49e22582011-08-09 16:12:27 -05005127static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
5128{
5129 int objects = 0;
5130 int pages = 0;
5131 int cpu;
5132 int len;
5133
5134 for_each_online_cpu(cpu) {
Wei Yanga93cf072017-07-06 15:36:31 -07005135 struct page *page;
5136
5137 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
Christoph Lameter49e22582011-08-09 16:12:27 -05005138
5139 if (page) {
5140 pages += page->pages;
5141 objects += page->pobjects;
5142 }
5143 }
5144
5145 len = sprintf(buf, "%d(%d)", objects, pages);
5146
5147#ifdef CONFIG_SMP
5148 for_each_online_cpu(cpu) {
Wei Yanga93cf072017-07-06 15:36:31 -07005149 struct page *page;
5150
5151 page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu));
Christoph Lameter49e22582011-08-09 16:12:27 -05005152
5153 if (page && len < PAGE_SIZE - 20)
5154 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
5155 page->pobjects, page->pages);
5156 }
5157#endif
5158 return len + sprintf(buf + len, "\n");
5159}
5160SLAB_ATTR_RO(slabs_cpu_partial);
5161
Christoph Lameter81819f02007-05-06 14:49:36 -07005162static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
5163{
5164 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
5165}
Vlastimil Babka8f58119a2020-08-06 23:18:48 -07005166SLAB_ATTR_RO(reclaim_account);
Christoph Lameter81819f02007-05-06 14:49:36 -07005167
5168static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
5169{
Christoph Lameter5af60832007-05-06 14:49:56 -07005170 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07005171}
5172SLAB_ATTR_RO(hwcache_align);
5173
5174#ifdef CONFIG_ZONE_DMA
5175static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
5176{
5177 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
5178}
5179SLAB_ATTR_RO(cache_dma);
5180#endif
5181
David Windsor8eb82842017-06-10 22:50:28 -04005182static ssize_t usersize_show(struct kmem_cache *s, char *buf)
5183{
Alexey Dobriyan7bbdb812018-04-05 16:21:31 -07005184 return sprintf(buf, "%u\n", s->usersize);
David Windsor8eb82842017-06-10 22:50:28 -04005185}
5186SLAB_ATTR_RO(usersize);
5187
Christoph Lameter81819f02007-05-06 14:49:36 -07005188static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
5189{
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08005190 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU));
Christoph Lameter81819f02007-05-06 14:49:36 -07005191}
5192SLAB_ATTR_RO(destroy_by_rcu);
5193
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005194#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005195static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5196{
5197 return show_slab_objects(s, buf, SO_ALL);
5198}
5199SLAB_ATTR_RO(slabs);
5200
5201static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5202{
5203 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5204}
5205SLAB_ATTR_RO(total_objects);
5206
5207static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5208{
Laura Abbottbecfda62016-03-15 14:55:06 -07005209 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
Christoph Lametera5a84752010-10-05 13:57:27 -05005210}
Vlastimil Babka060807f2020-08-06 23:18:45 -07005211SLAB_ATTR_RO(sanity_checks);
Christoph Lametera5a84752010-10-05 13:57:27 -05005212
5213static ssize_t trace_show(struct kmem_cache *s, char *buf)
5214{
5215 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5216}
Vlastimil Babka060807f2020-08-06 23:18:45 -07005217SLAB_ATTR_RO(trace);
Christoph Lametera5a84752010-10-05 13:57:27 -05005218
Christoph Lameter81819f02007-05-06 14:49:36 -07005219static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5220{
5221 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5222}
5223
Vlastimil Babkaad38b5b2020-08-06 23:18:38 -07005224SLAB_ATTR_RO(red_zone);
Christoph Lameter81819f02007-05-06 14:49:36 -07005225
5226static ssize_t poison_show(struct kmem_cache *s, char *buf)
5227{
5228 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5229}
5230
Vlastimil Babkaad38b5b2020-08-06 23:18:38 -07005231SLAB_ATTR_RO(poison);
Christoph Lameter81819f02007-05-06 14:49:36 -07005232
5233static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5234{
5235 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5236}
5237
Vlastimil Babkaad38b5b2020-08-06 23:18:38 -07005238SLAB_ATTR_RO(store_user);
Christoph Lameter81819f02007-05-06 14:49:36 -07005239
Christoph Lameter53e15af2007-05-06 14:49:43 -07005240static ssize_t validate_show(struct kmem_cache *s, char *buf)
5241{
5242 return 0;
5243}
5244
5245static ssize_t validate_store(struct kmem_cache *s,
5246 const char *buf, size_t length)
5247{
Christoph Lameter434e2452007-07-17 04:03:30 -07005248 int ret = -EINVAL;
5249
5250 if (buf[0] == '1') {
5251 ret = validate_slab_cache(s);
5252 if (ret >= 0)
5253 ret = length;
5254 }
5255 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07005256}
5257SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05005258
Christoph Lametera5a84752010-10-05 13:57:27 -05005259#endif /* CONFIG_SLUB_DEBUG */
5260
5261#ifdef CONFIG_FAILSLAB
5262static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5263{
5264 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5265}
Vlastimil Babka060807f2020-08-06 23:18:45 -07005266SLAB_ATTR_RO(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005267#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07005268
Christoph Lameter2086d262007-05-06 14:49:46 -07005269static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5270{
5271 return 0;
5272}
5273
5274static ssize_t shrink_store(struct kmem_cache *s,
5275 const char *buf, size_t length)
5276{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005277 if (buf[0] == '1')
Roman Gushchin10befea2020-08-06 23:21:27 -07005278 kmem_cache_shrink(s);
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005279 else
Christoph Lameter2086d262007-05-06 14:49:46 -07005280 return -EINVAL;
5281 return length;
5282}
5283SLAB_ATTR(shrink);
5284
Christoph Lameter81819f02007-05-06 14:49:36 -07005285#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005286static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07005287{
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005288 return sprintf(buf, "%u\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07005289}
5290
Christoph Lameter98246012008-01-07 23:20:26 -08005291static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07005292 const char *buf, size_t length)
5293{
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005294 unsigned int ratio;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005295 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005296
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005297 err = kstrtouint(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07005298 if (err)
5299 return err;
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005300 if (ratio > 100)
5301 return -ERANGE;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005302
Alexey Dobriyaneb7235e2018-04-05 16:20:48 -07005303 s->remote_node_defrag_ratio = ratio * 10;
Christoph Lameter0121c6192008-04-29 16:11:12 -07005304
Christoph Lameter81819f02007-05-06 14:49:36 -07005305 return length;
5306}
Christoph Lameter98246012008-01-07 23:20:26 -08005307SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005308#endif
5309
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005310#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005311static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5312{
5313 unsigned long sum = 0;
5314 int cpu;
5315 int len;
Kees Cook6da2ec52018-06-12 13:55:00 -07005316 int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005317
5318 if (!data)
5319 return -ENOMEM;
5320
5321 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005322 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005323
5324 data[cpu] = x;
5325 sum += x;
5326 }
5327
5328 len = sprintf(buf, "%lu", sum);
5329
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005330#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005331 for_each_online_cpu(cpu) {
5332 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005333 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005334 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005335#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005336 kfree(data);
5337 return len + sprintf(buf + len, "\n");
5338}
5339
David Rientjes78eb00c2009-10-15 02:20:22 -07005340static void clear_stat(struct kmem_cache *s, enum stat_item si)
5341{
5342 int cpu;
5343
5344 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005345 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005346}
5347
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005348#define STAT_ATTR(si, text) \
5349static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5350{ \
5351 return show_stat(s, buf, si); \
5352} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005353static ssize_t text##_store(struct kmem_cache *s, \
5354 const char *buf, size_t length) \
5355{ \
5356 if (buf[0] != '0') \
5357 return -EINVAL; \
5358 clear_stat(s, si); \
5359 return length; \
5360} \
5361SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005362
5363STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5364STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5365STAT_ATTR(FREE_FASTPATH, free_fastpath);
5366STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5367STAT_ATTR(FREE_FROZEN, free_frozen);
5368STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5369STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5370STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5371STAT_ATTR(ALLOC_SLAB, alloc_slab);
5372STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005373STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005374STAT_ATTR(FREE_SLAB, free_slab);
5375STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5376STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5377STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5378STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5379STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5380STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005381STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005382STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005383STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5384STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005385STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5386STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005387STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5388STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Tobin C. Harding6dfd1b62019-05-13 17:16:09 -07005389#endif /* CONFIG_SLUB_STATS */
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005390
Pekka Enberg06428782008-01-07 23:20:27 -08005391static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005392 &slab_size_attr.attr,
5393 &object_size_attr.attr,
5394 &objs_per_slab_attr.attr,
5395 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005396 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005397 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005398 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005399 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005400 &partial_attr.attr,
5401 &cpu_slabs_attr.attr,
5402 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005403 &aliases_attr.attr,
5404 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005405 &hwcache_align_attr.attr,
5406 &reclaim_account_attr.attr,
5407 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005408 &shrink_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005409 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005410#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005411 &total_objects_attr.attr,
5412 &slabs_attr.attr,
5413 &sanity_checks_attr.attr,
5414 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005415 &red_zone_attr.attr,
5416 &poison_attr.attr,
5417 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005418 &validate_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005419#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005420#ifdef CONFIG_ZONE_DMA
5421 &cache_dma_attr.attr,
5422#endif
5423#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005424 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005425#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005426#ifdef CONFIG_SLUB_STATS
5427 &alloc_fastpath_attr.attr,
5428 &alloc_slowpath_attr.attr,
5429 &free_fastpath_attr.attr,
5430 &free_slowpath_attr.attr,
5431 &free_frozen_attr.attr,
5432 &free_add_partial_attr.attr,
5433 &free_remove_partial_attr.attr,
5434 &alloc_from_partial_attr.attr,
5435 &alloc_slab_attr.attr,
5436 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005437 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005438 &free_slab_attr.attr,
5439 &cpuslab_flush_attr.attr,
5440 &deactivate_full_attr.attr,
5441 &deactivate_empty_attr.attr,
5442 &deactivate_to_head_attr.attr,
5443 &deactivate_to_tail_attr.attr,
5444 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005445 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005446 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005447 &cmpxchg_double_fail_attr.attr,
5448 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005449 &cpu_partial_alloc_attr.attr,
5450 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005451 &cpu_partial_node_attr.attr,
5452 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005453#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005454#ifdef CONFIG_FAILSLAB
5455 &failslab_attr.attr,
5456#endif
David Windsor8eb82842017-06-10 22:50:28 -04005457 &usersize_attr.attr,
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005458
Christoph Lameter81819f02007-05-06 14:49:36 -07005459 NULL
5460};
5461
Arvind Yadav1fdaaa22017-09-06 16:21:56 -07005462static const struct attribute_group slab_attr_group = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005463 .attrs = slab_attrs,
5464};
5465
5466static ssize_t slab_attr_show(struct kobject *kobj,
5467 struct attribute *attr,
5468 char *buf)
5469{
5470 struct slab_attribute *attribute;
5471 struct kmem_cache *s;
5472 int err;
5473
5474 attribute = to_slab_attr(attr);
5475 s = to_slab(kobj);
5476
5477 if (!attribute->show)
5478 return -EIO;
5479
5480 err = attribute->show(s, buf);
5481
5482 return err;
5483}
5484
5485static ssize_t slab_attr_store(struct kobject *kobj,
5486 struct attribute *attr,
5487 const char *buf, size_t len)
5488{
5489 struct slab_attribute *attribute;
5490 struct kmem_cache *s;
5491 int err;
5492
5493 attribute = to_slab_attr(attr);
5494 s = to_slab(kobj);
5495
5496 if (!attribute->store)
5497 return -EIO;
5498
5499 err = attribute->store(s, buf, len);
Christoph Lameter81819f02007-05-06 14:49:36 -07005500 return err;
5501}
5502
Christoph Lameter41a21282014-05-06 12:50:08 -07005503static void kmem_cache_release(struct kobject *k)
5504{
5505 slab_kmem_cache_release(to_slab(k));
5506}
5507
Emese Revfy52cf25d2010-01-19 02:58:23 +01005508static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005509 .show = slab_attr_show,
5510 .store = slab_attr_store,
5511};
5512
5513static struct kobj_type slab_ktype = {
5514 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005515 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005516};
5517
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005518static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005519
Vladimir Davydov9a417072014-04-07 15:39:31 -07005520static inline struct kset *cache_kset(struct kmem_cache *s)
5521{
Vladimir Davydov9a417072014-04-07 15:39:31 -07005522 return slab_kset;
5523}
5524
Christoph Lameter81819f02007-05-06 14:49:36 -07005525#define ID_STR_LENGTH 64
5526
5527/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005528 *
5529 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005530 */
5531static char *create_unique_id(struct kmem_cache *s)
5532{
5533 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5534 char *p = name;
5535
5536 BUG_ON(!name);
5537
5538 *p++ = ':';
5539 /*
5540 * First flags affecting slabcache operations. We will only
5541 * get here for aliasable slabs so we do not need to support
5542 * too many flags. The flags here must cover all flags that
5543 * are matched during merging to guarantee that the id is
5544 * unique.
5545 */
5546 if (s->flags & SLAB_CACHE_DMA)
5547 *p++ = 'd';
Nicolas Boichat6d6ea1e2019-03-28 20:43:42 -07005548 if (s->flags & SLAB_CACHE_DMA32)
5549 *p++ = 'D';
Christoph Lameter81819f02007-05-06 14:49:36 -07005550 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5551 *p++ = 'a';
Laura Abbottbecfda62016-03-15 14:55:06 -07005552 if (s->flags & SLAB_CONSISTENCY_CHECKS)
Christoph Lameter81819f02007-05-06 14:49:36 -07005553 *p++ = 'F';
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08005554 if (s->flags & SLAB_ACCOUNT)
5555 *p++ = 'A';
Christoph Lameter81819f02007-05-06 14:49:36 -07005556 if (p != name + 1)
5557 *p++ = '-';
Alexey Dobriyan44065b22018-04-05 16:21:20 -07005558 p += sprintf(p, "%07u", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005559
Christoph Lameter81819f02007-05-06 14:49:36 -07005560 BUG_ON(p > name + ID_STR_LENGTH - 1);
5561 return name;
5562}
5563
5564static int sysfs_slab_add(struct kmem_cache *s)
5565{
5566 int err;
5567 const char *name;
Tejun Heo1663f262017-02-22 15:41:39 -08005568 struct kset *kset = cache_kset(s);
Christoph Lameter45530c42012-11-28 16:23:07 +00005569 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005570
Tejun Heo1663f262017-02-22 15:41:39 -08005571 if (!kset) {
5572 kobject_init(&s->kobj, &slab_ktype);
5573 return 0;
5574 }
5575
Miles Chen11066382017-11-15 17:32:25 -08005576 if (!unmergeable && disable_higher_order_debug &&
5577 (slub_debug & DEBUG_METADATA_FLAGS))
5578 unmergeable = 1;
5579
Christoph Lameter81819f02007-05-06 14:49:36 -07005580 if (unmergeable) {
5581 /*
5582 * Slabcache can never be merged so we can use the name proper.
5583 * This is typically the case for debug situations. In that
5584 * case we can catch duplicate names easily.
5585 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005586 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005587 name = s->name;
5588 } else {
5589 /*
5590 * Create a unique name for the slab as a target
5591 * for the symlinks.
5592 */
5593 name = create_unique_id(s);
5594 }
5595
Tejun Heo1663f262017-02-22 15:41:39 -08005596 s->kobj.kset = kset;
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005597 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Wang Haibf5eb7d2021-01-28 19:32:50 +08005598 if (err)
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005599 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005600
5601 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005602 if (err)
5603 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005604
Christoph Lameter81819f02007-05-06 14:49:36 -07005605 if (!unmergeable) {
5606 /* Setup first alias */
5607 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005608 }
Dave Jones54b6a732014-04-07 15:39:32 -07005609out:
5610 if (!unmergeable)
5611 kfree(name);
5612 return err;
5613out_del_kobj:
5614 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005615 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005616}
5617
Mikulas Patockad50d82f2018-06-27 23:26:09 -07005618void sysfs_slab_unlink(struct kmem_cache *s)
5619{
5620 if (slab_state >= FULL)
5621 kobject_del(&s->kobj);
5622}
5623
Tejun Heobf5eb3d2017-02-22 15:41:11 -08005624void sysfs_slab_release(struct kmem_cache *s)
5625{
5626 if (slab_state >= FULL)
5627 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005628}
5629
5630/*
5631 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005632 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005633 */
5634struct saved_alias {
5635 struct kmem_cache *s;
5636 const char *name;
5637 struct saved_alias *next;
5638};
5639
Adrian Bunk5af328a2007-07-17 04:03:27 -07005640static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005641
5642static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5643{
5644 struct saved_alias *al;
5645
Christoph Lameter97d06602012-07-06 15:25:11 -05005646 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005647 /*
5648 * If we have a leftover link then remove it.
5649 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005650 sysfs_remove_link(&slab_kset->kobj, name);
5651 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005652 }
5653
5654 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5655 if (!al)
5656 return -ENOMEM;
5657
5658 al->s = s;
5659 al->name = name;
5660 al->next = alias_list;
5661 alias_list = al;
5662 return 0;
5663}
5664
5665static int __init slab_sysfs_init(void)
5666{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005667 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005668 int err;
5669
Christoph Lameter18004c52012-07-06 15:25:12 -05005670 mutex_lock(&slab_mutex);
Christoph Lameter2bce64852010-07-19 11:39:11 -05005671
Christoph Lameterd7660ce2020-06-01 21:45:50 -07005672 slab_kset = kset_create_and_add("slab", NULL, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005673 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005674 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005675 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005676 return -ENOSYS;
5677 }
5678
Christoph Lameter97d06602012-07-06 15:25:11 -05005679 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005680
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005681 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005682 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005683 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005684 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5685 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005686 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005687
5688 while (alias_list) {
5689 struct saved_alias *al = alias_list;
5690
5691 alias_list = alias_list->next;
5692 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005693 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005694 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5695 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005696 kfree(al);
5697 }
5698
Christoph Lameter18004c52012-07-06 15:25:12 -05005699 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005700 resiliency_test();
5701 return 0;
5702}
5703
5704__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005705#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005706
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305707#if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS)
5708static int slab_debugfs_show(struct seq_file *seq, void *v)
5709{
5710
5711 struct location *l;
5712 unsigned int idx = *(unsigned int *)v;
5713 struct loc_track *t = seq->private;
5714
5715 if (idx < t->count) {
5716 l = &t->loc[idx];
5717
5718 seq_printf(seq, "%7ld ", l->count);
5719
5720 if (l->addr)
5721 seq_printf(seq, "%pS", (void *)l->addr);
5722 else
5723 seq_puts(seq, "<not-available>");
5724
5725 if (l->sum_time != l->min_time) {
5726 seq_printf(seq, " age=%ld/%llu/%ld",
5727 l->min_time, div_u64(l->sum_time, l->count),
5728 l->max_time);
5729 } else
5730 seq_printf(seq, " age=%ld", l->min_time);
5731
5732 if (l->min_pid != l->max_pid)
5733 seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid);
5734 else
5735 seq_printf(seq, " pid=%ld",
5736 l->min_pid);
5737
5738 if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus)))
5739 seq_printf(seq, " cpus=%*pbl",
5740 cpumask_pr_args(to_cpumask(l->cpus)));
5741
5742 if (nr_online_nodes > 1 && !nodes_empty(l->nodes))
5743 seq_printf(seq, " nodes=%*pbl",
5744 nodemask_pr_args(&l->nodes));
5745
5746 seq_puts(seq, "\n");
5747 }
5748
5749 if (!idx && !t->count)
5750 seq_puts(seq, "No data\n");
5751
5752 return 0;
5753}
5754
5755static void slab_debugfs_stop(struct seq_file *seq, void *v)
5756{
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305757}
5758
5759static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos)
5760{
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305761 struct loc_track *t = seq->private;
5762
Faiyaz Mohammedb007e432021-06-23 10:50:06 +05305763 v = ppos;
5764 ++*ppos;
5765 if (*ppos <= t->count)
5766 return v;
5767
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305768 return NULL;
5769}
5770
5771static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos)
5772{
Faiyaz Mohammedb007e432021-06-23 10:50:06 +05305773 return ppos;
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305774}
5775
5776static const struct seq_operations slab_debugfs_sops = {
5777 .start = slab_debugfs_start,
5778 .next = slab_debugfs_next,
5779 .stop = slab_debugfs_stop,
5780 .show = slab_debugfs_show,
5781};
5782
5783static int slab_debug_trace_open(struct inode *inode, struct file *filep)
5784{
5785
5786 struct kmem_cache_node *n;
5787 enum track_item alloc;
5788 int node;
5789 struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops,
5790 sizeof(struct loc_track));
5791 struct kmem_cache *s = file_inode(filep)->i_private;
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005792 unsigned long *obj_map;
5793
Miaohe Lin6b6725f2021-10-18 15:15:52 -07005794 if (!t)
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005795 return -ENOMEM;
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305796
Miaohe Lin6b6725f2021-10-18 15:15:52 -07005797 obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL);
5798 if (!obj_map) {
5799 seq_release_private(inode, filep);
5800 return -ENOMEM;
5801 }
5802
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305803 if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0)
5804 alloc = TRACK_ALLOC;
5805 else
5806 alloc = TRACK_FREE;
5807
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005808 if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) {
5809 bitmap_free(obj_map);
Miaohe Lin6b6725f2021-10-18 15:15:52 -07005810 seq_release_private(inode, filep);
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305811 return -ENOMEM;
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005812 }
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305813
5814 /* Push back cpu slabs */
5815 flush_all(s);
5816
5817 for_each_kmem_cache_node(s, node, n) {
5818 unsigned long flags;
5819 struct page *page;
5820
5821 if (!atomic_long_read(&n->nr_slabs))
5822 continue;
5823
5824 spin_lock_irqsave(&n->list_lock, flags);
5825 list_for_each_entry(page, &n->partial, slab_list)
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005826 process_slab(t, s, page, alloc, obj_map);
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305827 list_for_each_entry(page, &n->full, slab_list)
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005828 process_slab(t, s, page, alloc, obj_map);
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305829 spin_unlock_irqrestore(&n->list_lock, flags);
5830 }
5831
Vlastimil Babka791f85d2021-05-23 01:28:37 +02005832 bitmap_free(obj_map);
Faiyaz Mohammed54e74122021-06-15 09:19:37 +05305833 return 0;
5834}
5835
5836static int slab_debug_trace_release(struct inode *inode, struct file *file)
5837{
5838 struct seq_file *seq = file->private_data;
5839 struct loc_track *t = seq->private;
5840
5841 free_loc_track(t);
5842 return seq_release_private(inode, file);
5843}
5844
5845static const struct file_operations slab_debugfs_fops = {
5846 .open = slab_debug_trace_open,
5847 .read = seq_read,
5848 .llseek = seq_lseek,
5849 .release = slab_debug_trace_release,
5850};
5851
5852static void debugfs_slab_add(struct kmem_cache *s)
5853{
5854 struct dentry *slab_cache_dir;
5855
5856 if (unlikely(!slab_debugfs_root))
5857 return;
5858
5859 slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root);
5860
5861 debugfs_create_file("alloc_traces", 0400,
5862 slab_cache_dir, s, &slab_debugfs_fops);
5863
5864 debugfs_create_file("free_traces", 0400,
5865 slab_cache_dir, s, &slab_debugfs_fops);
5866}
5867
5868void debugfs_slab_release(struct kmem_cache *s)
5869{
5870 debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root));
5871}
5872
5873static int __init slab_debugfs_init(void)
5874{
5875 struct kmem_cache *s;
5876
5877 slab_debugfs_root = debugfs_create_dir("slab", NULL);
5878
5879 list_for_each_entry(s, &slab_caches, list)
5880 if (s->flags & SLAB_STORE_USER)
5881 debugfs_slab_add(s);
5882
5883 return 0;
5884
5885}
5886__initcall(slab_debugfs_init);
5887#endif
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005888/*
5889 * The /proc/slabinfo ABI
5890 */
Yang Shi5b365772017-11-15 17:32:03 -08005891#ifdef CONFIG_SLUB_DEBUG
Glauber Costa0d7561c2012-10-19 18:20:27 +04005892void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005893{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005894 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005895 unsigned long nr_objs = 0;
5896 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005897 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005898 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005899
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005900 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005901 nr_slabs += node_nr_slabs(n);
5902 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005903 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005904 }
5905
Glauber Costa0d7561c2012-10-19 18:20:27 +04005906 sinfo->active_objs = nr_objs - nr_free;
5907 sinfo->num_objs = nr_objs;
5908 sinfo->active_slabs = nr_slabs;
5909 sinfo->num_slabs = nr_slabs;
5910 sinfo->objects_per_slab = oo_objects(s->oo);
5911 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005912}
Vijayanand Jittadf2e5752020-12-29 16:07:29 +05305913EXPORT_SYMBOL_GPL(get_slabinfo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005914
Glauber Costa0d7561c2012-10-19 18:20:27 +04005915void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005916{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005917}
5918
Glauber Costab7454ad2012-10-19 18:20:25 +04005919ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5920 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005921{
Glauber Costab7454ad2012-10-19 18:20:25 +04005922 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005923}
Yang Shi5b365772017-11-15 17:32:03 -08005924#endif /* CONFIG_SLUB_DEBUG */