blob: 2f308253c3d7689c677dbdc0b8dd7a686689bb7b [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
Eric Dumazet6a2d7a92006-12-13 00:34:27 -0800409/*
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500410 * We want to avoid an expensive divide : (offset / cache->size)
411 * Using the fact that size is a constant for a particular cache,
412 * we can replace (offset / cache->size) by
Eric Dumazet6a2d7a92006-12-13 00:34:27 -0800413 * reciprocal_divide(offset, cache->reciprocal_buffer_size)
414 */
415static inline unsigned int obj_to_index(const struct kmem_cache *cache,
Joonsoo Kim8456a642013-10-24 10:07:49 +0900416 const struct page *page, void *obj)
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800417{
Joonsoo Kim8456a642013-10-24 10:07:49 +0900418 u32 offset = (obj - page->s_mem);
Eric Dumazet6a2d7a92006-12-13 00:34:27 -0800419 return reciprocal_divide(offset, cache->reciprocal_buffer_size);
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800420}
421
Joonsoo Kim6fb92432016-03-15 14:54:09 -0700422#define BOOT_CPUCACHE_ENTRIES 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423/* internal cache of cache description objs */
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000424static struct kmem_cache kmem_cache_boot = {
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800425 .batchcount = 1,
426 .limit = BOOT_CPUCACHE_ENTRIES,
427 .shared = 1,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500428 .size = sizeof(struct kmem_cache),
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800429 .name = "kmem_cache",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430};
431
Tejun Heo1871e522009-10-29 22:34:13 +0900432static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433
Pekka Enberg343e0d72006-02-01 03:05:50 -0800434static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700436 return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
Andrew Mortona737b3e2006-03-22 00:08:11 -0800439/*
440 * Calculate the number of objects and left-over bytes for a given buffer size.
441 */
Joonsoo Kim70f75062016-03-15 14:54:53 -0700442static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800443 slab_flags_t flags, size_t *left_over)
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800444{
Joonsoo Kim70f75062016-03-15 14:54:53 -0700445 unsigned int num;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800446 size_t slab_size = PAGE_SIZE << gfporder;
447
448 /*
449 * The slab management structure can be either off the slab or
450 * on it. For the latter case, the memory allocated for a
451 * slab is used for:
452 *
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800453 * - @buffer_size bytes for each object
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700454 * - One freelist_idx_t for each object
455 *
456 * We don't need to consider alignment of freelist because
457 * freelist will be at the end of slab page. The objects will be
458 * at the correct alignment.
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800459 *
460 * If the slab management structure is off the slab, then the
461 * alignment will already be calculated into the size. Because
462 * the slabs are all pages aligned, the objects will be at the
463 * correct alignment when allocated.
464 */
Joonsoo Kimb03a017b2016-03-15 14:54:50 -0700465 if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700466 num = slab_size / buffer_size;
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700467 *left_over = slab_size % buffer_size;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800468 } else {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700469 num = slab_size / (buffer_size + sizeof(freelist_idx_t));
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700470 *left_over = slab_size %
471 (buffer_size + sizeof(freelist_idx_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 }
Joonsoo Kim70f75062016-03-15 14:54:53 -0700473
474 return num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
Christoph Lameterf28510d2012-09-11 19:49:38 +0000477#if DEBUG
Harvey Harrisond40cee22008-04-30 00:55:07 -0700478#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Andrew Mortona737b3e2006-03-22 00:08:11 -0800480static void __slab_error(const char *function, struct kmem_cache *cachep,
481 char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
Joe Perches11705322016-03-17 14:19:50 -0700483 pr_err("slab error in %s(): cache `%s': %s\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800484 function, cachep->name, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030486 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
Christoph Lameterf28510d2012-09-11 19:49:38 +0000488#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Paul Menage3395ee02006-12-06 20:32:16 -0800490/*
491 * By default on NUMA we use alien caches to stage the freeing of
492 * objects allocated from other nodes. This causes massive memory
493 * inefficiencies when using fake NUMA setup to split memory into a
494 * large number of small nodes, so it can be disabled on the command
495 * line
496 */
497
498static int use_alien_caches __read_mostly = 1;
499static int __init noaliencache_setup(char *s)
500{
501 use_alien_caches = 0;
502 return 1;
503}
504__setup("noaliencache", noaliencache_setup);
505
David Rientjes3df1ccc2011-10-18 22:09:28 -0700506static int __init slab_max_order_setup(char *str)
507{
508 get_option(&str, &slab_max_order);
509 slab_max_order = slab_max_order < 0 ? 0 :
510 min(slab_max_order, MAX_ORDER - 1);
511 slab_max_order_set = true;
512
513 return 1;
514}
515__setup("slab_max_order=", slab_max_order_setup);
516
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800517#ifdef CONFIG_NUMA
518/*
519 * Special reaping functions for NUMA systems called from cache_reap().
520 * These take care of doing round robin flushing of alien caches (containing
521 * objects freed on different nodes from which they were allocated) and the
522 * flushing of remote pcps by calling drain_node_pages.
523 */
Tejun Heo1871e522009-10-29 22:34:13 +0900524static DEFINE_PER_CPU(unsigned long, slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800525
526static void init_reap_node(int cpu)
527{
Andrew Morton0edaf862016-05-19 17:10:58 -0700528 per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
529 node_online_map);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800530}
531
532static void next_reap_node(void)
533{
Christoph Lameter909ea962010-12-08 16:22:55 +0100534 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800535
Andrew Morton0edaf862016-05-19 17:10:58 -0700536 node = next_node_in(node, node_online_map);
Christoph Lameter909ea962010-12-08 16:22:55 +0100537 __this_cpu_write(slab_reap_node, node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800538}
539
540#else
541#define init_reap_node(cpu) do { } while (0)
542#define next_reap_node(void) do { } while (0)
543#endif
544
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545/*
546 * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz
547 * via the workqueue/eventd.
548 * Add the CPU number into the expiration time to minimize the possibility of
549 * the CPUs getting into lockstep and contending for the global cache chain
550 * lock.
551 */
Paul Gortmaker0db06282013-06-19 14:53:51 -0400552static void start_cpu_timer(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
Tejun Heo1871e522009-10-29 22:34:13 +0900554 struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Tejun Heoeac03372016-09-16 15:49:34 -0400556 if (reap_work->work.func == NULL) {
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800557 init_reap_node(cpu);
Tejun Heo203b42f2012-08-21 13:18:23 -0700558 INIT_DEFERRABLE_WORK(reap_work, cache_reap);
Arjan van de Ven2b284212006-12-10 02:21:28 -0800559 schedule_delayed_work_on(cpu, reap_work,
560 __round_jiffies_relative(HZ, cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 }
562}
563
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700564static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
Catalin Marinasd5cff632009-06-11 13:22:40 +0100566 /*
567 * The array_cache structures contain pointers to free object.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300568 * However, when such objects are allocated or transferred to another
Catalin Marinasd5cff632009-06-11 13:22:40 +0100569 * cache the pointers are not cleared and they could be counted as
570 * valid references during a kmemleak scan. Therefore, kmemleak must
571 * not scan such objects.
572 */
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700573 kmemleak_no_scan(ac);
574 if (ac) {
575 ac->avail = 0;
576 ac->limit = limit;
577 ac->batchcount = batch;
578 ac->touched = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700580}
581
582static struct array_cache *alloc_arraycache(int node, int entries,
583 int batchcount, gfp_t gfp)
584{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700585 size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700586 struct array_cache *ac = NULL;
587
588 ac = kmalloc_node(memsize, gfp, node);
589 init_arraycache(ac, entries, batchcount);
590 return ac;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591}
592
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700593static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
594 struct page *page, void *objp)
Mel Gorman072bb0a2012-07-31 16:43:58 -0700595{
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700596 struct kmem_cache_node *n;
597 int page_node;
598 LIST_HEAD(list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700599
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700600 page_node = page_to_nid(page);
601 n = get_node(cachep, page_node);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700602
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700603 spin_lock(&n->list_lock);
604 free_block(cachep, &objp, 1, page_node, &list);
605 spin_unlock(&n->list_lock);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700606
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700607 slabs_destroy(cachep, &list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700608}
609
Christoph Lameter3ded1752006-03-25 03:06:44 -0800610/*
611 * Transfer objects in one arraycache to another.
612 * Locking must be handled by the caller.
613 *
614 * Return the number of entries transferred.
615 */
616static int transfer_objects(struct array_cache *to,
617 struct array_cache *from, unsigned int max)
618{
619 /* Figure out how many entries to transfer */
Hagen Paul Pfeifer732eacc2010-10-26 14:22:23 -0700620 int nr = min3(from->avail, max, to->limit - to->avail);
Christoph Lameter3ded1752006-03-25 03:06:44 -0800621
622 if (!nr)
623 return 0;
624
625 memcpy(to->entry + to->avail, from->entry + from->avail -nr,
626 sizeof(void *) *nr);
627
628 from->avail -= nr;
629 to->avail += nr;
Christoph Lameter3ded1752006-03-25 03:06:44 -0800630 return nr;
631}
632
Christoph Lameter765c4502006-09-27 01:50:08 -0700633#ifndef CONFIG_NUMA
634
635#define drain_alien_cache(cachep, alien) do { } while (0)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000636#define reap_alien(cachep, n) do { } while (0)
Christoph Lameter765c4502006-09-27 01:50:08 -0700637
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700638static inline struct alien_cache **alloc_alien_cache(int node,
639 int limit, gfp_t gfp)
Christoph Lameter765c4502006-09-27 01:50:08 -0700640{
Joonsoo Kim88881772016-05-19 17:10:05 -0700641 return NULL;
Christoph Lameter765c4502006-09-27 01:50:08 -0700642}
643
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700644static inline void free_alien_cache(struct alien_cache **ac_ptr)
Christoph Lameter765c4502006-09-27 01:50:08 -0700645{
646}
647
648static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
649{
650 return 0;
651}
652
653static inline void *alternate_node_alloc(struct kmem_cache *cachep,
654 gfp_t flags)
655{
656 return NULL;
657}
658
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800659static inline void *____cache_alloc_node(struct kmem_cache *cachep,
Christoph Lameter765c4502006-09-27 01:50:08 -0700660 gfp_t flags, int nodeid)
661{
662 return NULL;
663}
664
David Rientjes4167e9b2015-04-14 15:46:55 -0700665static inline gfp_t gfp_exact_node(gfp_t flags)
666{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700667 return flags & ~__GFP_NOFAIL;
David Rientjes4167e9b2015-04-14 15:46:55 -0700668}
669
Christoph Lameter765c4502006-09-27 01:50:08 -0700670#else /* CONFIG_NUMA */
671
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800672static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
Paul Jacksonc61afb12006-03-24 03:16:08 -0800673static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
Christoph Lameterdc85da12006-01-18 17:42:36 -0800674
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700675static struct alien_cache *__alloc_alien_cache(int node, int entries,
676 int batch, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -0700677{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700678 size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700679 struct alien_cache *alc = NULL;
680
681 alc = kmalloc_node(memsize, gfp, node);
682 init_arraycache(&alc->ac, entries, batch);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700683 spin_lock_init(&alc->lock);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700684 return alc;
685}
686
687static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
688{
689 struct alien_cache **alc_ptr;
Joonsoo Kim5e804782014-08-06 16:04:40 -0700690 size_t memsize = sizeof(void *) * nr_node_ids;
Christoph Lametere498be72005-09-09 13:03:32 -0700691 int i;
692
693 if (limit > 1)
694 limit = 12;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700695 alc_ptr = kzalloc_node(memsize, gfp, node);
696 if (!alc_ptr)
697 return NULL;
698
699 for_each_node(i) {
700 if (i == node || !node_online(i))
701 continue;
702 alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
703 if (!alc_ptr[i]) {
704 for (i--; i >= 0; i--)
705 kfree(alc_ptr[i]);
706 kfree(alc_ptr);
707 return NULL;
Christoph Lametere498be72005-09-09 13:03:32 -0700708 }
709 }
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700710 return alc_ptr;
Christoph Lametere498be72005-09-09 13:03:32 -0700711}
712
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700713static void free_alien_cache(struct alien_cache **alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700714{
715 int i;
716
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700717 if (!alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700718 return;
Christoph Lametere498be72005-09-09 13:03:32 -0700719 for_each_node(i)
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700720 kfree(alc_ptr[i]);
721 kfree(alc_ptr);
Christoph Lametere498be72005-09-09 13:03:32 -0700722}
723
Pekka Enberg343e0d72006-02-01 03:05:50 -0800724static void __drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim833b7062014-08-06 16:04:33 -0700725 struct array_cache *ac, int node,
726 struct list_head *list)
Christoph Lametere498be72005-09-09 13:03:32 -0700727{
Christoph Lameter18bf8542014-08-06 16:04:11 -0700728 struct kmem_cache_node *n = get_node(cachep, node);
Christoph Lametere498be72005-09-09 13:03:32 -0700729
730 if (ac->avail) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000731 spin_lock(&n->list_lock);
Christoph Lametere00946f2006-03-25 03:06:45 -0800732 /*
733 * Stuff objects into the remote nodes shared array first.
734 * That way we could avoid the overhead of putting the objects
735 * into the free lists and getting them back later.
736 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000737 if (n->shared)
738 transfer_objects(n->shared, ac, ac->limit);
Christoph Lametere00946f2006-03-25 03:06:45 -0800739
Joonsoo Kim833b7062014-08-06 16:04:33 -0700740 free_block(cachep, ac->entry, ac->avail, node, list);
Christoph Lametere498be72005-09-09 13:03:32 -0700741 ac->avail = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000742 spin_unlock(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -0700743 }
744}
745
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800746/*
747 * Called from cache_reap() to regularly drain alien caches round robin.
748 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000749static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800750{
Christoph Lameter909ea962010-12-08 16:22:55 +0100751 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800752
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000753 if (n->alien) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700754 struct alien_cache *alc = n->alien[node];
755 struct array_cache *ac;
Christoph Lametere00946f2006-03-25 03:06:45 -0800756
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700757 if (alc) {
758 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700759 if (ac->avail && spin_trylock_irq(&alc->lock)) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700760 LIST_HEAD(list);
761
762 __drain_alien_cache(cachep, ac, node, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700763 spin_unlock_irq(&alc->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700764 slabs_destroy(cachep, &list);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700765 }
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800766 }
767 }
768}
769
Andrew Mortona737b3e2006-03-22 00:08:11 -0800770static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700771 struct alien_cache **alien)
Christoph Lametere498be72005-09-09 13:03:32 -0700772{
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800773 int i = 0;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700774 struct alien_cache *alc;
Christoph Lametere498be72005-09-09 13:03:32 -0700775 struct array_cache *ac;
776 unsigned long flags;
777
778 for_each_online_node(i) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700779 alc = alien[i];
780 if (alc) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700781 LIST_HEAD(list);
782
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700783 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700784 spin_lock_irqsave(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700785 __drain_alien_cache(cachep, ac, i, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700786 spin_unlock_irqrestore(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700787 slabs_destroy(cachep, &list);
Christoph Lametere498be72005-09-09 13:03:32 -0700788 }
789 }
790}
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700791
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700792static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
793 int node, int page_node)
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700794{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000795 struct kmem_cache_node *n;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700796 struct alien_cache *alien = NULL;
797 struct array_cache *ac;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700798 LIST_HEAD(list);
Pekka Enberg1ca4cb22006-10-06 00:43:52 -0700799
Christoph Lameter18bf8542014-08-06 16:04:11 -0700800 n = get_node(cachep, node);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700801 STATS_INC_NODEFREES(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700802 if (n->alien && n->alien[page_node]) {
803 alien = n->alien[page_node];
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700804 ac = &alien->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700805 spin_lock(&alien->lock);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700806 if (unlikely(ac->avail == ac->limit)) {
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700807 STATS_INC_ACOVERFLOW(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700808 __drain_alien_cache(cachep, ac, page_node, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700809 }
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700810 ac->entry[ac->avail++] = objp;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700811 spin_unlock(&alien->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700812 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700813 } else {
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700814 n = get_node(cachep, page_node);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700815 spin_lock(&n->list_lock);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700816 free_block(cachep, &objp, 1, page_node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700817 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -0700818 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700819 }
820 return 1;
821}
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700822
823static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
824{
825 int page_node = page_to_nid(virt_to_page(objp));
826 int node = numa_mem_id();
827 /*
828 * Make sure we are not freeing a object from another node to the array
829 * cache on this cpu.
830 */
831 if (likely(node == page_node))
832 return 0;
833
834 return __cache_free_alien(cachep, objp, node, page_node);
835}
David Rientjes4167e9b2015-04-14 15:46:55 -0700836
837/*
Mel Gorman444eb2a42016-03-17 14:19:23 -0700838 * Construct gfp mask to allocate from a specific node but do not reclaim or
839 * warn about failures.
David Rientjes4167e9b2015-04-14 15:46:55 -0700840 */
841static inline gfp_t gfp_exact_node(gfp_t flags)
842{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700843 return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes4167e9b2015-04-14 15:46:55 -0700844}
Christoph Lametere498be72005-09-09 13:03:32 -0700845#endif
846
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700847static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
848{
849 struct kmem_cache_node *n;
850
851 /*
852 * Set up the kmem_cache_node for cpu before we can
853 * begin anything. Make sure some other cpu on this
854 * node has not already allocated this
855 */
856 n = get_node(cachep, node);
857 if (n) {
858 spin_lock_irq(&n->list_lock);
859 n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
860 cachep->num;
861 spin_unlock_irq(&n->list_lock);
862
863 return 0;
864 }
865
866 n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
867 if (!n)
868 return -ENOMEM;
869
870 kmem_cache_node_init(n);
871 n->next_reap = jiffies + REAPTIMEOUT_NODE +
872 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
873
874 n->free_limit =
875 (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;
876
877 /*
878 * The kmem_cache_nodes don't come and go as CPUs
879 * come and go. slab_mutex is sufficient
880 * protection here.
881 */
882 cachep->node[node] = n;
883
884 return 0;
885}
886
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200887#if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700888/*
Christoph Lameter6a673682013-01-10 19:14:19 +0000889 * Allocates and initializes node for a node on each slab cache, used for
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000890 * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node
David Rientjes8f9f8d92010-03-27 19:40:47 -0700891 * will be allocated off-node since memory is not yet online for the new node.
Christoph Lameter6a673682013-01-10 19:14:19 +0000892 * When hotplugging memory or a cpu, existing node are not replaced if
David Rientjes8f9f8d92010-03-27 19:40:47 -0700893 * already in use.
894 *
Christoph Lameter18004c52012-07-06 15:25:12 -0500895 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -0700896 */
Christoph Lameter6a673682013-01-10 19:14:19 +0000897static int init_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700898{
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700899 int ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700900 struct kmem_cache *cachep;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700901
Christoph Lameter18004c52012-07-06 15:25:12 -0500902 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700903 ret = init_cache_node(cachep, node, GFP_KERNEL);
904 if (ret)
905 return ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700906 }
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700907
David Rientjes8f9f8d92010-03-27 19:40:47 -0700908 return 0;
909}
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200910#endif
David Rientjes8f9f8d92010-03-27 19:40:47 -0700911
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700912static int setup_kmem_cache_node(struct kmem_cache *cachep,
913 int node, gfp_t gfp, bool force_change)
914{
915 int ret = -ENOMEM;
916 struct kmem_cache_node *n;
917 struct array_cache *old_shared = NULL;
918 struct array_cache *new_shared = NULL;
919 struct alien_cache **new_alien = NULL;
920 LIST_HEAD(list);
921
922 if (use_alien_caches) {
923 new_alien = alloc_alien_cache(node, cachep->limit, gfp);
924 if (!new_alien)
925 goto fail;
926 }
927
928 if (cachep->shared) {
929 new_shared = alloc_arraycache(node,
930 cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
931 if (!new_shared)
932 goto fail;
933 }
934
935 ret = init_cache_node(cachep, node, gfp);
936 if (ret)
937 goto fail;
938
939 n = get_node(cachep, node);
940 spin_lock_irq(&n->list_lock);
941 if (n->shared && force_change) {
942 free_block(cachep, n->shared->entry,
943 n->shared->avail, node, &list);
944 n->shared->avail = 0;
945 }
946
947 if (!n->shared || force_change) {
948 old_shared = n->shared;
949 n->shared = new_shared;
950 new_shared = NULL;
951 }
952
953 if (!n->alien) {
954 n->alien = new_alien;
955 new_alien = NULL;
956 }
957
958 spin_unlock_irq(&n->list_lock);
959 slabs_destroy(cachep, &list);
960
Joonsoo Kim801faf02016-05-19 17:10:31 -0700961 /*
962 * To protect lockless access to n->shared during irq disabled context.
963 * If n->shared isn't NULL in irq disabled context, accessing to it is
964 * guaranteed to be valid until irq is re-enabled, because it will be
965 * freed after synchronize_sched().
966 */
Joonsoo Kim86d9f482016-10-27 17:46:18 -0700967 if (old_shared && force_change)
Joonsoo Kim801faf02016-05-19 17:10:31 -0700968 synchronize_sched();
969
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700970fail:
971 kfree(old_shared);
972 kfree(new_shared);
973 free_alien_cache(new_alien);
974
975 return ret;
976}
977
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200978#ifdef CONFIG_SMP
979
Paul Gortmaker0db06282013-06-19 14:53:51 -0400980static void cpuup_canceled(long cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700982 struct kmem_cache *cachep;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000983 struct kmem_cache_node *n = NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -0700984 int node = cpu_to_mem(cpu);
Rusty Russella70f7302009-03-13 14:49:46 +1030985 const struct cpumask *mask = cpumask_of_node(node);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700986
Christoph Lameter18004c52012-07-06 15:25:12 -0500987 list_for_each_entry(cachep, &slab_caches, list) {
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700988 struct array_cache *nc;
989 struct array_cache *shared;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700990 struct alien_cache **alien;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700991 LIST_HEAD(list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700992
Christoph Lameter18bf8542014-08-06 16:04:11 -0700993 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000994 if (!n)
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700995 continue;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700996
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000997 spin_lock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700998
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000999 /* Free limit for this kmem_cache_node */
1000 n->free_limit -= cachep->batchcount;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001001
1002 /* cpu is dead; no one can alloc from it. */
1003 nc = per_cpu_ptr(cachep->cpu_cache, cpu);
1004 if (nc) {
Joonsoo Kim97654df2014-08-06 16:04:25 -07001005 free_block(cachep, nc->entry, nc->avail, node, &list);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001006 nc->avail = 0;
1007 }
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001008
Rusty Russell58463c12009-12-17 11:43:12 -06001009 if (!cpumask_empty(mask)) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001010 spin_unlock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001011 goto free_slab;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001012 }
1013
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001014 shared = n->shared;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001015 if (shared) {
1016 free_block(cachep, shared->entry,
Joonsoo Kim97654df2014-08-06 16:04:25 -07001017 shared->avail, node, &list);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001018 n->shared = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001019 }
1020
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001021 alien = n->alien;
1022 n->alien = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001023
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001024 spin_unlock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001025
1026 kfree(shared);
1027 if (alien) {
1028 drain_alien_cache(cachep, alien);
1029 free_alien_cache(alien);
1030 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001031
1032free_slab:
Joonsoo Kim97654df2014-08-06 16:04:25 -07001033 slabs_destroy(cachep, &list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001034 }
1035 /*
1036 * In the previous loop, all the objects were freed to
1037 * the respective cache's slabs, now we can go ahead and
1038 * shrink each nodelist to its limit.
1039 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001040 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07001041 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001042 if (!n)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001043 continue;
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001044 drain_freelist(cachep, n, INT_MAX);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001045 }
1046}
1047
Paul Gortmaker0db06282013-06-19 14:53:51 -04001048static int cpuup_prepare(long cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001049{
Pekka Enberg343e0d72006-02-01 03:05:50 -08001050 struct kmem_cache *cachep;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07001051 int node = cpu_to_mem(cpu);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001052 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001054 /*
1055 * We need to do this right in the beginning since
1056 * alloc_arraycache's are going to use this list.
1057 * kmalloc_node allows us to add the slab to the right
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001058 * kmem_cache_node and not this cpu's kmem_cache_node
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001059 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001060 err = init_cache_node_node(node);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001061 if (err < 0)
1062 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001063
1064 /*
1065 * Now we can go ahead with allocating the shared arrays and
1066 * array caches
1067 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001068 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07001069 err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
1070 if (err)
1071 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001072 }
Pekka Enbergce79ddc2009-11-23 22:01:15 +02001073
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001074 return 0;
1075bad:
Akinobu Mita12d00f62007-10-18 03:05:11 -07001076 cpuup_canceled(cpu);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001077 return -ENOMEM;
1078}
1079
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001080int slab_prepare_cpu(unsigned int cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001081{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001082 int err;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001083
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001084 mutex_lock(&slab_mutex);
1085 err = cpuup_prepare(cpu);
1086 mutex_unlock(&slab_mutex);
1087 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001090/*
1091 * This is called for a failed online attempt and for a successful
1092 * offline.
1093 *
1094 * Even if all the cpus of a node are down, we don't free the
1095 * kmem_list3 of any cache. This to avoid a race between cpu_down, and
1096 * a kmalloc allocation from another cpu for memory from the node of
1097 * the cpu going down. The list3 structure is usually allocated from
1098 * kmem_cache_create() and gets destroyed at kmem_cache_destroy().
1099 */
1100int slab_dead_cpu(unsigned int cpu)
1101{
1102 mutex_lock(&slab_mutex);
1103 cpuup_canceled(cpu);
1104 mutex_unlock(&slab_mutex);
1105 return 0;
1106}
1107#endif
1108
1109static int slab_online_cpu(unsigned int cpu)
1110{
1111 start_cpu_timer(cpu);
1112 return 0;
1113}
1114
1115static int slab_offline_cpu(unsigned int cpu)
1116{
1117 /*
1118 * Shutdown cache reaper. Note that the slab_mutex is held so
1119 * that if cache_reap() is invoked it cannot do anything
1120 * expensive but will only modify reap_work and reschedule the
1121 * timer.
1122 */
1123 cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu));
1124 /* Now the cache_reaper is guaranteed to be not running. */
1125 per_cpu(slab_reap_work, cpu).work.func = NULL;
1126 return 0;
1127}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128
David Rientjes8f9f8d92010-03-27 19:40:47 -07001129#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
1130/*
1131 * Drains freelist for a node on each slab cache, used for memory hot-remove.
1132 * Returns -EBUSY if all objects cannot be drained so that the node is not
1133 * removed.
1134 *
Christoph Lameter18004c52012-07-06 15:25:12 -05001135 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001136 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001137static int __meminit drain_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001138{
1139 struct kmem_cache *cachep;
1140 int ret = 0;
1141
Christoph Lameter18004c52012-07-06 15:25:12 -05001142 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001143 struct kmem_cache_node *n;
David Rientjes8f9f8d92010-03-27 19:40:47 -07001144
Christoph Lameter18bf8542014-08-06 16:04:11 -07001145 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001146 if (!n)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001147 continue;
1148
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001149 drain_freelist(cachep, n, INT_MAX);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001150
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001151 if (!list_empty(&n->slabs_full) ||
1152 !list_empty(&n->slabs_partial)) {
David Rientjes8f9f8d92010-03-27 19:40:47 -07001153 ret = -EBUSY;
1154 break;
1155 }
1156 }
1157 return ret;
1158}
1159
1160static int __meminit slab_memory_callback(struct notifier_block *self,
1161 unsigned long action, void *arg)
1162{
1163 struct memory_notify *mnb = arg;
1164 int ret = 0;
1165 int nid;
1166
1167 nid = mnb->status_change_nid;
1168 if (nid < 0)
1169 goto out;
1170
1171 switch (action) {
1172 case MEM_GOING_ONLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001173 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001174 ret = init_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001175 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001176 break;
1177 case MEM_GOING_OFFLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001178 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001179 ret = drain_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001180 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001181 break;
1182 case MEM_ONLINE:
1183 case MEM_OFFLINE:
1184 case MEM_CANCEL_ONLINE:
1185 case MEM_CANCEL_OFFLINE:
1186 break;
1187 }
1188out:
Prarit Bhargava5fda1bd2011-03-22 16:30:49 -07001189 return notifier_from_errno(ret);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001190}
1191#endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */
1192
Christoph Lametere498be72005-09-09 13:03:32 -07001193/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001194 * swap the static kmem_cache_node with kmalloced memory
Christoph Lametere498be72005-09-09 13:03:32 -07001195 */
Christoph Lameter6744f082013-01-10 19:12:17 +00001196static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list,
David Rientjes8f9f8d92010-03-27 19:40:47 -07001197 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07001198{
Christoph Lameter6744f082013-01-10 19:12:17 +00001199 struct kmem_cache_node *ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001200
Christoph Lameter6744f082013-01-10 19:12:17 +00001201 ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07001202 BUG_ON(!ptr);
1203
Christoph Lameter6744f082013-01-10 19:12:17 +00001204 memcpy(ptr, list, sizeof(struct kmem_cache_node));
Ingo Molnar2b2d5492006-07-03 00:25:28 -07001205 /*
1206 * Do not assume that spinlocks can be initialized via memcpy:
1207 */
1208 spin_lock_init(&ptr->list_lock);
1209
Christoph Lametere498be72005-09-09 13:03:32 -07001210 MAKE_ALL_LISTS(cachep, ptr, nodeid);
Christoph Lameter6a673682013-01-10 19:14:19 +00001211 cachep->node[nodeid] = ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001212}
1213
Andrew Mortona737b3e2006-03-22 00:08:11 -08001214/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001215 * For setting up all the kmem_cache_node for cache whose buffer_size is same as
1216 * size of kmem_cache_node.
Pekka Enberg556a1692008-01-25 08:20:51 +02001217 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001218static void __init set_up_node(struct kmem_cache *cachep, int index)
Pekka Enberg556a1692008-01-25 08:20:51 +02001219{
1220 int node;
1221
1222 for_each_online_node(node) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001223 cachep->node[node] = &init_kmem_cache_node[index + node];
Christoph Lameter6a673682013-01-10 19:14:19 +00001224 cachep->node[node]->next_reap = jiffies +
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001225 REAPTIMEOUT_NODE +
1226 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enberg556a1692008-01-25 08:20:51 +02001227 }
1228}
1229
1230/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08001231 * Initialisation. Called after the page allocator have been initialised and
1232 * before smp_init().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 */
1234void __init kmem_cache_init(void)
1235{
Christoph Lametere498be72005-09-09 13:03:32 -07001236 int i;
1237
Joonsoo Kim68126702013-10-24 10:07:42 +09001238 BUILD_BUG_ON(sizeof(((struct page *)NULL)->lru) <
1239 sizeof(struct rcu_head));
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001240 kmem_cache = &kmem_cache_boot;
1241
Joonsoo Kim88881772016-05-19 17:10:05 -07001242 if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1)
Siddha, Suresh B62918a02007-05-02 19:27:18 +02001243 use_alien_caches = 0;
1244
Christoph Lameter3c583462012-11-28 16:23:01 +00001245 for (i = 0; i < NUM_INIT_LISTS; i++)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001246 kmem_cache_node_init(&init_kmem_cache_node[i]);
Christoph Lameter3c583462012-11-28 16:23:01 +00001247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 /*
1249 * Fragmentation resistance on low memory - only use bigger
David Rientjes3df1ccc2011-10-18 22:09:28 -07001250 * page orders on machines with more than 32MB of memory if
1251 * not overridden on the command line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 */
David Rientjes3df1ccc2011-10-18 22:09:28 -07001253 if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT)
David Rientjes543585c2011-10-18 22:09:24 -07001254 slab_max_order = SLAB_MAX_ORDER_HI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 /* Bootstrap is tricky, because several objects are allocated
1257 * from caches that do not exist yet:
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001258 * 1) initialize the kmem_cache cache: it contains the struct
1259 * kmem_cache structures of all caches, except kmem_cache itself:
1260 * kmem_cache is statically allocated.
Christoph Lametere498be72005-09-09 13:03:32 -07001261 * Initially an __init data area is used for the head array and the
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001262 * kmem_cache_node structures, it's replaced with a kmalloc allocated
Christoph Lametere498be72005-09-09 13:03:32 -07001263 * array at the end of the bootstrap.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 * 2) Create the first kmalloc cache.
Pekka Enberg343e0d72006-02-01 03:05:50 -08001265 * The struct kmem_cache for the new cache is allocated normally.
Christoph Lametere498be72005-09-09 13:03:32 -07001266 * An __init data area is used for the head array.
1267 * 3) Create the remaining kmalloc caches, with minimally sized
1268 * head arrays.
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001269 * 4) Replace the __init data head arrays for kmem_cache and the first
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 * kmalloc cache with kmalloc allocated arrays.
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001271 * 5) Replace the __init data for kmem_cache_node for kmem_cache and
Christoph Lametere498be72005-09-09 13:03:32 -07001272 * the other cache's with kmalloc allocated memory.
1273 * 6) Resize the head arrays of the kmalloc caches to their final sizes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 */
1275
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001276 /* 1) create the kmem_cache */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
Eric Dumazet8da34302007-05-06 14:49:29 -07001278 /*
Eric Dumazetb56efcf2011-07-20 19:04:23 +02001279 * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids
Eric Dumazet8da34302007-05-06 14:49:29 -07001280 */
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001281 create_boot_cache(kmem_cache, "kmem_cache",
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001282 offsetof(struct kmem_cache, node) +
Christoph Lameter6744f082013-01-10 19:12:17 +00001283 nr_node_ids * sizeof(struct kmem_cache_node *),
David Windsor8eb82842017-06-10 22:50:28 -04001284 SLAB_HWCACHE_ALIGN, 0, 0);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001285 list_add(&kmem_cache->list, &slab_caches);
Shakeel Butt880cd272018-03-28 16:00:57 -07001286 memcg_link_cache(kmem_cache);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001287 slab_state = PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288
Andrew Mortona737b3e2006-03-22 00:08:11 -08001289 /*
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001290 * Initialize the caches that provide memory for the kmem_cache_node
1291 * structures first. Without this, further allocations will bug.
Christoph Lametere498be72005-09-09 13:03:32 -07001292 */
Vlastimil Babkaaf3b5f82017-02-22 15:41:05 -08001293 kmalloc_caches[INDEX_NODE] = create_kmalloc_cache(
1294 kmalloc_info[INDEX_NODE].name,
David Windsor6c0c21a2017-06-10 22:50:47 -04001295 kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS,
1296 0, kmalloc_size(INDEX_NODE));
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001297 slab_state = PARTIAL_NODE;
Daniel Sanders34cc6992015-06-24 16:55:57 -07001298 setup_kmalloc_cache_index_table();
Christoph Lametere498be72005-09-09 13:03:32 -07001299
Ingo Molnare0a42722006-06-23 02:03:46 -07001300 slab_early_init = 0;
1301
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001302 /* 5) Replace the bootstrap kmem_cache_node */
Christoph Lametere498be72005-09-09 13:03:32 -07001303 {
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07001304 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
Mel Gorman9c09a952008-01-24 05:49:54 -08001306 for_each_online_node(nid) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001307 init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid);
Pekka Enberg556a1692008-01-25 08:20:51 +02001308
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001309 init_list(kmalloc_caches[INDEX_NODE],
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001310 &init_kmem_cache_node[SIZE_NODE + nid], nid);
Christoph Lametere498be72005-09-09 13:03:32 -07001311 }
1312 }
1313
Christoph Lameterf97d5f62013-01-10 19:12:17 +00001314 create_kmalloc_caches(ARCH_KMALLOC_FLAGS);
Pekka Enberg8429db52009-06-12 15:58:59 +03001315}
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001316
Pekka Enberg8429db52009-06-12 15:58:59 +03001317void __init kmem_cache_init_late(void)
1318{
1319 struct kmem_cache *cachep;
1320
Pekka Enberg8429db52009-06-12 15:58:59 +03001321 /* 6) resize the head arrays to their final sizes */
Christoph Lameter18004c52012-07-06 15:25:12 -05001322 mutex_lock(&slab_mutex);
1323 list_for_each_entry(cachep, &slab_caches, list)
Pekka Enberg8429db52009-06-12 15:58:59 +03001324 if (enable_cpucache(cachep, GFP_NOWAIT))
1325 BUG();
Christoph Lameter18004c52012-07-06 15:25:12 -05001326 mutex_unlock(&slab_mutex);
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001327
Christoph Lameter97d06602012-07-06 15:25:11 -05001328 /* Done! */
1329 slab_state = FULL;
1330
David Rientjes8f9f8d92010-03-27 19:40:47 -07001331#ifdef CONFIG_NUMA
1332 /*
1333 * Register a memory hotplug callback that initializes and frees
Christoph Lameter6a673682013-01-10 19:14:19 +00001334 * node.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001335 */
1336 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
1337#endif
1338
Andrew Mortona737b3e2006-03-22 00:08:11 -08001339 /*
1340 * The reap timers are started later, with a module init call: That part
1341 * of the kernel is not yet operational.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 */
1343}
1344
1345static int __init cpucache_init(void)
1346{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001347 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Andrew Mortona737b3e2006-03-22 00:08:11 -08001349 /*
1350 * Register the timers that return unneeded pages to the page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 */
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001352 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online",
1353 slab_online_cpu, slab_offline_cpu);
1354 WARN_ON(ret < 0);
Glauber Costaa164f8962012-06-21 00:59:18 +04001355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return 0;
1357}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358__initcall(cpucache_init);
1359
Rafael Aquini8bdec192012-03-09 17:27:27 -03001360static noinline void
1361slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
1362{
David Rientjes9a02d692014-06-04 16:06:36 -07001363#if DEBUG
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001364 struct kmem_cache_node *n;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001365 unsigned long flags;
1366 int node;
David Rientjes9a02d692014-06-04 16:06:36 -07001367 static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
1368 DEFAULT_RATELIMIT_BURST);
1369
1370 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs))
1371 return;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001372
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07001373 pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
1374 nodeid, gfpflags, &gfpflags);
1375 pr_warn(" cache: %s, object size: %d, order: %d\n",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001376 cachep->name, cachep->size, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001377
Christoph Lameter18bf8542014-08-06 16:04:11 -07001378 for_each_kmem_cache_node(cachep, node, n) {
David Rientjesbf00bd32016-12-12 16:41:44 -08001379 unsigned long total_slabs, free_slabs, free_objs;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001380
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001381 spin_lock_irqsave(&n->list_lock, flags);
David Rientjesbf00bd32016-12-12 16:41:44 -08001382 total_slabs = n->total_slabs;
1383 free_slabs = n->free_slabs;
1384 free_objs = n->free_objects;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001385 spin_unlock_irqrestore(&n->list_lock, flags);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001386
David Rientjesbf00bd32016-12-12 16:41:44 -08001387 pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n",
1388 node, total_slabs - free_slabs, total_slabs,
1389 (total_slabs * cachep->num) - free_objs,
1390 total_slabs * cachep->num);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001391 }
David Rientjes9a02d692014-06-04 16:06:36 -07001392#endif
Rafael Aquini8bdec192012-03-09 17:27:27 -03001393}
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395/*
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001396 * Interface to system's page allocator. No need to hold the
1397 * kmem_cache_node ->list_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 *
1399 * If we requested dmaable memory, we will get it. Even if we
1400 * did not request dmaable memory, we might get it, but that
1401 * would be relatively rare and ignorable.
1402 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001403static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
1404 int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405{
1406 struct page *page;
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001407 int nr_pages;
Christoph Lameter765c4502006-09-27 01:50:08 -07001408
Glauber Costaa618e892012-06-14 16:17:21 +04001409 flags |= cachep->allocflags;
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001410
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -08001411 page = __alloc_pages_node(nodeid, flags, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001412 if (!page) {
David Rientjes9a02d692014-06-04 16:06:36 -07001413 slab_out_of_memory(cachep, flags, nodeid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 return NULL;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001415 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001417 if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) {
1418 __free_pages(page, cachep->gfporder);
1419 return NULL;
1420 }
1421
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001422 nr_pages = (1 << cachep->gfporder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
Johannes Weiner7779f212017-07-06 15:40:55 -07001424 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, nr_pages);
Christoph Lameter972d1a72006-09-25 23:31:51 -07001425 else
Johannes Weiner7779f212017-07-06 15:40:55 -07001426 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, nr_pages);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07001427
Joonsoo Kima57a4982013-10-24 10:07:44 +09001428 __SetPageSlab(page);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07001429 /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */
1430 if (sk_memalloc_socks() && page_is_pfmemalloc(page))
Joonsoo Kima57a4982013-10-24 10:07:44 +09001431 SetPageSlabPfmemalloc(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001432
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001433 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434}
1435
1436/*
1437 * Interface to system's page release.
1438 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001439static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440{
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001441 int order = cachep->gfporder;
1442 unsigned long nr_freed = (1 << order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Christoph Lameter972d1a72006-09-25 23:31:51 -07001444 if (cachep->flags & SLAB_RECLAIM_ACCOUNT)
Johannes Weiner7779f212017-07-06 15:40:55 -07001445 mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, -nr_freed);
Christoph Lameter972d1a72006-09-25 23:31:51 -07001446 else
Johannes Weiner7779f212017-07-06 15:40:55 -07001447 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, -nr_freed);
Joonsoo Kim73293c22013-10-24 10:07:37 +09001448
Joonsoo Kima57a4982013-10-24 10:07:44 +09001449 BUG_ON(!PageSlab(page));
Joonsoo Kim73293c22013-10-24 10:07:37 +09001450 __ClearPageSlabPfmemalloc(page);
Joonsoo Kima57a4982013-10-24 10:07:44 +09001451 __ClearPageSlab(page);
Joonsoo Kim8456a642013-10-24 10:07:49 +09001452 page_mapcount_reset(page);
1453 page->mapping = NULL;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 if (current->reclaim_state)
1456 current->reclaim_state->reclaimed_slab += nr_freed;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001457 memcg_uncharge_slab(page, order, cachep);
1458 __free_pages(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459}
1460
1461static void kmem_rcu_free(struct rcu_head *head)
1462{
Joonsoo Kim68126702013-10-24 10:07:42 +09001463 struct kmem_cache *cachep;
1464 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
Joonsoo Kim68126702013-10-24 10:07:42 +09001466 page = container_of(head, struct page, rcu_head);
1467 cachep = page->slab_cache;
1468
1469 kmem_freepages(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
1472#if DEBUG
Joonsoo Kim40b44132016-03-15 14:54:21 -07001473static bool is_debug_pagealloc_cache(struct kmem_cache *cachep)
1474{
1475 if (debug_pagealloc_enabled() && OFF_SLAB(cachep) &&
1476 (cachep->size % PAGE_SIZE) == 0)
1477 return true;
1478
1479 return false;
1480}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481
1482#ifdef CONFIG_DEBUG_PAGEALLOC
Pekka Enberg343e0d72006-02-01 03:05:50 -08001483static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr,
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001484 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001486 int size = cachep->object_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001488 addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001490 if (size < 5 * sizeof(unsigned long))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 return;
1492
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001493 *addr++ = 0x12345678;
1494 *addr++ = caller;
1495 *addr++ = smp_processor_id();
1496 size -= 3 * sizeof(unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 {
1498 unsigned long *sptr = &caller;
1499 unsigned long svalue;
1500
1501 while (!kstack_end(sptr)) {
1502 svalue = *sptr++;
1503 if (kernel_text_address(svalue)) {
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001504 *addr++ = svalue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 size -= sizeof(unsigned long);
1506 if (size <= sizeof(unsigned long))
1507 break;
1508 }
1509 }
1510
1511 }
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001512 *addr++ = 0x87654321;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513}
Joonsoo Kim40b44132016-03-15 14:54:21 -07001514
1515static void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1516 int map, unsigned long caller)
1517{
1518 if (!is_debug_pagealloc_cache(cachep))
1519 return;
1520
1521 if (caller)
1522 store_stackinfo(cachep, objp, caller);
1523
1524 kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map);
1525}
1526
1527#else
1528static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp,
1529 int map, unsigned long caller) {}
1530
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531#endif
1532
Pekka Enberg343e0d72006-02-01 03:05:50 -08001533static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001535 int size = cachep->object_size;
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001536 addr = &((char *)addr)[obj_offset(cachep)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
1538 memset(addr, val, size);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001539 *(unsigned char *)(addr + size - 1) = POISON_END;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540}
1541
1542static void dump_line(char *data, int offset, int limit)
1543{
1544 int i;
Dave Jonesaa83aa42006-09-29 01:59:51 -07001545 unsigned char error = 0;
1546 int bad_count = 0;
1547
Joe Perches11705322016-03-17 14:19:50 -07001548 pr_err("%03x: ", offset);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001549 for (i = 0; i < limit; i++) {
1550 if (data[offset + i] != POISON_FREE) {
1551 error = data[offset + i];
1552 bad_count++;
1553 }
Dave Jonesaa83aa42006-09-29 01:59:51 -07001554 }
Sebastian Andrzej Siewiorfdde6ab2011-07-29 18:22:13 +02001555 print_hex_dump(KERN_CONT, "", 0, 16, 1,
1556 &data[offset], limit, 1);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001557
1558 if (bad_count == 1) {
1559 error ^= POISON_FREE;
1560 if (!(error & (error - 1))) {
Joe Perches11705322016-03-17 14:19:50 -07001561 pr_err("Single bit error detected. Probably bad RAM.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001562#ifdef CONFIG_X86
Joe Perches11705322016-03-17 14:19:50 -07001563 pr_err("Run memtest86+ or a similar memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001564#else
Joe Perches11705322016-03-17 14:19:50 -07001565 pr_err("Run a memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001566#endif
1567 }
1568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570#endif
1571
1572#if DEBUG
1573
Pekka Enberg343e0d72006-02-01 03:05:50 -08001574static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575{
1576 int i, size;
1577 char *realobj;
1578
1579 if (cachep->flags & SLAB_RED_ZONE) {
Joe Perches11705322016-03-17 14:19:50 -07001580 pr_err("Redzone: 0x%llx/0x%llx\n",
1581 *dbg_redzone1(cachep, objp),
1582 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 }
1584
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001585 if (cachep->flags & SLAB_STORE_USER)
1586 pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp));
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001587 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001588 size = cachep->object_size;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001589 for (i = 0; i < size && lines; i += 16, lines--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 int limit;
1591 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001592 if (i + limit > size)
1593 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 dump_line(realobj, i, limit);
1595 }
1596}
1597
Pekka Enberg343e0d72006-02-01 03:05:50 -08001598static void check_poison_obj(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599{
1600 char *realobj;
1601 int size, i;
1602 int lines = 0;
1603
Joonsoo Kim40b44132016-03-15 14:54:21 -07001604 if (is_debug_pagealloc_cache(cachep))
1605 return;
1606
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001607 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001608 size = cachep->object_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001610 for (i = 0; i < size; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 char exp = POISON_FREE;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001612 if (i == size - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 exp = POISON_END;
1614 if (realobj[i] != exp) {
1615 int limit;
1616 /* Mismatch ! */
1617 /* Print header */
1618 if (lines == 0) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001619 pr_err("Slab corruption (%s): %s start=%px, len=%d\n",
Joe Perches11705322016-03-17 14:19:50 -07001620 print_tainted(), cachep->name,
1621 realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 print_objinfo(cachep, objp, 0);
1623 }
1624 /* Hexdump the affected line */
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001625 i = (i / 16) * 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001627 if (i + limit > size)
1628 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 dump_line(realobj, i, limit);
1630 i += 16;
1631 lines++;
1632 /* Limit to 5 lines */
1633 if (lines > 5)
1634 break;
1635 }
1636 }
1637 if (lines != 0) {
1638 /* Print some data about the neighboring objects, if they
1639 * exist:
1640 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001641 struct page *page = virt_to_head_page(objp);
Pekka Enberg8fea4e92006-03-22 00:08:10 -08001642 unsigned int objnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
Joonsoo Kim8456a642013-10-24 10:07:49 +09001644 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 if (objnr) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001646 objp = index_to_obj(cachep, page, objnr - 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001647 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001648 pr_err("Prev obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 print_objinfo(cachep, objp, 2);
1650 }
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001651 if (objnr + 1 < cachep->num) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001652 objp = index_to_obj(cachep, page, objnr + 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001653 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001654 pr_err("Next obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 print_objinfo(cachep, objp, 2);
1656 }
1657 }
1658}
1659#endif
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661#if DEBUG
Joonsoo Kim8456a642013-10-24 10:07:49 +09001662static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1663 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001664{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 int i;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001666
1667 if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) {
1668 poison_obj(cachep, page->freelist - obj_offset(cachep),
1669 POISON_FREE);
1670 }
1671
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001673 void *objp = index_to_obj(cachep, page, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 check_poison_obj(cachep, objp);
Joonsoo Kim40b44132016-03-15 14:54:21 -07001677 slab_kernel_map(cachep, objp, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 }
1679 if (cachep->flags & SLAB_RED_ZONE) {
1680 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001681 slab_error(cachep, "start of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001683 slab_error(cachep, "end of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 }
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001686}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687#else
Joonsoo Kim8456a642013-10-24 10:07:49 +09001688static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1689 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001690{
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001691}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692#endif
1693
Randy Dunlap911851e2006-03-22 00:08:14 -08001694/**
1695 * slab_destroy - destroy and release all objects in a slab
1696 * @cachep: cache pointer being destroyed
Masanari Iidacb8ee1a2014-01-28 02:57:08 +09001697 * @page: page pointer being destroyed
Randy Dunlap911851e2006-03-22 00:08:14 -08001698 *
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001699 * Destroy all the objs in a slab page, and release the mem back to the system.
1700 * Before calling the slab page must have been unlinked from the cache. The
1701 * kmem_cache_node ->list_lock is not held/needed.
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001702 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001703static void slab_destroy(struct kmem_cache *cachep, struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001704{
Joonsoo Kim7e007352013-10-30 19:04:01 +09001705 void *freelist;
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001706
Joonsoo Kim8456a642013-10-24 10:07:49 +09001707 freelist = page->freelist;
1708 slab_destroy_debugcheck(cachep, page);
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001709 if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU))
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001710 call_rcu(&page->rcu_head, kmem_rcu_free);
1711 else
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001712 kmem_freepages(cachep, page);
Joonsoo Kim68126702013-10-24 10:07:42 +09001713
1714 /*
Joonsoo Kim8456a642013-10-24 10:07:49 +09001715 * From now on, we don't use freelist
Joonsoo Kim68126702013-10-24 10:07:42 +09001716 * although actual page can be freed in rcu context
1717 */
1718 if (OFF_SLAB(cachep))
Joonsoo Kim8456a642013-10-24 10:07:49 +09001719 kmem_cache_free(cachep->freelist_cache, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720}
1721
Joonsoo Kim97654df2014-08-06 16:04:25 -07001722static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list)
1723{
1724 struct page *page, *n;
1725
1726 list_for_each_entry_safe(page, n, list, lru) {
1727 list_del(&page->lru);
1728 slab_destroy(cachep, page);
1729 }
1730}
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732/**
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001733 * calculate_slab_order - calculate size (page order) of slabs
1734 * @cachep: pointer to the cache that is being created
1735 * @size: size of objects to be created in this cache.
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001736 * @flags: slab allocation flags
1737 *
1738 * Also calculates the number of objects per slab.
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001739 *
1740 * This could be made much more intelligent. For now, try to avoid using
1741 * high order pages for slabs. When the gfp() functions are more friendly
1742 * towards high-order requests, this should be changed.
1743 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001744static size_t calculate_slab_order(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001745 size_t size, slab_flags_t flags)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001746{
1747 size_t left_over = 0;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001748 int gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001749
Christoph Lameter0aa817f2007-05-16 22:11:01 -07001750 for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001751 unsigned int num;
1752 size_t remainder;
1753
Joonsoo Kim70f75062016-03-15 14:54:53 -07001754 num = cache_estimate(gfporder, size, flags, &remainder);
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001755 if (!num)
1756 continue;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001757
Joonsoo Kimf315e3f2013-12-02 17:49:41 +09001758 /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */
1759 if (num > SLAB_OBJ_MAX_NUM)
1760 break;
1761
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001762 if (flags & CFLGS_OFF_SLAB) {
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001763 struct kmem_cache *freelist_cache;
1764 size_t freelist_size;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001765
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001766 freelist_size = num * sizeof(freelist_idx_t);
1767 freelist_cache = kmalloc_slab(freelist_size, 0u);
1768 if (!freelist_cache)
1769 continue;
1770
1771 /*
1772 * Needed to avoid possible looping condition
Joonsoo Kim76b342b2016-05-19 17:10:26 -07001773 * in cache_grow_begin()
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001774 */
1775 if (OFF_SLAB(freelist_cache))
1776 continue;
1777
1778 /* check if off slab has enough benefit */
1779 if (freelist_cache->size > cachep->size / 2)
1780 continue;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001781 }
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001782
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001783 /* Found something acceptable - save it away */
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001784 cachep->num = num;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001785 cachep->gfporder = gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001786 left_over = remainder;
1787
1788 /*
Linus Torvaldsf78bb8a2006-03-08 10:33:05 -08001789 * A VFS-reclaimable slab tends to have most allocations
1790 * as GFP_NOFS and we really don't want to have to be allocating
1791 * higher-order pages when we are unable to shrink dcache.
1792 */
1793 if (flags & SLAB_RECLAIM_ACCOUNT)
1794 break;
1795
1796 /*
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001797 * Large number of objects is good, but very large slabs are
1798 * currently bad for the gfp()s.
1799 */
David Rientjes543585c2011-10-18 22:09:24 -07001800 if (gfporder >= slab_max_order)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001801 break;
1802
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001803 /*
1804 * Acceptable internal fragmentation?
1805 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001806 if (left_over * 8 <= (PAGE_SIZE << gfporder))
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001807 break;
1808 }
1809 return left_over;
1810}
1811
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001812static struct array_cache __percpu *alloc_kmem_cache_cpus(
1813 struct kmem_cache *cachep, int entries, int batchcount)
1814{
1815 int cpu;
1816 size_t size;
1817 struct array_cache __percpu *cpu_cache;
1818
1819 size = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim85c9f4b2014-10-13 15:51:01 -07001820 cpu_cache = __alloc_percpu(size, sizeof(void *));
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001821
1822 if (!cpu_cache)
1823 return NULL;
1824
1825 for_each_possible_cpu(cpu) {
1826 init_arraycache(per_cpu_ptr(cpu_cache, cpu),
1827 entries, batchcount);
1828 }
1829
1830 return cpu_cache;
1831}
1832
Fabian Frederickbd721ea2016-08-02 14:03:33 -07001833static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001834{
Christoph Lameter97d06602012-07-06 15:25:11 -05001835 if (slab_state >= FULL)
Pekka Enberg83b519e2009-06-10 19:40:04 +03001836 return enable_cpucache(cachep, gfp);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001837
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001838 cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1);
1839 if (!cachep->cpu_cache)
1840 return 1;
1841
Christoph Lameter97d06602012-07-06 15:25:11 -05001842 if (slab_state == DOWN) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001843 /* Creation of first cache (kmem_cache). */
1844 set_up_node(kmem_cache, CACHE_CACHE);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001845 } else if (slab_state == PARTIAL) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001846 /* For kmem_cache_node */
1847 set_up_node(cachep, SIZE_NODE);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001848 } else {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001849 int node;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001850
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001851 for_each_online_node(node) {
1852 cachep->node[node] = kmalloc_node(
1853 sizeof(struct kmem_cache_node), gfp, node);
1854 BUG_ON(!cachep->node[node]);
1855 kmem_cache_node_init(cachep->node[node]);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001856 }
1857 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001858
Christoph Lameter6a673682013-01-10 19:14:19 +00001859 cachep->node[numa_mem_id()]->next_reap =
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001860 jiffies + REAPTIMEOUT_NODE +
1861 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001862
1863 cpu_cache_get(cachep)->avail = 0;
1864 cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
1865 cpu_cache_get(cachep)->batchcount = 1;
1866 cpu_cache_get(cachep)->touched = 0;
1867 cachep->batchcount = 1;
1868 cachep->limit = BOOT_CPUCACHE_ENTRIES;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001869 return 0;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001870}
1871
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001872slab_flags_t kmem_cache_flags(unsigned int object_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001873 slab_flags_t flags, const char *name,
Joonsoo Kim12220de2014-10-09 15:26:24 -07001874 void (*ctor)(void *))
1875{
1876 return flags;
1877}
1878
1879struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -07001880__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001881 slab_flags_t flags, void (*ctor)(void *))
Joonsoo Kim12220de2014-10-09 15:26:24 -07001882{
1883 struct kmem_cache *cachep;
1884
1885 cachep = find_mergeable(size, align, flags, name, ctor);
1886 if (cachep) {
1887 cachep->refcount++;
1888
1889 /*
1890 * Adjust the object sizes so that we clear
1891 * the complete object on kzalloc.
1892 */
1893 cachep->object_size = max_t(int, cachep->object_size, size);
1894 }
1895 return cachep;
1896}
1897
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001898static bool set_objfreelist_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001899 size_t size, slab_flags_t flags)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001900{
1901 size_t left;
1902
1903 cachep->num = 0;
1904
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001905 if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001906 return false;
1907
1908 left = calculate_slab_order(cachep, size,
1909 flags | CFLGS_OBJFREELIST_SLAB);
1910 if (!cachep->num)
1911 return false;
1912
1913 if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size)
1914 return false;
1915
1916 cachep->colour = left / cachep->colour_off;
1917
1918 return true;
1919}
1920
Joonsoo Kim158e3192016-03-15 14:54:35 -07001921static bool set_off_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001922 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001923{
1924 size_t left;
1925
1926 cachep->num = 0;
1927
1928 /*
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001929 * Always use on-slab management when SLAB_NOLEAKTRACE
1930 * to avoid recursive calls into kmemleak.
Joonsoo Kim158e3192016-03-15 14:54:35 -07001931 */
Joonsoo Kim158e3192016-03-15 14:54:35 -07001932 if (flags & SLAB_NOLEAKTRACE)
1933 return false;
1934
1935 /*
1936 * Size is large, assume best to place the slab management obj
1937 * off-slab (should allow better packing of objs).
1938 */
1939 left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB);
1940 if (!cachep->num)
1941 return false;
1942
1943 /*
1944 * If the slab has been placed off-slab, and we have enough space then
1945 * move it on-slab. This is at the expense of any extra colouring.
1946 */
1947 if (left >= cachep->num * sizeof(freelist_idx_t))
1948 return false;
1949
1950 cachep->colour = left / cachep->colour_off;
1951
1952 return true;
1953}
1954
1955static bool set_on_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001956 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001957{
1958 size_t left;
1959
1960 cachep->num = 0;
1961
1962 left = calculate_slab_order(cachep, size, flags);
1963 if (!cachep->num)
1964 return false;
1965
1966 cachep->colour = left / cachep->colour_off;
1967
1968 return true;
1969}
1970
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001971/**
Christoph Lameter039363f2012-07-06 15:25:10 -05001972 * __kmem_cache_create - Create a cache.
Randy Dunlapa755b762012-11-06 17:10:10 -08001973 * @cachep: cache management descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 * @flags: SLAB flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 *
1976 * Returns a ptr to the cache on success, NULL on failure.
1977 * Cannot be called within a int, but can be interrupted.
Paul Mundt20c2df82007-07-20 10:11:58 +09001978 * The @ctor is run when new pages are allocated by the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 * The flags are
1981 *
1982 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
1983 * to catch references to uninitialised memory.
1984 *
1985 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
1986 * for buffer overruns.
1987 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
1989 * cacheline. This can be beneficial if you're counting cycles as closely
1990 * as davem.
1991 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001992int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993{
David Rientjesd4a5fca2014-09-25 16:05:20 -07001994 size_t ralign = BYTES_PER_WORD;
Pekka Enberg83b519e2009-06-10 19:40:04 +03001995 gfp_t gfp;
Christoph Lameter278b1bb2012-09-05 00:20:34 +00001996 int err;
Alexey Dobriyanbe4a7982018-04-05 16:21:28 -07001997 unsigned int size = cachep->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000#if FORCED_DEBUG
2001 /*
2002 * Enable redzoning and last user accounting, except for caches with
2003 * large objects, if the increased size would increase the object size
2004 * above the next power of two: caches with object sizes just above a
2005 * power of two have a significant amount of internal fragmentation.
2006 */
David Woodhouse87a927c2007-07-04 21:26:44 -04002007 if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
2008 2 * sizeof(unsigned long long)))
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002009 flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08002010 if (!(flags & SLAB_TYPESAFE_BY_RCU))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 flags |= SLAB_POISON;
2012#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Andrew Mortona737b3e2006-03-22 00:08:11 -08002015 /*
2016 * Check that size is in terms of words. This is needed to avoid
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 * unaligned accesses for some archs when redzoning is used, and makes
2018 * sure any on-slab bufctl's are also correctly aligned.
2019 */
Canjiang Lue0771952017-07-06 15:36:37 -07002020 size = ALIGN(size, BYTES_PER_WORD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
David Woodhouse87a927c2007-07-04 21:26:44 -04002022 if (flags & SLAB_RED_ZONE) {
2023 ralign = REDZONE_ALIGN;
2024 /* If redzoning, ensure that the second redzone is suitably
2025 * aligned, by adjusting the object size accordingly. */
Canjiang Lue0771952017-07-06 15:36:37 -07002026 size = ALIGN(size, REDZONE_ALIGN);
David Woodhouse87a927c2007-07-04 21:26:44 -04002027 }
Pekka Enbergca5f9702006-09-25 23:31:25 -07002028
Kevin Hilmana44b56d2006-12-06 20:32:11 -08002029 /* 3) caller mandated alignment */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002030 if (ralign < cachep->align) {
2031 ralign = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
Pekka Enberg3ff84a72011-02-14 17:46:21 +02002033 /* disable debug if necessary */
2034 if (ralign > __alignof__(unsigned long long))
Kevin Hilmana44b56d2006-12-06 20:32:11 -08002035 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
Andrew Mortona737b3e2006-03-22 00:08:11 -08002036 /*
Pekka Enbergca5f9702006-09-25 23:31:25 -07002037 * 4) Store it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002039 cachep->align = ralign;
Joonsoo Kim158e3192016-03-15 14:54:35 -07002040 cachep->colour_off = cache_line_size();
2041 /* Offset must be a multiple of the alignment. */
2042 if (cachep->colour_off < cachep->align)
2043 cachep->colour_off = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Pekka Enberg83b519e2009-06-10 19:40:04 +03002045 if (slab_is_available())
2046 gfp = GFP_KERNEL;
2047 else
2048 gfp = GFP_NOWAIT;
2049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051
Pekka Enbergca5f9702006-09-25 23:31:25 -07002052 /*
2053 * Both debugging options require word-alignment which is calculated
2054 * into align above.
2055 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 if (flags & SLAB_RED_ZONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 /* add space for red zone words */
Pekka Enberg3ff84a72011-02-14 17:46:21 +02002058 cachep->obj_offset += sizeof(unsigned long long);
2059 size += 2 * sizeof(unsigned long long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 }
2061 if (flags & SLAB_STORE_USER) {
Pekka Enbergca5f9702006-09-25 23:31:25 -07002062 /* user store requires one word storage behind the end of
David Woodhouse87a927c2007-07-04 21:26:44 -04002063 * the real object. But if the second red zone needs to be
2064 * aligned to 64 bits, we must allow that much space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 */
David Woodhouse87a927c2007-07-04 21:26:44 -04002066 if (flags & SLAB_RED_ZONE)
2067 size += REDZONE_ALIGN;
2068 else
2069 size += BYTES_PER_WORD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 }
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002071#endif
2072
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002073 kasan_cache_create(cachep, &size, &flags);
2074
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002075 size = ALIGN(size, cachep->align);
2076 /*
2077 * We should restrict the number of objects in a slab to implement
2078 * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition.
2079 */
2080 if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE)
2081 size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align);
2082
2083#if DEBUG
Joonsoo Kim03a2d2a2015-10-01 15:36:54 -07002084 /*
2085 * To activate debug pagealloc, off-slab management is necessary
2086 * requirement. In early phase of initialization, small sized slab
2087 * doesn't get initialized so it would not be possible. So, we need
2088 * to check size >= 256. It guarantees that all necessary small
2089 * sized slab is initialized in current slab initialization sequence.
2090 */
Joonsoo Kim40323272016-03-15 14:54:18 -07002091 if (debug_pagealloc_enabled() && (flags & SLAB_POISON) &&
Joonsoo Kimf3a3c322016-03-15 14:54:38 -07002092 size >= 256 && cachep->object_size > cache_line_size()) {
2093 if (size < PAGE_SIZE || size % PAGE_SIZE == 0) {
2094 size_t tmp_size = ALIGN(size, PAGE_SIZE);
2095
2096 if (set_off_slab_cache(cachep, tmp_size, flags)) {
2097 flags |= CFLGS_OFF_SLAB;
2098 cachep->obj_offset += tmp_size - size;
2099 size = tmp_size;
2100 goto done;
2101 }
2102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 }
2104#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002106 if (set_objfreelist_slab_cache(cachep, size, flags)) {
2107 flags |= CFLGS_OBJFREELIST_SLAB;
2108 goto done;
2109 }
2110
Joonsoo Kim158e3192016-03-15 14:54:35 -07002111 if (set_off_slab_cache(cachep, size, flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 flags |= CFLGS_OFF_SLAB;
Joonsoo Kim158e3192016-03-15 14:54:35 -07002113 goto done;
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002114 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115
Joonsoo Kim158e3192016-03-15 14:54:35 -07002116 if (set_on_slab_cache(cachep, size, flags))
2117 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Joonsoo Kim158e3192016-03-15 14:54:35 -07002119 return -E2BIG;
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002120
Joonsoo Kim158e3192016-03-15 14:54:35 -07002121done:
2122 cachep->freelist_size = cachep->num * sizeof(freelist_idx_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 cachep->flags = flags;
Joonsoo Kima57a4982013-10-24 10:07:44 +09002124 cachep->allocflags = __GFP_COMP;
Yang Shia3187e42016-05-19 17:10:41 -07002125 if (flags & SLAB_CACHE_DMA)
Glauber Costaa618e892012-06-14 16:17:21 +04002126 cachep->allocflags |= GFP_DMA;
David Rientjesa3ba0742017-11-15 17:32:14 -08002127 if (flags & SLAB_RECLAIM_ACCOUNT)
2128 cachep->allocflags |= __GFP_RECLAIMABLE;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002129 cachep->size = size;
Eric Dumazet6a2d7a92006-12-13 00:34:27 -08002130 cachep->reciprocal_buffer_size = reciprocal_value(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Joonsoo Kim40b44132016-03-15 14:54:21 -07002132#if DEBUG
2133 /*
2134 * If we're going to use the generic kernel_map_pages()
2135 * poisoning, then it's going to smash the contents of
2136 * the redzone and userword anyhow, so switch them off.
2137 */
2138 if (IS_ENABLED(CONFIG_PAGE_POISONING) &&
2139 (cachep->flags & SLAB_POISON) &&
2140 is_debug_pagealloc_cache(cachep))
2141 cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2142#endif
2143
2144 if (OFF_SLAB(cachep)) {
Joonsoo Kim158e3192016-03-15 14:54:35 -07002145 cachep->freelist_cache =
2146 kmalloc_slab(cachep->freelist_size, 0u);
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002149 err = setup_cpu_cache(cachep, gfp);
2150 if (err) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002151 __kmem_cache_release(cachep);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002152 return err;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07002153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
2158#if DEBUG
2159static void check_irq_off(void)
2160{
2161 BUG_ON(!irqs_disabled());
2162}
2163
2164static void check_irq_on(void)
2165{
2166 BUG_ON(irqs_disabled());
2167}
2168
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002169static void check_mutex_acquired(void)
2170{
2171 BUG_ON(!mutex_is_locked(&slab_mutex));
2172}
2173
Pekka Enberg343e0d72006-02-01 03:05:50 -08002174static void check_spinlock_acquired(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175{
2176#ifdef CONFIG_SMP
2177 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002178 assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179#endif
2180}
Christoph Lametere498be72005-09-09 13:03:32 -07002181
Pekka Enberg343e0d72006-02-01 03:05:50 -08002182static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
Christoph Lametere498be72005-09-09 13:03:32 -07002183{
2184#ifdef CONFIG_SMP
2185 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002186 assert_spin_locked(&get_node(cachep, node)->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07002187#endif
2188}
2189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190#else
2191#define check_irq_off() do { } while(0)
2192#define check_irq_on() do { } while(0)
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002193#define check_mutex_acquired() do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194#define check_spinlock_acquired(x) do { } while(0)
Christoph Lametere498be72005-09-09 13:03:32 -07002195#define check_spinlock_acquired_node(x, y) do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196#endif
2197
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002198static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac,
2199 int node, bool free_all, struct list_head *list)
2200{
2201 int tofree;
2202
2203 if (!ac || !ac->avail)
2204 return;
2205
2206 tofree = free_all ? ac->avail : (ac->limit + 4) / 5;
2207 if (tofree > ac->avail)
2208 tofree = (ac->avail + 1) / 2;
2209
2210 free_block(cachep, ac->entry, tofree, node, list);
2211 ac->avail -= tofree;
2212 memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail);
2213}
Christoph Lameteraab22072006-03-22 00:09:06 -08002214
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215static void do_drain(void *arg)
2216{
Andrew Mortona737b3e2006-03-22 00:08:11 -08002217 struct kmem_cache *cachep = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 struct array_cache *ac;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002219 int node = numa_mem_id();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002220 struct kmem_cache_node *n;
Joonsoo Kim97654df2014-08-06 16:04:25 -07002221 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222
2223 check_irq_off();
Pekka Enberg9a2dba42006-02-01 03:05:49 -08002224 ac = cpu_cache_get(cachep);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002225 n = get_node(cachep, node);
2226 spin_lock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07002227 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002228 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07002229 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230 ac->avail = 0;
2231}
2232
Pekka Enberg343e0d72006-02-01 03:05:50 -08002233static void drain_cpu_caches(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002235 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002236 int node;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002237 LIST_HEAD(list);
Christoph Lametere498be72005-09-09 13:03:32 -07002238
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002239 on_each_cpu(do_drain, cachep, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002241 for_each_kmem_cache_node(cachep, node, n)
2242 if (n->alien)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002243 drain_alien_cache(cachep, n->alien);
Roland Dreiera4523a82006-05-15 11:41:00 -07002244
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002245 for_each_kmem_cache_node(cachep, node, n) {
2246 spin_lock_irq(&n->list_lock);
2247 drain_array_locked(cachep, n->shared, node, true, &list);
2248 spin_unlock_irq(&n->list_lock);
2249
2250 slabs_destroy(cachep, &list);
2251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252}
2253
Christoph Lametered11d9e2006-06-30 01:55:45 -07002254/*
2255 * Remove slabs from the list of free slabs.
2256 * Specify the number of slabs to drain in tofree.
2257 *
2258 * Returns the actual number of slabs released.
2259 */
2260static int drain_freelist(struct kmem_cache *cache,
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002261 struct kmem_cache_node *n, int tofree)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
Christoph Lametered11d9e2006-06-30 01:55:45 -07002263 struct list_head *p;
2264 int nr_freed;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002265 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Christoph Lametered11d9e2006-06-30 01:55:45 -07002267 nr_freed = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002268 while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002270 spin_lock_irq(&n->list_lock);
2271 p = n->slabs_free.prev;
2272 if (p == &n->slabs_free) {
2273 spin_unlock_irq(&n->list_lock);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002274 goto out;
2275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276
Joonsoo Kim8456a642013-10-24 10:07:49 +09002277 page = list_entry(p, struct page, lru);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002278 list_del(&page->lru);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002279 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08002280 n->total_slabs--;
Christoph Lametered11d9e2006-06-30 01:55:45 -07002281 /*
2282 * Safe to drop the lock. The slab is no longer linked
2283 * to the cache.
2284 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002285 n->free_objects -= cache->num;
2286 spin_unlock_irq(&n->list_lock);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002287 slab_destroy(cache, page);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002288 nr_freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 }
Christoph Lametered11d9e2006-06-30 01:55:45 -07002290out:
2291 return nr_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292}
2293
Shakeel Buttf9e13c02018-04-05 16:21:57 -07002294bool __kmem_cache_empty(struct kmem_cache *s)
2295{
2296 int node;
2297 struct kmem_cache_node *n;
2298
2299 for_each_kmem_cache_node(s, node, n)
2300 if (!list_empty(&n->slabs_full) ||
2301 !list_empty(&n->slabs_partial))
2302 return false;
2303 return true;
2304}
2305
Tejun Heoc9fc5862017-02-22 15:41:27 -08002306int __kmem_cache_shrink(struct kmem_cache *cachep)
Christoph Lametere498be72005-09-09 13:03:32 -07002307{
Christoph Lameter18bf8542014-08-06 16:04:11 -07002308 int ret = 0;
2309 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002310 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002311
2312 drain_cpu_caches(cachep);
2313
2314 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002315 for_each_kmem_cache_node(cachep, node, n) {
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07002316 drain_freelist(cachep, n, INT_MAX);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002317
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002318 ret += !list_empty(&n->slabs_full) ||
2319 !list_empty(&n->slabs_partial);
Christoph Lametere498be72005-09-09 13:03:32 -07002320 }
2321 return (ret ? 1 : 0);
2322}
2323
Tejun Heoc9fc5862017-02-22 15:41:27 -08002324#ifdef CONFIG_MEMCG
2325void __kmemcg_cache_deactivate(struct kmem_cache *cachep)
2326{
2327 __kmem_cache_shrink(cachep);
2328}
2329#endif
2330
Christoph Lameter945cf2b2012-09-04 23:18:33 +00002331int __kmem_cache_shutdown(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332{
Tejun Heoc9fc5862017-02-22 15:41:27 -08002333 return __kmem_cache_shrink(cachep);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002334}
2335
2336void __kmem_cache_release(struct kmem_cache *cachep)
2337{
Christoph Lameter12c36672012-09-04 23:38:33 +00002338 int i;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002339 struct kmem_cache_node *n;
Christoph Lameter12c36672012-09-04 23:38:33 +00002340
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002341 cache_random_seq_destroy(cachep);
2342
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07002343 free_percpu(cachep->cpu_cache);
Christoph Lameter12c36672012-09-04 23:38:33 +00002344
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002345 /* NUMA: free the node structures */
Christoph Lameter18bf8542014-08-06 16:04:11 -07002346 for_each_kmem_cache_node(cachep, i, n) {
2347 kfree(n->shared);
2348 free_alien_cache(n->alien);
2349 kfree(n);
2350 cachep->node[i] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002354/*
2355 * Get the memory for a slab management obj.
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08002356 *
2357 * For a slab cache when the slab descriptor is off-slab, the
2358 * slab descriptor can't come from the same cache which is being created,
2359 * Because if it is the case, that means we defer the creation of
2360 * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point.
2361 * And we eventually call down to __kmem_cache_create(), which
2362 * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one.
2363 * This is a "chicken-and-egg" problem.
2364 *
2365 * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches,
2366 * which are all initialized during kmem_cache_init().
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002367 */
Joonsoo Kim7e007352013-10-30 19:04:01 +09002368static void *alloc_slabmgmt(struct kmem_cache *cachep,
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002369 struct page *page, int colour_off,
2370 gfp_t local_flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002372 void *freelist;
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002373 void *addr = page_address(page);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002374
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002375 page->s_mem = addr + colour_off;
2376 page->active = 0;
2377
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002378 if (OBJFREELIST_SLAB(cachep))
2379 freelist = NULL;
2380 else if (OFF_SLAB(cachep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 /* Slab management obj is off-slab. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002382 freelist = kmem_cache_alloc_node(cachep->freelist_cache,
Pekka Enberg8759ec52008-11-26 10:01:31 +02002383 local_flags, nodeid);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002384 if (!freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 return NULL;
2386 } else {
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002387 /* We will use last bytes at the slab for freelist */
2388 freelist = addr + (PAGE_SIZE << cachep->gfporder) -
2389 cachep->freelist_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 }
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002391
Joonsoo Kim8456a642013-10-24 10:07:49 +09002392 return freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393}
2394
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002395static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002397 return ((freelist_idx_t *)page->freelist)[idx];
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002398}
2399
2400static inline void set_free_obj(struct page *page,
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002401 unsigned int idx, freelist_idx_t val)
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002402{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002403 ((freelist_idx_t *)(page->freelist))[idx] = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404}
2405
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002406static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407{
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002408#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 int i;
2410
2411 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09002412 void *objp = index_to_obj(cachep, page, i);
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002413
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 if (cachep->flags & SLAB_STORE_USER)
2415 *dbg_userword(cachep, objp) = NULL;
2416
2417 if (cachep->flags & SLAB_RED_ZONE) {
2418 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2419 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2420 }
2421 /*
Andrew Mortona737b3e2006-03-22 00:08:11 -08002422 * Constructors are not allowed to allocate memory from the same
2423 * cache which they are a constructor for. Otherwise, deadlock.
2424 * They must also be threaded.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002426 if (cachep->ctor && !(cachep->flags & SLAB_POISON)) {
2427 kasan_unpoison_object_data(cachep,
2428 objp + obj_offset(cachep));
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002429 cachep->ctor(objp + obj_offset(cachep));
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002430 kasan_poison_object_data(
2431 cachep, objp + obj_offset(cachep));
2432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433
2434 if (cachep->flags & SLAB_RED_ZONE) {
2435 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002436 slab_error(cachep, "constructor overwrote the end of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002438 slab_error(cachep, "constructor overwrote the start of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 }
Joonsoo Kim40b44132016-03-15 14:54:21 -07002440 /* need to poison the objs? */
2441 if (cachep->flags & SLAB_POISON) {
2442 poison_obj(cachep, objp, POISON_FREE);
2443 slab_kernel_map(cachep, objp, 0, 0);
2444 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002445 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446#endif
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002447}
2448
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002449#ifdef CONFIG_SLAB_FREELIST_RANDOM
2450/* Hold information during a freelist initialization */
2451union freelist_init_state {
2452 struct {
2453 unsigned int pos;
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002454 unsigned int *list;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002455 unsigned int count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002456 };
2457 struct rnd_state rnd_state;
2458};
2459
2460/*
2461 * Initialize the state based on the randomization methode available.
2462 * return true if the pre-computed list is available, false otherwize.
2463 */
2464static bool freelist_state_initialize(union freelist_init_state *state,
2465 struct kmem_cache *cachep,
2466 unsigned int count)
2467{
2468 bool ret;
2469 unsigned int rand;
2470
2471 /* Use best entropy available to define a random shift */
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002472 rand = get_random_int();
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002473
2474 /* Use a random state if the pre-computed list is not available */
2475 if (!cachep->random_seq) {
2476 prandom_seed_state(&state->rnd_state, rand);
2477 ret = false;
2478 } else {
2479 state->list = cachep->random_seq;
2480 state->count = count;
John Sperbeckc4e490c2017-01-10 16:58:24 -08002481 state->pos = rand % count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002482 ret = true;
2483 }
2484 return ret;
2485}
2486
2487/* Get the next entry on the list and randomize it using a random shift */
2488static freelist_idx_t next_random_slot(union freelist_init_state *state)
2489{
John Sperbeckc4e490c2017-01-10 16:58:24 -08002490 if (state->pos >= state->count)
2491 state->pos = 0;
2492 return state->list[state->pos++];
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002493}
2494
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002495/* Swap two freelist entries */
2496static void swap_free_obj(struct page *page, unsigned int a, unsigned int b)
2497{
2498 swap(((freelist_idx_t *)page->freelist)[a],
2499 ((freelist_idx_t *)page->freelist)[b]);
2500}
2501
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002502/*
2503 * Shuffle the freelist initialization state based on pre-computed lists.
2504 * return true if the list was successfully shuffled, false otherwise.
2505 */
2506static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page)
2507{
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002508 unsigned int objfreelist = 0, i, rand, count = cachep->num;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002509 union freelist_init_state state;
2510 bool precomputed;
2511
2512 if (count < 2)
2513 return false;
2514
2515 precomputed = freelist_state_initialize(&state, cachep, count);
2516
2517 /* Take a random entry as the objfreelist */
2518 if (OBJFREELIST_SLAB(cachep)) {
2519 if (!precomputed)
2520 objfreelist = count - 1;
2521 else
2522 objfreelist = next_random_slot(&state);
2523 page->freelist = index_to_obj(cachep, page, objfreelist) +
2524 obj_offset(cachep);
2525 count--;
2526 }
2527
2528 /*
2529 * On early boot, generate the list dynamically.
2530 * Later use a pre-computed list for speed.
2531 */
2532 if (!precomputed) {
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002533 for (i = 0; i < count; i++)
2534 set_free_obj(page, i, i);
2535
2536 /* Fisher-Yates shuffle */
2537 for (i = count - 1; i > 0; i--) {
2538 rand = prandom_u32_state(&state.rnd_state);
2539 rand %= (i + 1);
2540 swap_free_obj(page, i, rand);
2541 }
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002542 } else {
2543 for (i = 0; i < count; i++)
2544 set_free_obj(page, i, next_random_slot(&state));
2545 }
2546
2547 if (OBJFREELIST_SLAB(cachep))
2548 set_free_obj(page, cachep->num - 1, objfreelist);
2549
2550 return true;
2551}
2552#else
2553static inline bool shuffle_freelist(struct kmem_cache *cachep,
2554 struct page *page)
2555{
2556 return false;
2557}
2558#endif /* CONFIG_SLAB_FREELIST_RANDOM */
2559
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002560static void cache_init_objs(struct kmem_cache *cachep,
2561 struct page *page)
2562{
2563 int i;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002564 void *objp;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002565 bool shuffled;
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002566
2567 cache_init_objs_debug(cachep, page);
2568
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002569 /* Try to randomize the freelist if enabled */
2570 shuffled = shuffle_freelist(cachep, page);
2571
2572 if (!shuffled && OBJFREELIST_SLAB(cachep)) {
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002573 page->freelist = index_to_obj(cachep, page, cachep->num - 1) +
2574 obj_offset(cachep);
2575 }
2576
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002577 for (i = 0; i < cachep->num; i++) {
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07002578 objp = index_to_obj(cachep, page, i);
2579 kasan_init_slab_obj(cachep, objp);
2580
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002581 /* constructor could break poison info */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002582 if (DEBUG == 0 && cachep->ctor) {
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002583 kasan_unpoison_object_data(cachep, objp);
2584 cachep->ctor(objp);
2585 kasan_poison_object_data(cachep, objp);
2586 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002587
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002588 if (!shuffled)
2589 set_free_obj(page, i, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591}
2592
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002593static void *slab_get_obj(struct kmem_cache *cachep, struct page *page)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002594{
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002595 void *objp;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002596
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002597 objp = index_to_obj(cachep, page, get_free_obj(page, page->active));
Joonsoo Kim8456a642013-10-24 10:07:49 +09002598 page->active++;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002599
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002600#if DEBUG
2601 if (cachep->flags & SLAB_STORE_USER)
2602 set_store_user_dirty(cachep);
2603#endif
2604
Matthew Dobson78d382d2006-02-01 03:05:47 -08002605 return objp;
2606}
2607
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002608static void slab_put_obj(struct kmem_cache *cachep,
2609 struct page *page, void *objp)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002610{
Joonsoo Kim8456a642013-10-24 10:07:49 +09002611 unsigned int objnr = obj_to_index(cachep, page, objp);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002612#if DEBUG
Joonsoo Kim16025172013-10-24 10:07:46 +09002613 unsigned int i;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002614
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002615 /* Verify double free bug */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002616 for (i = page->active; i < cachep->num; i++) {
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002617 if (get_free_obj(page, i) == objnr) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002618 pr_err("slab: double free detected in cache '%s', objp %px\n",
Joe Perches756a0252016-03-17 14:19:47 -07002619 cachep->name, objp);
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002620 BUG();
2621 }
Matthew Dobson78d382d2006-02-01 03:05:47 -08002622 }
2623#endif
Joonsoo Kim8456a642013-10-24 10:07:49 +09002624 page->active--;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002625 if (!page->freelist)
2626 page->freelist = objp + obj_offset(cachep);
2627
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002628 set_free_obj(page, page->active, objnr);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002629}
2630
Pekka Enberg47768742006-06-23 02:03:07 -07002631/*
2632 * Map pages beginning at addr to the given cache and slab. This is required
2633 * for the slab allocator to be able to lookup the cache and slab of a
Nick Pigginccd35fb2011-01-07 17:49:17 +11002634 * virtual address for kfree, ksize, and slab debugging.
Pekka Enberg47768742006-06-23 02:03:07 -07002635 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002636static void slab_map_pages(struct kmem_cache *cache, struct page *page,
Joonsoo Kim7e007352013-10-30 19:04:01 +09002637 void *freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638{
Joonsoo Kima57a4982013-10-24 10:07:44 +09002639 page->slab_cache = cache;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002640 page->freelist = freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641}
2642
2643/*
2644 * Grow (by 1) the number of slabs within a cache. This is called by
2645 * kmem_cache_alloc() when there are no active objs left in a cache.
2646 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002647static struct page *cache_grow_begin(struct kmem_cache *cachep,
2648 gfp_t flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002650 void *freelist;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002651 size_t offset;
2652 gfp_t local_flags;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002653 int page_node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002654 struct kmem_cache_node *n;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002655 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656
Andrew Mortona737b3e2006-03-22 00:08:11 -08002657 /*
2658 * Be lazy and only check for valid flags here, keeping it out of the
2659 * critical path in kmem_cache_alloc().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 */
Andrew Mortonc871ac42014-12-10 15:42:25 -08002661 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07002662 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07002663 flags &= ~GFP_SLAB_BUG_MASK;
2664 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
2665 invalid_mask, &invalid_mask, flags, &flags);
2666 dump_stack();
Andrew Mortonc871ac42014-12-10 15:42:25 -08002667 }
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696 /* Get slab management. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002697 freelist = alloc_slabmgmt(cachep, page, offset,
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002698 local_flags & ~GFP_CONSTRAINT_MASK, page_node);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002699 if (OFF_SLAB(cachep) && !freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 goto opps1;
2701
Joonsoo Kim8456a642013-10-24 10:07:49 +09002702 slab_map_pages(cachep, page, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002704 kasan_poison_slab(page);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002705 cache_init_objs(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Mel Gormand0164ad2015-11-06 16:28:21 -08002707 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002710 return page;
2711
Andrew Mortona737b3e2006-03-22 00:08:11 -08002712opps1:
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002713 kmem_freepages(cachep, page);
Andrew Mortona737b3e2006-03-22 00:08:11 -08002714failed:
Mel Gormand0164ad2015-11-06 16:28:21 -08002715 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 local_irq_disable();
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002717 return NULL;
2718}
2719
2720static void cache_grow_end(struct kmem_cache *cachep, struct page *page)
2721{
2722 struct kmem_cache_node *n;
2723 void *list = NULL;
2724
2725 check_irq_off();
2726
2727 if (!page)
2728 return;
2729
2730 INIT_LIST_HEAD(&page->lru);
2731 n = get_node(cachep, page_to_nid(page));
2732
2733 spin_lock(&n->list_lock);
David Rientjesbf00bd32016-12-12 16:41:44 -08002734 n->total_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002735 if (!page->active) {
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002736 list_add_tail(&page->lru, &(n->slabs_free));
Greg Thelenf728b0a2016-12-12 16:41:41 -08002737 n->free_slabs++;
David Rientjesbf00bd32016-12-12 16:41:44 -08002738 } else
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002739 fixup_slab_list(cachep, n, page, &list);
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07002740
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002741 STATS_INC_GROWN(cachep);
2742 n->free_objects += cachep->num - page->active;
2743 spin_unlock(&n->list_lock);
2744
2745 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746}
2747
2748#if DEBUG
2749
2750/*
2751 * Perform extra freeing checks:
2752 * - detect bad pointers.
2753 * - POISON/RED_ZONE checking
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 */
2755static void kfree_debugcheck(const void *objp)
2756{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 if (!virt_addr_valid(objp)) {
Joe Perches11705322016-03-17 14:19:50 -07002758 pr_err("kfree_debugcheck: out of range ptr %lxh\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002759 (unsigned long)objp);
2760 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762}
2763
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002764static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
2765{
David Woodhouseb46b8f12007-05-08 00:22:59 -07002766 unsigned long long redzone1, redzone2;
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002767
2768 redzone1 = *dbg_redzone1(cache, obj);
2769 redzone2 = *dbg_redzone2(cache, obj);
2770
2771 /*
2772 * Redzone is ok.
2773 */
2774 if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
2775 return;
2776
2777 if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
2778 slab_error(cache, "double free detected");
2779 else
2780 slab_error(cache, "memory outside object was overwritten");
2781
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002782 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07002783 obj, redzone1, redzone2);
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002784}
2785
Pekka Enberg343e0d72006-02-01 03:05:50 -08002786static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002787 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 unsigned int objnr;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002790 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Matthew Wilcox80cbd912007-11-29 12:05:13 -07002792 BUG_ON(virt_to_cache(objp) != cachep);
2793
Manfred Spraul3dafccf2006-02-01 03:05:42 -08002794 objp -= obj_offset(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795 kfree_debugcheck(objp);
Christoph Lameterb49af682007-05-06 14:49:41 -07002796 page = virt_to_head_page(objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 if (cachep->flags & SLAB_RED_ZONE) {
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002799 verify_redzone_free(cachep, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2801 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2802 }
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002803 if (cachep->flags & SLAB_STORE_USER) {
2804 set_store_user_dirty(cachep);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002805 *dbg_userword(cachep, objp) = (void *)caller;
Joonsoo Kimd31676d2016-03-15 14:54:24 -07002806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807
Joonsoo Kim8456a642013-10-24 10:07:49 +09002808 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
2810 BUG_ON(objnr >= cachep->num);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002811 BUG_ON(objp != index_to_obj(cachep, page, objnr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 poison_obj(cachep, objp, POISON_FREE);
Joonsoo Kim40b44132016-03-15 14:54:21 -07002815 slab_kernel_map(cachep, objp, 0, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 }
2817 return objp;
2818}
2819
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820#else
2821#define kfree_debugcheck(x) do { } while(0)
2822#define cache_free_debugcheck(x,objp,z) (objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823#endif
2824
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002825static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
2826 void **list)
2827{
2828#if DEBUG
2829 void *next = *list;
2830 void *objp;
2831
2832 while (next) {
2833 objp = next - obj_offset(cachep);
2834 next = *(void **)next;
2835 poison_obj(cachep, objp, POISON_FREE);
2836 }
2837#endif
2838}
2839
Joonsoo Kimd8410232016-03-15 14:54:44 -07002840static inline void fixup_slab_list(struct kmem_cache *cachep,
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002841 struct kmem_cache_node *n, struct page *page,
2842 void **list)
Joonsoo Kimd8410232016-03-15 14:54:44 -07002843{
2844 /* move slabp to correct slabp list: */
2845 list_del(&page->lru);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002846 if (page->active == cachep->num) {
Joonsoo Kimd8410232016-03-15 14:54:44 -07002847 list_add(&page->lru, &n->slabs_full);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002848 if (OBJFREELIST_SLAB(cachep)) {
2849#if DEBUG
2850 /* Poisoning will be done without holding the lock */
2851 if (cachep->flags & SLAB_POISON) {
2852 void **objp = page->freelist;
2853
2854 *objp = *list;
2855 *list = objp;
2856 }
2857#endif
2858 page->freelist = NULL;
2859 }
2860 } else
Joonsoo Kimd8410232016-03-15 14:54:44 -07002861 list_add(&page->lru, &n->slabs_partial);
2862}
2863
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002864/* Try to find non-pfmemalloc slab if needed */
2865static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n,
David Rientjesbf00bd32016-12-12 16:41:44 -08002866 struct page *page, bool pfmemalloc)
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002867{
2868 if (!page)
2869 return NULL;
2870
2871 if (pfmemalloc)
2872 return page;
2873
2874 if (!PageSlabPfmemalloc(page))
2875 return page;
2876
2877 /* No need to keep pfmemalloc slab if we have enough free objects */
2878 if (n->free_objects > n->free_limit) {
2879 ClearPageSlabPfmemalloc(page);
2880 return page;
2881 }
2882
2883 /* Move pfmemalloc slab to the end of list to speed up next search */
2884 list_del(&page->lru);
David Rientjesbf00bd32016-12-12 16:41:44 -08002885 if (!page->active) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002886 list_add_tail(&page->lru, &n->slabs_free);
David Rientjesbf00bd32016-12-12 16:41:44 -08002887 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002888 } else
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002889 list_add_tail(&page->lru, &n->slabs_partial);
2890
2891 list_for_each_entry(page, &n->slabs_partial, lru) {
2892 if (!PageSlabPfmemalloc(page))
2893 return page;
2894 }
2895
Greg Thelenf728b0a2016-12-12 16:41:41 -08002896 n->free_touched = 1;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002897 list_for_each_entry(page, &n->slabs_free, lru) {
Greg Thelenf728b0a2016-12-12 16:41:41 -08002898 if (!PageSlabPfmemalloc(page)) {
David Rientjesbf00bd32016-12-12 16:41:44 -08002899 n->free_slabs--;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002900 return page;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002901 }
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002902 }
2903
2904 return NULL;
2905}
2906
2907static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc)
Geliang Tang7aa0d222016-01-14 15:18:02 -08002908{
2909 struct page *page;
2910
Greg Thelenf728b0a2016-12-12 16:41:41 -08002911 assert_spin_locked(&n->list_lock);
David Rientjesbf00bd32016-12-12 16:41:44 -08002912 page = list_first_entry_or_null(&n->slabs_partial, struct page, lru);
Geliang Tang7aa0d222016-01-14 15:18:02 -08002913 if (!page) {
2914 n->free_touched = 1;
David Rientjesbf00bd32016-12-12 16:41:44 -08002915 page = list_first_entry_or_null(&n->slabs_free, struct page,
2916 lru);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002917 if (page)
David Rientjesbf00bd32016-12-12 16:41:44 -08002918 n->free_slabs--;
Geliang Tang7aa0d222016-01-14 15:18:02 -08002919 }
2920
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002921 if (sk_memalloc_socks())
David Rientjesbf00bd32016-12-12 16:41:44 -08002922 page = get_valid_first_slab(n, page, pfmemalloc);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002923
Geliang Tang7aa0d222016-01-14 15:18:02 -08002924 return page;
2925}
2926
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002927static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep,
2928 struct kmem_cache_node *n, gfp_t flags)
2929{
2930 struct page *page;
2931 void *obj;
2932 void *list = NULL;
2933
2934 if (!gfp_pfmemalloc_allowed(flags))
2935 return NULL;
2936
2937 spin_lock(&n->list_lock);
2938 page = get_first_slab(n, true);
2939 if (!page) {
2940 spin_unlock(&n->list_lock);
2941 return NULL;
2942 }
2943
2944 obj = slab_get_obj(cachep, page);
2945 n->free_objects--;
2946
2947 fixup_slab_list(cachep, n, page, &list);
2948
2949 spin_unlock(&n->list_lock);
2950 fixup_objfreelist_debug(cachep, &list);
2951
2952 return obj;
2953}
2954
Joonsoo Kim213b4692016-05-19 17:10:29 -07002955/*
2956 * Slab list should be fixed up by fixup_slab_list() for existing slab
2957 * or cache_grow_end() for new slab
2958 */
2959static __always_inline int alloc_block(struct kmem_cache *cachep,
2960 struct array_cache *ac, struct page *page, int batchcount)
2961{
2962 /*
2963 * There must be at least one object available for
2964 * allocation.
2965 */
2966 BUG_ON(page->active >= cachep->num);
2967
2968 while (page->active < cachep->num && batchcount--) {
2969 STATS_INC_ALLOCED(cachep);
2970 STATS_INC_ACTIVE(cachep);
2971 STATS_SET_HIGH(cachep);
2972
2973 ac->entry[ac->avail++] = slab_get_obj(cachep, page);
2974 }
2975
2976 return batchcount;
2977}
2978
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002979static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980{
2981 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002982 struct kmem_cache_node *n;
Joonsoo Kim801faf02016-05-19 17:10:31 -07002983 struct array_cache *ac, *shared;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002984 int node;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002985 void *list = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002986 struct page *page;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002987
Joe Korty6d2144d2008-03-05 15:04:59 -08002988 check_irq_off();
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002989 node = numa_mem_id();
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002990
Joe Korty6d2144d2008-03-05 15:04:59 -08002991 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992 batchcount = ac->batchcount;
2993 if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08002994 /*
2995 * If there was little recent activity on this cache, then
2996 * perform only a partial refill. Otherwise we could generate
2997 * refill bouncing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998 */
2999 batchcount = BATCHREFILL_LIMIT;
3000 }
Christoph Lameter18bf8542014-08-06 16:04:11 -07003001 n = get_node(cachep, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003003 BUG_ON(ac->avail > 0 || !n);
Joonsoo Kim801faf02016-05-19 17:10:31 -07003004 shared = READ_ONCE(n->shared);
3005 if (!n->free_objects && (!shared || !shared->avail))
3006 goto direct_grow;
3007
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003008 spin_lock(&n->list_lock);
Joonsoo Kim801faf02016-05-19 17:10:31 -07003009 shared = READ_ONCE(n->shared);
Christoph Lametere498be72005-09-09 13:03:32 -07003010
Christoph Lameter3ded1752006-03-25 03:06:44 -08003011 /* See if we can refill from the shared array */
Joonsoo Kim801faf02016-05-19 17:10:31 -07003012 if (shared && transfer_objects(ac, shared, batchcount)) {
3013 shared->touched = 1;
Christoph Lameter3ded1752006-03-25 03:06:44 -08003014 goto alloc_done;
Nick Piggin44b57f12010-01-27 22:27:40 +11003015 }
Christoph Lameter3ded1752006-03-25 03:06:44 -08003016
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 while (batchcount > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 /* Get slab alloc is to come from. */
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003019 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08003020 if (!page)
3021 goto must_grow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023 check_spinlock_acquired(cachep);
Pekka Enberg714b81712007-05-06 14:49:03 -07003024
Joonsoo Kim213b4692016-05-19 17:10:29 -07003025 batchcount = alloc_block(cachep, ac, page, batchcount);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003026 fixup_slab_list(cachep, n, page, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 }
3028
Andrew Mortona737b3e2006-03-22 00:08:11 -08003029must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003030 n->free_objects -= ac->avail;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003031alloc_done:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003032 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003033 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034
Joonsoo Kim801faf02016-05-19 17:10:31 -07003035direct_grow:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 if (unlikely(!ac->avail)) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003037 /* Check if we can use obj in pfmemalloc slab */
3038 if (sk_memalloc_socks()) {
3039 void *obj = cache_alloc_pfmemalloc(cachep, n, flags);
3040
3041 if (obj)
3042 return obj;
3043 }
3044
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003045 page = cache_grow_begin(cachep, gfp_exact_node(flags), node);
Christoph Lametere498be72005-09-09 13:03:32 -07003046
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003047 /*
3048 * cache_grow_begin() can reenable interrupts,
3049 * then ac could change.
3050 */
Pekka Enberg9a2dba42006-02-01 03:05:49 -08003051 ac = cpu_cache_get(cachep);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003052 if (!ac->avail && page)
3053 alloc_block(cachep, ac, page, batchcount);
3054 cache_grow_end(cachep, page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003055
Joonsoo Kim213b4692016-05-19 17:10:29 -07003056 if (!ac->avail)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 }
3059 ac->touched = 1;
Mel Gorman072bb0a2012-07-31 16:43:58 -07003060
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003061 return ac->entry[--ac->avail];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062}
3063
Andrew Mortona737b3e2006-03-22 00:08:11 -08003064static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
3065 gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066{
Mel Gormand0164ad2015-11-06 16:28:21 -08003067 might_sleep_if(gfpflags_allow_blocking(flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068}
3069
3070#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08003071static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003072 gfp_t flags, void *objp, unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003074 if (!objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075 return objp;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003076 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 check_poison_obj(cachep, objp);
Joonsoo Kim40b44132016-03-15 14:54:21 -07003078 slab_kernel_map(cachep, objp, 1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079 poison_obj(cachep, objp, POISON_INUSE);
3080 }
3081 if (cachep->flags & SLAB_STORE_USER)
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003082 *dbg_userword(cachep, objp) = (void *)caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
3084 if (cachep->flags & SLAB_RED_ZONE) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08003085 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
3086 *dbg_redzone2(cachep, objp) != RED_INACTIVE) {
Joe Perches756a0252016-03-17 14:19:47 -07003087 slab_error(cachep, "double free, or memory outside object was overwritten");
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003088 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07003089 objp, *dbg_redzone1(cachep, objp),
3090 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 }
3092 *dbg_redzone1(cachep, objp) = RED_ACTIVE;
3093 *dbg_redzone2(cachep, objp) = RED_ACTIVE;
3094 }
Joonsoo Kim03787302014-06-23 13:22:06 -07003095
Manfred Spraul3dafccf2006-02-01 03:05:42 -08003096 objp += obj_offset(cachep);
Christoph Lameter4f104932007-05-06 14:50:17 -07003097 if (cachep->ctor && cachep->flags & SLAB_POISON)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003098 cachep->ctor(objp);
Tetsuo Handa7ea466f2011-07-21 09:42:45 +09003099 if (ARCH_SLAB_MINALIGN &&
3100 ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003101 pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
Hugh Dickinsc2251502011-07-11 13:35:08 -07003102 objp, (int)ARCH_SLAB_MINALIGN);
Kevin Hilmana44b56d2006-12-06 20:32:11 -08003103 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104 return objp;
3105}
3106#else
3107#define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
3108#endif
3109
Pekka Enberg343e0d72006-02-01 03:05:50 -08003110static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003112 void *objp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 struct array_cache *ac;
3114
Alok N Kataria5c382302005-09-27 21:45:46 -07003115 check_irq_off();
Akinobu Mita8a8b6502006-12-08 02:39:44 -08003116
Pekka Enberg9a2dba42006-02-01 03:05:49 -08003117 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 if (likely(ac->avail)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003119 ac->touched = 1;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003120 objp = ac->entry[--ac->avail];
Mel Gorman072bb0a2012-07-31 16:43:58 -07003121
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003122 STATS_INC_ALLOCHIT(cachep);
3123 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 }
Mel Gorman072bb0a2012-07-31 16:43:58 -07003125
3126 STATS_INC_ALLOCMISS(cachep);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003127 objp = cache_alloc_refill(cachep, flags);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003128 /*
3129 * the 'ac' may be updated by cache_alloc_refill(),
3130 * and kmemleak_erase() requires its correct value.
3131 */
3132 ac = cpu_cache_get(cachep);
3133
3134out:
Catalin Marinasd5cff632009-06-11 13:22:40 +01003135 /*
3136 * To avoid a false negative, if an object that is in one of the
3137 * per-CPU caches is leaked, we need to make sure kmemleak doesn't
3138 * treat the array pointers as a reference to the object.
3139 */
J. R. Okajimaf3d8b532009-12-02 16:55:49 +09003140 if (objp)
3141 kmemleak_erase(&ac->entry[ac->avail]);
Alok N Kataria5c382302005-09-27 21:45:46 -07003142 return objp;
3143}
3144
Christoph Lametere498be72005-09-09 13:03:32 -07003145#ifdef CONFIG_NUMA
3146/*
Zefan Li2ad654b2014-09-25 09:41:02 +08003147 * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set.
Paul Jacksonc61afb12006-03-24 03:16:08 -08003148 *
3149 * If we are in_interrupt, then process context, including cpusets and
3150 * mempolicy, may not apply and should not be used for allocation policy.
3151 */
3152static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
3153{
3154 int nid_alloc, nid_here;
3155
Christoph Lameter765c4502006-09-27 01:50:08 -07003156 if (in_interrupt() || (flags & __GFP_THISNODE))
Paul Jacksonc61afb12006-03-24 03:16:08 -08003157 return NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003158 nid_alloc = nid_here = numa_mem_id();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003159 if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
Jack Steiner6adef3e2010-05-26 14:42:49 -07003160 nid_alloc = cpuset_slab_spread_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003161 else if (current->mempolicy)
David Rientjes2a389612014-04-07 15:37:29 -07003162 nid_alloc = mempolicy_slab_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003163 if (nid_alloc != nid_here)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003164 return ____cache_alloc_node(cachep, flags, nid_alloc);
Paul Jacksonc61afb12006-03-24 03:16:08 -08003165 return NULL;
3166}
3167
3168/*
Christoph Lameter765c4502006-09-27 01:50:08 -07003169 * Fallback function if there was no memory available and no objects on a
Christoph Lameter3c517a62006-12-06 20:33:29 -08003170 * certain node and fall back is permitted. First we scan all the
Christoph Lameter6a673682013-01-10 19:14:19 +00003171 * available node for available objects. If that fails then we
Christoph Lameter3c517a62006-12-06 20:33:29 -08003172 * perform an allocation without specifying a node. This allows the page
3173 * allocator to do its reclaim / fallback magic. We then insert the
3174 * slab into the proper nodelist and then allocate from it.
Christoph Lameter765c4502006-09-27 01:50:08 -07003175 */
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003176static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
Christoph Lameter765c4502006-09-27 01:50:08 -07003177{
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003178 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07003179 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003180 struct zone *zone;
3181 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter765c4502006-09-27 01:50:08 -07003182 void *obj = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003183 struct page *page;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003184 int nid;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003185 unsigned int cpuset_mems_cookie;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003186
3187 if (flags & __GFP_THISNODE)
3188 return NULL;
3189
Mel Gormancc9a6c82012-03-21 16:34:11 -07003190retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003191 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07003192 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003193
Christoph Lameter3c517a62006-12-06 20:33:29 -08003194retry:
3195 /*
3196 * Look through allowed nodes for objects available
3197 * from existing per node queues.
3198 */
Mel Gorman54a6eb52008-04-28 02:12:16 -07003199 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
3200 nid = zone_to_nid(zone);
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003201
Vladimir Davydov061d7072014-12-12 16:58:25 -08003202 if (cpuset_zone_allowed(zone, flags) &&
Christoph Lameter18bf8542014-08-06 16:04:11 -07003203 get_node(cache, nid) &&
3204 get_node(cache, nid)->free_objects) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003205 obj = ____cache_alloc_node(cache,
David Rientjes4167e9b2015-04-14 15:46:55 -07003206 gfp_exact_node(flags), nid);
Christoph Lameter481c5342008-06-21 16:46:35 -07003207 if (obj)
3208 break;
3209 }
Christoph Lameter3c517a62006-12-06 20:33:29 -08003210 }
3211
Christoph Lametercfce6602007-05-06 14:50:17 -07003212 if (!obj) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003213 /*
3214 * This allocation will be performed within the constraints
3215 * of the current cpuset / memory policy requirements.
3216 * We may trigger various forms of reclaim on the allowed
3217 * set and go into memory reserves if necessary.
3218 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003219 page = cache_grow_begin(cache, flags, numa_mem_id());
3220 cache_grow_end(cache, page);
3221 if (page) {
3222 nid = page_to_nid(page);
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003223 obj = ____cache_alloc_node(cache,
3224 gfp_exact_node(flags), nid);
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09003225
Christoph Lameter3c517a62006-12-06 20:33:29 -08003226 /*
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003227 * Another processor may allocate the objects in
3228 * the slab since we are not holding any locks.
Christoph Lameter3c517a62006-12-06 20:33:29 -08003229 */
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003230 if (!obj)
3231 goto retry;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003232 }
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003233 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003234
Mel Gormand26914d2014-04-03 14:47:24 -07003235 if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003236 goto retry_cpuset;
Christoph Lameter765c4502006-09-27 01:50:08 -07003237 return obj;
3238}
3239
3240/*
Christoph Lametere498be72005-09-09 13:03:32 -07003241 * A interface to enable slab creation on nodeid
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003243static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
Andrew Mortona737b3e2006-03-22 00:08:11 -08003244 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07003245{
Joonsoo Kim8456a642013-10-24 10:07:49 +09003246 struct page *page;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003247 struct kmem_cache_node *n;
Joonsoo Kim213b4692016-05-19 17:10:29 -07003248 void *obj = NULL;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003249 void *list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003250
Paul Mackerras7c3fbbd2014-12-02 15:59:48 -08003251 VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003252 n = get_node(cachep, nodeid);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003253 BUG_ON(!n);
Christoph Lametere498be72005-09-09 13:03:32 -07003254
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08003255 check_irq_off();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003256 spin_lock(&n->list_lock);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003257 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08003258 if (!page)
3259 goto must_grow;
Christoph Lametere498be72005-09-09 13:03:32 -07003260
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003261 check_spinlock_acquired_node(cachep, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07003262
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003263 STATS_INC_NODEALLOCS(cachep);
3264 STATS_INC_ACTIVE(cachep);
3265 STATS_SET_HIGH(cachep);
Christoph Lametere498be72005-09-09 13:03:32 -07003266
Joonsoo Kim8456a642013-10-24 10:07:49 +09003267 BUG_ON(page->active == cachep->num);
Christoph Lametere498be72005-09-09 13:03:32 -07003268
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003269 obj = slab_get_obj(cachep, page);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003270 n->free_objects--;
Christoph Lametere498be72005-09-09 13:03:32 -07003271
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003272 fixup_slab_list(cachep, n, page, &list);
Christoph Lametere498be72005-09-09 13:03:32 -07003273
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003274 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003275 fixup_objfreelist_debug(cachep, &list);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003276 return obj;
Christoph Lametere498be72005-09-09 13:03:32 -07003277
Andrew Mortona737b3e2006-03-22 00:08:11 -08003278must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003279 spin_unlock(&n->list_lock);
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003280 page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003281 if (page) {
3282 /* This slab isn't counted yet so don't update free_objects */
3283 obj = slab_get_obj(cachep, page);
3284 }
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003285 cache_grow_end(cachep, page);
Christoph Lametere498be72005-09-09 13:03:32 -07003286
Joonsoo Kim213b4692016-05-19 17:10:29 -07003287 return obj ? obj : fallback_alloc(cachep, flags);
Christoph Lametere498be72005-09-09 13:03:32 -07003288}
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003289
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003290static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003291slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003292 unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003293{
3294 unsigned long save_flags;
3295 void *ptr;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003296 int slab_node = numa_mem_id();
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003297
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003298 flags &= gfp_allowed_mask;
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003299 cachep = slab_pre_alloc_hook(cachep, flags);
3300 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003301 return NULL;
3302
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003303 cache_alloc_debugcheck_before(cachep, flags);
3304 local_irq_save(save_flags);
3305
Andrew Mortoneacbbae2011-07-28 13:59:49 -07003306 if (nodeid == NUMA_NO_NODE)
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003307 nodeid = slab_node;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003308
Christoph Lameter18bf8542014-08-06 16:04:11 -07003309 if (unlikely(!get_node(cachep, nodeid))) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003310 /* Node not bootstrapped yet */
3311 ptr = fallback_alloc(cachep, flags);
3312 goto out;
3313 }
3314
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003315 if (nodeid == slab_node) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003316 /*
3317 * Use the locally cached objects if possible.
3318 * However ____cache_alloc does not allow fallback
3319 * to other nodes. It may fail while we still have
3320 * objects on other nodes available.
3321 */
3322 ptr = ____cache_alloc(cachep, flags);
3323 if (ptr)
3324 goto out;
3325 }
3326 /* ___cache_alloc_node can fall back to other nodes */
3327 ptr = ____cache_alloc_node(cachep, flags, nodeid);
3328 out:
3329 local_irq_restore(save_flags);
3330 ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
3331
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003332 if (unlikely(flags & __GFP_ZERO) && ptr)
3333 memset(ptr, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003334
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003335 slab_post_alloc_hook(cachep, flags, 1, &ptr);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003336 return ptr;
3337}
3338
3339static __always_inline void *
3340__do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
3341{
3342 void *objp;
3343
Zefan Li2ad654b2014-09-25 09:41:02 +08003344 if (current->mempolicy || cpuset_do_slab_mem_spread()) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003345 objp = alternate_node_alloc(cache, flags);
3346 if (objp)
3347 goto out;
3348 }
3349 objp = ____cache_alloc(cache, flags);
3350
3351 /*
3352 * We may just have run out of memory on the local node.
3353 * ____cache_alloc_node() knows how to locate memory on other nodes
3354 */
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003355 if (!objp)
3356 objp = ____cache_alloc_node(cache, flags, numa_mem_id());
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003357
3358 out:
3359 return objp;
3360}
3361#else
3362
3363static __always_inline void *
3364__do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3365{
3366 return ____cache_alloc(cachep, flags);
3367}
3368
3369#endif /* CONFIG_NUMA */
3370
3371static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003372slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003373{
3374 unsigned long save_flags;
3375 void *objp;
3376
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003377 flags &= gfp_allowed_mask;
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003378 cachep = slab_pre_alloc_hook(cachep, flags);
3379 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003380 return NULL;
3381
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003382 cache_alloc_debugcheck_before(cachep, flags);
3383 local_irq_save(save_flags);
3384 objp = __do_cache_alloc(cachep, flags);
3385 local_irq_restore(save_flags);
3386 objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
3387 prefetchw(objp);
3388
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003389 if (unlikely(flags & __GFP_ZERO) && objp)
3390 memset(objp, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003391
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003392 slab_post_alloc_hook(cachep, flags, 1, &objp);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003393 return objp;
3394}
Christoph Lametere498be72005-09-09 13:03:32 -07003395
3396/*
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08003397 * Caller needs to acquire correct kmem_cache_node's list_lock
Joonsoo Kim97654df2014-08-06 16:04:25 -07003398 * @list: List of detached free slabs should be freed by caller
Christoph Lametere498be72005-09-09 13:03:32 -07003399 */
Joonsoo Kim97654df2014-08-06 16:04:25 -07003400static void free_block(struct kmem_cache *cachep, void **objpp,
3401 int nr_objects, int node, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402{
3403 int i;
Joonsoo Kim25c063f2014-08-06 16:04:22 -07003404 struct kmem_cache_node *n = get_node(cachep, node);
Joonsoo Kim6052b782016-05-19 17:10:17 -07003405 struct page *page;
3406
3407 n->free_objects += nr_objects;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408
3409 for (i = 0; i < nr_objects; i++) {
Mel Gorman072bb0a2012-07-31 16:43:58 -07003410 void *objp;
Joonsoo Kim8456a642013-10-24 10:07:49 +09003411 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Mel Gorman072bb0a2012-07-31 16:43:58 -07003413 objp = objpp[i];
3414
Joonsoo Kim8456a642013-10-24 10:07:49 +09003415 page = virt_to_head_page(objp);
Joonsoo Kim8456a642013-10-24 10:07:49 +09003416 list_del(&page->lru);
Christoph Lameterff694162005-09-22 21:44:02 -07003417 check_spinlock_acquired_node(cachep, node);
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003418 slab_put_obj(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 STATS_DEC_ACTIVE(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003420
3421 /* fixup slab chains */
Greg Thelenf728b0a2016-12-12 16:41:41 -08003422 if (page->active == 0) {
Joonsoo Kim6052b782016-05-19 17:10:17 -07003423 list_add(&page->lru, &n->slabs_free);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003424 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08003425 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 /* Unconditionally move a slab to the end of the
3427 * partial list on free - maximum time for the
3428 * other objects to be freed, too.
3429 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09003430 list_add_tail(&page->lru, &n->slabs_partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 }
3432 }
Joonsoo Kim6052b782016-05-19 17:10:17 -07003433
3434 while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) {
3435 n->free_objects -= cachep->num;
3436
3437 page = list_last_entry(&n->slabs_free, struct page, lru);
Wei Yongjunde24bae2016-07-26 15:22:11 -07003438 list_move(&page->lru, list);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003439 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08003440 n->total_slabs--;
Joonsoo Kim6052b782016-05-19 17:10:17 -07003441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442}
3443
Pekka Enberg343e0d72006-02-01 03:05:50 -08003444static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445{
3446 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003447 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003448 int node = numa_mem_id();
Joonsoo Kim97654df2014-08-06 16:04:25 -07003449 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
3451 batchcount = ac->batchcount;
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003452
Linus Torvalds1da177e2005-04-16 15:20:36 -07003453 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07003454 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003455 spin_lock(&n->list_lock);
3456 if (n->shared) {
3457 struct array_cache *shared_array = n->shared;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003458 int max = shared_array->limit - shared_array->avail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 if (max) {
3460 if (batchcount > max)
3461 batchcount = max;
Christoph Lametere498be72005-09-09 13:03:32 -07003462 memcpy(&(shared_array->entry[shared_array->avail]),
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003463 ac->entry, sizeof(void *) * batchcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 shared_array->avail += batchcount;
3465 goto free_done;
3466 }
3467 }
3468
Joonsoo Kim97654df2014-08-06 16:04:25 -07003469 free_block(cachep, ac->entry, batchcount, node, &list);
Andrew Mortona737b3e2006-03-22 00:08:11 -08003470free_done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003471#if STATS
3472 {
3473 int i = 0;
Geliang Tang73c02192016-01-14 15:17:59 -08003474 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Geliang Tang73c02192016-01-14 15:17:59 -08003476 list_for_each_entry(page, &n->slabs_free, lru) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09003477 BUG_ON(page->active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478
3479 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 }
3481 STATS_SET_FREEABLE(cachep, i);
3482 }
3483#endif
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003484 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07003485 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486 ac->avail -= batchcount;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003487 memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488}
3489
3490/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08003491 * Release an obj back to its cache. If the obj has a constructed state, it must
3492 * be in this state _before_ it is released. Called with disabled ints.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003494static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp,
3495 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496{
Alexander Potapenko55834c52016-05-20 16:59:11 -07003497 /* Put the object into the quarantine, don't touch it for now. */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003498 if (kasan_slab_free(cachep, objp, _RET_IP_))
Alexander Potapenko55834c52016-05-20 16:59:11 -07003499 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500
Alexander Potapenko55834c52016-05-20 16:59:11 -07003501 ___cache_free(cachep, objp, caller);
3502}
3503
3504void ___cache_free(struct kmem_cache *cachep, void *objp,
3505 unsigned long caller)
3506{
3507 struct array_cache *ac = cpu_cache_get(cachep);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003508
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509 check_irq_off();
Catalin Marinasd5cff632009-06-11 13:22:40 +01003510 kmemleak_free_recursive(objp, cachep->flags);
Suleiman Souhlala947eb92011-06-02 00:16:42 -07003511 objp = cache_free_debugcheck(cachep, objp, caller);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512
Siddha, Suresh B1807a1a2007-08-22 14:01:49 -07003513 /*
3514 * Skip calling cache_free_alien() when the platform is not numa.
3515 * This will avoid cache misses that happen while accessing slabp (which
3516 * is per page memory reference) to get nodeid. Instead use a global
3517 * variable to skip the call, which is mostly likely to be present in
3518 * the cache.
3519 */
Mel Gormanb6e68bc2009-06-16 15:32:16 -07003520 if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
Pekka Enberg729bd0b2006-06-23 02:03:05 -07003521 return;
Christoph Lametere498be72005-09-09 13:03:32 -07003522
Joonsoo Kim3d880192014-10-09 15:26:04 -07003523 if (ac->avail < ac->limit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003524 STATS_INC_FREEHIT(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 } else {
3526 STATS_INC_FREEMISS(cachep);
3527 cache_flusharray(cachep, ac);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 }
Zhao Jin42c8c992011-08-27 00:26:17 +08003529
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003530 if (sk_memalloc_socks()) {
3531 struct page *page = virt_to_head_page(objp);
3532
3533 if (unlikely(PageSlabPfmemalloc(page))) {
3534 cache_free_pfmemalloc(cachep, page, objp);
3535 return;
3536 }
3537 }
3538
3539 ac->entry[ac->avail++] = objp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540}
3541
3542/**
3543 * kmem_cache_alloc - Allocate an object
3544 * @cachep: The cache to allocate from.
3545 * @flags: See kmalloc().
3546 *
3547 * Allocate an object from this cache. The flags are only relevant
3548 * if the cache has no available objects.
3549 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003550void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003551{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003552 void *ret = slab_alloc(cachep, flags, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003553
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003554 kasan_slab_alloc(cachep, ret, flags);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003555 trace_kmem_cache_alloc(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003556 cachep->object_size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003557
3558 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559}
3560EXPORT_SYMBOL(kmem_cache_alloc);
3561
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003562static __always_inline void
3563cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags,
3564 size_t size, void **p, unsigned long caller)
3565{
3566 size_t i;
3567
3568 for (i = 0; i < size; i++)
3569 p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller);
3570}
3571
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003572int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003573 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003574{
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003575 size_t i;
3576
3577 s = slab_pre_alloc_hook(s, flags);
3578 if (!s)
3579 return 0;
3580
3581 cache_alloc_debugcheck_before(s, flags);
3582
3583 local_irq_disable();
3584 for (i = 0; i < size; i++) {
3585 void *objp = __do_cache_alloc(s, flags);
3586
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003587 if (unlikely(!objp))
3588 goto error;
3589 p[i] = objp;
3590 }
3591 local_irq_enable();
3592
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003593 cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_);
3594
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003595 /* Clear memory outside IRQ disabled section */
3596 if (unlikely(flags & __GFP_ZERO))
3597 for (i = 0; i < size; i++)
3598 memset(p[i], 0, s->object_size);
3599
3600 slab_post_alloc_hook(s, flags, size, p);
3601 /* FIXME: Trace call missing. Christoph would like a bulk variant */
3602 return size;
3603error:
3604 local_irq_enable();
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003605 cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_);
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003606 slab_post_alloc_hook(s, flags, i, p);
3607 __kmem_cache_free_bulk(s, i, p);
3608 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003609}
3610EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3611
Li Zefan0f24f122009-12-11 15:45:30 +08003612#ifdef CONFIG_TRACING
Steven Rostedt85beb582010-11-24 16:23:34 -05003613void *
Ezequiel Garcia40521472012-09-08 17:47:56 -03003614kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003615{
Steven Rostedt85beb582010-11-24 16:23:34 -05003616 void *ret;
3617
Ezequiel Garcia48356302012-09-08 17:47:57 -03003618 ret = slab_alloc(cachep, flags, _RET_IP_);
Steven Rostedt85beb582010-11-24 16:23:34 -05003619
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003620 kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003621 trace_kmalloc(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003622 size, cachep->size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003623 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003624}
Steven Rostedt85beb582010-11-24 16:23:34 -05003625EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003626#endif
3627
Linus Torvalds1da177e2005-04-16 15:20:36 -07003628#ifdef CONFIG_NUMA
Zhouping Liud0d04b72013-05-16 11:36:23 +08003629/**
3630 * kmem_cache_alloc_node - Allocate an object on the specified node
3631 * @cachep: The cache to allocate from.
3632 * @flags: See kmalloc().
3633 * @nodeid: node number of the target node.
3634 *
3635 * Identical to kmem_cache_alloc but it will allocate memory on the given
3636 * node, which can improve the performance for cpu bound structures.
3637 *
3638 * Fallback to other node is possible if __GFP_THISNODE is not set.
3639 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003640void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
3641{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003642 void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003643
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003644 kasan_slab_alloc(cachep, ret, flags);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003645 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003646 cachep->object_size, cachep->size,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003647 flags, nodeid);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003648
3649 return ret;
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003650}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003651EXPORT_SYMBOL(kmem_cache_alloc_node);
3652
Li Zefan0f24f122009-12-11 15:45:30 +08003653#ifdef CONFIG_TRACING
Ezequiel Garcia40521472012-09-08 17:47:56 -03003654void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep,
Steven Rostedt85beb582010-11-24 16:23:34 -05003655 gfp_t flags,
Ezequiel Garcia40521472012-09-08 17:47:56 -03003656 int nodeid,
3657 size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003658{
Steven Rostedt85beb582010-11-24 16:23:34 -05003659 void *ret;
3660
Ezequiel Garcia592f4142012-09-25 08:07:08 -03003661 ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003662
3663 kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003664 trace_kmalloc_node(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003665 size, cachep->size,
Steven Rostedt85beb582010-11-24 16:23:34 -05003666 flags, nodeid);
3667 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003668}
Steven Rostedt85beb582010-11-24 16:23:34 -05003669EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003670#endif
3671
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003672static __always_inline void *
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003673__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003674{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003675 struct kmem_cache *cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003676 void *ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003677
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003678 cachep = kmalloc_slab(size, flags);
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003679 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3680 return cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003681 ret = kmem_cache_alloc_node_trace(cachep, flags, node, size);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003682 kasan_kmalloc(cachep, ret, size, flags);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003683
3684 return ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003685}
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003686
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003687void *__kmalloc_node(size_t size, gfp_t flags, int node)
3688{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003689 return __do_kmalloc_node(size, flags, node, _RET_IP_);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003690}
Christoph Hellwigdbe5e692006-09-25 23:31:36 -07003691EXPORT_SYMBOL(__kmalloc_node);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003692
3693void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003694 int node, unsigned long caller)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003695{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003696 return __do_kmalloc_node(size, flags, node, caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003697}
3698EXPORT_SYMBOL(__kmalloc_node_track_caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003699#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003700
3701/**
Paul Drynoff800590f2006-06-23 02:03:48 -07003702 * __do_kmalloc - allocate memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07003703 * @size: how many bytes of memory are required.
Paul Drynoff800590f2006-06-23 02:03:48 -07003704 * @flags: the type of memory to allocate (see kmalloc).
Randy Dunlap911851e2006-03-22 00:08:14 -08003705 * @caller: function caller for debug tracking of the caller
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706 */
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003707static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003708 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003710 struct kmem_cache *cachep;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003711 void *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003712
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003713 cachep = kmalloc_slab(size, flags);
Linus Torvaldsa5c96d82007-07-19 13:17:15 -07003714 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3715 return cachep;
Ezequiel Garcia48356302012-09-08 17:47:57 -03003716 ret = slab_alloc(cachep, flags, caller);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003717
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003718 kasan_kmalloc(cachep, ret, size, flags);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003719 trace_kmalloc(caller, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003720 size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003721
3722 return ret;
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003723}
3724
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003725void *__kmalloc(size_t size, gfp_t flags)
3726{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003727 return __do_kmalloc(size, flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728}
3729EXPORT_SYMBOL(__kmalloc);
3730
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003731void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003732{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003733 return __do_kmalloc(size, flags, caller);
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003734}
3735EXPORT_SYMBOL(__kmalloc_track_caller);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -07003736
Linus Torvalds1da177e2005-04-16 15:20:36 -07003737/**
3738 * kmem_cache_free - Deallocate an object
3739 * @cachep: The cache the allocation was from.
3740 * @objp: The previously allocated object.
3741 *
3742 * Free an object which was previously allocated from this
3743 * cache.
3744 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003745void kmem_cache_free(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003746{
3747 unsigned long flags;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003748 cachep = cache_from_obj(cachep, objp);
3749 if (!cachep)
3750 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003751
3752 local_irq_save(flags);
Feng Tangd97d4762012-07-02 14:29:10 +08003753 debug_check_no_locks_freed(objp, cachep->object_size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07003754 if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003755 debug_check_no_obj_freed(objp, cachep->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003756 __cache_free(cachep, objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003757 local_irq_restore(flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003758
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003759 trace_kmem_cache_free(_RET_IP_, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003760}
3761EXPORT_SYMBOL(kmem_cache_free);
3762
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003763void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p)
3764{
3765 struct kmem_cache *s;
3766 size_t i;
3767
3768 local_irq_disable();
3769 for (i = 0; i < size; i++) {
3770 void *objp = p[i];
3771
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003772 if (!orig_s) /* called via kfree_bulk */
3773 s = virt_to_cache(objp);
3774 else
3775 s = cache_from_obj(orig_s, objp);
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003776
3777 debug_check_no_locks_freed(objp, s->object_size);
3778 if (!(s->flags & SLAB_DEBUG_OBJECTS))
3779 debug_check_no_obj_freed(objp, s->object_size);
3780
3781 __cache_free(s, objp, _RET_IP_);
3782 }
3783 local_irq_enable();
3784
3785 /* FIXME: add tracing */
3786}
3787EXPORT_SYMBOL(kmem_cache_free_bulk);
3788
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003790 * kfree - free previously allocated memory
3791 * @objp: pointer returned by kmalloc.
3792 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07003793 * If @objp is NULL, no operation is performed.
3794 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003795 * Don't free memory not originally allocated by kmalloc()
3796 * or you will run into trouble.
3797 */
3798void kfree(const void *objp)
3799{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003800 struct kmem_cache *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801 unsigned long flags;
3802
Pekka Enberg2121db72009-03-25 11:05:57 +02003803 trace_kfree(_RET_IP_, objp);
3804
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003805 if (unlikely(ZERO_OR_NULL_PTR(objp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806 return;
3807 local_irq_save(flags);
3808 kfree_debugcheck(objp);
Pekka Enberg6ed5eb2212006-02-01 03:05:49 -08003809 c = virt_to_cache(objp);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003810 debug_check_no_locks_freed(objp, c->object_size);
3811
3812 debug_check_no_obj_freed(objp, c->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003813 __cache_free(c, (void *)objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003814 local_irq_restore(flags);
3815}
3816EXPORT_SYMBOL(kfree);
3817
Christoph Lametere498be72005-09-09 13:03:32 -07003818/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003819 * This initializes kmem_cache_node or resizes various caches for all nodes.
Christoph Lametere498be72005-09-09 13:03:32 -07003820 */
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003821static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -07003822{
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003823 int ret;
Christoph Lametere498be72005-09-09 13:03:32 -07003824 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003825 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07003826
Mel Gorman9c09a952008-01-24 05:49:54 -08003827 for_each_online_node(node) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003828 ret = setup_kmem_cache_node(cachep, node, gfp, true);
3829 if (ret)
Christoph Lametere498be72005-09-09 13:03:32 -07003830 goto fail;
3831
Christoph Lametere498be72005-09-09 13:03:32 -07003832 }
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003833
Christoph Lametercafeb022006-03-25 03:06:46 -08003834 return 0;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003835
Andrew Mortona737b3e2006-03-22 00:08:11 -08003836fail:
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003837 if (!cachep->list.next) {
Christoph Lameter0718dc22006-03-25 03:06:47 -08003838 /* Cache is not active yet. Roll back what we did */
3839 node--;
3840 while (node >= 0) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07003841 n = get_node(cachep, node);
3842 if (n) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003843 kfree(n->shared);
3844 free_alien_cache(n->alien);
3845 kfree(n);
Christoph Lameter6a673682013-01-10 19:14:19 +00003846 cachep->node[node] = NULL;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003847 }
3848 node--;
3849 }
3850 }
Christoph Lametercafeb022006-03-25 03:06:46 -08003851 return -ENOMEM;
Christoph Lametere498be72005-09-09 13:03:32 -07003852}
3853
Christoph Lameter18004c52012-07-06 15:25:12 -05003854/* Always called with the slab_mutex held */
Glauber Costa943a4512012-12-18 14:23:03 -08003855static int __do_tune_cpucache(struct kmem_cache *cachep, int limit,
Pekka Enberg83b519e2009-06-10 19:40:04 +03003856 int batchcount, int shared, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003858 struct array_cache __percpu *cpu_cache, *prev;
3859 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003861 cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount);
3862 if (!cpu_cache)
Siddha, Suresh Bd2e7b7d2006-09-25 23:31:47 -07003863 return -ENOMEM;
3864
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003865 prev = cachep->cpu_cache;
3866 cachep->cpu_cache = cpu_cache;
Greg Thelena87c75f2017-05-03 14:51:47 -07003867 /*
3868 * Without a previous cpu_cache there's no need to synchronize remote
3869 * cpus, so skip the IPIs.
3870 */
3871 if (prev)
3872 kick_all_cpus_sync();
Christoph Lametere498be72005-09-09 13:03:32 -07003873
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 check_irq_on();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 cachep->batchcount = batchcount;
3876 cachep->limit = limit;
Christoph Lametere498be72005-09-09 13:03:32 -07003877 cachep->shared = shared;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003879 if (!prev)
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003880 goto setup_node;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003881
3882 for_each_online_cpu(cpu) {
Joonsoo Kim97654df2014-08-06 16:04:25 -07003883 LIST_HEAD(list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003884 int node;
3885 struct kmem_cache_node *n;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003886 struct array_cache *ac = per_cpu_ptr(prev, cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003887
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003888 node = cpu_to_mem(cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003889 n = get_node(cachep, node);
3890 spin_lock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003891 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003892 spin_unlock_irq(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07003893 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003894 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003895 free_percpu(prev);
3896
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003897setup_node:
3898 return setup_kmem_cache_nodes(cachep, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899}
3900
Glauber Costa943a4512012-12-18 14:23:03 -08003901static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
3902 int batchcount, int shared, gfp_t gfp)
3903{
3904 int ret;
Vladimir Davydov426589f2015-02-12 14:59:23 -08003905 struct kmem_cache *c;
Glauber Costa943a4512012-12-18 14:23:03 -08003906
3907 ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
3908
3909 if (slab_state < FULL)
3910 return ret;
3911
3912 if ((ret < 0) || !is_root_cache(cachep))
3913 return ret;
3914
Vladimir Davydov426589f2015-02-12 14:59:23 -08003915 lockdep_assert_held(&slab_mutex);
3916 for_each_memcg_cache(c, cachep) {
3917 /* return value determined by the root cache only */
3918 __do_tune_cpucache(c, limit, batchcount, shared, gfp);
Glauber Costa943a4512012-12-18 14:23:03 -08003919 }
3920
3921 return ret;
3922}
3923
Christoph Lameter18004c52012-07-06 15:25:12 -05003924/* Called with slab_mutex held always */
Pekka Enberg83b519e2009-06-10 19:40:04 +03003925static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926{
3927 int err;
Glauber Costa943a4512012-12-18 14:23:03 -08003928 int limit = 0;
3929 int shared = 0;
3930 int batchcount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931
Thomas Garnier7c00fce2016-07-26 15:21:56 -07003932 err = cache_random_seq_create(cachep, cachep->num, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07003933 if (err)
3934 goto end;
3935
Glauber Costa943a4512012-12-18 14:23:03 -08003936 if (!is_root_cache(cachep)) {
3937 struct kmem_cache *root = memcg_root_cache(cachep);
3938 limit = root->limit;
3939 shared = root->shared;
3940 batchcount = root->batchcount;
3941 }
3942
3943 if (limit && shared && batchcount)
3944 goto skip_setup;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003945 /*
3946 * The head array serves three purposes:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 * - create a LIFO ordering, i.e. return objects that are cache-warm
3948 * - reduce the number of spinlock operations.
Andrew Mortona737b3e2006-03-22 00:08:11 -08003949 * - reduce the number of linked list operations on the slab and
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 * bufctl chains: array operations are cheaper.
3951 * The numbers are guessed, we should auto-tune as described by
3952 * Bonwick.
3953 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003954 if (cachep->size > 131072)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003955 limit = 1;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003956 else if (cachep->size > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 limit = 8;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003958 else if (cachep->size > 1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 limit = 24;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003960 else if (cachep->size > 256)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961 limit = 54;
3962 else
3963 limit = 120;
3964
Andrew Mortona737b3e2006-03-22 00:08:11 -08003965 /*
3966 * CPU bound tasks (e.g. network routing) can exhibit cpu bound
Linus Torvalds1da177e2005-04-16 15:20:36 -07003967 * allocation behaviour: Most allocs on one cpu, most free operations
3968 * on another cpu. For these cases, an efficient object passing between
3969 * cpus is necessary. This is provided by a shared array. The array
3970 * replaces Bonwick's magazine layer.
3971 * On uniprocessor, it's functionally equivalent (but less efficient)
3972 * to a larger limit. Thus disabled by default.
3973 */
3974 shared = 0;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003975 if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003976 shared = 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003977
3978#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08003979 /*
3980 * With debugging enabled, large batchcount lead to excessively long
3981 * periods with disabled local interrupts. Limit the batchcount
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 */
3983 if (limit > 32)
3984 limit = 32;
3985#endif
Glauber Costa943a4512012-12-18 14:23:03 -08003986 batchcount = (limit + 1) / 2;
3987skip_setup:
3988 err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07003989end:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003990 if (err)
Joe Perches11705322016-03-17 14:19:50 -07003991 pr_err("enable_cpucache failed for %s, error %d\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003992 cachep->name, -err);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07003993 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994}
3995
Christoph Lameter1b552532006-03-22 00:09:07 -08003996/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003997 * Drain an array if it contains any elements taking the node lock only if
3998 * necessary. Note that the node listlock also protects the array_cache
Christoph Lameterb18e7e62006-03-22 00:09:07 -08003999 * if drain_array() is used on the shared array.
Christoph Lameter1b552532006-03-22 00:09:07 -08004000 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004001static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n,
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004002 struct array_cache *ac, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004003{
Joonsoo Kim97654df2014-08-06 16:04:25 -07004004 LIST_HEAD(list);
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004005
4006 /* ac from n->shared can be freed if we don't hold the slab_mutex. */
4007 check_mutex_acquired();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004008
Christoph Lameter1b552532006-03-22 00:09:07 -08004009 if (!ac || !ac->avail)
4010 return;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004011
4012 if (ac->touched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004013 ac->touched = 0;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004014 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015 }
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004016
4017 spin_lock_irq(&n->list_lock);
4018 drain_array_locked(cachep, ac, node, false, &list);
4019 spin_unlock_irq(&n->list_lock);
4020
4021 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004022}
4023
4024/**
4025 * cache_reap - Reclaim memory from caches.
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08004026 * @w: work descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07004027 *
4028 * Called from workqueue/eventd every few seconds.
4029 * Purpose:
4030 * - clear the per-cpu caches for this CPU.
4031 * - return freeable pages to the main free memory pool.
4032 *
Andrew Mortona737b3e2006-03-22 00:08:11 -08004033 * If we cannot acquire the cache chain mutex then just give up - we'll try
4034 * again on the next iteration.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004035 */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004036static void cache_reap(struct work_struct *w)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037{
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07004038 struct kmem_cache *searchp;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004039 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07004040 int node = numa_mem_id();
Jean Delvarebf6aede2009-04-02 16:56:54 -07004041 struct delayed_work *work = to_delayed_work(w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042
Christoph Lameter18004c52012-07-06 15:25:12 -05004043 if (!mutex_trylock(&slab_mutex))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044 /* Give up. Setup the next iteration. */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004045 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004046
Christoph Lameter18004c52012-07-06 15:25:12 -05004047 list_for_each_entry(searchp, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 check_irq_on();
4049
Christoph Lameter35386e32006-03-22 00:09:05 -08004050 /*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004051 * We only take the node lock if absolutely necessary and we
Christoph Lameter35386e32006-03-22 00:09:05 -08004052 * have established with reasonable certainty that
4053 * we can do some work if the lock was obtained.
4054 */
Christoph Lameter18bf8542014-08-06 16:04:11 -07004055 n = get_node(searchp, node);
Christoph Lameter35386e32006-03-22 00:09:05 -08004056
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004057 reap_alien(searchp, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004059 drain_array(searchp, n, cpu_cache_get(searchp), node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004060
Christoph Lameter35386e32006-03-22 00:09:05 -08004061 /*
4062 * These are racy checks but it does not matter
4063 * if we skip one check or scan twice.
4064 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004065 if (time_after(n->next_reap, jiffies))
Christoph Lameter35386e32006-03-22 00:09:05 -08004066 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08004068 n->next_reap = jiffies + REAPTIMEOUT_NODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004069
Joonsoo Kim18726ca2016-05-19 17:10:02 -07004070 drain_array(searchp, n, n->shared, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004071
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004072 if (n->free_touched)
4073 n->free_touched = 0;
Christoph Lametered11d9e2006-06-30 01:55:45 -07004074 else {
4075 int freed;
4076
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004077 freed = drain_freelist(searchp, n, (n->free_limit +
Christoph Lametered11d9e2006-06-30 01:55:45 -07004078 5 * searchp->num - 1) / (5 * searchp->num));
4079 STATS_ADD_REAPED(searchp, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004080 }
Christoph Lameter35386e32006-03-22 00:09:05 -08004081next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082 cond_resched();
4083 }
4084 check_irq_on();
Christoph Lameter18004c52012-07-06 15:25:12 -05004085 mutex_unlock(&slab_mutex);
Christoph Lameter8fce4d82006-03-09 17:33:54 -08004086 next_reap_node();
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004087out:
Andrew Mortona737b3e2006-03-22 00:08:11 -08004088 /* Set up the next iteration */
Vlastimil Babkaa9f2a842018-04-13 15:35:38 -07004089 schedule_delayed_work_on(smp_processor_id(), work,
4090 round_jiffies_relative(REAPTIMEOUT_AC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091}
4092
Glauber Costa0d7561c2012-10-19 18:20:27 +04004093void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094{
Greg Thelenf728b0a2016-12-12 16:41:41 -08004095 unsigned long active_objs, num_objs, active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004096 unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0;
4097 unsigned long free_slabs = 0;
Christoph Lametere498be72005-09-09 13:03:32 -07004098 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004099 struct kmem_cache_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004100
Christoph Lameter18bf8542014-08-06 16:04:11 -07004101 for_each_kmem_cache_node(cachep, node, n) {
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08004102 check_irq_on();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004103 spin_lock_irq(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07004104
David Rientjesbf00bd32016-12-12 16:41:44 -08004105 total_slabs += n->total_slabs;
4106 free_slabs += n->free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004107 free_objs += n->free_objects;
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07004108
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004109 if (n->shared)
4110 shared_avail += n->shared->avail;
Christoph Lametere498be72005-09-09 13:03:32 -07004111
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004112 spin_unlock_irq(&n->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004113 }
David Rientjesbf00bd32016-12-12 16:41:44 -08004114 num_objs = total_slabs * cachep->num;
4115 active_slabs = total_slabs - free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004116 active_objs = num_objs - free_objs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004117
Glauber Costa0d7561c2012-10-19 18:20:27 +04004118 sinfo->active_objs = active_objs;
4119 sinfo->num_objs = num_objs;
4120 sinfo->active_slabs = active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004121 sinfo->num_slabs = total_slabs;
Glauber Costa0d7561c2012-10-19 18:20:27 +04004122 sinfo->shared_avail = shared_avail;
4123 sinfo->limit = cachep->limit;
4124 sinfo->batchcount = cachep->batchcount;
4125 sinfo->shared = cachep->shared;
4126 sinfo->objects_per_slab = cachep->num;
4127 sinfo->cache_order = cachep->gfporder;
4128}
4129
4130void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep)
4131{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004132#if STATS
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004133 { /* node stats */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004134 unsigned long high = cachep->high_mark;
4135 unsigned long allocs = cachep->num_allocations;
4136 unsigned long grown = cachep->grown;
4137 unsigned long reaped = cachep->reaped;
4138 unsigned long errors = cachep->errors;
4139 unsigned long max_freeable = cachep->max_freeable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 unsigned long node_allocs = cachep->node_allocs;
Christoph Lametere498be72005-09-09 13:03:32 -07004141 unsigned long node_frees = cachep->node_frees;
Ravikiran G Thirumalaifb7faf32006-04-10 22:52:54 -07004142 unsigned long overflows = cachep->node_overflow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004143
Joe Perches756a0252016-03-17 14:19:47 -07004144 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu",
Joe Perchese92dd4f2010-03-26 19:27:58 -07004145 allocs, high, grown,
4146 reaped, errors, max_freeable, node_allocs,
4147 node_frees, overflows);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 }
4149 /* cpu stats */
4150 {
4151 unsigned long allochit = atomic_read(&cachep->allochit);
4152 unsigned long allocmiss = atomic_read(&cachep->allocmiss);
4153 unsigned long freehit = atomic_read(&cachep->freehit);
4154 unsigned long freemiss = atomic_read(&cachep->freemiss);
4155
4156 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004157 allochit, allocmiss, freehit, freemiss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004158 }
4159#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160}
4161
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162#define MAX_SLABINFO_WRITE 128
4163/**
4164 * slabinfo_write - Tuning for the slab allocator
4165 * @file: unused
4166 * @buffer: user buffer
4167 * @count: data length
4168 * @ppos: unused
4169 */
Glauber Costab7454ad2012-10-19 18:20:25 +04004170ssize_t slabinfo_write(struct file *file, const char __user *buffer,
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004171 size_t count, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004172{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004173 char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004174 int limit, batchcount, shared, res;
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07004175 struct kmem_cache *cachep;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004176
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177 if (count > MAX_SLABINFO_WRITE)
4178 return -EINVAL;
4179 if (copy_from_user(&kbuf, buffer, count))
4180 return -EFAULT;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004181 kbuf[MAX_SLABINFO_WRITE] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07004182
4183 tmp = strchr(kbuf, ' ');
4184 if (!tmp)
4185 return -EINVAL;
4186 *tmp = '\0';
4187 tmp++;
4188 if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
4189 return -EINVAL;
4190
4191 /* Find the cache in the chain of caches. */
Christoph Lameter18004c52012-07-06 15:25:12 -05004192 mutex_lock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 res = -EINVAL;
Christoph Lameter18004c52012-07-06 15:25:12 -05004194 list_for_each_entry(cachep, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 if (!strcmp(cachep->name, kbuf)) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08004196 if (limit < 1 || batchcount < 1 ||
4197 batchcount > limit || shared < 0) {
Christoph Lametere498be72005-09-09 13:03:32 -07004198 res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004199 } else {
Christoph Lametere498be72005-09-09 13:03:32 -07004200 res = do_tune_cpucache(cachep, limit,
Pekka Enberg83b519e2009-06-10 19:40:04 +03004201 batchcount, shared,
4202 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004203 }
4204 break;
4205 }
4206 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004207 mutex_unlock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004208 if (res >= 0)
4209 res = count;
4210 return res;
4211}
Al Viro871751e2006-03-25 03:06:39 -08004212
4213#ifdef CONFIG_DEBUG_SLAB_LEAK
4214
Al Viro871751e2006-03-25 03:06:39 -08004215static inline int add_caller(unsigned long *n, unsigned long v)
4216{
4217 unsigned long *p;
4218 int l;
4219 if (!v)
4220 return 1;
4221 l = n[1];
4222 p = n + 2;
4223 while (l) {
4224 int i = l/2;
4225 unsigned long *q = p + 2 * i;
4226 if (*q == v) {
4227 q[1]++;
4228 return 1;
4229 }
4230 if (*q > v) {
4231 l = i;
4232 } else {
4233 p = q + 2;
4234 l -= i + 1;
4235 }
4236 }
4237 if (++n[1] == n[0])
4238 return 0;
4239 memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n));
4240 p[0] = v;
4241 p[1] = 1;
4242 return 1;
4243}
4244
Joonsoo Kim8456a642013-10-24 10:07:49 +09004245static void handle_slab(unsigned long *n, struct kmem_cache *c,
4246 struct page *page)
Al Viro871751e2006-03-25 03:06:39 -08004247{
4248 void *p;
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004249 int i, j;
4250 unsigned long v;
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09004251
Al Viro871751e2006-03-25 03:06:39 -08004252 if (n[0] == n[1])
4253 return;
Joonsoo Kim8456a642013-10-24 10:07:49 +09004254 for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) {
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004255 bool active = true;
4256
4257 for (j = page->active; j < c->num; j++) {
4258 if (get_free_obj(page, j) == i) {
4259 active = false;
4260 break;
4261 }
4262 }
4263
4264 if (!active)
Al Viro871751e2006-03-25 03:06:39 -08004265 continue;
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09004266
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004267 /*
4268 * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table
4269 * mapping is established when actual object allocation and
4270 * we could mistakenly access the unmapped object in the cpu
4271 * cache.
4272 */
4273 if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v)))
4274 continue;
4275
4276 if (!add_caller(n, v))
Al Viro871751e2006-03-25 03:06:39 -08004277 return;
4278 }
4279}
4280
4281static void show_symbol(struct seq_file *m, unsigned long address)
4282{
4283#ifdef CONFIG_KALLSYMS
Al Viro871751e2006-03-25 03:06:39 -08004284 unsigned long offset, size;
Tejun Heo9281ace2007-07-17 04:03:51 -07004285 char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN];
Al Viro871751e2006-03-25 03:06:39 -08004286
Alexey Dobriyana5c43da2007-05-08 00:28:47 -07004287 if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) {
Al Viro871751e2006-03-25 03:06:39 -08004288 seq_printf(m, "%s+%#lx/%#lx", name, offset, size);
Alexey Dobriyana5c43da2007-05-08 00:28:47 -07004289 if (modname[0])
Al Viro871751e2006-03-25 03:06:39 -08004290 seq_printf(m, " [%s]", modname);
4291 return;
4292 }
4293#endif
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08004294 seq_printf(m, "%px", (void *)address);
Al Viro871751e2006-03-25 03:06:39 -08004295}
4296
4297static int leaks_show(struct seq_file *m, void *p)
4298{
Thierry Reding0672aa72012-06-22 19:42:49 +02004299 struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list);
Joonsoo Kim8456a642013-10-24 10:07:49 +09004300 struct page *page;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004301 struct kmem_cache_node *n;
Al Viro871751e2006-03-25 03:06:39 -08004302 const char *name;
Christoph Lameterdb845062013-02-05 18:45:23 +00004303 unsigned long *x = m->private;
Al Viro871751e2006-03-25 03:06:39 -08004304 int node;
4305 int i;
4306
4307 if (!(cachep->flags & SLAB_STORE_USER))
4308 return 0;
4309 if (!(cachep->flags & SLAB_RED_ZONE))
4310 return 0;
4311
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004312 /*
4313 * Set store_user_clean and start to grab stored user information
4314 * for all objects on this cache. If some alloc/free requests comes
4315 * during the processing, information would be wrong so restart
4316 * whole processing.
4317 */
4318 do {
4319 set_store_user_clean(cachep);
4320 drain_cpu_caches(cachep);
Al Viro871751e2006-03-25 03:06:39 -08004321
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004322 x[1] = 0;
Al Viro871751e2006-03-25 03:06:39 -08004323
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004324 for_each_kmem_cache_node(cachep, node, n) {
Al Viro871751e2006-03-25 03:06:39 -08004325
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004326 check_irq_on();
4327 spin_lock_irq(&n->list_lock);
Al Viro871751e2006-03-25 03:06:39 -08004328
Joonsoo Kimd31676d2016-03-15 14:54:24 -07004329 list_for_each_entry(page, &n->slabs_full, lru)
4330 handle_slab(x, cachep, page);
4331 list_for_each_entry(page, &n->slabs_partial, lru)
4332 handle_slab(x, cachep, page);
4333 spin_unlock_irq(&n->list_lock);
4334 }
4335 } while (!is_store_user_clean(cachep));
4336
Al Viro871751e2006-03-25 03:06:39 -08004337 name = cachep->name;
Christoph Lameterdb845062013-02-05 18:45:23 +00004338 if (x[0] == x[1]) {
Al Viro871751e2006-03-25 03:06:39 -08004339 /* Increase the buffer size */
Christoph Lameter18004c52012-07-06 15:25:12 -05004340 mutex_unlock(&slab_mutex);
Christoph Lameterdb845062013-02-05 18:45:23 +00004341 m->private = kzalloc(x[0] * 4 * sizeof(unsigned long), GFP_KERNEL);
Al Viro871751e2006-03-25 03:06:39 -08004342 if (!m->private) {
4343 /* Too bad, we are really out */
Christoph Lameterdb845062013-02-05 18:45:23 +00004344 m->private = x;
Christoph Lameter18004c52012-07-06 15:25:12 -05004345 mutex_lock(&slab_mutex);
Al Viro871751e2006-03-25 03:06:39 -08004346 return -ENOMEM;
4347 }
Christoph Lameterdb845062013-02-05 18:45:23 +00004348 *(unsigned long *)m->private = x[0] * 2;
4349 kfree(x);
Christoph Lameter18004c52012-07-06 15:25:12 -05004350 mutex_lock(&slab_mutex);
Al Viro871751e2006-03-25 03:06:39 -08004351 /* Now make sure this entry will be retried */
4352 m->count = m->size;
4353 return 0;
4354 }
Christoph Lameterdb845062013-02-05 18:45:23 +00004355 for (i = 0; i < x[1]; i++) {
4356 seq_printf(m, "%s: %lu ", name, x[2*i+3]);
4357 show_symbol(m, x[2*i+2]);
Al Viro871751e2006-03-25 03:06:39 -08004358 seq_putc(m, '\n');
4359 }
Siddha, Suresh Bd2e7b7d2006-09-25 23:31:47 -07004360
Al Viro871751e2006-03-25 03:06:39 -08004361 return 0;
4362}
4363
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004364static const struct seq_operations slabstats_op = {
Vladimir Davydov1df3b262014-12-10 15:42:16 -08004365 .start = slab_start,
Wanpeng Li276a2432013-07-08 08:08:28 +08004366 .next = slab_next,
4367 .stop = slab_stop,
Al Viro871751e2006-03-25 03:06:39 -08004368 .show = leaks_show,
4369};
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004370
4371static int slabstats_open(struct inode *inode, struct file *file)
4372{
Rob Jonesb208ce32014-10-09 15:28:03 -07004373 unsigned long *n;
4374
4375 n = __seq_open_private(file, &slabstats_op, PAGE_SIZE);
4376 if (!n)
4377 return -ENOMEM;
4378
4379 *n = PAGE_SIZE / (2 * sizeof(unsigned long));
4380
4381 return 0;
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004382}
4383
4384static const struct file_operations proc_slabstats_operations = {
4385 .open = slabstats_open,
4386 .read = seq_read,
4387 .llseek = seq_lseek,
4388 .release = seq_release_private,
4389};
Al Viro871751e2006-03-25 03:06:39 -08004390#endif
Alexey Dobriyana0ec95a2008-10-06 00:59:10 +04004391
4392static int __init slab_proc_init(void)
4393{
4394#ifdef CONFIG_DEBUG_SLAB_LEAK
4395 proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations);
4396#endif
4397 return 0;
4398}
4399module_init(slab_proc_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004400
Kees Cook04385fc2016-06-23 15:20:59 -07004401#ifdef CONFIG_HARDENED_USERCOPY
4402/*
Kees Cookafcc90f82018-01-10 15:17:01 -08004403 * Rejects incorrectly sized objects and objects that are to be copied
4404 * to/from userspace but do not fall entirely within the containing slab
4405 * cache's usercopy region.
Kees Cook04385fc2016-06-23 15:20:59 -07004406 *
4407 * Returns NULL if check passes, otherwise const char * to name of cache
4408 * to indicate an error.
4409 */
Kees Cookf4e6e282018-01-10 14:48:22 -08004410void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
4411 bool to_user)
Kees Cook04385fc2016-06-23 15:20:59 -07004412{
4413 struct kmem_cache *cachep;
4414 unsigned int objnr;
4415 unsigned long offset;
4416
4417 /* Find and validate object. */
4418 cachep = page->slab_cache;
4419 objnr = obj_to_index(cachep, page, (void *)ptr);
4420 BUG_ON(objnr >= cachep->num);
4421
4422 /* Find offset within object. */
4423 offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep);
4424
Kees Cookafcc90f82018-01-10 15:17:01 -08004425 /* Allow address range falling entirely within usercopy region. */
4426 if (offset >= cachep->useroffset &&
4427 offset - cachep->useroffset <= cachep->usersize &&
4428 n <= cachep->useroffset - offset + cachep->usersize)
Kees Cookf4e6e282018-01-10 14:48:22 -08004429 return;
Kees Cook04385fc2016-06-23 15:20:59 -07004430
Kees Cookafcc90f82018-01-10 15:17:01 -08004431 /*
4432 * If the copy is still within the allocated object, produce
4433 * a warning instead of rejecting the copy. This is intended
4434 * to be a temporary method to find any missing usercopy
4435 * whitelists.
4436 */
Kees Cook2d891fb2017-11-30 13:04:32 -08004437 if (usercopy_fallback &&
4438 offset <= cachep->object_size &&
Kees Cookafcc90f82018-01-10 15:17:01 -08004439 n <= cachep->object_size - offset) {
4440 usercopy_warn("SLAB object", cachep->name, to_user, offset, n);
4441 return;
4442 }
4443
Kees Cookf4e6e282018-01-10 14:48:22 -08004444 usercopy_abort("SLAB object", cachep->name, to_user, offset, n);
Kees Cook04385fc2016-06-23 15:20:59 -07004445}
4446#endif /* CONFIG_HARDENED_USERCOPY */
4447
Manfred Spraul00e145b2005-09-03 15:55:07 -07004448/**
4449 * ksize - get the actual amount of memory allocated for a given object
4450 * @objp: Pointer to the object
4451 *
4452 * kmalloc may internally round up allocations and return more memory
4453 * than requested. ksize() can be used to determine the actual amount of
4454 * memory allocated. The caller may use this additional memory, even though
4455 * a smaller amount of memory was initially specified with the kmalloc call.
4456 * The caller must guarantee that objp points to a valid object previously
4457 * allocated with either kmalloc() or kmem_cache_alloc(). The object
4458 * must not be freed during the duration of the call.
4459 */
Pekka Enbergfd76bab2007-05-06 14:48:40 -07004460size_t ksize(const void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004461{
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004462 size_t size;
4463
Christoph Lameteref8b4522007-10-16 01:24:46 -07004464 BUG_ON(!objp);
4465 if (unlikely(objp == ZERO_SIZE_PTR))
Manfred Spraul00e145b2005-09-03 15:55:07 -07004466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004467
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004468 size = virt_to_cache(objp)->object_size;
4469 /* We assume that ksize callers could use the whole allocated area,
4470 * so we need to unpoison this area.
4471 */
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07004472 kasan_unpoison_shadow(objp, size);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004473
4474 return size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004475}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02004476EXPORT_SYMBOL(ksize);