blob: fa53bb09495d65312df7cbab2d46e121a74d58cc [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/mm/slab.c
4 * Written by Mark Hemment, 1996/97.
5 * (markhe@nextd.demon.co.uk)
6 *
7 * kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
8 *
9 * Major cleanup, different bufctl logic, per-cpu arrays
10 * (c) 2000 Manfred Spraul
11 *
12 * Cleanup, make the head arrays unconditional, preparation for NUMA
13 * (c) 2002 Manfred Spraul
14 *
15 * An implementation of the Slab Allocator as described in outline in;
16 * UNIX Internals: The New Frontiers by Uresh Vahalia
17 * Pub: Prentice Hall ISBN 0-13-101908-2
18 * or with a little more detail in;
19 * The Slab Allocator: An Object-Caching Kernel Memory Allocator
20 * Jeff Bonwick (Sun Microsystems).
21 * Presented at: USENIX Summer 1994 Technical Conference
22 *
23 * The memory is organized in caches, one cache for each object type.
24 * (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
25 * Each cache consists out of many slabs (they are small (usually one
26 * page long) and always contiguous), and each slab contains multiple
27 * initialized objects.
28 *
29 * This means, that your constructor is used only for newly allocated
Simon Arlott183ff222007-10-20 01:27:18 +020030 * slabs and you must pass objects with the same initializations to
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 * kmem_cache_free.
32 *
33 * Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
34 * normal). If you need a special memory type, then must create a new
35 * cache for that memory type.
36 *
37 * In order to reduce fragmentation, the slabs are sorted in 3 groups:
38 * full slabs with 0 free objects
39 * partial slabs
40 * empty slabs with no allocated objects
41 *
42 * If partial slabs exist, then new allocations come from these slabs,
43 * otherwise from empty slabs or new slabs are allocated.
44 *
45 * kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
46 * during kmem_cache_destroy(). The caller must prevent concurrent allocs.
47 *
48 * Each cache has a short per-cpu head array, most allocs
49 * and frees go into that array, and if that array overflows, then 1/2
50 * of the entries in the array are given back into the global cache.
51 * The head array is strictly LIFO and should improve the cache hit rates.
52 * On SMP, it additionally reduces the spinlock operations.
53 *
Andrew Mortona737b3e2006-03-22 00:08:11 -080054 * The c_cpuarray may not be read with enabled local interrupts -
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * it's changed with a smp_call_function().
56 *
57 * SMP synchronization:
58 * constructors and destructors are called without any locking.
Pekka Enberg343e0d72006-02-01 03:05:50 -080059 * Several members in struct kmem_cache and struct slab never change, they
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * are accessed without any locking.
61 * The per-cpu arrays are never accessed from the wrong cpu, no locking,
62 * and local interrupts are disabled so slab code is preempt-safe.
63 * The non-constant members are protected with a per-cache irq spinlock.
64 *
65 * Many thanks to Mark Hemment, who wrote another per-cpu slab patch
66 * in 2000 - many ideas in the current implementation are derived from
67 * his patch.
68 *
69 * Further notes from the original documentation:
70 *
71 * 11 April '97. Started multi-threading - markhe
Christoph Lameter18004c52012-07-06 15:25:12 -050072 * The global cache-chain is protected by the mutex 'slab_mutex'.
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * The sem is only needed when accessing/extending the cache-chain, which
74 * can never happen inside an interrupt (kmem_cache_create(),
75 * kmem_cache_shrink() and kmem_cache_reap()).
76 *
77 * At present, each engine can be growing a cache. This should be blocked.
78 *
Christoph Lametere498be72005-09-09 13:03:32 -070079 * 15 March 2005. NUMA slab allocator.
80 * Shai Fultheim <shai@scalex86.org>.
81 * Shobhit Dayal <shobhit@calsoftinc.com>
82 * Alok N Kataria <alokk@calsoftinc.com>
83 * Christoph Lameter <christoph@lameter.com>
84 *
85 * Modified the slab allocator to be node aware on NUMA systems.
86 * Each node has its own list of partial, free and full slabs.
87 * All object allocations for a node occur from node specific slab lists.
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 */
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090#include <linux/slab.h>
91#include <linux/mm.h>
Randy Dunlapc9cf5522006-06-27 02:53:52 -070092#include <linux/poison.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070093#include <linux/swap.h>
94#include <linux/cache.h>
95#include <linux/interrupt.h>
96#include <linux/init.h>
97#include <linux/compiler.h>
Paul Jackson101a5002006-03-24 03:16:07 -080098#include <linux/cpuset.h>
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +040099#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100#include <linux/seq_file.h>
101#include <linux/notifier.h>
102#include <linux/kallsyms.h>
103#include <linux/cpu.h>
104#include <linux/sysctl.h>
105#include <linux/module.h>
106#include <linux/rcupdate.h>
Paulo Marques543537b2005-06-23 00:09:02 -0700107#include <linux/string.h>
Andrew Morton138ae662006-12-06 20:36:41 -0800108#include <linux/uaccess.h>
Christoph Lametere498be72005-09-09 13:03:32 -0700109#include <linux/nodemask.h>
Catalin Marinasd5cff632009-06-11 13:22:40 +0100110#include <linux/kmemleak.h>
Christoph Lameterdc85da12006-01-18 17:42:36 -0800111#include <linux/mempolicy.h>
Ingo Molnarfc0abb12006-01-18 17:42:33 -0800112#include <linux/mutex.h>
Akinobu Mita8a8b6502006-12-08 02:39:44 -0800113#include <linux/fault-inject.h>
Ingo Molnare7eebaf2006-06-27 02:54:55 -0700114#include <linux/rtmutex.h>
Eric Dumazet6a2d7a92006-12-13 00:34:27 -0800115#include <linux/reciprocal_div.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700116#include <linux/debugobjects.h>
David Rientjes8f9f8d92010-03-27 19:40:47 -0700117#include <linux/memory.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -0700118#include <linux/prefetch.h>
Ingo Molnar3f8c2452017-02-05 14:31:22 +0100119#include <linux/sched/task_stack.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
Mel Gorman381760e2012-07-31 16:44:30 -0700121#include <net/sock.h>
122
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123#include <asm/cacheflush.h>
124#include <asm/tlbflush.h>
125#include <asm/page.h>
126
Steven Rostedt4dee6b62012-01-09 17:15:42 -0500127#include <trace/events/kmem.h>
128
Mel Gorman072bb0a2012-07-31 16:43:58 -0700129#include "internal.h"
130
Glauber Costab9ce5ef2012-12-18 14:22:46 -0800131#include "slab.h"
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133/*
Christoph Lameter50953fe2007-05-06 14:50:16 -0700134 * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 * 0 for faster, smaller code (especially in the critical paths).
136 *
137 * STATS - 1 to collect stats for /proc/slabinfo.
138 * 0 for faster, smaller code (especially in the critical paths).
139 *
140 * FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
141 */
142
143#ifdef CONFIG_DEBUG_SLAB
144#define DEBUG 1
145#define STATS 1
146#define FORCED_DEBUG 1
147#else
148#define DEBUG 0
149#define STATS 0
150#define FORCED_DEBUG 0
151#endif
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/* Shouldn't this be in a header file somewhere? */
154#define BYTES_PER_WORD sizeof(void *)
David Woodhouse87a927c2007-07-04 21:26:44 -0400155#define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157#ifndef ARCH_KMALLOC_FLAGS
158#define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
159#endif
160
Joonsoo Kimf315e3f2013-12-02 17:49:41 +0900161#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
162 <= SLAB_OBJ_MIN_SIZE) ? 1 : 0)
163
164#if FREELIST_BYTE_INDEX
165typedef unsigned char freelist_idx_t;
166#else
167typedef unsigned short freelist_idx_t;
168#endif
169
David Miller30321c72014-05-05 16:20:04 -0400170#define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
Joonsoo Kimf315e3f2013-12-02 17:49:41 +0900171
Mel Gorman072bb0a2012-07-31 16:43:58 -0700172/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * struct array_cache
174 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 * Purpose:
176 * - LIFO ordering, to hand out cache-warm objects from _alloc
177 * - reduce the number of linked list operations
178 * - reduce spinlock operations
179 *
180 * The limit is stored in the per-cpu structure to reduce the data cache
181 * footprint.
182 *
183 */
184struct array_cache {
185 unsigned int avail;
186 unsigned int limit;
187 unsigned int batchcount;
188 unsigned int touched;
Robert P. J. Daybda5b652007-10-16 23:30:05 -0700189 void *entry[]; /*
Andrew Mortona737b3e2006-03-22 00:08:11 -0800190 * Must have this definition in here for the proper
191 * alignment of array_cache. Also simplifies accessing
192 * the entries.
Andrew Mortona737b3e2006-03-22 00:08:11 -0800193 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194};
195
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700196struct alien_cache {
197 spinlock_t lock;
198 struct array_cache ac;
199};
200
Andrew Mortona737b3e2006-03-22 00:08:11 -0800201/*
Christoph Lametere498be72005-09-09 13:03:32 -0700202 * Need this for bootstrapping a per node allocator.
203 */
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700204#define NUM_INIT_LISTS (2 * MAX_NUMNODES)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000205static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
Christoph Lametere498be72005-09-09 13:03:32 -0700206#define CACHE_CACHE 0
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700207#define SIZE_NODE (MAX_NUMNODES)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Christoph Lametered11d9e2006-06-30 01:55:45 -0700209static int drain_freelist(struct kmem_cache *cache,
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000210 struct kmem_cache_node *n, int tofree);
Christoph Lametered11d9e2006-06-30 01:55:45 -0700211static void free_block(struct kmem_cache *cachep, void **objpp, int len,
Joonsoo Kim97654df2014-08-06 16:04:25 -0700212 int node, struct list_head *list);
213static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300214static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
David Howells65f27f32006-11-22 14:55:48 +0000215static void cache_reap(struct work_struct *unused);
Christoph Lametered11d9e2006-06-30 01:55:45 -0700216
Joonsoo Kim76b342b2016-05-19 17:10:26 -0700217static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
218 void **list);
219static inline void fixup_slab_list(struct kmem_cache *cachep,
220 struct kmem_cache_node *n, struct page *page,
221 void **list);
Ingo Molnare0a42722006-06-23 02:03:46 -0700222static int slab_early_init = 1;
223
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000224#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
Christoph Lametere498be72005-09-09 13:03:32 -0700225
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000226static void kmem_cache_node_init(struct kmem_cache_node *parent)
Christoph Lametere498be72005-09-09 13:03:32 -0700227{
228 INIT_LIST_HEAD(&parent->slabs_full);
229 INIT_LIST_HEAD(&parent->slabs_partial);
230 INIT_LIST_HEAD(&parent->slabs_free);
David Rientjesbf00bd32016-12-12 16:41:44 -0800231 parent->total_slabs = 0;
Greg Thelenf728b0a2016-12-12 16:41:41 -0800232 parent->free_slabs = 0;
Christoph Lametere498be72005-09-09 13:03:32 -0700233 parent->shared = NULL;
234 parent->alien = NULL;
Ravikiran G Thirumalai2e1217c2006-02-04 23:27:56 -0800235 parent->colour_next = 0;
Christoph Lametere498be72005-09-09 13:03:32 -0700236 spin_lock_init(&parent->list_lock);
237 parent->free_objects = 0;
238 parent->free_touched = 0;
239}
240
Andrew Mortona737b3e2006-03-22 00:08:11 -0800241#define MAKE_LIST(cachep, listp, slab, nodeid) \
242 do { \
243 INIT_LIST_HEAD(listp); \
Christoph Lameter18bf8542014-08-06 16:04:11 -0700244 list_splice(&get_node(cachep, nodeid)->slab, listp); \
Christoph Lametere498be72005-09-09 13:03:32 -0700245 } while (0)
246
Andrew Mortona737b3e2006-03-22 00:08:11 -0800247#define MAKE_ALL_LISTS(cachep, ptr, nodeid) \
248 do { \
Christoph Lametere498be72005-09-09 13:03:32 -0700249 MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \
250 MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
251 MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \
252 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Alexey Dobriyan4fd0b462017-11-15 17:32:21 -0800254#define CFLGS_OBJFREELIST_SLAB ((slab_flags_t __force)0x40000000U)
255#define CFLGS_OFF_SLAB ((slab_flags_t __force)0x80000000U)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -0700256#define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257#define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB)
258
259#define BATCHREFILL_LIMIT 16
Andrew Mortona737b3e2006-03-22 00:08:11 -0800260/*
261 * Optimization question: fewer reaps means less probability for unnessary
262 * cpucache drain/refill cycles.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 *
Adrian Bunkdc6f3f22005-11-08 16:44:08 +0100264 * OTOH the cpuarrays can contain lots of objects,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 * which could lock up otherwise freeable slabs.
266 */
Jianyu Zhan5f0985b2014-03-30 17:02:20 +0800267#define REAPTIMEOUT_AC (2*HZ)
268#define REAPTIMEOUT_NODE (4*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270#if STATS
271#define STATS_INC_ACTIVE(x) ((x)->num_active++)
272#define STATS_DEC_ACTIVE(x) ((x)->num_active--)
273#define STATS_INC_ALLOCED(x) ((x)->num_allocations++)
274#define STATS_INC_GROWN(x) ((x)->grown++)
Christoph Lametered11d9e2006-06-30 01:55:45 -0700275#define STATS_ADD_REAPED(x,y) ((x)->reaped += (y))
Andrew Mortona737b3e2006-03-22 00:08:11 -0800276#define STATS_SET_HIGH(x) \
277 do { \
278 if ((x)->num_active > (x)->high_mark) \
279 (x)->high_mark = (x)->num_active; \
280 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281#define STATS_INC_ERR(x) ((x)->errors++)
282#define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++)
Christoph Lametere498be72005-09-09 13:03:32 -0700283#define STATS_INC_NODEFREES(x) ((x)->node_frees++)
Ravikiran G Thirumalaifb7faf32006-04-10 22:52:54 -0700284#define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++)
Andrew Mortona737b3e2006-03-22 00:08:11 -0800285#define STATS_SET_FREEABLE(x, i) \
286 do { \
287 if ((x)->max_freeable < i) \
288 (x)->max_freeable = i; \
289 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290#define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit)
291#define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss)
292#define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit)
293#define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss)
294#else
295#define STATS_INC_ACTIVE(x) do { } while (0)
296#define STATS_DEC_ACTIVE(x) do { } while (0)
297#define STATS_INC_ALLOCED(x) do { } while (0)
298#define STATS_INC_GROWN(x) do { } while (0)
Andi Kleen4e60c862010-08-09 17:19:03 -0700299#define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300#define STATS_SET_HIGH(x) do { } while (0)
301#define STATS_INC_ERR(x) do { } while (0)
302#define STATS_INC_NODEALLOCS(x) do { } while (0)
Christoph Lametere498be72005-09-09 13:03:32 -0700303#define STATS_INC_NODEFREES(x) do { } while (0)
Ravikiran G Thirumalaifb7faf32006-04-10 22:52:54 -0700304#define STATS_INC_ACOVERFLOW(x) do { } while (0)
Andrew Mortona737b3e2006-03-22 00:08:11 -0800305#define STATS_SET_FREEABLE(x, i) do { } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306#define STATS_INC_ALLOCHIT(x) do { } while (0)
307#define STATS_INC_ALLOCMISS(x) do { } while (0)
308#define STATS_INC_FREEHIT(x) do { } while (0)
309#define STATS_INC_FREEMISS(x) do { } while (0)
310#endif
311
312#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Andrew Mortona737b3e2006-03-22 00:08:11 -0800314/*
315 * memory layout of objects:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * 0 : objp
Manfred Spraul3dafccf2006-02-01 03:05:42 -0800317 * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * the end of an object is aligned with the end of the real
319 * allocation. Catches writes behind the end of the allocation.
Manfred Spraul3dafccf2006-02-01 03:05:42 -0800320 * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * redzone word.
Manfred Spraul3dafccf2006-02-01 03:05:42 -0800322 * cachep->obj_offset: The real object.
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500323 * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
324 * cachep->size - 1* BYTES_PER_WORD: last caller address
Andrew Mortona737b3e2006-03-22 00:08:11 -0800325 * [BYTES_PER_WORD long]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 */
Pekka Enberg343e0d72006-02-01 03:05:50 -0800327static int obj_offset(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
Manfred Spraul3dafccf2006-02-01 03:05:42 -0800329 return cachep->obj_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
David Woodhouseb46b8f12007-05-08 00:22:59 -0700332static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
334 BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
David Woodhouseb46b8f12007-05-08 00:22:59 -0700335 return (unsigned long long*) (objp + obj_offset(cachep) -
336 sizeof(unsigned long long));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
David Woodhouseb46b8f12007-05-08 00:22:59 -0700339static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
341 BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
342 if (cachep->flags & SLAB_STORE_USER)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500343 return (unsigned long long *)(objp + cachep->size -
David Woodhouseb46b8f12007-05-08 00:22:59 -0700344 sizeof(unsigned long long) -
David Woodhouse87a927c2007-07-04 21:26:44 -0400345 REDZONE_ALIGN);
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500346 return (unsigned long long *) (objp + cachep->size -
David Woodhouseb46b8f12007-05-08 00:22:59 -0700347 sizeof(unsigned long long));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
Pekka Enberg343e0d72006-02-01 03:05:50 -0800350static void **dbg_userword(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351{
352 BUG_ON(!(cachep->flags & SLAB_STORE_USER));
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500353 return (void **)(objp + cachep->size - BYTES_PER_WORD);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
356#else
357
Manfred Spraul3dafccf2006-02-01 03:05:42 -0800358#define obj_offset(x) 0
David Woodhouseb46b8f12007-05-08 00:22:59 -0700359#define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
360#define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361#define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;})
362
363#endif
364
Joonsoo Kim03787302014-06-23 13:22:06 -0700365#ifdef CONFIG_DEBUG_SLAB_LEAK
366
Joonsoo Kimd31676d2016-03-15 14:54:24 -0700367static inline bool is_store_user_clean(struct kmem_cache *cachep)
Joonsoo Kim03787302014-06-23 13:22:06 -0700368{
Joonsoo Kimd31676d2016-03-15 14:54:24 -0700369 return atomic_read(&cachep->store_user_clean) == 1;
370}
Joonsoo Kim03787302014-06-23 13:22:06 -0700371
Joonsoo Kimd31676d2016-03-15 14:54:24 -0700372static inline void set_store_user_clean(struct kmem_cache *cachep)
373{
374 atomic_set(&cachep->store_user_clean, 1);
375}
Joonsoo Kim03787302014-06-23 13:22:06 -0700376
Joonsoo Kimd31676d2016-03-15 14:54:24 -0700377static inline void set_store_user_dirty(struct kmem_cache *cachep)
378{
379 if (is_store_user_clean(cachep))
380 atomic_set(&cachep->store_user_clean, 0);
Joonsoo Kim03787302014-06-23 13:22:06 -0700381}
382
383#else
Joonsoo Kimd31676d2016-03-15 14:54:24 -0700384static inline void set_store_user_dirty(struct kmem_cache *cachep) {}
Joonsoo Kim03787302014-06-23 13:22:06 -0700385
386#endif
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388/*
David Rientjes3df1ccc2011-10-18 22:09:28 -0700389 * Do not go above this order unless 0 objects fit into the slab or
390 * overridden on the command line.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 */
David Rientjes543585c2011-10-18 22:09:24 -0700392#define SLAB_MAX_ORDER_HI 1
393#define SLAB_MAX_ORDER_LO 0
394static int slab_max_order = SLAB_MAX_ORDER_LO;
David Rientjes3df1ccc2011-10-18 22:09:28 -0700395static bool slab_max_order_set __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Pekka Enberg6ed5eb2212006-02-01 03:05:49 -0800397static inline struct kmem_cache *virt_to_cache(const void *obj)
398{
Christoph Lameterb49af682007-05-06 14:49:41 -0700399 struct page *page = virt_to_head_page(obj);
Christoph Lameter35026082012-06-13 10:24:56 -0500400 return page->slab_cache;
Pekka Enberg6ed5eb2212006-02-01 03:05:49 -0800401}
402
Joonsoo Kim8456a642013-10-24 10:07:49 +0900403static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800404 unsigned int idx)
405{
Joonsoo Kim8456a642013-10-24 10:07:49 +0900406 return page->s_mem + cache->size * idx;
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800407}
408
Joonsoo Kim6fb92432016-03-15 14:54:09 -0700409#define BOOT_CPUCACHE_ENTRIES 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410/* internal cache of cache description objs */
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000411static struct kmem_cache kmem_cache_boot = {
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800412 .batchcount = 1,
413 .limit = BOOT_CPUCACHE_ENTRIES,
414 .shared = 1,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500415 .size = sizeof(struct kmem_cache),
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800416 .name = "kmem_cache",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417};
418
Tejun Heo1871e522009-10-29 22:34:13 +0900419static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
Pekka Enberg343e0d72006-02-01 03:05:50 -0800421static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700423 return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
Andrew Mortona737b3e2006-03-22 00:08:11 -0800426/*
427 * Calculate the number of objects and left-over bytes for a given buffer size.
428 */
Joonsoo Kim70f75062016-03-15 14:54:53 -0700429static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800430 slab_flags_t flags, size_t *left_over)
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800431{
Joonsoo Kim70f75062016-03-15 14:54:53 -0700432 unsigned int num;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800433 size_t slab_size = PAGE_SIZE << gfporder;
434
435 /*
436 * The slab management structure can be either off the slab or
437 * on it. For the latter case, the memory allocated for a
438 * slab is used for:
439 *
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800440 * - @buffer_size bytes for each object
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700441 * - One freelist_idx_t for each object
442 *
443 * We don't need to consider alignment of freelist because
444 * freelist will be at the end of slab page. The objects will be
445 * at the correct alignment.
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800446 *
447 * If the slab management structure is off the slab, then the
448 * alignment will already be calculated into the size. Because
449 * the slabs are all pages aligned, the objects will be at the
450 * correct alignment when allocated.
451 */
Joonsoo Kimb03a017b2016-03-15 14:54:50 -0700452 if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700453 num = slab_size / buffer_size;
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700454 *left_over = slab_size % buffer_size;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800455 } else {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700456 num = slab_size / (buffer_size + sizeof(freelist_idx_t));
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700457 *left_over = slab_size %
458 (buffer_size + sizeof(freelist_idx_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 }
Joonsoo Kim70f75062016-03-15 14:54:53 -0700460
461 return num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
Christoph Lameterf28510d2012-09-11 19:49:38 +0000464#if DEBUG
Harvey Harrisond40cee22008-04-30 00:55:07 -0700465#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Andrew Mortona737b3e2006-03-22 00:08:11 -0800467static void __slab_error(const char *function, struct kmem_cache *cachep,
468 char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Joe Perches11705322016-03-17 14:19:50 -0700470 pr_err("slab error in %s(): cache `%s': %s\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800471 function, cachep->name, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030473 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
Christoph Lameterf28510d2012-09-11 19:49:38 +0000475#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Paul Menage3395ee02006-12-06 20:32:16 -0800477/*
478 * By default on NUMA we use alien caches to stage the freeing of
479 * objects allocated from other nodes. This causes massive memory
480 * inefficiencies when using fake NUMA setup to split memory into a
481 * large number of small nodes, so it can be disabled on the command
482 * line
483 */
484
485static int use_alien_caches __read_mostly = 1;
486static int __init noaliencache_setup(char *s)
487{
488 use_alien_caches = 0;
489 return 1;
490}
491__setup("noaliencache", noaliencache_setup);
492
David Rientjes3df1ccc2011-10-18 22:09:28 -0700493static int __init slab_max_order_setup(char *str)
494{
495 get_option(&str, &slab_max_order);
496 slab_max_order = slab_max_order < 0 ? 0 :
497 min(slab_max_order, MAX_ORDER - 1);
498 slab_max_order_set = true;
499
500 return 1;
501}
502__setup("slab_max_order=", slab_max_order_setup);
503
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800504#ifdef CONFIG_NUMA
505/*
506 * Special reaping functions for NUMA systems called from cache_reap().
507 * These take care of doing round robin flushing of alien caches (containing
508 * objects freed on different nodes from which they were allocated) and the
509 * flushing of remote pcps by calling drain_node_pages.
510 */
Tejun Heo1871e522009-10-29 22:34:13 +0900511static DEFINE_PER_CPU(unsigned long, slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800512
513static void init_reap_node(int cpu)
514{
Andrew Morton0edaf862016-05-19 17:10:58 -0700515 per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
516 node_online_map);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800517}
518
519static void next_reap_node(void)
520{
Christoph Lameter909ea962010-12-08 16:22:55 +0100521 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800522
Andrew Morton0edaf862016-05-19 17:10:58 -0700523 node = next_node_in(node, node_online_map);
Christoph Lameter909ea962010-12-08 16:22:55 +0100524 __this_cpu_write(slab_reap_node, node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800525}
526
527#else
528#define init_reap_node(cpu) do { } while (0)
529#define next_reap_node(void) do { } while (0)
530#endif
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532/*
533 * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz
534 * via the workqueue/eventd.
535 * Add the CPU number into the expiration time to minimize the possibility of
536 * the CPUs getting into lockstep and contending for the global cache chain
537 * lock.
538 */
Paul Gortmaker0db06282013-06-19 14:53:51 -0400539static void start_cpu_timer(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Tejun Heo1871e522009-10-29 22:34:13 +0900541 struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Tejun Heoeac03372016-09-16 15:49:34 -0400543 if (reap_work->work.func == NULL) {
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800544 init_reap_node(cpu);
Tejun Heo203b42f2012-08-21 13:18:23 -0700545 INIT_DEFERRABLE_WORK(reap_work, cache_reap);
Arjan van de Ven2b284212006-12-10 02:21:28 -0800546 schedule_delayed_work_on(cpu, reap_work,
547 __round_jiffies_relative(HZ, cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 }
549}
550
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700551static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700553 if (ac) {
554 ac->avail = 0;
555 ac->limit = limit;
556 ac->batchcount = batch;
557 ac->touched = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 }
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700559}
560
561static struct array_cache *alloc_arraycache(int node, int entries,
562 int batchcount, gfp_t gfp)
563{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700564 size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700565 struct array_cache *ac = NULL;
566
567 ac = kmalloc_node(memsize, gfp, node);
Qian Caif09c4242019-03-05 15:42:03 -0800568 /*
569 * The array_cache structures contain pointers to free object.
570 * However, when such objects are allocated or transferred to another
571 * cache the pointers are not cleared and they could be counted as
572 * valid references during a kmemleak scan. Therefore, kmemleak must
573 * not scan such objects.
574 */
575 kmemleak_no_scan(ac);
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700576 init_arraycache(ac, entries, batchcount);
577 return ac;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700580static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
581 struct page *page, void *objp)
Mel Gorman072bb0a2012-07-31 16:43:58 -0700582{
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700583 struct kmem_cache_node *n;
584 int page_node;
585 LIST_HEAD(list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700586
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700587 page_node = page_to_nid(page);
588 n = get_node(cachep, page_node);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700589
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700590 spin_lock(&n->list_lock);
591 free_block(cachep, &objp, 1, page_node, &list);
592 spin_unlock(&n->list_lock);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700593
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700594 slabs_destroy(cachep, &list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700595}
596
Christoph Lameter3ded1752006-03-25 03:06:44 -0800597/*
598 * Transfer objects in one arraycache to another.
599 * Locking must be handled by the caller.
600 *
601 * Return the number of entries transferred.
602 */
603static int transfer_objects(struct array_cache *to,
604 struct array_cache *from, unsigned int max)
605{
606 /* Figure out how many entries to transfer */
Hagen Paul Pfeifer732eacc2010-10-26 14:22:23 -0700607 int nr = min3(from->avail, max, to->limit - to->avail);
Christoph Lameter3ded1752006-03-25 03:06:44 -0800608
609 if (!nr)
610 return 0;
611
612 memcpy(to->entry + to->avail, from->entry + from->avail -nr,
613 sizeof(void *) *nr);
614
615 from->avail -= nr;
616 to->avail += nr;
Christoph Lameter3ded1752006-03-25 03:06:44 -0800617 return nr;
618}
619
Christoph Lameter765c4502006-09-27 01:50:08 -0700620#ifndef CONFIG_NUMA
621
622#define drain_alien_cache(cachep, alien) do { } while (0)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000623#define reap_alien(cachep, n) do { } while (0)
Christoph Lameter765c4502006-09-27 01:50:08 -0700624
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700625static inline struct alien_cache **alloc_alien_cache(int node,
626 int limit, gfp_t gfp)
Christoph Lameter765c4502006-09-27 01:50:08 -0700627{
Joonsoo Kim88881772016-05-19 17:10:05 -0700628 return NULL;
Christoph Lameter765c4502006-09-27 01:50:08 -0700629}
630
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700631static inline void free_alien_cache(struct alien_cache **ac_ptr)
Christoph Lameter765c4502006-09-27 01:50:08 -0700632{
633}
634
635static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
636{
637 return 0;
638}
639
640static inline void *alternate_node_alloc(struct kmem_cache *cachep,
641 gfp_t flags)
642{
643 return NULL;
644}
645
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800646static inline void *____cache_alloc_node(struct kmem_cache *cachep,
Christoph Lameter765c4502006-09-27 01:50:08 -0700647 gfp_t flags, int nodeid)
648{
649 return NULL;
650}
651
David Rientjes4167e9b2015-04-14 15:46:55 -0700652static inline gfp_t gfp_exact_node(gfp_t flags)
653{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700654 return flags & ~__GFP_NOFAIL;
David Rientjes4167e9b2015-04-14 15:46:55 -0700655}
656
Christoph Lameter765c4502006-09-27 01:50:08 -0700657#else /* CONFIG_NUMA */
658
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800659static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
Paul Jacksonc61afb12006-03-24 03:16:08 -0800660static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
Christoph Lameterdc85da12006-01-18 17:42:36 -0800661
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700662static struct alien_cache *__alloc_alien_cache(int node, int entries,
663 int batch, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -0700664{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700665 size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700666 struct alien_cache *alc = NULL;
667
668 alc = kmalloc_node(memsize, gfp, node);
Christoph Lameterf928ca32019-01-08 15:23:00 -0800669 if (alc) {
Qian Caif09c4242019-03-05 15:42:03 -0800670 kmemleak_no_scan(alc);
Christoph Lameterf928ca32019-01-08 15:23:00 -0800671 init_arraycache(&alc->ac, entries, batch);
672 spin_lock_init(&alc->lock);
673 }
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700674 return alc;
675}
676
677static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
678{
679 struct alien_cache **alc_ptr;
Joonsoo Kim5e804782014-08-06 16:04:40 -0700680 size_t memsize = sizeof(void *) * nr_node_ids;
Christoph Lametere498be72005-09-09 13:03:32 -0700681 int i;
682
683 if (limit > 1)
684 limit = 12;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700685 alc_ptr = kzalloc_node(memsize, gfp, node);
686 if (!alc_ptr)
687 return NULL;
688
689 for_each_node(i) {
690 if (i == node || !node_online(i))
691 continue;
692 alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
693 if (!alc_ptr[i]) {
694 for (i--; i >= 0; i--)
695 kfree(alc_ptr[i]);
696 kfree(alc_ptr);
697 return NULL;
Christoph Lametere498be72005-09-09 13:03:32 -0700698 }
699 }
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700700 return alc_ptr;
Christoph Lametere498be72005-09-09 13:03:32 -0700701}
702
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700703static void free_alien_cache(struct alien_cache **alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700704{
705 int i;
706
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700707 if (!alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700708 return;
Christoph Lametere498be72005-09-09 13:03:32 -0700709 for_each_node(i)
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700710 kfree(alc_ptr[i]);
711 kfree(alc_ptr);
Christoph Lametere498be72005-09-09 13:03:32 -0700712}
713
Pekka Enberg343e0d72006-02-01 03:05:50 -0800714static void __drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim833b7062014-08-06 16:04:33 -0700715 struct array_cache *ac, int node,
716 struct list_head *list)
Christoph Lametere498be72005-09-09 13:03:32 -0700717{
Christoph Lameter18bf8542014-08-06 16:04:11 -0700718 struct kmem_cache_node *n = get_node(cachep, node);
Christoph Lametere498be72005-09-09 13:03:32 -0700719
720 if (ac->avail) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000721 spin_lock(&n->list_lock);
Christoph Lametere00946f2006-03-25 03:06:45 -0800722 /*
723 * Stuff objects into the remote nodes shared array first.
724 * That way we could avoid the overhead of putting the objects
725 * into the free lists and getting them back later.
726 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000727 if (n->shared)
728 transfer_objects(n->shared, ac, ac->limit);
Christoph Lametere00946f2006-03-25 03:06:45 -0800729
Joonsoo Kim833b7062014-08-06 16:04:33 -0700730 free_block(cachep, ac->entry, ac->avail, node, list);
Christoph Lametere498be72005-09-09 13:03:32 -0700731 ac->avail = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000732 spin_unlock(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -0700733 }
734}
735
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800736/*
737 * Called from cache_reap() to regularly drain alien caches round robin.
738 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000739static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800740{
Christoph Lameter909ea962010-12-08 16:22:55 +0100741 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800742
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000743 if (n->alien) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700744 struct alien_cache *alc = n->alien[node];
745 struct array_cache *ac;
Christoph Lametere00946f2006-03-25 03:06:45 -0800746
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700747 if (alc) {
748 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700749 if (ac->avail && spin_trylock_irq(&alc->lock)) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700750 LIST_HEAD(list);
751
752 __drain_alien_cache(cachep, ac, node, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700753 spin_unlock_irq(&alc->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700754 slabs_destroy(cachep, &list);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700755 }
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800756 }
757 }
758}
759
Andrew Mortona737b3e2006-03-22 00:08:11 -0800760static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700761 struct alien_cache **alien)
Christoph Lametere498be72005-09-09 13:03:32 -0700762{
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800763 int i = 0;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700764 struct alien_cache *alc;
Christoph Lametere498be72005-09-09 13:03:32 -0700765 struct array_cache *ac;
766 unsigned long flags;
767
768 for_each_online_node(i) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700769 alc = alien[i];
770 if (alc) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700771 LIST_HEAD(list);
772
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700773 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700774 spin_lock_irqsave(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700775 __drain_alien_cache(cachep, ac, i, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700776 spin_unlock_irqrestore(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700777 slabs_destroy(cachep, &list);
Christoph Lametere498be72005-09-09 13:03:32 -0700778 }
779 }
780}
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700781
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700782static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
783 int node, int page_node)
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700784{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000785 struct kmem_cache_node *n;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700786 struct alien_cache *alien = NULL;
787 struct array_cache *ac;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700788 LIST_HEAD(list);
Pekka Enberg1ca4cb22006-10-06 00:43:52 -0700789
Christoph Lameter18bf8542014-08-06 16:04:11 -0700790 n = get_node(cachep, node);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700791 STATS_INC_NODEFREES(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700792 if (n->alien && n->alien[page_node]) {
793 alien = n->alien[page_node];
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700794 ac = &alien->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700795 spin_lock(&alien->lock);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700796 if (unlikely(ac->avail == ac->limit)) {
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700797 STATS_INC_ACOVERFLOW(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700798 __drain_alien_cache(cachep, ac, page_node, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700799 }
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700800 ac->entry[ac->avail++] = objp;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700801 spin_unlock(&alien->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700802 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700803 } else {
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700804 n = get_node(cachep, page_node);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700805 spin_lock(&n->list_lock);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700806 free_block(cachep, &objp, 1, page_node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700807 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -0700808 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700809 }
810 return 1;
811}
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700812
813static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
814{
815 int page_node = page_to_nid(virt_to_page(objp));
816 int node = numa_mem_id();
817 /*
818 * Make sure we are not freeing a object from another node to the array
819 * cache on this cpu.
820 */
821 if (likely(node == page_node))
822 return 0;
823
824 return __cache_free_alien(cachep, objp, node, page_node);
825}
David Rientjes4167e9b2015-04-14 15:46:55 -0700826
827/*
Mel Gorman444eb2a42016-03-17 14:19:23 -0700828 * Construct gfp mask to allocate from a specific node but do not reclaim or
829 * warn about failures.
David Rientjes4167e9b2015-04-14 15:46:55 -0700830 */
831static inline gfp_t gfp_exact_node(gfp_t flags)
832{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700833 return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes4167e9b2015-04-14 15:46:55 -0700834}
Christoph Lametere498be72005-09-09 13:03:32 -0700835#endif
836
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700837static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
838{
839 struct kmem_cache_node *n;
840
841 /*
842 * Set up the kmem_cache_node for cpu before we can
843 * begin anything. Make sure some other cpu on this
844 * node has not already allocated this
845 */
846 n = get_node(cachep, node);
847 if (n) {
848 spin_lock_irq(&n->list_lock);
849 n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
850 cachep->num;
851 spin_unlock_irq(&n->list_lock);
852
853 return 0;
854 }
855
856 n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
857 if (!n)
858 return -ENOMEM;
859
860 kmem_cache_node_init(n);
861 n->next_reap = jiffies + REAPTIMEOUT_NODE +
862 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
863
864 n->free_limit =
865 (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;
866
867 /*
868 * The kmem_cache_nodes don't come and go as CPUs
869 * come and go. slab_mutex is sufficient
870 * protection here.
871 */
872 cachep->node[node] = n;
873
874 return 0;
875}
876
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200877#if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700878/*
Christoph Lameter6a673682013-01-10 19:14:19 +0000879 * Allocates and initializes node for a node on each slab cache, used for
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000880 * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node
David Rientjes8f9f8d92010-03-27 19:40:47 -0700881 * will be allocated off-node since memory is not yet online for the new node.
Christoph Lameter6a673682013-01-10 19:14:19 +0000882 * When hotplugging memory or a cpu, existing node are not replaced if
David Rientjes8f9f8d92010-03-27 19:40:47 -0700883 * already in use.
884 *
Christoph Lameter18004c52012-07-06 15:25:12 -0500885 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -0700886 */
Christoph Lameter6a673682013-01-10 19:14:19 +0000887static int init_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700888{
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700889 int ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700890 struct kmem_cache *cachep;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700891
Christoph Lameter18004c52012-07-06 15:25:12 -0500892 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700893 ret = init_cache_node(cachep, node, GFP_KERNEL);
894 if (ret)
895 return ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700896 }
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700897
David Rientjes8f9f8d92010-03-27 19:40:47 -0700898 return 0;
899}
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200900#endif
David Rientjes8f9f8d92010-03-27 19:40:47 -0700901
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700902static int setup_kmem_cache_node(struct kmem_cache *cachep,
903 int node, gfp_t gfp, bool force_change)
904{
905 int ret = -ENOMEM;
906 struct kmem_cache_node *n;
907 struct array_cache *old_shared = NULL;
908 struct array_cache *new_shared = NULL;
909 struct alien_cache **new_alien = NULL;
910 LIST_HEAD(list);
911
912 if (use_alien_caches) {
913 new_alien = alloc_alien_cache(node, cachep->limit, gfp);
914 if (!new_alien)
915 goto fail;
916 }
917
918 if (cachep->shared) {
919 new_shared = alloc_arraycache(node,
920 cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
921 if (!new_shared)
922 goto fail;
923 }
924
925 ret = init_cache_node(cachep, node, gfp);
926 if (ret)
927 goto fail;
928
929 n = get_node(cachep, node);
930 spin_lock_irq(&n->list_lock);
931 if (n->shared && force_change) {
932 free_block(cachep, n->shared->entry,
933 n->shared->avail, node, &list);
934 n->shared->avail = 0;
935 }
936
937 if (!n->shared || force_change) {
938 old_shared = n->shared;
939 n->shared = new_shared;
940 new_shared = NULL;
941 }
942
943 if (!n->alien) {
944 n->alien = new_alien;
945 new_alien = NULL;
946 }
947
948 spin_unlock_irq(&n->list_lock);
949 slabs_destroy(cachep, &list);
950
Joonsoo Kim801faf02016-05-19 17:10:31 -0700951 /*
952 * To protect lockless access to n->shared during irq disabled context.
953 * If n->shared isn't NULL in irq disabled context, accessing to it is
954 * guaranteed to be valid until irq is re-enabled, because it will be
955 * freed after synchronize_sched().
956 */
Joonsoo Kim86d9f482016-10-27 17:46:18 -0700957 if (old_shared && force_change)
Joonsoo Kim801faf02016-05-19 17:10:31 -0700958 synchronize_sched();
959
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700960fail:
961 kfree(old_shared);
962 kfree(new_shared);
963 free_alien_cache(new_alien);
964
965 return ret;
966}
967
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200968#ifdef CONFIG_SMP
969
Paul Gortmaker0db06282013-06-19 14:53:51 -0400970static void cpuup_canceled(long cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971{
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700972 struct kmem_cache *cachep;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000973 struct kmem_cache_node *n = NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -0700974 int node = cpu_to_mem(cpu);
Rusty Russella70f7302009-03-13 14:49:46 +1030975 const struct cpumask *mask = cpumask_of_node(node);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700976
Christoph Lameter18004c52012-07-06 15:25:12 -0500977 list_for_each_entry(cachep, &slab_caches, list) {
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700978 struct array_cache *nc;
979 struct array_cache *shared;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700980 struct alien_cache **alien;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700981 LIST_HEAD(list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700982
Christoph Lameter18bf8542014-08-06 16:04:11 -0700983 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000984 if (!n)
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700985 continue;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700986
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000987 spin_lock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700988
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000989 /* Free limit for this kmem_cache_node */
990 n->free_limit -= cachep->batchcount;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700991
992 /* cpu is dead; no one can alloc from it. */
993 nc = per_cpu_ptr(cachep->cpu_cache, cpu);
994 if (nc) {
Joonsoo Kim97654df2014-08-06 16:04:25 -0700995 free_block(cachep, nc->entry, nc->avail, node, &list);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700996 nc->avail = 0;
997 }
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700998
Rusty Russell58463c12009-12-17 11:43:12 -0600999 if (!cpumask_empty(mask)) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001000 spin_unlock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001001 goto free_slab;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001002 }
1003
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001004 shared = n->shared;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001005 if (shared) {
1006 free_block(cachep, shared->entry,
Joonsoo Kim97654df2014-08-06 16:04:25 -07001007 shared->avail, node, &list);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001008 n->shared = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001009 }
1010
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001011 alien = n->alien;
1012 n->alien = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001013
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001014 spin_unlock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001015
1016 kfree(shared);
1017 if (alien) {
1018 drain_alien_cache(cachep, alien);
1019 free_alien_cache(alien);
1020 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001021
1022free_slab:
Joonsoo Kim97654df2014-08-06 16:04:25 -07001023 slabs_destroy(cachep, &list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001024 }
1025 /*
1026 * In the previous loop, all the objects were freed to
1027 * the respective cache's slabs, now we can go ahead and
1028 * shrink each nodelist to its limit.
1029 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001030 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07001031 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001032 if (!n)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001033 continue;
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001034 drain_freelist(cachep, n, INT_MAX);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001035 }
1036}
1037
Paul Gortmaker0db06282013-06-19 14:53:51 -04001038static int cpuup_prepare(long cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001039{
Pekka Enberg343e0d72006-02-01 03:05:50 -08001040 struct kmem_cache *cachep;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07001041 int node = cpu_to_mem(cpu);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001042 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001044 /*
1045 * We need to do this right in the beginning since
1046 * alloc_arraycache's are going to use this list.
1047 * kmalloc_node allows us to add the slab to the right
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001048 * kmem_cache_node and not this cpu's kmem_cache_node
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001049 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001050 err = init_cache_node_node(node);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001051 if (err < 0)
1052 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001053
1054 /*
1055 * Now we can go ahead with allocating the shared arrays and
1056 * array caches
1057 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001058 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07001059 err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
1060 if (err)
1061 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001062 }
Pekka Enbergce79ddc2009-11-23 22:01:15 +02001063
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001064 return 0;
1065bad:
Akinobu Mita12d00f62007-10-18 03:05:11 -07001066 cpuup_canceled(cpu);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001067 return -ENOMEM;
1068}
1069
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001070int slab_prepare_cpu(unsigned int cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001071{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001072 int err;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001073
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001074 mutex_lock(&slab_mutex);
1075 err = cpuup_prepare(cpu);
1076 mutex_unlock(&slab_mutex);
1077 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001080/*
1081 * This is called for a failed online attempt and for a successful
1082 * offline.
1083 *
1084 * Even if all the cpus of a node are down, we don't free the
1085 * kmem_list3 of any cache. This to avoid a race between cpu_down, and
1086 * a kmalloc allocation from another cpu for memory from the node of
1087 * the cpu going down. The list3 structure is usually allocated from
1088 * kmem_cache_create() and gets destroyed at kmem_cache_destroy().
1089 */
1090int slab_dead_cpu(unsigned int cpu)
1091{
1092 mutex_lock(&slab_mutex);
1093 cpuup_canceled(cpu);
1094 mutex_unlock(&slab_mutex);
1095 return 0;
1096}
1097#endif
1098
1099static int slab_online_cpu(unsigned int cpu)
1100{
1101 start_cpu_timer(cpu);
1102 return 0;
1103}
1104
1105static int slab_offline_cpu(unsigned int cpu)
1106{
1107 /*
1108 * Shutdown cache reaper. Note that the slab_mutex is held so
1109 * that if cache_reap() is invoked it cannot do anything
1110 * expensive but will only modify reap_work and reschedule the
1111 * timer.
1112 */
1113 cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu));
1114 /* Now the cache_reaper is guaranteed to be not running. */
1115 per_cpu(slab_reap_work, cpu).work.func = NULL;
1116 return 0;
1117}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
David Rientjes8f9f8d92010-03-27 19:40:47 -07001119#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
1120/*
1121 * Drains freelist for a node on each slab cache, used for memory hot-remove.
1122 * Returns -EBUSY if all objects cannot be drained so that the node is not
1123 * removed.
1124 *
Christoph Lameter18004c52012-07-06 15:25:12 -05001125 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001126 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001127static int __meminit drain_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001128{
1129 struct kmem_cache *cachep;
1130 int ret = 0;
1131
Christoph Lameter18004c52012-07-06 15:25:12 -05001132 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001133 struct kmem_cache_node *n;
David Rientjes8f9f8d92010-03-27 19:40:47 -07001134
Christoph Lameter18bf8542014-08-06 16:04:11 -07001135 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001136 if (!n)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001137 continue;
1138
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001139 drain_freelist(cachep, n, INT_MAX);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001140
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001141 if (!list_empty(&n->slabs_full) ||
1142 !list_empty(&n->slabs_partial)) {
David Rientjes8f9f8d92010-03-27 19:40:47 -07001143 ret = -EBUSY;
1144 break;
1145 }
1146 }
1147 return ret;
1148}
1149
1150static int __meminit slab_memory_callback(struct notifier_block *self,
1151 unsigned long action, void *arg)
1152{
1153 struct memory_notify *mnb = arg;
1154 int ret = 0;
1155 int nid;
1156
1157 nid = mnb->status_change_nid;
1158 if (nid < 0)
1159 goto out;
1160
1161 switch (action) {
1162 case MEM_GOING_ONLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001163 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001164 ret = init_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001165 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001166 break;
1167 case MEM_GOING_OFFLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001168 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001169 ret = drain_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001170 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001171 break;
1172 case MEM_ONLINE:
1173 case MEM_OFFLINE:
1174 case MEM_CANCEL_ONLINE:
1175 case MEM_CANCEL_OFFLINE:
1176 break;
1177 }
1178out:
Prarit Bhargava5fda1bd2011-03-22 16:30:49 -07001179 return notifier_from_errno(ret);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001180}
1181#endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */
1182
Christoph Lametere498be72005-09-09 13:03:32 -07001183/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001184 * swap the static kmem_cache_node with kmalloced memory
Christoph Lametere498be72005-09-09 13:03:32 -07001185 */
Christoph Lameter6744f082013-01-10 19:12:17 +00001186static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list,
David Rientjes8f9f8d92010-03-27 19:40:47 -07001187 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07001188{
Christoph Lameter6744f082013-01-10 19:12:17 +00001189 struct kmem_cache_node *ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001190
Christoph Lameter6744f082013-01-10 19:12:17 +00001191 ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07001192 BUG_ON(!ptr);
1193
Christoph Lameter6744f082013-01-10 19:12:17 +00001194 memcpy(ptr, list, sizeof(struct kmem_cache_node));
Ingo Molnar2b2d5492006-07-03 00:25:28 -07001195 /*
1196 * Do not assume that spinlocks can be initialized via memcpy:
1197 */
1198 spin_lock_init(&ptr->list_lock);
1199
Christoph Lametere498be72005-09-09 13:03:32 -07001200 MAKE_ALL_LISTS(cachep, ptr, nodeid);
Christoph Lameter6a673682013-01-10 19:14:19 +00001201 cachep->node[nodeid] = ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001202}
1203
Andrew Mortona737b3e2006-03-22 00:08:11 -08001204/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001205 * For setting up all the kmem_cache_node for cache whose buffer_size is same as
1206 * size of kmem_cache_node.
Pekka Enberg556a1692008-01-25 08:20:51 +02001207 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001208static void __init set_up_node(struct kmem_cache *cachep, int index)
Pekka Enberg556a1692008-01-25 08:20:51 +02001209{
1210 int node;
1211
1212 for_each_online_node(node) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001213 cachep->node[node] = &init_kmem_cache_node[index + node];
Christoph Lameter6a673682013-01-10 19:14:19 +00001214 cachep->node[node]->next_reap = jiffies +
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001215 REAPTIMEOUT_NODE +
1216 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enberg556a1692008-01-25 08:20:51 +02001217 }
1218}
1219
1220/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08001221 * Initialisation. Called after the page allocator have been initialised and
1222 * before smp_init().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 */
1224void __init kmem_cache_init(void)
1225{
Christoph Lametere498be72005-09-09 13:03:32 -07001226 int i;
1227
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001228 kmem_cache = &kmem_cache_boot;
1229
Joonsoo Kim88881772016-05-19 17:10:05 -07001230 if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1)
Siddha, Suresh B62918a02007-05-02 19:27:18 +02001231 use_alien_caches = 0;
1232
Christoph Lameter3c583462012-11-28 16:23:01 +00001233 for (i = 0; i < NUM_INIT_LISTS; i++)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001234 kmem_cache_node_init(&init_kmem_cache_node[i]);
Christoph Lameter3c583462012-11-28 16:23:01 +00001235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 /*
1237 * Fragmentation resistance on low memory - only use bigger
David Rientjes3df1ccc2011-10-18 22:09:28 -07001238 * page orders on machines with more than 32MB of memory if
1239 * not overridden on the command line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 */
David Rientjes3df1ccc2011-10-18 22:09:28 -07001241 if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT)
David Rientjes543585c2011-10-18 22:09:24 -07001242 slab_max_order = SLAB_MAX_ORDER_HI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 /* Bootstrap is tricky, because several objects are allocated
1245 * from caches that do not exist yet:
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001246 * 1) initialize the kmem_cache cache: it contains the struct
1247 * kmem_cache structures of all caches, except kmem_cache itself:
1248 * kmem_cache is statically allocated.
Christoph Lametere498be72005-09-09 13:03:32 -07001249 * Initially an __init data area is used for the head array and the
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001250 * kmem_cache_node structures, it's replaced with a kmalloc allocated
Christoph Lametere498be72005-09-09 13:03:32 -07001251 * array at the end of the bootstrap.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 * 2) Create the first kmalloc cache.
Pekka Enberg343e0d72006-02-01 03:05:50 -08001253 * The struct kmem_cache for the new cache is allocated normally.
Christoph Lametere498be72005-09-09 13:03:32 -07001254 * An __init data area is used for the head array.
1255 * 3) Create the remaining kmalloc caches, with minimally sized
1256 * head arrays.
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001257 * 4) Replace the __init data head arrays for kmem_cache and the first
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 * kmalloc cache with kmalloc allocated arrays.
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001259 * 5) Replace the __init data for kmem_cache_node for kmem_cache and
Christoph Lametere498be72005-09-09 13:03:32 -07001260 * the other cache's with kmalloc allocated memory.
1261 * 6) Resize the head arrays of the kmalloc caches to their final sizes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 */
1263
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001264 /* 1) create the kmem_cache */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
Eric Dumazet8da34302007-05-06 14:49:29 -07001266 /*
Eric Dumazetb56efcf2011-07-20 19:04:23 +02001267 * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids
Eric Dumazet8da34302007-05-06 14:49:29 -07001268 */
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001269 create_boot_cache(kmem_cache, "kmem_cache",
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001270 offsetof(struct kmem_cache, node) +
Christoph Lameter6744f082013-01-10 19:12:17 +00001271 nr_node_ids * sizeof(struct kmem_cache_node *),
David Windsor8eb82842017-06-10 22:50:28 -04001272 SLAB_HWCACHE_ALIGN, 0, 0);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001273 list_add(&kmem_cache->list, &slab_caches);
Shakeel Butt880cd272018-03-28 16:00:57 -07001274 memcg_link_cache(kmem_cache);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001275 slab_state = PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Andrew Mortona737b3e2006-03-22 00:08:11 -08001277 /*
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001278 * Initialize the caches that provide memory for the kmem_cache_node
1279 * structures first. Without this, further allocations will bug.
Christoph Lametere498be72005-09-09 13:03:32 -07001280 */
Vlastimil Babka9d8edd92018-10-26 15:05:34 -07001281 kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE] = create_kmalloc_cache(
Vlastimil Babkaaf3b5f82017-02-22 15:41:05 -08001282 kmalloc_info[INDEX_NODE].name,
David Windsor6c0c21a2017-06-10 22:50:47 -04001283 kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS,
1284 0, kmalloc_size(INDEX_NODE));
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001285 slab_state = PARTIAL_NODE;
Daniel Sanders34cc6992015-06-24 16:55:57 -07001286 setup_kmalloc_cache_index_table();
Christoph Lametere498be72005-09-09 13:03:32 -07001287
Ingo Molnare0a42722006-06-23 02:03:46 -07001288 slab_early_init = 0;
1289
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001290 /* 5) Replace the bootstrap kmem_cache_node */
Christoph Lametere498be72005-09-09 13:03:32 -07001291 {
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07001292 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
Mel Gorman9c09a952008-01-24 05:49:54 -08001294 for_each_online_node(nid) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001295 init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid);
Pekka Enberg556a1692008-01-25 08:20:51 +02001296
Vlastimil Babka9d8edd92018-10-26 15:05:34 -07001297 init_list(kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE],
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001298 &init_kmem_cache_node[SIZE_NODE + nid], nid);
Christoph Lametere498be72005-09-09 13:03:32 -07001299 }
1300 }
1301
Christoph Lameterf97d5f62013-01-10 19:12:17 +00001302 create_kmalloc_caches(ARCH_KMALLOC_FLAGS);
Pekka Enberg8429db52009-06-12 15:58:59 +03001303}
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001304
Pekka Enberg8429db52009-06-12 15:58:59 +03001305void __init kmem_cache_init_late(void)
1306{
1307 struct kmem_cache *cachep;
1308
Pekka Enberg8429db52009-06-12 15:58:59 +03001309 /* 6) resize the head arrays to their final sizes */
Christoph Lameter18004c52012-07-06 15:25:12 -05001310 mutex_lock(&slab_mutex);
1311 list_for_each_entry(cachep, &slab_caches, list)
Pekka Enberg8429db52009-06-12 15:58:59 +03001312 if (enable_cpucache(cachep, GFP_NOWAIT))
1313 BUG();
Christoph Lameter18004c52012-07-06 15:25:12 -05001314 mutex_unlock(&slab_mutex);
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001315
Christoph Lameter97d06602012-07-06 15:25:11 -05001316 /* Done! */
1317 slab_state = FULL;
1318
David Rientjes8f9f8d92010-03-27 19:40:47 -07001319#ifdef CONFIG_NUMA
1320 /*
1321 * Register a memory hotplug callback that initializes and frees
Christoph Lameter6a673682013-01-10 19:14:19 +00001322 * node.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001323 */
1324 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
1325#endif
1326
Andrew Mortona737b3e2006-03-22 00:08:11 -08001327 /*
1328 * The reap timers are started later, with a module init call: That part
1329 * of the kernel is not yet operational.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 */
1331}
1332
1333static int __init cpucache_init(void)
1334{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001335 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Andrew Mortona737b3e2006-03-22 00:08:11 -08001337 /*
1338 * Register the timers that return unneeded pages to the page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001340 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online",
1341 slab_online_cpu, slab_offline_cpu);
1342 WARN_ON(ret < 0);
Glauber Costaa164f8962012-06-21 00:59:18 +04001343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 return 0;
1345}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346__initcall(cpucache_init);
1347
Rafael Aquini8bdec192012-03-09 17:27:27 -03001348static noinline void
1349slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
1350{
David Rientjes9a02d692014-06-04 16:06:36 -07001351#if DEBUG
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001352 struct kmem_cache_node *n;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001353 unsigned long flags;
1354 int node;
David Rientjes9a02d692014-06-04 16:06:36 -07001355 static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
1356 DEFAULT_RATELIMIT_BURST);
1357
1358 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs))
1359 return;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001360
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07001361 pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
1362 nodeid, gfpflags, &gfpflags);
1363 pr_warn(" cache: %s, object size: %d, order: %d\n",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001364 cachep->name, cachep->size, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001365
Christoph Lameter18bf8542014-08-06 16:04:11 -07001366 for_each_kmem_cache_node(cachep, node, n) {
David Rientjesbf00bd32016-12-12 16:41:44 -08001367 unsigned long total_slabs, free_slabs, free_objs;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001368
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001369 spin_lock_irqsave(&n->list_lock, flags);
David Rientjesbf00bd32016-12-12 16:41:44 -08001370 total_slabs = n->total_slabs;
1371 free_slabs = n->free_slabs;
1372 free_objs = n->free_objects;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001373 spin_unlock_irqrestore(&n->list_lock, flags);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001374
David Rientjesbf00bd32016-12-12 16:41:44 -08001375 pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n",
1376 node, total_slabs - free_slabs, total_slabs,
1377 (total_slabs * cachep->num) - free_objs,
1378 total_slabs * cachep->num);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001379 }
David Rientjes9a02d692014-06-04 16:06:36 -07001380#endif
Rafael Aquini8bdec192012-03-09 17:27:27 -03001381}
1382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383/*
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001384 * Interface to system's page allocator. No need to hold the
1385 * kmem_cache_node ->list_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 *
1387 * If we requested dmaable memory, we will get it. Even if we
1388 * did not request dmaable memory, we might get it, but that
1389 * would be relatively rare and ignorable.
1390 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001391static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
1392 int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
1394 struct page *page;
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001395 int nr_pages;
Christoph Lameter765c4502006-09-27 01:50:08 -07001396
Glauber Costaa618e892012-06-14 16:17:21 +04001397 flags |= cachep->allocflags;
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001398
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -08001399 page = __alloc_pages_node(nodeid, flags, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001400 if (!page) {
David Rientjes9a02d692014-06-04 16:06:36 -07001401 slab_out_of_memory(cachep, flags, nodeid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 return NULL;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001403 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001405 if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) {
1406 __free_pages(page, cachep->gfporder);
1407 return NULL;
1408 }
1409
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001410 nr_pages = (1 << cachep->gfporder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
Johannes Weiner7779f212017-07-06 15:40:55 -07001412 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, nr_pages);
Christoph Lameter972d1a72006-09-25 23:31:51 -07001413 else
Johannes Weiner7779f212017-07-06 15:40:55 -07001414 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, nr_pages);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07001415
Joonsoo Kima57a4982013-10-24 10:07:44 +09001416 __SetPageSlab(page);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07001417 /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */
1418 if (sk_memalloc_socks() && page_is_pfmemalloc(page))
Joonsoo Kima57a4982013-10-24 10:07:44 +09001419 SetPageSlabPfmemalloc(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001420
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001421 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422}
1423
1424/*
1425 * Interface to system's page release.
1426 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001427static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428{
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001429 int order = cachep->gfporder;
1430 unsigned long nr_freed = (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Christoph Lameter972d1a72006-09-25 23:31:51 -07001432 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
Johannes Weiner7779f212017-07-06 15:40:55 -07001433 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, -nr_freed);
Christoph Lameter972d1a72006-09-25 23:31:51 -07001434 else
Johannes Weiner7779f212017-07-06 15:40:55 -07001435 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, -nr_freed);
Joonsoo Kim73293c22013-10-24 10:07:37 +09001436
Joonsoo Kima57a4982013-10-24 10:07:44 +09001437 BUG_ON(!PageSlab(page));
Joonsoo Kim73293c22013-10-24 10:07:37 +09001438 __ClearPageSlabPfmemalloc(page);
Joonsoo Kima57a4982013-10-24 10:07:44 +09001439 __ClearPageSlab(page);
Joonsoo Kim8456a642013-10-24 10:07:49 +09001440 page_mapcount_reset(page);
1441 page->mapping = NULL;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (current->reclaim_state)
1444 current->reclaim_state->reclaimed_slab += nr_freed;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001445 memcg_uncharge_slab(page, order, cachep);
1446 __free_pages(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447}
1448
1449static void kmem_rcu_free(struct rcu_head *head)
1450{
Joonsoo Kim68126702013-10-24 10:07:42 +09001451 struct kmem_cache *cachep;
1452 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Joonsoo Kim68126702013-10-24 10:07:42 +09001454 page = container_of(head, struct page, rcu_head);
1455 cachep = page->slab_cache;
1456
1457 kmem_freepages(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458}
1459
1460#if DEBUG
Joonsoo Kim40b44132016-03-15 14:54:21 -07001461static bool is_debug_pagealloc_cache(struct kmem_cache *cachep)
1462{
1463 if (debug_pagealloc_enabled() && OFF_SLAB(cachep) &&
1464 (cachep->size % PAGE_SIZE) == 0)
1465 return true;
1466
1467 return false;
1468}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
1470#ifdef CONFIG_DEBUG_PAGEALLOC
Pekka Enberg343e0d72006-02-01 03:05:50 -08001471static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr,
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001472 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001474 int size = cachep->object_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001476 addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001478 if (size < 5 * sizeof(unsigned long))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 return;
1480
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001481 *addr++ = 0x12345678;
1482 *addr++ = caller;
1483 *addr++ = smp_processor_id();
1484 size -= 3 * sizeof(unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 {
1486 unsigned long *sptr = &caller;
1487 unsigned long svalue;
1488
1489 while (!kstack_end(sptr)) {
1490 svalue = *sptr++;
1491 if (kernel_text_address(svalue)) {
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001492 *addr++ = svalue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 size -= sizeof(unsigned long);
1494 if (size <= sizeof(unsigned long))
1495 break;
1496 }
1497 }
1498
1499 }
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001500 *addr++ = 0x87654321;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
Joonsoo Kim40b44132016-03-15 14:54:21 -07001502
1503static void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1504 int map, unsigned long caller)
1505{
1506 if (!is_debug_pagealloc_cache(cachep))
1507 return;
1508
1509 if (caller)
1510 store_stackinfo(cachep, objp, caller);
1511
1512 kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map);
1513}
1514
1515#else
1516static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1517 int map, unsigned long caller) {}
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519#endif
1520
Pekka Enberg343e0d72006-02-01 03:05:50 -08001521static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522{
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001523 int size = cachep->object_size;
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001524 addr = &((char *)addr)[obj_offset(cachep)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
1526 memset(addr, val, size);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001527 *(unsigned char *)(addr + size - 1) = POISON_END;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528}
1529
1530static void dump_line(char *data, int offset, int limit)
1531{
1532 int i;
Dave Jonesaa83aa42006-09-29 01:59:51 -07001533 unsigned char error = 0;
1534 int bad_count = 0;
1535
Joe Perches11705322016-03-17 14:19:50 -07001536 pr_err("%03x: ", offset);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001537 for (i = 0; i < limit; i++) {
1538 if (data[offset + i] != POISON_FREE) {
1539 error = data[offset + i];
1540 bad_count++;
1541 }
Dave Jonesaa83aa42006-09-29 01:59:51 -07001542 }
Sebastian Andrzej Siewiorfdde6ab2011-07-29 18:22:13 +02001543 print_hex_dump(KERN_CONT, "", 0, 16, 1,
1544 &data[offset], limit, 1);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001545
1546 if (bad_count == 1) {
1547 error ^= POISON_FREE;
1548 if (!(error & (error - 1))) {
Joe Perches11705322016-03-17 14:19:50 -07001549 pr_err("Single bit error detected. Probably bad RAM.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001550#ifdef CONFIG_X86
Joe Perches11705322016-03-17 14:19:50 -07001551 pr_err("Run memtest86+ or a similar memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001552#else
Joe Perches11705322016-03-17 14:19:50 -07001553 pr_err("Run a memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001554#endif
1555 }
1556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557}
1558#endif
1559
1560#if DEBUG
1561
Pekka Enberg343e0d72006-02-01 03:05:50 -08001562static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
1564 int i, size;
1565 char *realobj;
1566
1567 if (cachep->flags & SLAB_RED_ZONE) {
Joe Perches11705322016-03-17 14:19:50 -07001568 pr_err("Redzone: 0x%llx/0x%llx\n",
1569 *dbg_redzone1(cachep, objp),
1570 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
1572
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001573 if (cachep->flags & SLAB_STORE_USER)
1574 pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp));
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001575 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001576 size = cachep->object_size;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001577 for (i = 0; i < size && lines; i += 16, lines--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 int limit;
1579 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001580 if (i + limit > size)
1581 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 dump_line(realobj, i, limit);
1583 }
1584}
1585
Pekka Enberg343e0d72006-02-01 03:05:50 -08001586static void check_poison_obj(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587{
1588 char *realobj;
1589 int size, i;
1590 int lines = 0;
1591
Joonsoo Kim40b44132016-03-15 14:54:21 -07001592 if (is_debug_pagealloc_cache(cachep))
1593 return;
1594
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001595 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001596 size = cachep->object_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001598 for (i = 0; i < size; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 char exp = POISON_FREE;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001600 if (i == size - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 exp = POISON_END;
1602 if (realobj[i] != exp) {
1603 int limit;
1604 /* Mismatch ! */
1605 /* Print header */
1606 if (lines == 0) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001607 pr_err("Slab corruption (%s): %s start=%px, len=%d\n",
Joe Perches11705322016-03-17 14:19:50 -07001608 print_tainted(), cachep->name,
1609 realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 print_objinfo(cachep, objp, 0);
1611 }
1612 /* Hexdump the affected line */
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001613 i = (i / 16) * 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001615 if (i + limit > size)
1616 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 dump_line(realobj, i, limit);
1618 i += 16;
1619 lines++;
1620 /* Limit to 5 lines */
1621 if (lines > 5)
1622 break;
1623 }
1624 }
1625 if (lines != 0) {
1626 /* Print some data about the neighboring objects, if they
1627 * exist:
1628 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001629 struct page *page = virt_to_head_page(objp);
Pekka Enberg8fea4e92006-03-22 00:08:10 -08001630 unsigned int objnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
Joonsoo Kim8456a642013-10-24 10:07:49 +09001632 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (objnr) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001634 objp = index_to_obj(cachep, page, objnr - 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001635 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001636 pr_err("Prev obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 print_objinfo(cachep, objp, 2);
1638 }
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001639 if (objnr + 1 < cachep->num) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001640 objp = index_to_obj(cachep, page, objnr + 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001641 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001642 pr_err("Next obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 print_objinfo(cachep, objp, 2);
1644 }
1645 }
1646}
1647#endif
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649#if DEBUG
Joonsoo Kim8456a642013-10-24 10:07:49 +09001650static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1651 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001652{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 int i;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001654
1655 if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) {
1656 poison_obj(cachep, page->freelist - obj_offset(cachep),
1657 POISON_FREE);
1658 }
1659
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001661 void *objp = index_to_obj(cachep, page, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
1663 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 check_poison_obj(cachep, objp);
Joonsoo Kim40b44132016-03-15 14:54:21 -07001665 slab_kernel_map(cachep, objp, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 }
1667 if (cachep->flags & SLAB_RED_ZONE) {
1668 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001669 slab_error(cachep, "start of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001671 slab_error(cachep, "end of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001674}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675#else
Joonsoo Kim8456a642013-10-24 10:07:49 +09001676static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1677 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001678{
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001679}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680#endif
1681
Randy Dunlap911851e2006-03-22 00:08:14 -08001682/**
1683 * slab_destroy - destroy and release all objects in a slab
1684 * @cachep: cache pointer being destroyed
Masanari Iidacb8ee1a2014-01-28 02:57:08 +09001685 * @page: page pointer being destroyed
Randy Dunlap911851e2006-03-22 00:08:14 -08001686 *
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001687 * Destroy all the objs in a slab page, and release the mem back to the system.
1688 * Before calling the slab page must have been unlinked from the cache. The
1689 * kmem_cache_node ->list_lock is not held/needed.
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001690 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001691static void slab_destroy(struct kmem_cache *cachep, struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001692{
Joonsoo Kim7e007352013-10-30 19:04:01 +09001693 void *freelist;
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001694
Joonsoo Kim8456a642013-10-24 10:07:49 +09001695 freelist = page->freelist;
1696 slab_destroy_debugcheck(cachep, page);
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001697 if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU))
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001698 call_rcu(&page->rcu_head, kmem_rcu_free);
1699 else
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001700 kmem_freepages(cachep, page);
Joonsoo Kim68126702013-10-24 10:07:42 +09001701
1702 /*
Joonsoo Kim8456a642013-10-24 10:07:49 +09001703 * From now on, we don't use freelist
Joonsoo Kim68126702013-10-24 10:07:42 +09001704 * although actual page can be freed in rcu context
1705 */
1706 if (OFF_SLAB(cachep))
Joonsoo Kim8456a642013-10-24 10:07:49 +09001707 kmem_cache_free(cachep->freelist_cache, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708}
1709
Joonsoo Kim97654df2014-08-06 16:04:25 -07001710static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list)
1711{
1712 struct page *page, *n;
1713
1714 list_for_each_entry_safe(page, n, list, lru) {
1715 list_del(&page->lru);
1716 slab_destroy(cachep, page);
1717 }
1718}
1719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720/**
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001721 * calculate_slab_order - calculate size (page order) of slabs
1722 * @cachep: pointer to the cache that is being created
1723 * @size: size of objects to be created in this cache.
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001724 * @flags: slab allocation flags
1725 *
1726 * Also calculates the number of objects per slab.
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001727 *
1728 * This could be made much more intelligent. For now, try to avoid using
1729 * high order pages for slabs. When the gfp() functions are more friendly
1730 * towards high-order requests, this should be changed.
1731 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001732static size_t calculate_slab_order(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001733 size_t size, slab_flags_t flags)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001734{
1735 size_t left_over = 0;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001736 int gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001737
Christoph Lameter0aa817f2007-05-16 22:11:01 -07001738 for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001739 unsigned int num;
1740 size_t remainder;
1741
Joonsoo Kim70f75062016-03-15 14:54:53 -07001742 num = cache_estimate(gfporder, size, flags, &remainder);
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001743 if (!num)
1744 continue;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001745
Joonsoo Kimf315e3f2013-12-02 17:49:41 +09001746 /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */
1747 if (num > SLAB_OBJ_MAX_NUM)
1748 break;
1749
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001750 if (flags & CFLGS_OFF_SLAB) {
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001751 struct kmem_cache *freelist_cache;
1752 size_t freelist_size;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001753
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001754 freelist_size = num * sizeof(freelist_idx_t);
1755 freelist_cache = kmalloc_slab(freelist_size, 0u);
1756 if (!freelist_cache)
1757 continue;
1758
1759 /*
1760 * Needed to avoid possible looping condition
Joonsoo Kim76b342b2016-05-19 17:10:26 -07001761 * in cache_grow_begin()
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001762 */
1763 if (OFF_SLAB(freelist_cache))
1764 continue;
1765
1766 /* check if off slab has enough benefit */
1767 if (freelist_cache->size > cachep->size / 2)
1768 continue;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001769 }
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001770
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001771 /* Found something acceptable - save it away */
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001772 cachep->num = num;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001773 cachep->gfporder = gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001774 left_over = remainder;
1775
1776 /*
Linus Torvaldsf78bb8a2006-03-08 10:33:05 -08001777 * A VFS-reclaimable slab tends to have most allocations
1778 * as GFP_NOFS and we really don't want to have to be allocating
1779 * higher-order pages when we are unable to shrink dcache.
1780 */
1781 if (flags & SLAB_RECLAIM_ACCOUNT)
1782 break;
1783
1784 /*
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001785 * Large number of objects is good, but very large slabs are
1786 * currently bad for the gfp()s.
1787 */
David Rientjes543585c2011-10-18 22:09:24 -07001788 if (gfporder >= slab_max_order)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001789 break;
1790
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001791 /*
1792 * Acceptable internal fragmentation?
1793 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001794 if (left_over * 8 <= (PAGE_SIZE << gfporder))
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001795 break;
1796 }
1797 return left_over;
1798}
1799
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001800static struct array_cache __percpu *alloc_kmem_cache_cpus(
1801 struct kmem_cache *cachep, int entries, int batchcount)
1802{
1803 int cpu;
1804 size_t size;
1805 struct array_cache __percpu *cpu_cache;
1806
1807 size = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim85c9f4b2014-10-13 15:51:01 -07001808 cpu_cache = __alloc_percpu(size, sizeof(void *));
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001809
1810 if (!cpu_cache)
1811 return NULL;
1812
1813 for_each_possible_cpu(cpu) {
1814 init_arraycache(per_cpu_ptr(cpu_cache, cpu),
1815 entries, batchcount);
1816 }
1817
1818 return cpu_cache;
1819}
1820
Fabian Frederickbd721ea2016-08-02 14:03:33 -07001821static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001822{
Christoph Lameter97d06602012-07-06 15:25:11 -05001823 if (slab_state >= FULL)
Pekka Enberg83b519e2009-06-10 19:40:04 +03001824 return enable_cpucache(cachep, gfp);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001825
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001826 cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1);
1827 if (!cachep->cpu_cache)
1828 return 1;
1829
Christoph Lameter97d06602012-07-06 15:25:11 -05001830 if (slab_state == DOWN) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001831 /* Creation of first cache (kmem_cache). */
1832 set_up_node(kmem_cache, CACHE_CACHE);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001833 } else if (slab_state == PARTIAL) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001834 /* For kmem_cache_node */
1835 set_up_node(cachep, SIZE_NODE);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001836 } else {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001837 int node;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001838
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001839 for_each_online_node(node) {
1840 cachep->node[node] = kmalloc_node(
1841 sizeof(struct kmem_cache_node), gfp, node);
1842 BUG_ON(!cachep->node[node]);
1843 kmem_cache_node_init(cachep->node[node]);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001844 }
1845 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001846
Christoph Lameter6a673682013-01-10 19:14:19 +00001847 cachep->node[numa_mem_id()]->next_reap =
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001848 jiffies + REAPTIMEOUT_NODE +
1849 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001850
1851 cpu_cache_get(cachep)->avail = 0;
1852 cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
1853 cpu_cache_get(cachep)->batchcount = 1;
1854 cpu_cache_get(cachep)->touched = 0;
1855 cachep->batchcount = 1;
1856 cachep->limit = BOOT_CPUCACHE_ENTRIES;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001857 return 0;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001858}
1859
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001860slab_flags_t kmem_cache_flags(unsigned int object_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001861 slab_flags_t flags, const char *name,
Joonsoo Kim12220de2014-10-09 15:26:24 -07001862 void (*ctor)(void *))
1863{
1864 return flags;
1865}
1866
1867struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -07001868__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001869 slab_flags_t flags, void (*ctor)(void *))
Joonsoo Kim12220de2014-10-09 15:26:24 -07001870{
1871 struct kmem_cache *cachep;
1872
1873 cachep = find_mergeable(size, align, flags, name, ctor);
1874 if (cachep) {
1875 cachep->refcount++;
1876
1877 /*
1878 * Adjust the object sizes so that we clear
1879 * the complete object on kzalloc.
1880 */
1881 cachep->object_size = max_t(int, cachep->object_size, size);
1882 }
1883 return cachep;
1884}
1885
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001886static bool set_objfreelist_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001887 size_t size, slab_flags_t flags)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001888{
1889 size_t left;
1890
1891 cachep->num = 0;
1892
Alexander Potapenkoa5587d82019-07-11 20:59:19 -07001893 /*
1894 * If slab auto-initialization on free is enabled, store the freelist
1895 * off-slab, so that its contents don't end up in one of the allocated
1896 * objects.
1897 */
1898 if (unlikely(slab_want_init_on_free(cachep)))
1899 return false;
1900
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001901 if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001902 return false;
1903
1904 left = calculate_slab_order(cachep, size,
1905 flags | CFLGS_OBJFREELIST_SLAB);
1906 if (!cachep->num)
1907 return false;
1908
1909 if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size)
1910 return false;
1911
1912 cachep->colour = left / cachep->colour_off;
1913
1914 return true;
1915}
1916
Joonsoo Kim158e3192016-03-15 14:54:35 -07001917static bool set_off_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001918 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001919{
1920 size_t left;
1921
1922 cachep->num = 0;
1923
1924 /*
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001925 * Always use on-slab management when SLAB_NOLEAKTRACE
1926 * to avoid recursive calls into kmemleak.
Joonsoo Kim158e3192016-03-15 14:54:35 -07001927 */
Joonsoo Kim158e3192016-03-15 14:54:35 -07001928 if (flags & SLAB_NOLEAKTRACE)
1929 return false;
1930
1931 /*
1932 * Size is large, assume best to place the slab management obj
1933 * off-slab (should allow better packing of objs).
1934 */
1935 left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB);
1936 if (!cachep->num)
1937 return false;
1938
1939 /*
1940 * If the slab has been placed off-slab, and we have enough space then
1941 * move it on-slab. This is at the expense of any extra colouring.
1942 */
1943 if (left >= cachep->num * sizeof(freelist_idx_t))
1944 return false;
1945
1946 cachep->colour = left / cachep->colour_off;
1947
1948 return true;
1949}
1950
1951static bool set_on_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001952 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001953{
1954 size_t left;
1955
1956 cachep->num = 0;
1957
1958 left = calculate_slab_order(cachep, size, flags);
1959 if (!cachep->num)
1960 return false;
1961
1962 cachep->colour = left / cachep->colour_off;
1963
1964 return true;
1965}
1966
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001967/**
Christoph Lameter039363f2012-07-06 15:25:10 -05001968 * __kmem_cache_create - Create a cache.
Randy Dunlapa755b762012-11-06 17:10:10 -08001969 * @cachep: cache management descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 * @flags: SLAB flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 *
1972 * Returns a ptr to the cache on success, NULL on failure.
1973 * Cannot be called within a int, but can be interrupted.
Paul Mundt20c2df82007-07-20 10:11:58 +09001974 * The @ctor is run when new pages are allocated by the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 * The flags are
1977 *
1978 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
1979 * to catch references to uninitialised memory.
1980 *
1981 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
1982 * for buffer overruns.
1983 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
1985 * cacheline. This can be beneficial if you're counting cycles as closely
1986 * as davem.
1987 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001988int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
David Rientjesd4a5fca2014-09-25 16:05:20 -07001990 size_t ralign = BYTES_PER_WORD;
Pekka Enberg83b519e2009-06-10 19:40:04 +03001991 gfp_t gfp;
Christoph Lameter278b1bb2012-09-05 00:20:34 +00001992 int err;
Alexey Dobriyanbe4a7982018-04-05 16:21:28 -07001993 unsigned int size = cachep->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996#if FORCED_DEBUG
1997 /*
1998 * Enable redzoning and last user accounting, except for caches with
1999 * large objects, if the increased size would increase the object size
2000 * above the next power of two: caches with object sizes just above a
2001 * power of two have a significant amount of internal fragmentation.
2002 */
David Woodhouse87a927c2007-07-04 21:26:44 -04002003 if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
2004 2 * sizeof(unsigned long long)))
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002005 flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08002006 if (!(flags & SLAB_TYPESAFE_BY_RCU))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 flags |= SLAB_POISON;
2008#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Andrew Mortona737b3e2006-03-22 00:08:11 -08002011 /*
2012 * Check that size is in terms of words. This is needed to avoid
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 * unaligned accesses for some archs when redzoning is used, and makes
2014 * sure any on-slab bufctl's are also correctly aligned.
2015 */
Canjiang Lue0771952017-07-06 15:36:37 -07002016 size = ALIGN(size, BYTES_PER_WORD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017
David Woodhouse87a927c2007-07-04 21:26:44 -04002018 if (flags & SLAB_RED_ZONE) {
2019 ralign = REDZONE_ALIGN;
2020 /* If redzoning, ensure that the second redzone is suitably
2021 * aligned, by adjusting the object size accordingly. */
Canjiang Lue0771952017-07-06 15:36:37 -07002022 size = ALIGN(size, REDZONE_ALIGN);
David Woodhouse87a927c2007-07-04 21:26:44 -04002023 }
Pekka Enbergca5f9702006-09-25 23:31:25 -07002024
Kevin Hilmana44b56d2006-12-06 20:32:11 -08002025 /* 3) caller mandated alignment */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002026 if (ralign < cachep->align) {
2027 ralign = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
Pekka Enberg3ff84a72011-02-14 17:46:21 +02002029 /* disable debug if necessary */
2030 if (ralign > __alignof__(unsigned long long))
Kevin Hilmana44b56d2006-12-06 20:32:11 -08002031 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
Andrew Mortona737b3e2006-03-22 00:08:11 -08002032 /*
Pekka Enbergca5f9702006-09-25 23:31:25 -07002033 * 4) Store it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002035 cachep->align = ralign;
Joonsoo Kim158e3192016-03-15 14:54:35 -07002036 cachep->colour_off = cache_line_size();
2037 /* Offset must be a multiple of the alignment. */
2038 if (cachep->colour_off < cachep->align)
2039 cachep->colour_off = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
Pekka Enberg83b519e2009-06-10 19:40:04 +03002041 if (slab_is_available())
2042 gfp = GFP_KERNEL;
2043 else
2044 gfp = GFP_NOWAIT;
2045
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Pekka Enbergca5f9702006-09-25 23:31:25 -07002048 /*
2049 * Both debugging options require word-alignment which is calculated
2050 * into align above.
2051 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (flags & SLAB_RED_ZONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 /* add space for red zone words */
Pekka Enberg3ff84a72011-02-14 17:46:21 +02002054 cachep->obj_offset += sizeof(unsigned long long);
2055 size += 2 * sizeof(unsigned long long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 }
2057 if (flags & SLAB_STORE_USER) {
Pekka Enbergca5f9702006-09-25 23:31:25 -07002058 /* user store requires one word storage behind the end of
David Woodhouse87a927c2007-07-04 21:26:44 -04002059 * the real object. But if the second red zone needs to be
2060 * aligned to 64 bits, we must allow that much space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 */
David Woodhouse87a927c2007-07-04 21:26:44 -04002062 if (flags & SLAB_RED_ZONE)
2063 size += REDZONE_ALIGN;
2064 else
2065 size += BYTES_PER_WORD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 }
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002067#endif
2068
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002069 kasan_cache_create(cachep, &size, &flags);
2070
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002071 size = ALIGN(size, cachep->align);
2072 /*
2073 * We should restrict the number of objects in a slab to implement
2074 * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition.
2075 */
2076 if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE)
2077 size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align);
2078
2079#if DEBUG
Joonsoo Kim03a2d2a2015-10-01 15:36:54 -07002080 /*
2081 * To activate debug pagealloc, off-slab management is necessary
2082 * requirement. In early phase of initialization, small sized slab
2083 * doesn't get initialized so it would not be possible. So, we need
2084 * to check size >= 256. It guarantees that all necessary small
2085 * sized slab is initialized in current slab initialization sequence.
2086 */
Joonsoo Kim40323272016-03-15 14:54:18 -07002087 if (debug_pagealloc_enabled() && (flags & SLAB_POISON) &&
Joonsoo Kimf3a3c322016-03-15 14:54:38 -07002088 size >= 256 && cachep->object_size > cache_line_size()) {
2089 if (size < PAGE_SIZE || size % PAGE_SIZE == 0) {
2090 size_t tmp_size = ALIGN(size, PAGE_SIZE);
2091
2092 if (set_off_slab_cache(cachep, tmp_size, flags)) {
2093 flags |= CFLGS_OFF_SLAB;
2094 cachep->obj_offset += tmp_size - size;
2095 size = tmp_size;
2096 goto done;
2097 }
2098 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 }
2100#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002102 if (set_objfreelist_slab_cache(cachep, size, flags)) {
2103 flags |= CFLGS_OBJFREELIST_SLAB;
2104 goto done;
2105 }
2106
Joonsoo Kim158e3192016-03-15 14:54:35 -07002107 if (set_off_slab_cache(cachep, size, flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 flags |= CFLGS_OFF_SLAB;
Joonsoo Kim158e3192016-03-15 14:54:35 -07002109 goto done;
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
Joonsoo Kim158e3192016-03-15 14:54:35 -07002112 if (set_on_slab_cache(cachep, size, flags))
2113 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Joonsoo Kim158e3192016-03-15 14:54:35 -07002115 return -E2BIG;
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002116
Joonsoo Kim158e3192016-03-15 14:54:35 -07002117done:
2118 cachep->freelist_size = cachep->num * sizeof(freelist_idx_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 cachep->flags = flags;
Joonsoo Kima57a4982013-10-24 10:07:44 +09002120 cachep->allocflags = __GFP_COMP;
Yang Shia3187e42016-05-19 17:10:41 -07002121 if (flags & SLAB_CACHE_DMA)
Glauber Costaa618e892012-06-14 16:17:21 +04002122 cachep->allocflags |= GFP_DMA;
Nicolas Boichat62d342d2019-03-28 20:43:42 -07002123 if (flags & SLAB_CACHE_DMA32)
2124 cachep->allocflags |= GFP_DMA32;
David Rientjesa3ba0742017-11-15 17:32:14 -08002125 if (flags & SLAB_RECLAIM_ACCOUNT)
2126 cachep->allocflags |= __GFP_RECLAIMABLE;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002127 cachep->size = size;
Eric Dumazet6a2d7a92006-12-13 00:34:27 -08002128 cachep->reciprocal_buffer_size = reciprocal_value(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129
Joonsoo Kim40b44132016-03-15 14:54:21 -07002130#if DEBUG
2131 /*
2132 * If we're going to use the generic kernel_map_pages()
2133 * poisoning, then it's going to smash the contents of
2134 * the redzone and userword anyhow, so switch them off.
2135 */
2136 if (IS_ENABLED(CONFIG_PAGE_POISONING) &&
2137 (cachep->flags & SLAB_POISON) &&
2138 is_debug_pagealloc_cache(cachep))
2139 cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2140#endif
2141
2142 if (OFF_SLAB(cachep)) {
Joonsoo Kim158e3192016-03-15 14:54:35 -07002143 cachep->freelist_cache =
2144 kmalloc_slab(cachep->freelist_size, 0u);
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002145 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002147 err = setup_cpu_cache(cachep, gfp);
2148 if (err) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002149 __kmem_cache_release(cachep);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002150 return err;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07002151 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002153 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
2156#if DEBUG
2157static void check_irq_off(void)
2158{
2159 BUG_ON(!irqs_disabled());
2160}
2161
2162static void check_irq_on(void)
2163{
2164 BUG_ON(irqs_disabled());
2165}
2166
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002167static void check_mutex_acquired(void)
2168{
2169 BUG_ON(!mutex_is_locked(&slab_mutex));
2170}
2171
Pekka Enberg343e0d72006-02-01 03:05:50 -08002172static void check_spinlock_acquired(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173{
2174#ifdef CONFIG_SMP
2175 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002176 assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177#endif
2178}
Christoph Lametere498be72005-09-09 13:03:32 -07002179
Pekka Enberg343e0d72006-02-01 03:05:50 -08002180static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
Christoph Lametere498be72005-09-09 13:03:32 -07002181{
2182#ifdef CONFIG_SMP
2183 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002184 assert_spin_locked(&get_node(cachep, node)->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07002185#endif
2186}
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188#else
2189#define check_irq_off() do { } while(0)
2190#define check_irq_on() do { } while(0)
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002191#define check_mutex_acquired() do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192#define check_spinlock_acquired(x) do { } while(0)
Christoph Lametere498be72005-09-09 13:03:32 -07002193#define check_spinlock_acquired_node(x, y) do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194#endif
2195
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002196static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac,
2197 int node, bool free_all, struct list_head *list)
2198{
2199 int tofree;
2200
2201 if (!ac || !ac->avail)
2202 return;
2203
2204 tofree = free_all ? ac->avail : (ac->limit + 4) / 5;
2205 if (tofree > ac->avail)
2206 tofree = (ac->avail + 1) / 2;
2207
2208 free_block(cachep, ac->entry, tofree, node, list);
2209 ac->avail -= tofree;
2210 memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail);
2211}
Christoph Lameteraab22072006-03-22 00:09:06 -08002212
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213static void do_drain(void *arg)
2214{
Andrew Mortona737b3e2006-03-22 00:08:11 -08002215 struct kmem_cache *cachep = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 struct array_cache *ac;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002217 int node = numa_mem_id();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002218 struct kmem_cache_node *n;
Joonsoo Kim97654df2014-08-06 16:04:25 -07002219 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220
2221 check_irq_off();
Pekka Enberg9a2dba42006-02-01 03:05:49 -08002222 ac = cpu_cache_get(cachep);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002223 n = get_node(cachep, node);
2224 spin_lock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07002225 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002226 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07002227 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 ac->avail = 0;
2229}
2230
Pekka Enberg343e0d72006-02-01 03:05:50 -08002231static void drain_cpu_caches(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002233 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002234 int node;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002235 LIST_HEAD(list);
Christoph Lametere498be72005-09-09 13:03:32 -07002236
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002237 on_each_cpu(do_drain, cachep, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002239 for_each_kmem_cache_node(cachep, node, n)
2240 if (n->alien)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002241 drain_alien_cache(cachep, n->alien);
Roland Dreiera4523a82006-05-15 11:41:00 -07002242
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002243 for_each_kmem_cache_node(cachep, node, n) {
2244 spin_lock_irq(&n->list_lock);
2245 drain_array_locked(cachep, n->shared, node, true, &list);
2246 spin_unlock_irq(&n->list_lock);
2247
2248 slabs_destroy(cachep, &list);
2249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250}
2251
Christoph Lametered11d9e2006-06-30 01:55:45 -07002252/*
2253 * Remove slabs from the list of free slabs.
2254 * Specify the number of slabs to drain in tofree.
2255 *
2256 * Returns the actual number of slabs released.
2257 */
2258static int drain_freelist(struct kmem_cache *cache,
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002259 struct kmem_cache_node *n, int tofree)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260{
Christoph Lametered11d9e2006-06-30 01:55:45 -07002261 struct list_head *p;
2262 int nr_freed;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002263 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Christoph Lametered11d9e2006-06-30 01:55:45 -07002265 nr_freed = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002266 while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002268 spin_lock_irq(&n->list_lock);
2269 p = n->slabs_free.prev;
2270 if (p == &n->slabs_free) {
2271 spin_unlock_irq(&n->list_lock);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002272 goto out;
2273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274
Joonsoo Kim8456a642013-10-24 10:07:49 +09002275 page = list_entry(p, struct page, lru);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002276 list_del(&page->lru);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002277 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08002278 n->total_slabs--;
Christoph Lametered11d9e2006-06-30 01:55:45 -07002279 /*
2280 * Safe to drop the lock. The slab is no longer linked
2281 * to the cache.
2282 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002283 n->free_objects -= cache->num;
2284 spin_unlock_irq(&n->list_lock);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002285 slab_destroy(cache, page);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002286 nr_freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 }
Christoph Lametered11d9e2006-06-30 01:55:45 -07002288out:
2289 return nr_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290}
2291
Shakeel Buttf9e13c02018-04-05 16:21:57 -07002292bool __kmem_cache_empty(struct kmem_cache *s)
2293{
2294 int node;
2295 struct kmem_cache_node *n;
2296
2297 for_each_kmem_cache_node(s, node, n)
2298 if (!list_empty(&n->slabs_full) ||
2299 !list_empty(&n->slabs_partial))
2300 return false;
2301 return true;
2302}
2303
Tejun Heoc9fc5862017-02-22 15:41:27 -08002304int __kmem_cache_shrink(struct kmem_cache *cachep)
Christoph Lametere498be72005-09-09 13:03:32 -07002305{
Christoph Lameter18bf8542014-08-06 16:04:11 -07002306 int ret = 0;
2307 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002308 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002309
2310 drain_cpu_caches(cachep);
2311
2312 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002313 for_each_kmem_cache_node(cachep, node, n) {
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07002314 drain_freelist(cachep, n, INT_MAX);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002315
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002316 ret += !list_empty(&n->slabs_full) ||
2317 !list_empty(&n->slabs_partial);
Christoph Lametere498be72005-09-09 13:03:32 -07002318 }
2319 return (ret ? 1 : 0);
2320}
2321
Tejun Heoc9fc5862017-02-22 15:41:27 -08002322#ifdef CONFIG_MEMCG
2323void __kmemcg_cache_deactivate(struct kmem_cache *cachep)
2324{
2325 __kmem_cache_shrink(cachep);
2326}
2327#endif
2328
Christoph Lameter945cf2b2012-09-04 23:18:33 +00002329int __kmem_cache_shutdown(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330{
Tejun Heoc9fc5862017-02-22 15:41:27 -08002331 return __kmem_cache_shrink(cachep);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002332}
2333
2334void __kmem_cache_release(struct kmem_cache *cachep)
2335{
Christoph Lameter12c36672012-09-04 23:38:33 +00002336 int i;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002337 struct kmem_cache_node *n;
Christoph Lameter12c36672012-09-04 23:38:33 +00002338
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002339 cache_random_seq_destroy(cachep);
2340
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07002341 free_percpu(cachep->cpu_cache);
Christoph Lameter12c36672012-09-04 23:38:33 +00002342
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002343 /* NUMA: free the node structures */
Christoph Lameter18bf8542014-08-06 16:04:11 -07002344 for_each_kmem_cache_node(cachep, i, n) {
2345 kfree(n->shared);
2346 free_alien_cache(n->alien);
2347 kfree(n);
2348 cachep->node[i] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002352/*
2353 * Get the memory for a slab management obj.
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08002354 *
2355 * For a slab cache when the slab descriptor is off-slab, the
2356 * slab descriptor can't come from the same cache which is being created,
2357 * Because if it is the case, that means we defer the creation of
2358 * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point.
2359 * And we eventually call down to __kmem_cache_create(), which
2360 * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one.
2361 * This is a "chicken-and-egg" problem.
2362 *
2363 * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches,
2364 * which are all initialized during kmem_cache_init().
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002365 */
Joonsoo Kim7e007352013-10-30 19:04:01 +09002366static void *alloc_slabmgmt(struct kmem_cache *cachep,
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002367 struct page *page, int colour_off,
2368 gfp_t local_flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002370 void *freelist;
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002371 void *addr = page_address(page);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002372
Andrey Konovalov8607c712019-02-20 22:20:28 -08002373 page->s_mem = addr + colour_off;
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002374 page->active = 0;
2375
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002376 if (OBJFREELIST_SLAB(cachep))
2377 freelist = NULL;
2378 else if (OFF_SLAB(cachep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379 /* Slab management obj is off-slab. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002380 freelist = kmem_cache_alloc_node(cachep->freelist_cache,
Pekka Enberg8759ec52008-11-26 10:01:31 +02002381 local_flags, nodeid);
Andrey Konovalov8607c712019-02-20 22:20:28 -08002382 freelist = kasan_reset_tag(freelist);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002383 if (!freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 return NULL;
2385 } else {
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002386 /* We will use last bytes at the slab for freelist */
2387 freelist = addr + (PAGE_SIZE << cachep->gfporder) -
2388 cachep->freelist_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 }
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002390
Joonsoo Kim8456a642013-10-24 10:07:49 +09002391 return freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392}
2393
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002394static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002396 return ((freelist_idx_t *)page->freelist)[idx];
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002397}
2398
2399static inline void set_free_obj(struct page *page,
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002400 unsigned int idx, freelist_idx_t val)
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002401{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002402 ((freelist_idx_t *)(page->freelist))[idx] = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403}
2404
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002405static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406{
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002407#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 int i;
2409
2410 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09002411 void *objp = index_to_obj(cachep, page, i);
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002412
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 if (cachep->flags & SLAB_STORE_USER)
2414 *dbg_userword(cachep, objp) = NULL;
2415
2416 if (cachep->flags & SLAB_RED_ZONE) {
2417 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2418 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2419 }
2420 /*
Andrew Mortona737b3e2006-03-22 00:08:11 -08002421 * Constructors are not allowed to allocate memory from the same
2422 * cache which they are a constructor for. Otherwise, deadlock.
2423 * They must also be threaded.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002425 if (cachep->ctor && !(cachep->flags & SLAB_POISON)) {
2426 kasan_unpoison_object_data(cachep,
2427 objp + obj_offset(cachep));
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002428 cachep->ctor(objp + obj_offset(cachep));
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002429 kasan_poison_object_data(
2430 cachep, objp + obj_offset(cachep));
2431 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432
2433 if (cachep->flags & SLAB_RED_ZONE) {
2434 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002435 slab_error(cachep, "constructor overwrote the end of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002437 slab_error(cachep, "constructor overwrote the start of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 }
Joonsoo Kim40b44132016-03-15 14:54:21 -07002439 /* need to poison the objs? */
2440 if (cachep->flags & SLAB_POISON) {
2441 poison_obj(cachep, objp, POISON_FREE);
2442 slab_kernel_map(cachep, objp, 0, 0);
2443 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002444 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445#endif
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002446}
2447
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002448#ifdef CONFIG_SLAB_FREELIST_RANDOM
2449/* Hold information during a freelist initialization */
2450union freelist_init_state {
2451 struct {
2452 unsigned int pos;
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002453 unsigned int *list;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002454 unsigned int count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002455 };
2456 struct rnd_state rnd_state;
2457};
2458
2459/*
2460 * Initialize the state based on the randomization methode available.
2461 * return true if the pre-computed list is available, false otherwize.
2462 */
2463static bool freelist_state_initialize(union freelist_init_state *state,
2464 struct kmem_cache *cachep,
2465 unsigned int count)
2466{
2467 bool ret;
2468 unsigned int rand;
2469
2470 /* Use best entropy available to define a random shift */
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002471 rand = get_random_int();
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002472
2473 /* Use a random state if the pre-computed list is not available */
2474 if (!cachep->random_seq) {
2475 prandom_seed_state(&state->rnd_state, rand);
2476 ret = false;
2477 } else {
2478 state->list = cachep->random_seq;
2479 state->count = count;
John Sperbeckc4e490c2017-01-10 16:58:24 -08002480 state->pos = rand % count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002481 ret = true;
2482 }
2483 return ret;
2484}
2485
2486/* Get the next entry on the list and randomize it using a random shift */
2487static freelist_idx_t next_random_slot(union freelist_init_state *state)
2488{
John Sperbeckc4e490c2017-01-10 16:58:24 -08002489 if (state->pos >= state->count)
2490 state->pos = 0;
2491 return state->list[state->pos++];
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002492}
2493
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002494/* Swap two freelist entries */
2495static void swap_free_obj(struct page *page, unsigned int a, unsigned int b)
2496{
2497 swap(((freelist_idx_t *)page->freelist)[a],
2498 ((freelist_idx_t *)page->freelist)[b]);
2499}
2500
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002501/*
2502 * Shuffle the freelist initialization state based on pre-computed lists.
2503 * return true if the list was successfully shuffled, false otherwise.
2504 */
2505static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page)
2506{
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002507 unsigned int objfreelist = 0, i, rand, count = cachep->num;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002508 union freelist_init_state state;
2509 bool precomputed;
2510
2511 if (count < 2)
2512 return false;
2513
2514 precomputed = freelist_state_initialize(&state, cachep, count);
2515
2516 /* Take a random entry as the objfreelist */
2517 if (OBJFREELIST_SLAB(cachep)) {
2518 if (!precomputed)
2519 objfreelist = count - 1;
2520 else
2521 objfreelist = next_random_slot(&state);
2522 page->freelist = index_to_obj(cachep, page, objfreelist) +
2523 obj_offset(cachep);
2524 count--;
2525 }
2526
2527 /*
2528 * On early boot, generate the list dynamically.
2529 * Later use a pre-computed list for speed.
2530 */
2531 if (!precomputed) {
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002532 for (i = 0; i < count; i++)
2533 set_free_obj(page, i, i);
2534
2535 /* Fisher-Yates shuffle */
2536 for (i = count - 1; i > 0; i--) {
2537 rand = prandom_u32_state(&state.rnd_state);
2538 rand %= (i + 1);
2539 swap_free_obj(page, i, rand);
2540 }
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002541 } else {
2542 for (i = 0; i < count; i++)
2543 set_free_obj(page, i, next_random_slot(&state));
2544 }
2545
2546 if (OBJFREELIST_SLAB(cachep))
2547 set_free_obj(page, cachep->num - 1, objfreelist);
2548
2549 return true;
2550}
2551#else
2552static inline bool shuffle_freelist(struct kmem_cache *cachep,
2553 struct page *page)
2554{
2555 return false;
2556}
2557#endif /* CONFIG_SLAB_FREELIST_RANDOM */
2558
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002559static void cache_init_objs(struct kmem_cache *cachep,
2560 struct page *page)
2561{
2562 int i;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002563 void *objp;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002564 bool shuffled;
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002565
2566 cache_init_objs_debug(cachep, page);
2567
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002568 /* Try to randomize the freelist if enabled */
2569 shuffled = shuffle_freelist(cachep, page);
2570
2571 if (!shuffled && OBJFREELIST_SLAB(cachep)) {
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002572 page->freelist = index_to_obj(cachep, page, cachep->num - 1) +
2573 obj_offset(cachep);
2574 }
2575
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002576 for (i = 0; i < cachep->num; i++) {
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07002577 objp = index_to_obj(cachep, page, i);
Andrey Konovalov99009492018-12-28 00:30:23 -08002578 objp = kasan_init_slab_obj(cachep, objp);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07002579
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002580 /* constructor could break poison info */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002581 if (DEBUG == 0 && cachep->ctor) {
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002582 kasan_unpoison_object_data(cachep, objp);
2583 cachep->ctor(objp);
2584 kasan_poison_object_data(cachep, objp);
2585 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002586
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002587 if (!shuffled)
2588 set_free_obj(page, i, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590}
2591
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002592static void *slab_get_obj(struct kmem_cache *cachep, struct page *page)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002593{
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002594 void *objp;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002595
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002596 objp = index_to_obj(cachep, page, get_free_obj(page, page->active));
Joonsoo Kim8456a642013-10-24 10:07:49 +09002597 page->active++;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002598
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002599#if DEBUG
2600 if (cachep->flags & SLAB_STORE_USER)
2601 set_store_user_dirty(cachep);
2602#endif
2603
Matthew Dobson78d382d2006-02-01 03:05:47 -08002604 return objp;
2605}
2606
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002607static void slab_put_obj(struct kmem_cache *cachep,
2608 struct page *page, void *objp)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002609{
Joonsoo Kim8456a642013-10-24 10:07:49 +09002610 unsigned int objnr = obj_to_index(cachep, page, objp);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002611#if DEBUG
Joonsoo Kim16025172013-10-24 10:07:46 +09002612 unsigned int i;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002613
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002614 /* Verify double free bug */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002615 for (i = page->active; i < cachep->num; i++) {
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002616 if (get_free_obj(page, i) == objnr) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002617 pr_err("slab: double free detected in cache '%s', objp %px\n",
Joe Perches756a0252016-03-17 14:19:47 -07002618 cachep->name, objp);
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002619 BUG();
2620 }
Matthew Dobson78d382d2006-02-01 03:05:47 -08002621 }
2622#endif
Joonsoo Kim8456a642013-10-24 10:07:49 +09002623 page->active--;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002624 if (!page->freelist)
2625 page->freelist = objp + obj_offset(cachep);
2626
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002627 set_free_obj(page, page->active, objnr);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002628}
2629
Pekka Enberg47768742006-06-23 02:03:07 -07002630/*
2631 * Map pages beginning at addr to the given cache and slab. This is required
2632 * for the slab allocator to be able to lookup the cache and slab of a
Nick Pigginccd35fb2011-01-07 17:49:17 +11002633 * virtual address for kfree, ksize, and slab debugging.
Pekka Enberg47768742006-06-23 02:03:07 -07002634 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002635static void slab_map_pages(struct kmem_cache *cache, struct page *page,
Joonsoo Kim7e007352013-10-30 19:04:01 +09002636 void *freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002637{
Joonsoo Kima57a4982013-10-24 10:07:44 +09002638 page->slab_cache = cache;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002639 page->freelist = freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640}
2641
2642/*
2643 * Grow (by 1) the number of slabs within a cache. This is called by
2644 * kmem_cache_alloc() when there are no active objs left in a cache.
2645 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002646static struct page *cache_grow_begin(struct kmem_cache *cachep,
2647 gfp_t flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002649 void *freelist;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002650 size_t offset;
2651 gfp_t local_flags;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002652 int page_node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002653 struct kmem_cache_node *n;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002654 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
Andrew Mortona737b3e2006-03-22 00:08:11 -08002656 /*
2657 * Be lazy and only check for valid flags here, keeping it out of the
2658 * critical path in kmem_cache_alloc().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 */
Andrew Mortonc871ac42014-12-10 15:42:25 -08002660 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07002661 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07002662 flags &= ~GFP_SLAB_BUG_MASK;
2663 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
2664 invalid_mask, &invalid_mask, flags, &flags);
2665 dump_stack();
Andrew Mortonc871ac42014-12-10 15:42:25 -08002666 }
Matthew Wilcox128227e2018-06-07 17:05:13 -07002667 WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
Christoph Lameter6cb06222007-10-16 01:25:41 -07002668 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 check_irq_off();
Mel Gormand0164ad2015-11-06 16:28:21 -08002671 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 local_irq_enable();
2673
2674 /*
Andrew Mortona737b3e2006-03-22 00:08:11 -08002675 * Get mem for the objs. Attempt to allocate a physical page from
2676 * 'nodeid'.
Christoph Lametere498be72005-09-09 13:03:32 -07002677 */
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002678 page = kmem_getpages(cachep, local_flags, nodeid);
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002679 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 goto failed;
2681
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002682 page_node = page_to_nid(page);
2683 n = get_node(cachep, page_node);
Joonsoo Kim03d1d432016-05-19 17:10:20 -07002684
2685 /* Get colour for the slab, and cal the next value. */
2686 n->colour_next++;
2687 if (n->colour_next >= cachep->colour)
2688 n->colour_next = 0;
2689
2690 offset = n->colour_next;
2691 if (offset >= cachep->colour)
2692 offset = 0;
2693
2694 offset *= cachep->colour_off;
2695
Andrey Konovalov8607c712019-02-20 22:20:28 -08002696 /*
2697 * Call kasan_poison_slab() before calling alloc_slabmgmt(), so
2698 * page_address() in the latter returns a non-tagged pointer,
2699 * as it should be for slab pages.
2700 */
2701 kasan_poison_slab(page);
2702
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703 /* Get slab management. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002704 freelist = alloc_slabmgmt(cachep, page, offset,
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002705 local_flags & ~GFP_CONSTRAINT_MASK, page_node);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002706 if (OFF_SLAB(cachep) && !freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 goto opps1;
2708
Joonsoo Kim8456a642013-10-24 10:07:49 +09002709 slab_map_pages(cachep, page, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
Joonsoo Kim8456a642013-10-24 10:07:49 +09002711 cache_init_objs(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
Mel Gormand0164ad2015-11-06 16:28:21 -08002713 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002714 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002716 return page;
2717
Andrew Mortona737b3e2006-03-22 00:08:11 -08002718opps1:
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002719 kmem_freepages(cachep, page);
Andrew Mortona737b3e2006-03-22 00:08:11 -08002720failed:
Mel Gormand0164ad2015-11-06 16:28:21 -08002721 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 local_irq_disable();
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002723 return NULL;
2724}
2725
2726static void cache_grow_end(struct kmem_cache *cachep, struct page *page)
2727{
2728 struct kmem_cache_node *n;
2729 void *list = NULL;
2730
2731 check_irq_off();
2732
2733 if (!page)
2734 return;
2735
2736 INIT_LIST_HEAD(&page->lru);
2737 n = get_node(cachep, page_to_nid(page));
2738
2739 spin_lock(&n->list_lock);
David Rientjesbf00bd32016-12-12 16:41:44 -08002740 n->total_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002741 if (!page->active) {
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002742 list_add_tail(&page->lru, &(n->slabs_free));
Greg Thelenf728b0a2016-12-12 16:41:41 -08002743 n->free_slabs++;
David Rientjesbf00bd32016-12-12 16:41:44 -08002744 } else
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002745 fixup_slab_list(cachep, n, page, &list);
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07002746
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002747 STATS_INC_GROWN(cachep);
2748 n->free_objects += cachep->num - page->active;
2749 spin_unlock(&n->list_lock);
2750
2751 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752}
2753
2754#if DEBUG
2755
2756/*
2757 * Perform extra freeing checks:
2758 * - detect bad pointers.
2759 * - POISON/RED_ZONE checking
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 */
2761static void kfree_debugcheck(const void *objp)
2762{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 if (!virt_addr_valid(objp)) {
Joe Perches11705322016-03-17 14:19:50 -07002764 pr_err("kfree_debugcheck: out of range ptr %lxh\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002765 (unsigned long)objp);
2766 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768}
2769
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002770static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
2771{
David Woodhouseb46b8f12007-05-08 00:22:59 -07002772 unsigned long long redzone1, redzone2;
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002773
2774 redzone1 = *dbg_redzone1(cache, obj);
2775 redzone2 = *dbg_redzone2(cache, obj);
2776
2777 /*
2778 * Redzone is ok.
2779 */
2780 if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
2781 return;
2782
2783 if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
2784 slab_error(cache, "double free detected");
2785 else
2786 slab_error(cache, "memory outside object was overwritten");
2787
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002788 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07002789 obj, redzone1, redzone2);
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002790}
2791
Pekka Enberg343e0d72006-02-01 03:05:50 -08002792static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002793 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 unsigned int objnr;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002796 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Matthew Wilcox80cbd912007-11-29 12:05:13 -07002798 BUG_ON(virt_to_cache(objp) != cachep);
2799
Manfred Spraul3dafccf2006-02-01 03:05:42 -08002800 objp -= obj_offset(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 kfree_debugcheck(objp);
Christoph Lameterb49af682007-05-06 14:49:41 -07002802 page = virt_to_head_page(objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 if (cachep->flags & SLAB_RED_ZONE) {
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002805 verify_redzone_free(cachep, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2807 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2808 }
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002809 if (cachep->flags & SLAB_STORE_USER) {
2810 set_store_user_dirty(cachep);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002811 *dbg_userword(cachep, objp) = (void *)caller;
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813
Joonsoo Kim8456a642013-10-24 10:07:49 +09002814 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816 BUG_ON(objnr >= cachep->num);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002817 BUG_ON(objp != index_to_obj(cachep, page, objnr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 poison_obj(cachep, objp, POISON_FREE);
Joonsoo Kim40b44132016-03-15 14:54:21 -07002821 slab_kernel_map(cachep, objp, 0, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 }
2823 return objp;
2824}
2825
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826#else
2827#define kfree_debugcheck(x) do { } while(0)
2828#define cache_free_debugcheck(x,objp,z) (objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829#endif
2830
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002831static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
2832 void **list)
2833{
2834#if DEBUG
2835 void *next = *list;
2836 void *objp;
2837
2838 while (next) {
2839 objp = next - obj_offset(cachep);
2840 next = *(void **)next;
2841 poison_obj(cachep, objp, POISON_FREE);
2842 }
2843#endif
2844}
2845
Joonsoo Kimd8410232016-03-15 14:54:44 -07002846static inline void fixup_slab_list(struct kmem_cache *cachep,
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002847 struct kmem_cache_node *n, struct page *page,
2848 void **list)
Joonsoo Kimd8410232016-03-15 14:54:44 -07002849{
2850 /* move slabp to correct slabp list: */
2851 list_del(&page->lru);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002852 if (page->active == cachep->num) {
Joonsoo Kimd8410232016-03-15 14:54:44 -07002853 list_add(&page->lru, &n->slabs_full);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002854 if (OBJFREELIST_SLAB(cachep)) {
2855#if DEBUG
2856 /* Poisoning will be done without holding the lock */
2857 if (cachep->flags & SLAB_POISON) {
2858 void **objp = page->freelist;
2859
2860 *objp = *list;
2861 *list = objp;
2862 }
2863#endif
2864 page->freelist = NULL;
2865 }
2866 } else
Joonsoo Kimd8410232016-03-15 14:54:44 -07002867 list_add(&page->lru, &n->slabs_partial);
2868}
2869
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002870/* Try to find non-pfmemalloc slab if needed */
2871static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n,
David Rientjesbf00bd32016-12-12 16:41:44 -08002872 struct page *page, bool pfmemalloc)
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002873{
2874 if (!page)
2875 return NULL;
2876
2877 if (pfmemalloc)
2878 return page;
2879
2880 if (!PageSlabPfmemalloc(page))
2881 return page;
2882
2883 /* No need to keep pfmemalloc slab if we have enough free objects */
2884 if (n->free_objects > n->free_limit) {
2885 ClearPageSlabPfmemalloc(page);
2886 return page;
2887 }
2888
2889 /* Move pfmemalloc slab to the end of list to speed up next search */
2890 list_del(&page->lru);
David Rientjesbf00bd32016-12-12 16:41:44 -08002891 if (!page->active) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002892 list_add_tail(&page->lru, &n->slabs_free);
David Rientjesbf00bd32016-12-12 16:41:44 -08002893 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002894 } else
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002895 list_add_tail(&page->lru, &n->slabs_partial);
2896
2897 list_for_each_entry(page, &n->slabs_partial, lru) {
2898 if (!PageSlabPfmemalloc(page))
2899 return page;
2900 }
2901
Greg Thelenf728b0a2016-12-12 16:41:41 -08002902 n->free_touched = 1;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002903 list_for_each_entry(page, &n->slabs_free, lru) {
Greg Thelenf728b0a2016-12-12 16:41:41 -08002904 if (!PageSlabPfmemalloc(page)) {
David Rientjesbf00bd32016-12-12 16:41:44 -08002905 n->free_slabs--;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002906 return page;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002907 }
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002908 }
2909
2910 return NULL;
2911}
2912
2913static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc)
Geliang Tang7aa0d222016-01-14 15:18:02 -08002914{
2915 struct page *page;
2916
Greg Thelenf728b0a2016-12-12 16:41:41 -08002917 assert_spin_locked(&n->list_lock);
David Rientjesbf00bd32016-12-12 16:41:44 -08002918 page = list_first_entry_or_null(&n->slabs_partial, struct page, lru);
Geliang Tang7aa0d222016-01-14 15:18:02 -08002919 if (!page) {
2920 n->free_touched = 1;
David Rientjesbf00bd32016-12-12 16:41:44 -08002921 page = list_first_entry_or_null(&n->slabs_free, struct page,
2922 lru);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002923 if (page)
David Rientjesbf00bd32016-12-12 16:41:44 -08002924 n->free_slabs--;
Geliang Tang7aa0d222016-01-14 15:18:02 -08002925 }
2926
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002927 if (sk_memalloc_socks())
David Rientjesbf00bd32016-12-12 16:41:44 -08002928 page = get_valid_first_slab(n, page, pfmemalloc);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002929
Geliang Tang7aa0d222016-01-14 15:18:02 -08002930 return page;
2931}
2932
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002933static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep,
2934 struct kmem_cache_node *n, gfp_t flags)
2935{
2936 struct page *page;
2937 void *obj;
2938 void *list = NULL;
2939
2940 if (!gfp_pfmemalloc_allowed(flags))
2941 return NULL;
2942
2943 spin_lock(&n->list_lock);
2944 page = get_first_slab(n, true);
2945 if (!page) {
2946 spin_unlock(&n->list_lock);
2947 return NULL;
2948 }
2949
2950 obj = slab_get_obj(cachep, page);
2951 n->free_objects--;
2952
2953 fixup_slab_list(cachep, n, page, &list);
2954
2955 spin_unlock(&n->list_lock);
2956 fixup_objfreelist_debug(cachep, &list);
2957
2958 return obj;
2959}
2960
Joonsoo Kim213b4692016-05-19 17:10:29 -07002961/*
2962 * Slab list should be fixed up by fixup_slab_list() for existing slab
2963 * or cache_grow_end() for new slab
2964 */
2965static __always_inline int alloc_block(struct kmem_cache *cachep,
2966 struct array_cache *ac, struct page *page, int batchcount)
2967{
2968 /*
2969 * There must be at least one object available for
2970 * allocation.
2971 */
2972 BUG_ON(page->active >= cachep->num);
2973
2974 while (page->active < cachep->num && batchcount--) {
2975 STATS_INC_ALLOCED(cachep);
2976 STATS_INC_ACTIVE(cachep);
2977 STATS_SET_HIGH(cachep);
2978
2979 ac->entry[ac->avail++] = slab_get_obj(cachep, page);
2980 }
2981
2982 return batchcount;
2983}
2984
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002985static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986{
2987 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002988 struct kmem_cache_node *n;
Joonsoo Kim801faf02016-05-19 17:10:31 -07002989 struct array_cache *ac, *shared;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002990 int node;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002991 void *list = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002992 struct page *page;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002993
Joe Korty6d2144d2008-03-05 15:04:59 -08002994 check_irq_off();
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002995 node = numa_mem_id();
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002996
Joe Korty6d2144d2008-03-05 15:04:59 -08002997 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 batchcount = ac->batchcount;
2999 if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08003000 /*
3001 * If there was little recent activity on this cache, then
3002 * perform only a partial refill. Otherwise we could generate
3003 * refill bouncing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 */
3005 batchcount = BATCHREFILL_LIMIT;
3006 }
Christoph Lameter18bf8542014-08-06 16:04:11 -07003007 n = get_node(cachep, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003009 BUG_ON(ac->avail > 0 || !n);
Joonsoo Kim801faf02016-05-19 17:10:31 -07003010 shared = READ_ONCE(n->shared);
3011 if (!n->free_objects && (!shared || !shared->avail))
3012 goto direct_grow;
3013
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003014 spin_lock(&n->list_lock);
Joonsoo Kim801faf02016-05-19 17:10:31 -07003015 shared = READ_ONCE(n->shared);
Christoph Lametere498be72005-09-09 13:03:32 -07003016
Christoph Lameter3ded1752006-03-25 03:06:44 -08003017 /* See if we can refill from the shared array */
Joonsoo Kim801faf02016-05-19 17:10:31 -07003018 if (shared && transfer_objects(ac, shared, batchcount)) {
3019 shared->touched = 1;
Christoph Lameter3ded1752006-03-25 03:06:44 -08003020 goto alloc_done;
Nick Piggin44b57f12010-01-27 22:27:40 +11003021 }
Christoph Lameter3ded1752006-03-25 03:06:44 -08003022
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 while (batchcount > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 /* Get slab alloc is to come from. */
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003025 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08003026 if (!page)
3027 goto must_grow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029 check_spinlock_acquired(cachep);
Pekka Enberg714b81712007-05-06 14:49:03 -07003030
Joonsoo Kim213b4692016-05-19 17:10:29 -07003031 batchcount = alloc_block(cachep, ac, page, batchcount);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003032 fixup_slab_list(cachep, n, page, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 }
3034
Andrew Mortona737b3e2006-03-22 00:08:11 -08003035must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003036 n->free_objects -= ac->avail;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003037alloc_done:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003038 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003039 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040
Joonsoo Kim801faf02016-05-19 17:10:31 -07003041direct_grow:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 if (unlikely(!ac->avail)) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003043 /* Check if we can use obj in pfmemalloc slab */
3044 if (sk_memalloc_socks()) {
3045 void *obj = cache_alloc_pfmemalloc(cachep, n, flags);
3046
3047 if (obj)
3048 return obj;
3049 }
3050
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003051 page = cache_grow_begin(cachep, gfp_exact_node(flags), node);
Christoph Lametere498be72005-09-09 13:03:32 -07003052
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003053 /*
3054 * cache_grow_begin() can reenable interrupts,
3055 * then ac could change.
3056 */
Pekka Enberg9a2dba42006-02-01 03:05:49 -08003057 ac = cpu_cache_get(cachep);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003058 if (!ac->avail && page)
3059 alloc_block(cachep, ac, page, batchcount);
3060 cache_grow_end(cachep, page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003061
Joonsoo Kim213b4692016-05-19 17:10:29 -07003062 if (!ac->avail)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 }
3065 ac->touched = 1;
Mel Gorman072bb0a2012-07-31 16:43:58 -07003066
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003067 return ac->entry[--ac->avail];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068}
3069
Andrew Mortona737b3e2006-03-22 00:08:11 -08003070static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
3071 gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072{
Mel Gormand0164ad2015-11-06 16:28:21 -08003073 might_sleep_if(gfpflags_allow_blocking(flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074}
3075
3076#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08003077static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003078 gfp_t flags, void *objp, unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079{
Matthew Wilcox128227e2018-06-07 17:05:13 -07003080 WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003081 if (!objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082 return objp;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003083 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 check_poison_obj(cachep, objp);
Joonsoo Kim40b44132016-03-15 14:54:21 -07003085 slab_kernel_map(cachep, objp, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 poison_obj(cachep, objp, POISON_INUSE);
3087 }
3088 if (cachep->flags & SLAB_STORE_USER)
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003089 *dbg_userword(cachep, objp) = (void *)caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090
3091 if (cachep->flags & SLAB_RED_ZONE) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08003092 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
3093 *dbg_redzone2(cachep, objp) != RED_INACTIVE) {
Joe Perches756a0252016-03-17 14:19:47 -07003094 slab_error(cachep, "double free, or memory outside object was overwritten");
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003095 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07003096 objp, *dbg_redzone1(cachep, objp),
3097 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 }
3099 *dbg_redzone1(cachep, objp) = RED_ACTIVE;
3100 *dbg_redzone2(cachep, objp) = RED_ACTIVE;
3101 }
Joonsoo Kim03787302014-06-23 13:22:06 -07003102
Manfred Spraul3dafccf2006-02-01 03:05:42 -08003103 objp += obj_offset(cachep);
Christoph Lameter4f104932007-05-06 14:50:17 -07003104 if (cachep->ctor && cachep->flags & SLAB_POISON)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003105 cachep->ctor(objp);
Tetsuo Handa7ea466f2011-07-21 09:42:45 +09003106 if (ARCH_SLAB_MINALIGN &&
3107 ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003108 pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
Hugh Dickinsc2251502011-07-11 13:35:08 -07003109 objp, (int)ARCH_SLAB_MINALIGN);
Kevin Hilmana44b56d2006-12-06 20:32:11 -08003110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 return objp;
3112}
3113#else
3114#define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
3115#endif
3116
Pekka Enberg343e0d72006-02-01 03:05:50 -08003117static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003119 void *objp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 struct array_cache *ac;
3121
Alok N Kataria5c382302005-09-27 21:45:46 -07003122 check_irq_off();
Akinobu Mita8a8b6502006-12-08 02:39:44 -08003123
Pekka Enberg9a2dba42006-02-01 03:05:49 -08003124 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 if (likely(ac->avail)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 ac->touched = 1;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003127 objp = ac->entry[--ac->avail];
Mel Gorman072bb0a2012-07-31 16:43:58 -07003128
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003129 STATS_INC_ALLOCHIT(cachep);
3130 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 }
Mel Gorman072bb0a2012-07-31 16:43:58 -07003132
3133 STATS_INC_ALLOCMISS(cachep);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003134 objp = cache_alloc_refill(cachep, flags);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003135 /*
3136 * the 'ac' may be updated by cache_alloc_refill(),
3137 * and kmemleak_erase() requires its correct value.
3138 */
3139 ac = cpu_cache_get(cachep);
3140
3141out:
Catalin Marinasd5cff632009-06-11 13:22:40 +01003142 /*
3143 * To avoid a false negative, if an object that is in one of the
3144 * per-CPU caches is leaked, we need to make sure kmemleak doesn't
3145 * treat the array pointers as a reference to the object.
3146 */
J. R. Okajimaf3d8b532009-12-02 16:55:49 +09003147 if (objp)
3148 kmemleak_erase(&ac->entry[ac->avail]);
Alok N Kataria5c382302005-09-27 21:45:46 -07003149 return objp;
3150}
3151
Christoph Lametere498be72005-09-09 13:03:32 -07003152#ifdef CONFIG_NUMA
3153/*
Zefan Li2ad654b2014-09-25 09:41:02 +08003154 * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set.
Paul Jacksonc61afb12006-03-24 03:16:08 -08003155 *
3156 * If we are in_interrupt, then process context, including cpusets and
3157 * mempolicy, may not apply and should not be used for allocation policy.
3158 */
3159static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
3160{
3161 int nid_alloc, nid_here;
3162
Christoph Lameter765c4502006-09-27 01:50:08 -07003163 if (in_interrupt() || (flags & __GFP_THISNODE))
Paul Jacksonc61afb12006-03-24 03:16:08 -08003164 return NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003165 nid_alloc = nid_here = numa_mem_id();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003166 if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
Jack Steiner6adef3e2010-05-26 14:42:49 -07003167 nid_alloc = cpuset_slab_spread_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003168 else if (current->mempolicy)
David Rientjes2a389612014-04-07 15:37:29 -07003169 nid_alloc = mempolicy_slab_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003170 if (nid_alloc != nid_here)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003171 return ____cache_alloc_node(cachep, flags, nid_alloc);
Paul Jacksonc61afb12006-03-24 03:16:08 -08003172 return NULL;
3173}
3174
3175/*
Christoph Lameter765c4502006-09-27 01:50:08 -07003176 * Fallback function if there was no memory available and no objects on a
Christoph Lameter3c517a62006-12-06 20:33:29 -08003177 * certain node and fall back is permitted. First we scan all the
Christoph Lameter6a673682013-01-10 19:14:19 +00003178 * available node for available objects. If that fails then we
Christoph Lameter3c517a62006-12-06 20:33:29 -08003179 * perform an allocation without specifying a node. This allows the page
3180 * allocator to do its reclaim / fallback magic. We then insert the
3181 * slab into the proper nodelist and then allocate from it.
Christoph Lameter765c4502006-09-27 01:50:08 -07003182 */
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003183static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
Christoph Lameter765c4502006-09-27 01:50:08 -07003184{
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003185 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07003186 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003187 struct zone *zone;
3188 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter765c4502006-09-27 01:50:08 -07003189 void *obj = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003190 struct page *page;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003191 int nid;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003192 unsigned int cpuset_mems_cookie;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003193
3194 if (flags & __GFP_THISNODE)
3195 return NULL;
3196
Mel Gormancc9a6c82012-03-21 16:34:11 -07003197retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003198 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07003199 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003200
Christoph Lameter3c517a62006-12-06 20:33:29 -08003201retry:
3202 /*
3203 * Look through allowed nodes for objects available
3204 * from existing per node queues.
3205 */
Mel Gorman54a6eb52008-04-28 02:12:16 -07003206 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
3207 nid = zone_to_nid(zone);
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003208
Vladimir Davydov061d7072014-12-12 16:58:25 -08003209 if (cpuset_zone_allowed(zone, flags) &&
Christoph Lameter18bf8542014-08-06 16:04:11 -07003210 get_node(cache, nid) &&
3211 get_node(cache, nid)->free_objects) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003212 obj = ____cache_alloc_node(cache,
David Rientjes4167e9b2015-04-14 15:46:55 -07003213 gfp_exact_node(flags), nid);
Christoph Lameter481c5342008-06-21 16:46:35 -07003214 if (obj)
3215 break;
3216 }
Christoph Lameter3c517a62006-12-06 20:33:29 -08003217 }
3218
Christoph Lametercfce6602007-05-06 14:50:17 -07003219 if (!obj) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003220 /*
3221 * This allocation will be performed within the constraints
3222 * of the current cpuset / memory policy requirements.
3223 * We may trigger various forms of reclaim on the allowed
3224 * set and go into memory reserves if necessary.
3225 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003226 page = cache_grow_begin(cache, flags, numa_mem_id());
3227 cache_grow_end(cache, page);
3228 if (page) {
3229 nid = page_to_nid(page);
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003230 obj = ____cache_alloc_node(cache,
3231 gfp_exact_node(flags), nid);
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09003232
Christoph Lameter3c517a62006-12-06 20:33:29 -08003233 /*
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003234 * Another processor may allocate the objects in
3235 * the slab since we are not holding any locks.
Christoph Lameter3c517a62006-12-06 20:33:29 -08003236 */
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003237 if (!obj)
3238 goto retry;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003239 }
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003240 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003241
Mel Gormand26914d2014-04-03 14:47:24 -07003242 if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003243 goto retry_cpuset;
Christoph Lameter765c4502006-09-27 01:50:08 -07003244 return obj;
3245}
3246
3247/*
Christoph Lametere498be72005-09-09 13:03:32 -07003248 * A interface to enable slab creation on nodeid
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003250static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
Andrew Mortona737b3e2006-03-22 00:08:11 -08003251 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07003252{
Joonsoo Kim8456a642013-10-24 10:07:49 +09003253 struct page *page;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003254 struct kmem_cache_node *n;
Joonsoo Kim213b4692016-05-19 17:10:29 -07003255 void *obj = NULL;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003256 void *list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Paul Mackerras7c3fbbd2014-12-02 15:59:48 -08003258 VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003259 n = get_node(cachep, nodeid);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003260 BUG_ON(!n);
Christoph Lametere498be72005-09-09 13:03:32 -07003261
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08003262 check_irq_off();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003263 spin_lock(&n->list_lock);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003264 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08003265 if (!page)
3266 goto must_grow;
Christoph Lametere498be72005-09-09 13:03:32 -07003267
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003268 check_spinlock_acquired_node(cachep, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07003269
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003270 STATS_INC_NODEALLOCS(cachep);
3271 STATS_INC_ACTIVE(cachep);
3272 STATS_SET_HIGH(cachep);
Christoph Lametere498be72005-09-09 13:03:32 -07003273
Joonsoo Kim8456a642013-10-24 10:07:49 +09003274 BUG_ON(page->active == cachep->num);
Christoph Lametere498be72005-09-09 13:03:32 -07003275
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003276 obj = slab_get_obj(cachep, page);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003277 n->free_objects--;
Christoph Lametere498be72005-09-09 13:03:32 -07003278
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003279 fixup_slab_list(cachep, n, page, &list);
Christoph Lametere498be72005-09-09 13:03:32 -07003280
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003281 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003282 fixup_objfreelist_debug(cachep, &list);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003283 return obj;
Christoph Lametere498be72005-09-09 13:03:32 -07003284
Andrew Mortona737b3e2006-03-22 00:08:11 -08003285must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003286 spin_unlock(&n->list_lock);
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003287 page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003288 if (page) {
3289 /* This slab isn't counted yet so don't update free_objects */
3290 obj = slab_get_obj(cachep, page);
3291 }
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003292 cache_grow_end(cachep, page);
Christoph Lametere498be72005-09-09 13:03:32 -07003293
Joonsoo Kim213b4692016-05-19 17:10:29 -07003294 return obj ? obj : fallback_alloc(cachep, flags);
Christoph Lametere498be72005-09-09 13:03:32 -07003295}
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003296
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003297static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003298slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003299 unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003300{
3301 unsigned long save_flags;
3302 void *ptr;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003303 int slab_node = numa_mem_id();
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003304
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003305 flags &= gfp_allowed_mask;
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003306 cachep = slab_pre_alloc_hook(cachep, flags);
3307 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003308 return NULL;
3309
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003310 cache_alloc_debugcheck_before(cachep, flags);
3311 local_irq_save(save_flags);
3312
Andrew Mortoneacbbae2011-07-28 13:59:49 -07003313 if (nodeid == NUMA_NO_NODE)
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003314 nodeid = slab_node;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003315
Christoph Lameter18bf8542014-08-06 16:04:11 -07003316 if (unlikely(!get_node(cachep, nodeid))) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003317 /* Node not bootstrapped yet */
3318 ptr = fallback_alloc(cachep, flags);
3319 goto out;
3320 }
3321
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003322 if (nodeid == slab_node) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003323 /*
3324 * Use the locally cached objects if possible.
3325 * However ____cache_alloc does not allow fallback
3326 * to other nodes. It may fail while we still have
3327 * objects on other nodes available.
3328 */
3329 ptr = ____cache_alloc(cachep, flags);
3330 if (ptr)
3331 goto out;
3332 }
3333 /* ___cache_alloc_node can fall back to other nodes */
3334 ptr = ____cache_alloc_node(cachep, flags, nodeid);
3335 out:
3336 local_irq_restore(save_flags);
3337 ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
3338
Alexander Potapenkoa5587d82019-07-11 20:59:19 -07003339 if (unlikely(slab_want_init_on_alloc(flags, cachep)) && ptr)
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003340 memset(ptr, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003341
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003342 slab_post_alloc_hook(cachep, flags, 1, &ptr);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003343 return ptr;
3344}
3345
3346static __always_inline void *
3347__do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
3348{
3349 void *objp;
3350
Zefan Li2ad654b2014-09-25 09:41:02 +08003351 if (current->mempolicy || cpuset_do_slab_mem_spread()) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003352 objp = alternate_node_alloc(cache, flags);
3353 if (objp)
3354 goto out;
3355 }
3356 objp = ____cache_alloc(cache, flags);
3357
3358 /*
3359 * We may just have run out of memory on the local node.
3360 * ____cache_alloc_node() knows how to locate memory on other nodes
3361 */
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003362 if (!objp)
3363 objp = ____cache_alloc_node(cache, flags, numa_mem_id());
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003364
3365 out:
3366 return objp;
3367}
3368#else
3369
3370static __always_inline void *
3371__do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3372{
3373 return ____cache_alloc(cachep, flags);
3374}
3375
3376#endif /* CONFIG_NUMA */
3377
3378static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003379slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003380{
3381 unsigned long save_flags;
3382 void *objp;
3383
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003384 flags &= gfp_allowed_mask;
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003385 cachep = slab_pre_alloc_hook(cachep, flags);
3386 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003387 return NULL;
3388
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003389 cache_alloc_debugcheck_before(cachep, flags);
3390 local_irq_save(save_flags);
3391 objp = __do_cache_alloc(cachep, flags);
3392 local_irq_restore(save_flags);
3393 objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
3394 prefetchw(objp);
3395
Alexander Potapenkoa5587d82019-07-11 20:59:19 -07003396 if (unlikely(slab_want_init_on_alloc(flags, cachep)) && objp)
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003397 memset(objp, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003398
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003399 slab_post_alloc_hook(cachep, flags, 1, &objp);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003400 return objp;
3401}
Christoph Lametere498be72005-09-09 13:03:32 -07003402
3403/*
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08003404 * Caller needs to acquire correct kmem_cache_node's list_lock
Joonsoo Kim97654df2014-08-06 16:04:25 -07003405 * @list: List of detached free slabs should be freed by caller
Christoph Lametere498be72005-09-09 13:03:32 -07003406 */
Joonsoo Kim97654df2014-08-06 16:04:25 -07003407static void free_block(struct kmem_cache *cachep, void **objpp,
3408 int nr_objects, int node, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409{
3410 int i;
Joonsoo Kim25c063f2014-08-06 16:04:22 -07003411 struct kmem_cache_node *n = get_node(cachep, node);
Joonsoo Kim6052b782016-05-19 17:10:17 -07003412 struct page *page;
3413
3414 n->free_objects += nr_objects;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415
3416 for (i = 0; i < nr_objects; i++) {
Mel Gorman072bb0a2012-07-31 16:43:58 -07003417 void *objp;
Joonsoo Kim8456a642013-10-24 10:07:49 +09003418 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
Mel Gorman072bb0a2012-07-31 16:43:58 -07003420 objp = objpp[i];
3421
Joonsoo Kim8456a642013-10-24 10:07:49 +09003422 page = virt_to_head_page(objp);
Joonsoo Kim8456a642013-10-24 10:07:49 +09003423 list_del(&page->lru);
Christoph Lameterff694162005-09-22 21:44:02 -07003424 check_spinlock_acquired_node(cachep, node);
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003425 slab_put_obj(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 STATS_DEC_ACTIVE(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
3428 /* fixup slab chains */
Greg Thelenf728b0a2016-12-12 16:41:41 -08003429 if (page->active == 0) {
Joonsoo Kim6052b782016-05-19 17:10:17 -07003430 list_add(&page->lru, &n->slabs_free);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003431 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08003432 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433 /* Unconditionally move a slab to the end of the
3434 * partial list on free - maximum time for the
3435 * other objects to be freed, too.
3436 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09003437 list_add_tail(&page->lru, &n->slabs_partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438 }
3439 }
Joonsoo Kim6052b782016-05-19 17:10:17 -07003440
3441 while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) {
3442 n->free_objects -= cachep->num;
3443
3444 page = list_last_entry(&n->slabs_free, struct page, lru);
Wei Yongjunde24bae2016-07-26 15:22:11 -07003445 list_move(&page->lru, list);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003446 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08003447 n->total_slabs--;
Joonsoo Kim6052b782016-05-19 17:10:17 -07003448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449}
3450
Pekka Enberg343e0d72006-02-01 03:05:50 -08003451static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452{
3453 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003454 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003455 int node = numa_mem_id();
Joonsoo Kim97654df2014-08-06 16:04:25 -07003456 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457
3458 batchcount = ac->batchcount;
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003459
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07003461 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003462 spin_lock(&n->list_lock);
3463 if (n->shared) {
3464 struct array_cache *shared_array = n->shared;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003465 int max = shared_array->limit - shared_array->avail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466 if (max) {
3467 if (batchcount > max)
3468 batchcount = max;
Christoph Lametere498be72005-09-09 13:03:32 -07003469 memcpy(&(shared_array->entry[shared_array->avail]),
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003470 ac->entry, sizeof(void *) * batchcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471 shared_array->avail += batchcount;
3472 goto free_done;
3473 }
3474 }
3475
Joonsoo Kim97654df2014-08-06 16:04:25 -07003476 free_block(cachep, ac->entry, batchcount, node, &list);
Andrew Mortona737b3e2006-03-22 00:08:11 -08003477free_done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478#if STATS
3479 {
3480 int i = 0;
Geliang Tang73c02192016-01-14 15:17:59 -08003481 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482
Geliang Tang73c02192016-01-14 15:17:59 -08003483 list_for_each_entry(page, &n->slabs_free, lru) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09003484 BUG_ON(page->active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003485
3486 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 }
3488 STATS_SET_FREEABLE(cachep, i);
3489 }
3490#endif
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003491 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07003492 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 ac->avail -= batchcount;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003494 memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495}
3496
3497/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08003498 * Release an obj back to its cache. If the obj has a constructed state, it must
3499 * be in this state _before_ it is released. Called with disabled ints.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003501static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp,
3502 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503{
Alexander Potapenko55834c52016-05-20 16:59:11 -07003504 /* Put the object into the quarantine, don't touch it for now. */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003505 if (kasan_slab_free(cachep, objp, _RET_IP_))
Alexander Potapenko55834c52016-05-20 16:59:11 -07003506 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003507
Alexander Potapenko55834c52016-05-20 16:59:11 -07003508 ___cache_free(cachep, objp, caller);
3509}
3510
3511void ___cache_free(struct kmem_cache *cachep, void *objp,
3512 unsigned long caller)
3513{
3514 struct array_cache *ac = cpu_cache_get(cachep);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003515
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516 check_irq_off();
Alexander Potapenkoa5587d82019-07-11 20:59:19 -07003517 if (unlikely(slab_want_init_on_free(cachep)))
3518 memset(objp, 0, cachep->object_size);
Catalin Marinasd5cff632009-06-11 13:22:40 +01003519 kmemleak_free_recursive(objp, cachep->flags);
Suleiman Souhlala947eb92011-06-02 00:16:42 -07003520 objp = cache_free_debugcheck(cachep, objp, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521
Siddha, Suresh B1807a1a2007-08-22 14:01:49 -07003522 /*
3523 * Skip calling cache_free_alien() when the platform is not numa.
3524 * This will avoid cache misses that happen while accessing slabp (which
3525 * is per page memory reference) to get nodeid. Instead use a global
3526 * variable to skip the call, which is mostly likely to be present in
3527 * the cache.
3528 */
Mel Gormanb6e68bc2009-06-16 15:32:16 -07003529 if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
Pekka Enberg729bd0b2006-06-23 02:03:05 -07003530 return;
Christoph Lametere498be72005-09-09 13:03:32 -07003531
Joonsoo Kim3d880192014-10-09 15:26:04 -07003532 if (ac->avail < ac->limit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003533 STATS_INC_FREEHIT(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003534 } else {
3535 STATS_INC_FREEMISS(cachep);
3536 cache_flusharray(cachep, ac);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 }
Zhao Jin42c8c992011-08-27 00:26:17 +08003538
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003539 if (sk_memalloc_socks()) {
3540 struct page *page = virt_to_head_page(objp);
3541
3542 if (unlikely(PageSlabPfmemalloc(page))) {
3543 cache_free_pfmemalloc(cachep, page, objp);
3544 return;
3545 }
3546 }
3547
3548 ac->entry[ac->avail++] = objp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003549}
3550
3551/**
3552 * kmem_cache_alloc - Allocate an object
3553 * @cachep: The cache to allocate from.
3554 * @flags: See kmalloc().
3555 *
3556 * Allocate an object from this cache. The flags are only relevant
3557 * if the cache has no available objects.
3558 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003559void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003561 void *ret = slab_alloc(cachep, flags, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003562
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003563 trace_kmem_cache_alloc(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003564 cachep->object_size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003565
3566 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567}
3568EXPORT_SYMBOL(kmem_cache_alloc);
3569
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003570static __always_inline void
3571cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags,
3572 size_t size, void **p, unsigned long caller)
3573{
3574 size_t i;
3575
3576 for (i = 0; i < size; i++)
3577 p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller);
3578}
3579
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003580int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003581 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003582{
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003583 size_t i;
3584
3585 s = slab_pre_alloc_hook(s, flags);
3586 if (!s)
3587 return 0;
3588
3589 cache_alloc_debugcheck_before(s, flags);
3590
3591 local_irq_disable();
3592 for (i = 0; i < size; i++) {
3593 void *objp = __do_cache_alloc(s, flags);
3594
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003595 if (unlikely(!objp))
3596 goto error;
3597 p[i] = objp;
3598 }
3599 local_irq_enable();
3600
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003601 cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_);
3602
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003603 /* Clear memory outside IRQ disabled section */
Alexander Potapenkoa5587d82019-07-11 20:59:19 -07003604 if (unlikely(slab_want_init_on_alloc(flags, s)))
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003605 for (i = 0; i < size; i++)
3606 memset(p[i], 0, s->object_size);
3607
3608 slab_post_alloc_hook(s, flags, size, p);
3609 /* FIXME: Trace call missing. Christoph would like a bulk variant */
3610 return size;
3611error:
3612 local_irq_enable();
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003613 cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_);
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003614 slab_post_alloc_hook(s, flags, i, p);
3615 __kmem_cache_free_bulk(s, i, p);
3616 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003617}
3618EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3619
Li Zefan0f24f122009-12-11 15:45:30 +08003620#ifdef CONFIG_TRACING
Steven Rostedt85beb582010-11-24 16:23:34 -05003621void *
Ezequiel Garcia40521472012-09-08 17:47:56 -03003622kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003623{
Steven Rostedt85beb582010-11-24 16:23:34 -05003624 void *ret;
3625
Ezequiel Garcia48356302012-09-08 17:47:57 -03003626 ret = slab_alloc(cachep, flags, _RET_IP_);
Steven Rostedt85beb582010-11-24 16:23:34 -05003627
Andrey Konovalov0697e142018-12-28 00:29:37 -08003628 ret = kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003629 trace_kmalloc(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003630 size, cachep->size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003631 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003632}
Steven Rostedt85beb582010-11-24 16:23:34 -05003633EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003634#endif
3635
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636#ifdef CONFIG_NUMA
Zhouping Liud0d04b72013-05-16 11:36:23 +08003637/**
3638 * kmem_cache_alloc_node - Allocate an object on the specified node
3639 * @cachep: The cache to allocate from.
3640 * @flags: See kmalloc().
3641 * @nodeid: node number of the target node.
3642 *
3643 * Identical to kmem_cache_alloc but it will allocate memory on the given
3644 * node, which can improve the performance for cpu bound structures.
3645 *
3646 * Fallback to other node is possible if __GFP_THISNODE is not set.
3647 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003648void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
3649{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003650 void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003651
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003652 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003653 cachep->object_size, cachep->size,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003654 flags, nodeid);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003655
3656 return ret;
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003657}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003658EXPORT_SYMBOL(kmem_cache_alloc_node);
3659
Li Zefan0f24f122009-12-11 15:45:30 +08003660#ifdef CONFIG_TRACING
Ezequiel Garcia40521472012-09-08 17:47:56 -03003661void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep,
Steven Rostedt85beb582010-11-24 16:23:34 -05003662 gfp_t flags,
Ezequiel Garcia40521472012-09-08 17:47:56 -03003663 int nodeid,
3664 size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003665{
Steven Rostedt85beb582010-11-24 16:23:34 -05003666 void *ret;
3667
Ezequiel Garcia592f4142012-09-25 08:07:08 -03003668 ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003669
Andrey Konovalov0697e142018-12-28 00:29:37 -08003670 ret = kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003671 trace_kmalloc_node(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003672 size, cachep->size,
Steven Rostedt85beb582010-11-24 16:23:34 -05003673 flags, nodeid);
3674 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003675}
Steven Rostedt85beb582010-11-24 16:23:34 -05003676EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003677#endif
3678
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003679static __always_inline void *
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003680__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003681{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003682 struct kmem_cache *cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003683 void *ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003684
Dmitry Vyukov3996e892018-10-26 15:03:12 -07003685 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3686 return NULL;
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003687 cachep = kmalloc_slab(size, flags);
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003688 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3689 return cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003690 ret = kmem_cache_alloc_node_trace(cachep, flags, node, size);
Andrey Konovalov0697e142018-12-28 00:29:37 -08003691 ret = kasan_kmalloc(cachep, ret, size, flags);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003692
3693 return ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003694}
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003695
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003696void *__kmalloc_node(size_t size, gfp_t flags, int node)
3697{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003698 return __do_kmalloc_node(size, flags, node, _RET_IP_);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003699}
Christoph Hellwigdbe5e692006-09-25 23:31:36 -07003700EXPORT_SYMBOL(__kmalloc_node);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003701
3702void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003703 int node, unsigned long caller)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003704{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003705 return __do_kmalloc_node(size, flags, node, caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003706}
3707EXPORT_SYMBOL(__kmalloc_node_track_caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003708#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709
3710/**
Paul Drynoff800590f2006-06-23 02:03:48 -07003711 * __do_kmalloc - allocate memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712 * @size: how many bytes of memory are required.
Paul Drynoff800590f2006-06-23 02:03:48 -07003713 * @flags: the type of memory to allocate (see kmalloc).
Randy Dunlap911851e2006-03-22 00:08:14 -08003714 * @caller: function caller for debug tracking of the caller
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715 */
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003716static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003717 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003719 struct kmem_cache *cachep;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003720 void *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003721
Dmitry Vyukov3996e892018-10-26 15:03:12 -07003722 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3723 return NULL;
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003724 cachep = kmalloc_slab(size, flags);
Linus Torvaldsa5c96d82007-07-19 13:17:15 -07003725 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3726 return cachep;
Ezequiel Garcia48356302012-09-08 17:47:57 -03003727 ret = slab_alloc(cachep, flags, caller);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003728
Andrey Konovalov0697e142018-12-28 00:29:37 -08003729 ret = kasan_kmalloc(cachep, ret, size, flags);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003730 trace_kmalloc(caller, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003731 size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003732
3733 return ret;
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003734}
3735
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003736void *__kmalloc(size_t size, gfp_t flags)
3737{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003738 return __do_kmalloc(size, flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739}
3740EXPORT_SYMBOL(__kmalloc);
3741
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003742void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003743{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003744 return __do_kmalloc(size, flags, caller);
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003745}
3746EXPORT_SYMBOL(__kmalloc_track_caller);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -07003747
Linus Torvalds1da177e2005-04-16 15:20:36 -07003748/**
3749 * kmem_cache_free - Deallocate an object
3750 * @cachep: The cache the allocation was from.
3751 * @objp: The previously allocated object.
3752 *
3753 * Free an object which was previously allocated from this
3754 * cache.
3755 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003756void kmem_cache_free(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757{
3758 unsigned long flags;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003759 cachep = cache_from_obj(cachep, objp);
3760 if (!cachep)
3761 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003762
3763 local_irq_save(flags);
Feng Tangd97d4762012-07-02 14:29:10 +08003764 debug_check_no_locks_freed(objp, cachep->object_size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07003765 if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003766 debug_check_no_obj_freed(objp, cachep->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003767 __cache_free(cachep, objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 local_irq_restore(flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003769
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003770 trace_kmem_cache_free(_RET_IP_, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771}
3772EXPORT_SYMBOL(kmem_cache_free);
3773
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003774void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p)
3775{
3776 struct kmem_cache *s;
3777 size_t i;
3778
3779 local_irq_disable();
3780 for (i = 0; i < size; i++) {
3781 void *objp = p[i];
3782
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003783 if (!orig_s) /* called via kfree_bulk */
3784 s = virt_to_cache(objp);
3785 else
3786 s = cache_from_obj(orig_s, objp);
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003787
3788 debug_check_no_locks_freed(objp, s->object_size);
3789 if (!(s->flags & SLAB_DEBUG_OBJECTS))
3790 debug_check_no_obj_freed(objp, s->object_size);
3791
3792 __cache_free(s, objp, _RET_IP_);
3793 }
3794 local_irq_enable();
3795
3796 /* FIXME: add tracing */
3797}
3798EXPORT_SYMBOL(kmem_cache_free_bulk);
3799
Linus Torvalds1da177e2005-04-16 15:20:36 -07003800/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 * kfree - free previously allocated memory
3802 * @objp: pointer returned by kmalloc.
3803 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07003804 * If @objp is NULL, no operation is performed.
3805 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 * Don't free memory not originally allocated by kmalloc()
3807 * or you will run into trouble.
3808 */
3809void kfree(const void *objp)
3810{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003811 struct kmem_cache *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 unsigned long flags;
3813
Pekka Enberg2121db72009-03-25 11:05:57 +02003814 trace_kfree(_RET_IP_, objp);
3815
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003816 if (unlikely(ZERO_OR_NULL_PTR(objp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817 return;
3818 local_irq_save(flags);
3819 kfree_debugcheck(objp);
Pekka Enberg6ed5eb2212006-02-01 03:05:49 -08003820 c = virt_to_cache(objp);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003821 debug_check_no_locks_freed(objp, c->object_size);
3822
3823 debug_check_no_obj_freed(objp, c->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003824 __cache_free(c, (void *)objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825 local_irq_restore(flags);
3826}
3827EXPORT_SYMBOL(kfree);
3828
Christoph Lametere498be72005-09-09 13:03:32 -07003829/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003830 * This initializes kmem_cache_node or resizes various caches for all nodes.
Christoph Lametere498be72005-09-09 13:03:32 -07003831 */
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003832static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -07003833{
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003834 int ret;
Christoph Lametere498be72005-09-09 13:03:32 -07003835 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003836 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07003837
Mel Gorman9c09a952008-01-24 05:49:54 -08003838 for_each_online_node(node) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003839 ret = setup_kmem_cache_node(cachep, node, gfp, true);
3840 if (ret)
Christoph Lametere498be72005-09-09 13:03:32 -07003841 goto fail;
3842
Christoph Lametere498be72005-09-09 13:03:32 -07003843 }
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003844
Christoph Lametercafeb022006-03-25 03:06:46 -08003845 return 0;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003846
Andrew Mortona737b3e2006-03-22 00:08:11 -08003847fail:
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003848 if (!cachep->list.next) {
Christoph Lameter0718dc22006-03-25 03:06:47 -08003849 /* Cache is not active yet. Roll back what we did */
3850 node--;
3851 while (node >= 0) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07003852 n = get_node(cachep, node);
3853 if (n) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003854 kfree(n->shared);
3855 free_alien_cache(n->alien);
3856 kfree(n);
Christoph Lameter6a673682013-01-10 19:14:19 +00003857 cachep->node[node] = NULL;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003858 }
3859 node--;
3860 }
3861 }
Christoph Lametercafeb022006-03-25 03:06:46 -08003862 return -ENOMEM;
Christoph Lametere498be72005-09-09 13:03:32 -07003863}
3864
Christoph Lameter18004c52012-07-06 15:25:12 -05003865/* Always called with the slab_mutex held */
Glauber Costa943a4512012-12-18 14:23:03 -08003866static int __do_tune_cpucache(struct kmem_cache *cachep, int limit,
Pekka Enberg83b519e2009-06-10 19:40:04 +03003867 int batchcount, int shared, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003868{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003869 struct array_cache __percpu *cpu_cache, *prev;
3870 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003872 cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount);
3873 if (!cpu_cache)
Siddha, Suresh Bd2e7b7d2006-09-25 23:31:47 -07003874 return -ENOMEM;
3875
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003876 prev = cachep->cpu_cache;
3877 cachep->cpu_cache = cpu_cache;
Greg Thelena87c75f2017-05-03 14:51:47 -07003878 /*
3879 * Without a previous cpu_cache there's no need to synchronize remote
3880 * cpus, so skip the IPIs.
3881 */
3882 if (prev)
3883 kick_all_cpus_sync();
Christoph Lametere498be72005-09-09 13:03:32 -07003884
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885 check_irq_on();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 cachep->batchcount = batchcount;
3887 cachep->limit = limit;
Christoph Lametere498be72005-09-09 13:03:32 -07003888 cachep->shared = shared;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003889
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003890 if (!prev)
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003891 goto setup_node;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003892
3893 for_each_online_cpu(cpu) {
Joonsoo Kim97654df2014-08-06 16:04:25 -07003894 LIST_HEAD(list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003895 int node;
3896 struct kmem_cache_node *n;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003897 struct array_cache *ac = per_cpu_ptr(prev, cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003898
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003899 node = cpu_to_mem(cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003900 n = get_node(cachep, node);
3901 spin_lock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003902 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003903 spin_unlock_irq(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07003904 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003906 free_percpu(prev);
3907
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003908setup_node:
3909 return setup_kmem_cache_nodes(cachep, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910}
3911
Glauber Costa943a4512012-12-18 14:23:03 -08003912static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
3913 int batchcount, int shared, gfp_t gfp)
3914{
3915 int ret;
Vladimir Davydov426589f2015-02-12 14:59:23 -08003916 struct kmem_cache *c;
Glauber Costa943a4512012-12-18 14:23:03 -08003917
3918 ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
3919
3920 if (slab_state < FULL)
3921 return ret;
3922
3923 if ((ret < 0) || !is_root_cache(cachep))
3924 return ret;
3925
Vladimir Davydov426589f2015-02-12 14:59:23 -08003926 lockdep_assert_held(&slab_mutex);
3927 for_each_memcg_cache(c, cachep) {
3928 /* return value determined by the root cache only */
3929 __do_tune_cpucache(c, limit, batchcount, shared, gfp);
Glauber Costa943a4512012-12-18 14:23:03 -08003930 }
3931
3932 return ret;
3933}
3934
Christoph Lameter18004c52012-07-06 15:25:12 -05003935/* Called with slab_mutex held always */
Pekka Enberg83b519e2009-06-10 19:40:04 +03003936static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937{
3938 int err;
Glauber Costa943a4512012-12-18 14:23:03 -08003939 int limit = 0;
3940 int shared = 0;
3941 int batchcount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942
Thomas Garnier7c00fce2016-07-26 15:21:56 -07003943 err = cache_random_seq_create(cachep, cachep->num, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07003944 if (err)
3945 goto end;
3946
Glauber Costa943a4512012-12-18 14:23:03 -08003947 if (!is_root_cache(cachep)) {
3948 struct kmem_cache *root = memcg_root_cache(cachep);
3949 limit = root->limit;
3950 shared = root->shared;
3951 batchcount = root->batchcount;
3952 }
3953
3954 if (limit && shared && batchcount)
3955 goto skip_setup;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003956 /*
3957 * The head array serves three purposes:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 * - create a LIFO ordering, i.e. return objects that are cache-warm
3959 * - reduce the number of spinlock operations.
Andrew Mortona737b3e2006-03-22 00:08:11 -08003960 * - reduce the number of linked list operations on the slab and
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 * bufctl chains: array operations are cheaper.
3962 * The numbers are guessed, we should auto-tune as described by
3963 * Bonwick.
3964 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003965 if (cachep->size > 131072)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 limit = 1;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003967 else if (cachep->size > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968 limit = 8;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003969 else if (cachep->size > 1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 limit = 24;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003971 else if (cachep->size > 256)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 limit = 54;
3973 else
3974 limit = 120;
3975
Andrew Mortona737b3e2006-03-22 00:08:11 -08003976 /*
3977 * CPU bound tasks (e.g. network routing) can exhibit cpu bound
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 * allocation behaviour: Most allocs on one cpu, most free operations
3979 * on another cpu. For these cases, an efficient object passing between
3980 * cpus is necessary. This is provided by a shared array. The array
3981 * replaces Bonwick's magazine layer.
3982 * On uniprocessor, it's functionally equivalent (but less efficient)
3983 * to a larger limit. Thus disabled by default.
3984 */
3985 shared = 0;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003986 if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003987 shared = 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003988
3989#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08003990 /*
3991 * With debugging enabled, large batchcount lead to excessively long
3992 * periods with disabled local interrupts. Limit the batchcount
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993 */
3994 if (limit > 32)
3995 limit = 32;
3996#endif
Glauber Costa943a4512012-12-18 14:23:03 -08003997 batchcount = (limit + 1) / 2;
3998skip_setup:
3999 err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07004000end:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001 if (err)
Joe Perches11705322016-03-17 14:19:50 -07004002 pr_err("enable_cpucache failed for %s, error %d\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004003 cachep->name, -err);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07004004 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005}
4006
Christoph Lameter1b552532006-03-22 00:09:07 -08004007/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004008 * Drain an array if it contains any elements taking the node lock only if
4009 * necessary. Note that the node listlock also protects the array_cache
Christoph Lameterb18e7e62006-03-22 00:09:07 -08004010 * if drain_array() is used on the shared array.
Christoph Lameter1b552532006-03-22 00:09:07 -08004011 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004012static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n,
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004013 struct array_cache *ac, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014{
Joonsoo Kim97654df2014-08-06 16:04:25 -07004015 LIST_HEAD(list);
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004016
4017 /* ac from n->shared can be freed if we don't hold the slab_mutex. */
4018 check_mutex_acquired();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019
Christoph Lameter1b552532006-03-22 00:09:07 -08004020 if (!ac || !ac->avail)
4021 return;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004022
4023 if (ac->touched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 ac->touched = 0;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004025 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004026 }
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004027
4028 spin_lock_irq(&n->list_lock);
4029 drain_array_locked(cachep, ac, node, false, &list);
4030 spin_unlock_irq(&n->list_lock);
4031
4032 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033}
4034
4035/**
4036 * cache_reap - Reclaim memory from caches.
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08004037 * @w: work descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07004038 *
4039 * Called from workqueue/eventd every few seconds.
4040 * Purpose:
4041 * - clear the per-cpu caches for this CPU.
4042 * - return freeable pages to the main free memory pool.
4043 *
Andrew Mortona737b3e2006-03-22 00:08:11 -08004044 * If we cannot acquire the cache chain mutex then just give up - we'll try
4045 * again on the next iteration.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046 */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004047static void cache_reap(struct work_struct *w)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048{
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07004049 struct kmem_cache *searchp;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004050 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07004051 int node = numa_mem_id();
Jean Delvarebf6aede2009-04-02 16:56:54 -07004052 struct delayed_work *work = to_delayed_work(w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004053
Christoph Lameter18004c52012-07-06 15:25:12 -05004054 if (!mutex_trylock(&slab_mutex))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 /* Give up. Setup the next iteration. */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004056 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004057
Christoph Lameter18004c52012-07-06 15:25:12 -05004058 list_for_each_entry(searchp, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004059 check_irq_on();
4060
Christoph Lameter35386e32006-03-22 00:09:05 -08004061 /*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004062 * We only take the node lock if absolutely necessary and we
Christoph Lameter35386e32006-03-22 00:09:05 -08004063 * have established with reasonable certainty that
4064 * we can do some work if the lock was obtained.
4065 */
Christoph Lameter18bf8542014-08-06 16:04:11 -07004066 n = get_node(searchp, node);
Christoph Lameter35386e32006-03-22 00:09:05 -08004067
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004068 reap_alien(searchp, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004070 drain_array(searchp, n, cpu_cache_get(searchp), node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Christoph Lameter35386e32006-03-22 00:09:05 -08004072 /*
4073 * These are racy checks but it does not matter
4074 * if we skip one check or scan twice.
4075 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004076 if (time_after(n->next_reap, jiffies))
Christoph Lameter35386e32006-03-22 00:09:05 -08004077 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08004079 n->next_reap = jiffies + REAPTIMEOUT_NODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004081 drain_array(searchp, n, n->shared, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004083 if (n->free_touched)
4084 n->free_touched = 0;
Christoph Lametered11d9e2006-06-30 01:55:45 -07004085 else {
4086 int freed;
4087
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004088 freed = drain_freelist(searchp, n, (n->free_limit +
Christoph Lametered11d9e2006-06-30 01:55:45 -07004089 5 * searchp->num - 1) / (5 * searchp->num));
4090 STATS_ADD_REAPED(searchp, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 }
Christoph Lameter35386e32006-03-22 00:09:05 -08004092next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004093 cond_resched();
4094 }
4095 check_irq_on();
Christoph Lameter18004c52012-07-06 15:25:12 -05004096 mutex_unlock(&slab_mutex);
Christoph Lameter8fce4d82006-03-09 17:33:54 -08004097 next_reap_node();
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004098out:
Andrew Mortona737b3e2006-03-22 00:08:11 -08004099 /* Set up the next iteration */
Vlastimil Babkaa9f2a842018-04-13 15:35:38 -07004100 schedule_delayed_work_on(smp_processor_id(), work,
4101 round_jiffies_relative(REAPTIMEOUT_AC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004102}
4103
Glauber Costa0d7561c2012-10-19 18:20:27 +04004104void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004105{
Greg Thelenf728b0a2016-12-12 16:41:41 -08004106 unsigned long active_objs, num_objs, active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004107 unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0;
4108 unsigned long free_slabs = 0;
Christoph Lametere498be72005-09-09 13:03:32 -07004109 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004110 struct kmem_cache_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004111
Christoph Lameter18bf8542014-08-06 16:04:11 -07004112 for_each_kmem_cache_node(cachep, node, n) {
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08004113 check_irq_on();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004114 spin_lock_irq(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07004115
David Rientjesbf00bd32016-12-12 16:41:44 -08004116 total_slabs += n->total_slabs;
4117 free_slabs += n->free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004118 free_objs += n->free_objects;
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07004119
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004120 if (n->shared)
4121 shared_avail += n->shared->avail;
Christoph Lametere498be72005-09-09 13:03:32 -07004122
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004123 spin_unlock_irq(&n->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004124 }
David Rientjesbf00bd32016-12-12 16:41:44 -08004125 num_objs = total_slabs * cachep->num;
4126 active_slabs = total_slabs - free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004127 active_objs = num_objs - free_objs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128
Glauber Costa0d7561c2012-10-19 18:20:27 +04004129 sinfo->active_objs = active_objs;
4130 sinfo->num_objs = num_objs;
4131 sinfo->active_slabs = active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004132 sinfo->num_slabs = total_slabs;
Glauber Costa0d7561c2012-10-19 18:20:27 +04004133 sinfo->shared_avail = shared_avail;
4134 sinfo->limit = cachep->limit;
4135 sinfo->batchcount = cachep->batchcount;
4136 sinfo->shared = cachep->shared;
4137 sinfo->objects_per_slab = cachep->num;
4138 sinfo->cache_order = cachep->gfporder;
4139}
4140
4141void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep)
4142{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143#if STATS
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004144 { /* node stats */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004145 unsigned long high = cachep->high_mark;
4146 unsigned long allocs = cachep->num_allocations;
4147 unsigned long grown = cachep->grown;
4148 unsigned long reaped = cachep->reaped;
4149 unsigned long errors = cachep->errors;
4150 unsigned long max_freeable = cachep->max_freeable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 unsigned long node_allocs = cachep->node_allocs;
Christoph Lametere498be72005-09-09 13:03:32 -07004152 unsigned long node_frees = cachep->node_frees;
Ravikiran G Thirumalaifb7faf32006-04-10 22:52:54 -07004153 unsigned long overflows = cachep->node_overflow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154
Joe Perches756a0252016-03-17 14:19:47 -07004155 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu",
Joe Perchese92dd4f2010-03-26 19:27:58 -07004156 allocs, high, grown,
4157 reaped, errors, max_freeable, node_allocs,
4158 node_frees, overflows);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004159 }
4160 /* cpu stats */
4161 {
4162 unsigned long allochit = atomic_read(&cachep->allochit);
4163 unsigned long allocmiss = atomic_read(&cachep->allocmiss);
4164 unsigned long freehit = atomic_read(&cachep->freehit);
4165 unsigned long freemiss = atomic_read(&cachep->freemiss);
4166
4167 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004168 allochit, allocmiss, freehit, freemiss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 }
4170#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171}
4172
Linus Torvalds1da177e2005-04-16 15:20:36 -07004173#define MAX_SLABINFO_WRITE 128
4174/**
4175 * slabinfo_write - Tuning for the slab allocator
4176 * @file: unused
4177 * @buffer: user buffer
4178 * @count: data length
4179 * @ppos: unused
4180 */
Glauber Costab7454ad2012-10-19 18:20:25 +04004181ssize_t slabinfo_write(struct file *file, const char __user *buffer,
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004182 size_t count, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004184 char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 int limit, batchcount, shared, res;
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07004186 struct kmem_cache *cachep;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004187
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 if (count > MAX_SLABINFO_WRITE)
4189 return -EINVAL;
4190 if (copy_from_user(&kbuf, buffer, count))
4191 return -EFAULT;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004192 kbuf[MAX_SLABINFO_WRITE] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193
4194 tmp = strchr(kbuf, ' ');
4195 if (!tmp)
4196 return -EINVAL;
4197 *tmp = '\0';
4198 tmp++;
4199 if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
4200 return -EINVAL;
4201
4202 /* Find the cache in the chain of caches. */
Christoph Lameter18004c52012-07-06 15:25:12 -05004203 mutex_lock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 res = -EINVAL;
Christoph Lameter18004c52012-07-06 15:25:12 -05004205 list_for_each_entry(cachep, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004206 if (!strcmp(cachep->name, kbuf)) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08004207 if (limit < 1 || batchcount < 1 ||
4208 batchcount > limit || shared < 0) {
Christoph Lametere498be72005-09-09 13:03:32 -07004209 res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004210 } else {
Christoph Lametere498be72005-09-09 13:03:32 -07004211 res = do_tune_cpucache(cachep, limit,
Pekka Enberg83b519e2009-06-10 19:40:04 +03004212 batchcount, shared,
4213 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004214 }
4215 break;
4216 }
4217 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004218 mutex_unlock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219 if (res >= 0)
4220 res = count;
4221 return res;
4222}
Al Viro871751e2006-03-25 03:06:39 -08004223
4224#ifdef CONFIG_DEBUG_SLAB_LEAK
4225
Al Viro871751e2006-03-25 03:06:39 -08004226static inline int add_caller(unsigned long *n, unsigned long v)
4227{
4228 unsigned long *p;
4229 int l;
4230 if (!v)
4231 return 1;
4232 l = n[1];
4233 p = n + 2;
4234 while (l) {
4235 int i = l/2;
4236 unsigned long *q = p + 2 * i;
4237 if (*q == v) {
4238 q[1]++;
4239 return 1;
4240 }
4241 if (*q > v) {
4242 l = i;
4243 } else {
4244 p = q + 2;
4245 l -= i + 1;
4246 }
4247 }
4248 if (++n[1] == n[0])
4249 return 0;
4250 memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n));
4251 p[0] = v;
4252 p[1] = 1;
4253 return 1;
4254}
4255
Joonsoo Kim8456a642013-10-24 10:07:49 +09004256static void handle_slab(unsigned long *n, struct kmem_cache *c,
4257 struct page *page)
Al Viro871751e2006-03-25 03:06:39 -08004258{
4259 void *p;
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004260 int i, j;
4261 unsigned long v;
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09004262
Al Viro871751e2006-03-25 03:06:39 -08004263 if (n[0] == n[1])
4264 return;
Joonsoo Kim8456a642013-10-24 10:07:49 +09004265 for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) {
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004266 bool active = true;
4267
4268 for (j = page->active; j < c->num; j++) {
4269 if (get_free_obj(page, j) == i) {
4270 active = false;
4271 break;
4272 }
4273 }
4274
4275 if (!active)
Al Viro871751e2006-03-25 03:06:39 -08004276 continue;
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09004277
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004278 /*
4279 * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table
4280 * mapping is established when actual object allocation and
4281 * we could mistakenly access the unmapped object in the cpu
4282 * cache.
4283 */
4284 if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v)))
4285 continue;
4286
4287 if (!add_caller(n, v))
Al Viro871751e2006-03-25 03:06:39 -08004288 return;
4289 }
4290}
4291
4292static void show_symbol(struct seq_file *m, unsigned long address)
4293{
4294#ifdef CONFIG_KALLSYMS
Al Viro871751e2006-03-25 03:06:39 -08004295 unsigned long offset, size;
Tejun Heo9281ace2007-07-17 04:03:51 -07004296 char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN];
Al Viro871751e2006-03-25 03:06:39 -08004297
Alexey Dobriyana5c43da2007-05-08 00:28:47 -07004298 if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) {
Al Viro871751e2006-03-25 03:06:39 -08004299 seq_printf(m, "%s+%#lx/%#lx", name, offset, size);
Alexey Dobriyana5c43da2007-05-08 00:28:47 -07004300 if (modname[0])
Al Viro871751e2006-03-25 03:06:39 -08004301 seq_printf(m, " [%s]", modname);
4302 return;
4303 }
4304#endif
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08004305 seq_printf(m, "%px", (void *)address);
Al Viro871751e2006-03-25 03:06:39 -08004306}
4307
4308static int leaks_show(struct seq_file *m, void *p)
4309{
Qian Cai78bc9822019-04-06 18:59:01 -04004310 struct kmem_cache *cachep = list_entry(p, struct kmem_cache,
4311 root_caches_node);
Joonsoo Kim8456a642013-10-24 10:07:49 +09004312 struct page *page;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004313 struct kmem_cache_node *n;
Al Viro871751e2006-03-25 03:06:39 -08004314 const char *name;
Christoph Lameterdb845062013-02-05 18:45:23 +00004315 unsigned long *x = m->private;
Al Viro871751e2006-03-25 03:06:39 -08004316 int node;
4317 int i;
4318
4319 if (!(cachep->flags & SLAB_STORE_USER))
4320 return 0;
4321 if (!(cachep->flags & SLAB_RED_ZONE))
4322 return 0;
4323
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004324 /*
4325 * Set store_user_clean and start to grab stored user information
4326 * for all objects on this cache. If some alloc/free requests comes
4327 * during the processing, information would be wrong so restart
4328 * whole processing.
4329 */
4330 do {
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004331 drain_cpu_caches(cachep);
Qian Cai515d18c2019-05-13 17:16:31 -07004332 /*
4333 * drain_cpu_caches() could make kmemleak_object and
4334 * debug_objects_cache dirty, so reset afterwards.
4335 */
4336 set_store_user_clean(cachep);
Al Viro871751e2006-03-25 03:06:39 -08004337
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004338 x[1] = 0;
Al Viro871751e2006-03-25 03:06:39 -08004339
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004340 for_each_kmem_cache_node(cachep, node, n) {
Al Viro871751e2006-03-25 03:06:39 -08004341
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004342 check_irq_on();
4343 spin_lock_irq(&n->list_lock);
Al Viro871751e2006-03-25 03:06:39 -08004344
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004345 list_for_each_entry(page, &n->slabs_full, lru)
4346 handle_slab(x, cachep, page);
4347 list_for_each_entry(page, &n->slabs_partial, lru)
4348 handle_slab(x, cachep, page);
4349 spin_unlock_irq(&n->list_lock);
4350 }
4351 } while (!is_store_user_clean(cachep));
4352
Al Viro871751e2006-03-25 03:06:39 -08004353 name = cachep->name;
Christoph Lameterdb845062013-02-05 18:45:23 +00004354 if (x[0] == x[1]) {
Al Viro871751e2006-03-25 03:06:39 -08004355 /* Increase the buffer size */
Christoph Lameter18004c52012-07-06 15:25:12 -05004356 mutex_unlock(&slab_mutex);
Kees Cook6396bb22018-06-12 14:03:40 -07004357 m->private = kcalloc(x[0] * 4, sizeof(unsigned long),
4358 GFP_KERNEL);
Al Viro871751e2006-03-25 03:06:39 -08004359 if (!m->private) {
4360 /* Too bad, we are really out */
Christoph Lameterdb845062013-02-05 18:45:23 +00004361 m->private = x;
Christoph Lameter18004c52012-07-06 15:25:12 -05004362 mutex_lock(&slab_mutex);
Al Viro871751e2006-03-25 03:06:39 -08004363 return -ENOMEM;
4364 }
Christoph Lameterdb845062013-02-05 18:45:23 +00004365 *(unsigned long *)m->private = x[0] * 2;
4366 kfree(x);
Christoph Lameter18004c52012-07-06 15:25:12 -05004367 mutex_lock(&slab_mutex);
Al Viro871751e2006-03-25 03:06:39 -08004368 /* Now make sure this entry will be retried */
4369 m->count = m->size;
4370 return 0;
4371 }
Christoph Lameterdb845062013-02-05 18:45:23 +00004372 for (i = 0; i < x[1]; i++) {
4373 seq_printf(m, "%s: %lu ", name, x[2*i+3]);
4374 show_symbol(m, x[2*i+2]);
Al Viro871751e2006-03-25 03:06:39 -08004375 seq_putc(m, '\n');
4376 }
Siddha, Suresh Bd2e7b7d2006-09-25 23:31:47 -07004377
Al Viro871751e2006-03-25 03:06:39 -08004378 return 0;
4379}
4380
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004381static const struct seq_operations slabstats_op = {
Vladimir Davydov1df3b262014-12-10 15:42:16 -08004382 .start = slab_start,
Wanpeng Li276a2432013-07-08 08:08:28 +08004383 .next = slab_next,
4384 .stop = slab_stop,
Al Viro871751e2006-03-25 03:06:39 -08004385 .show = leaks_show,
4386};
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004387
4388static int slabstats_open(struct inode *inode, struct file *file)
4389{
Rob Jonesb208ce32014-10-09 15:28:03 -07004390 unsigned long *n;
4391
4392 n = __seq_open_private(file, &slabstats_op, PAGE_SIZE);
4393 if (!n)
4394 return -ENOMEM;
4395
4396 *n = PAGE_SIZE / (2 * sizeof(unsigned long));
4397
4398 return 0;
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004399}
4400
4401static const struct file_operations proc_slabstats_operations = {
4402 .open = slabstats_open,
4403 .read = seq_read,
4404 .llseek = seq_lseek,
4405 .release = seq_release_private,
4406};
Al Viro871751e2006-03-25 03:06:39 -08004407#endif
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004408
4409static int __init slab_proc_init(void)
4410{
4411#ifdef CONFIG_DEBUG_SLAB_LEAK
4412 proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations);
4413#endif
4414 return 0;
4415}
4416module_init(slab_proc_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004417
Kees Cook04385fc2016-06-23 15:20:59 -07004418#ifdef CONFIG_HARDENED_USERCOPY
4419/*
Kees Cookafcc90f82018-01-10 15:17:01 -08004420 * Rejects incorrectly sized objects and objects that are to be copied
4421 * to/from userspace but do not fall entirely within the containing slab
4422 * cache's usercopy region.
Kees Cook04385fc2016-06-23 15:20:59 -07004423 *
4424 * Returns NULL if check passes, otherwise const char * to name of cache
4425 * to indicate an error.
4426 */
Kees Cookf4e6e282018-01-10 14:48:22 -08004427void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
4428 bool to_user)
Kees Cook04385fc2016-06-23 15:20:59 -07004429{
4430 struct kmem_cache *cachep;
4431 unsigned int objnr;
4432 unsigned long offset;
4433
Andrey Konovalov137a7d42019-02-20 22:20:25 -08004434 ptr = kasan_reset_tag(ptr);
4435
Kees Cook04385fc2016-06-23 15:20:59 -07004436 /* Find and validate object. */
4437 cachep = page->slab_cache;
4438 objnr = obj_to_index(cachep, page, (void *)ptr);
4439 BUG_ON(objnr >= cachep->num);
4440
4441 /* Find offset within object. */
4442 offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep);
4443
Kees Cookafcc90f82018-01-10 15:17:01 -08004444 /* Allow address range falling entirely within usercopy region. */
4445 if (offset >= cachep->useroffset &&
4446 offset - cachep->useroffset <= cachep->usersize &&
4447 n <= cachep->useroffset - offset + cachep->usersize)
Kees Cookf4e6e282018-01-10 14:48:22 -08004448 return;
Kees Cook04385fc2016-06-23 15:20:59 -07004449
Kees Cookafcc90f82018-01-10 15:17:01 -08004450 /*
4451 * If the copy is still within the allocated object, produce
4452 * a warning instead of rejecting the copy. This is intended
4453 * to be a temporary method to find any missing usercopy
4454 * whitelists.
4455 */
Kees Cook2d891fb2017-11-30 13:04:32 -08004456 if (usercopy_fallback &&
4457 offset <= cachep->object_size &&
Kees Cookafcc90f82018-01-10 15:17:01 -08004458 n <= cachep->object_size - offset) {
4459 usercopy_warn("SLAB object", cachep->name, to_user, offset, n);
4460 return;
4461 }
4462
Kees Cookf4e6e282018-01-10 14:48:22 -08004463 usercopy_abort("SLAB object", cachep->name, to_user, offset, n);
Kees Cook04385fc2016-06-23 15:20:59 -07004464}
4465#endif /* CONFIG_HARDENED_USERCOPY */
4466
Manfred Spraul00e145b2005-09-03 15:55:07 -07004467/**
4468 * ksize - get the actual amount of memory allocated for a given object
4469 * @objp: Pointer to the object
4470 *
4471 * kmalloc may internally round up allocations and return more memory
4472 * than requested. ksize() can be used to determine the actual amount of
4473 * memory allocated. The caller may use this additional memory, even though
4474 * a smaller amount of memory was initially specified with the kmalloc call.
4475 * The caller must guarantee that objp points to a valid object previously
4476 * allocated with either kmalloc() or kmem_cache_alloc(). The object
4477 * must not be freed during the duration of the call.
4478 */
Pekka Enbergfd76bab2007-05-06 14:48:40 -07004479size_t ksize(const void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004480{
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004481 size_t size;
4482
Christoph Lameteref8b4522007-10-16 01:24:46 -07004483 BUG_ON(!objp);
4484 if (unlikely(objp == ZERO_SIZE_PTR))
Manfred Spraul00e145b2005-09-03 15:55:07 -07004485 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004486
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004487 size = virt_to_cache(objp)->object_size;
4488 /* We assume that ksize callers could use the whole allocated area,
4489 * so we need to unpoison this area.
4490 */
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07004491 kasan_unpoison_shadow(objp, size);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004492
4493 return size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004494}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02004495EXPORT_SYMBOL(ksize);