blob: b1113561b98b17e76a509c667a16351c467e69bd [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
365/*
David Rientjes3df1ccc2011-10-18 22:09:28 -0700366 * Do not go above this order unless 0 objects fit into the slab or
367 * overridden on the command line.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 */
David Rientjes543585c2011-10-18 22:09:24 -0700369#define SLAB_MAX_ORDER_HI 1
370#define SLAB_MAX_ORDER_LO 0
371static int slab_max_order = SLAB_MAX_ORDER_LO;
David Rientjes3df1ccc2011-10-18 22:09:28 -0700372static bool slab_max_order_set __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Joonsoo Kim8456a642013-10-24 10:07:49 +0900374static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800375 unsigned int idx)
376{
Joonsoo Kim8456a642013-10-24 10:07:49 +0900377 return page->s_mem + cache->size * idx;
Pekka Enberg8fea4e92006-03-22 00:08:10 -0800378}
379
Joonsoo Kim6fb92432016-03-15 14:54:09 -0700380#define BOOT_CPUCACHE_ENTRIES 1
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381/* internal cache of cache description objs */
Christoph Lameter9b030cb2012-09-05 00:20:33 +0000382static struct kmem_cache kmem_cache_boot = {
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800383 .batchcount = 1,
384 .limit = BOOT_CPUCACHE_ENTRIES,
385 .shared = 1,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500386 .size = sizeof(struct kmem_cache),
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800387 .name = "kmem_cache",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388};
389
Tejun Heo1871e522009-10-29 22:34:13 +0900390static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Pekka Enberg343e0d72006-02-01 03:05:50 -0800392static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700394 return this_cpu_ptr(cachep->cpu_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Andrew Mortona737b3e2006-03-22 00:08:11 -0800397/*
398 * Calculate the number of objects and left-over bytes for a given buffer size.
399 */
Joonsoo Kim70f75062016-03-15 14:54:53 -0700400static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -0800401 slab_flags_t flags, size_t *left_over)
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800402{
Joonsoo Kim70f75062016-03-15 14:54:53 -0700403 unsigned int num;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800404 size_t slab_size = PAGE_SIZE << gfporder;
405
406 /*
407 * The slab management structure can be either off the slab or
408 * on it. For the latter case, the memory allocated for a
409 * slab is used for:
410 *
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800411 * - @buffer_size bytes for each object
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700412 * - One freelist_idx_t for each object
413 *
414 * We don't need to consider alignment of freelist because
415 * freelist will be at the end of slab page. The objects will be
416 * at the correct alignment.
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800417 *
418 * If the slab management structure is off the slab, then the
419 * alignment will already be calculated into the size. Because
420 * the slabs are all pages aligned, the objects will be at the
421 * correct alignment when allocated.
422 */
Joonsoo Kimb03a017b2016-03-15 14:54:50 -0700423 if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700424 num = slab_size / buffer_size;
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700425 *left_over = slab_size % buffer_size;
Steven Rostedtfbaccac2006-02-01 03:05:45 -0800426 } else {
Joonsoo Kim70f75062016-03-15 14:54:53 -0700427 num = slab_size / (buffer_size + sizeof(freelist_idx_t));
Joonsoo Kim2e6b3602016-03-15 14:54:30 -0700428 *left_over = slab_size %
429 (buffer_size + sizeof(freelist_idx_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
Joonsoo Kim70f75062016-03-15 14:54:53 -0700431
432 return num;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Christoph Lameterf28510d2012-09-11 19:49:38 +0000435#if DEBUG
Harvey Harrisond40cee22008-04-30 00:55:07 -0700436#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
Andrew Mortona737b3e2006-03-22 00:08:11 -0800438static void __slab_error(const char *function, struct kmem_cache *cachep,
439 char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
Joe Perches11705322016-03-17 14:19:50 -0700441 pr_err("slab error in %s(): cache `%s': %s\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800442 function, cachep->name, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 dump_stack();
Rusty Russell373d4d02013-01-21 17:17:39 +1030444 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445}
Christoph Lameterf28510d2012-09-11 19:49:38 +0000446#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Paul Menage3395ee02006-12-06 20:32:16 -0800448/*
449 * By default on NUMA we use alien caches to stage the freeing of
450 * objects allocated from other nodes. This causes massive memory
451 * inefficiencies when using fake NUMA setup to split memory into a
452 * large number of small nodes, so it can be disabled on the command
453 * line
454 */
455
456static int use_alien_caches __read_mostly = 1;
457static int __init noaliencache_setup(char *s)
458{
459 use_alien_caches = 0;
460 return 1;
461}
462__setup("noaliencache", noaliencache_setup);
463
David Rientjes3df1ccc2011-10-18 22:09:28 -0700464static int __init slab_max_order_setup(char *str)
465{
466 get_option(&str, &slab_max_order);
467 slab_max_order = slab_max_order < 0 ? 0 :
468 min(slab_max_order, MAX_ORDER - 1);
469 slab_max_order_set = true;
470
471 return 1;
472}
473__setup("slab_max_order=", slab_max_order_setup);
474
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800475#ifdef CONFIG_NUMA
476/*
477 * Special reaping functions for NUMA systems called from cache_reap().
478 * These take care of doing round robin flushing of alien caches (containing
479 * objects freed on different nodes from which they were allocated) and the
480 * flushing of remote pcps by calling drain_node_pages.
481 */
Tejun Heo1871e522009-10-29 22:34:13 +0900482static DEFINE_PER_CPU(unsigned long, slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800483
484static void init_reap_node(int cpu)
485{
Andrew Morton0edaf862016-05-19 17:10:58 -0700486 per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
487 node_online_map);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800488}
489
490static void next_reap_node(void)
491{
Christoph Lameter909ea962010-12-08 16:22:55 +0100492 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800493
Andrew Morton0edaf862016-05-19 17:10:58 -0700494 node = next_node_in(node, node_online_map);
Christoph Lameter909ea962010-12-08 16:22:55 +0100495 __this_cpu_write(slab_reap_node, node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800496}
497
498#else
499#define init_reap_node(cpu) do { } while (0)
500#define next_reap_node(void) do { } while (0)
501#endif
502
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503/*
504 * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz
505 * via the workqueue/eventd.
506 * Add the CPU number into the expiration time to minimize the possibility of
507 * the CPUs getting into lockstep and contending for the global cache chain
508 * lock.
509 */
Paul Gortmaker0db06282013-06-19 14:53:51 -0400510static void start_cpu_timer(int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
Tejun Heo1871e522009-10-29 22:34:13 +0900512 struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Tejun Heoeac03372016-09-16 15:49:34 -0400514 if (reap_work->work.func == NULL) {
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800515 init_reap_node(cpu);
Tejun Heo203b42f2012-08-21 13:18:23 -0700516 INIT_DEFERRABLE_WORK(reap_work, cache_reap);
Arjan van de Ven2b284212006-12-10 02:21:28 -0800517 schedule_delayed_work_on(cpu, reap_work,
518 __round_jiffies_relative(HZ, cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 }
520}
521
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700522static void init_arraycache(struct array_cache *ac, int limit, int batch)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700524 if (ac) {
525 ac->avail = 0;
526 ac->limit = limit;
527 ac->batchcount = batch;
528 ac->touched = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 }
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700530}
531
532static struct array_cache *alloc_arraycache(int node, int entries,
533 int batchcount, gfp_t gfp)
534{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700535 size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700536 struct array_cache *ac = NULL;
537
538 ac = kmalloc_node(memsize, gfp, node);
Qian Cai92d1d072019-03-05 15:42:03 -0800539 /*
540 * The array_cache structures contain pointers to free object.
541 * However, when such objects are allocated or transferred to another
542 * cache the pointers are not cleared and they could be counted as
543 * valid references during a kmemleak scan. Therefore, kmemleak must
544 * not scan such objects.
545 */
546 kmemleak_no_scan(ac);
Joonsoo Kim1fe00d52014-08-06 16:04:27 -0700547 init_arraycache(ac, entries, batchcount);
548 return ac;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700551static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
552 struct page *page, void *objp)
Mel Gorman072bb0a2012-07-31 16:43:58 -0700553{
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700554 struct kmem_cache_node *n;
555 int page_node;
556 LIST_HEAD(list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700557
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700558 page_node = page_to_nid(page);
559 n = get_node(cachep, page_node);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700560
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700561 spin_lock(&n->list_lock);
562 free_block(cachep, &objp, 1, page_node, &list);
563 spin_unlock(&n->list_lock);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700564
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -0700565 slabs_destroy(cachep, &list);
Mel Gorman072bb0a2012-07-31 16:43:58 -0700566}
567
Christoph Lameter3ded1752006-03-25 03:06:44 -0800568/*
569 * Transfer objects in one arraycache to another.
570 * Locking must be handled by the caller.
571 *
572 * Return the number of entries transferred.
573 */
574static int transfer_objects(struct array_cache *to,
575 struct array_cache *from, unsigned int max)
576{
577 /* Figure out how many entries to transfer */
Hagen Paul Pfeifer732eacc2010-10-26 14:22:23 -0700578 int nr = min3(from->avail, max, to->limit - to->avail);
Christoph Lameter3ded1752006-03-25 03:06:44 -0800579
580 if (!nr)
581 return 0;
582
583 memcpy(to->entry + to->avail, from->entry + from->avail -nr,
584 sizeof(void *) *nr);
585
586 from->avail -= nr;
587 to->avail += nr;
Christoph Lameter3ded1752006-03-25 03:06:44 -0800588 return nr;
589}
590
Kees Cookdabc3e22020-08-06 23:18:24 -0700591/* &alien->lock must be held by alien callers. */
592static __always_inline void __free_one(struct array_cache *ac, void *objp)
593{
594 /* Avoid trivial double-free. */
595 if (IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
596 WARN_ON_ONCE(ac->avail > 0 && ac->entry[ac->avail - 1] == objp))
597 return;
598 ac->entry[ac->avail++] = objp;
599}
600
Christoph Lameter765c4502006-09-27 01:50:08 -0700601#ifndef CONFIG_NUMA
602
603#define drain_alien_cache(cachep, alien) do { } while (0)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000604#define reap_alien(cachep, n) do { } while (0)
Christoph Lameter765c4502006-09-27 01:50:08 -0700605
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700606static inline struct alien_cache **alloc_alien_cache(int node,
607 int limit, gfp_t gfp)
Christoph Lameter765c4502006-09-27 01:50:08 -0700608{
Joonsoo Kim88881772016-05-19 17:10:05 -0700609 return NULL;
Christoph Lameter765c4502006-09-27 01:50:08 -0700610}
611
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700612static inline void free_alien_cache(struct alien_cache **ac_ptr)
Christoph Lameter765c4502006-09-27 01:50:08 -0700613{
614}
615
616static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
617{
618 return 0;
619}
620
621static inline void *alternate_node_alloc(struct kmem_cache *cachep,
622 gfp_t flags)
623{
624 return NULL;
625}
626
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800627static inline void *____cache_alloc_node(struct kmem_cache *cachep,
Christoph Lameter765c4502006-09-27 01:50:08 -0700628 gfp_t flags, int nodeid)
629{
630 return NULL;
631}
632
David Rientjes4167e9b2015-04-14 15:46:55 -0700633static inline gfp_t gfp_exact_node(gfp_t flags)
634{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700635 return flags & ~__GFP_NOFAIL;
David Rientjes4167e9b2015-04-14 15:46:55 -0700636}
637
Christoph Lameter765c4502006-09-27 01:50:08 -0700638#else /* CONFIG_NUMA */
639
Christoph Hellwig8b98c162006-12-06 20:32:30 -0800640static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
Paul Jacksonc61afb12006-03-24 03:16:08 -0800641static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
Christoph Lameterdc85da12006-01-18 17:42:36 -0800642
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700643static struct alien_cache *__alloc_alien_cache(int node, int entries,
644 int batch, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -0700645{
Joonsoo Kim5e804782014-08-06 16:04:40 -0700646 size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700647 struct alien_cache *alc = NULL;
648
649 alc = kmalloc_node(memsize, gfp, node);
Christoph Lameter09c2e762019-01-08 15:23:00 -0800650 if (alc) {
Qian Cai92d1d072019-03-05 15:42:03 -0800651 kmemleak_no_scan(alc);
Christoph Lameter09c2e762019-01-08 15:23:00 -0800652 init_arraycache(&alc->ac, entries, batch);
653 spin_lock_init(&alc->lock);
654 }
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700655 return alc;
656}
657
658static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
659{
660 struct alien_cache **alc_ptr;
Christoph Lametere498be72005-09-09 13:03:32 -0700661 int i;
662
663 if (limit > 1)
664 limit = 12;
Alexey Dobriyanb9726c22019-03-05 15:48:26 -0800665 alc_ptr = kcalloc_node(nr_node_ids, sizeof(void *), gfp, node);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700666 if (!alc_ptr)
667 return NULL;
668
669 for_each_node(i) {
670 if (i == node || !node_online(i))
671 continue;
672 alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
673 if (!alc_ptr[i]) {
674 for (i--; i >= 0; i--)
675 kfree(alc_ptr[i]);
676 kfree(alc_ptr);
677 return NULL;
Christoph Lametere498be72005-09-09 13:03:32 -0700678 }
679 }
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700680 return alc_ptr;
Christoph Lametere498be72005-09-09 13:03:32 -0700681}
682
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700683static void free_alien_cache(struct alien_cache **alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700684{
685 int i;
686
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700687 if (!alc_ptr)
Christoph Lametere498be72005-09-09 13:03:32 -0700688 return;
Christoph Lametere498be72005-09-09 13:03:32 -0700689 for_each_node(i)
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700690 kfree(alc_ptr[i]);
691 kfree(alc_ptr);
Christoph Lametere498be72005-09-09 13:03:32 -0700692}
693
Pekka Enberg343e0d72006-02-01 03:05:50 -0800694static void __drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kim833b7062014-08-06 16:04:33 -0700695 struct array_cache *ac, int node,
696 struct list_head *list)
Christoph Lametere498be72005-09-09 13:03:32 -0700697{
Christoph Lameter18bf8542014-08-06 16:04:11 -0700698 struct kmem_cache_node *n = get_node(cachep, node);
Christoph Lametere498be72005-09-09 13:03:32 -0700699
700 if (ac->avail) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000701 spin_lock(&n->list_lock);
Christoph Lametere00946f2006-03-25 03:06:45 -0800702 /*
703 * Stuff objects into the remote nodes shared array first.
704 * That way we could avoid the overhead of putting the objects
705 * into the free lists and getting them back later.
706 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000707 if (n->shared)
708 transfer_objects(n->shared, ac, ac->limit);
Christoph Lametere00946f2006-03-25 03:06:45 -0800709
Joonsoo Kim833b7062014-08-06 16:04:33 -0700710 free_block(cachep, ac->entry, ac->avail, node, list);
Christoph Lametere498be72005-09-09 13:03:32 -0700711 ac->avail = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000712 spin_unlock(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -0700713 }
714}
715
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800716/*
717 * Called from cache_reap() to regularly drain alien caches round robin.
718 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000719static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800720{
Christoph Lameter909ea962010-12-08 16:22:55 +0100721 int node = __this_cpu_read(slab_reap_node);
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800722
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000723 if (n->alien) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700724 struct alien_cache *alc = n->alien[node];
725 struct array_cache *ac;
Christoph Lametere00946f2006-03-25 03:06:45 -0800726
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700727 if (alc) {
728 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700729 if (ac->avail && spin_trylock_irq(&alc->lock)) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700730 LIST_HEAD(list);
731
732 __drain_alien_cache(cachep, ac, node, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700733 spin_unlock_irq(&alc->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700734 slabs_destroy(cachep, &list);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700735 }
Christoph Lameter8fce4d82006-03-09 17:33:54 -0800736 }
737 }
738}
739
Andrew Mortona737b3e2006-03-22 00:08:11 -0800740static void drain_alien_cache(struct kmem_cache *cachep,
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700741 struct alien_cache **alien)
Christoph Lametere498be72005-09-09 13:03:32 -0700742{
Pekka Enbergb28a02d2006-01-08 01:00:37 -0800743 int i = 0;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700744 struct alien_cache *alc;
Christoph Lametere498be72005-09-09 13:03:32 -0700745 struct array_cache *ac;
746 unsigned long flags;
747
748 for_each_online_node(i) {
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700749 alc = alien[i];
750 if (alc) {
Joonsoo Kim833b7062014-08-06 16:04:33 -0700751 LIST_HEAD(list);
752
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700753 ac = &alc->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700754 spin_lock_irqsave(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700755 __drain_alien_cache(cachep, ac, i, &list);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700756 spin_unlock_irqrestore(&alc->lock, flags);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700757 slabs_destroy(cachep, &list);
Christoph Lametere498be72005-09-09 13:03:32 -0700758 }
759 }
760}
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700761
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700762static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
763 int node, int page_node)
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700764{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000765 struct kmem_cache_node *n;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700766 struct alien_cache *alien = NULL;
767 struct array_cache *ac;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700768 LIST_HEAD(list);
Pekka Enberg1ca4cb22006-10-06 00:43:52 -0700769
Christoph Lameter18bf8542014-08-06 16:04:11 -0700770 n = get_node(cachep, node);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700771 STATS_INC_NODEFREES(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700772 if (n->alien && n->alien[page_node]) {
773 alien = n->alien[page_node];
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700774 ac = &alien->ac;
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700775 spin_lock(&alien->lock);
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700776 if (unlikely(ac->avail == ac->limit)) {
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700777 STATS_INC_ACOVERFLOW(cachep);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700778 __drain_alien_cache(cachep, ac, page_node, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700779 }
Kees Cookdabc3e22020-08-06 23:18:24 -0700780 __free_one(ac, objp);
Joonsoo Kim49dfc302014-08-06 16:04:31 -0700781 spin_unlock(&alien->lock);
Joonsoo Kim833b7062014-08-06 16:04:33 -0700782 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700783 } else {
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700784 n = get_node(cachep, page_node);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700785 spin_lock(&n->list_lock);
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700786 free_block(cachep, &objp, 1, page_node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -0700787 spin_unlock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -0700788 slabs_destroy(cachep, &list);
Pekka Enberg729bd0b2006-06-23 02:03:05 -0700789 }
790 return 1;
791}
Joonsoo Kim25c4f302014-10-09 15:26:09 -0700792
793static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
794{
795 int page_node = page_to_nid(virt_to_page(objp));
796 int node = numa_mem_id();
797 /*
798 * Make sure we are not freeing a object from another node to the array
799 * cache on this cpu.
800 */
801 if (likely(node == page_node))
802 return 0;
803
804 return __cache_free_alien(cachep, objp, node, page_node);
805}
David Rientjes4167e9b2015-04-14 15:46:55 -0700806
807/*
Mel Gorman444eb2a42016-03-17 14:19:23 -0700808 * Construct gfp mask to allocate from a specific node but do not reclaim or
809 * warn about failures.
David Rientjes4167e9b2015-04-14 15:46:55 -0700810 */
811static inline gfp_t gfp_exact_node(gfp_t flags)
812{
Mel Gorman444eb2a42016-03-17 14:19:23 -0700813 return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
David Rientjes4167e9b2015-04-14 15:46:55 -0700814}
Christoph Lametere498be72005-09-09 13:03:32 -0700815#endif
816
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700817static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
818{
819 struct kmem_cache_node *n;
820
821 /*
822 * Set up the kmem_cache_node for cpu before we can
823 * begin anything. Make sure some other cpu on this
824 * node has not already allocated this
825 */
826 n = get_node(cachep, node);
827 if (n) {
828 spin_lock_irq(&n->list_lock);
829 n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
830 cachep->num;
831 spin_unlock_irq(&n->list_lock);
832
833 return 0;
834 }
835
836 n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
837 if (!n)
838 return -ENOMEM;
839
840 kmem_cache_node_init(n);
841 n->next_reap = jiffies + REAPTIMEOUT_NODE +
842 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
843
844 n->free_limit =
845 (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;
846
847 /*
848 * The kmem_cache_nodes don't come and go as CPUs
849 * come and go. slab_mutex is sufficient
850 * protection here.
851 */
852 cachep->node[node] = n;
853
854 return 0;
855}
856
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200857#if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700858/*
Christoph Lameter6a673682013-01-10 19:14:19 +0000859 * Allocates and initializes node for a node on each slab cache, used for
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000860 * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node
David Rientjes8f9f8d92010-03-27 19:40:47 -0700861 * will be allocated off-node since memory is not yet online for the new node.
Christoph Lameter6a673682013-01-10 19:14:19 +0000862 * When hotplugging memory or a cpu, existing node are not replaced if
David Rientjes8f9f8d92010-03-27 19:40:47 -0700863 * already in use.
864 *
Christoph Lameter18004c52012-07-06 15:25:12 -0500865 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -0700866 */
Christoph Lameter6a673682013-01-10 19:14:19 +0000867static int init_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -0700868{
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700869 int ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700870 struct kmem_cache *cachep;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700871
Christoph Lameter18004c52012-07-06 15:25:12 -0500872 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700873 ret = init_cache_node(cachep, node, GFP_KERNEL);
874 if (ret)
875 return ret;
David Rientjes8f9f8d92010-03-27 19:40:47 -0700876 }
Joonsoo Kimded0ecf2016-05-19 17:10:11 -0700877
David Rientjes8f9f8d92010-03-27 19:40:47 -0700878 return 0;
879}
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200880#endif
David Rientjes8f9f8d92010-03-27 19:40:47 -0700881
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700882static int setup_kmem_cache_node(struct kmem_cache *cachep,
883 int node, gfp_t gfp, bool force_change)
884{
885 int ret = -ENOMEM;
886 struct kmem_cache_node *n;
887 struct array_cache *old_shared = NULL;
888 struct array_cache *new_shared = NULL;
889 struct alien_cache **new_alien = NULL;
890 LIST_HEAD(list);
891
892 if (use_alien_caches) {
893 new_alien = alloc_alien_cache(node, cachep->limit, gfp);
894 if (!new_alien)
895 goto fail;
896 }
897
898 if (cachep->shared) {
899 new_shared = alloc_arraycache(node,
900 cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
901 if (!new_shared)
902 goto fail;
903 }
904
905 ret = init_cache_node(cachep, node, gfp);
906 if (ret)
907 goto fail;
908
909 n = get_node(cachep, node);
910 spin_lock_irq(&n->list_lock);
911 if (n->shared && force_change) {
912 free_block(cachep, n->shared->entry,
913 n->shared->avail, node, &list);
914 n->shared->avail = 0;
915 }
916
917 if (!n->shared || force_change) {
918 old_shared = n->shared;
919 n->shared = new_shared;
920 new_shared = NULL;
921 }
922
923 if (!n->alien) {
924 n->alien = new_alien;
925 new_alien = NULL;
926 }
927
928 spin_unlock_irq(&n->list_lock);
929 slabs_destroy(cachep, &list);
930
Joonsoo Kim801faf02016-05-19 17:10:31 -0700931 /*
932 * To protect lockless access to n->shared during irq disabled context.
933 * If n->shared isn't NULL in irq disabled context, accessing to it is
934 * guaranteed to be valid until irq is re-enabled, because it will be
Paul E. McKenney6564a252018-11-06 19:24:33 -0800935 * freed after synchronize_rcu().
Joonsoo Kim801faf02016-05-19 17:10:31 -0700936 */
Joonsoo Kim86d9f482016-10-27 17:46:18 -0700937 if (old_shared && force_change)
Paul E. McKenney6564a252018-11-06 19:24:33 -0800938 synchronize_rcu();
Joonsoo Kim801faf02016-05-19 17:10:31 -0700939
Joonsoo Kimc3d332b2016-05-19 17:10:14 -0700940fail:
941 kfree(old_shared);
942 kfree(new_shared);
943 free_alien_cache(new_alien);
944
945 return ret;
946}
947
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +0200948#ifdef CONFIG_SMP
949
Paul Gortmaker0db06282013-06-19 14:53:51 -0400950static void cpuup_canceled(long cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951{
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700952 struct kmem_cache *cachep;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000953 struct kmem_cache_node *n = NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -0700954 int node = cpu_to_mem(cpu);
Rusty Russella70f7302009-03-13 14:49:46 +1030955 const struct cpumask *mask = cpumask_of_node(node);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700956
Christoph Lameter18004c52012-07-06 15:25:12 -0500957 list_for_each_entry(cachep, &slab_caches, list) {
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700958 struct array_cache *nc;
959 struct array_cache *shared;
Joonsoo Kimc8522a32014-08-06 16:04:29 -0700960 struct alien_cache **alien;
Joonsoo Kim97654df2014-08-06 16:04:25 -0700961 LIST_HEAD(list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700962
Christoph Lameter18bf8542014-08-06 16:04:11 -0700963 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000964 if (!n)
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700965 continue;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700966
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000967 spin_lock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700968
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000969 /* Free limit for this kmem_cache_node */
970 n->free_limit -= cachep->batchcount;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700971
972 /* cpu is dead; no one can alloc from it. */
973 nc = per_cpu_ptr(cachep->cpu_cache, cpu);
Li RongQing517f9f12019-05-13 17:16:25 -0700974 free_block(cachep, nc->entry, nc->avail, node, &list);
975 nc->avail = 0;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700976
Rusty Russell58463c12009-12-17 11:43:12 -0600977 if (!cpumask_empty(mask)) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000978 spin_unlock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700979 goto free_slab;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700980 }
981
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000982 shared = n->shared;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700983 if (shared) {
984 free_block(cachep, shared->entry,
Joonsoo Kim97654df2014-08-06 16:04:25 -0700985 shared->avail, node, &list);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000986 n->shared = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700987 }
988
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000989 alien = n->alien;
990 n->alien = NULL;
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700991
Christoph Lameterce8eb6c2013-01-10 19:14:19 +0000992 spin_unlock_irq(&n->list_lock);
Akinobu Mitafbf1e472007-10-18 03:05:09 -0700993
994 kfree(shared);
995 if (alien) {
996 drain_alien_cache(cachep, alien);
997 free_alien_cache(alien);
998 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -0700999
1000free_slab:
Joonsoo Kim97654df2014-08-06 16:04:25 -07001001 slabs_destroy(cachep, &list);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001002 }
1003 /*
1004 * In the previous loop, all the objects were freed to
1005 * the respective cache's slabs, now we can go ahead and
1006 * shrink each nodelist to its limit.
1007 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001008 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07001009 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001010 if (!n)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001011 continue;
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001012 drain_freelist(cachep, n, INT_MAX);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001013 }
1014}
1015
Paul Gortmaker0db06282013-06-19 14:53:51 -04001016static int cpuup_prepare(long cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001017{
Pekka Enberg343e0d72006-02-01 03:05:50 -08001018 struct kmem_cache *cachep;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07001019 int node = cpu_to_mem(cpu);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001020 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001022 /*
1023 * We need to do this right in the beginning since
1024 * alloc_arraycache's are going to use this list.
1025 * kmalloc_node allows us to add the slab to the right
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001026 * kmem_cache_node and not this cpu's kmem_cache_node
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001027 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001028 err = init_cache_node_node(node);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001029 if (err < 0)
1030 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001031
1032 /*
1033 * Now we can go ahead with allocating the shared arrays and
1034 * array caches
1035 */
Christoph Lameter18004c52012-07-06 15:25:12 -05001036 list_for_each_entry(cachep, &slab_caches, list) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07001037 err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
1038 if (err)
1039 goto bad;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001040 }
Pekka Enbergce79ddc2009-11-23 22:01:15 +02001041
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001042 return 0;
1043bad:
Akinobu Mita12d00f62007-10-18 03:05:11 -07001044 cpuup_canceled(cpu);
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001045 return -ENOMEM;
1046}
1047
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001048int slab_prepare_cpu(unsigned int cpu)
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001049{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001050 int err;
Akinobu Mitafbf1e472007-10-18 03:05:09 -07001051
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001052 mutex_lock(&slab_mutex);
1053 err = cpuup_prepare(cpu);
1054 mutex_unlock(&slab_mutex);
1055 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056}
1057
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001058/*
1059 * This is called for a failed online attempt and for a successful
1060 * offline.
1061 *
1062 * Even if all the cpus of a node are down, we don't free the
Xiao Yang221503e2020-08-06 23:18:31 -07001063 * kmem_cache_node of any cache. This to avoid a race between cpu_down, and
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001064 * a kmalloc allocation from another cpu for memory from the node of
Chen Tao70b6d252020-10-15 20:10:01 -07001065 * the cpu going down. The kmem_cache_node structure is usually allocated from
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001066 * kmem_cache_create() and gets destroyed at kmem_cache_destroy().
1067 */
1068int slab_dead_cpu(unsigned int cpu)
1069{
1070 mutex_lock(&slab_mutex);
1071 cpuup_canceled(cpu);
1072 mutex_unlock(&slab_mutex);
1073 return 0;
1074}
1075#endif
1076
1077static int slab_online_cpu(unsigned int cpu)
1078{
1079 start_cpu_timer(cpu);
1080 return 0;
1081}
1082
1083static int slab_offline_cpu(unsigned int cpu)
1084{
1085 /*
1086 * Shutdown cache reaper. Note that the slab_mutex is held so
1087 * that if cache_reap() is invoked it cannot do anything
1088 * expensive but will only modify reap_work and reschedule the
1089 * timer.
1090 */
1091 cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu));
1092 /* Now the cache_reaper is guaranteed to be not running. */
1093 per_cpu(slab_reap_work, cpu).work.func = NULL;
1094 return 0;
1095}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
David Rientjes8f9f8d92010-03-27 19:40:47 -07001097#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
1098/*
1099 * Drains freelist for a node on each slab cache, used for memory hot-remove.
1100 * Returns -EBUSY if all objects cannot be drained so that the node is not
1101 * removed.
1102 *
Christoph Lameter18004c52012-07-06 15:25:12 -05001103 * Must hold slab_mutex.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001104 */
Christoph Lameter6a673682013-01-10 19:14:19 +00001105static int __meminit drain_cache_node_node(int node)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001106{
1107 struct kmem_cache *cachep;
1108 int ret = 0;
1109
Christoph Lameter18004c52012-07-06 15:25:12 -05001110 list_for_each_entry(cachep, &slab_caches, list) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001111 struct kmem_cache_node *n;
David Rientjes8f9f8d92010-03-27 19:40:47 -07001112
Christoph Lameter18bf8542014-08-06 16:04:11 -07001113 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001114 if (!n)
David Rientjes8f9f8d92010-03-27 19:40:47 -07001115 continue;
1116
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07001117 drain_freelist(cachep, n, INT_MAX);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001118
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001119 if (!list_empty(&n->slabs_full) ||
1120 !list_empty(&n->slabs_partial)) {
David Rientjes8f9f8d92010-03-27 19:40:47 -07001121 ret = -EBUSY;
1122 break;
1123 }
1124 }
1125 return ret;
1126}
1127
1128static int __meminit slab_memory_callback(struct notifier_block *self,
1129 unsigned long action, void *arg)
1130{
1131 struct memory_notify *mnb = arg;
1132 int ret = 0;
1133 int nid;
1134
1135 nid = mnb->status_change_nid;
1136 if (nid < 0)
1137 goto out;
1138
1139 switch (action) {
1140 case MEM_GOING_ONLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001141 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001142 ret = init_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001143 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001144 break;
1145 case MEM_GOING_OFFLINE:
Christoph Lameter18004c52012-07-06 15:25:12 -05001146 mutex_lock(&slab_mutex);
Christoph Lameter6a673682013-01-10 19:14:19 +00001147 ret = drain_cache_node_node(nid);
Christoph Lameter18004c52012-07-06 15:25:12 -05001148 mutex_unlock(&slab_mutex);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001149 break;
1150 case MEM_ONLINE:
1151 case MEM_OFFLINE:
1152 case MEM_CANCEL_ONLINE:
1153 case MEM_CANCEL_OFFLINE:
1154 break;
1155 }
1156out:
Prarit Bhargava5fda1bd2011-03-22 16:30:49 -07001157 return notifier_from_errno(ret);
David Rientjes8f9f8d92010-03-27 19:40:47 -07001158}
1159#endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */
1160
Christoph Lametere498be72005-09-09 13:03:32 -07001161/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001162 * swap the static kmem_cache_node with kmalloced memory
Christoph Lametere498be72005-09-09 13:03:32 -07001163 */
Christoph Lameter6744f082013-01-10 19:12:17 +00001164static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list,
David Rientjes8f9f8d92010-03-27 19:40:47 -07001165 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07001166{
Christoph Lameter6744f082013-01-10 19:12:17 +00001167 struct kmem_cache_node *ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001168
Christoph Lameter6744f082013-01-10 19:12:17 +00001169 ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07001170 BUG_ON(!ptr);
1171
Christoph Lameter6744f082013-01-10 19:12:17 +00001172 memcpy(ptr, list, sizeof(struct kmem_cache_node));
Ingo Molnar2b2d5492006-07-03 00:25:28 -07001173 /*
1174 * Do not assume that spinlocks can be initialized via memcpy:
1175 */
1176 spin_lock_init(&ptr->list_lock);
1177
Christoph Lametere498be72005-09-09 13:03:32 -07001178 MAKE_ALL_LISTS(cachep, ptr, nodeid);
Christoph Lameter6a673682013-01-10 19:14:19 +00001179 cachep->node[nodeid] = ptr;
Christoph Lametere498be72005-09-09 13:03:32 -07001180}
1181
Andrew Mortona737b3e2006-03-22 00:08:11 -08001182/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001183 * For setting up all the kmem_cache_node for cache whose buffer_size is same as
1184 * size of kmem_cache_node.
Pekka Enberg556a1692008-01-25 08:20:51 +02001185 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001186static void __init set_up_node(struct kmem_cache *cachep, int index)
Pekka Enberg556a1692008-01-25 08:20:51 +02001187{
1188 int node;
1189
1190 for_each_online_node(node) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001191 cachep->node[node] = &init_kmem_cache_node[index + node];
Christoph Lameter6a673682013-01-10 19:14:19 +00001192 cachep->node[node]->next_reap = jiffies +
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001193 REAPTIMEOUT_NODE +
1194 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enberg556a1692008-01-25 08:20:51 +02001195 }
1196}
1197
1198/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08001199 * Initialisation. Called after the page allocator have been initialised and
1200 * before smp_init().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 */
1202void __init kmem_cache_init(void)
1203{
Christoph Lametere498be72005-09-09 13:03:32 -07001204 int i;
1205
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001206 kmem_cache = &kmem_cache_boot;
1207
Joonsoo Kim88881772016-05-19 17:10:05 -07001208 if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1)
Siddha, Suresh B62918a02007-05-02 19:27:18 +02001209 use_alien_caches = 0;
1210
Christoph Lameter3c583462012-11-28 16:23:01 +00001211 for (i = 0; i < NUM_INIT_LISTS; i++)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001212 kmem_cache_node_init(&init_kmem_cache_node[i]);
Christoph Lameter3c583462012-11-28 16:23:01 +00001213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 /*
1215 * Fragmentation resistance on low memory - only use bigger
David Rientjes3df1ccc2011-10-18 22:09:28 -07001216 * page orders on machines with more than 32MB of memory if
1217 * not overridden on the command line.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 */
Arun KSca79b0c2018-12-28 00:34:29 -08001219 if (!slab_max_order_set && totalram_pages() > (32 << 20) >> PAGE_SHIFT)
David Rientjes543585c2011-10-18 22:09:24 -07001220 slab_max_order = SLAB_MAX_ORDER_HI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 /* Bootstrap is tricky, because several objects are allocated
1223 * from caches that do not exist yet:
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001224 * 1) initialize the kmem_cache cache: it contains the struct
1225 * kmem_cache structures of all caches, except kmem_cache itself:
1226 * kmem_cache is statically allocated.
Christoph Lametere498be72005-09-09 13:03:32 -07001227 * Initially an __init data area is used for the head array and the
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001228 * kmem_cache_node structures, it's replaced with a kmalloc allocated
Christoph Lametere498be72005-09-09 13:03:32 -07001229 * array at the end of the bootstrap.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 * 2) Create the first kmalloc cache.
Pekka Enberg343e0d72006-02-01 03:05:50 -08001231 * The struct kmem_cache for the new cache is allocated normally.
Christoph Lametere498be72005-09-09 13:03:32 -07001232 * An __init data area is used for the head array.
1233 * 3) Create the remaining kmalloc caches, with minimally sized
1234 * head arrays.
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001235 * 4) Replace the __init data head arrays for kmem_cache and the first
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 * kmalloc cache with kmalloc allocated arrays.
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001237 * 5) Replace the __init data for kmem_cache_node for kmem_cache and
Christoph Lametere498be72005-09-09 13:03:32 -07001238 * the other cache's with kmalloc allocated memory.
1239 * 6) Resize the head arrays of the kmalloc caches to their final sizes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 */
1241
Christoph Lameter9b030cb2012-09-05 00:20:33 +00001242 /* 1) create the kmem_cache */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243
Eric Dumazet8da34302007-05-06 14:49:29 -07001244 /*
Eric Dumazetb56efcf2011-07-20 19:04:23 +02001245 * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids
Eric Dumazet8da34302007-05-06 14:49:29 -07001246 */
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001247 create_boot_cache(kmem_cache, "kmem_cache",
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001248 offsetof(struct kmem_cache, node) +
Christoph Lameter6744f082013-01-10 19:12:17 +00001249 nr_node_ids * sizeof(struct kmem_cache_node *),
David Windsor8eb82842017-06-10 22:50:28 -04001250 SLAB_HWCACHE_ALIGN, 0, 0);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001251 list_add(&kmem_cache->list, &slab_caches);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001252 slab_state = PARTIAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
Andrew Mortona737b3e2006-03-22 00:08:11 -08001254 /*
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001255 * Initialize the caches that provide memory for the kmem_cache_node
1256 * structures first. Without this, further allocations will bug.
Christoph Lametere498be72005-09-09 13:03:32 -07001257 */
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07001258 kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE] = create_kmalloc_cache(
Pengfei Licb5d9fb2019-11-30 17:49:21 -08001259 kmalloc_info[INDEX_NODE].name[KMALLOC_NORMAL],
Pengfei Lidc0a7f72019-11-30 17:49:25 -08001260 kmalloc_info[INDEX_NODE].size,
1261 ARCH_KMALLOC_FLAGS, 0,
1262 kmalloc_info[INDEX_NODE].size);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001263 slab_state = PARTIAL_NODE;
Daniel Sanders34cc6992015-06-24 16:55:57 -07001264 setup_kmalloc_cache_index_table();
Christoph Lametere498be72005-09-09 13:03:32 -07001265
Ingo Molnare0a42722006-06-23 02:03:46 -07001266 slab_early_init = 0;
1267
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001268 /* 5) Replace the bootstrap kmem_cache_node */
Christoph Lametere498be72005-09-09 13:03:32 -07001269 {
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07001270 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271
Mel Gorman9c09a952008-01-24 05:49:54 -08001272 for_each_online_node(nid) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001273 init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid);
Pekka Enberg556a1692008-01-25 08:20:51 +02001274
Vlastimil Babkacc252ea2018-10-26 15:05:34 -07001275 init_list(kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE],
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001276 &init_kmem_cache_node[SIZE_NODE + nid], nid);
Christoph Lametere498be72005-09-09 13:03:32 -07001277 }
1278 }
1279
Christoph Lameterf97d5f632013-01-10 19:12:17 +00001280 create_kmalloc_caches(ARCH_KMALLOC_FLAGS);
Pekka Enberg8429db52009-06-12 15:58:59 +03001281}
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001282
Pekka Enberg8429db52009-06-12 15:58:59 +03001283void __init kmem_cache_init_late(void)
1284{
1285 struct kmem_cache *cachep;
1286
Pekka Enberg8429db52009-06-12 15:58:59 +03001287 /* 6) resize the head arrays to their final sizes */
Christoph Lameter18004c52012-07-06 15:25:12 -05001288 mutex_lock(&slab_mutex);
1289 list_for_each_entry(cachep, &slab_caches, list)
Pekka Enberg8429db52009-06-12 15:58:59 +03001290 if (enable_cpucache(cachep, GFP_NOWAIT))
1291 BUG();
Christoph Lameter18004c52012-07-06 15:25:12 -05001292 mutex_unlock(&slab_mutex);
Ravikiran G Thirumalai056c6242006-09-25 23:31:38 -07001293
Christoph Lameter97d06602012-07-06 15:25:11 -05001294 /* Done! */
1295 slab_state = FULL;
1296
David Rientjes8f9f8d92010-03-27 19:40:47 -07001297#ifdef CONFIG_NUMA
1298 /*
1299 * Register a memory hotplug callback that initializes and frees
Christoph Lameter6a673682013-01-10 19:14:19 +00001300 * node.
David Rientjes8f9f8d92010-03-27 19:40:47 -07001301 */
1302 hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
1303#endif
1304
Andrew Mortona737b3e2006-03-22 00:08:11 -08001305 /*
1306 * The reap timers are started later, with a module init call: That part
1307 * of the kernel is not yet operational.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 */
1309}
1310
1311static int __init cpucache_init(void)
1312{
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001313 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Andrew Mortona737b3e2006-03-22 00:08:11 -08001315 /*
1316 * Register the timers that return unneeded pages to the page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 */
Sebastian Andrzej Siewior6731d4f2016-08-23 14:53:19 +02001318 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online",
1319 slab_online_cpu, slab_offline_cpu);
1320 WARN_ON(ret < 0);
Glauber Costaa164f8962012-06-21 00:59:18 +04001321
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 return 0;
1323}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324__initcall(cpucache_init);
1325
Rafael Aquini8bdec192012-03-09 17:27:27 -03001326static noinline void
1327slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
1328{
David Rientjes9a02d692014-06-04 16:06:36 -07001329#if DEBUG
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001330 struct kmem_cache_node *n;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001331 unsigned long flags;
1332 int node;
David Rientjes9a02d692014-06-04 16:06:36 -07001333 static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
1334 DEFAULT_RATELIMIT_BURST);
1335
1336 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs))
1337 return;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001338
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07001339 pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
1340 nodeid, gfpflags, &gfpflags);
1341 pr_warn(" cache: %s, object size: %d, order: %d\n",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001342 cachep->name, cachep->size, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001343
Christoph Lameter18bf8542014-08-06 16:04:11 -07001344 for_each_kmem_cache_node(cachep, node, n) {
David Rientjesbf00bd32016-12-12 16:41:44 -08001345 unsigned long total_slabs, free_slabs, free_objs;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001346
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001347 spin_lock_irqsave(&n->list_lock, flags);
David Rientjesbf00bd32016-12-12 16:41:44 -08001348 total_slabs = n->total_slabs;
1349 free_slabs = n->free_slabs;
1350 free_objs = n->free_objects;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00001351 spin_unlock_irqrestore(&n->list_lock, flags);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001352
David Rientjesbf00bd32016-12-12 16:41:44 -08001353 pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n",
1354 node, total_slabs - free_slabs, total_slabs,
1355 (total_slabs * cachep->num) - free_objs,
1356 total_slabs * cachep->num);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001357 }
David Rientjes9a02d692014-06-04 16:06:36 -07001358#endif
Rafael Aquini8bdec192012-03-09 17:27:27 -03001359}
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361/*
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001362 * Interface to system's page allocator. No need to hold the
1363 * kmem_cache_node ->list_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 *
1365 * If we requested dmaable memory, we will get it. Even if we
1366 * did not request dmaable memory, we might get it, but that
1367 * would be relatively rare and ignorable.
1368 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001369static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
1370 int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{
1372 struct page *page;
Christoph Lameter765c4502006-09-27 01:50:08 -07001373
Glauber Costaa618e892012-06-14 16:17:21 +04001374 flags |= cachep->allocflags;
Christoph Hellwige1b6aa62006-06-23 02:03:17 -07001375
Levin, Alexander (Sasha Levin)75f296d2017-11-15 17:35:54 -08001376 page = __alloc_pages_node(nodeid, flags, cachep->gfporder);
Rafael Aquini8bdec192012-03-09 17:27:27 -03001377 if (!page) {
David Rientjes9a02d692014-06-04 16:06:36 -07001378 slab_out_of_memory(cachep, flags, nodeid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return NULL;
Rafael Aquini8bdec192012-03-09 17:27:27 -03001380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381
Roman Gushchin74d555b2020-08-06 23:21:44 -07001382 account_slab_page(page, cachep->gfporder, cachep);
Joonsoo Kima57a4982013-10-24 10:07:44 +09001383 __SetPageSlab(page);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07001384 /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */
1385 if (sk_memalloc_socks() && page_is_pfmemalloc(page))
Joonsoo Kima57a4982013-10-24 10:07:44 +09001386 SetPageSlabPfmemalloc(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001387
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001388 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390
1391/*
1392 * Interface to system's page release.
1393 */
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001394static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001396 int order = cachep->gfporder;
Joonsoo Kim73293c22013-10-24 10:07:37 +09001397
Joonsoo Kima57a4982013-10-24 10:07:44 +09001398 BUG_ON(!PageSlab(page));
Joonsoo Kim73293c22013-10-24 10:07:37 +09001399 __ClearPageSlabPfmemalloc(page);
Joonsoo Kima57a4982013-10-24 10:07:44 +09001400 __ClearPageSlab(page);
Joonsoo Kim8456a642013-10-24 10:07:49 +09001401 page_mapcount_reset(page);
1402 page->mapping = NULL;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (current->reclaim_state)
Roman Gushchin6cea1d52019-07-11 20:56:16 -07001405 current->reclaim_state->reclaimed_slab += 1 << order;
Roman Gushchin74d555b2020-08-06 23:21:44 -07001406 unaccount_slab_page(page, order, cachep);
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001407 __free_pages(page, order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408}
1409
1410static void kmem_rcu_free(struct rcu_head *head)
1411{
Joonsoo Kim68126702013-10-24 10:07:42 +09001412 struct kmem_cache *cachep;
1413 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Joonsoo Kim68126702013-10-24 10:07:42 +09001415 page = container_of(head, struct page, rcu_head);
1416 cachep = page->slab_cache;
1417
1418 kmem_freepages(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419}
1420
1421#if DEBUG
Joonsoo Kim40b44132016-03-15 14:54:21 -07001422static bool is_debug_pagealloc_cache(struct kmem_cache *cachep)
1423{
Vlastimil Babka8e57f8a2020-01-13 16:29:20 -08001424 if (debug_pagealloc_enabled_static() && OFF_SLAB(cachep) &&
Joonsoo Kim40b44132016-03-15 14:54:21 -07001425 (cachep->size % PAGE_SIZE) == 0)
1426 return true;
1427
1428 return false;
1429}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431#ifdef CONFIG_DEBUG_PAGEALLOC
Qian Cai80552f02019-04-16 10:22:57 -04001432static void slab_kernel_map(struct kmem_cache *cachep, void *objp, int map)
Joonsoo Kim40b44132016-03-15 14:54:21 -07001433{
1434 if (!is_debug_pagealloc_cache(cachep))
1435 return;
1436
Joonsoo Kim40b44132016-03-15 14:54:21 -07001437 kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map);
1438}
1439
1440#else
1441static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp,
Qian Cai80552f02019-04-16 10:22:57 -04001442 int map) {}
Joonsoo Kim40b44132016-03-15 14:54:21 -07001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444#endif
1445
Pekka Enberg343e0d72006-02-01 03:05:50 -08001446static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447{
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001448 int size = cachep->object_size;
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001449 addr = &((char *)addr)[obj_offset(cachep)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
1451 memset(addr, val, size);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001452 *(unsigned char *)(addr + size - 1) = POISON_END;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
1454
1455static void dump_line(char *data, int offset, int limit)
1456{
1457 int i;
Dave Jonesaa83aa42006-09-29 01:59:51 -07001458 unsigned char error = 0;
1459 int bad_count = 0;
1460
Joe Perches11705322016-03-17 14:19:50 -07001461 pr_err("%03x: ", offset);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001462 for (i = 0; i < limit; i++) {
1463 if (data[offset + i] != POISON_FREE) {
1464 error = data[offset + i];
1465 bad_count++;
1466 }
Dave Jonesaa83aa42006-09-29 01:59:51 -07001467 }
Sebastian Andrzej Siewiorfdde6ab2011-07-29 18:22:13 +02001468 print_hex_dump(KERN_CONT, "", 0, 16, 1,
1469 &data[offset], limit, 1);
Dave Jonesaa83aa42006-09-29 01:59:51 -07001470
1471 if (bad_count == 1) {
1472 error ^= POISON_FREE;
1473 if (!(error & (error - 1))) {
Joe Perches11705322016-03-17 14:19:50 -07001474 pr_err("Single bit error detected. Probably bad RAM.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001475#ifdef CONFIG_X86
Joe Perches11705322016-03-17 14:19:50 -07001476 pr_err("Run memtest86+ or a similar memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001477#else
Joe Perches11705322016-03-17 14:19:50 -07001478 pr_err("Run a memory test tool.\n");
Dave Jonesaa83aa42006-09-29 01:59:51 -07001479#endif
1480 }
1481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482}
1483#endif
1484
1485#if DEBUG
1486
Pekka Enberg343e0d72006-02-01 03:05:50 -08001487static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
1489 int i, size;
1490 char *realobj;
1491
1492 if (cachep->flags & SLAB_RED_ZONE) {
Joe Perches11705322016-03-17 14:19:50 -07001493 pr_err("Redzone: 0x%llx/0x%llx\n",
1494 *dbg_redzone1(cachep, objp),
1495 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 }
1497
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001498 if (cachep->flags & SLAB_STORE_USER)
1499 pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp));
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001500 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001501 size = cachep->object_size;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001502 for (i = 0; i < size && lines; i += 16, lines--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 int limit;
1504 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001505 if (i + limit > size)
1506 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 dump_line(realobj, i, limit);
1508 }
1509}
1510
Pekka Enberg343e0d72006-02-01 03:05:50 -08001511static void check_poison_obj(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512{
1513 char *realobj;
1514 int size, i;
1515 int lines = 0;
1516
Joonsoo Kim40b44132016-03-15 14:54:21 -07001517 if (is_debug_pagealloc_cache(cachep))
1518 return;
1519
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001520 realobj = (char *)objp + obj_offset(cachep);
Christoph Lameter8c138bc2012-06-13 10:24:58 -05001521 size = cachep->object_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001523 for (i = 0; i < size; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 char exp = POISON_FREE;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001525 if (i == size - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 exp = POISON_END;
1527 if (realobj[i] != exp) {
1528 int limit;
1529 /* Mismatch ! */
1530 /* Print header */
1531 if (lines == 0) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001532 pr_err("Slab corruption (%s): %s start=%px, len=%d\n",
Joe Perches11705322016-03-17 14:19:50 -07001533 print_tainted(), cachep->name,
1534 realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 print_objinfo(cachep, objp, 0);
1536 }
1537 /* Hexdump the affected line */
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001538 i = (i / 16) * 16;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 limit = 16;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001540 if (i + limit > size)
1541 limit = size - i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 dump_line(realobj, i, limit);
1543 i += 16;
1544 lines++;
1545 /* Limit to 5 lines */
1546 if (lines > 5)
1547 break;
1548 }
1549 }
1550 if (lines != 0) {
1551 /* Print some data about the neighboring objects, if they
1552 * exist:
1553 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001554 struct page *page = virt_to_head_page(objp);
Pekka Enberg8fea4e92006-03-22 00:08:10 -08001555 unsigned int objnr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556
Joonsoo Kim8456a642013-10-24 10:07:49 +09001557 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 if (objnr) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001559 objp = index_to_obj(cachep, page, objnr - 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001560 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001561 pr_err("Prev obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 print_objinfo(cachep, objp, 2);
1563 }
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001564 if (objnr + 1 < cachep->num) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001565 objp = index_to_obj(cachep, page, objnr + 1);
Manfred Spraul3dafccf2006-02-01 03:05:42 -08001566 realobj = (char *)objp + obj_offset(cachep);
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08001567 pr_err("Next obj: start=%px, len=%d\n", realobj, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 print_objinfo(cachep, objp, 2);
1569 }
1570 }
1571}
1572#endif
1573
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574#if DEBUG
Joonsoo Kim8456a642013-10-24 10:07:49 +09001575static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1576 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001577{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 int i;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001579
1580 if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) {
1581 poison_obj(cachep, page->freelist - obj_offset(cachep),
1582 POISON_FREE);
1583 }
1584
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09001586 void *objp = index_to_obj(cachep, page, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
1588 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 check_poison_obj(cachep, objp);
Qian Cai80552f02019-04-16 10:22:57 -04001590 slab_kernel_map(cachep, objp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 }
1592 if (cachep->flags & SLAB_RED_ZONE) {
1593 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001594 slab_error(cachep, "start of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07001596 slab_error(cachep, "end of a freed object was overwritten");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001599}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600#else
Joonsoo Kim8456a642013-10-24 10:07:49 +09001601static void slab_destroy_debugcheck(struct kmem_cache *cachep,
1602 struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001603{
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001604}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605#endif
1606
Randy Dunlap911851e2006-03-22 00:08:14 -08001607/**
1608 * slab_destroy - destroy and release all objects in a slab
1609 * @cachep: cache pointer being destroyed
Masanari Iidacb8ee1a2014-01-28 02:57:08 +09001610 * @page: page pointer being destroyed
Randy Dunlap911851e2006-03-22 00:08:14 -08001611 *
Wang Sheng-Hui8a7d9b42014-08-06 16:04:46 -07001612 * Destroy all the objs in a slab page, and release the mem back to the system.
1613 * Before calling the slab page must have been unlinked from the cache. The
1614 * kmem_cache_node ->list_lock is not held/needed.
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001615 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09001616static void slab_destroy(struct kmem_cache *cachep, struct page *page)
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001617{
Joonsoo Kim7e007352013-10-30 19:04:01 +09001618 void *freelist;
Matthew Dobson12dd36f2006-02-01 03:05:46 -08001619
Joonsoo Kim8456a642013-10-24 10:07:49 +09001620 freelist = page->freelist;
1621 slab_destroy_debugcheck(cachep, page);
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001622 if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU))
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001623 call_rcu(&page->rcu_head, kmem_rcu_free);
1624 else
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09001625 kmem_freepages(cachep, page);
Joonsoo Kim68126702013-10-24 10:07:42 +09001626
1627 /*
Joonsoo Kim8456a642013-10-24 10:07:49 +09001628 * From now on, we don't use freelist
Joonsoo Kim68126702013-10-24 10:07:42 +09001629 * although actual page can be freed in rcu context
1630 */
1631 if (OFF_SLAB(cachep))
Joonsoo Kim8456a642013-10-24 10:07:49 +09001632 kmem_cache_free(cachep->freelist_cache, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633}
1634
Shakeel Butt678ff6a2020-09-26 07:13:41 -07001635/*
1636 * Update the size of the caches before calling slabs_destroy as it may
1637 * recursively call kfree.
1638 */
Joonsoo Kim97654df2014-08-06 16:04:25 -07001639static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list)
1640{
1641 struct page *page, *n;
1642
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07001643 list_for_each_entry_safe(page, n, list, slab_list) {
1644 list_del(&page->slab_list);
Joonsoo Kim97654df2014-08-06 16:04:25 -07001645 slab_destroy(cachep, page);
1646 }
1647}
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649/**
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001650 * calculate_slab_order - calculate size (page order) of slabs
1651 * @cachep: pointer to the cache that is being created
1652 * @size: size of objects to be created in this cache.
Randy.Dunlapa70773d2006-02-01 03:05:52 -08001653 * @flags: slab allocation flags
1654 *
1655 * Also calculates the number of objects per slab.
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001656 *
1657 * This could be made much more intelligent. For now, try to avoid using
1658 * high order pages for slabs. When the gfp() functions are more friendly
1659 * towards high-order requests, this should be changed.
Mike Rapoporta862f682019-03-05 15:48:42 -08001660 *
1661 * Return: number of left-over bytes in a slab
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001662 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001663static size_t calculate_slab_order(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001664 size_t size, slab_flags_t flags)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001665{
1666 size_t left_over = 0;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001667 int gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001668
Christoph Lameter0aa817f2007-05-16 22:11:01 -07001669 for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001670 unsigned int num;
1671 size_t remainder;
1672
Joonsoo Kim70f75062016-03-15 14:54:53 -07001673 num = cache_estimate(gfporder, size, flags, &remainder);
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001674 if (!num)
1675 continue;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001676
Joonsoo Kimf315e3f2013-12-02 17:49:41 +09001677 /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */
1678 if (num > SLAB_OBJ_MAX_NUM)
1679 break;
1680
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001681 if (flags & CFLGS_OFF_SLAB) {
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001682 struct kmem_cache *freelist_cache;
1683 size_t freelist_size;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001684
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001685 freelist_size = num * sizeof(freelist_idx_t);
1686 freelist_cache = kmalloc_slab(freelist_size, 0u);
1687 if (!freelist_cache)
1688 continue;
1689
1690 /*
1691 * Needed to avoid possible looping condition
Joonsoo Kim76b342b2016-05-19 17:10:26 -07001692 * in cache_grow_begin()
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001693 */
1694 if (OFF_SLAB(freelist_cache))
1695 continue;
1696
1697 /* check if off slab has enough benefit */
1698 if (freelist_cache->size > cachep->size / 2)
1699 continue;
Ingo Molnarb1ab41c2006-06-02 15:44:58 +02001700 }
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001701
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001702 /* Found something acceptable - save it away */
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001703 cachep->num = num;
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001704 cachep->gfporder = gfporder;
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001705 left_over = remainder;
1706
1707 /*
Linus Torvaldsf78bb8a2006-03-08 10:33:05 -08001708 * A VFS-reclaimable slab tends to have most allocations
1709 * as GFP_NOFS and we really don't want to have to be allocating
1710 * higher-order pages when we are unable to shrink dcache.
1711 */
1712 if (flags & SLAB_RECLAIM_ACCOUNT)
1713 break;
1714
1715 /*
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001716 * Large number of objects is good, but very large slabs are
1717 * currently bad for the gfp()s.
1718 */
David Rientjes543585c2011-10-18 22:09:24 -07001719 if (gfporder >= slab_max_order)
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001720 break;
1721
Linus Torvalds9888e6f2006-03-06 17:44:43 -08001722 /*
1723 * Acceptable internal fragmentation?
1724 */
Andrew Mortona737b3e2006-03-22 00:08:11 -08001725 if (left_over * 8 <= (PAGE_SIZE << gfporder))
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001726 break;
1727 }
1728 return left_over;
1729}
1730
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001731static struct array_cache __percpu *alloc_kmem_cache_cpus(
1732 struct kmem_cache *cachep, int entries, int batchcount)
1733{
1734 int cpu;
1735 size_t size;
1736 struct array_cache __percpu *cpu_cache;
1737
1738 size = sizeof(void *) * entries + sizeof(struct array_cache);
Joonsoo Kim85c9f4b2014-10-13 15:51:01 -07001739 cpu_cache = __alloc_percpu(size, sizeof(void *));
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001740
1741 if (!cpu_cache)
1742 return NULL;
1743
1744 for_each_possible_cpu(cpu) {
1745 init_arraycache(per_cpu_ptr(cpu_cache, cpu),
1746 entries, batchcount);
1747 }
1748
1749 return cpu_cache;
1750}
1751
Fabian Frederickbd721ea2016-08-02 14:03:33 -07001752static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001753{
Christoph Lameter97d06602012-07-06 15:25:11 -05001754 if (slab_state >= FULL)
Pekka Enberg83b519e2009-06-10 19:40:04 +03001755 return enable_cpucache(cachep, gfp);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001756
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001757 cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1);
1758 if (!cachep->cpu_cache)
1759 return 1;
1760
Christoph Lameter97d06602012-07-06 15:25:11 -05001761 if (slab_state == DOWN) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001762 /* Creation of first cache (kmem_cache). */
1763 set_up_node(kmem_cache, CACHE_CACHE);
Christoph Lameter2f9baa92012-11-28 16:23:09 +00001764 } else if (slab_state == PARTIAL) {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001765 /* For kmem_cache_node */
1766 set_up_node(cachep, SIZE_NODE);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001767 } else {
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001768 int node;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001769
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001770 for_each_online_node(node) {
1771 cachep->node[node] = kmalloc_node(
1772 sizeof(struct kmem_cache_node), gfp, node);
1773 BUG_ON(!cachep->node[node]);
1774 kmem_cache_node_init(cachep->node[node]);
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001775 }
1776 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07001777
Christoph Lameter6a673682013-01-10 19:14:19 +00001778 cachep->node[numa_mem_id()]->next_reap =
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08001779 jiffies + REAPTIMEOUT_NODE +
1780 ((unsigned long)cachep) % REAPTIMEOUT_NODE;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001781
1782 cpu_cache_get(cachep)->avail = 0;
1783 cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
1784 cpu_cache_get(cachep)->batchcount = 1;
1785 cpu_cache_get(cachep)->touched = 0;
1786 cachep->batchcount = 1;
1787 cachep->limit = BOOT_CPUCACHE_ENTRIES;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07001788 return 0;
Pekka Enbergf30cf7d2006-03-22 00:08:11 -08001789}
1790
Alexey Dobriyan0293d1f2018-04-05 16:21:24 -07001791slab_flags_t kmem_cache_flags(unsigned int object_size,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001792 slab_flags_t flags, const char *name,
Joonsoo Kim12220de2014-10-09 15:26:24 -07001793 void (*ctor)(void *))
1794{
1795 return flags;
1796}
1797
1798struct kmem_cache *
Alexey Dobriyanf4957d52018-04-05 16:20:37 -07001799__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001800 slab_flags_t flags, void (*ctor)(void *))
Joonsoo Kim12220de2014-10-09 15:26:24 -07001801{
1802 struct kmem_cache *cachep;
1803
1804 cachep = find_mergeable(size, align, flags, name, ctor);
1805 if (cachep) {
1806 cachep->refcount++;
1807
1808 /*
1809 * Adjust the object sizes so that we clear
1810 * the complete object on kzalloc.
1811 */
1812 cachep->object_size = max_t(int, cachep->object_size, size);
1813 }
1814 return cachep;
1815}
1816
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001817static bool set_objfreelist_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001818 size_t size, slab_flags_t flags)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001819{
1820 size_t left;
1821
1822 cachep->num = 0;
1823
Alexander Potapenko64713842019-07-11 20:59:19 -07001824 /*
1825 * If slab auto-initialization on free is enabled, store the freelist
1826 * off-slab, so that its contents don't end up in one of the allocated
1827 * objects.
1828 */
1829 if (unlikely(slab_want_init_on_free(cachep)))
1830 return false;
1831
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001832 if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU)
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07001833 return false;
1834
1835 left = calculate_slab_order(cachep, size,
1836 flags | CFLGS_OBJFREELIST_SLAB);
1837 if (!cachep->num)
1838 return false;
1839
1840 if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size)
1841 return false;
1842
1843 cachep->colour = left / cachep->colour_off;
1844
1845 return true;
1846}
1847
Joonsoo Kim158e3192016-03-15 14:54:35 -07001848static bool set_off_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001849 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001850{
1851 size_t left;
1852
1853 cachep->num = 0;
1854
1855 /*
Joonsoo Kim3217fd92016-03-15 14:54:41 -07001856 * Always use on-slab management when SLAB_NOLEAKTRACE
1857 * to avoid recursive calls into kmemleak.
Joonsoo Kim158e3192016-03-15 14:54:35 -07001858 */
Joonsoo Kim158e3192016-03-15 14:54:35 -07001859 if (flags & SLAB_NOLEAKTRACE)
1860 return false;
1861
1862 /*
1863 * Size is large, assume best to place the slab management obj
1864 * off-slab (should allow better packing of objs).
1865 */
1866 left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB);
1867 if (!cachep->num)
1868 return false;
1869
1870 /*
1871 * If the slab has been placed off-slab, and we have enough space then
1872 * move it on-slab. This is at the expense of any extra colouring.
1873 */
1874 if (left >= cachep->num * sizeof(freelist_idx_t))
1875 return false;
1876
1877 cachep->colour = left / cachep->colour_off;
1878
1879 return true;
1880}
1881
1882static bool set_on_slab_cache(struct kmem_cache *cachep,
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001883 size_t size, slab_flags_t flags)
Joonsoo Kim158e3192016-03-15 14:54:35 -07001884{
1885 size_t left;
1886
1887 cachep->num = 0;
1888
1889 left = calculate_slab_order(cachep, size, flags);
1890 if (!cachep->num)
1891 return false;
1892
1893 cachep->colour = left / cachep->colour_off;
1894
1895 return true;
1896}
1897
Pekka Enberg4d268eb2006-01-08 01:00:36 -08001898/**
Christoph Lameter039363f2012-07-06 15:25:10 -05001899 * __kmem_cache_create - Create a cache.
Randy Dunlapa755b762012-11-06 17:10:10 -08001900 * @cachep: cache management descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 * @flags: SLAB flags
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 *
1903 * Returns a ptr to the cache on success, NULL on failure.
1904 * Cannot be called within a int, but can be interrupted.
Paul Mundt20c2df82007-07-20 10:11:58 +09001905 * The @ctor is run when new pages are allocated by the cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 * The flags are
1908 *
1909 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
1910 * to catch references to uninitialised memory.
1911 *
1912 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
1913 * for buffer overruns.
1914 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
1916 * cacheline. This can be beneficial if you're counting cycles as closely
1917 * as davem.
Mike Rapoporta862f682019-03-05 15:48:42 -08001918 *
1919 * Return: a pointer to the created cache or %NULL in case of error
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 */
Alexey Dobriyand50112e2017-11-15 17:32:18 -08001921int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
David Rientjesd4a5fca52014-09-25 16:05:20 -07001923 size_t ralign = BYTES_PER_WORD;
Pekka Enberg83b519e2009-06-10 19:40:04 +03001924 gfp_t gfp;
Christoph Lameter278b1bb2012-09-05 00:20:34 +00001925 int err;
Alexey Dobriyanbe4a7982018-04-05 16:21:28 -07001926 unsigned int size = cachep->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929#if FORCED_DEBUG
1930 /*
1931 * Enable redzoning and last user accounting, except for caches with
1932 * large objects, if the increased size would increase the object size
1933 * above the next power of two: caches with object sizes just above a
1934 * power of two have a significant amount of internal fragmentation.
1935 */
David Woodhouse87a927c2007-07-04 21:26:44 -04001936 if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
1937 2 * sizeof(unsigned long long)))
Pekka Enbergb28a02d2006-01-08 01:00:37 -08001938 flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
Paul E. McKenney5f0d5a32017-01-18 02:53:44 -08001939 if (!(flags & SLAB_TYPESAFE_BY_RCU))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 flags |= SLAB_POISON;
1941#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
Andrew Mortona737b3e2006-03-22 00:08:11 -08001944 /*
1945 * Check that size is in terms of words. This is needed to avoid
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 * unaligned accesses for some archs when redzoning is used, and makes
1947 * sure any on-slab bufctl's are also correctly aligned.
1948 */
Canjiang Lue0771952017-07-06 15:36:37 -07001949 size = ALIGN(size, BYTES_PER_WORD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
David Woodhouse87a927c2007-07-04 21:26:44 -04001951 if (flags & SLAB_RED_ZONE) {
1952 ralign = REDZONE_ALIGN;
1953 /* If redzoning, ensure that the second redzone is suitably
1954 * aligned, by adjusting the object size accordingly. */
Canjiang Lue0771952017-07-06 15:36:37 -07001955 size = ALIGN(size, REDZONE_ALIGN);
David Woodhouse87a927c2007-07-04 21:26:44 -04001956 }
Pekka Enbergca5f9702006-09-25 23:31:25 -07001957
Kevin Hilmana44b56d2006-12-06 20:32:11 -08001958 /* 3) caller mandated alignment */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00001959 if (ralign < cachep->align) {
1960 ralign = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
Pekka Enberg3ff84a72011-02-14 17:46:21 +02001962 /* disable debug if necessary */
1963 if (ralign > __alignof__(unsigned long long))
Kevin Hilmana44b56d2006-12-06 20:32:11 -08001964 flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
Andrew Mortona737b3e2006-03-22 00:08:11 -08001965 /*
Pekka Enbergca5f9702006-09-25 23:31:25 -07001966 * 4) Store it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 */
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00001968 cachep->align = ralign;
Joonsoo Kim158e3192016-03-15 14:54:35 -07001969 cachep->colour_off = cache_line_size();
1970 /* Offset must be a multiple of the alignment. */
1971 if (cachep->colour_off < cachep->align)
1972 cachep->colour_off = cachep->align;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973
Pekka Enberg83b519e2009-06-10 19:40:04 +03001974 if (slab_is_available())
1975 gfp = GFP_KERNEL;
1976 else
1977 gfp = GFP_NOWAIT;
1978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Pekka Enbergca5f9702006-09-25 23:31:25 -07001981 /*
1982 * Both debugging options require word-alignment which is calculated
1983 * into align above.
1984 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 if (flags & SLAB_RED_ZONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 /* add space for red zone words */
Pekka Enberg3ff84a72011-02-14 17:46:21 +02001987 cachep->obj_offset += sizeof(unsigned long long);
1988 size += 2 * sizeof(unsigned long long);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 }
1990 if (flags & SLAB_STORE_USER) {
Pekka Enbergca5f9702006-09-25 23:31:25 -07001991 /* user store requires one word storage behind the end of
David Woodhouse87a927c2007-07-04 21:26:44 -04001992 * the real object. But if the second red zone needs to be
1993 * aligned to 64 bits, we must allow that much space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 */
David Woodhouse87a927c2007-07-04 21:26:44 -04001995 if (flags & SLAB_RED_ZONE)
1996 size += REDZONE_ALIGN;
1997 else
1998 size += BYTES_PER_WORD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002000#endif
2001
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002002 kasan_cache_create(cachep, &size, &flags);
2003
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002004 size = ALIGN(size, cachep->align);
2005 /*
2006 * We should restrict the number of objects in a slab to implement
2007 * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition.
2008 */
2009 if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE)
2010 size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align);
2011
2012#if DEBUG
Joonsoo Kim03a2d2a2015-10-01 15:36:54 -07002013 /*
2014 * To activate debug pagealloc, off-slab management is necessary
2015 * requirement. In early phase of initialization, small sized slab
2016 * doesn't get initialized so it would not be possible. So, we need
2017 * to check size >= 256. It guarantees that all necessary small
2018 * sized slab is initialized in current slab initialization sequence.
2019 */
Vlastimil Babka8e57f8a2020-01-13 16:29:20 -08002020 if (debug_pagealloc_enabled_static() && (flags & SLAB_POISON) &&
Joonsoo Kimf3a3c322016-03-15 14:54:38 -07002021 size >= 256 && cachep->object_size > cache_line_size()) {
2022 if (size < PAGE_SIZE || size % PAGE_SIZE == 0) {
2023 size_t tmp_size = ALIGN(size, PAGE_SIZE);
2024
2025 if (set_off_slab_cache(cachep, tmp_size, flags)) {
2026 flags |= CFLGS_OFF_SLAB;
2027 cachep->obj_offset += tmp_size - size;
2028 size = tmp_size;
2029 goto done;
2030 }
2031 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 }
2033#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002035 if (set_objfreelist_slab_cache(cachep, size, flags)) {
2036 flags |= CFLGS_OBJFREELIST_SLAB;
2037 goto done;
2038 }
2039
Joonsoo Kim158e3192016-03-15 14:54:35 -07002040 if (set_off_slab_cache(cachep, size, flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 flags |= CFLGS_OFF_SLAB;
Joonsoo Kim158e3192016-03-15 14:54:35 -07002042 goto done;
Joonsoo Kim832a15d2016-03-15 14:54:33 -07002043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
Joonsoo Kim158e3192016-03-15 14:54:35 -07002045 if (set_on_slab_cache(cachep, size, flags))
2046 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Joonsoo Kim158e3192016-03-15 14:54:35 -07002048 return -E2BIG;
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00002049
Joonsoo Kim158e3192016-03-15 14:54:35 -07002050done:
2051 cachep->freelist_size = cachep->num * sizeof(freelist_idx_t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 cachep->flags = flags;
Joonsoo Kima57a4982013-10-24 10:07:44 +09002053 cachep->allocflags = __GFP_COMP;
Yang Shia3187e42016-05-19 17:10:41 -07002054 if (flags & SLAB_CACHE_DMA)
Glauber Costaa618e892012-06-14 16:17:21 +04002055 cachep->allocflags |= GFP_DMA;
Nicolas Boichat6d6ea1e2019-03-28 20:43:42 -07002056 if (flags & SLAB_CACHE_DMA32)
2057 cachep->allocflags |= GFP_DMA32;
David Rientjesa3ba0742017-11-15 17:32:14 -08002058 if (flags & SLAB_RECLAIM_ACCOUNT)
2059 cachep->allocflags |= __GFP_RECLAIMABLE;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002060 cachep->size = size;
Eric Dumazet6a2d7a92006-12-13 00:34:27 -08002061 cachep->reciprocal_buffer_size = reciprocal_value(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
Joonsoo Kim40b44132016-03-15 14:54:21 -07002063#if DEBUG
2064 /*
2065 * If we're going to use the generic kernel_map_pages()
2066 * poisoning, then it's going to smash the contents of
2067 * the redzone and userword anyhow, so switch them off.
2068 */
2069 if (IS_ENABLED(CONFIG_PAGE_POISONING) &&
2070 (cachep->flags & SLAB_POISON) &&
2071 is_debug_pagealloc_cache(cachep))
2072 cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
2073#endif
2074
2075 if (OFF_SLAB(cachep)) {
Joonsoo Kim158e3192016-03-15 14:54:35 -07002076 cachep->freelist_cache =
2077 kmalloc_slab(cachep->freelist_size, 0u);
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002078 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002080 err = setup_cpu_cache(cachep, gfp);
2081 if (err) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002082 __kmem_cache_release(cachep);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002083 return err;
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07002084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085
Christoph Lameter278b1bb2012-09-05 00:20:34 +00002086 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
2089#if DEBUG
2090static void check_irq_off(void)
2091{
2092 BUG_ON(!irqs_disabled());
2093}
2094
2095static void check_irq_on(void)
2096{
2097 BUG_ON(irqs_disabled());
2098}
2099
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002100static void check_mutex_acquired(void)
2101{
2102 BUG_ON(!mutex_is_locked(&slab_mutex));
2103}
2104
Pekka Enberg343e0d72006-02-01 03:05:50 -08002105static void check_spinlock_acquired(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
2107#ifdef CONFIG_SMP
2108 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002109 assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110#endif
2111}
Christoph Lametere498be72005-09-09 13:03:32 -07002112
Pekka Enberg343e0d72006-02-01 03:05:50 -08002113static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
Christoph Lametere498be72005-09-09 13:03:32 -07002114{
2115#ifdef CONFIG_SMP
2116 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002117 assert_spin_locked(&get_node(cachep, node)->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07002118#endif
2119}
2120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121#else
2122#define check_irq_off() do { } while(0)
2123#define check_irq_on() do { } while(0)
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002124#define check_mutex_acquired() do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125#define check_spinlock_acquired(x) do { } while(0)
Christoph Lametere498be72005-09-09 13:03:32 -07002126#define check_spinlock_acquired_node(x, y) do { } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127#endif
2128
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002129static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac,
2130 int node, bool free_all, struct list_head *list)
2131{
2132 int tofree;
2133
2134 if (!ac || !ac->avail)
2135 return;
2136
2137 tofree = free_all ? ac->avail : (ac->limit + 4) / 5;
2138 if (tofree > ac->avail)
2139 tofree = (ac->avail + 1) / 2;
2140
2141 free_block(cachep, ac->entry, tofree, node, list);
2142 ac->avail -= tofree;
2143 memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail);
2144}
Christoph Lameteraab22072006-03-22 00:09:06 -08002145
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146static void do_drain(void *arg)
2147{
Andrew Mortona737b3e2006-03-22 00:08:11 -08002148 struct kmem_cache *cachep = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 struct array_cache *ac;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002150 int node = numa_mem_id();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002151 struct kmem_cache_node *n;
Joonsoo Kim97654df2014-08-06 16:04:25 -07002152 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 check_irq_off();
Pekka Enberg9a2dba42006-02-01 03:05:49 -08002155 ac = cpu_cache_get(cachep);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002156 n = get_node(cachep, node);
2157 spin_lock(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07002158 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07002159 spin_unlock(&n->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 ac->avail = 0;
Shakeel Butt678ff6a2020-09-26 07:13:41 -07002161 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162}
2163
Pekka Enberg343e0d72006-02-01 03:05:50 -08002164static void drain_cpu_caches(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165{
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002166 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002167 int node;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002168 LIST_HEAD(list);
Christoph Lametere498be72005-09-09 13:03:32 -07002169
Jens Axboe15c8b6c2008-05-09 09:39:44 +02002170 on_each_cpu(do_drain, cachep, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002172 for_each_kmem_cache_node(cachep, node, n)
2173 if (n->alien)
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002174 drain_alien_cache(cachep, n->alien);
Roland Dreiera4523a82006-05-15 11:41:00 -07002175
Joonsoo Kim18726ca2016-05-19 17:10:02 -07002176 for_each_kmem_cache_node(cachep, node, n) {
2177 spin_lock_irq(&n->list_lock);
2178 drain_array_locked(cachep, n->shared, node, true, &list);
2179 spin_unlock_irq(&n->list_lock);
2180
2181 slabs_destroy(cachep, &list);
2182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183}
2184
Christoph Lametered11d9e2006-06-30 01:55:45 -07002185/*
2186 * Remove slabs from the list of free slabs.
2187 * Specify the number of slabs to drain in tofree.
2188 *
2189 * Returns the actual number of slabs released.
2190 */
2191static int drain_freelist(struct kmem_cache *cache,
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002192 struct kmem_cache_node *n, int tofree)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193{
Christoph Lametered11d9e2006-06-30 01:55:45 -07002194 struct list_head *p;
2195 int nr_freed;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002196 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197
Christoph Lametered11d9e2006-06-30 01:55:45 -07002198 nr_freed = 0;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002199 while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002201 spin_lock_irq(&n->list_lock);
2202 p = n->slabs_free.prev;
2203 if (p == &n->slabs_free) {
2204 spin_unlock_irq(&n->list_lock);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002205 goto out;
2206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002208 page = list_entry(p, struct page, slab_list);
2209 list_del(&page->slab_list);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002210 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08002211 n->total_slabs--;
Christoph Lametered11d9e2006-06-30 01:55:45 -07002212 /*
2213 * Safe to drop the lock. The slab is no longer linked
2214 * to the cache.
2215 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002216 n->free_objects -= cache->num;
2217 spin_unlock_irq(&n->list_lock);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002218 slab_destroy(cache, page);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002219 nr_freed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 }
Christoph Lametered11d9e2006-06-30 01:55:45 -07002221out:
2222 return nr_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223}
2224
Shakeel Buttf9e13c02018-04-05 16:21:57 -07002225bool __kmem_cache_empty(struct kmem_cache *s)
2226{
2227 int node;
2228 struct kmem_cache_node *n;
2229
2230 for_each_kmem_cache_node(s, node, n)
2231 if (!list_empty(&n->slabs_full) ||
2232 !list_empty(&n->slabs_partial))
2233 return false;
2234 return true;
2235}
2236
Tejun Heoc9fc5862017-02-22 15:41:27 -08002237int __kmem_cache_shrink(struct kmem_cache *cachep)
Christoph Lametere498be72005-09-09 13:03:32 -07002238{
Christoph Lameter18bf8542014-08-06 16:04:11 -07002239 int ret = 0;
2240 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002241 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07002242
2243 drain_cpu_caches(cachep);
2244
2245 check_irq_on();
Christoph Lameter18bf8542014-08-06 16:04:11 -07002246 for_each_kmem_cache_node(cachep, node, n) {
Joonsoo Kima5aa63a2016-05-19 17:10:08 -07002247 drain_freelist(cachep, n, INT_MAX);
Christoph Lametered11d9e2006-06-30 01:55:45 -07002248
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002249 ret += !list_empty(&n->slabs_full) ||
2250 !list_empty(&n->slabs_partial);
Christoph Lametere498be72005-09-09 13:03:32 -07002251 }
2252 return (ret ? 1 : 0);
2253}
2254
Christoph Lameter945cf2b2012-09-04 23:18:33 +00002255int __kmem_cache_shutdown(struct kmem_cache *cachep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256{
Tejun Heoc9fc5862017-02-22 15:41:27 -08002257 return __kmem_cache_shrink(cachep);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08002258}
2259
2260void __kmem_cache_release(struct kmem_cache *cachep)
2261{
Christoph Lameter12c36672012-09-04 23:38:33 +00002262 int i;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002263 struct kmem_cache_node *n;
Christoph Lameter12c36672012-09-04 23:38:33 +00002264
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002265 cache_random_seq_destroy(cachep);
2266
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07002267 free_percpu(cachep->cpu_cache);
Christoph Lameter12c36672012-09-04 23:38:33 +00002268
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002269 /* NUMA: free the node structures */
Christoph Lameter18bf8542014-08-06 16:04:11 -07002270 for_each_kmem_cache_node(cachep, i, n) {
2271 kfree(n->shared);
2272 free_alien_cache(n->alien);
2273 kfree(n);
2274 cachep->node[i] = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002278/*
2279 * Get the memory for a slab management obj.
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08002280 *
2281 * For a slab cache when the slab descriptor is off-slab, the
2282 * slab descriptor can't come from the same cache which is being created,
2283 * Because if it is the case, that means we defer the creation of
2284 * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point.
2285 * And we eventually call down to __kmem_cache_create(), which
2286 * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one.
2287 * This is a "chicken-and-egg" problem.
2288 *
2289 * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches,
2290 * which are all initialized during kmem_cache_init().
Ravikiran G Thirumalaie5ac9c52006-09-25 23:31:34 -07002291 */
Joonsoo Kim7e007352013-10-30 19:04:01 +09002292static void *alloc_slabmgmt(struct kmem_cache *cachep,
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002293 struct page *page, int colour_off,
2294 gfp_t local_flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002296 void *freelist;
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002297 void *addr = page_address(page);
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002298
Andrey Konovalov51dedad2019-02-20 22:20:28 -08002299 page->s_mem = addr + colour_off;
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002300 page->active = 0;
2301
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002302 if (OBJFREELIST_SLAB(cachep))
2303 freelist = NULL;
2304 else if (OFF_SLAB(cachep)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 /* Slab management obj is off-slab. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002306 freelist = kmem_cache_alloc_node(cachep->freelist_cache,
Pekka Enberg8759ec52008-11-26 10:01:31 +02002307 local_flags, nodeid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 } else {
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002309 /* We will use last bytes at the slab for freelist */
2310 freelist = addr + (PAGE_SIZE << cachep->gfporder) -
2311 cachep->freelist_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 }
Joonsoo Kim2e6b3602016-03-15 14:54:30 -07002313
Joonsoo Kim8456a642013-10-24 10:07:49 +09002314 return freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315}
2316
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002317static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002319 return ((freelist_idx_t *)page->freelist)[idx];
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002320}
2321
2322static inline void set_free_obj(struct page *page,
Joonsoo Kim7cc68972014-04-18 16:24:09 +09002323 unsigned int idx, freelist_idx_t val)
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002324{
Joonsoo Kima41adfa2013-12-02 17:49:42 +09002325 ((freelist_idx_t *)(page->freelist))[idx] = val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326}
2327
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002328static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002330#if DEBUG
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 int i;
2332
2333 for (i = 0; i < cachep->num; i++) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09002334 void *objp = index_to_obj(cachep, page, i);
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002335
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 if (cachep->flags & SLAB_STORE_USER)
2337 *dbg_userword(cachep, objp) = NULL;
2338
2339 if (cachep->flags & SLAB_RED_ZONE) {
2340 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2341 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2342 }
2343 /*
Andrew Mortona737b3e2006-03-22 00:08:11 -08002344 * Constructors are not allowed to allocate memory from the same
2345 * cache which they are a constructor for. Otherwise, deadlock.
2346 * They must also be threaded.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002348 if (cachep->ctor && !(cachep->flags & SLAB_POISON)) {
2349 kasan_unpoison_object_data(cachep,
2350 objp + obj_offset(cachep));
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07002351 cachep->ctor(objp + obj_offset(cachep));
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002352 kasan_poison_object_data(
2353 cachep, objp + obj_offset(cachep));
2354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355
2356 if (cachep->flags & SLAB_RED_ZONE) {
2357 if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002358 slab_error(cachep, "constructor overwrote the end of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
Joe Perches756a0252016-03-17 14:19:47 -07002360 slab_error(cachep, "constructor overwrote the start of an object");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 }
Joonsoo Kim40b44132016-03-15 14:54:21 -07002362 /* need to poison the objs? */
2363 if (cachep->flags & SLAB_POISON) {
2364 poison_obj(cachep, objp, POISON_FREE);
Qian Cai80552f02019-04-16 10:22:57 -04002365 slab_kernel_map(cachep, objp, 0);
Joonsoo Kim40b44132016-03-15 14:54:21 -07002366 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002367 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368#endif
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002369}
2370
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002371#ifdef CONFIG_SLAB_FREELIST_RANDOM
2372/* Hold information during a freelist initialization */
2373union freelist_init_state {
2374 struct {
2375 unsigned int pos;
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002376 unsigned int *list;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002377 unsigned int count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002378 };
2379 struct rnd_state rnd_state;
2380};
2381
2382/*
2383 * Initialize the state based on the randomization methode available.
2384 * return true if the pre-computed list is available, false otherwize.
2385 */
2386static bool freelist_state_initialize(union freelist_init_state *state,
2387 struct kmem_cache *cachep,
2388 unsigned int count)
2389{
2390 bool ret;
2391 unsigned int rand;
2392
2393 /* Use best entropy available to define a random shift */
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002394 rand = get_random_int();
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002395
2396 /* Use a random state if the pre-computed list is not available */
2397 if (!cachep->random_seq) {
2398 prandom_seed_state(&state->rnd_state, rand);
2399 ret = false;
2400 } else {
2401 state->list = cachep->random_seq;
2402 state->count = count;
John Sperbeckc4e490c2017-01-10 16:58:24 -08002403 state->pos = rand % count;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002404 ret = true;
2405 }
2406 return ret;
2407}
2408
2409/* Get the next entry on the list and randomize it using a random shift */
2410static freelist_idx_t next_random_slot(union freelist_init_state *state)
2411{
John Sperbeckc4e490c2017-01-10 16:58:24 -08002412 if (state->pos >= state->count)
2413 state->pos = 0;
2414 return state->list[state->pos++];
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002415}
2416
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002417/* Swap two freelist entries */
2418static void swap_free_obj(struct page *page, unsigned int a, unsigned int b)
2419{
2420 swap(((freelist_idx_t *)page->freelist)[a],
2421 ((freelist_idx_t *)page->freelist)[b]);
2422}
2423
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002424/*
2425 * Shuffle the freelist initialization state based on pre-computed lists.
2426 * return true if the list was successfully shuffled, false otherwise.
2427 */
2428static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page)
2429{
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002430 unsigned int objfreelist = 0, i, rand, count = cachep->num;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002431 union freelist_init_state state;
2432 bool precomputed;
2433
2434 if (count < 2)
2435 return false;
2436
2437 precomputed = freelist_state_initialize(&state, cachep, count);
2438
2439 /* Take a random entry as the objfreelist */
2440 if (OBJFREELIST_SLAB(cachep)) {
2441 if (!precomputed)
2442 objfreelist = count - 1;
2443 else
2444 objfreelist = next_random_slot(&state);
2445 page->freelist = index_to_obj(cachep, page, objfreelist) +
2446 obj_offset(cachep);
2447 count--;
2448 }
2449
2450 /*
2451 * On early boot, generate the list dynamically.
2452 * Later use a pre-computed list for speed.
2453 */
2454 if (!precomputed) {
Thomas Garnier7c00fce2016-07-26 15:21:56 -07002455 for (i = 0; i < count; i++)
2456 set_free_obj(page, i, i);
2457
2458 /* Fisher-Yates shuffle */
2459 for (i = count - 1; i > 0; i--) {
2460 rand = prandom_u32_state(&state.rnd_state);
2461 rand %= (i + 1);
2462 swap_free_obj(page, i, rand);
2463 }
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002464 } else {
2465 for (i = 0; i < count; i++)
2466 set_free_obj(page, i, next_random_slot(&state));
2467 }
2468
2469 if (OBJFREELIST_SLAB(cachep))
2470 set_free_obj(page, cachep->num - 1, objfreelist);
2471
2472 return true;
2473}
2474#else
2475static inline bool shuffle_freelist(struct kmem_cache *cachep,
2476 struct page *page)
2477{
2478 return false;
2479}
2480#endif /* CONFIG_SLAB_FREELIST_RANDOM */
2481
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002482static void cache_init_objs(struct kmem_cache *cachep,
2483 struct page *page)
2484{
2485 int i;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002486 void *objp;
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002487 bool shuffled;
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002488
2489 cache_init_objs_debug(cachep, page);
2490
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002491 /* Try to randomize the freelist if enabled */
2492 shuffled = shuffle_freelist(cachep, page);
2493
2494 if (!shuffled && OBJFREELIST_SLAB(cachep)) {
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002495 page->freelist = index_to_obj(cachep, page, cachep->num - 1) +
2496 obj_offset(cachep);
2497 }
2498
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002499 for (i = 0; i < cachep->num; i++) {
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07002500 objp = index_to_obj(cachep, page, i);
Andrey Konovalov4d176712018-12-28 00:30:23 -08002501 objp = kasan_init_slab_obj(cachep, objp);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07002502
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002503 /* constructor could break poison info */
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002504 if (DEBUG == 0 && cachep->ctor) {
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07002505 kasan_unpoison_object_data(cachep, objp);
2506 cachep->ctor(objp);
2507 kasan_poison_object_data(cachep, objp);
2508 }
Joonsoo Kim10b2e9e2016-03-15 14:54:47 -07002509
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07002510 if (!shuffled)
2511 set_free_obj(page, i, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513}
2514
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002515static void *slab_get_obj(struct kmem_cache *cachep, struct page *page)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002516{
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002517 void *objp;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002518
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002519 objp = index_to_obj(cachep, page, get_free_obj(page, page->active));
Joonsoo Kim8456a642013-10-24 10:07:49 +09002520 page->active++;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002521
2522 return objp;
2523}
2524
Joonsoo Kim260b61d2016-03-15 14:54:12 -07002525static void slab_put_obj(struct kmem_cache *cachep,
2526 struct page *page, void *objp)
Matthew Dobson78d382d2006-02-01 03:05:47 -08002527{
Joonsoo Kim8456a642013-10-24 10:07:49 +09002528 unsigned int objnr = obj_to_index(cachep, page, objp);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002529#if DEBUG
Joonsoo Kim16025172013-10-24 10:07:46 +09002530 unsigned int i;
Matthew Dobson78d382d2006-02-01 03:05:47 -08002531
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002532 /* Verify double free bug */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002533 for (i = page->active; i < cachep->num; i++) {
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002534 if (get_free_obj(page, i) == objnr) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002535 pr_err("slab: double free detected in cache '%s', objp %px\n",
Joe Perches756a0252016-03-17 14:19:47 -07002536 cachep->name, objp);
Joonsoo Kimb1cb0982013-10-24 10:07:45 +09002537 BUG();
2538 }
Matthew Dobson78d382d2006-02-01 03:05:47 -08002539 }
2540#endif
Joonsoo Kim8456a642013-10-24 10:07:49 +09002541 page->active--;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002542 if (!page->freelist)
2543 page->freelist = objp + obj_offset(cachep);
2544
Joonsoo Kime5c58df2013-12-02 17:49:40 +09002545 set_free_obj(page, page->active, objnr);
Matthew Dobson78d382d2006-02-01 03:05:47 -08002546}
2547
Pekka Enberg47768742006-06-23 02:03:07 -07002548/*
2549 * Map pages beginning at addr to the given cache and slab. This is required
2550 * for the slab allocator to be able to lookup the cache and slab of a
Nick Pigginccd35fb2011-01-07 17:49:17 +11002551 * virtual address for kfree, ksize, and slab debugging.
Pekka Enberg47768742006-06-23 02:03:07 -07002552 */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002553static void slab_map_pages(struct kmem_cache *cache, struct page *page,
Joonsoo Kim7e007352013-10-30 19:04:01 +09002554 void *freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555{
Joonsoo Kima57a4982013-10-24 10:07:44 +09002556 page->slab_cache = cache;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002557 page->freelist = freelist;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558}
2559
2560/*
2561 * Grow (by 1) the number of slabs within a cache. This is called by
2562 * kmem_cache_alloc() when there are no active objs left in a cache.
2563 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002564static struct page *cache_grow_begin(struct kmem_cache *cachep,
2565 gfp_t flags, int nodeid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
Joonsoo Kim7e007352013-10-30 19:04:01 +09002567 void *freelist;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002568 size_t offset;
2569 gfp_t local_flags;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002570 int page_node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002571 struct kmem_cache_node *n;
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002572 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573
Andrew Mortona737b3e2006-03-22 00:08:11 -08002574 /*
2575 * Be lazy and only check for valid flags here, keeping it out of the
2576 * critical path in kmem_cache_alloc().
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 */
Long Li44405092020-08-06 23:18:28 -07002578 if (unlikely(flags & GFP_SLAB_BUG_MASK))
2579 flags = kmalloc_fix_flags(flags);
2580
Matthew Wilcox128227e2018-06-07 17:05:13 -07002581 WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
Christoph Lameter6cb06222007-10-16 01:25:41 -07002582 local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 check_irq_off();
Mel Gormand0164ad2015-11-06 16:28:21 -08002585 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 local_irq_enable();
2587
2588 /*
Andrew Mortona737b3e2006-03-22 00:08:11 -08002589 * Get mem for the objs. Attempt to allocate a physical page from
2590 * 'nodeid'.
Christoph Lametere498be72005-09-09 13:03:32 -07002591 */
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002592 page = kmem_getpages(cachep, local_flags, nodeid);
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002593 if (!page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 goto failed;
2595
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002596 page_node = page_to_nid(page);
2597 n = get_node(cachep, page_node);
Joonsoo Kim03d1d432016-05-19 17:10:20 -07002598
2599 /* Get colour for the slab, and cal the next value. */
2600 n->colour_next++;
2601 if (n->colour_next >= cachep->colour)
2602 n->colour_next = 0;
2603
2604 offset = n->colour_next;
2605 if (offset >= cachep->colour)
2606 offset = 0;
2607
2608 offset *= cachep->colour_off;
2609
Andrey Konovalov51dedad2019-02-20 22:20:28 -08002610 /*
2611 * Call kasan_poison_slab() before calling alloc_slabmgmt(), so
2612 * page_address() in the latter returns a non-tagged pointer,
2613 * as it should be for slab pages.
2614 */
2615 kasan_poison_slab(page);
2616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 /* Get slab management. */
Joonsoo Kim8456a642013-10-24 10:07:49 +09002618 freelist = alloc_slabmgmt(cachep, page, offset,
Joonsoo Kim511e3a052016-05-19 17:10:23 -07002619 local_flags & ~GFP_CONSTRAINT_MASK, page_node);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002620 if (OFF_SLAB(cachep) && !freelist)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 goto opps1;
2622
Joonsoo Kim8456a642013-10-24 10:07:49 +09002623 slab_map_pages(cachep, page, freelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Joonsoo Kim8456a642013-10-24 10:07:49 +09002625 cache_init_objs(cachep, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
Mel Gormand0164ad2015-11-06 16:28:21 -08002627 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 local_irq_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002630 return page;
2631
Andrew Mortona737b3e2006-03-22 00:08:11 -08002632opps1:
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09002633 kmem_freepages(cachep, page);
Andrew Mortona737b3e2006-03-22 00:08:11 -08002634failed:
Mel Gormand0164ad2015-11-06 16:28:21 -08002635 if (gfpflags_allow_blocking(local_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 local_irq_disable();
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002637 return NULL;
2638}
2639
2640static void cache_grow_end(struct kmem_cache *cachep, struct page *page)
2641{
2642 struct kmem_cache_node *n;
2643 void *list = NULL;
2644
2645 check_irq_off();
2646
2647 if (!page)
2648 return;
2649
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002650 INIT_LIST_HEAD(&page->slab_list);
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002651 n = get_node(cachep, page_to_nid(page));
2652
2653 spin_lock(&n->list_lock);
David Rientjesbf00bd32016-12-12 16:41:44 -08002654 n->total_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002655 if (!page->active) {
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002656 list_add_tail(&page->slab_list, &n->slabs_free);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002657 n->free_slabs++;
David Rientjesbf00bd32016-12-12 16:41:44 -08002658 } else
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002659 fixup_slab_list(cachep, n, page, &list);
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07002660
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002661 STATS_INC_GROWN(cachep);
2662 n->free_objects += cachep->num - page->active;
2663 spin_unlock(&n->list_lock);
2664
2665 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666}
2667
2668#if DEBUG
2669
2670/*
2671 * Perform extra freeing checks:
2672 * - detect bad pointers.
2673 * - POISON/RED_ZONE checking
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 */
2675static void kfree_debugcheck(const void *objp)
2676{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 if (!virt_addr_valid(objp)) {
Joe Perches11705322016-03-17 14:19:50 -07002678 pr_err("kfree_debugcheck: out of range ptr %lxh\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002679 (unsigned long)objp);
2680 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682}
2683
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002684static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
2685{
David Woodhouseb46b8f12007-05-08 00:22:59 -07002686 unsigned long long redzone1, redzone2;
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002687
2688 redzone1 = *dbg_redzone1(cache, obj);
2689 redzone2 = *dbg_redzone2(cache, obj);
2690
2691 /*
2692 * Redzone is ok.
2693 */
2694 if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
2695 return;
2696
2697 if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
2698 slab_error(cache, "double free detected");
2699 else
2700 slab_error(cache, "memory outside object was overwritten");
2701
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08002702 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07002703 obj, redzone1, redzone2);
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002704}
2705
Pekka Enberg343e0d72006-02-01 03:05:50 -08002706static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002707 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 unsigned int objnr;
Joonsoo Kim8456a642013-10-24 10:07:49 +09002710 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
Matthew Wilcox80cbd912007-11-29 12:05:13 -07002712 BUG_ON(virt_to_cache(objp) != cachep);
2713
Manfred Spraul3dafccf2006-02-01 03:05:42 -08002714 objp -= obj_offset(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 kfree_debugcheck(objp);
Christoph Lameterb49af682007-05-06 14:49:41 -07002716 page = virt_to_head_page(objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 if (cachep->flags & SLAB_RED_ZONE) {
Pekka Enberg58ce1fd2006-06-23 02:03:24 -07002719 verify_redzone_free(cachep, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 *dbg_redzone1(cachep, objp) = RED_INACTIVE;
2721 *dbg_redzone2(cachep, objp) = RED_INACTIVE;
2722 }
Qian Cai7878c232019-05-16 15:57:41 -04002723 if (cachep->flags & SLAB_STORE_USER)
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002724 *dbg_userword(cachep, objp) = (void *)caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725
Joonsoo Kim8456a642013-10-24 10:07:49 +09002726 objnr = obj_to_index(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727
2728 BUG_ON(objnr >= cachep->num);
Joonsoo Kim8456a642013-10-24 10:07:49 +09002729 BUG_ON(objp != index_to_obj(cachep, page, objnr));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 poison_obj(cachep, objp, POISON_FREE);
Qian Cai80552f02019-04-16 10:22:57 -04002733 slab_kernel_map(cachep, objp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734 }
2735 return objp;
2736}
2737
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738#else
2739#define kfree_debugcheck(x) do { } while(0)
2740#define cache_free_debugcheck(x,objp,z) (objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741#endif
2742
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002743static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
2744 void **list)
2745{
2746#if DEBUG
2747 void *next = *list;
2748 void *objp;
2749
2750 while (next) {
2751 objp = next - obj_offset(cachep);
2752 next = *(void **)next;
2753 poison_obj(cachep, objp, POISON_FREE);
2754 }
2755#endif
2756}
2757
Joonsoo Kimd8410232016-03-15 14:54:44 -07002758static inline void fixup_slab_list(struct kmem_cache *cachep,
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002759 struct kmem_cache_node *n, struct page *page,
2760 void **list)
Joonsoo Kimd8410232016-03-15 14:54:44 -07002761{
2762 /* move slabp to correct slabp list: */
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002763 list_del(&page->slab_list);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002764 if (page->active == cachep->num) {
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002765 list_add(&page->slab_list, &n->slabs_full);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002766 if (OBJFREELIST_SLAB(cachep)) {
2767#if DEBUG
2768 /* Poisoning will be done without holding the lock */
2769 if (cachep->flags & SLAB_POISON) {
2770 void **objp = page->freelist;
2771
2772 *objp = *list;
2773 *list = objp;
2774 }
2775#endif
2776 page->freelist = NULL;
2777 }
2778 } else
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002779 list_add(&page->slab_list, &n->slabs_partial);
Joonsoo Kimd8410232016-03-15 14:54:44 -07002780}
2781
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002782/* Try to find non-pfmemalloc slab if needed */
2783static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n,
David Rientjesbf00bd32016-12-12 16:41:44 -08002784 struct page *page, bool pfmemalloc)
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002785{
2786 if (!page)
2787 return NULL;
2788
2789 if (pfmemalloc)
2790 return page;
2791
2792 if (!PageSlabPfmemalloc(page))
2793 return page;
2794
2795 /* No need to keep pfmemalloc slab if we have enough free objects */
2796 if (n->free_objects > n->free_limit) {
2797 ClearPageSlabPfmemalloc(page);
2798 return page;
2799 }
2800
2801 /* Move pfmemalloc slab to the end of list to speed up next search */
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002802 list_del(&page->slab_list);
David Rientjesbf00bd32016-12-12 16:41:44 -08002803 if (!page->active) {
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002804 list_add_tail(&page->slab_list, &n->slabs_free);
David Rientjesbf00bd32016-12-12 16:41:44 -08002805 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002806 } else
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002807 list_add_tail(&page->slab_list, &n->slabs_partial);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002808
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002809 list_for_each_entry(page, &n->slabs_partial, slab_list) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002810 if (!PageSlabPfmemalloc(page))
2811 return page;
2812 }
2813
Greg Thelenf728b0a2016-12-12 16:41:41 -08002814 n->free_touched = 1;
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002815 list_for_each_entry(page, &n->slabs_free, slab_list) {
Greg Thelenf728b0a2016-12-12 16:41:41 -08002816 if (!PageSlabPfmemalloc(page)) {
David Rientjesbf00bd32016-12-12 16:41:44 -08002817 n->free_slabs--;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002818 return page;
Greg Thelenf728b0a2016-12-12 16:41:41 -08002819 }
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002820 }
2821
2822 return NULL;
2823}
2824
2825static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc)
Geliang Tang7aa0d222016-01-14 15:18:02 -08002826{
2827 struct page *page;
2828
Greg Thelenf728b0a2016-12-12 16:41:41 -08002829 assert_spin_locked(&n->list_lock);
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002830 page = list_first_entry_or_null(&n->slabs_partial, struct page,
2831 slab_list);
Geliang Tang7aa0d222016-01-14 15:18:02 -08002832 if (!page) {
2833 n->free_touched = 1;
David Rientjesbf00bd32016-12-12 16:41:44 -08002834 page = list_first_entry_or_null(&n->slabs_free, struct page,
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07002835 slab_list);
Greg Thelenf728b0a2016-12-12 16:41:41 -08002836 if (page)
David Rientjesbf00bd32016-12-12 16:41:44 -08002837 n->free_slabs--;
Geliang Tang7aa0d222016-01-14 15:18:02 -08002838 }
2839
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002840 if (sk_memalloc_socks())
David Rientjesbf00bd32016-12-12 16:41:44 -08002841 page = get_valid_first_slab(n, page, pfmemalloc);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002842
Geliang Tang7aa0d222016-01-14 15:18:02 -08002843 return page;
2844}
2845
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002846static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep,
2847 struct kmem_cache_node *n, gfp_t flags)
2848{
2849 struct page *page;
2850 void *obj;
2851 void *list = NULL;
2852
2853 if (!gfp_pfmemalloc_allowed(flags))
2854 return NULL;
2855
2856 spin_lock(&n->list_lock);
2857 page = get_first_slab(n, true);
2858 if (!page) {
2859 spin_unlock(&n->list_lock);
2860 return NULL;
2861 }
2862
2863 obj = slab_get_obj(cachep, page);
2864 n->free_objects--;
2865
2866 fixup_slab_list(cachep, n, page, &list);
2867
2868 spin_unlock(&n->list_lock);
2869 fixup_objfreelist_debug(cachep, &list);
2870
2871 return obj;
2872}
2873
Joonsoo Kim213b4692016-05-19 17:10:29 -07002874/*
2875 * Slab list should be fixed up by fixup_slab_list() for existing slab
2876 * or cache_grow_end() for new slab
2877 */
2878static __always_inline int alloc_block(struct kmem_cache *cachep,
2879 struct array_cache *ac, struct page *page, int batchcount)
2880{
2881 /*
2882 * There must be at least one object available for
2883 * allocation.
2884 */
2885 BUG_ON(page->active >= cachep->num);
2886
2887 while (page->active < cachep->num && batchcount--) {
2888 STATS_INC_ALLOCED(cachep);
2889 STATS_INC_ACTIVE(cachep);
2890 STATS_SET_HIGH(cachep);
2891
2892 ac->entry[ac->avail++] = slab_get_obj(cachep, page);
2893 }
2894
2895 return batchcount;
2896}
2897
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002898static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899{
2900 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002901 struct kmem_cache_node *n;
Joonsoo Kim801faf02016-05-19 17:10:31 -07002902 struct array_cache *ac, *shared;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002903 int node;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002904 void *list = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002905 struct page *page;
Pekka Enberg1ca4cb22006-10-06 00:43:52 -07002906
Joe Korty6d2144d2008-03-05 15:04:59 -08002907 check_irq_off();
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07002908 node = numa_mem_id();
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002909
Joe Korty6d2144d2008-03-05 15:04:59 -08002910 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 batchcount = ac->batchcount;
2912 if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08002913 /*
2914 * If there was little recent activity on this cache, then
2915 * perform only a partial refill. Otherwise we could generate
2916 * refill bouncing.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 */
2918 batchcount = BATCHREFILL_LIMIT;
2919 }
Christoph Lameter18bf8542014-08-06 16:04:11 -07002920 n = get_node(cachep, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002922 BUG_ON(ac->avail > 0 || !n);
Joonsoo Kim801faf02016-05-19 17:10:31 -07002923 shared = READ_ONCE(n->shared);
2924 if (!n->free_objects && (!shared || !shared->avail))
2925 goto direct_grow;
2926
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002927 spin_lock(&n->list_lock);
Joonsoo Kim801faf02016-05-19 17:10:31 -07002928 shared = READ_ONCE(n->shared);
Christoph Lametere498be72005-09-09 13:03:32 -07002929
Christoph Lameter3ded1752006-03-25 03:06:44 -08002930 /* See if we can refill from the shared array */
Joonsoo Kim801faf02016-05-19 17:10:31 -07002931 if (shared && transfer_objects(ac, shared, batchcount)) {
2932 shared->touched = 1;
Christoph Lameter3ded1752006-03-25 03:06:44 -08002933 goto alloc_done;
Nick Piggin44b57f12010-01-27 22:27:40 +11002934 }
Christoph Lameter3ded1752006-03-25 03:06:44 -08002935
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936 while (batchcount > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 /* Get slab alloc is to come from. */
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002938 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08002939 if (!page)
2940 goto must_grow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942 check_spinlock_acquired(cachep);
Pekka Enberg714b81712007-05-06 14:49:03 -07002943
Joonsoo Kim213b4692016-05-19 17:10:29 -07002944 batchcount = alloc_block(cachep, ac, page, batchcount);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002945 fixup_slab_list(cachep, n, page, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 }
2947
Andrew Mortona737b3e2006-03-22 00:08:11 -08002948must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002949 n->free_objects -= ac->avail;
Andrew Mortona737b3e2006-03-22 00:08:11 -08002950alloc_done:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00002951 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07002952 fixup_objfreelist_debug(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
Joonsoo Kim801faf02016-05-19 17:10:31 -07002954direct_grow:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 if (unlikely(!ac->avail)) {
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002956 /* Check if we can use obj in pfmemalloc slab */
2957 if (sk_memalloc_socks()) {
2958 void *obj = cache_alloc_pfmemalloc(cachep, n, flags);
2959
2960 if (obj)
2961 return obj;
2962 }
2963
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002964 page = cache_grow_begin(cachep, gfp_exact_node(flags), node);
Christoph Lametere498be72005-09-09 13:03:32 -07002965
Joonsoo Kim76b342b2016-05-19 17:10:26 -07002966 /*
2967 * cache_grow_begin() can reenable interrupts,
2968 * then ac could change.
2969 */
Pekka Enberg9a2dba42006-02-01 03:05:49 -08002970 ac = cpu_cache_get(cachep);
Joonsoo Kim213b4692016-05-19 17:10:29 -07002971 if (!ac->avail && page)
2972 alloc_block(cachep, ac, page, batchcount);
2973 cache_grow_end(cachep, page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07002974
Joonsoo Kim213b4692016-05-19 17:10:29 -07002975 if (!ac->avail)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 }
2978 ac->touched = 1;
Mel Gorman072bb0a2012-07-31 16:43:58 -07002979
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07002980 return ac->entry[--ac->avail];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981}
2982
Andrew Mortona737b3e2006-03-22 00:08:11 -08002983static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
2984 gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985{
Mel Gormand0164ad2015-11-06 16:28:21 -08002986 might_sleep_if(gfpflags_allow_blocking(flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987}
2988
2989#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08002990static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03002991 gfp_t flags, void *objp, unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992{
Matthew Wilcox128227e2018-06-07 17:05:13 -07002993 WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002994 if (!objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 return objp;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08002996 if (cachep->flags & SLAB_POISON) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 check_poison_obj(cachep, objp);
Qian Cai80552f02019-04-16 10:22:57 -04002998 slab_kernel_map(cachep, objp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999 poison_obj(cachep, objp, POISON_INUSE);
3000 }
3001 if (cachep->flags & SLAB_STORE_USER)
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003002 *dbg_userword(cachep, objp) = (void *)caller;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003
3004 if (cachep->flags & SLAB_RED_ZONE) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08003005 if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
3006 *dbg_redzone2(cachep, objp) != RED_INACTIVE) {
Joe Perches756a0252016-03-17 14:19:47 -07003007 slab_error(cachep, "double free, or memory outside object was overwritten");
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003008 pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
Joe Perches11705322016-03-17 14:19:50 -07003009 objp, *dbg_redzone1(cachep, objp),
3010 *dbg_redzone2(cachep, objp));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 }
3012 *dbg_redzone1(cachep, objp) = RED_ACTIVE;
3013 *dbg_redzone2(cachep, objp) = RED_ACTIVE;
3014 }
Joonsoo Kim03787302014-06-23 13:22:06 -07003015
Manfred Spraul3dafccf2006-02-01 03:05:42 -08003016 objp += obj_offset(cachep);
Christoph Lameter4f104932007-05-06 14:50:17 -07003017 if (cachep->ctor && cachep->flags & SLAB_POISON)
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07003018 cachep->ctor(objp);
Tetsuo Handa7ea466f2011-07-21 09:42:45 +09003019 if (ARCH_SLAB_MINALIGN &&
3020 ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
Geert Uytterhoeven85c3e4a2017-12-14 15:32:58 -08003021 pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
Hugh Dickinsc2251502011-07-11 13:35:08 -07003022 objp, (int)ARCH_SLAB_MINALIGN);
Kevin Hilmana44b56d2006-12-06 20:32:11 -08003023 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 return objp;
3025}
3026#else
3027#define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
3028#endif
3029
Pekka Enberg343e0d72006-02-01 03:05:50 -08003030static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003031{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003032 void *objp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 struct array_cache *ac;
3034
Alok N Kataria5c382302005-09-27 21:45:46 -07003035 check_irq_off();
Akinobu Mita8a8b6502006-12-08 02:39:44 -08003036
Pekka Enberg9a2dba42006-02-01 03:05:49 -08003037 ac = cpu_cache_get(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038 if (likely(ac->avail)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003039 ac->touched = 1;
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003040 objp = ac->entry[--ac->avail];
Mel Gorman072bb0a2012-07-31 16:43:58 -07003041
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003042 STATS_INC_ALLOCHIT(cachep);
3043 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044 }
Mel Gorman072bb0a2012-07-31 16:43:58 -07003045
3046 STATS_INC_ALLOCMISS(cachep);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003047 objp = cache_alloc_refill(cachep, flags);
Mel Gorman072bb0a2012-07-31 16:43:58 -07003048 /*
3049 * the 'ac' may be updated by cache_alloc_refill(),
3050 * and kmemleak_erase() requires its correct value.
3051 */
3052 ac = cpu_cache_get(cachep);
3053
3054out:
Catalin Marinasd5cff632009-06-11 13:22:40 +01003055 /*
3056 * To avoid a false negative, if an object that is in one of the
3057 * per-CPU caches is leaked, we need to make sure kmemleak doesn't
3058 * treat the array pointers as a reference to the object.
3059 */
J. R. Okajimaf3d8b532009-12-02 16:55:49 +09003060 if (objp)
3061 kmemleak_erase(&ac->entry[ac->avail]);
Alok N Kataria5c382302005-09-27 21:45:46 -07003062 return objp;
3063}
3064
Christoph Lametere498be72005-09-09 13:03:32 -07003065#ifdef CONFIG_NUMA
3066/*
Zefan Li2ad654b2014-09-25 09:41:02 +08003067 * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set.
Paul Jacksonc61afb12006-03-24 03:16:08 -08003068 *
3069 * If we are in_interrupt, then process context, including cpusets and
3070 * mempolicy, may not apply and should not be used for allocation policy.
3071 */
3072static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
3073{
3074 int nid_alloc, nid_here;
3075
Christoph Lameter765c4502006-09-27 01:50:08 -07003076 if (in_interrupt() || (flags & __GFP_THISNODE))
Paul Jacksonc61afb12006-03-24 03:16:08 -08003077 return NULL;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003078 nid_alloc = nid_here = numa_mem_id();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003079 if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
Jack Steiner6adef3e2010-05-26 14:42:49 -07003080 nid_alloc = cpuset_slab_spread_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003081 else if (current->mempolicy)
David Rientjes2a389612014-04-07 15:37:29 -07003082 nid_alloc = mempolicy_slab_node();
Paul Jacksonc61afb12006-03-24 03:16:08 -08003083 if (nid_alloc != nid_here)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003084 return ____cache_alloc_node(cachep, flags, nid_alloc);
Paul Jacksonc61afb12006-03-24 03:16:08 -08003085 return NULL;
3086}
3087
3088/*
Christoph Lameter765c4502006-09-27 01:50:08 -07003089 * Fallback function if there was no memory available and no objects on a
Christoph Lameter3c517a62006-12-06 20:33:29 -08003090 * certain node and fall back is permitted. First we scan all the
Christoph Lameter6a673682013-01-10 19:14:19 +00003091 * available node for available objects. If that fails then we
Christoph Lameter3c517a62006-12-06 20:33:29 -08003092 * perform an allocation without specifying a node. This allows the page
3093 * allocator to do its reclaim / fallback magic. We then insert the
3094 * slab into the proper nodelist and then allocate from it.
Christoph Lameter765c4502006-09-27 01:50:08 -07003095 */
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003096static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
Christoph Lameter765c4502006-09-27 01:50:08 -07003097{
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003098 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07003099 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07003100 struct zone *zone;
Joonsoo Kim97a225e2020-06-03 15:59:01 -07003101 enum zone_type highest_zoneidx = gfp_zone(flags);
Christoph Lameter765c4502006-09-27 01:50:08 -07003102 void *obj = NULL;
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003103 struct page *page;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003104 int nid;
Mel Gormancc9a6c82012-03-21 16:34:11 -07003105 unsigned int cpuset_mems_cookie;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003106
3107 if (flags & __GFP_THISNODE)
3108 return NULL;
3109
Mel Gormancc9a6c82012-03-21 16:34:11 -07003110retry_cpuset:
Mel Gormand26914d2014-04-03 14:47:24 -07003111 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07003112 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07003113
Christoph Lameter3c517a62006-12-06 20:33:29 -08003114retry:
3115 /*
3116 * Look through allowed nodes for objects available
3117 * from existing per node queues.
3118 */
Joonsoo Kim97a225e2020-06-03 15:59:01 -07003119 for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) {
Mel Gorman54a6eb52008-04-28 02:12:16 -07003120 nid = zone_to_nid(zone);
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003121
Vladimir Davydov061d70742014-12-12 16:58:25 -08003122 if (cpuset_zone_allowed(zone, flags) &&
Christoph Lameter18bf8542014-08-06 16:04:11 -07003123 get_node(cache, nid) &&
3124 get_node(cache, nid)->free_objects) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003125 obj = ____cache_alloc_node(cache,
David Rientjes4167e9b2015-04-14 15:46:55 -07003126 gfp_exact_node(flags), nid);
Christoph Lameter481c5342008-06-21 16:46:35 -07003127 if (obj)
3128 break;
3129 }
Christoph Lameter3c517a62006-12-06 20:33:29 -08003130 }
3131
Christoph Lametercfce6602007-05-06 14:50:17 -07003132 if (!obj) {
Christoph Lameter3c517a62006-12-06 20:33:29 -08003133 /*
3134 * This allocation will be performed within the constraints
3135 * of the current cpuset / memory policy requirements.
3136 * We may trigger various forms of reclaim on the allowed
3137 * set and go into memory reserves if necessary.
3138 */
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003139 page = cache_grow_begin(cache, flags, numa_mem_id());
3140 cache_grow_end(cache, page);
3141 if (page) {
3142 nid = page_to_nid(page);
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003143 obj = ____cache_alloc_node(cache,
3144 gfp_exact_node(flags), nid);
Joonsoo Kim0c3aa832013-10-24 10:07:38 +09003145
Christoph Lameter3c517a62006-12-06 20:33:29 -08003146 /*
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003147 * Another processor may allocate the objects in
3148 * the slab since we are not holding any locks.
Christoph Lameter3c517a62006-12-06 20:33:29 -08003149 */
Joonsoo Kim511e3a052016-05-19 17:10:23 -07003150 if (!obj)
3151 goto retry;
Christoph Lameter3c517a62006-12-06 20:33:29 -08003152 }
Christoph Lameteraedb0eb2006-10-21 10:24:16 -07003153 }
Mel Gormancc9a6c82012-03-21 16:34:11 -07003154
Mel Gormand26914d2014-04-03 14:47:24 -07003155 if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie)))
Mel Gormancc9a6c82012-03-21 16:34:11 -07003156 goto retry_cpuset;
Christoph Lameter765c4502006-09-27 01:50:08 -07003157 return obj;
3158}
3159
3160/*
Christoph Lametere498be72005-09-09 13:03:32 -07003161 * A interface to enable slab creation on nodeid
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003163static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
Andrew Mortona737b3e2006-03-22 00:08:11 -08003164 int nodeid)
Christoph Lametere498be72005-09-09 13:03:32 -07003165{
Joonsoo Kim8456a642013-10-24 10:07:49 +09003166 struct page *page;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003167 struct kmem_cache_node *n;
Joonsoo Kim213b4692016-05-19 17:10:29 -07003168 void *obj = NULL;
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003169 void *list = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170
Paul Mackerras7c3fbbd2014-12-02 15:59:48 -08003171 VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003172 n = get_node(cachep, nodeid);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003173 BUG_ON(!n);
Christoph Lametere498be72005-09-09 13:03:32 -07003174
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08003175 check_irq_off();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003176 spin_lock(&n->list_lock);
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003177 page = get_first_slab(n, false);
Geliang Tang7aa0d222016-01-14 15:18:02 -08003178 if (!page)
3179 goto must_grow;
Christoph Lametere498be72005-09-09 13:03:32 -07003180
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003181 check_spinlock_acquired_node(cachep, nodeid);
Christoph Lametere498be72005-09-09 13:03:32 -07003182
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003183 STATS_INC_NODEALLOCS(cachep);
3184 STATS_INC_ACTIVE(cachep);
3185 STATS_SET_HIGH(cachep);
Christoph Lametere498be72005-09-09 13:03:32 -07003186
Joonsoo Kim8456a642013-10-24 10:07:49 +09003187 BUG_ON(page->active == cachep->num);
Christoph Lametere498be72005-09-09 13:03:32 -07003188
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003189 obj = slab_get_obj(cachep, page);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003190 n->free_objects--;
Christoph Lametere498be72005-09-09 13:03:32 -07003191
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003192 fixup_slab_list(cachep, n, page, &list);
Christoph Lametere498be72005-09-09 13:03:32 -07003193
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003194 spin_unlock(&n->list_lock);
Joonsoo Kimb03a017b2016-03-15 14:54:50 -07003195 fixup_objfreelist_debug(cachep, &list);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003196 return obj;
Christoph Lametere498be72005-09-09 13:03:32 -07003197
Andrew Mortona737b3e2006-03-22 00:08:11 -08003198must_grow:
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003199 spin_unlock(&n->list_lock);
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003200 page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid);
Joonsoo Kim213b4692016-05-19 17:10:29 -07003201 if (page) {
3202 /* This slab isn't counted yet so don't update free_objects */
3203 obj = slab_get_obj(cachep, page);
3204 }
Joonsoo Kim76b342b2016-05-19 17:10:26 -07003205 cache_grow_end(cachep, page);
Christoph Lametere498be72005-09-09 13:03:32 -07003206
Joonsoo Kim213b4692016-05-19 17:10:29 -07003207 return obj ? obj : fallback_alloc(cachep, flags);
Christoph Lametere498be72005-09-09 13:03:32 -07003208}
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003209
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003210static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003211slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003212 unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003213{
3214 unsigned long save_flags;
3215 void *ptr;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003216 int slab_node = numa_mem_id();
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003217 struct obj_cgroup *objcg = NULL;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003218
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003219 flags &= gfp_allowed_mask;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003220 cachep = slab_pre_alloc_hook(cachep, &objcg, 1, flags);
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003221 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003222 return NULL;
3223
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003224 cache_alloc_debugcheck_before(cachep, flags);
3225 local_irq_save(save_flags);
3226
Andrew Mortoneacbbae2011-07-28 13:59:49 -07003227 if (nodeid == NUMA_NO_NODE)
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003228 nodeid = slab_node;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003229
Christoph Lameter18bf8542014-08-06 16:04:11 -07003230 if (unlikely(!get_node(cachep, nodeid))) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003231 /* Node not bootstrapped yet */
3232 ptr = fallback_alloc(cachep, flags);
3233 goto out;
3234 }
3235
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003236 if (nodeid == slab_node) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003237 /*
3238 * Use the locally cached objects if possible.
3239 * However ____cache_alloc does not allow fallback
3240 * to other nodes. It may fail while we still have
3241 * objects on other nodes available.
3242 */
3243 ptr = ____cache_alloc(cachep, flags);
3244 if (ptr)
3245 goto out;
3246 }
3247 /* ___cache_alloc_node can fall back to other nodes */
3248 ptr = ____cache_alloc_node(cachep, flags, nodeid);
3249 out:
3250 local_irq_restore(save_flags);
3251 ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
3252
Alexander Potapenko64713842019-07-11 20:59:19 -07003253 if (unlikely(slab_want_init_on_alloc(flags, cachep)) && ptr)
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003254 memset(ptr, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003255
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003256 slab_post_alloc_hook(cachep, objcg, flags, 1, &ptr);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003257 return ptr;
3258}
3259
3260static __always_inline void *
3261__do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
3262{
3263 void *objp;
3264
Zefan Li2ad654b2014-09-25 09:41:02 +08003265 if (current->mempolicy || cpuset_do_slab_mem_spread()) {
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003266 objp = alternate_node_alloc(cache, flags);
3267 if (objp)
3268 goto out;
3269 }
3270 objp = ____cache_alloc(cache, flags);
3271
3272 /*
3273 * We may just have run out of memory on the local node.
3274 * ____cache_alloc_node() knows how to locate memory on other nodes
3275 */
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003276 if (!objp)
3277 objp = ____cache_alloc_node(cache, flags, numa_mem_id());
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003278
3279 out:
3280 return objp;
3281}
3282#else
3283
3284static __always_inline void *
3285__do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
3286{
3287 return ____cache_alloc(cachep, flags);
3288}
3289
3290#endif /* CONFIG_NUMA */
3291
3292static __always_inline void *
Ezequiel Garcia48356302012-09-08 17:47:57 -03003293slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller)
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003294{
3295 unsigned long save_flags;
3296 void *objp;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003297 struct obj_cgroup *objcg = NULL;
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003298
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +10003299 flags &= gfp_allowed_mask;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003300 cachep = slab_pre_alloc_hook(cachep, &objcg, 1, flags);
Jesper Dangaard Brouer011ecea2016-03-15 14:53:41 -07003301 if (unlikely(!cachep))
Akinobu Mita824ebef2007-05-06 14:49:58 -07003302 return NULL;
3303
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003304 cache_alloc_debugcheck_before(cachep, flags);
3305 local_irq_save(save_flags);
3306 objp = __do_cache_alloc(cachep, flags);
3307 local_irq_restore(save_flags);
3308 objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
3309 prefetchw(objp);
3310
Alexander Potapenko64713842019-07-11 20:59:19 -07003311 if (unlikely(slab_want_init_on_alloc(flags, cachep)) && objp)
Jesper Dangaard Brouerd5e3ed62016-03-15 14:53:47 -07003312 memset(objp, 0, cachep->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07003313
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003314 slab_post_alloc_hook(cachep, objcg, flags, 1, &objp);
Pekka Enberg8c8cc2c2007-02-10 01:42:53 -08003315 return objp;
3316}
Christoph Lametere498be72005-09-09 13:03:32 -07003317
3318/*
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08003319 * Caller needs to acquire correct kmem_cache_node's list_lock
Joonsoo Kim97654df2014-08-06 16:04:25 -07003320 * @list: List of detached free slabs should be freed by caller
Christoph Lametere498be72005-09-09 13:03:32 -07003321 */
Joonsoo Kim97654df2014-08-06 16:04:25 -07003322static void free_block(struct kmem_cache *cachep, void **objpp,
3323 int nr_objects, int node, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324{
3325 int i;
Joonsoo Kim25c063f2014-08-06 16:04:22 -07003326 struct kmem_cache_node *n = get_node(cachep, node);
Joonsoo Kim6052b782016-05-19 17:10:17 -07003327 struct page *page;
3328
3329 n->free_objects += nr_objects;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330
3331 for (i = 0; i < nr_objects; i++) {
Mel Gorman072bb0a2012-07-31 16:43:58 -07003332 void *objp;
Joonsoo Kim8456a642013-10-24 10:07:49 +09003333 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003334
Mel Gorman072bb0a2012-07-31 16:43:58 -07003335 objp = objpp[i];
3336
Joonsoo Kim8456a642013-10-24 10:07:49 +09003337 page = virt_to_head_page(objp);
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07003338 list_del(&page->slab_list);
Christoph Lameterff694162005-09-22 21:44:02 -07003339 check_spinlock_acquired_node(cachep, node);
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003340 slab_put_obj(cachep, page, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003341 STATS_DEC_ACTIVE(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003342
3343 /* fixup slab chains */
Greg Thelenf728b0a2016-12-12 16:41:41 -08003344 if (page->active == 0) {
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07003345 list_add(&page->slab_list, &n->slabs_free);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003346 n->free_slabs++;
Greg Thelenf728b0a2016-12-12 16:41:41 -08003347 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 /* Unconditionally move a slab to the end of the
3349 * partial list on free - maximum time for the
3350 * other objects to be freed, too.
3351 */
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07003352 list_add_tail(&page->slab_list, &n->slabs_partial);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 }
3354 }
Joonsoo Kim6052b782016-05-19 17:10:17 -07003355
3356 while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) {
3357 n->free_objects -= cachep->num;
3358
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07003359 page = list_last_entry(&n->slabs_free, struct page, slab_list);
3360 list_move(&page->slab_list, list);
Greg Thelenf728b0a2016-12-12 16:41:41 -08003361 n->free_slabs--;
David Rientjesbf00bd32016-12-12 16:41:44 -08003362 n->total_slabs--;
Joonsoo Kim6052b782016-05-19 17:10:17 -07003363 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364}
3365
Pekka Enberg343e0d72006-02-01 03:05:50 -08003366static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367{
3368 int batchcount;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003369 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003370 int node = numa_mem_id();
Joonsoo Kim97654df2014-08-06 16:04:25 -07003371 LIST_HEAD(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372
3373 batchcount = ac->batchcount;
Joonsoo Kim260b61d2016-03-15 14:54:12 -07003374
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 check_irq_off();
Christoph Lameter18bf8542014-08-06 16:04:11 -07003376 n = get_node(cachep, node);
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003377 spin_lock(&n->list_lock);
3378 if (n->shared) {
3379 struct array_cache *shared_array = n->shared;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003380 int max = shared_array->limit - shared_array->avail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381 if (max) {
3382 if (batchcount > max)
3383 batchcount = max;
Christoph Lametere498be72005-09-09 13:03:32 -07003384 memcpy(&(shared_array->entry[shared_array->avail]),
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003385 ac->entry, sizeof(void *) * batchcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 shared_array->avail += batchcount;
3387 goto free_done;
3388 }
3389 }
3390
Joonsoo Kim97654df2014-08-06 16:04:25 -07003391 free_block(cachep, ac->entry, batchcount, node, &list);
Andrew Mortona737b3e2006-03-22 00:08:11 -08003392free_done:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393#if STATS
3394 {
3395 int i = 0;
Geliang Tang73c02192016-01-14 15:17:59 -08003396 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003397
Tobin C. Harding16cb0ec72019-05-13 17:16:15 -07003398 list_for_each_entry(page, &n->slabs_free, slab_list) {
Joonsoo Kim8456a642013-10-24 10:07:49 +09003399 BUG_ON(page->active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400
3401 i++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 }
3403 STATS_SET_FREEABLE(cachep, i);
3404 }
3405#endif
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003406 spin_unlock(&n->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407 ac->avail -= batchcount;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003408 memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
Shakeel Butt678ff6a2020-09-26 07:13:41 -07003409 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003410}
3411
3412/*
Andrew Mortona737b3e2006-03-22 00:08:11 -08003413 * Release an obj back to its cache. If the obj has a constructed state, it must
3414 * be in this state _before_ it is released. Called with disabled ints.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003416static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp,
3417 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418{
Alexander Potapenko55834c52016-05-20 16:59:11 -07003419 /* Put the object into the quarantine, don't touch it for now. */
Dmitry Vyukovee3ce772018-02-06 15:36:27 -08003420 if (kasan_slab_free(cachep, objp, _RET_IP_))
Alexander Potapenko55834c52016-05-20 16:59:11 -07003421 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422
Marco Elvercfbe1632020-08-06 23:19:12 -07003423 /* Use KCSAN to help debug racy use-after-free. */
3424 if (!(cachep->flags & SLAB_TYPESAFE_BY_RCU))
3425 __kcsan_check_access(objp, cachep->object_size,
3426 KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
3427
Alexander Potapenko55834c52016-05-20 16:59:11 -07003428 ___cache_free(cachep, objp, caller);
3429}
3430
3431void ___cache_free(struct kmem_cache *cachep, void *objp,
3432 unsigned long caller)
3433{
3434 struct array_cache *ac = cpu_cache_get(cachep);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003435
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436 check_irq_off();
Alexander Potapenko64713842019-07-11 20:59:19 -07003437 if (unlikely(slab_want_init_on_free(cachep)))
3438 memset(objp, 0, cachep->object_size);
Catalin Marinasd5cff632009-06-11 13:22:40 +01003439 kmemleak_free_recursive(objp, cachep->flags);
Suleiman Souhlala947eb92011-06-02 00:16:42 -07003440 objp = cache_free_debugcheck(cachep, objp, caller);
Bharata B Raod1b2cf62020-10-13 16:53:09 -07003441 memcg_slab_free_hook(cachep, &objp, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442
Siddha, Suresh B1807a1a2007-08-22 14:01:49 -07003443 /*
3444 * Skip calling cache_free_alien() when the platform is not numa.
3445 * This will avoid cache misses that happen while accessing slabp (which
3446 * is per page memory reference) to get nodeid. Instead use a global
3447 * variable to skip the call, which is mostly likely to be present in
3448 * the cache.
3449 */
Mel Gormanb6e68bc12009-06-16 15:32:16 -07003450 if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
Pekka Enberg729bd0b2006-06-23 02:03:05 -07003451 return;
Christoph Lametere498be72005-09-09 13:03:32 -07003452
Joonsoo Kim3d880192014-10-09 15:26:04 -07003453 if (ac->avail < ac->limit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454 STATS_INC_FREEHIT(cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003455 } else {
3456 STATS_INC_FREEMISS(cachep);
3457 cache_flusharray(cachep, ac);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 }
Zhao Jin42c8c992011-08-27 00:26:17 +08003459
Joonsoo Kimf68f8dd2016-03-15 14:54:56 -07003460 if (sk_memalloc_socks()) {
3461 struct page *page = virt_to_head_page(objp);
3462
3463 if (unlikely(PageSlabPfmemalloc(page))) {
3464 cache_free_pfmemalloc(cachep, page, objp);
3465 return;
3466 }
3467 }
3468
Kees Cookdabc3e22020-08-06 23:18:24 -07003469 __free_one(ac, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003470}
3471
3472/**
3473 * kmem_cache_alloc - Allocate an object
3474 * @cachep: The cache to allocate from.
3475 * @flags: See kmalloc().
3476 *
3477 * Allocate an object from this cache. The flags are only relevant
3478 * if the cache has no available objects.
Mike Rapoporta862f682019-03-05 15:48:42 -08003479 *
3480 * Return: pointer to the new object or %NULL in case of error
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003482void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003484 void *ret = slab_alloc(cachep, flags, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003485
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003486 trace_kmem_cache_alloc(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003487 cachep->object_size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003488
3489 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490}
3491EXPORT_SYMBOL(kmem_cache_alloc);
3492
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003493static __always_inline void
3494cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags,
3495 size_t size, void **p, unsigned long caller)
3496{
3497 size_t i;
3498
3499 for (i = 0; i < size; i++)
3500 p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller);
3501}
3502
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003503int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003504 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003505{
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003506 size_t i;
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003507 struct obj_cgroup *objcg = NULL;
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003508
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003509 s = slab_pre_alloc_hook(s, &objcg, size, flags);
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003510 if (!s)
3511 return 0;
3512
3513 cache_alloc_debugcheck_before(s, flags);
3514
3515 local_irq_disable();
3516 for (i = 0; i < size; i++) {
3517 void *objp = __do_cache_alloc(s, flags);
3518
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003519 if (unlikely(!objp))
3520 goto error;
3521 p[i] = objp;
3522 }
3523 local_irq_enable();
3524
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003525 cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_);
3526
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003527 /* Clear memory outside IRQ disabled section */
Alexander Potapenko64713842019-07-11 20:59:19 -07003528 if (unlikely(slab_want_init_on_alloc(flags, s)))
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003529 for (i = 0; i < size; i++)
3530 memset(p[i], 0, s->object_size);
3531
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003532 slab_post_alloc_hook(s, objcg, flags, size, p);
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003533 /* FIXME: Trace call missing. Christoph would like a bulk variant */
3534 return size;
3535error:
3536 local_irq_enable();
Jesper Dangaard Brouer7b0501d2016-03-15 14:53:53 -07003537 cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_);
Roman Gushchin964d4bd2020-08-06 23:20:56 -07003538 slab_post_alloc_hook(s, objcg, flags, i, p);
Jesper Dangaard Brouer2a777ea2016-03-15 14:53:50 -07003539 __kmem_cache_free_bulk(s, i, p);
3540 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003541}
3542EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3543
Li Zefan0f24f122009-12-11 15:45:30 +08003544#ifdef CONFIG_TRACING
Steven Rostedt85beb582010-11-24 16:23:34 -05003545void *
Ezequiel Garcia40521472012-09-08 17:47:56 -03003546kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003547{
Steven Rostedt85beb582010-11-24 16:23:34 -05003548 void *ret;
3549
Ezequiel Garcia48356302012-09-08 17:47:57 -03003550 ret = slab_alloc(cachep, flags, _RET_IP_);
Steven Rostedt85beb582010-11-24 16:23:34 -05003551
Andrey Konovalov0116523c2018-12-28 00:29:37 -08003552 ret = kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003553 trace_kmalloc(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003554 size, cachep->size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003555 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003556}
Steven Rostedt85beb582010-11-24 16:23:34 -05003557EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003558#endif
3559
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560#ifdef CONFIG_NUMA
Zhouping Liud0d04b72013-05-16 11:36:23 +08003561/**
3562 * kmem_cache_alloc_node - Allocate an object on the specified node
3563 * @cachep: The cache to allocate from.
3564 * @flags: See kmalloc().
3565 * @nodeid: node number of the target node.
3566 *
3567 * Identical to kmem_cache_alloc but it will allocate memory on the given
3568 * node, which can improve the performance for cpu bound structures.
3569 *
3570 * Fallback to other node is possible if __GFP_THISNODE is not set.
Mike Rapoporta862f682019-03-05 15:48:42 -08003571 *
3572 * Return: pointer to the new object or %NULL in case of error
Zhouping Liud0d04b72013-05-16 11:36:23 +08003573 */
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003574void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
3575{
Ezequiel Garcia48356302012-09-08 17:47:57 -03003576 void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003577
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003578 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003579 cachep->object_size, cachep->size,
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003580 flags, nodeid);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003581
3582 return ret;
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003583}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584EXPORT_SYMBOL(kmem_cache_alloc_node);
3585
Li Zefan0f24f122009-12-11 15:45:30 +08003586#ifdef CONFIG_TRACING
Ezequiel Garcia40521472012-09-08 17:47:56 -03003587void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep,
Steven Rostedt85beb582010-11-24 16:23:34 -05003588 gfp_t flags,
Ezequiel Garcia40521472012-09-08 17:47:56 -03003589 int nodeid,
3590 size_t size)
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003591{
Steven Rostedt85beb582010-11-24 16:23:34 -05003592 void *ret;
3593
Ezequiel Garcia592f4142012-09-25 08:07:08 -03003594 ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003595
Andrey Konovalov0116523c2018-12-28 00:29:37 -08003596 ret = kasan_kmalloc(cachep, ret, size, flags);
Steven Rostedt85beb582010-11-24 16:23:34 -05003597 trace_kmalloc_node(_RET_IP_, ret,
Ezequiel Garciaff4fcd02012-09-08 17:47:52 -03003598 size, cachep->size,
Steven Rostedt85beb582010-11-24 16:23:34 -05003599 flags, nodeid);
3600 return ret;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003601}
Steven Rostedt85beb582010-11-24 16:23:34 -05003602EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003603#endif
3604
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003605static __always_inline void *
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003606__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003607{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003608 struct kmem_cache *cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003609 void *ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003610
Dmitry Vyukov61448472018-10-26 15:03:12 -07003611 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3612 return NULL;
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003613 cachep = kmalloc_slab(size, flags);
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003614 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3615 return cachep;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003616 ret = kmem_cache_alloc_node_trace(cachep, flags, node, size);
Andrey Konovalov0116523c2018-12-28 00:29:37 -08003617 ret = kasan_kmalloc(cachep, ret, size, flags);
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07003618
3619 return ret;
Manfred Spraul97e2bde2005-05-01 08:58:38 -07003620}
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003621
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003622void *__kmalloc_node(size_t size, gfp_t flags, int node)
3623{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003624 return __do_kmalloc_node(size, flags, node, _RET_IP_);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003625}
Christoph Hellwigdbe5e692006-09-25 23:31:36 -07003626EXPORT_SYMBOL(__kmalloc_node);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003627
3628void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003629 int node, unsigned long caller)
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003630{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003631 return __do_kmalloc_node(size, flags, node, caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003632}
3633EXPORT_SYMBOL(__kmalloc_node_track_caller);
Christoph Hellwig8b98c162006-12-06 20:32:30 -08003634#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635
3636/**
Paul Drynoff800590f2006-06-23 02:03:48 -07003637 * __do_kmalloc - allocate memory
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 * @size: how many bytes of memory are required.
Paul Drynoff800590f2006-06-23 02:03:48 -07003639 * @flags: the type of memory to allocate (see kmalloc).
Randy Dunlap911851e2006-03-22 00:08:14 -08003640 * @caller: function caller for debug tracking of the caller
Mike Rapoporta862f682019-03-05 15:48:42 -08003641 *
3642 * Return: pointer to the allocated memory or %NULL in case of error
Linus Torvalds1da177e2005-04-16 15:20:36 -07003643 */
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003644static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003645 unsigned long caller)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003646{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003647 struct kmem_cache *cachep;
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003648 void *ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003649
Dmitry Vyukov61448472018-10-26 15:03:12 -07003650 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
3651 return NULL;
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003652 cachep = kmalloc_slab(size, flags);
Linus Torvaldsa5c96d82007-07-19 13:17:15 -07003653 if (unlikely(ZERO_OR_NULL_PTR(cachep)))
3654 return cachep;
Ezequiel Garcia48356302012-09-08 17:47:57 -03003655 ret = slab_alloc(cachep, flags, caller);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003656
Andrey Konovalov0116523c2018-12-28 00:29:37 -08003657 ret = kasan_kmalloc(cachep, ret, size, flags);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003658 trace_kmalloc(caller, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003659 size, cachep->size, flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003660
3661 return ret;
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003662}
3663
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003664void *__kmalloc(size_t size, gfp_t flags)
3665{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003666 return __do_kmalloc(size, flags, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003667}
3668EXPORT_SYMBOL(__kmalloc);
3669
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003670void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003671{
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003672 return __do_kmalloc(size, flags, caller);
Pekka Enberg7fd6b142006-02-01 03:05:52 -08003673}
3674EXPORT_SYMBOL(__kmalloc_track_caller);
Christoph Hellwig1d2c8ee2006-10-04 02:15:25 -07003675
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676/**
3677 * kmem_cache_free - Deallocate an object
3678 * @cachep: The cache the allocation was from.
3679 * @objp: The previously allocated object.
3680 *
3681 * Free an object which was previously allocated from this
3682 * cache.
3683 */
Pekka Enberg343e0d72006-02-01 03:05:50 -08003684void kmem_cache_free(struct kmem_cache *cachep, void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685{
3686 unsigned long flags;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08003687 cachep = cache_from_obj(cachep, objp);
3688 if (!cachep)
3689 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003690
3691 local_irq_save(flags);
Feng Tangd97d4762012-07-02 14:29:10 +08003692 debug_check_no_locks_freed(objp, cachep->object_size);
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -07003693 if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003694 debug_check_no_obj_freed(objp, cachep->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003695 __cache_free(cachep, objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696 local_irq_restore(flags);
Eduard - Gabriel Munteanu36555752008-08-10 20:14:05 +03003697
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003698 trace_kmem_cache_free(_RET_IP_, objp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699}
3700EXPORT_SYMBOL(kmem_cache_free);
3701
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003702void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p)
3703{
3704 struct kmem_cache *s;
3705 size_t i;
3706
3707 local_irq_disable();
3708 for (i = 0; i < size; i++) {
3709 void *objp = p[i];
3710
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003711 if (!orig_s) /* called via kfree_bulk */
3712 s = virt_to_cache(objp);
3713 else
3714 s = cache_from_obj(orig_s, objp);
Kees Cooka64b5372019-07-11 20:53:26 -07003715 if (!s)
3716 continue;
Jesper Dangaard Brouere6cdb582016-03-15 14:53:56 -07003717
3718 debug_check_no_locks_freed(objp, s->object_size);
3719 if (!(s->flags & SLAB_DEBUG_OBJECTS))
3720 debug_check_no_obj_freed(objp, s->object_size);
3721
3722 __cache_free(s, objp, _RET_IP_);
3723 }
3724 local_irq_enable();
3725
3726 /* FIXME: add tracing */
3727}
3728EXPORT_SYMBOL(kmem_cache_free_bulk);
3729
Linus Torvalds1da177e2005-04-16 15:20:36 -07003730/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 * kfree - free previously allocated memory
3732 * @objp: pointer returned by kmalloc.
3733 *
Pekka Enberg80e93ef2005-09-09 13:10:16 -07003734 * If @objp is NULL, no operation is performed.
3735 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07003736 * Don't free memory not originally allocated by kmalloc()
3737 * or you will run into trouble.
3738 */
3739void kfree(const void *objp)
3740{
Pekka Enberg343e0d72006-02-01 03:05:50 -08003741 struct kmem_cache *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742 unsigned long flags;
3743
Pekka Enberg2121db72009-03-25 11:05:57 +02003744 trace_kfree(_RET_IP_, objp);
3745
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003746 if (unlikely(ZERO_OR_NULL_PTR(objp)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003747 return;
3748 local_irq_save(flags);
3749 kfree_debugcheck(objp);
Pekka Enberg6ed5eb2212006-02-01 03:05:49 -08003750 c = virt_to_cache(objp);
Kees Cooka64b5372019-07-11 20:53:26 -07003751 if (!c) {
3752 local_irq_restore(flags);
3753 return;
3754 }
Christoph Lameter8c138bc2012-06-13 10:24:58 -05003755 debug_check_no_locks_freed(objp, c->object_size);
3756
3757 debug_check_no_obj_freed(objp, c->object_size);
Ezequiel Garcia7c0cb9c2012-09-08 17:47:55 -03003758 __cache_free(c, (void *)objp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003759 local_irq_restore(flags);
3760}
3761EXPORT_SYMBOL(kfree);
3762
Christoph Lametere498be72005-09-09 13:03:32 -07003763/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003764 * This initializes kmem_cache_node or resizes various caches for all nodes.
Christoph Lametere498be72005-09-09 13:03:32 -07003765 */
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003766static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp)
Christoph Lametere498be72005-09-09 13:03:32 -07003767{
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003768 int ret;
Christoph Lametere498be72005-09-09 13:03:32 -07003769 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003770 struct kmem_cache_node *n;
Christoph Lametere498be72005-09-09 13:03:32 -07003771
Mel Gorman9c09a952008-01-24 05:49:54 -08003772 for_each_online_node(node) {
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003773 ret = setup_kmem_cache_node(cachep, node, gfp, true);
3774 if (ret)
Christoph Lametere498be72005-09-09 13:03:32 -07003775 goto fail;
3776
Christoph Lametere498be72005-09-09 13:03:32 -07003777 }
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003778
Christoph Lametercafeb022006-03-25 03:06:46 -08003779 return 0;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003780
Andrew Mortona737b3e2006-03-22 00:08:11 -08003781fail:
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003782 if (!cachep->list.next) {
Christoph Lameter0718dc22006-03-25 03:06:47 -08003783 /* Cache is not active yet. Roll back what we did */
3784 node--;
3785 while (node >= 0) {
Christoph Lameter18bf8542014-08-06 16:04:11 -07003786 n = get_node(cachep, node);
3787 if (n) {
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003788 kfree(n->shared);
3789 free_alien_cache(n->alien);
3790 kfree(n);
Christoph Lameter6a673682013-01-10 19:14:19 +00003791 cachep->node[node] = NULL;
Christoph Lameter0718dc22006-03-25 03:06:47 -08003792 }
3793 node--;
3794 }
3795 }
Christoph Lametercafeb022006-03-25 03:06:46 -08003796 return -ENOMEM;
Christoph Lametere498be72005-09-09 13:03:32 -07003797}
3798
Christoph Lameter18004c52012-07-06 15:25:12 -05003799/* Always called with the slab_mutex held */
Roman Gushchin10befea2020-08-06 23:21:27 -07003800static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
3801 int batchcount, int shared, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003802{
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003803 struct array_cache __percpu *cpu_cache, *prev;
3804 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003806 cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount);
3807 if (!cpu_cache)
Siddha, Suresh Bd2e7b7d2006-09-25 23:31:47 -07003808 return -ENOMEM;
3809
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003810 prev = cachep->cpu_cache;
3811 cachep->cpu_cache = cpu_cache;
Greg Thelena87c75f2017-05-03 14:51:47 -07003812 /*
3813 * Without a previous cpu_cache there's no need to synchronize remote
3814 * cpus, so skip the IPIs.
3815 */
3816 if (prev)
3817 kick_all_cpus_sync();
Christoph Lametere498be72005-09-09 13:03:32 -07003818
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 check_irq_on();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820 cachep->batchcount = batchcount;
3821 cachep->limit = limit;
Christoph Lametere498be72005-09-09 13:03:32 -07003822 cachep->shared = shared;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003824 if (!prev)
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003825 goto setup_node;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003826
3827 for_each_online_cpu(cpu) {
Joonsoo Kim97654df2014-08-06 16:04:25 -07003828 LIST_HEAD(list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003829 int node;
3830 struct kmem_cache_node *n;
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003831 struct array_cache *ac = per_cpu_ptr(prev, cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003832
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003833 node = cpu_to_mem(cpu);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003834 n = get_node(cachep, node);
3835 spin_lock_irq(&n->list_lock);
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003836 free_block(cachep, ac->entry, ac->avail, node, &list);
Christoph Lameter18bf8542014-08-06 16:04:11 -07003837 spin_unlock_irq(&n->list_lock);
Joonsoo Kim97654df2014-08-06 16:04:25 -07003838 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839 }
Joonsoo Kimbf0dea22014-10-09 15:26:27 -07003840 free_percpu(prev);
3841
Joonsoo Kimc3d332b2016-05-19 17:10:14 -07003842setup_node:
3843 return setup_kmem_cache_nodes(cachep, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003844}
3845
Christoph Lameter18004c52012-07-06 15:25:12 -05003846/* Called with slab_mutex held always */
Pekka Enberg83b519e2009-06-10 19:40:04 +03003847static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003848{
3849 int err;
Glauber Costa943a4512012-12-18 14:23:03 -08003850 int limit = 0;
3851 int shared = 0;
3852 int batchcount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853
Thomas Garnier7c00fce2016-07-26 15:21:56 -07003854 err = cache_random_seq_create(cachep, cachep->num, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07003855 if (err)
3856 goto end;
3857
Glauber Costa943a4512012-12-18 14:23:03 -08003858 if (limit && shared && batchcount)
3859 goto skip_setup;
Andrew Mortona737b3e2006-03-22 00:08:11 -08003860 /*
3861 * The head array serves three purposes:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862 * - create a LIFO ordering, i.e. return objects that are cache-warm
3863 * - reduce the number of spinlock operations.
Andrew Mortona737b3e2006-03-22 00:08:11 -08003864 * - reduce the number of linked list operations on the slab and
Linus Torvalds1da177e2005-04-16 15:20:36 -07003865 * bufctl chains: array operations are cheaper.
3866 * The numbers are guessed, we should auto-tune as described by
3867 * Bonwick.
3868 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003869 if (cachep->size > 131072)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003870 limit = 1;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003871 else if (cachep->size > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003872 limit = 8;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003873 else if (cachep->size > 1024)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 limit = 24;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003875 else if (cachep->size > 256)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003876 limit = 54;
3877 else
3878 limit = 120;
3879
Andrew Mortona737b3e2006-03-22 00:08:11 -08003880 /*
3881 * CPU bound tasks (e.g. network routing) can exhibit cpu bound
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 * allocation behaviour: Most allocs on one cpu, most free operations
3883 * on another cpu. For these cases, an efficient object passing between
3884 * cpus is necessary. This is provided by a shared array. The array
3885 * replaces Bonwick's magazine layer.
3886 * On uniprocessor, it's functionally equivalent (but less efficient)
3887 * to a larger limit. Thus disabled by default.
3888 */
3889 shared = 0;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003890 if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 shared = 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003892
3893#if DEBUG
Andrew Mortona737b3e2006-03-22 00:08:11 -08003894 /*
3895 * With debugging enabled, large batchcount lead to excessively long
3896 * periods with disabled local interrupts. Limit the batchcount
Linus Torvalds1da177e2005-04-16 15:20:36 -07003897 */
3898 if (limit > 32)
3899 limit = 32;
3900#endif
Glauber Costa943a4512012-12-18 14:23:03 -08003901 batchcount = (limit + 1) / 2;
3902skip_setup:
3903 err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
Thomas Garnierc7ce4f602016-05-19 17:10:37 -07003904end:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003905 if (err)
Joe Perches11705322016-03-17 14:19:50 -07003906 pr_err("enable_cpucache failed for %s, error %d\n",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08003907 cachep->name, -err);
Christoph Lameter2ed3a4e2006-09-25 23:31:38 -07003908 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003909}
3910
Christoph Lameter1b552532006-03-22 00:09:07 -08003911/*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003912 * Drain an array if it contains any elements taking the node lock only if
3913 * necessary. Note that the node listlock also protects the array_cache
Christoph Lameterb18e7e62006-03-22 00:09:07 -08003914 * if drain_array() is used on the shared array.
Christoph Lameter1b552532006-03-22 00:09:07 -08003915 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003916static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n,
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003917 struct array_cache *ac, int node)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918{
Joonsoo Kim97654df2014-08-06 16:04:25 -07003919 LIST_HEAD(list);
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003920
3921 /* ac from n->shared can be freed if we don't hold the slab_mutex. */
3922 check_mutex_acquired();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923
Christoph Lameter1b552532006-03-22 00:09:07 -08003924 if (!ac || !ac->avail)
3925 return;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003926
3927 if (ac->touched) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 ac->touched = 0;
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003929 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930 }
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003931
3932 spin_lock_irq(&n->list_lock);
3933 drain_array_locked(cachep, ac, node, false, &list);
3934 spin_unlock_irq(&n->list_lock);
3935
3936 slabs_destroy(cachep, &list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937}
3938
3939/**
3940 * cache_reap - Reclaim memory from caches.
Randy Dunlap05fb6bf2007-02-28 20:12:13 -08003941 * @w: work descriptor
Linus Torvalds1da177e2005-04-16 15:20:36 -07003942 *
3943 * Called from workqueue/eventd every few seconds.
3944 * Purpose:
3945 * - clear the per-cpu caches for this CPU.
3946 * - return freeable pages to the main free memory pool.
3947 *
Andrew Mortona737b3e2006-03-22 00:08:11 -08003948 * If we cannot acquire the cache chain mutex then just give up - we'll try
3949 * again on the next iteration.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08003951static void cache_reap(struct work_struct *w)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003952{
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07003953 struct kmem_cache *searchp;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003954 struct kmem_cache_node *n;
Lee Schermerhorn7d6e6d02010-05-26 14:45:03 -07003955 int node = numa_mem_id();
Jean Delvarebf6aede2009-04-02 16:56:54 -07003956 struct delayed_work *work = to_delayed_work(w);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957
Christoph Lameter18004c52012-07-06 15:25:12 -05003958 if (!mutex_trylock(&slab_mutex))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003959 /* Give up. Setup the next iteration. */
Christoph Lameter7c5cae32007-02-10 01:42:55 -08003960 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003961
Christoph Lameter18004c52012-07-06 15:25:12 -05003962 list_for_each_entry(searchp, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003963 check_irq_on();
3964
Christoph Lameter35386e32006-03-22 00:09:05 -08003965 /*
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003966 * We only take the node lock if absolutely necessary and we
Christoph Lameter35386e32006-03-22 00:09:05 -08003967 * have established with reasonable certainty that
3968 * we can do some work if the lock was obtained.
3969 */
Christoph Lameter18bf8542014-08-06 16:04:11 -07003970 n = get_node(searchp, node);
Christoph Lameter35386e32006-03-22 00:09:05 -08003971
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003972 reap_alien(searchp, n);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003973
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003974 drain_array(searchp, n, cpu_cache_get(searchp), node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003975
Christoph Lameter35386e32006-03-22 00:09:05 -08003976 /*
3977 * These are racy checks but it does not matter
3978 * if we skip one check or scan twice.
3979 */
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003980 if (time_after(n->next_reap, jiffies))
Christoph Lameter35386e32006-03-22 00:09:05 -08003981 goto next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982
Jianyu Zhan5f0985b2014-03-30 17:02:20 +08003983 n->next_reap = jiffies + REAPTIMEOUT_NODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003984
Joonsoo Kim18726ca2016-05-19 17:10:02 -07003985 drain_array(searchp, n, n->shared, node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003987 if (n->free_touched)
3988 n->free_touched = 0;
Christoph Lametered11d9e2006-06-30 01:55:45 -07003989 else {
3990 int freed;
3991
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00003992 freed = drain_freelist(searchp, n, (n->free_limit +
Christoph Lametered11d9e2006-06-30 01:55:45 -07003993 5 * searchp->num - 1) / (5 * searchp->num));
3994 STATS_ADD_REAPED(searchp, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 }
Christoph Lameter35386e32006-03-22 00:09:05 -08003996next:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997 cond_resched();
3998 }
3999 check_irq_on();
Christoph Lameter18004c52012-07-06 15:25:12 -05004000 mutex_unlock(&slab_mutex);
Christoph Lameter8fce4d82006-03-09 17:33:54 -08004001 next_reap_node();
Christoph Lameter7c5cae32007-02-10 01:42:55 -08004002out:
Andrew Mortona737b3e2006-03-22 00:08:11 -08004003 /* Set up the next iteration */
Vlastimil Babkaa9f2a8462018-04-13 15:35:38 -07004004 schedule_delayed_work_on(smp_processor_id(), work,
4005 round_jiffies_relative(REAPTIMEOUT_AC));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006}
4007
Glauber Costa0d7561c2012-10-19 18:20:27 +04004008void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009{
Greg Thelenf728b0a2016-12-12 16:41:41 -08004010 unsigned long active_objs, num_objs, active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004011 unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0;
4012 unsigned long free_slabs = 0;
Christoph Lametere498be72005-09-09 13:03:32 -07004013 int node;
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004014 struct kmem_cache_node *n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015
Christoph Lameter18bf8542014-08-06 16:04:11 -07004016 for_each_kmem_cache_node(cachep, node, n) {
Ravikiran G Thirumalaica3b9b92006-02-04 23:27:58 -08004017 check_irq_on();
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004018 spin_lock_irq(&n->list_lock);
Christoph Lametere498be72005-09-09 13:03:32 -07004019
David Rientjesbf00bd32016-12-12 16:41:44 -08004020 total_slabs += n->total_slabs;
4021 free_slabs += n->free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004022 free_objs += n->free_objects;
Aruna Ramakrishna07a63c42016-10-27 17:46:32 -07004023
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004024 if (n->shared)
4025 shared_avail += n->shared->avail;
Christoph Lametere498be72005-09-09 13:03:32 -07004026
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004027 spin_unlock_irq(&n->list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004028 }
David Rientjesbf00bd32016-12-12 16:41:44 -08004029 num_objs = total_slabs * cachep->num;
4030 active_slabs = total_slabs - free_slabs;
Greg Thelenf728b0a2016-12-12 16:41:41 -08004031 active_objs = num_objs - free_objs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004032
Glauber Costa0d7561c2012-10-19 18:20:27 +04004033 sinfo->active_objs = active_objs;
4034 sinfo->num_objs = num_objs;
4035 sinfo->active_slabs = active_slabs;
David Rientjesbf00bd32016-12-12 16:41:44 -08004036 sinfo->num_slabs = total_slabs;
Glauber Costa0d7561c2012-10-19 18:20:27 +04004037 sinfo->shared_avail = shared_avail;
4038 sinfo->limit = cachep->limit;
4039 sinfo->batchcount = cachep->batchcount;
4040 sinfo->shared = cachep->shared;
4041 sinfo->objects_per_slab = cachep->num;
4042 sinfo->cache_order = cachep->gfporder;
4043}
4044
4045void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep)
4046{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047#if STATS
Christoph Lameterce8eb6c2013-01-10 19:14:19 +00004048 { /* node stats */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004049 unsigned long high = cachep->high_mark;
4050 unsigned long allocs = cachep->num_allocations;
4051 unsigned long grown = cachep->grown;
4052 unsigned long reaped = cachep->reaped;
4053 unsigned long errors = cachep->errors;
4054 unsigned long max_freeable = cachep->max_freeable;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055 unsigned long node_allocs = cachep->node_allocs;
Christoph Lametere498be72005-09-09 13:03:32 -07004056 unsigned long node_frees = cachep->node_frees;
Ravikiran G Thirumalaifb7faf32006-04-10 22:52:54 -07004057 unsigned long overflows = cachep->node_overflow;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058
Joe Perches756a0252016-03-17 14:19:47 -07004059 seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu",
Joe Perchese92dd4f2010-03-26 19:27:58 -07004060 allocs, high, grown,
4061 reaped, errors, max_freeable, node_allocs,
4062 node_frees, overflows);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 }
4064 /* cpu stats */
4065 {
4066 unsigned long allochit = atomic_read(&cachep->allochit);
4067 unsigned long allocmiss = atomic_read(&cachep->allocmiss);
4068 unsigned long freehit = atomic_read(&cachep->freehit);
4069 unsigned long freemiss = atomic_read(&cachep->freemiss);
4070
4071 seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004072 allochit, allocmiss, freehit, freemiss);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004073 }
4074#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075}
4076
Linus Torvalds1da177e2005-04-16 15:20:36 -07004077#define MAX_SLABINFO_WRITE 128
4078/**
4079 * slabinfo_write - Tuning for the slab allocator
4080 * @file: unused
4081 * @buffer: user buffer
4082 * @count: data length
4083 * @ppos: unused
Mike Rapoporta862f682019-03-05 15:48:42 -08004084 *
4085 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004086 */
Glauber Costab7454ad2012-10-19 18:20:25 +04004087ssize_t slabinfo_write(struct file *file, const char __user *buffer,
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004088 size_t count, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089{
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004090 char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 int limit, batchcount, shared, res;
Christoph Hellwig7a7c3812006-06-23 02:03:17 -07004092 struct kmem_cache *cachep;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004093
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 if (count > MAX_SLABINFO_WRITE)
4095 return -EINVAL;
4096 if (copy_from_user(&kbuf, buffer, count))
4097 return -EFAULT;
Pekka Enbergb28a02d2006-01-08 01:00:37 -08004098 kbuf[MAX_SLABINFO_WRITE] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -07004099
4100 tmp = strchr(kbuf, ' ');
4101 if (!tmp)
4102 return -EINVAL;
4103 *tmp = '\0';
4104 tmp++;
4105 if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
4106 return -EINVAL;
4107
4108 /* Find the cache in the chain of caches. */
Christoph Lameter18004c52012-07-06 15:25:12 -05004109 mutex_lock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110 res = -EINVAL;
Christoph Lameter18004c52012-07-06 15:25:12 -05004111 list_for_each_entry(cachep, &slab_caches, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112 if (!strcmp(cachep->name, kbuf)) {
Andrew Mortona737b3e2006-03-22 00:08:11 -08004113 if (limit < 1 || batchcount < 1 ||
4114 batchcount > limit || shared < 0) {
Christoph Lametere498be72005-09-09 13:03:32 -07004115 res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004116 } else {
Christoph Lametere498be72005-09-09 13:03:32 -07004117 res = do_tune_cpucache(cachep, limit,
Pekka Enberg83b519e2009-06-10 19:40:04 +03004118 batchcount, shared,
4119 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004120 }
4121 break;
4122 }
4123 }
Christoph Lameter18004c52012-07-06 15:25:12 -05004124 mutex_unlock(&slab_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004125 if (res >= 0)
4126 res = count;
4127 return res;
4128}
Al Viro871751e2006-03-25 03:06:39 -08004129
Kees Cook04385fc2016-06-23 15:20:59 -07004130#ifdef CONFIG_HARDENED_USERCOPY
4131/*
Kees Cookafcc90f82018-01-10 15:17:01 -08004132 * Rejects incorrectly sized objects and objects that are to be copied
4133 * to/from userspace but do not fall entirely within the containing slab
4134 * cache's usercopy region.
Kees Cook04385fc2016-06-23 15:20:59 -07004135 *
4136 * Returns NULL if check passes, otherwise const char * to name of cache
4137 * to indicate an error.
4138 */
Kees Cookf4e6e282018-01-10 14:48:22 -08004139void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
4140 bool to_user)
Kees Cook04385fc2016-06-23 15:20:59 -07004141{
4142 struct kmem_cache *cachep;
4143 unsigned int objnr;
4144 unsigned long offset;
4145
Andrey Konovalov219667c2019-02-20 22:20:25 -08004146 ptr = kasan_reset_tag(ptr);
4147
Kees Cook04385fc2016-06-23 15:20:59 -07004148 /* Find and validate object. */
4149 cachep = page->slab_cache;
4150 objnr = obj_to_index(cachep, page, (void *)ptr);
4151 BUG_ON(objnr >= cachep->num);
4152
4153 /* Find offset within object. */
4154 offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep);
4155
Kees Cookafcc90f82018-01-10 15:17:01 -08004156 /* Allow address range falling entirely within usercopy region. */
4157 if (offset >= cachep->useroffset &&
4158 offset - cachep->useroffset <= cachep->usersize &&
4159 n <= cachep->useroffset - offset + cachep->usersize)
Kees Cookf4e6e282018-01-10 14:48:22 -08004160 return;
Kees Cook04385fc2016-06-23 15:20:59 -07004161
Kees Cookafcc90f82018-01-10 15:17:01 -08004162 /*
4163 * If the copy is still within the allocated object, produce
4164 * a warning instead of rejecting the copy. This is intended
4165 * to be a temporary method to find any missing usercopy
4166 * whitelists.
4167 */
Kees Cook2d891fb2017-11-30 13:04:32 -08004168 if (usercopy_fallback &&
4169 offset <= cachep->object_size &&
Kees Cookafcc90f82018-01-10 15:17:01 -08004170 n <= cachep->object_size - offset) {
4171 usercopy_warn("SLAB object", cachep->name, to_user, offset, n);
4172 return;
4173 }
4174
Kees Cookf4e6e282018-01-10 14:48:22 -08004175 usercopy_abort("SLAB object", cachep->name, to_user, offset, n);
Kees Cook04385fc2016-06-23 15:20:59 -07004176}
4177#endif /* CONFIG_HARDENED_USERCOPY */
4178
Manfred Spraul00e145b2005-09-03 15:55:07 -07004179/**
Marco Elver10d1f8c2019-07-11 20:54:14 -07004180 * __ksize -- Uninstrumented ksize.
Randy Dunlap87bf4f72019-10-14 14:12:26 -07004181 * @objp: pointer to the object
Manfred Spraul00e145b2005-09-03 15:55:07 -07004182 *
Marco Elver10d1f8c2019-07-11 20:54:14 -07004183 * Unlike ksize(), __ksize() is uninstrumented, and does not provide the same
4184 * safety checks as ksize() with KASAN instrumentation enabled.
Randy Dunlap87bf4f72019-10-14 14:12:26 -07004185 *
4186 * Return: size of the actual memory used by @objp in bytes
Manfred Spraul00e145b2005-09-03 15:55:07 -07004187 */
Marco Elver10d1f8c2019-07-11 20:54:14 -07004188size_t __ksize(const void *objp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189{
Kees Cooka64b5372019-07-11 20:53:26 -07004190 struct kmem_cache *c;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004191 size_t size;
4192
Christoph Lameteref8b4522007-10-16 01:24:46 -07004193 BUG_ON(!objp);
4194 if (unlikely(objp == ZERO_SIZE_PTR))
Manfred Spraul00e145b2005-09-03 15:55:07 -07004195 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004196
Kees Cooka64b5372019-07-11 20:53:26 -07004197 c = virt_to_cache(objp);
4198 size = c ? c->object_size : 0;
Alexander Potapenko7ed2f9e2016-03-25 14:21:59 -07004199
4200 return size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004201}
Marco Elver10d1f8c2019-07-11 20:54:14 -07004202EXPORT_SYMBOL(__ksize);