]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
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 | |
183ff22b | 30 | * slabs and you must pass objects with the same initializations to |
1da177e4 LT |
31 | * 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 | * | |
a737b3e2 | 54 | * The c_cpuarray may not be read with enabled local interrupts - |
1da177e4 LT |
55 | * it's changed with a smp_call_function(). |
56 | * | |
57 | * SMP synchronization: | |
58 | * constructors and destructors are called without any locking. | |
343e0d7a | 59 | * Several members in struct kmem_cache and struct slab never change, they |
1da177e4 LT |
60 | * 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 | |
18004c5d | 72 | * The global cache-chain is protected by the mutex 'slab_mutex'. |
1da177e4 LT |
73 | * 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 | * | |
e498be7d CL |
79 | * 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. | |
1da177e4 LT |
88 | */ |
89 | ||
1da177e4 LT |
90 | #include <linux/slab.h> |
91 | #include <linux/mm.h> | |
c9cf5528 | 92 | #include <linux/poison.h> |
1da177e4 LT |
93 | #include <linux/swap.h> |
94 | #include <linux/cache.h> | |
95 | #include <linux/interrupt.h> | |
96 | #include <linux/init.h> | |
97 | #include <linux/compiler.h> | |
101a5001 | 98 | #include <linux/cpuset.h> |
a0ec95a8 | 99 | #include <linux/proc_fs.h> |
1da177e4 LT |
100 | #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> | |
543537bd | 107 | #include <linux/string.h> |
138ae663 | 108 | #include <linux/uaccess.h> |
e498be7d | 109 | #include <linux/nodemask.h> |
d5cff635 | 110 | #include <linux/kmemleak.h> |
dc85da15 | 111 | #include <linux/mempolicy.h> |
fc0abb14 | 112 | #include <linux/mutex.h> |
8a8b6502 | 113 | #include <linux/fault-inject.h> |
e7eebaf6 | 114 | #include <linux/rtmutex.h> |
6a2d7a95 | 115 | #include <linux/reciprocal_div.h> |
3ac7fe5a | 116 | #include <linux/debugobjects.h> |
8f9f8d9e | 117 | #include <linux/memory.h> |
268bb0ce | 118 | #include <linux/prefetch.h> |
3f8c2452 | 119 | #include <linux/sched/task_stack.h> |
1da177e4 | 120 | |
381760ea MG |
121 | #include <net/sock.h> |
122 | ||
1da177e4 LT |
123 | #include <asm/cacheflush.h> |
124 | #include <asm/tlbflush.h> | |
125 | #include <asm/page.h> | |
126 | ||
4dee6b64 SR |
127 | #include <trace/events/kmem.h> |
128 | ||
072bb0aa MG |
129 | #include "internal.h" |
130 | ||
b9ce5ef4 GC |
131 | #include "slab.h" |
132 | ||
1da177e4 | 133 | /* |
50953fe9 | 134 | * DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON. |
1da177e4 LT |
135 | * 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 | ||
1da177e4 LT |
153 | /* Shouldn't this be in a header file somewhere? */ |
154 | #define BYTES_PER_WORD sizeof(void *) | |
87a927c7 | 155 | #define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long)) |
1da177e4 | 156 | |
1da177e4 LT |
157 | #ifndef ARCH_KMALLOC_FLAGS |
158 | #define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN | |
159 | #endif | |
160 | ||
f315e3fa JK |
161 | #define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \ |
162 | <= SLAB_OBJ_MIN_SIZE) ? 1 : 0) | |
163 | ||
164 | #if FREELIST_BYTE_INDEX | |
165 | typedef unsigned char freelist_idx_t; | |
166 | #else | |
167 | typedef unsigned short freelist_idx_t; | |
168 | #endif | |
169 | ||
30321c7b | 170 | #define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1) |
f315e3fa | 171 | |
1da177e4 LT |
172 | /* |
173 | * struct array_cache | |
174 | * | |
1da177e4 LT |
175 | * 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 | */ | |
184 | struct array_cache { | |
185 | unsigned int avail; | |
186 | unsigned int limit; | |
187 | unsigned int batchcount; | |
188 | unsigned int touched; | |
bda5b655 | 189 | void *entry[]; /* |
a737b3e2 AM |
190 | * Must have this definition in here for the proper |
191 | * alignment of array_cache. Also simplifies accessing | |
192 | * the entries. | |
a737b3e2 | 193 | */ |
1da177e4 LT |
194 | }; |
195 | ||
c8522a3a JK |
196 | struct alien_cache { |
197 | spinlock_t lock; | |
198 | struct array_cache ac; | |
199 | }; | |
200 | ||
e498be7d CL |
201 | /* |
202 | * Need this for bootstrapping a per node allocator. | |
203 | */ | |
bf0dea23 | 204 | #define NUM_INIT_LISTS (2 * MAX_NUMNODES) |
ce8eb6c4 | 205 | static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS]; |
e498be7d | 206 | #define CACHE_CACHE 0 |
bf0dea23 | 207 | #define SIZE_NODE (MAX_NUMNODES) |
e498be7d | 208 | |
ed11d9eb | 209 | static int drain_freelist(struct kmem_cache *cache, |
ce8eb6c4 | 210 | struct kmem_cache_node *n, int tofree); |
ed11d9eb | 211 | static void free_block(struct kmem_cache *cachep, void **objpp, int len, |
97654dfa JK |
212 | int node, struct list_head *list); |
213 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list); | |
83b519e8 | 214 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp); |
65f27f38 | 215 | static void cache_reap(struct work_struct *unused); |
ed11d9eb | 216 | |
76b342bd JK |
217 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
218 | void **list); | |
219 | static inline void fixup_slab_list(struct kmem_cache *cachep, | |
220 | struct kmem_cache_node *n, struct page *page, | |
221 | void **list); | |
e0a42726 IM |
222 | static int slab_early_init = 1; |
223 | ||
ce8eb6c4 | 224 | #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) |
1da177e4 | 225 | |
ce8eb6c4 | 226 | static void kmem_cache_node_init(struct kmem_cache_node *parent) |
e498be7d CL |
227 | { |
228 | INIT_LIST_HEAD(&parent->slabs_full); | |
229 | INIT_LIST_HEAD(&parent->slabs_partial); | |
230 | INIT_LIST_HEAD(&parent->slabs_free); | |
bf00bd34 | 231 | parent->total_slabs = 0; |
f728b0a5 | 232 | parent->free_slabs = 0; |
e498be7d CL |
233 | parent->shared = NULL; |
234 | parent->alien = NULL; | |
2e1217cf | 235 | parent->colour_next = 0; |
e498be7d CL |
236 | spin_lock_init(&parent->list_lock); |
237 | parent->free_objects = 0; | |
238 | parent->free_touched = 0; | |
239 | } | |
240 | ||
a737b3e2 AM |
241 | #define MAKE_LIST(cachep, listp, slab, nodeid) \ |
242 | do { \ | |
243 | INIT_LIST_HEAD(listp); \ | |
18bf8541 | 244 | list_splice(&get_node(cachep, nodeid)->slab, listp); \ |
e498be7d CL |
245 | } while (0) |
246 | ||
a737b3e2 AM |
247 | #define MAKE_ALL_LISTS(cachep, ptr, nodeid) \ |
248 | do { \ | |
e498be7d CL |
249 | 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) | |
1da177e4 | 253 | |
4fd0b46e AD |
254 | #define CFLGS_OBJFREELIST_SLAB ((slab_flags_t __force)0x40000000U) |
255 | #define CFLGS_OFF_SLAB ((slab_flags_t __force)0x80000000U) | |
b03a017b | 256 | #define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB) |
1da177e4 LT |
257 | #define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB) |
258 | ||
259 | #define BATCHREFILL_LIMIT 16 | |
a737b3e2 AM |
260 | /* |
261 | * Optimization question: fewer reaps means less probability for unnessary | |
262 | * cpucache drain/refill cycles. | |
1da177e4 | 263 | * |
dc6f3f27 | 264 | * OTOH the cpuarrays can contain lots of objects, |
1da177e4 LT |
265 | * which could lock up otherwise freeable slabs. |
266 | */ | |
5f0985bb JZ |
267 | #define REAPTIMEOUT_AC (2*HZ) |
268 | #define REAPTIMEOUT_NODE (4*HZ) | |
1da177e4 LT |
269 | |
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++) | |
ed11d9eb | 275 | #define STATS_ADD_REAPED(x,y) ((x)->reaped += (y)) |
a737b3e2 AM |
276 | #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) | |
1da177e4 LT |
281 | #define STATS_INC_ERR(x) ((x)->errors++) |
282 | #define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++) | |
e498be7d | 283 | #define STATS_INC_NODEFREES(x) ((x)->node_frees++) |
fb7faf33 | 284 | #define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++) |
a737b3e2 AM |
285 | #define STATS_SET_FREEABLE(x, i) \ |
286 | do { \ | |
287 | if ((x)->max_freeable < i) \ | |
288 | (x)->max_freeable = i; \ | |
289 | } while (0) | |
1da177e4 LT |
290 | #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) | |
4e60c86b | 299 | #define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0) |
1da177e4 LT |
300 | #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) | |
e498be7d | 303 | #define STATS_INC_NODEFREES(x) do { } while (0) |
fb7faf33 | 304 | #define STATS_INC_ACOVERFLOW(x) do { } while (0) |
a737b3e2 | 305 | #define STATS_SET_FREEABLE(x, i) do { } while (0) |
1da177e4 LT |
306 | #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 | |
1da177e4 | 313 | |
a737b3e2 AM |
314 | /* |
315 | * memory layout of objects: | |
1da177e4 | 316 | * 0 : objp |
3dafccf2 | 317 | * 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that |
1da177e4 LT |
318 | * the end of an object is aligned with the end of the real |
319 | * allocation. Catches writes behind the end of the allocation. | |
3dafccf2 | 320 | * cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1: |
1da177e4 | 321 | * redzone word. |
3dafccf2 | 322 | * cachep->obj_offset: The real object. |
3b0efdfa CL |
323 | * cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long] |
324 | * cachep->size - 1* BYTES_PER_WORD: last caller address | |
a737b3e2 | 325 | * [BYTES_PER_WORD long] |
1da177e4 | 326 | */ |
343e0d7a | 327 | static int obj_offset(struct kmem_cache *cachep) |
1da177e4 | 328 | { |
3dafccf2 | 329 | return cachep->obj_offset; |
1da177e4 LT |
330 | } |
331 | ||
b46b8f19 | 332 | static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
333 | { |
334 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
b46b8f19 DW |
335 | return (unsigned long long*) (objp + obj_offset(cachep) - |
336 | sizeof(unsigned long long)); | |
1da177e4 LT |
337 | } |
338 | ||
b46b8f19 | 339 | static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
340 | { |
341 | BUG_ON(!(cachep->flags & SLAB_RED_ZONE)); | |
342 | if (cachep->flags & SLAB_STORE_USER) | |
3b0efdfa | 343 | return (unsigned long long *)(objp + cachep->size - |
b46b8f19 | 344 | sizeof(unsigned long long) - |
87a927c7 | 345 | REDZONE_ALIGN); |
3b0efdfa | 346 | return (unsigned long long *) (objp + cachep->size - |
b46b8f19 | 347 | sizeof(unsigned long long)); |
1da177e4 LT |
348 | } |
349 | ||
343e0d7a | 350 | static void **dbg_userword(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
351 | { |
352 | BUG_ON(!(cachep->flags & SLAB_STORE_USER)); | |
3b0efdfa | 353 | return (void **)(objp + cachep->size - BYTES_PER_WORD); |
1da177e4 LT |
354 | } |
355 | ||
356 | #else | |
357 | ||
3dafccf2 | 358 | #define obj_offset(x) 0 |
b46b8f19 DW |
359 | #define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) |
360 | #define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;}) | |
1da177e4 LT |
361 | #define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;}) |
362 | ||
363 | #endif | |
364 | ||
03787301 JK |
365 | #ifdef CONFIG_DEBUG_SLAB_LEAK |
366 | ||
d31676df | 367 | static inline bool is_store_user_clean(struct kmem_cache *cachep) |
03787301 | 368 | { |
d31676df JK |
369 | return atomic_read(&cachep->store_user_clean) == 1; |
370 | } | |
03787301 | 371 | |
d31676df JK |
372 | static inline void set_store_user_clean(struct kmem_cache *cachep) |
373 | { | |
374 | atomic_set(&cachep->store_user_clean, 1); | |
375 | } | |
03787301 | 376 | |
d31676df JK |
377 | static inline void set_store_user_dirty(struct kmem_cache *cachep) |
378 | { | |
379 | if (is_store_user_clean(cachep)) | |
380 | atomic_set(&cachep->store_user_clean, 0); | |
03787301 JK |
381 | } |
382 | ||
383 | #else | |
d31676df | 384 | static inline void set_store_user_dirty(struct kmem_cache *cachep) {} |
03787301 JK |
385 | |
386 | #endif | |
387 | ||
1da177e4 | 388 | /* |
3df1cccd DR |
389 | * Do not go above this order unless 0 objects fit into the slab or |
390 | * overridden on the command line. | |
1da177e4 | 391 | */ |
543585cc DR |
392 | #define SLAB_MAX_ORDER_HI 1 |
393 | #define SLAB_MAX_ORDER_LO 0 | |
394 | static int slab_max_order = SLAB_MAX_ORDER_LO; | |
3df1cccd | 395 | static bool slab_max_order_set __initdata; |
1da177e4 | 396 | |
6ed5eb22 PE |
397 | static inline struct kmem_cache *virt_to_cache(const void *obj) |
398 | { | |
b49af68f | 399 | struct page *page = virt_to_head_page(obj); |
35026088 | 400 | return page->slab_cache; |
6ed5eb22 PE |
401 | } |
402 | ||
8456a648 | 403 | static inline void *index_to_obj(struct kmem_cache *cache, struct page *page, |
8fea4e96 PE |
404 | unsigned int idx) |
405 | { | |
8456a648 | 406 | return page->s_mem + cache->size * idx; |
8fea4e96 PE |
407 | } |
408 | ||
6a2d7a95 | 409 | /* |
3b0efdfa CL |
410 | * We want to avoid an expensive divide : (offset / cache->size) |
411 | * Using the fact that size is a constant for a particular cache, | |
412 | * we can replace (offset / cache->size) by | |
6a2d7a95 ED |
413 | * reciprocal_divide(offset, cache->reciprocal_buffer_size) |
414 | */ | |
415 | static inline unsigned int obj_to_index(const struct kmem_cache *cache, | |
8456a648 | 416 | const struct page *page, void *obj) |
8fea4e96 | 417 | { |
8456a648 | 418 | u32 offset = (obj - page->s_mem); |
6a2d7a95 | 419 | return reciprocal_divide(offset, cache->reciprocal_buffer_size); |
8fea4e96 PE |
420 | } |
421 | ||
6fb92430 | 422 | #define BOOT_CPUCACHE_ENTRIES 1 |
1da177e4 | 423 | /* internal cache of cache description objs */ |
9b030cb8 | 424 | static struct kmem_cache kmem_cache_boot = { |
b28a02de PE |
425 | .batchcount = 1, |
426 | .limit = BOOT_CPUCACHE_ENTRIES, | |
427 | .shared = 1, | |
3b0efdfa | 428 | .size = sizeof(struct kmem_cache), |
b28a02de | 429 | .name = "kmem_cache", |
1da177e4 LT |
430 | }; |
431 | ||
1871e52c | 432 | static DEFINE_PER_CPU(struct delayed_work, slab_reap_work); |
1da177e4 | 433 | |
343e0d7a | 434 | static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep) |
1da177e4 | 435 | { |
bf0dea23 | 436 | return this_cpu_ptr(cachep->cpu_cache); |
1da177e4 LT |
437 | } |
438 | ||
a737b3e2 AM |
439 | /* |
440 | * Calculate the number of objects and left-over bytes for a given buffer size. | |
441 | */ | |
70f75067 | 442 | static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size, |
d50112ed | 443 | slab_flags_t flags, size_t *left_over) |
fbaccacf | 444 | { |
70f75067 | 445 | unsigned int num; |
fbaccacf | 446 | size_t slab_size = PAGE_SIZE << gfporder; |
1da177e4 | 447 | |
fbaccacf SR |
448 | /* |
449 | * The slab management structure can be either off the slab or | |
450 | * on it. For the latter case, the memory allocated for a | |
451 | * slab is used for: | |
452 | * | |
fbaccacf | 453 | * - @buffer_size bytes for each object |
2e6b3602 JK |
454 | * - One freelist_idx_t for each object |
455 | * | |
456 | * We don't need to consider alignment of freelist because | |
457 | * freelist will be at the end of slab page. The objects will be | |
458 | * at the correct alignment. | |
fbaccacf SR |
459 | * |
460 | * If the slab management structure is off the slab, then the | |
461 | * alignment will already be calculated into the size. Because | |
462 | * the slabs are all pages aligned, the objects will be at the | |
463 | * correct alignment when allocated. | |
464 | */ | |
b03a017b | 465 | if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) { |
70f75067 | 466 | num = slab_size / buffer_size; |
2e6b3602 | 467 | *left_over = slab_size % buffer_size; |
fbaccacf | 468 | } else { |
70f75067 | 469 | num = slab_size / (buffer_size + sizeof(freelist_idx_t)); |
2e6b3602 JK |
470 | *left_over = slab_size % |
471 | (buffer_size + sizeof(freelist_idx_t)); | |
fbaccacf | 472 | } |
70f75067 JK |
473 | |
474 | return num; | |
1da177e4 LT |
475 | } |
476 | ||
f28510d3 | 477 | #if DEBUG |
d40cee24 | 478 | #define slab_error(cachep, msg) __slab_error(__func__, cachep, msg) |
1da177e4 | 479 | |
a737b3e2 AM |
480 | static void __slab_error(const char *function, struct kmem_cache *cachep, |
481 | char *msg) | |
1da177e4 | 482 | { |
1170532b | 483 | pr_err("slab error in %s(): cache `%s': %s\n", |
b28a02de | 484 | function, cachep->name, msg); |
1da177e4 | 485 | dump_stack(); |
373d4d09 | 486 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
1da177e4 | 487 | } |
f28510d3 | 488 | #endif |
1da177e4 | 489 | |
3395ee05 PM |
490 | /* |
491 | * By default on NUMA we use alien caches to stage the freeing of | |
492 | * objects allocated from other nodes. This causes massive memory | |
493 | * inefficiencies when using fake NUMA setup to split memory into a | |
494 | * large number of small nodes, so it can be disabled on the command | |
495 | * line | |
496 | */ | |
497 | ||
498 | static int use_alien_caches __read_mostly = 1; | |
499 | static int __init noaliencache_setup(char *s) | |
500 | { | |
501 | use_alien_caches = 0; | |
502 | return 1; | |
503 | } | |
504 | __setup("noaliencache", noaliencache_setup); | |
505 | ||
3df1cccd DR |
506 | static int __init slab_max_order_setup(char *str) |
507 | { | |
508 | get_option(&str, &slab_max_order); | |
509 | slab_max_order = slab_max_order < 0 ? 0 : | |
510 | min(slab_max_order, MAX_ORDER - 1); | |
511 | slab_max_order_set = true; | |
512 | ||
513 | return 1; | |
514 | } | |
515 | __setup("slab_max_order=", slab_max_order_setup); | |
516 | ||
8fce4d8e CL |
517 | #ifdef CONFIG_NUMA |
518 | /* | |
519 | * Special reaping functions for NUMA systems called from cache_reap(). | |
520 | * These take care of doing round robin flushing of alien caches (containing | |
521 | * objects freed on different nodes from which they were allocated) and the | |
522 | * flushing of remote pcps by calling drain_node_pages. | |
523 | */ | |
1871e52c | 524 | static DEFINE_PER_CPU(unsigned long, slab_reap_node); |
8fce4d8e CL |
525 | |
526 | static void init_reap_node(int cpu) | |
527 | { | |
0edaf86c AM |
528 | per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu), |
529 | node_online_map); | |
8fce4d8e CL |
530 | } |
531 | ||
532 | static void next_reap_node(void) | |
533 | { | |
909ea964 | 534 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 535 | |
0edaf86c | 536 | node = next_node_in(node, node_online_map); |
909ea964 | 537 | __this_cpu_write(slab_reap_node, node); |
8fce4d8e CL |
538 | } |
539 | ||
540 | #else | |
541 | #define init_reap_node(cpu) do { } while (0) | |
542 | #define next_reap_node(void) do { } while (0) | |
543 | #endif | |
544 | ||
1da177e4 LT |
545 | /* |
546 | * Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz | |
547 | * via the workqueue/eventd. | |
548 | * Add the CPU number into the expiration time to minimize the possibility of | |
549 | * the CPUs getting into lockstep and contending for the global cache chain | |
550 | * lock. | |
551 | */ | |
0db0628d | 552 | static void start_cpu_timer(int cpu) |
1da177e4 | 553 | { |
1871e52c | 554 | struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu); |
1da177e4 | 555 | |
eac0337a | 556 | if (reap_work->work.func == NULL) { |
8fce4d8e | 557 | init_reap_node(cpu); |
203b42f7 | 558 | INIT_DEFERRABLE_WORK(reap_work, cache_reap); |
2b284214 AV |
559 | schedule_delayed_work_on(cpu, reap_work, |
560 | __round_jiffies_relative(HZ, cpu)); | |
1da177e4 LT |
561 | } |
562 | } | |
563 | ||
1fe00d50 | 564 | static void init_arraycache(struct array_cache *ac, int limit, int batch) |
1da177e4 | 565 | { |
d5cff635 CM |
566 | /* |
567 | * The array_cache structures contain pointers to free object. | |
25985edc | 568 | * However, when such objects are allocated or transferred to another |
d5cff635 CM |
569 | * cache the pointers are not cleared and they could be counted as |
570 | * valid references during a kmemleak scan. Therefore, kmemleak must | |
571 | * not scan such objects. | |
572 | */ | |
1fe00d50 JK |
573 | kmemleak_no_scan(ac); |
574 | if (ac) { | |
575 | ac->avail = 0; | |
576 | ac->limit = limit; | |
577 | ac->batchcount = batch; | |
578 | ac->touched = 0; | |
1da177e4 | 579 | } |
1fe00d50 JK |
580 | } |
581 | ||
582 | static struct array_cache *alloc_arraycache(int node, int entries, | |
583 | int batchcount, gfp_t gfp) | |
584 | { | |
5e804789 | 585 | size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache); |
1fe00d50 JK |
586 | struct array_cache *ac = NULL; |
587 | ||
588 | ac = kmalloc_node(memsize, gfp, node); | |
589 | init_arraycache(ac, entries, batchcount); | |
590 | return ac; | |
1da177e4 LT |
591 | } |
592 | ||
f68f8ddd JK |
593 | static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep, |
594 | struct page *page, void *objp) | |
072bb0aa | 595 | { |
f68f8ddd JK |
596 | struct kmem_cache_node *n; |
597 | int page_node; | |
598 | LIST_HEAD(list); | |
072bb0aa | 599 | |
f68f8ddd JK |
600 | page_node = page_to_nid(page); |
601 | n = get_node(cachep, page_node); | |
381760ea | 602 | |
f68f8ddd JK |
603 | spin_lock(&n->list_lock); |
604 | free_block(cachep, &objp, 1, page_node, &list); | |
605 | spin_unlock(&n->list_lock); | |
381760ea | 606 | |
f68f8ddd | 607 | slabs_destroy(cachep, &list); |
072bb0aa MG |
608 | } |
609 | ||
3ded175a CL |
610 | /* |
611 | * Transfer objects in one arraycache to another. | |
612 | * Locking must be handled by the caller. | |
613 | * | |
614 | * Return the number of entries transferred. | |
615 | */ | |
616 | static int transfer_objects(struct array_cache *to, | |
617 | struct array_cache *from, unsigned int max) | |
618 | { | |
619 | /* Figure out how many entries to transfer */ | |
732eacc0 | 620 | int nr = min3(from->avail, max, to->limit - to->avail); |
3ded175a CL |
621 | |
622 | if (!nr) | |
623 | return 0; | |
624 | ||
625 | memcpy(to->entry + to->avail, from->entry + from->avail -nr, | |
626 | sizeof(void *) *nr); | |
627 | ||
628 | from->avail -= nr; | |
629 | to->avail += nr; | |
3ded175a CL |
630 | return nr; |
631 | } | |
632 | ||
765c4507 CL |
633 | #ifndef CONFIG_NUMA |
634 | ||
635 | #define drain_alien_cache(cachep, alien) do { } while (0) | |
ce8eb6c4 | 636 | #define reap_alien(cachep, n) do { } while (0) |
765c4507 | 637 | |
c8522a3a JK |
638 | static inline struct alien_cache **alloc_alien_cache(int node, |
639 | int limit, gfp_t gfp) | |
765c4507 | 640 | { |
8888177e | 641 | return NULL; |
765c4507 CL |
642 | } |
643 | ||
c8522a3a | 644 | static inline void free_alien_cache(struct alien_cache **ac_ptr) |
765c4507 CL |
645 | { |
646 | } | |
647 | ||
648 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
649 | { | |
650 | return 0; | |
651 | } | |
652 | ||
653 | static inline void *alternate_node_alloc(struct kmem_cache *cachep, | |
654 | gfp_t flags) | |
655 | { | |
656 | return NULL; | |
657 | } | |
658 | ||
8b98c169 | 659 | static inline void *____cache_alloc_node(struct kmem_cache *cachep, |
765c4507 CL |
660 | gfp_t flags, int nodeid) |
661 | { | |
662 | return NULL; | |
663 | } | |
664 | ||
4167e9b2 DR |
665 | static inline gfp_t gfp_exact_node(gfp_t flags) |
666 | { | |
444eb2a4 | 667 | return flags & ~__GFP_NOFAIL; |
4167e9b2 DR |
668 | } |
669 | ||
765c4507 CL |
670 | #else /* CONFIG_NUMA */ |
671 | ||
8b98c169 | 672 | static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int); |
c61afb18 | 673 | static void *alternate_node_alloc(struct kmem_cache *, gfp_t); |
dc85da15 | 674 | |
c8522a3a JK |
675 | static struct alien_cache *__alloc_alien_cache(int node, int entries, |
676 | int batch, gfp_t gfp) | |
677 | { | |
5e804789 | 678 | size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache); |
c8522a3a JK |
679 | struct alien_cache *alc = NULL; |
680 | ||
681 | alc = kmalloc_node(memsize, gfp, node); | |
682 | init_arraycache(&alc->ac, entries, batch); | |
49dfc304 | 683 | spin_lock_init(&alc->lock); |
c8522a3a JK |
684 | return alc; |
685 | } | |
686 | ||
687 | static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp) | |
e498be7d | 688 | { |
c8522a3a | 689 | struct alien_cache **alc_ptr; |
5e804789 | 690 | size_t memsize = sizeof(void *) * nr_node_ids; |
e498be7d CL |
691 | int i; |
692 | ||
693 | if (limit > 1) | |
694 | limit = 12; | |
c8522a3a JK |
695 | alc_ptr = kzalloc_node(memsize, gfp, node); |
696 | if (!alc_ptr) | |
697 | return NULL; | |
698 | ||
699 | for_each_node(i) { | |
700 | if (i == node || !node_online(i)) | |
701 | continue; | |
702 | alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp); | |
703 | if (!alc_ptr[i]) { | |
704 | for (i--; i >= 0; i--) | |
705 | kfree(alc_ptr[i]); | |
706 | kfree(alc_ptr); | |
707 | return NULL; | |
e498be7d CL |
708 | } |
709 | } | |
c8522a3a | 710 | return alc_ptr; |
e498be7d CL |
711 | } |
712 | ||
c8522a3a | 713 | static void free_alien_cache(struct alien_cache **alc_ptr) |
e498be7d CL |
714 | { |
715 | int i; | |
716 | ||
c8522a3a | 717 | if (!alc_ptr) |
e498be7d | 718 | return; |
e498be7d | 719 | for_each_node(i) |
c8522a3a JK |
720 | kfree(alc_ptr[i]); |
721 | kfree(alc_ptr); | |
e498be7d CL |
722 | } |
723 | ||
343e0d7a | 724 | static void __drain_alien_cache(struct kmem_cache *cachep, |
833b706c JK |
725 | struct array_cache *ac, int node, |
726 | struct list_head *list) | |
e498be7d | 727 | { |
18bf8541 | 728 | struct kmem_cache_node *n = get_node(cachep, node); |
e498be7d CL |
729 | |
730 | if (ac->avail) { | |
ce8eb6c4 | 731 | spin_lock(&n->list_lock); |
e00946fe CL |
732 | /* |
733 | * Stuff objects into the remote nodes shared array first. | |
734 | * That way we could avoid the overhead of putting the objects | |
735 | * into the free lists and getting them back later. | |
736 | */ | |
ce8eb6c4 CL |
737 | if (n->shared) |
738 | transfer_objects(n->shared, ac, ac->limit); | |
e00946fe | 739 | |
833b706c | 740 | free_block(cachep, ac->entry, ac->avail, node, list); |
e498be7d | 741 | ac->avail = 0; |
ce8eb6c4 | 742 | spin_unlock(&n->list_lock); |
e498be7d CL |
743 | } |
744 | } | |
745 | ||
8fce4d8e CL |
746 | /* |
747 | * Called from cache_reap() to regularly drain alien caches round robin. | |
748 | */ | |
ce8eb6c4 | 749 | static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n) |
8fce4d8e | 750 | { |
909ea964 | 751 | int node = __this_cpu_read(slab_reap_node); |
8fce4d8e | 752 | |
ce8eb6c4 | 753 | if (n->alien) { |
c8522a3a JK |
754 | struct alien_cache *alc = n->alien[node]; |
755 | struct array_cache *ac; | |
756 | ||
757 | if (alc) { | |
758 | ac = &alc->ac; | |
49dfc304 | 759 | if (ac->avail && spin_trylock_irq(&alc->lock)) { |
833b706c JK |
760 | LIST_HEAD(list); |
761 | ||
762 | __drain_alien_cache(cachep, ac, node, &list); | |
49dfc304 | 763 | spin_unlock_irq(&alc->lock); |
833b706c | 764 | slabs_destroy(cachep, &list); |
c8522a3a | 765 | } |
8fce4d8e CL |
766 | } |
767 | } | |
768 | } | |
769 | ||
a737b3e2 | 770 | static void drain_alien_cache(struct kmem_cache *cachep, |
c8522a3a | 771 | struct alien_cache **alien) |
e498be7d | 772 | { |
b28a02de | 773 | int i = 0; |
c8522a3a | 774 | struct alien_cache *alc; |
e498be7d CL |
775 | struct array_cache *ac; |
776 | unsigned long flags; | |
777 | ||
778 | for_each_online_node(i) { | |
c8522a3a JK |
779 | alc = alien[i]; |
780 | if (alc) { | |
833b706c JK |
781 | LIST_HEAD(list); |
782 | ||
c8522a3a | 783 | ac = &alc->ac; |
49dfc304 | 784 | spin_lock_irqsave(&alc->lock, flags); |
833b706c | 785 | __drain_alien_cache(cachep, ac, i, &list); |
49dfc304 | 786 | spin_unlock_irqrestore(&alc->lock, flags); |
833b706c | 787 | slabs_destroy(cachep, &list); |
e498be7d CL |
788 | } |
789 | } | |
790 | } | |
729bd0b7 | 791 | |
25c4f304 JK |
792 | static int __cache_free_alien(struct kmem_cache *cachep, void *objp, |
793 | int node, int page_node) | |
729bd0b7 | 794 | { |
ce8eb6c4 | 795 | struct kmem_cache_node *n; |
c8522a3a JK |
796 | struct alien_cache *alien = NULL; |
797 | struct array_cache *ac; | |
97654dfa | 798 | LIST_HEAD(list); |
1ca4cb24 | 799 | |
18bf8541 | 800 | n = get_node(cachep, node); |
729bd0b7 | 801 | STATS_INC_NODEFREES(cachep); |
25c4f304 JK |
802 | if (n->alien && n->alien[page_node]) { |
803 | alien = n->alien[page_node]; | |
c8522a3a | 804 | ac = &alien->ac; |
49dfc304 | 805 | spin_lock(&alien->lock); |
c8522a3a | 806 | if (unlikely(ac->avail == ac->limit)) { |
729bd0b7 | 807 | STATS_INC_ACOVERFLOW(cachep); |
25c4f304 | 808 | __drain_alien_cache(cachep, ac, page_node, &list); |
729bd0b7 | 809 | } |
f68f8ddd | 810 | ac->entry[ac->avail++] = objp; |
49dfc304 | 811 | spin_unlock(&alien->lock); |
833b706c | 812 | slabs_destroy(cachep, &list); |
729bd0b7 | 813 | } else { |
25c4f304 | 814 | n = get_node(cachep, page_node); |
18bf8541 | 815 | spin_lock(&n->list_lock); |
25c4f304 | 816 | free_block(cachep, &objp, 1, page_node, &list); |
18bf8541 | 817 | spin_unlock(&n->list_lock); |
97654dfa | 818 | slabs_destroy(cachep, &list); |
729bd0b7 PE |
819 | } |
820 | return 1; | |
821 | } | |
25c4f304 JK |
822 | |
823 | static inline int cache_free_alien(struct kmem_cache *cachep, void *objp) | |
824 | { | |
825 | int page_node = page_to_nid(virt_to_page(objp)); | |
826 | int node = numa_mem_id(); | |
827 | /* | |
828 | * Make sure we are not freeing a object from another node to the array | |
829 | * cache on this cpu. | |
830 | */ | |
831 | if (likely(node == page_node)) | |
832 | return 0; | |
833 | ||
834 | return __cache_free_alien(cachep, objp, node, page_node); | |
835 | } | |
4167e9b2 DR |
836 | |
837 | /* | |
444eb2a4 MG |
838 | * Construct gfp mask to allocate from a specific node but do not reclaim or |
839 | * warn about failures. | |
4167e9b2 DR |
840 | */ |
841 | static inline gfp_t gfp_exact_node(gfp_t flags) | |
842 | { | |
444eb2a4 | 843 | return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
4167e9b2 | 844 | } |
e498be7d CL |
845 | #endif |
846 | ||
ded0ecf6 JK |
847 | static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp) |
848 | { | |
849 | struct kmem_cache_node *n; | |
850 | ||
851 | /* | |
852 | * Set up the kmem_cache_node for cpu before we can | |
853 | * begin anything. Make sure some other cpu on this | |
854 | * node has not already allocated this | |
855 | */ | |
856 | n = get_node(cachep, node); | |
857 | if (n) { | |
858 | spin_lock_irq(&n->list_lock); | |
859 | n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount + | |
860 | cachep->num; | |
861 | spin_unlock_irq(&n->list_lock); | |
862 | ||
863 | return 0; | |
864 | } | |
865 | ||
866 | n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node); | |
867 | if (!n) | |
868 | return -ENOMEM; | |
869 | ||
870 | kmem_cache_node_init(n); | |
871 | n->next_reap = jiffies + REAPTIMEOUT_NODE + | |
872 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
873 | ||
874 | n->free_limit = | |
875 | (1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num; | |
876 | ||
877 | /* | |
878 | * The kmem_cache_nodes don't come and go as CPUs | |
879 | * come and go. slab_mutex is sufficient | |
880 | * protection here. | |
881 | */ | |
882 | cachep->node[node] = n; | |
883 | ||
884 | return 0; | |
885 | } | |
886 | ||
6731d4f1 | 887 | #if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP) |
8f9f8d9e | 888 | /* |
6a67368c | 889 | * Allocates and initializes node for a node on each slab cache, used for |
ce8eb6c4 | 890 | * either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node |
8f9f8d9e | 891 | * will be allocated off-node since memory is not yet online for the new node. |
6a67368c | 892 | * When hotplugging memory or a cpu, existing node are not replaced if |
8f9f8d9e DR |
893 | * already in use. |
894 | * | |
18004c5d | 895 | * Must hold slab_mutex. |
8f9f8d9e | 896 | */ |
6a67368c | 897 | static int init_cache_node_node(int node) |
8f9f8d9e | 898 | { |
ded0ecf6 | 899 | int ret; |
8f9f8d9e | 900 | struct kmem_cache *cachep; |
8f9f8d9e | 901 | |
18004c5d | 902 | list_for_each_entry(cachep, &slab_caches, list) { |
ded0ecf6 JK |
903 | ret = init_cache_node(cachep, node, GFP_KERNEL); |
904 | if (ret) | |
905 | return ret; | |
8f9f8d9e | 906 | } |
ded0ecf6 | 907 | |
8f9f8d9e DR |
908 | return 0; |
909 | } | |
6731d4f1 | 910 | #endif |
8f9f8d9e | 911 | |
c3d332b6 JK |
912 | static int setup_kmem_cache_node(struct kmem_cache *cachep, |
913 | int node, gfp_t gfp, bool force_change) | |
914 | { | |
915 | int ret = -ENOMEM; | |
916 | struct kmem_cache_node *n; | |
917 | struct array_cache *old_shared = NULL; | |
918 | struct array_cache *new_shared = NULL; | |
919 | struct alien_cache **new_alien = NULL; | |
920 | LIST_HEAD(list); | |
921 | ||
922 | if (use_alien_caches) { | |
923 | new_alien = alloc_alien_cache(node, cachep->limit, gfp); | |
924 | if (!new_alien) | |
925 | goto fail; | |
926 | } | |
927 | ||
928 | if (cachep->shared) { | |
929 | new_shared = alloc_arraycache(node, | |
930 | cachep->shared * cachep->batchcount, 0xbaadf00d, gfp); | |
931 | if (!new_shared) | |
932 | goto fail; | |
933 | } | |
934 | ||
935 | ret = init_cache_node(cachep, node, gfp); | |
936 | if (ret) | |
937 | goto fail; | |
938 | ||
939 | n = get_node(cachep, node); | |
940 | spin_lock_irq(&n->list_lock); | |
941 | if (n->shared && force_change) { | |
942 | free_block(cachep, n->shared->entry, | |
943 | n->shared->avail, node, &list); | |
944 | n->shared->avail = 0; | |
945 | } | |
946 | ||
947 | if (!n->shared || force_change) { | |
948 | old_shared = n->shared; | |
949 | n->shared = new_shared; | |
950 | new_shared = NULL; | |
951 | } | |
952 | ||
953 | if (!n->alien) { | |
954 | n->alien = new_alien; | |
955 | new_alien = NULL; | |
956 | } | |
957 | ||
958 | spin_unlock_irq(&n->list_lock); | |
959 | slabs_destroy(cachep, &list); | |
960 | ||
801faf0d JK |
961 | /* |
962 | * To protect lockless access to n->shared during irq disabled context. | |
963 | * If n->shared isn't NULL in irq disabled context, accessing to it is | |
964 | * guaranteed to be valid until irq is re-enabled, because it will be | |
6564a25e | 965 | * freed after synchronize_rcu(). |
801faf0d | 966 | */ |
86d9f485 | 967 | if (old_shared && force_change) |
6564a25e | 968 | synchronize_rcu(); |
801faf0d | 969 | |
c3d332b6 JK |
970 | fail: |
971 | kfree(old_shared); | |
972 | kfree(new_shared); | |
973 | free_alien_cache(new_alien); | |
974 | ||
975 | return ret; | |
976 | } | |
977 | ||
6731d4f1 SAS |
978 | #ifdef CONFIG_SMP |
979 | ||
0db0628d | 980 | static void cpuup_canceled(long cpu) |
fbf1e473 AM |
981 | { |
982 | struct kmem_cache *cachep; | |
ce8eb6c4 | 983 | struct kmem_cache_node *n = NULL; |
7d6e6d09 | 984 | int node = cpu_to_mem(cpu); |
a70f7302 | 985 | const struct cpumask *mask = cpumask_of_node(node); |
fbf1e473 | 986 | |
18004c5d | 987 | list_for_each_entry(cachep, &slab_caches, list) { |
fbf1e473 AM |
988 | struct array_cache *nc; |
989 | struct array_cache *shared; | |
c8522a3a | 990 | struct alien_cache **alien; |
97654dfa | 991 | LIST_HEAD(list); |
fbf1e473 | 992 | |
18bf8541 | 993 | n = get_node(cachep, node); |
ce8eb6c4 | 994 | if (!n) |
bf0dea23 | 995 | continue; |
fbf1e473 | 996 | |
ce8eb6c4 | 997 | spin_lock_irq(&n->list_lock); |
fbf1e473 | 998 | |
ce8eb6c4 CL |
999 | /* Free limit for this kmem_cache_node */ |
1000 | n->free_limit -= cachep->batchcount; | |
bf0dea23 JK |
1001 | |
1002 | /* cpu is dead; no one can alloc from it. */ | |
1003 | nc = per_cpu_ptr(cachep->cpu_cache, cpu); | |
1004 | if (nc) { | |
97654dfa | 1005 | free_block(cachep, nc->entry, nc->avail, node, &list); |
bf0dea23 JK |
1006 | nc->avail = 0; |
1007 | } | |
fbf1e473 | 1008 | |
58463c1f | 1009 | if (!cpumask_empty(mask)) { |
ce8eb6c4 | 1010 | spin_unlock_irq(&n->list_lock); |
bf0dea23 | 1011 | goto free_slab; |
fbf1e473 AM |
1012 | } |
1013 | ||
ce8eb6c4 | 1014 | shared = n->shared; |
fbf1e473 AM |
1015 | if (shared) { |
1016 | free_block(cachep, shared->entry, | |
97654dfa | 1017 | shared->avail, node, &list); |
ce8eb6c4 | 1018 | n->shared = NULL; |
fbf1e473 AM |
1019 | } |
1020 | ||
ce8eb6c4 CL |
1021 | alien = n->alien; |
1022 | n->alien = NULL; | |
fbf1e473 | 1023 | |
ce8eb6c4 | 1024 | spin_unlock_irq(&n->list_lock); |
fbf1e473 AM |
1025 | |
1026 | kfree(shared); | |
1027 | if (alien) { | |
1028 | drain_alien_cache(cachep, alien); | |
1029 | free_alien_cache(alien); | |
1030 | } | |
bf0dea23 JK |
1031 | |
1032 | free_slab: | |
97654dfa | 1033 | slabs_destroy(cachep, &list); |
fbf1e473 AM |
1034 | } |
1035 | /* | |
1036 | * In the previous loop, all the objects were freed to | |
1037 | * the respective cache's slabs, now we can go ahead and | |
1038 | * shrink each nodelist to its limit. | |
1039 | */ | |
18004c5d | 1040 | list_for_each_entry(cachep, &slab_caches, list) { |
18bf8541 | 1041 | n = get_node(cachep, node); |
ce8eb6c4 | 1042 | if (!n) |
fbf1e473 | 1043 | continue; |
a5aa63a5 | 1044 | drain_freelist(cachep, n, INT_MAX); |
fbf1e473 AM |
1045 | } |
1046 | } | |
1047 | ||
0db0628d | 1048 | static int cpuup_prepare(long cpu) |
1da177e4 | 1049 | { |
343e0d7a | 1050 | struct kmem_cache *cachep; |
7d6e6d09 | 1051 | int node = cpu_to_mem(cpu); |
8f9f8d9e | 1052 | int err; |
1da177e4 | 1053 | |
fbf1e473 AM |
1054 | /* |
1055 | * We need to do this right in the beginning since | |
1056 | * alloc_arraycache's are going to use this list. | |
1057 | * kmalloc_node allows us to add the slab to the right | |
ce8eb6c4 | 1058 | * kmem_cache_node and not this cpu's kmem_cache_node |
fbf1e473 | 1059 | */ |
6a67368c | 1060 | err = init_cache_node_node(node); |
8f9f8d9e DR |
1061 | if (err < 0) |
1062 | goto bad; | |
fbf1e473 AM |
1063 | |
1064 | /* | |
1065 | * Now we can go ahead with allocating the shared arrays and | |
1066 | * array caches | |
1067 | */ | |
18004c5d | 1068 | list_for_each_entry(cachep, &slab_caches, list) { |
c3d332b6 JK |
1069 | err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false); |
1070 | if (err) | |
1071 | goto bad; | |
fbf1e473 | 1072 | } |
ce79ddc8 | 1073 | |
fbf1e473 AM |
1074 | return 0; |
1075 | bad: | |
12d00f6a | 1076 | cpuup_canceled(cpu); |
fbf1e473 AM |
1077 | return -ENOMEM; |
1078 | } | |
1079 | ||
6731d4f1 | 1080 | int slab_prepare_cpu(unsigned int cpu) |
fbf1e473 | 1081 | { |
6731d4f1 | 1082 | int err; |
fbf1e473 | 1083 | |
6731d4f1 SAS |
1084 | mutex_lock(&slab_mutex); |
1085 | err = cpuup_prepare(cpu); | |
1086 | mutex_unlock(&slab_mutex); | |
1087 | return err; | |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * This is called for a failed online attempt and for a successful | |
1092 | * offline. | |
1093 | * | |
1094 | * Even if all the cpus of a node are down, we don't free the | |
1095 | * kmem_list3 of any cache. This to avoid a race between cpu_down, and | |
1096 | * a kmalloc allocation from another cpu for memory from the node of | |
1097 | * the cpu going down. The list3 structure is usually allocated from | |
1098 | * kmem_cache_create() and gets destroyed at kmem_cache_destroy(). | |
1099 | */ | |
1100 | int slab_dead_cpu(unsigned int cpu) | |
1101 | { | |
1102 | mutex_lock(&slab_mutex); | |
1103 | cpuup_canceled(cpu); | |
1104 | mutex_unlock(&slab_mutex); | |
1105 | return 0; | |
1106 | } | |
8f5be20b | 1107 | #endif |
6731d4f1 SAS |
1108 | |
1109 | static int slab_online_cpu(unsigned int cpu) | |
1110 | { | |
1111 | start_cpu_timer(cpu); | |
1112 | return 0; | |
1da177e4 LT |
1113 | } |
1114 | ||
6731d4f1 SAS |
1115 | static int slab_offline_cpu(unsigned int cpu) |
1116 | { | |
1117 | /* | |
1118 | * Shutdown cache reaper. Note that the slab_mutex is held so | |
1119 | * that if cache_reap() is invoked it cannot do anything | |
1120 | * expensive but will only modify reap_work and reschedule the | |
1121 | * timer. | |
1122 | */ | |
1123 | cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu)); | |
1124 | /* Now the cache_reaper is guaranteed to be not running. */ | |
1125 | per_cpu(slab_reap_work, cpu).work.func = NULL; | |
1126 | return 0; | |
1127 | } | |
1da177e4 | 1128 | |
8f9f8d9e DR |
1129 | #if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG) |
1130 | /* | |
1131 | * Drains freelist for a node on each slab cache, used for memory hot-remove. | |
1132 | * Returns -EBUSY if all objects cannot be drained so that the node is not | |
1133 | * removed. | |
1134 | * | |
18004c5d | 1135 | * Must hold slab_mutex. |
8f9f8d9e | 1136 | */ |
6a67368c | 1137 | static int __meminit drain_cache_node_node(int node) |
8f9f8d9e DR |
1138 | { |
1139 | struct kmem_cache *cachep; | |
1140 | int ret = 0; | |
1141 | ||
18004c5d | 1142 | list_for_each_entry(cachep, &slab_caches, list) { |
ce8eb6c4 | 1143 | struct kmem_cache_node *n; |
8f9f8d9e | 1144 | |
18bf8541 | 1145 | n = get_node(cachep, node); |
ce8eb6c4 | 1146 | if (!n) |
8f9f8d9e DR |
1147 | continue; |
1148 | ||
a5aa63a5 | 1149 | drain_freelist(cachep, n, INT_MAX); |
8f9f8d9e | 1150 | |
ce8eb6c4 CL |
1151 | if (!list_empty(&n->slabs_full) || |
1152 | !list_empty(&n->slabs_partial)) { | |
8f9f8d9e DR |
1153 | ret = -EBUSY; |
1154 | break; | |
1155 | } | |
1156 | } | |
1157 | return ret; | |
1158 | } | |
1159 | ||
1160 | static int __meminit slab_memory_callback(struct notifier_block *self, | |
1161 | unsigned long action, void *arg) | |
1162 | { | |
1163 | struct memory_notify *mnb = arg; | |
1164 | int ret = 0; | |
1165 | int nid; | |
1166 | ||
1167 | nid = mnb->status_change_nid; | |
1168 | if (nid < 0) | |
1169 | goto out; | |
1170 | ||
1171 | switch (action) { | |
1172 | case MEM_GOING_ONLINE: | |
18004c5d | 1173 | mutex_lock(&slab_mutex); |
6a67368c | 1174 | ret = init_cache_node_node(nid); |
18004c5d | 1175 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1176 | break; |
1177 | case MEM_GOING_OFFLINE: | |
18004c5d | 1178 | mutex_lock(&slab_mutex); |
6a67368c | 1179 | ret = drain_cache_node_node(nid); |
18004c5d | 1180 | mutex_unlock(&slab_mutex); |
8f9f8d9e DR |
1181 | break; |
1182 | case MEM_ONLINE: | |
1183 | case MEM_OFFLINE: | |
1184 | case MEM_CANCEL_ONLINE: | |
1185 | case MEM_CANCEL_OFFLINE: | |
1186 | break; | |
1187 | } | |
1188 | out: | |
5fda1bd5 | 1189 | return notifier_from_errno(ret); |
8f9f8d9e DR |
1190 | } |
1191 | #endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */ | |
1192 | ||
e498be7d | 1193 | /* |
ce8eb6c4 | 1194 | * swap the static kmem_cache_node with kmalloced memory |
e498be7d | 1195 | */ |
6744f087 | 1196 | static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list, |
8f9f8d9e | 1197 | int nodeid) |
e498be7d | 1198 | { |
6744f087 | 1199 | struct kmem_cache_node *ptr; |
e498be7d | 1200 | |
6744f087 | 1201 | ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid); |
e498be7d CL |
1202 | BUG_ON(!ptr); |
1203 | ||
6744f087 | 1204 | memcpy(ptr, list, sizeof(struct kmem_cache_node)); |
2b2d5493 IM |
1205 | /* |
1206 | * Do not assume that spinlocks can be initialized via memcpy: | |
1207 | */ | |
1208 | spin_lock_init(&ptr->list_lock); | |
1209 | ||
e498be7d | 1210 | MAKE_ALL_LISTS(cachep, ptr, nodeid); |
6a67368c | 1211 | cachep->node[nodeid] = ptr; |
e498be7d CL |
1212 | } |
1213 | ||
556a169d | 1214 | /* |
ce8eb6c4 CL |
1215 | * For setting up all the kmem_cache_node for cache whose buffer_size is same as |
1216 | * size of kmem_cache_node. | |
556a169d | 1217 | */ |
ce8eb6c4 | 1218 | static void __init set_up_node(struct kmem_cache *cachep, int index) |
556a169d PE |
1219 | { |
1220 | int node; | |
1221 | ||
1222 | for_each_online_node(node) { | |
ce8eb6c4 | 1223 | cachep->node[node] = &init_kmem_cache_node[index + node]; |
6a67368c | 1224 | cachep->node[node]->next_reap = jiffies + |
5f0985bb JZ |
1225 | REAPTIMEOUT_NODE + |
1226 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
556a169d PE |
1227 | } |
1228 | } | |
1229 | ||
a737b3e2 AM |
1230 | /* |
1231 | * Initialisation. Called after the page allocator have been initialised and | |
1232 | * before smp_init(). | |
1da177e4 LT |
1233 | */ |
1234 | void __init kmem_cache_init(void) | |
1235 | { | |
e498be7d CL |
1236 | int i; |
1237 | ||
9b030cb8 CL |
1238 | kmem_cache = &kmem_cache_boot; |
1239 | ||
8888177e | 1240 | if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1) |
62918a03 SS |
1241 | use_alien_caches = 0; |
1242 | ||
3c583465 | 1243 | for (i = 0; i < NUM_INIT_LISTS; i++) |
ce8eb6c4 | 1244 | kmem_cache_node_init(&init_kmem_cache_node[i]); |
3c583465 | 1245 | |
1da177e4 LT |
1246 | /* |
1247 | * Fragmentation resistance on low memory - only use bigger | |
3df1cccd DR |
1248 | * page orders on machines with more than 32MB of memory if |
1249 | * not overridden on the command line. | |
1da177e4 | 1250 | */ |
3df1cccd | 1251 | if (!slab_max_order_set && totalram_pages > (32 << 20) >> PAGE_SHIFT) |
543585cc | 1252 | slab_max_order = SLAB_MAX_ORDER_HI; |
1da177e4 | 1253 | |
1da177e4 LT |
1254 | /* Bootstrap is tricky, because several objects are allocated |
1255 | * from caches that do not exist yet: | |
9b030cb8 CL |
1256 | * 1) initialize the kmem_cache cache: it contains the struct |
1257 | * kmem_cache structures of all caches, except kmem_cache itself: | |
1258 | * kmem_cache is statically allocated. | |
e498be7d | 1259 | * Initially an __init data area is used for the head array and the |
ce8eb6c4 | 1260 | * kmem_cache_node structures, it's replaced with a kmalloc allocated |
e498be7d | 1261 | * array at the end of the bootstrap. |
1da177e4 | 1262 | * 2) Create the first kmalloc cache. |
343e0d7a | 1263 | * The struct kmem_cache for the new cache is allocated normally. |
e498be7d CL |
1264 | * An __init data area is used for the head array. |
1265 | * 3) Create the remaining kmalloc caches, with minimally sized | |
1266 | * head arrays. | |
9b030cb8 | 1267 | * 4) Replace the __init data head arrays for kmem_cache and the first |
1da177e4 | 1268 | * kmalloc cache with kmalloc allocated arrays. |
ce8eb6c4 | 1269 | * 5) Replace the __init data for kmem_cache_node for kmem_cache and |
e498be7d CL |
1270 | * the other cache's with kmalloc allocated memory. |
1271 | * 6) Resize the head arrays of the kmalloc caches to their final sizes. | |
1da177e4 LT |
1272 | */ |
1273 | ||
9b030cb8 | 1274 | /* 1) create the kmem_cache */ |
1da177e4 | 1275 | |
8da3430d | 1276 | /* |
b56efcf0 | 1277 | * struct kmem_cache size depends on nr_node_ids & nr_cpu_ids |
8da3430d | 1278 | */ |
2f9baa9f | 1279 | create_boot_cache(kmem_cache, "kmem_cache", |
bf0dea23 | 1280 | offsetof(struct kmem_cache, node) + |
6744f087 | 1281 | nr_node_ids * sizeof(struct kmem_cache_node *), |
8eb8284b | 1282 | SLAB_HWCACHE_ALIGN, 0, 0); |
2f9baa9f | 1283 | list_add(&kmem_cache->list, &slab_caches); |
880cd276 | 1284 | memcg_link_cache(kmem_cache); |
bf0dea23 | 1285 | slab_state = PARTIAL; |
1da177e4 | 1286 | |
a737b3e2 | 1287 | /* |
bf0dea23 JK |
1288 | * Initialize the caches that provide memory for the kmem_cache_node |
1289 | * structures first. Without this, further allocations will bug. | |
e498be7d | 1290 | */ |
cc252eae | 1291 | kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE] = create_kmalloc_cache( |
af3b5f87 | 1292 | kmalloc_info[INDEX_NODE].name, |
6c0c21ad DW |
1293 | kmalloc_size(INDEX_NODE), ARCH_KMALLOC_FLAGS, |
1294 | 0, kmalloc_size(INDEX_NODE)); | |
bf0dea23 | 1295 | slab_state = PARTIAL_NODE; |
34cc6990 | 1296 | setup_kmalloc_cache_index_table(); |
e498be7d | 1297 | |
e0a42726 IM |
1298 | slab_early_init = 0; |
1299 | ||
ce8eb6c4 | 1300 | /* 5) Replace the bootstrap kmem_cache_node */ |
e498be7d | 1301 | { |
1ca4cb24 PE |
1302 | int nid; |
1303 | ||
9c09a95c | 1304 | for_each_online_node(nid) { |
ce8eb6c4 | 1305 | init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid); |
556a169d | 1306 | |
cc252eae | 1307 | init_list(kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE], |
ce8eb6c4 | 1308 | &init_kmem_cache_node[SIZE_NODE + nid], nid); |
e498be7d CL |
1309 | } |
1310 | } | |
1da177e4 | 1311 | |
f97d5f63 | 1312 | create_kmalloc_caches(ARCH_KMALLOC_FLAGS); |
8429db5c PE |
1313 | } |
1314 | ||
1315 | void __init kmem_cache_init_late(void) | |
1316 | { | |
1317 | struct kmem_cache *cachep; | |
1318 | ||
8429db5c | 1319 | /* 6) resize the head arrays to their final sizes */ |
18004c5d CL |
1320 | mutex_lock(&slab_mutex); |
1321 | list_for_each_entry(cachep, &slab_caches, list) | |
8429db5c PE |
1322 | if (enable_cpucache(cachep, GFP_NOWAIT)) |
1323 | BUG(); | |
18004c5d | 1324 | mutex_unlock(&slab_mutex); |
056c6241 | 1325 | |
97d06609 CL |
1326 | /* Done! */ |
1327 | slab_state = FULL; | |
1328 | ||
8f9f8d9e DR |
1329 | #ifdef CONFIG_NUMA |
1330 | /* | |
1331 | * Register a memory hotplug callback that initializes and frees | |
6a67368c | 1332 | * node. |
8f9f8d9e DR |
1333 | */ |
1334 | hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI); | |
1335 | #endif | |
1336 | ||
a737b3e2 AM |
1337 | /* |
1338 | * The reap timers are started later, with a module init call: That part | |
1339 | * of the kernel is not yet operational. | |
1da177e4 LT |
1340 | */ |
1341 | } | |
1342 | ||
1343 | static int __init cpucache_init(void) | |
1344 | { | |
6731d4f1 | 1345 | int ret; |
1da177e4 | 1346 | |
a737b3e2 AM |
1347 | /* |
1348 | * Register the timers that return unneeded pages to the page allocator | |
1da177e4 | 1349 | */ |
6731d4f1 SAS |
1350 | ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online", |
1351 | slab_online_cpu, slab_offline_cpu); | |
1352 | WARN_ON(ret < 0); | |
a164f896 | 1353 | |
1da177e4 LT |
1354 | return 0; |
1355 | } | |
1da177e4 LT |
1356 | __initcall(cpucache_init); |
1357 | ||
8bdec192 RA |
1358 | static noinline void |
1359 | slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid) | |
1360 | { | |
9a02d699 | 1361 | #if DEBUG |
ce8eb6c4 | 1362 | struct kmem_cache_node *n; |
8bdec192 RA |
1363 | unsigned long flags; |
1364 | int node; | |
9a02d699 DR |
1365 | static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
1366 | DEFAULT_RATELIMIT_BURST); | |
1367 | ||
1368 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs)) | |
1369 | return; | |
8bdec192 | 1370 | |
5b3810e5 VB |
1371 | pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
1372 | nodeid, gfpflags, &gfpflags); | |
1373 | pr_warn(" cache: %s, object size: %d, order: %d\n", | |
3b0efdfa | 1374 | cachep->name, cachep->size, cachep->gfporder); |
8bdec192 | 1375 | |
18bf8541 | 1376 | for_each_kmem_cache_node(cachep, node, n) { |
bf00bd34 | 1377 | unsigned long total_slabs, free_slabs, free_objs; |
8bdec192 | 1378 | |
ce8eb6c4 | 1379 | spin_lock_irqsave(&n->list_lock, flags); |
bf00bd34 DR |
1380 | total_slabs = n->total_slabs; |
1381 | free_slabs = n->free_slabs; | |
1382 | free_objs = n->free_objects; | |
ce8eb6c4 | 1383 | spin_unlock_irqrestore(&n->list_lock, flags); |
8bdec192 | 1384 | |
bf00bd34 DR |
1385 | pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n", |
1386 | node, total_slabs - free_slabs, total_slabs, | |
1387 | (total_slabs * cachep->num) - free_objs, | |
1388 | total_slabs * cachep->num); | |
8bdec192 | 1389 | } |
9a02d699 | 1390 | #endif |
8bdec192 RA |
1391 | } |
1392 | ||
1da177e4 | 1393 | /* |
8a7d9b43 WSH |
1394 | * Interface to system's page allocator. No need to hold the |
1395 | * kmem_cache_node ->list_lock. | |
1da177e4 LT |
1396 | * |
1397 | * If we requested dmaable memory, we will get it. Even if we | |
1398 | * did not request dmaable memory, we might get it, but that | |
1399 | * would be relatively rare and ignorable. | |
1400 | */ | |
0c3aa83e JK |
1401 | static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags, |
1402 | int nodeid) | |
1da177e4 LT |
1403 | { |
1404 | struct page *page; | |
e1b6aa6f | 1405 | int nr_pages; |
765c4507 | 1406 | |
a618e89f | 1407 | flags |= cachep->allocflags; |
e1b6aa6f | 1408 | |
75f296d9 | 1409 | page = __alloc_pages_node(nodeid, flags, cachep->gfporder); |
8bdec192 | 1410 | if (!page) { |
9a02d699 | 1411 | slab_out_of_memory(cachep, flags, nodeid); |
1da177e4 | 1412 | return NULL; |
8bdec192 | 1413 | } |
1da177e4 | 1414 | |
f3ccb2c4 VD |
1415 | if (memcg_charge_slab(page, flags, cachep->gfporder, cachep)) { |
1416 | __free_pages(page, cachep->gfporder); | |
1417 | return NULL; | |
1418 | } | |
1419 | ||
e1b6aa6f | 1420 | nr_pages = (1 << cachep->gfporder); |
1da177e4 | 1421 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
7779f212 | 1422 | mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, nr_pages); |
972d1a7b | 1423 | else |
7779f212 | 1424 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, nr_pages); |
f68f8ddd | 1425 | |
a57a4988 | 1426 | __SetPageSlab(page); |
f68f8ddd JK |
1427 | /* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */ |
1428 | if (sk_memalloc_socks() && page_is_pfmemalloc(page)) | |
a57a4988 | 1429 | SetPageSlabPfmemalloc(page); |
072bb0aa | 1430 | |
0c3aa83e | 1431 | return page; |
1da177e4 LT |
1432 | } |
1433 | ||
1434 | /* | |
1435 | * Interface to system's page release. | |
1436 | */ | |
0c3aa83e | 1437 | static void kmem_freepages(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 1438 | { |
27ee57c9 VD |
1439 | int order = cachep->gfporder; |
1440 | unsigned long nr_freed = (1 << order); | |
1da177e4 | 1441 | |
972d1a7b | 1442 | if (cachep->flags & SLAB_RECLAIM_ACCOUNT) |
7779f212 | 1443 | mod_lruvec_page_state(page, NR_SLAB_RECLAIMABLE, -nr_freed); |
972d1a7b | 1444 | else |
7779f212 | 1445 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE, -nr_freed); |
73293c2f | 1446 | |
a57a4988 | 1447 | BUG_ON(!PageSlab(page)); |
73293c2f | 1448 | __ClearPageSlabPfmemalloc(page); |
a57a4988 | 1449 | __ClearPageSlab(page); |
8456a648 JK |
1450 | page_mapcount_reset(page); |
1451 | page->mapping = NULL; | |
1f458cbf | 1452 | |
1da177e4 LT |
1453 | if (current->reclaim_state) |
1454 | current->reclaim_state->reclaimed_slab += nr_freed; | |
27ee57c9 VD |
1455 | memcg_uncharge_slab(page, order, cachep); |
1456 | __free_pages(page, order); | |
1da177e4 LT |
1457 | } |
1458 | ||
1459 | static void kmem_rcu_free(struct rcu_head *head) | |
1460 | { | |
68126702 JK |
1461 | struct kmem_cache *cachep; |
1462 | struct page *page; | |
1da177e4 | 1463 | |
68126702 JK |
1464 | page = container_of(head, struct page, rcu_head); |
1465 | cachep = page->slab_cache; | |
1466 | ||
1467 | kmem_freepages(cachep, page); | |
1da177e4 LT |
1468 | } |
1469 | ||
1470 | #if DEBUG | |
40b44137 JK |
1471 | static bool is_debug_pagealloc_cache(struct kmem_cache *cachep) |
1472 | { | |
1473 | if (debug_pagealloc_enabled() && OFF_SLAB(cachep) && | |
1474 | (cachep->size % PAGE_SIZE) == 0) | |
1475 | return true; | |
1476 | ||
1477 | return false; | |
1478 | } | |
1da177e4 LT |
1479 | |
1480 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
343e0d7a | 1481 | static void store_stackinfo(struct kmem_cache *cachep, unsigned long *addr, |
b28a02de | 1482 | unsigned long caller) |
1da177e4 | 1483 | { |
8c138bc0 | 1484 | int size = cachep->object_size; |
1da177e4 | 1485 | |
3dafccf2 | 1486 | addr = (unsigned long *)&((char *)addr)[obj_offset(cachep)]; |
1da177e4 | 1487 | |
b28a02de | 1488 | if (size < 5 * sizeof(unsigned long)) |
1da177e4 LT |
1489 | return; |
1490 | ||
b28a02de PE |
1491 | *addr++ = 0x12345678; |
1492 | *addr++ = caller; | |
1493 | *addr++ = smp_processor_id(); | |
1494 | size -= 3 * sizeof(unsigned long); | |
1da177e4 LT |
1495 | { |
1496 | unsigned long *sptr = &caller; | |
1497 | unsigned long svalue; | |
1498 | ||
1499 | while (!kstack_end(sptr)) { | |
1500 | svalue = *sptr++; | |
1501 | if (kernel_text_address(svalue)) { | |
b28a02de | 1502 | *addr++ = svalue; |
1da177e4 LT |
1503 | size -= sizeof(unsigned long); |
1504 | if (size <= sizeof(unsigned long)) | |
1505 | break; | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | } | |
b28a02de | 1510 | *addr++ = 0x87654321; |
1da177e4 | 1511 | } |
40b44137 JK |
1512 | |
1513 | static void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1514 | int map, unsigned long caller) | |
1515 | { | |
1516 | if (!is_debug_pagealloc_cache(cachep)) | |
1517 | return; | |
1518 | ||
1519 | if (caller) | |
1520 | store_stackinfo(cachep, objp, caller); | |
1521 | ||
1522 | kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map); | |
1523 | } | |
1524 | ||
1525 | #else | |
1526 | static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp, | |
1527 | int map, unsigned long caller) {} | |
1528 | ||
1da177e4 LT |
1529 | #endif |
1530 | ||
343e0d7a | 1531 | static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val) |
1da177e4 | 1532 | { |
8c138bc0 | 1533 | int size = cachep->object_size; |
3dafccf2 | 1534 | addr = &((char *)addr)[obj_offset(cachep)]; |
1da177e4 LT |
1535 | |
1536 | memset(addr, val, size); | |
b28a02de | 1537 | *(unsigned char *)(addr + size - 1) = POISON_END; |
1da177e4 LT |
1538 | } |
1539 | ||
1540 | static void dump_line(char *data, int offset, int limit) | |
1541 | { | |
1542 | int i; | |
aa83aa40 DJ |
1543 | unsigned char error = 0; |
1544 | int bad_count = 0; | |
1545 | ||
1170532b | 1546 | pr_err("%03x: ", offset); |
aa83aa40 DJ |
1547 | for (i = 0; i < limit; i++) { |
1548 | if (data[offset + i] != POISON_FREE) { | |
1549 | error = data[offset + i]; | |
1550 | bad_count++; | |
1551 | } | |
aa83aa40 | 1552 | } |
fdde6abb SAS |
1553 | print_hex_dump(KERN_CONT, "", 0, 16, 1, |
1554 | &data[offset], limit, 1); | |
aa83aa40 DJ |
1555 | |
1556 | if (bad_count == 1) { | |
1557 | error ^= POISON_FREE; | |
1558 | if (!(error & (error - 1))) { | |
1170532b | 1559 | pr_err("Single bit error detected. Probably bad RAM.\n"); |
aa83aa40 | 1560 | #ifdef CONFIG_X86 |
1170532b | 1561 | pr_err("Run memtest86+ or a similar memory test tool.\n"); |
aa83aa40 | 1562 | #else |
1170532b | 1563 | pr_err("Run a memory test tool.\n"); |
aa83aa40 DJ |
1564 | #endif |
1565 | } | |
1566 | } | |
1da177e4 LT |
1567 | } |
1568 | #endif | |
1569 | ||
1570 | #if DEBUG | |
1571 | ||
343e0d7a | 1572 | static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines) |
1da177e4 LT |
1573 | { |
1574 | int i, size; | |
1575 | char *realobj; | |
1576 | ||
1577 | if (cachep->flags & SLAB_RED_ZONE) { | |
1170532b JP |
1578 | pr_err("Redzone: 0x%llx/0x%llx\n", |
1579 | *dbg_redzone1(cachep, objp), | |
1580 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
1581 | } |
1582 | ||
85c3e4a5 GU |
1583 | if (cachep->flags & SLAB_STORE_USER) |
1584 | pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp)); | |
3dafccf2 | 1585 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1586 | size = cachep->object_size; |
b28a02de | 1587 | for (i = 0; i < size && lines; i += 16, lines--) { |
1da177e4 LT |
1588 | int limit; |
1589 | limit = 16; | |
b28a02de PE |
1590 | if (i + limit > size) |
1591 | limit = size - i; | |
1da177e4 LT |
1592 | dump_line(realobj, i, limit); |
1593 | } | |
1594 | } | |
1595 | ||
343e0d7a | 1596 | static void check_poison_obj(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
1597 | { |
1598 | char *realobj; | |
1599 | int size, i; | |
1600 | int lines = 0; | |
1601 | ||
40b44137 JK |
1602 | if (is_debug_pagealloc_cache(cachep)) |
1603 | return; | |
1604 | ||
3dafccf2 | 1605 | realobj = (char *)objp + obj_offset(cachep); |
8c138bc0 | 1606 | size = cachep->object_size; |
1da177e4 | 1607 | |
b28a02de | 1608 | for (i = 0; i < size; i++) { |
1da177e4 | 1609 | char exp = POISON_FREE; |
b28a02de | 1610 | if (i == size - 1) |
1da177e4 LT |
1611 | exp = POISON_END; |
1612 | if (realobj[i] != exp) { | |
1613 | int limit; | |
1614 | /* Mismatch ! */ | |
1615 | /* Print header */ | |
1616 | if (lines == 0) { | |
85c3e4a5 | 1617 | pr_err("Slab corruption (%s): %s start=%px, len=%d\n", |
1170532b JP |
1618 | print_tainted(), cachep->name, |
1619 | realobj, size); | |
1da177e4 LT |
1620 | print_objinfo(cachep, objp, 0); |
1621 | } | |
1622 | /* Hexdump the affected line */ | |
b28a02de | 1623 | i = (i / 16) * 16; |
1da177e4 | 1624 | limit = 16; |
b28a02de PE |
1625 | if (i + limit > size) |
1626 | limit = size - i; | |
1da177e4 LT |
1627 | dump_line(realobj, i, limit); |
1628 | i += 16; | |
1629 | lines++; | |
1630 | /* Limit to 5 lines */ | |
1631 | if (lines > 5) | |
1632 | break; | |
1633 | } | |
1634 | } | |
1635 | if (lines != 0) { | |
1636 | /* Print some data about the neighboring objects, if they | |
1637 | * exist: | |
1638 | */ | |
8456a648 | 1639 | struct page *page = virt_to_head_page(objp); |
8fea4e96 | 1640 | unsigned int objnr; |
1da177e4 | 1641 | |
8456a648 | 1642 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 | 1643 | if (objnr) { |
8456a648 | 1644 | objp = index_to_obj(cachep, page, objnr - 1); |
3dafccf2 | 1645 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1646 | pr_err("Prev obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1647 | print_objinfo(cachep, objp, 2); |
1648 | } | |
b28a02de | 1649 | if (objnr + 1 < cachep->num) { |
8456a648 | 1650 | objp = index_to_obj(cachep, page, objnr + 1); |
3dafccf2 | 1651 | realobj = (char *)objp + obj_offset(cachep); |
85c3e4a5 | 1652 | pr_err("Next obj: start=%px, len=%d\n", realobj, size); |
1da177e4 LT |
1653 | print_objinfo(cachep, objp, 2); |
1654 | } | |
1655 | } | |
1656 | } | |
1657 | #endif | |
1658 | ||
12dd36fa | 1659 | #if DEBUG |
8456a648 JK |
1660 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1661 | struct page *page) | |
1da177e4 | 1662 | { |
1da177e4 | 1663 | int i; |
b03a017b JK |
1664 | |
1665 | if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) { | |
1666 | poison_obj(cachep, page->freelist - obj_offset(cachep), | |
1667 | POISON_FREE); | |
1668 | } | |
1669 | ||
1da177e4 | 1670 | for (i = 0; i < cachep->num; i++) { |
8456a648 | 1671 | void *objp = index_to_obj(cachep, page, i); |
1da177e4 LT |
1672 | |
1673 | if (cachep->flags & SLAB_POISON) { | |
1da177e4 | 1674 | check_poison_obj(cachep, objp); |
40b44137 | 1675 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
1676 | } |
1677 | if (cachep->flags & SLAB_RED_ZONE) { | |
1678 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) | |
756a025f | 1679 | slab_error(cachep, "start of a freed object was overwritten"); |
1da177e4 | 1680 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) |
756a025f | 1681 | slab_error(cachep, "end of a freed object was overwritten"); |
1da177e4 | 1682 | } |
1da177e4 | 1683 | } |
12dd36fa | 1684 | } |
1da177e4 | 1685 | #else |
8456a648 JK |
1686 | static void slab_destroy_debugcheck(struct kmem_cache *cachep, |
1687 | struct page *page) | |
12dd36fa | 1688 | { |
12dd36fa | 1689 | } |
1da177e4 LT |
1690 | #endif |
1691 | ||
911851e6 RD |
1692 | /** |
1693 | * slab_destroy - destroy and release all objects in a slab | |
1694 | * @cachep: cache pointer being destroyed | |
cb8ee1a3 | 1695 | * @page: page pointer being destroyed |
911851e6 | 1696 | * |
8a7d9b43 WSH |
1697 | * Destroy all the objs in a slab page, and release the mem back to the system. |
1698 | * Before calling the slab page must have been unlinked from the cache. The | |
1699 | * kmem_cache_node ->list_lock is not held/needed. | |
12dd36fa | 1700 | */ |
8456a648 | 1701 | static void slab_destroy(struct kmem_cache *cachep, struct page *page) |
12dd36fa | 1702 | { |
7e007355 | 1703 | void *freelist; |
12dd36fa | 1704 | |
8456a648 JK |
1705 | freelist = page->freelist; |
1706 | slab_destroy_debugcheck(cachep, page); | |
5f0d5a3a | 1707 | if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU)) |
bc4f610d KS |
1708 | call_rcu(&page->rcu_head, kmem_rcu_free); |
1709 | else | |
0c3aa83e | 1710 | kmem_freepages(cachep, page); |
68126702 JK |
1711 | |
1712 | /* | |
8456a648 | 1713 | * From now on, we don't use freelist |
68126702 JK |
1714 | * although actual page can be freed in rcu context |
1715 | */ | |
1716 | if (OFF_SLAB(cachep)) | |
8456a648 | 1717 | kmem_cache_free(cachep->freelist_cache, freelist); |
1da177e4 LT |
1718 | } |
1719 | ||
97654dfa JK |
1720 | static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list) |
1721 | { | |
1722 | struct page *page, *n; | |
1723 | ||
1724 | list_for_each_entry_safe(page, n, list, lru) { | |
1725 | list_del(&page->lru); | |
1726 | slab_destroy(cachep, page); | |
1727 | } | |
1728 | } | |
1729 | ||
4d268eba | 1730 | /** |
a70773dd RD |
1731 | * calculate_slab_order - calculate size (page order) of slabs |
1732 | * @cachep: pointer to the cache that is being created | |
1733 | * @size: size of objects to be created in this cache. | |
a70773dd RD |
1734 | * @flags: slab allocation flags |
1735 | * | |
1736 | * Also calculates the number of objects per slab. | |
4d268eba PE |
1737 | * |
1738 | * This could be made much more intelligent. For now, try to avoid using | |
1739 | * high order pages for slabs. When the gfp() functions are more friendly | |
1740 | * towards high-order requests, this should be changed. | |
1741 | */ | |
a737b3e2 | 1742 | static size_t calculate_slab_order(struct kmem_cache *cachep, |
d50112ed | 1743 | size_t size, slab_flags_t flags) |
4d268eba PE |
1744 | { |
1745 | size_t left_over = 0; | |
9888e6fa | 1746 | int gfporder; |
4d268eba | 1747 | |
0aa817f0 | 1748 | for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) { |
4d268eba PE |
1749 | unsigned int num; |
1750 | size_t remainder; | |
1751 | ||
70f75067 | 1752 | num = cache_estimate(gfporder, size, flags, &remainder); |
4d268eba PE |
1753 | if (!num) |
1754 | continue; | |
9888e6fa | 1755 | |
f315e3fa JK |
1756 | /* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */ |
1757 | if (num > SLAB_OBJ_MAX_NUM) | |
1758 | break; | |
1759 | ||
b1ab41c4 | 1760 | if (flags & CFLGS_OFF_SLAB) { |
3217fd9b JK |
1761 | struct kmem_cache *freelist_cache; |
1762 | size_t freelist_size; | |
1763 | ||
1764 | freelist_size = num * sizeof(freelist_idx_t); | |
1765 | freelist_cache = kmalloc_slab(freelist_size, 0u); | |
1766 | if (!freelist_cache) | |
1767 | continue; | |
1768 | ||
b1ab41c4 | 1769 | /* |
3217fd9b | 1770 | * Needed to avoid possible looping condition |
76b342bd | 1771 | * in cache_grow_begin() |
b1ab41c4 | 1772 | */ |
3217fd9b JK |
1773 | if (OFF_SLAB(freelist_cache)) |
1774 | continue; | |
b1ab41c4 | 1775 | |
3217fd9b JK |
1776 | /* check if off slab has enough benefit */ |
1777 | if (freelist_cache->size > cachep->size / 2) | |
1778 | continue; | |
b1ab41c4 | 1779 | } |
4d268eba | 1780 | |
9888e6fa | 1781 | /* Found something acceptable - save it away */ |
4d268eba | 1782 | cachep->num = num; |
9888e6fa | 1783 | cachep->gfporder = gfporder; |
4d268eba PE |
1784 | left_over = remainder; |
1785 | ||
f78bb8ad LT |
1786 | /* |
1787 | * A VFS-reclaimable slab tends to have most allocations | |
1788 | * as GFP_NOFS and we really don't want to have to be allocating | |
1789 | * higher-order pages when we are unable to shrink dcache. | |
1790 | */ | |
1791 | if (flags & SLAB_RECLAIM_ACCOUNT) | |
1792 | break; | |
1793 | ||
4d268eba PE |
1794 | /* |
1795 | * Large number of objects is good, but very large slabs are | |
1796 | * currently bad for the gfp()s. | |
1797 | */ | |
543585cc | 1798 | if (gfporder >= slab_max_order) |
4d268eba PE |
1799 | break; |
1800 | ||
9888e6fa LT |
1801 | /* |
1802 | * Acceptable internal fragmentation? | |
1803 | */ | |
a737b3e2 | 1804 | if (left_over * 8 <= (PAGE_SIZE << gfporder)) |
4d268eba PE |
1805 | break; |
1806 | } | |
1807 | return left_over; | |
1808 | } | |
1809 | ||
bf0dea23 JK |
1810 | static struct array_cache __percpu *alloc_kmem_cache_cpus( |
1811 | struct kmem_cache *cachep, int entries, int batchcount) | |
1812 | { | |
1813 | int cpu; | |
1814 | size_t size; | |
1815 | struct array_cache __percpu *cpu_cache; | |
1816 | ||
1817 | size = sizeof(void *) * entries + sizeof(struct array_cache); | |
85c9f4b0 | 1818 | cpu_cache = __alloc_percpu(size, sizeof(void *)); |
bf0dea23 JK |
1819 | |
1820 | if (!cpu_cache) | |
1821 | return NULL; | |
1822 | ||
1823 | for_each_possible_cpu(cpu) { | |
1824 | init_arraycache(per_cpu_ptr(cpu_cache, cpu), | |
1825 | entries, batchcount); | |
1826 | } | |
1827 | ||
1828 | return cpu_cache; | |
1829 | } | |
1830 | ||
bd721ea7 | 1831 | static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp) |
f30cf7d1 | 1832 | { |
97d06609 | 1833 | if (slab_state >= FULL) |
83b519e8 | 1834 | return enable_cpucache(cachep, gfp); |
2ed3a4ef | 1835 | |
bf0dea23 JK |
1836 | cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1); |
1837 | if (!cachep->cpu_cache) | |
1838 | return 1; | |
1839 | ||
97d06609 | 1840 | if (slab_state == DOWN) { |
bf0dea23 JK |
1841 | /* Creation of first cache (kmem_cache). */ |
1842 | set_up_node(kmem_cache, CACHE_CACHE); | |
2f9baa9f | 1843 | } else if (slab_state == PARTIAL) { |
bf0dea23 JK |
1844 | /* For kmem_cache_node */ |
1845 | set_up_node(cachep, SIZE_NODE); | |
f30cf7d1 | 1846 | } else { |
bf0dea23 | 1847 | int node; |
f30cf7d1 | 1848 | |
bf0dea23 JK |
1849 | for_each_online_node(node) { |
1850 | cachep->node[node] = kmalloc_node( | |
1851 | sizeof(struct kmem_cache_node), gfp, node); | |
1852 | BUG_ON(!cachep->node[node]); | |
1853 | kmem_cache_node_init(cachep->node[node]); | |
f30cf7d1 PE |
1854 | } |
1855 | } | |
bf0dea23 | 1856 | |
6a67368c | 1857 | cachep->node[numa_mem_id()]->next_reap = |
5f0985bb JZ |
1858 | jiffies + REAPTIMEOUT_NODE + |
1859 | ((unsigned long)cachep) % REAPTIMEOUT_NODE; | |
f30cf7d1 PE |
1860 | |
1861 | cpu_cache_get(cachep)->avail = 0; | |
1862 | cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES; | |
1863 | cpu_cache_get(cachep)->batchcount = 1; | |
1864 | cpu_cache_get(cachep)->touched = 0; | |
1865 | cachep->batchcount = 1; | |
1866 | cachep->limit = BOOT_CPUCACHE_ENTRIES; | |
2ed3a4ef | 1867 | return 0; |
f30cf7d1 PE |
1868 | } |
1869 | ||
0293d1fd | 1870 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
d50112ed | 1871 | slab_flags_t flags, const char *name, |
12220dea JK |
1872 | void (*ctor)(void *)) |
1873 | { | |
1874 | return flags; | |
1875 | } | |
1876 | ||
1877 | struct kmem_cache * | |
f4957d5b | 1878 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 1879 | slab_flags_t flags, void (*ctor)(void *)) |
12220dea JK |
1880 | { |
1881 | struct kmem_cache *cachep; | |
1882 | ||
1883 | cachep = find_mergeable(size, align, flags, name, ctor); | |
1884 | if (cachep) { | |
1885 | cachep->refcount++; | |
1886 | ||
1887 | /* | |
1888 | * Adjust the object sizes so that we clear | |
1889 | * the complete object on kzalloc. | |
1890 | */ | |
1891 | cachep->object_size = max_t(int, cachep->object_size, size); | |
1892 | } | |
1893 | return cachep; | |
1894 | } | |
1895 | ||
b03a017b | 1896 | static bool set_objfreelist_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1897 | size_t size, slab_flags_t flags) |
b03a017b JK |
1898 | { |
1899 | size_t left; | |
1900 | ||
1901 | cachep->num = 0; | |
1902 | ||
5f0d5a3a | 1903 | if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU) |
b03a017b JK |
1904 | return false; |
1905 | ||
1906 | left = calculate_slab_order(cachep, size, | |
1907 | flags | CFLGS_OBJFREELIST_SLAB); | |
1908 | if (!cachep->num) | |
1909 | return false; | |
1910 | ||
1911 | if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size) | |
1912 | return false; | |
1913 | ||
1914 | cachep->colour = left / cachep->colour_off; | |
1915 | ||
1916 | return true; | |
1917 | } | |
1918 | ||
158e319b | 1919 | static bool set_off_slab_cache(struct kmem_cache *cachep, |
d50112ed | 1920 | size_t size, slab_flags_t flags) |
158e319b JK |
1921 | { |
1922 | size_t left; | |
1923 | ||
1924 | cachep->num = 0; | |
1925 | ||
1926 | /* | |
3217fd9b JK |
1927 | * Always use on-slab management when SLAB_NOLEAKTRACE |
1928 | * to avoid recursive calls into kmemleak. | |
158e319b | 1929 | */ |
158e319b JK |
1930 | if (flags & SLAB_NOLEAKTRACE) |
1931 | return false; | |
1932 | ||
1933 | /* | |
1934 | * Size is large, assume best to place the slab management obj | |
1935 | * off-slab (should allow better packing of objs). | |
1936 | */ | |
1937 | left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB); | |
1938 | if (!cachep->num) | |
1939 | return false; | |
1940 | ||
1941 | /* | |
1942 | * If the slab has been placed off-slab, and we have enough space then | |
1943 | * move it on-slab. This is at the expense of any extra colouring. | |
1944 | */ | |
1945 | if (left >= cachep->num * sizeof(freelist_idx_t)) | |
1946 | return false; | |
1947 | ||
1948 | cachep->colour = left / cachep->colour_off; | |
1949 | ||
1950 | return true; | |
1951 | } | |
1952 | ||
1953 | static bool set_on_slab_cache(struct kmem_cache *cachep, | |
d50112ed | 1954 | size_t size, slab_flags_t flags) |
158e319b JK |
1955 | { |
1956 | size_t left; | |
1957 | ||
1958 | cachep->num = 0; | |
1959 | ||
1960 | left = calculate_slab_order(cachep, size, flags); | |
1961 | if (!cachep->num) | |
1962 | return false; | |
1963 | ||
1964 | cachep->colour = left / cachep->colour_off; | |
1965 | ||
1966 | return true; | |
1967 | } | |
1968 | ||
1da177e4 | 1969 | /** |
039363f3 | 1970 | * __kmem_cache_create - Create a cache. |
a755b76a | 1971 | * @cachep: cache management descriptor |
1da177e4 | 1972 | * @flags: SLAB flags |
1da177e4 LT |
1973 | * |
1974 | * Returns a ptr to the cache on success, NULL on failure. | |
1975 | * Cannot be called within a int, but can be interrupted. | |
20c2df83 | 1976 | * The @ctor is run when new pages are allocated by the cache. |
1da177e4 | 1977 | * |
1da177e4 LT |
1978 | * The flags are |
1979 | * | |
1980 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
1981 | * to catch references to uninitialised memory. | |
1982 | * | |
1983 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
1984 | * for buffer overruns. | |
1985 | * | |
1da177e4 LT |
1986 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware |
1987 | * cacheline. This can be beneficial if you're counting cycles as closely | |
1988 | * as davem. | |
1989 | */ | |
d50112ed | 1990 | int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags) |
1da177e4 | 1991 | { |
d4a5fca5 | 1992 | size_t ralign = BYTES_PER_WORD; |
83b519e8 | 1993 | gfp_t gfp; |
278b1bb1 | 1994 | int err; |
be4a7988 | 1995 | unsigned int size = cachep->size; |
1da177e4 | 1996 | |
1da177e4 | 1997 | #if DEBUG |
1da177e4 LT |
1998 | #if FORCED_DEBUG |
1999 | /* | |
2000 | * Enable redzoning and last user accounting, except for caches with | |
2001 | * large objects, if the increased size would increase the object size | |
2002 | * above the next power of two: caches with object sizes just above a | |
2003 | * power of two have a significant amount of internal fragmentation. | |
2004 | */ | |
87a927c7 DW |
2005 | if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN + |
2006 | 2 * sizeof(unsigned long long))) | |
b28a02de | 2007 | flags |= SLAB_RED_ZONE | SLAB_STORE_USER; |
5f0d5a3a | 2008 | if (!(flags & SLAB_TYPESAFE_BY_RCU)) |
1da177e4 LT |
2009 | flags |= SLAB_POISON; |
2010 | #endif | |
1da177e4 | 2011 | #endif |
1da177e4 | 2012 | |
a737b3e2 AM |
2013 | /* |
2014 | * Check that size is in terms of words. This is needed to avoid | |
1da177e4 LT |
2015 | * unaligned accesses for some archs when redzoning is used, and makes |
2016 | * sure any on-slab bufctl's are also correctly aligned. | |
2017 | */ | |
e0771950 | 2018 | size = ALIGN(size, BYTES_PER_WORD); |
1da177e4 | 2019 | |
87a927c7 DW |
2020 | if (flags & SLAB_RED_ZONE) { |
2021 | ralign = REDZONE_ALIGN; | |
2022 | /* If redzoning, ensure that the second redzone is suitably | |
2023 | * aligned, by adjusting the object size accordingly. */ | |
e0771950 | 2024 | size = ALIGN(size, REDZONE_ALIGN); |
87a927c7 | 2025 | } |
ca5f9703 | 2026 | |
a44b56d3 | 2027 | /* 3) caller mandated alignment */ |
8a13a4cc CL |
2028 | if (ralign < cachep->align) { |
2029 | ralign = cachep->align; | |
1da177e4 | 2030 | } |
3ff84a7f PE |
2031 | /* disable debug if necessary */ |
2032 | if (ralign > __alignof__(unsigned long long)) | |
a44b56d3 | 2033 | flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); |
a737b3e2 | 2034 | /* |
ca5f9703 | 2035 | * 4) Store it. |
1da177e4 | 2036 | */ |
8a13a4cc | 2037 | cachep->align = ralign; |
158e319b JK |
2038 | cachep->colour_off = cache_line_size(); |
2039 | /* Offset must be a multiple of the alignment. */ | |
2040 | if (cachep->colour_off < cachep->align) | |
2041 | cachep->colour_off = cachep->align; | |
1da177e4 | 2042 | |
83b519e8 PE |
2043 | if (slab_is_available()) |
2044 | gfp = GFP_KERNEL; | |
2045 | else | |
2046 | gfp = GFP_NOWAIT; | |
2047 | ||
1da177e4 | 2048 | #if DEBUG |
1da177e4 | 2049 | |
ca5f9703 PE |
2050 | /* |
2051 | * Both debugging options require word-alignment which is calculated | |
2052 | * into align above. | |
2053 | */ | |
1da177e4 | 2054 | if (flags & SLAB_RED_ZONE) { |
1da177e4 | 2055 | /* add space for red zone words */ |
3ff84a7f PE |
2056 | cachep->obj_offset += sizeof(unsigned long long); |
2057 | size += 2 * sizeof(unsigned long long); | |
1da177e4 LT |
2058 | } |
2059 | if (flags & SLAB_STORE_USER) { | |
ca5f9703 | 2060 | /* user store requires one word storage behind the end of |
87a927c7 DW |
2061 | * the real object. But if the second red zone needs to be |
2062 | * aligned to 64 bits, we must allow that much space. | |
1da177e4 | 2063 | */ |
87a927c7 DW |
2064 | if (flags & SLAB_RED_ZONE) |
2065 | size += REDZONE_ALIGN; | |
2066 | else | |
2067 | size += BYTES_PER_WORD; | |
1da177e4 | 2068 | } |
832a15d2 JK |
2069 | #endif |
2070 | ||
7ed2f9e6 AP |
2071 | kasan_cache_create(cachep, &size, &flags); |
2072 | ||
832a15d2 JK |
2073 | size = ALIGN(size, cachep->align); |
2074 | /* | |
2075 | * We should restrict the number of objects in a slab to implement | |
2076 | * byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition. | |
2077 | */ | |
2078 | if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE) | |
2079 | size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align); | |
2080 | ||
2081 | #if DEBUG | |
03a2d2a3 JK |
2082 | /* |
2083 | * To activate debug pagealloc, off-slab management is necessary | |
2084 | * requirement. In early phase of initialization, small sized slab | |
2085 | * doesn't get initialized so it would not be possible. So, we need | |
2086 | * to check size >= 256. It guarantees that all necessary small | |
2087 | * sized slab is initialized in current slab initialization sequence. | |
2088 | */ | |
40323278 | 2089 | if (debug_pagealloc_enabled() && (flags & SLAB_POISON) && |
f3a3c320 JK |
2090 | size >= 256 && cachep->object_size > cache_line_size()) { |
2091 | if (size < PAGE_SIZE || size % PAGE_SIZE == 0) { | |
2092 | size_t tmp_size = ALIGN(size, PAGE_SIZE); | |
2093 | ||
2094 | if (set_off_slab_cache(cachep, tmp_size, flags)) { | |
2095 | flags |= CFLGS_OFF_SLAB; | |
2096 | cachep->obj_offset += tmp_size - size; | |
2097 | size = tmp_size; | |
2098 | goto done; | |
2099 | } | |
2100 | } | |
1da177e4 | 2101 | } |
1da177e4 LT |
2102 | #endif |
2103 | ||
b03a017b JK |
2104 | if (set_objfreelist_slab_cache(cachep, size, flags)) { |
2105 | flags |= CFLGS_OBJFREELIST_SLAB; | |
2106 | goto done; | |
2107 | } | |
2108 | ||
158e319b | 2109 | if (set_off_slab_cache(cachep, size, flags)) { |
1da177e4 | 2110 | flags |= CFLGS_OFF_SLAB; |
158e319b | 2111 | goto done; |
832a15d2 | 2112 | } |
1da177e4 | 2113 | |
158e319b JK |
2114 | if (set_on_slab_cache(cachep, size, flags)) |
2115 | goto done; | |
1da177e4 | 2116 | |
158e319b | 2117 | return -E2BIG; |
1da177e4 | 2118 | |
158e319b JK |
2119 | done: |
2120 | cachep->freelist_size = cachep->num * sizeof(freelist_idx_t); | |
1da177e4 | 2121 | cachep->flags = flags; |
a57a4988 | 2122 | cachep->allocflags = __GFP_COMP; |
a3187e43 | 2123 | if (flags & SLAB_CACHE_DMA) |
a618e89f | 2124 | cachep->allocflags |= GFP_DMA; |
a3ba0744 DR |
2125 | if (flags & SLAB_RECLAIM_ACCOUNT) |
2126 | cachep->allocflags |= __GFP_RECLAIMABLE; | |
3b0efdfa | 2127 | cachep->size = size; |
6a2d7a95 | 2128 | cachep->reciprocal_buffer_size = reciprocal_value(size); |
1da177e4 | 2129 | |
40b44137 JK |
2130 | #if DEBUG |
2131 | /* | |
2132 | * If we're going to use the generic kernel_map_pages() | |
2133 | * poisoning, then it's going to smash the contents of | |
2134 | * the redzone and userword anyhow, so switch them off. | |
2135 | */ | |
2136 | if (IS_ENABLED(CONFIG_PAGE_POISONING) && | |
2137 | (cachep->flags & SLAB_POISON) && | |
2138 | is_debug_pagealloc_cache(cachep)) | |
2139 | cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER); | |
2140 | #endif | |
2141 | ||
2142 | if (OFF_SLAB(cachep)) { | |
158e319b JK |
2143 | cachep->freelist_cache = |
2144 | kmalloc_slab(cachep->freelist_size, 0u); | |
e5ac9c5a | 2145 | } |
1da177e4 | 2146 | |
278b1bb1 CL |
2147 | err = setup_cpu_cache(cachep, gfp); |
2148 | if (err) { | |
52b4b950 | 2149 | __kmem_cache_release(cachep); |
278b1bb1 | 2150 | return err; |
2ed3a4ef | 2151 | } |
1da177e4 | 2152 | |
278b1bb1 | 2153 | return 0; |
1da177e4 | 2154 | } |
1da177e4 LT |
2155 | |
2156 | #if DEBUG | |
2157 | static void check_irq_off(void) | |
2158 | { | |
2159 | BUG_ON(!irqs_disabled()); | |
2160 | } | |
2161 | ||
2162 | static void check_irq_on(void) | |
2163 | { | |
2164 | BUG_ON(irqs_disabled()); | |
2165 | } | |
2166 | ||
18726ca8 JK |
2167 | static void check_mutex_acquired(void) |
2168 | { | |
2169 | BUG_ON(!mutex_is_locked(&slab_mutex)); | |
2170 | } | |
2171 | ||
343e0d7a | 2172 | static void check_spinlock_acquired(struct kmem_cache *cachep) |
1da177e4 LT |
2173 | { |
2174 | #ifdef CONFIG_SMP | |
2175 | check_irq_off(); | |
18bf8541 | 2176 | assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock); |
1da177e4 LT |
2177 | #endif |
2178 | } | |
e498be7d | 2179 | |
343e0d7a | 2180 | static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node) |
e498be7d CL |
2181 | { |
2182 | #ifdef CONFIG_SMP | |
2183 | check_irq_off(); | |
18bf8541 | 2184 | assert_spin_locked(&get_node(cachep, node)->list_lock); |
e498be7d CL |
2185 | #endif |
2186 | } | |
2187 | ||
1da177e4 LT |
2188 | #else |
2189 | #define check_irq_off() do { } while(0) | |
2190 | #define check_irq_on() do { } while(0) | |
18726ca8 | 2191 | #define check_mutex_acquired() do { } while(0) |
1da177e4 | 2192 | #define check_spinlock_acquired(x) do { } while(0) |
e498be7d | 2193 | #define check_spinlock_acquired_node(x, y) do { } while(0) |
1da177e4 LT |
2194 | #endif |
2195 | ||
18726ca8 JK |
2196 | static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac, |
2197 | int node, bool free_all, struct list_head *list) | |
2198 | { | |
2199 | int tofree; | |
2200 | ||
2201 | if (!ac || !ac->avail) | |
2202 | return; | |
2203 | ||
2204 | tofree = free_all ? ac->avail : (ac->limit + 4) / 5; | |
2205 | if (tofree > ac->avail) | |
2206 | tofree = (ac->avail + 1) / 2; | |
2207 | ||
2208 | free_block(cachep, ac->entry, tofree, node, list); | |
2209 | ac->avail -= tofree; | |
2210 | memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail); | |
2211 | } | |
aab2207c | 2212 | |
1da177e4 LT |
2213 | static void do_drain(void *arg) |
2214 | { | |
a737b3e2 | 2215 | struct kmem_cache *cachep = arg; |
1da177e4 | 2216 | struct array_cache *ac; |
7d6e6d09 | 2217 | int node = numa_mem_id(); |
18bf8541 | 2218 | struct kmem_cache_node *n; |
97654dfa | 2219 | LIST_HEAD(list); |
1da177e4 LT |
2220 | |
2221 | check_irq_off(); | |
9a2dba4b | 2222 | ac = cpu_cache_get(cachep); |
18bf8541 CL |
2223 | n = get_node(cachep, node); |
2224 | spin_lock(&n->list_lock); | |
97654dfa | 2225 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 2226 | spin_unlock(&n->list_lock); |
97654dfa | 2227 | slabs_destroy(cachep, &list); |
1da177e4 LT |
2228 | ac->avail = 0; |
2229 | } | |
2230 | ||
343e0d7a | 2231 | static void drain_cpu_caches(struct kmem_cache *cachep) |
1da177e4 | 2232 | { |
ce8eb6c4 | 2233 | struct kmem_cache_node *n; |
e498be7d | 2234 | int node; |
18726ca8 | 2235 | LIST_HEAD(list); |
e498be7d | 2236 | |
15c8b6c1 | 2237 | on_each_cpu(do_drain, cachep, 1); |
1da177e4 | 2238 | check_irq_on(); |
18bf8541 CL |
2239 | for_each_kmem_cache_node(cachep, node, n) |
2240 | if (n->alien) | |
ce8eb6c4 | 2241 | drain_alien_cache(cachep, n->alien); |
a4523a8b | 2242 | |
18726ca8 JK |
2243 | for_each_kmem_cache_node(cachep, node, n) { |
2244 | spin_lock_irq(&n->list_lock); | |
2245 | drain_array_locked(cachep, n->shared, node, true, &list); | |
2246 | spin_unlock_irq(&n->list_lock); | |
2247 | ||
2248 | slabs_destroy(cachep, &list); | |
2249 | } | |
1da177e4 LT |
2250 | } |
2251 | ||
ed11d9eb CL |
2252 | /* |
2253 | * Remove slabs from the list of free slabs. | |
2254 | * Specify the number of slabs to drain in tofree. | |
2255 | * | |
2256 | * Returns the actual number of slabs released. | |
2257 | */ | |
2258 | static int drain_freelist(struct kmem_cache *cache, | |
ce8eb6c4 | 2259 | struct kmem_cache_node *n, int tofree) |
1da177e4 | 2260 | { |
ed11d9eb CL |
2261 | struct list_head *p; |
2262 | int nr_freed; | |
8456a648 | 2263 | struct page *page; |
1da177e4 | 2264 | |
ed11d9eb | 2265 | nr_freed = 0; |
ce8eb6c4 | 2266 | while (nr_freed < tofree && !list_empty(&n->slabs_free)) { |
1da177e4 | 2267 | |
ce8eb6c4 CL |
2268 | spin_lock_irq(&n->list_lock); |
2269 | p = n->slabs_free.prev; | |
2270 | if (p == &n->slabs_free) { | |
2271 | spin_unlock_irq(&n->list_lock); | |
ed11d9eb CL |
2272 | goto out; |
2273 | } | |
1da177e4 | 2274 | |
8456a648 | 2275 | page = list_entry(p, struct page, lru); |
8456a648 | 2276 | list_del(&page->lru); |
f728b0a5 | 2277 | n->free_slabs--; |
bf00bd34 | 2278 | n->total_slabs--; |
ed11d9eb CL |
2279 | /* |
2280 | * Safe to drop the lock. The slab is no longer linked | |
2281 | * to the cache. | |
2282 | */ | |
ce8eb6c4 CL |
2283 | n->free_objects -= cache->num; |
2284 | spin_unlock_irq(&n->list_lock); | |
8456a648 | 2285 | slab_destroy(cache, page); |
ed11d9eb | 2286 | nr_freed++; |
1da177e4 | 2287 | } |
ed11d9eb CL |
2288 | out: |
2289 | return nr_freed; | |
1da177e4 LT |
2290 | } |
2291 | ||
f9e13c0a SB |
2292 | bool __kmem_cache_empty(struct kmem_cache *s) |
2293 | { | |
2294 | int node; | |
2295 | struct kmem_cache_node *n; | |
2296 | ||
2297 | for_each_kmem_cache_node(s, node, n) | |
2298 | if (!list_empty(&n->slabs_full) || | |
2299 | !list_empty(&n->slabs_partial)) | |
2300 | return false; | |
2301 | return true; | |
2302 | } | |
2303 | ||
c9fc5864 | 2304 | int __kmem_cache_shrink(struct kmem_cache *cachep) |
e498be7d | 2305 | { |
18bf8541 CL |
2306 | int ret = 0; |
2307 | int node; | |
ce8eb6c4 | 2308 | struct kmem_cache_node *n; |
e498be7d CL |
2309 | |
2310 | drain_cpu_caches(cachep); | |
2311 | ||
2312 | check_irq_on(); | |
18bf8541 | 2313 | for_each_kmem_cache_node(cachep, node, n) { |
a5aa63a5 | 2314 | drain_freelist(cachep, n, INT_MAX); |
ed11d9eb | 2315 | |
ce8eb6c4 CL |
2316 | ret += !list_empty(&n->slabs_full) || |
2317 | !list_empty(&n->slabs_partial); | |
e498be7d CL |
2318 | } |
2319 | return (ret ? 1 : 0); | |
2320 | } | |
2321 | ||
c9fc5864 TH |
2322 | #ifdef CONFIG_MEMCG |
2323 | void __kmemcg_cache_deactivate(struct kmem_cache *cachep) | |
2324 | { | |
2325 | __kmem_cache_shrink(cachep); | |
2326 | } | |
2327 | #endif | |
2328 | ||
945cf2b6 | 2329 | int __kmem_cache_shutdown(struct kmem_cache *cachep) |
52b4b950 | 2330 | { |
c9fc5864 | 2331 | return __kmem_cache_shrink(cachep); |
52b4b950 DS |
2332 | } |
2333 | ||
2334 | void __kmem_cache_release(struct kmem_cache *cachep) | |
1da177e4 | 2335 | { |
12c3667f | 2336 | int i; |
ce8eb6c4 | 2337 | struct kmem_cache_node *n; |
1da177e4 | 2338 | |
c7ce4f60 TG |
2339 | cache_random_seq_destroy(cachep); |
2340 | ||
bf0dea23 | 2341 | free_percpu(cachep->cpu_cache); |
1da177e4 | 2342 | |
ce8eb6c4 | 2343 | /* NUMA: free the node structures */ |
18bf8541 CL |
2344 | for_each_kmem_cache_node(cachep, i, n) { |
2345 | kfree(n->shared); | |
2346 | free_alien_cache(n->alien); | |
2347 | kfree(n); | |
2348 | cachep->node[i] = NULL; | |
12c3667f | 2349 | } |
1da177e4 | 2350 | } |
1da177e4 | 2351 | |
e5ac9c5a RT |
2352 | /* |
2353 | * Get the memory for a slab management obj. | |
5f0985bb JZ |
2354 | * |
2355 | * For a slab cache when the slab descriptor is off-slab, the | |
2356 | * slab descriptor can't come from the same cache which is being created, | |
2357 | * Because if it is the case, that means we defer the creation of | |
2358 | * the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point. | |
2359 | * And we eventually call down to __kmem_cache_create(), which | |
2360 | * in turn looks up in the kmalloc_{dma,}_caches for the disired-size one. | |
2361 | * This is a "chicken-and-egg" problem. | |
2362 | * | |
2363 | * So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches, | |
2364 | * which are all initialized during kmem_cache_init(). | |
e5ac9c5a | 2365 | */ |
7e007355 | 2366 | static void *alloc_slabmgmt(struct kmem_cache *cachep, |
0c3aa83e JK |
2367 | struct page *page, int colour_off, |
2368 | gfp_t local_flags, int nodeid) | |
1da177e4 | 2369 | { |
7e007355 | 2370 | void *freelist; |
0c3aa83e | 2371 | void *addr = page_address(page); |
b28a02de | 2372 | |
2e6b3602 JK |
2373 | page->s_mem = addr + colour_off; |
2374 | page->active = 0; | |
2375 | ||
b03a017b JK |
2376 | if (OBJFREELIST_SLAB(cachep)) |
2377 | freelist = NULL; | |
2378 | else if (OFF_SLAB(cachep)) { | |
1da177e4 | 2379 | /* Slab management obj is off-slab. */ |
8456a648 | 2380 | freelist = kmem_cache_alloc_node(cachep->freelist_cache, |
8759ec50 | 2381 | local_flags, nodeid); |
8456a648 | 2382 | if (!freelist) |
1da177e4 LT |
2383 | return NULL; |
2384 | } else { | |
2e6b3602 JK |
2385 | /* We will use last bytes at the slab for freelist */ |
2386 | freelist = addr + (PAGE_SIZE << cachep->gfporder) - | |
2387 | cachep->freelist_size; | |
1da177e4 | 2388 | } |
2e6b3602 | 2389 | |
8456a648 | 2390 | return freelist; |
1da177e4 LT |
2391 | } |
2392 | ||
7cc68973 | 2393 | static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx) |
1da177e4 | 2394 | { |
a41adfaa | 2395 | return ((freelist_idx_t *)page->freelist)[idx]; |
e5c58dfd JK |
2396 | } |
2397 | ||
2398 | static inline void set_free_obj(struct page *page, | |
7cc68973 | 2399 | unsigned int idx, freelist_idx_t val) |
e5c58dfd | 2400 | { |
a41adfaa | 2401 | ((freelist_idx_t *)(page->freelist))[idx] = val; |
1da177e4 LT |
2402 | } |
2403 | ||
10b2e9e8 | 2404 | static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page) |
1da177e4 | 2405 | { |
10b2e9e8 | 2406 | #if DEBUG |
1da177e4 LT |
2407 | int i; |
2408 | ||
2409 | for (i = 0; i < cachep->num; i++) { | |
8456a648 | 2410 | void *objp = index_to_obj(cachep, page, i); |
10b2e9e8 | 2411 | |
1da177e4 LT |
2412 | if (cachep->flags & SLAB_STORE_USER) |
2413 | *dbg_userword(cachep, objp) = NULL; | |
2414 | ||
2415 | if (cachep->flags & SLAB_RED_ZONE) { | |
2416 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; | |
2417 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2418 | } | |
2419 | /* | |
a737b3e2 AM |
2420 | * Constructors are not allowed to allocate memory from the same |
2421 | * cache which they are a constructor for. Otherwise, deadlock. | |
2422 | * They must also be threaded. | |
1da177e4 | 2423 | */ |
7ed2f9e6 AP |
2424 | if (cachep->ctor && !(cachep->flags & SLAB_POISON)) { |
2425 | kasan_unpoison_object_data(cachep, | |
2426 | objp + obj_offset(cachep)); | |
51cc5068 | 2427 | cachep->ctor(objp + obj_offset(cachep)); |
7ed2f9e6 AP |
2428 | kasan_poison_object_data( |
2429 | cachep, objp + obj_offset(cachep)); | |
2430 | } | |
1da177e4 LT |
2431 | |
2432 | if (cachep->flags & SLAB_RED_ZONE) { | |
2433 | if (*dbg_redzone2(cachep, objp) != RED_INACTIVE) | |
756a025f | 2434 | slab_error(cachep, "constructor overwrote the end of an object"); |
1da177e4 | 2435 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE) |
756a025f | 2436 | slab_error(cachep, "constructor overwrote the start of an object"); |
1da177e4 | 2437 | } |
40b44137 JK |
2438 | /* need to poison the objs? */ |
2439 | if (cachep->flags & SLAB_POISON) { | |
2440 | poison_obj(cachep, objp, POISON_FREE); | |
2441 | slab_kernel_map(cachep, objp, 0, 0); | |
2442 | } | |
10b2e9e8 | 2443 | } |
1da177e4 | 2444 | #endif |
10b2e9e8 JK |
2445 | } |
2446 | ||
c7ce4f60 TG |
2447 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
2448 | /* Hold information during a freelist initialization */ | |
2449 | union freelist_init_state { | |
2450 | struct { | |
2451 | unsigned int pos; | |
7c00fce9 | 2452 | unsigned int *list; |
c7ce4f60 | 2453 | unsigned int count; |
c7ce4f60 TG |
2454 | }; |
2455 | struct rnd_state rnd_state; | |
2456 | }; | |
2457 | ||
2458 | /* | |
2459 | * Initialize the state based on the randomization methode available. | |
2460 | * return true if the pre-computed list is available, false otherwize. | |
2461 | */ | |
2462 | static bool freelist_state_initialize(union freelist_init_state *state, | |
2463 | struct kmem_cache *cachep, | |
2464 | unsigned int count) | |
2465 | { | |
2466 | bool ret; | |
2467 | unsigned int rand; | |
2468 | ||
2469 | /* Use best entropy available to define a random shift */ | |
7c00fce9 | 2470 | rand = get_random_int(); |
c7ce4f60 TG |
2471 | |
2472 | /* Use a random state if the pre-computed list is not available */ | |
2473 | if (!cachep->random_seq) { | |
2474 | prandom_seed_state(&state->rnd_state, rand); | |
2475 | ret = false; | |
2476 | } else { | |
2477 | state->list = cachep->random_seq; | |
2478 | state->count = count; | |
c4e490cf | 2479 | state->pos = rand % count; |
c7ce4f60 TG |
2480 | ret = true; |
2481 | } | |
2482 | return ret; | |
2483 | } | |
2484 | ||
2485 | /* Get the next entry on the list and randomize it using a random shift */ | |
2486 | static freelist_idx_t next_random_slot(union freelist_init_state *state) | |
2487 | { | |
c4e490cf JS |
2488 | if (state->pos >= state->count) |
2489 | state->pos = 0; | |
2490 | return state->list[state->pos++]; | |
c7ce4f60 TG |
2491 | } |
2492 | ||
7c00fce9 TG |
2493 | /* Swap two freelist entries */ |
2494 | static void swap_free_obj(struct page *page, unsigned int a, unsigned int b) | |
2495 | { | |
2496 | swap(((freelist_idx_t *)page->freelist)[a], | |
2497 | ((freelist_idx_t *)page->freelist)[b]); | |
2498 | } | |
2499 | ||
c7ce4f60 TG |
2500 | /* |
2501 | * Shuffle the freelist initialization state based on pre-computed lists. | |
2502 | * return true if the list was successfully shuffled, false otherwise. | |
2503 | */ | |
2504 | static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page) | |
2505 | { | |
7c00fce9 | 2506 | unsigned int objfreelist = 0, i, rand, count = cachep->num; |
c7ce4f60 TG |
2507 | union freelist_init_state state; |
2508 | bool precomputed; | |
2509 | ||
2510 | if (count < 2) | |
2511 | return false; | |
2512 | ||
2513 | precomputed = freelist_state_initialize(&state, cachep, count); | |
2514 | ||
2515 | /* Take a random entry as the objfreelist */ | |
2516 | if (OBJFREELIST_SLAB(cachep)) { | |
2517 | if (!precomputed) | |
2518 | objfreelist = count - 1; | |
2519 | else | |
2520 | objfreelist = next_random_slot(&state); | |
2521 | page->freelist = index_to_obj(cachep, page, objfreelist) + | |
2522 | obj_offset(cachep); | |
2523 | count--; | |
2524 | } | |
2525 | ||
2526 | /* | |
2527 | * On early boot, generate the list dynamically. | |
2528 | * Later use a pre-computed list for speed. | |
2529 | */ | |
2530 | if (!precomputed) { | |
7c00fce9 TG |
2531 | for (i = 0; i < count; i++) |
2532 | set_free_obj(page, i, i); | |
2533 | ||
2534 | /* Fisher-Yates shuffle */ | |
2535 | for (i = count - 1; i > 0; i--) { | |
2536 | rand = prandom_u32_state(&state.rnd_state); | |
2537 | rand %= (i + 1); | |
2538 | swap_free_obj(page, i, rand); | |
2539 | } | |
c7ce4f60 TG |
2540 | } else { |
2541 | for (i = 0; i < count; i++) | |
2542 | set_free_obj(page, i, next_random_slot(&state)); | |
2543 | } | |
2544 | ||
2545 | if (OBJFREELIST_SLAB(cachep)) | |
2546 | set_free_obj(page, cachep->num - 1, objfreelist); | |
2547 | ||
2548 | return true; | |
2549 | } | |
2550 | #else | |
2551 | static inline bool shuffle_freelist(struct kmem_cache *cachep, | |
2552 | struct page *page) | |
2553 | { | |
2554 | return false; | |
2555 | } | |
2556 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
2557 | ||
10b2e9e8 JK |
2558 | static void cache_init_objs(struct kmem_cache *cachep, |
2559 | struct page *page) | |
2560 | { | |
2561 | int i; | |
7ed2f9e6 | 2562 | void *objp; |
c7ce4f60 | 2563 | bool shuffled; |
10b2e9e8 JK |
2564 | |
2565 | cache_init_objs_debug(cachep, page); | |
2566 | ||
c7ce4f60 TG |
2567 | /* Try to randomize the freelist if enabled */ |
2568 | shuffled = shuffle_freelist(cachep, page); | |
2569 | ||
2570 | if (!shuffled && OBJFREELIST_SLAB(cachep)) { | |
b03a017b JK |
2571 | page->freelist = index_to_obj(cachep, page, cachep->num - 1) + |
2572 | obj_offset(cachep); | |
2573 | } | |
2574 | ||
10b2e9e8 | 2575 | for (i = 0; i < cachep->num; i++) { |
b3cbd9bf AR |
2576 | objp = index_to_obj(cachep, page, i); |
2577 | kasan_init_slab_obj(cachep, objp); | |
2578 | ||
10b2e9e8 | 2579 | /* constructor could break poison info */ |
7ed2f9e6 | 2580 | if (DEBUG == 0 && cachep->ctor) { |
7ed2f9e6 AP |
2581 | kasan_unpoison_object_data(cachep, objp); |
2582 | cachep->ctor(objp); | |
2583 | kasan_poison_object_data(cachep, objp); | |
2584 | } | |
10b2e9e8 | 2585 | |
c7ce4f60 TG |
2586 | if (!shuffled) |
2587 | set_free_obj(page, i, i); | |
1da177e4 | 2588 | } |
1da177e4 LT |
2589 | } |
2590 | ||
260b61dd | 2591 | static void *slab_get_obj(struct kmem_cache *cachep, struct page *page) |
78d382d7 | 2592 | { |
b1cb0982 | 2593 | void *objp; |
78d382d7 | 2594 | |
e5c58dfd | 2595 | objp = index_to_obj(cachep, page, get_free_obj(page, page->active)); |
8456a648 | 2596 | page->active++; |
78d382d7 | 2597 | |
d31676df JK |
2598 | #if DEBUG |
2599 | if (cachep->flags & SLAB_STORE_USER) | |
2600 | set_store_user_dirty(cachep); | |
2601 | #endif | |
2602 | ||
78d382d7 MD |
2603 | return objp; |
2604 | } | |
2605 | ||
260b61dd JK |
2606 | static void slab_put_obj(struct kmem_cache *cachep, |
2607 | struct page *page, void *objp) | |
78d382d7 | 2608 | { |
8456a648 | 2609 | unsigned int objnr = obj_to_index(cachep, page, objp); |
78d382d7 | 2610 | #if DEBUG |
16025177 | 2611 | unsigned int i; |
b1cb0982 | 2612 | |
b1cb0982 | 2613 | /* Verify double free bug */ |
8456a648 | 2614 | for (i = page->active; i < cachep->num; i++) { |
e5c58dfd | 2615 | if (get_free_obj(page, i) == objnr) { |
85c3e4a5 | 2616 | pr_err("slab: double free detected in cache '%s', objp %px\n", |
756a025f | 2617 | cachep->name, objp); |
b1cb0982 JK |
2618 | BUG(); |
2619 | } | |
78d382d7 MD |
2620 | } |
2621 | #endif | |
8456a648 | 2622 | page->active--; |
b03a017b JK |
2623 | if (!page->freelist) |
2624 | page->freelist = objp + obj_offset(cachep); | |
2625 | ||
e5c58dfd | 2626 | set_free_obj(page, page->active, objnr); |
78d382d7 MD |
2627 | } |
2628 | ||
4776874f PE |
2629 | /* |
2630 | * Map pages beginning at addr to the given cache and slab. This is required | |
2631 | * for the slab allocator to be able to lookup the cache and slab of a | |
ccd35fb9 | 2632 | * virtual address for kfree, ksize, and slab debugging. |
4776874f | 2633 | */ |
8456a648 | 2634 | static void slab_map_pages(struct kmem_cache *cache, struct page *page, |
7e007355 | 2635 | void *freelist) |
1da177e4 | 2636 | { |
a57a4988 | 2637 | page->slab_cache = cache; |
8456a648 | 2638 | page->freelist = freelist; |
1da177e4 LT |
2639 | } |
2640 | ||
2641 | /* | |
2642 | * Grow (by 1) the number of slabs within a cache. This is called by | |
2643 | * kmem_cache_alloc() when there are no active objs left in a cache. | |
2644 | */ | |
76b342bd JK |
2645 | static struct page *cache_grow_begin(struct kmem_cache *cachep, |
2646 | gfp_t flags, int nodeid) | |
1da177e4 | 2647 | { |
7e007355 | 2648 | void *freelist; |
b28a02de PE |
2649 | size_t offset; |
2650 | gfp_t local_flags; | |
511e3a05 | 2651 | int page_node; |
ce8eb6c4 | 2652 | struct kmem_cache_node *n; |
511e3a05 | 2653 | struct page *page; |
1da177e4 | 2654 | |
a737b3e2 AM |
2655 | /* |
2656 | * Be lazy and only check for valid flags here, keeping it out of the | |
2657 | * critical path in kmem_cache_alloc(). | |
1da177e4 | 2658 | */ |
c871ac4e | 2659 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
bacdcb34 | 2660 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
2661 | flags &= ~GFP_SLAB_BUG_MASK; |
2662 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
2663 | invalid_mask, &invalid_mask, flags, &flags); | |
2664 | dump_stack(); | |
c871ac4e | 2665 | } |
128227e7 | 2666 | WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO)); |
6cb06229 | 2667 | local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK); |
1da177e4 | 2668 | |
1da177e4 | 2669 | check_irq_off(); |
d0164adc | 2670 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 LT |
2671 | local_irq_enable(); |
2672 | ||
a737b3e2 AM |
2673 | /* |
2674 | * Get mem for the objs. Attempt to allocate a physical page from | |
2675 | * 'nodeid'. | |
e498be7d | 2676 | */ |
511e3a05 | 2677 | page = kmem_getpages(cachep, local_flags, nodeid); |
0c3aa83e | 2678 | if (!page) |
1da177e4 LT |
2679 | goto failed; |
2680 | ||
511e3a05 JK |
2681 | page_node = page_to_nid(page); |
2682 | n = get_node(cachep, page_node); | |
03d1d43a JK |
2683 | |
2684 | /* Get colour for the slab, and cal the next value. */ | |
2685 | n->colour_next++; | |
2686 | if (n->colour_next >= cachep->colour) | |
2687 | n->colour_next = 0; | |
2688 | ||
2689 | offset = n->colour_next; | |
2690 | if (offset >= cachep->colour) | |
2691 | offset = 0; | |
2692 | ||
2693 | offset *= cachep->colour_off; | |
2694 | ||
1da177e4 | 2695 | /* Get slab management. */ |
8456a648 | 2696 | freelist = alloc_slabmgmt(cachep, page, offset, |
511e3a05 | 2697 | local_flags & ~GFP_CONSTRAINT_MASK, page_node); |
b03a017b | 2698 | if (OFF_SLAB(cachep) && !freelist) |
1da177e4 LT |
2699 | goto opps1; |
2700 | ||
8456a648 | 2701 | slab_map_pages(cachep, page, freelist); |
1da177e4 | 2702 | |
7ed2f9e6 | 2703 | kasan_poison_slab(page); |
8456a648 | 2704 | cache_init_objs(cachep, page); |
1da177e4 | 2705 | |
d0164adc | 2706 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2707 | local_irq_disable(); |
1da177e4 | 2708 | |
76b342bd JK |
2709 | return page; |
2710 | ||
a737b3e2 | 2711 | opps1: |
0c3aa83e | 2712 | kmem_freepages(cachep, page); |
a737b3e2 | 2713 | failed: |
d0164adc | 2714 | if (gfpflags_allow_blocking(local_flags)) |
1da177e4 | 2715 | local_irq_disable(); |
76b342bd JK |
2716 | return NULL; |
2717 | } | |
2718 | ||
2719 | static void cache_grow_end(struct kmem_cache *cachep, struct page *page) | |
2720 | { | |
2721 | struct kmem_cache_node *n; | |
2722 | void *list = NULL; | |
2723 | ||
2724 | check_irq_off(); | |
2725 | ||
2726 | if (!page) | |
2727 | return; | |
2728 | ||
2729 | INIT_LIST_HEAD(&page->lru); | |
2730 | n = get_node(cachep, page_to_nid(page)); | |
2731 | ||
2732 | spin_lock(&n->list_lock); | |
bf00bd34 | 2733 | n->total_slabs++; |
f728b0a5 | 2734 | if (!page->active) { |
76b342bd | 2735 | list_add_tail(&page->lru, &(n->slabs_free)); |
f728b0a5 | 2736 | n->free_slabs++; |
bf00bd34 | 2737 | } else |
76b342bd | 2738 | fixup_slab_list(cachep, n, page, &list); |
07a63c41 | 2739 | |
76b342bd JK |
2740 | STATS_INC_GROWN(cachep); |
2741 | n->free_objects += cachep->num - page->active; | |
2742 | spin_unlock(&n->list_lock); | |
2743 | ||
2744 | fixup_objfreelist_debug(cachep, &list); | |
1da177e4 LT |
2745 | } |
2746 | ||
2747 | #if DEBUG | |
2748 | ||
2749 | /* | |
2750 | * Perform extra freeing checks: | |
2751 | * - detect bad pointers. | |
2752 | * - POISON/RED_ZONE checking | |
1da177e4 LT |
2753 | */ |
2754 | static void kfree_debugcheck(const void *objp) | |
2755 | { | |
1da177e4 | 2756 | if (!virt_addr_valid(objp)) { |
1170532b | 2757 | pr_err("kfree_debugcheck: out of range ptr %lxh\n", |
b28a02de PE |
2758 | (unsigned long)objp); |
2759 | BUG(); | |
1da177e4 | 2760 | } |
1da177e4 LT |
2761 | } |
2762 | ||
58ce1fd5 PE |
2763 | static inline void verify_redzone_free(struct kmem_cache *cache, void *obj) |
2764 | { | |
b46b8f19 | 2765 | unsigned long long redzone1, redzone2; |
58ce1fd5 PE |
2766 | |
2767 | redzone1 = *dbg_redzone1(cache, obj); | |
2768 | redzone2 = *dbg_redzone2(cache, obj); | |
2769 | ||
2770 | /* | |
2771 | * Redzone is ok. | |
2772 | */ | |
2773 | if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE) | |
2774 | return; | |
2775 | ||
2776 | if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE) | |
2777 | slab_error(cache, "double free detected"); | |
2778 | else | |
2779 | slab_error(cache, "memory outside object was overwritten"); | |
2780 | ||
85c3e4a5 | 2781 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b | 2782 | obj, redzone1, redzone2); |
58ce1fd5 PE |
2783 | } |
2784 | ||
343e0d7a | 2785 | static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp, |
7c0cb9c6 | 2786 | unsigned long caller) |
1da177e4 | 2787 | { |
1da177e4 | 2788 | unsigned int objnr; |
8456a648 | 2789 | struct page *page; |
1da177e4 | 2790 | |
80cbd911 MW |
2791 | BUG_ON(virt_to_cache(objp) != cachep); |
2792 | ||
3dafccf2 | 2793 | objp -= obj_offset(cachep); |
1da177e4 | 2794 | kfree_debugcheck(objp); |
b49af68f | 2795 | page = virt_to_head_page(objp); |
1da177e4 | 2796 | |
1da177e4 | 2797 | if (cachep->flags & SLAB_RED_ZONE) { |
58ce1fd5 | 2798 | verify_redzone_free(cachep, objp); |
1da177e4 LT |
2799 | *dbg_redzone1(cachep, objp) = RED_INACTIVE; |
2800 | *dbg_redzone2(cachep, objp) = RED_INACTIVE; | |
2801 | } | |
d31676df JK |
2802 | if (cachep->flags & SLAB_STORE_USER) { |
2803 | set_store_user_dirty(cachep); | |
7c0cb9c6 | 2804 | *dbg_userword(cachep, objp) = (void *)caller; |
d31676df | 2805 | } |
1da177e4 | 2806 | |
8456a648 | 2807 | objnr = obj_to_index(cachep, page, objp); |
1da177e4 LT |
2808 | |
2809 | BUG_ON(objnr >= cachep->num); | |
8456a648 | 2810 | BUG_ON(objp != index_to_obj(cachep, page, objnr)); |
1da177e4 | 2811 | |
1da177e4 | 2812 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 2813 | poison_obj(cachep, objp, POISON_FREE); |
40b44137 | 2814 | slab_kernel_map(cachep, objp, 0, caller); |
1da177e4 LT |
2815 | } |
2816 | return objp; | |
2817 | } | |
2818 | ||
1da177e4 LT |
2819 | #else |
2820 | #define kfree_debugcheck(x) do { } while(0) | |
2821 | #define cache_free_debugcheck(x,objp,z) (objp) | |
1da177e4 LT |
2822 | #endif |
2823 | ||
b03a017b JK |
2824 | static inline void fixup_objfreelist_debug(struct kmem_cache *cachep, |
2825 | void **list) | |
2826 | { | |
2827 | #if DEBUG | |
2828 | void *next = *list; | |
2829 | void *objp; | |
2830 | ||
2831 | while (next) { | |
2832 | objp = next - obj_offset(cachep); | |
2833 | next = *(void **)next; | |
2834 | poison_obj(cachep, objp, POISON_FREE); | |
2835 | } | |
2836 | #endif | |
2837 | } | |
2838 | ||
d8410234 | 2839 | static inline void fixup_slab_list(struct kmem_cache *cachep, |
b03a017b JK |
2840 | struct kmem_cache_node *n, struct page *page, |
2841 | void **list) | |
d8410234 JK |
2842 | { |
2843 | /* move slabp to correct slabp list: */ | |
2844 | list_del(&page->lru); | |
b03a017b | 2845 | if (page->active == cachep->num) { |
d8410234 | 2846 | list_add(&page->lru, &n->slabs_full); |
b03a017b JK |
2847 | if (OBJFREELIST_SLAB(cachep)) { |
2848 | #if DEBUG | |
2849 | /* Poisoning will be done without holding the lock */ | |
2850 | if (cachep->flags & SLAB_POISON) { | |
2851 | void **objp = page->freelist; | |
2852 | ||
2853 | *objp = *list; | |
2854 | *list = objp; | |
2855 | } | |
2856 | #endif | |
2857 | page->freelist = NULL; | |
2858 | } | |
2859 | } else | |
d8410234 JK |
2860 | list_add(&page->lru, &n->slabs_partial); |
2861 | } | |
2862 | ||
f68f8ddd JK |
2863 | /* Try to find non-pfmemalloc slab if needed */ |
2864 | static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n, | |
bf00bd34 | 2865 | struct page *page, bool pfmemalloc) |
f68f8ddd JK |
2866 | { |
2867 | if (!page) | |
2868 | return NULL; | |
2869 | ||
2870 | if (pfmemalloc) | |
2871 | return page; | |
2872 | ||
2873 | if (!PageSlabPfmemalloc(page)) | |
2874 | return page; | |
2875 | ||
2876 | /* No need to keep pfmemalloc slab if we have enough free objects */ | |
2877 | if (n->free_objects > n->free_limit) { | |
2878 | ClearPageSlabPfmemalloc(page); | |
2879 | return page; | |
2880 | } | |
2881 | ||
2882 | /* Move pfmemalloc slab to the end of list to speed up next search */ | |
2883 | list_del(&page->lru); | |
bf00bd34 | 2884 | if (!page->active) { |
f68f8ddd | 2885 | list_add_tail(&page->lru, &n->slabs_free); |
bf00bd34 | 2886 | n->free_slabs++; |
f728b0a5 | 2887 | } else |
f68f8ddd JK |
2888 | list_add_tail(&page->lru, &n->slabs_partial); |
2889 | ||
2890 | list_for_each_entry(page, &n->slabs_partial, lru) { | |
2891 | if (!PageSlabPfmemalloc(page)) | |
2892 | return page; | |
2893 | } | |
2894 | ||
f728b0a5 | 2895 | n->free_touched = 1; |
f68f8ddd | 2896 | list_for_each_entry(page, &n->slabs_free, lru) { |
f728b0a5 | 2897 | if (!PageSlabPfmemalloc(page)) { |
bf00bd34 | 2898 | n->free_slabs--; |
f68f8ddd | 2899 | return page; |
f728b0a5 | 2900 | } |
f68f8ddd JK |
2901 | } |
2902 | ||
2903 | return NULL; | |
2904 | } | |
2905 | ||
2906 | static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc) | |
7aa0d227 GT |
2907 | { |
2908 | struct page *page; | |
2909 | ||
f728b0a5 | 2910 | assert_spin_locked(&n->list_lock); |
bf00bd34 | 2911 | page = list_first_entry_or_null(&n->slabs_partial, struct page, lru); |
7aa0d227 GT |
2912 | if (!page) { |
2913 | n->free_touched = 1; | |
bf00bd34 DR |
2914 | page = list_first_entry_or_null(&n->slabs_free, struct page, |
2915 | lru); | |
f728b0a5 | 2916 | if (page) |
bf00bd34 | 2917 | n->free_slabs--; |
7aa0d227 GT |
2918 | } |
2919 | ||
f68f8ddd | 2920 | if (sk_memalloc_socks()) |
bf00bd34 | 2921 | page = get_valid_first_slab(n, page, pfmemalloc); |
f68f8ddd | 2922 | |
7aa0d227 GT |
2923 | return page; |
2924 | } | |
2925 | ||
f68f8ddd JK |
2926 | static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep, |
2927 | struct kmem_cache_node *n, gfp_t flags) | |
2928 | { | |
2929 | struct page *page; | |
2930 | void *obj; | |
2931 | void *list = NULL; | |
2932 | ||
2933 | if (!gfp_pfmemalloc_allowed(flags)) | |
2934 | return NULL; | |
2935 | ||
2936 | spin_lock(&n->list_lock); | |
2937 | page = get_first_slab(n, true); | |
2938 | if (!page) { | |
2939 | spin_unlock(&n->list_lock); | |
2940 | return NULL; | |
2941 | } | |
2942 | ||
2943 | obj = slab_get_obj(cachep, page); | |
2944 | n->free_objects--; | |
2945 | ||
2946 | fixup_slab_list(cachep, n, page, &list); | |
2947 | ||
2948 | spin_unlock(&n->list_lock); | |
2949 | fixup_objfreelist_debug(cachep, &list); | |
2950 | ||
2951 | return obj; | |
2952 | } | |
2953 | ||
213b4695 JK |
2954 | /* |
2955 | * Slab list should be fixed up by fixup_slab_list() for existing slab | |
2956 | * or cache_grow_end() for new slab | |
2957 | */ | |
2958 | static __always_inline int alloc_block(struct kmem_cache *cachep, | |
2959 | struct array_cache *ac, struct page *page, int batchcount) | |
2960 | { | |
2961 | /* | |
2962 | * There must be at least one object available for | |
2963 | * allocation. | |
2964 | */ | |
2965 | BUG_ON(page->active >= cachep->num); | |
2966 | ||
2967 | while (page->active < cachep->num && batchcount--) { | |
2968 | STATS_INC_ALLOCED(cachep); | |
2969 | STATS_INC_ACTIVE(cachep); | |
2970 | STATS_SET_HIGH(cachep); | |
2971 | ||
2972 | ac->entry[ac->avail++] = slab_get_obj(cachep, page); | |
2973 | } | |
2974 | ||
2975 | return batchcount; | |
2976 | } | |
2977 | ||
f68f8ddd | 2978 | static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 LT |
2979 | { |
2980 | int batchcount; | |
ce8eb6c4 | 2981 | struct kmem_cache_node *n; |
801faf0d | 2982 | struct array_cache *ac, *shared; |
1ca4cb24 | 2983 | int node; |
b03a017b | 2984 | void *list = NULL; |
76b342bd | 2985 | struct page *page; |
1ca4cb24 | 2986 | |
1da177e4 | 2987 | check_irq_off(); |
7d6e6d09 | 2988 | node = numa_mem_id(); |
f68f8ddd | 2989 | |
9a2dba4b | 2990 | ac = cpu_cache_get(cachep); |
1da177e4 LT |
2991 | batchcount = ac->batchcount; |
2992 | if (!ac->touched && batchcount > BATCHREFILL_LIMIT) { | |
a737b3e2 AM |
2993 | /* |
2994 | * If there was little recent activity on this cache, then | |
2995 | * perform only a partial refill. Otherwise we could generate | |
2996 | * refill bouncing. | |
1da177e4 LT |
2997 | */ |
2998 | batchcount = BATCHREFILL_LIMIT; | |
2999 | } | |
18bf8541 | 3000 | n = get_node(cachep, node); |
e498be7d | 3001 | |
ce8eb6c4 | 3002 | BUG_ON(ac->avail > 0 || !n); |
801faf0d JK |
3003 | shared = READ_ONCE(n->shared); |
3004 | if (!n->free_objects && (!shared || !shared->avail)) | |
3005 | goto direct_grow; | |
3006 | ||
ce8eb6c4 | 3007 | spin_lock(&n->list_lock); |
801faf0d | 3008 | shared = READ_ONCE(n->shared); |
1da177e4 | 3009 | |
3ded175a | 3010 | /* See if we can refill from the shared array */ |
801faf0d JK |
3011 | if (shared && transfer_objects(ac, shared, batchcount)) { |
3012 | shared->touched = 1; | |
3ded175a | 3013 | goto alloc_done; |
44b57f1c | 3014 | } |
3ded175a | 3015 | |
1da177e4 | 3016 | while (batchcount > 0) { |
1da177e4 | 3017 | /* Get slab alloc is to come from. */ |
f68f8ddd | 3018 | page = get_first_slab(n, false); |
7aa0d227 GT |
3019 | if (!page) |
3020 | goto must_grow; | |
1da177e4 | 3021 | |
1da177e4 | 3022 | check_spinlock_acquired(cachep); |
714b8171 | 3023 | |
213b4695 | 3024 | batchcount = alloc_block(cachep, ac, page, batchcount); |
b03a017b | 3025 | fixup_slab_list(cachep, n, page, &list); |
1da177e4 LT |
3026 | } |
3027 | ||
a737b3e2 | 3028 | must_grow: |
ce8eb6c4 | 3029 | n->free_objects -= ac->avail; |
a737b3e2 | 3030 | alloc_done: |
ce8eb6c4 | 3031 | spin_unlock(&n->list_lock); |
b03a017b | 3032 | fixup_objfreelist_debug(cachep, &list); |
1da177e4 | 3033 | |
801faf0d | 3034 | direct_grow: |
1da177e4 | 3035 | if (unlikely(!ac->avail)) { |
f68f8ddd JK |
3036 | /* Check if we can use obj in pfmemalloc slab */ |
3037 | if (sk_memalloc_socks()) { | |
3038 | void *obj = cache_alloc_pfmemalloc(cachep, n, flags); | |
3039 | ||
3040 | if (obj) | |
3041 | return obj; | |
3042 | } | |
3043 | ||
76b342bd | 3044 | page = cache_grow_begin(cachep, gfp_exact_node(flags), node); |
e498be7d | 3045 | |
76b342bd JK |
3046 | /* |
3047 | * cache_grow_begin() can reenable interrupts, | |
3048 | * then ac could change. | |
3049 | */ | |
9a2dba4b | 3050 | ac = cpu_cache_get(cachep); |
213b4695 JK |
3051 | if (!ac->avail && page) |
3052 | alloc_block(cachep, ac, page, batchcount); | |
3053 | cache_grow_end(cachep, page); | |
072bb0aa | 3054 | |
213b4695 | 3055 | if (!ac->avail) |
1da177e4 | 3056 | return NULL; |
1da177e4 LT |
3057 | } |
3058 | ac->touched = 1; | |
072bb0aa | 3059 | |
f68f8ddd | 3060 | return ac->entry[--ac->avail]; |
1da177e4 LT |
3061 | } |
3062 | ||
a737b3e2 AM |
3063 | static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep, |
3064 | gfp_t flags) | |
1da177e4 | 3065 | { |
d0164adc | 3066 | might_sleep_if(gfpflags_allow_blocking(flags)); |
1da177e4 LT |
3067 | } |
3068 | ||
3069 | #if DEBUG | |
a737b3e2 | 3070 | static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep, |
7c0cb9c6 | 3071 | gfp_t flags, void *objp, unsigned long caller) |
1da177e4 | 3072 | { |
128227e7 | 3073 | WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO)); |
b28a02de | 3074 | if (!objp) |
1da177e4 | 3075 | return objp; |
b28a02de | 3076 | if (cachep->flags & SLAB_POISON) { |
1da177e4 | 3077 | check_poison_obj(cachep, objp); |
40b44137 | 3078 | slab_kernel_map(cachep, objp, 1, 0); |
1da177e4 LT |
3079 | poison_obj(cachep, objp, POISON_INUSE); |
3080 | } | |
3081 | if (cachep->flags & SLAB_STORE_USER) | |
7c0cb9c6 | 3082 | *dbg_userword(cachep, objp) = (void *)caller; |
1da177e4 LT |
3083 | |
3084 | if (cachep->flags & SLAB_RED_ZONE) { | |
a737b3e2 AM |
3085 | if (*dbg_redzone1(cachep, objp) != RED_INACTIVE || |
3086 | *dbg_redzone2(cachep, objp) != RED_INACTIVE) { | |
756a025f | 3087 | slab_error(cachep, "double free, or memory outside object was overwritten"); |
85c3e4a5 | 3088 | pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n", |
1170532b JP |
3089 | objp, *dbg_redzone1(cachep, objp), |
3090 | *dbg_redzone2(cachep, objp)); | |
1da177e4 LT |
3091 | } |
3092 | *dbg_redzone1(cachep, objp) = RED_ACTIVE; | |
3093 | *dbg_redzone2(cachep, objp) = RED_ACTIVE; | |
3094 | } | |
03787301 | 3095 | |
3dafccf2 | 3096 | objp += obj_offset(cachep); |
4f104934 | 3097 | if (cachep->ctor && cachep->flags & SLAB_POISON) |
51cc5068 | 3098 | cachep->ctor(objp); |
7ea466f2 TH |
3099 | if (ARCH_SLAB_MINALIGN && |
3100 | ((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) { | |
85c3e4a5 | 3101 | pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n", |
c225150b | 3102 | objp, (int)ARCH_SLAB_MINALIGN); |
a44b56d3 | 3103 | } |
1da177e4 LT |
3104 | return objp; |
3105 | } | |
3106 | #else | |
3107 | #define cache_alloc_debugcheck_after(a,b,objp,d) (objp) | |
3108 | #endif | |
3109 | ||
343e0d7a | 3110 | static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3111 | { |
b28a02de | 3112 | void *objp; |
1da177e4 LT |
3113 | struct array_cache *ac; |
3114 | ||
5c382300 | 3115 | check_irq_off(); |
8a8b6502 | 3116 | |
9a2dba4b | 3117 | ac = cpu_cache_get(cachep); |
1da177e4 | 3118 | if (likely(ac->avail)) { |
1da177e4 | 3119 | ac->touched = 1; |
f68f8ddd | 3120 | objp = ac->entry[--ac->avail]; |
072bb0aa | 3121 | |
f68f8ddd JK |
3122 | STATS_INC_ALLOCHIT(cachep); |
3123 | goto out; | |
1da177e4 | 3124 | } |
072bb0aa MG |
3125 | |
3126 | STATS_INC_ALLOCMISS(cachep); | |
f68f8ddd | 3127 | objp = cache_alloc_refill(cachep, flags); |
072bb0aa MG |
3128 | /* |
3129 | * the 'ac' may be updated by cache_alloc_refill(), | |
3130 | * and kmemleak_erase() requires its correct value. | |
3131 | */ | |
3132 | ac = cpu_cache_get(cachep); | |
3133 | ||
3134 | out: | |
d5cff635 CM |
3135 | /* |
3136 | * To avoid a false negative, if an object that is in one of the | |
3137 | * per-CPU caches is leaked, we need to make sure kmemleak doesn't | |
3138 | * treat the array pointers as a reference to the object. | |
3139 | */ | |
f3d8b53a O |
3140 | if (objp) |
3141 | kmemleak_erase(&ac->entry[ac->avail]); | |
5c382300 AK |
3142 | return objp; |
3143 | } | |
3144 | ||
e498be7d | 3145 | #ifdef CONFIG_NUMA |
c61afb18 | 3146 | /* |
2ad654bc | 3147 | * Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set. |
c61afb18 PJ |
3148 | * |
3149 | * If we are in_interrupt, then process context, including cpusets and | |
3150 | * mempolicy, may not apply and should not be used for allocation policy. | |
3151 | */ | |
3152 | static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3153 | { | |
3154 | int nid_alloc, nid_here; | |
3155 | ||
765c4507 | 3156 | if (in_interrupt() || (flags & __GFP_THISNODE)) |
c61afb18 | 3157 | return NULL; |
7d6e6d09 | 3158 | nid_alloc = nid_here = numa_mem_id(); |
c61afb18 | 3159 | if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD)) |
6adef3eb | 3160 | nid_alloc = cpuset_slab_spread_node(); |
c61afb18 | 3161 | else if (current->mempolicy) |
2a389610 | 3162 | nid_alloc = mempolicy_slab_node(); |
c61afb18 | 3163 | if (nid_alloc != nid_here) |
8b98c169 | 3164 | return ____cache_alloc_node(cachep, flags, nid_alloc); |
c61afb18 PJ |
3165 | return NULL; |
3166 | } | |
3167 | ||
765c4507 CL |
3168 | /* |
3169 | * Fallback function if there was no memory available and no objects on a | |
3c517a61 | 3170 | * certain node and fall back is permitted. First we scan all the |
6a67368c | 3171 | * available node for available objects. If that fails then we |
3c517a61 CL |
3172 | * perform an allocation without specifying a node. This allows the page |
3173 | * allocator to do its reclaim / fallback magic. We then insert the | |
3174 | * slab into the proper nodelist and then allocate from it. | |
765c4507 | 3175 | */ |
8c8cc2c1 | 3176 | static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags) |
765c4507 | 3177 | { |
8c8cc2c1 | 3178 | struct zonelist *zonelist; |
dd1a239f | 3179 | struct zoneref *z; |
54a6eb5c MG |
3180 | struct zone *zone; |
3181 | enum zone_type high_zoneidx = gfp_zone(flags); | |
765c4507 | 3182 | void *obj = NULL; |
76b342bd | 3183 | struct page *page; |
3c517a61 | 3184 | int nid; |
cc9a6c87 | 3185 | unsigned int cpuset_mems_cookie; |
8c8cc2c1 PE |
3186 | |
3187 | if (flags & __GFP_THISNODE) | |
3188 | return NULL; | |
3189 | ||
cc9a6c87 | 3190 | retry_cpuset: |
d26914d1 | 3191 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 3192 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 | 3193 | |
3c517a61 CL |
3194 | retry: |
3195 | /* | |
3196 | * Look through allowed nodes for objects available | |
3197 | * from existing per node queues. | |
3198 | */ | |
54a6eb5c MG |
3199 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
3200 | nid = zone_to_nid(zone); | |
aedb0eb1 | 3201 | |
061d7074 | 3202 | if (cpuset_zone_allowed(zone, flags) && |
18bf8541 CL |
3203 | get_node(cache, nid) && |
3204 | get_node(cache, nid)->free_objects) { | |
3c517a61 | 3205 | obj = ____cache_alloc_node(cache, |
4167e9b2 | 3206 | gfp_exact_node(flags), nid); |
481c5346 CL |
3207 | if (obj) |
3208 | break; | |
3209 | } | |
3c517a61 CL |
3210 | } |
3211 | ||
cfce6604 | 3212 | if (!obj) { |
3c517a61 CL |
3213 | /* |
3214 | * This allocation will be performed within the constraints | |
3215 | * of the current cpuset / memory policy requirements. | |
3216 | * We may trigger various forms of reclaim on the allowed | |
3217 | * set and go into memory reserves if necessary. | |
3218 | */ | |
76b342bd JK |
3219 | page = cache_grow_begin(cache, flags, numa_mem_id()); |
3220 | cache_grow_end(cache, page); | |
3221 | if (page) { | |
3222 | nid = page_to_nid(page); | |
511e3a05 JK |
3223 | obj = ____cache_alloc_node(cache, |
3224 | gfp_exact_node(flags), nid); | |
0c3aa83e | 3225 | |
3c517a61 | 3226 | /* |
511e3a05 JK |
3227 | * Another processor may allocate the objects in |
3228 | * the slab since we are not holding any locks. | |
3c517a61 | 3229 | */ |
511e3a05 JK |
3230 | if (!obj) |
3231 | goto retry; | |
3c517a61 | 3232 | } |
aedb0eb1 | 3233 | } |
cc9a6c87 | 3234 | |
d26914d1 | 3235 | if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie))) |
cc9a6c87 | 3236 | goto retry_cpuset; |
765c4507 CL |
3237 | return obj; |
3238 | } | |
3239 | ||
e498be7d CL |
3240 | /* |
3241 | * A interface to enable slab creation on nodeid | |
1da177e4 | 3242 | */ |
8b98c169 | 3243 | static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, |
a737b3e2 | 3244 | int nodeid) |
e498be7d | 3245 | { |
8456a648 | 3246 | struct page *page; |
ce8eb6c4 | 3247 | struct kmem_cache_node *n; |
213b4695 | 3248 | void *obj = NULL; |
b03a017b | 3249 | void *list = NULL; |
b28a02de | 3250 | |
7c3fbbdd | 3251 | VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES); |
18bf8541 | 3252 | n = get_node(cachep, nodeid); |
ce8eb6c4 | 3253 | BUG_ON(!n); |
b28a02de | 3254 | |
ca3b9b91 | 3255 | check_irq_off(); |
ce8eb6c4 | 3256 | spin_lock(&n->list_lock); |
f68f8ddd | 3257 | page = get_first_slab(n, false); |
7aa0d227 GT |
3258 | if (!page) |
3259 | goto must_grow; | |
b28a02de | 3260 | |
b28a02de | 3261 | check_spinlock_acquired_node(cachep, nodeid); |
b28a02de PE |
3262 | |
3263 | STATS_INC_NODEALLOCS(cachep); | |
3264 | STATS_INC_ACTIVE(cachep); | |
3265 | STATS_SET_HIGH(cachep); | |
3266 | ||
8456a648 | 3267 | BUG_ON(page->active == cachep->num); |
b28a02de | 3268 | |
260b61dd | 3269 | obj = slab_get_obj(cachep, page); |
ce8eb6c4 | 3270 | n->free_objects--; |
b28a02de | 3271 | |
b03a017b | 3272 | fixup_slab_list(cachep, n, page, &list); |
e498be7d | 3273 | |
ce8eb6c4 | 3274 | spin_unlock(&n->list_lock); |
b03a017b | 3275 | fixup_objfreelist_debug(cachep, &list); |
213b4695 | 3276 | return obj; |
e498be7d | 3277 | |
a737b3e2 | 3278 | must_grow: |
ce8eb6c4 | 3279 | spin_unlock(&n->list_lock); |
76b342bd | 3280 | page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid); |
213b4695 JK |
3281 | if (page) { |
3282 | /* This slab isn't counted yet so don't update free_objects */ | |
3283 | obj = slab_get_obj(cachep, page); | |
3284 | } | |
76b342bd | 3285 | cache_grow_end(cachep, page); |
1da177e4 | 3286 | |
213b4695 | 3287 | return obj ? obj : fallback_alloc(cachep, flags); |
e498be7d | 3288 | } |
8c8cc2c1 | 3289 | |
8c8cc2c1 | 3290 | static __always_inline void * |
48356303 | 3291 | slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid, |
7c0cb9c6 | 3292 | unsigned long caller) |
8c8cc2c1 PE |
3293 | { |
3294 | unsigned long save_flags; | |
3295 | void *ptr; | |
7d6e6d09 | 3296 | int slab_node = numa_mem_id(); |
8c8cc2c1 | 3297 | |
dcce284a | 3298 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3299 | cachep = slab_pre_alloc_hook(cachep, flags); |
3300 | if (unlikely(!cachep)) | |
824ebef1 AM |
3301 | return NULL; |
3302 | ||
8c8cc2c1 PE |
3303 | cache_alloc_debugcheck_before(cachep, flags); |
3304 | local_irq_save(save_flags); | |
3305 | ||
eacbbae3 | 3306 | if (nodeid == NUMA_NO_NODE) |
7d6e6d09 | 3307 | nodeid = slab_node; |
8c8cc2c1 | 3308 | |
18bf8541 | 3309 | if (unlikely(!get_node(cachep, nodeid))) { |
8c8cc2c1 PE |
3310 | /* Node not bootstrapped yet */ |
3311 | ptr = fallback_alloc(cachep, flags); | |
3312 | goto out; | |
3313 | } | |
3314 | ||
7d6e6d09 | 3315 | if (nodeid == slab_node) { |
8c8cc2c1 PE |
3316 | /* |
3317 | * Use the locally cached objects if possible. | |
3318 | * However ____cache_alloc does not allow fallback | |
3319 | * to other nodes. It may fail while we still have | |
3320 | * objects on other nodes available. | |
3321 | */ | |
3322 | ptr = ____cache_alloc(cachep, flags); | |
3323 | if (ptr) | |
3324 | goto out; | |
3325 | } | |
3326 | /* ___cache_alloc_node can fall back to other nodes */ | |
3327 | ptr = ____cache_alloc_node(cachep, flags, nodeid); | |
3328 | out: | |
3329 | local_irq_restore(save_flags); | |
3330 | ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller); | |
3331 | ||
d5e3ed66 JDB |
3332 | if (unlikely(flags & __GFP_ZERO) && ptr) |
3333 | memset(ptr, 0, cachep->object_size); | |
d07dbea4 | 3334 | |
d5e3ed66 | 3335 | slab_post_alloc_hook(cachep, flags, 1, &ptr); |
8c8cc2c1 PE |
3336 | return ptr; |
3337 | } | |
3338 | ||
3339 | static __always_inline void * | |
3340 | __do_cache_alloc(struct kmem_cache *cache, gfp_t flags) | |
3341 | { | |
3342 | void *objp; | |
3343 | ||
2ad654bc | 3344 | if (current->mempolicy || cpuset_do_slab_mem_spread()) { |
8c8cc2c1 PE |
3345 | objp = alternate_node_alloc(cache, flags); |
3346 | if (objp) | |
3347 | goto out; | |
3348 | } | |
3349 | objp = ____cache_alloc(cache, flags); | |
3350 | ||
3351 | /* | |
3352 | * We may just have run out of memory on the local node. | |
3353 | * ____cache_alloc_node() knows how to locate memory on other nodes | |
3354 | */ | |
7d6e6d09 LS |
3355 | if (!objp) |
3356 | objp = ____cache_alloc_node(cache, flags, numa_mem_id()); | |
8c8cc2c1 PE |
3357 | |
3358 | out: | |
3359 | return objp; | |
3360 | } | |
3361 | #else | |
3362 | ||
3363 | static __always_inline void * | |
3364 | __do_cache_alloc(struct kmem_cache *cachep, gfp_t flags) | |
3365 | { | |
3366 | return ____cache_alloc(cachep, flags); | |
3367 | } | |
3368 | ||
3369 | #endif /* CONFIG_NUMA */ | |
3370 | ||
3371 | static __always_inline void * | |
48356303 | 3372 | slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller) |
8c8cc2c1 PE |
3373 | { |
3374 | unsigned long save_flags; | |
3375 | void *objp; | |
3376 | ||
dcce284a | 3377 | flags &= gfp_allowed_mask; |
011eceaf JDB |
3378 | cachep = slab_pre_alloc_hook(cachep, flags); |
3379 | if (unlikely(!cachep)) | |
824ebef1 AM |
3380 | return NULL; |
3381 | ||
8c8cc2c1 PE |
3382 | cache_alloc_debugcheck_before(cachep, flags); |
3383 | local_irq_save(save_flags); | |
3384 | objp = __do_cache_alloc(cachep, flags); | |
3385 | local_irq_restore(save_flags); | |
3386 | objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller); | |
3387 | prefetchw(objp); | |
3388 | ||
d5e3ed66 JDB |
3389 | if (unlikely(flags & __GFP_ZERO) && objp) |
3390 | memset(objp, 0, cachep->object_size); | |
d07dbea4 | 3391 | |
d5e3ed66 | 3392 | slab_post_alloc_hook(cachep, flags, 1, &objp); |
8c8cc2c1 PE |
3393 | return objp; |
3394 | } | |
e498be7d CL |
3395 | |
3396 | /* | |
5f0985bb | 3397 | * Caller needs to acquire correct kmem_cache_node's list_lock |
97654dfa | 3398 | * @list: List of detached free slabs should be freed by caller |
e498be7d | 3399 | */ |
97654dfa JK |
3400 | static void free_block(struct kmem_cache *cachep, void **objpp, |
3401 | int nr_objects, int node, struct list_head *list) | |
1da177e4 LT |
3402 | { |
3403 | int i; | |
25c063fb | 3404 | struct kmem_cache_node *n = get_node(cachep, node); |
6052b788 JK |
3405 | struct page *page; |
3406 | ||
3407 | n->free_objects += nr_objects; | |
1da177e4 LT |
3408 | |
3409 | for (i = 0; i < nr_objects; i++) { | |
072bb0aa | 3410 | void *objp; |
8456a648 | 3411 | struct page *page; |
1da177e4 | 3412 | |
072bb0aa MG |
3413 | objp = objpp[i]; |
3414 | ||
8456a648 | 3415 | page = virt_to_head_page(objp); |
8456a648 | 3416 | list_del(&page->lru); |
ff69416e | 3417 | check_spinlock_acquired_node(cachep, node); |
260b61dd | 3418 | slab_put_obj(cachep, page, objp); |
1da177e4 | 3419 | STATS_DEC_ACTIVE(cachep); |
1da177e4 LT |
3420 | |
3421 | /* fixup slab chains */ | |
f728b0a5 | 3422 | if (page->active == 0) { |
6052b788 | 3423 | list_add(&page->lru, &n->slabs_free); |
f728b0a5 | 3424 | n->free_slabs++; |
f728b0a5 | 3425 | } else { |
1da177e4 LT |
3426 | /* Unconditionally move a slab to the end of the |
3427 | * partial list on free - maximum time for the | |
3428 | * other objects to be freed, too. | |
3429 | */ | |
8456a648 | 3430 | list_add_tail(&page->lru, &n->slabs_partial); |
1da177e4 LT |
3431 | } |
3432 | } | |
6052b788 JK |
3433 | |
3434 | while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) { | |
3435 | n->free_objects -= cachep->num; | |
3436 | ||
3437 | page = list_last_entry(&n->slabs_free, struct page, lru); | |
de24baec | 3438 | list_move(&page->lru, list); |
f728b0a5 | 3439 | n->free_slabs--; |
bf00bd34 | 3440 | n->total_slabs--; |
6052b788 | 3441 | } |
1da177e4 LT |
3442 | } |
3443 | ||
343e0d7a | 3444 | static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac) |
1da177e4 LT |
3445 | { |
3446 | int batchcount; | |
ce8eb6c4 | 3447 | struct kmem_cache_node *n; |
7d6e6d09 | 3448 | int node = numa_mem_id(); |
97654dfa | 3449 | LIST_HEAD(list); |
1da177e4 LT |
3450 | |
3451 | batchcount = ac->batchcount; | |
260b61dd | 3452 | |
1da177e4 | 3453 | check_irq_off(); |
18bf8541 | 3454 | n = get_node(cachep, node); |
ce8eb6c4 CL |
3455 | spin_lock(&n->list_lock); |
3456 | if (n->shared) { | |
3457 | struct array_cache *shared_array = n->shared; | |
b28a02de | 3458 | int max = shared_array->limit - shared_array->avail; |
1da177e4 LT |
3459 | if (max) { |
3460 | if (batchcount > max) | |
3461 | batchcount = max; | |
e498be7d | 3462 | memcpy(&(shared_array->entry[shared_array->avail]), |
b28a02de | 3463 | ac->entry, sizeof(void *) * batchcount); |
1da177e4 LT |
3464 | shared_array->avail += batchcount; |
3465 | goto free_done; | |
3466 | } | |
3467 | } | |
3468 | ||
97654dfa | 3469 | free_block(cachep, ac->entry, batchcount, node, &list); |
a737b3e2 | 3470 | free_done: |
1da177e4 LT |
3471 | #if STATS |
3472 | { | |
3473 | int i = 0; | |
73c0219d | 3474 | struct page *page; |
1da177e4 | 3475 | |
73c0219d | 3476 | list_for_each_entry(page, &n->slabs_free, lru) { |
8456a648 | 3477 | BUG_ON(page->active); |
1da177e4 LT |
3478 | |
3479 | i++; | |
1da177e4 LT |
3480 | } |
3481 | STATS_SET_FREEABLE(cachep, i); | |
3482 | } | |
3483 | #endif | |
ce8eb6c4 | 3484 | spin_unlock(&n->list_lock); |
97654dfa | 3485 | slabs_destroy(cachep, &list); |
1da177e4 | 3486 | ac->avail -= batchcount; |
a737b3e2 | 3487 | memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail); |
1da177e4 LT |
3488 | } |
3489 | ||
3490 | /* | |
a737b3e2 AM |
3491 | * Release an obj back to its cache. If the obj has a constructed state, it must |
3492 | * be in this state _before_ it is released. Called with disabled ints. | |
1da177e4 | 3493 | */ |
ee3ce779 DV |
3494 | static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp, |
3495 | unsigned long caller) | |
1da177e4 | 3496 | { |
55834c59 | 3497 | /* Put the object into the quarantine, don't touch it for now. */ |
ee3ce779 | 3498 | if (kasan_slab_free(cachep, objp, _RET_IP_)) |
55834c59 AP |
3499 | return; |
3500 | ||
3501 | ___cache_free(cachep, objp, caller); | |
3502 | } | |
1da177e4 | 3503 | |
55834c59 AP |
3504 | void ___cache_free(struct kmem_cache *cachep, void *objp, |
3505 | unsigned long caller) | |
3506 | { | |
3507 | struct array_cache *ac = cpu_cache_get(cachep); | |
7ed2f9e6 | 3508 | |
1da177e4 | 3509 | check_irq_off(); |
d5cff635 | 3510 | kmemleak_free_recursive(objp, cachep->flags); |
a947eb95 | 3511 | objp = cache_free_debugcheck(cachep, objp, caller); |
1da177e4 | 3512 | |
1807a1aa SS |
3513 | /* |
3514 | * Skip calling cache_free_alien() when the platform is not numa. | |
3515 | * This will avoid cache misses that happen while accessing slabp (which | |
3516 | * is per page memory reference) to get nodeid. Instead use a global | |
3517 | * variable to skip the call, which is mostly likely to be present in | |
3518 | * the cache. | |
3519 | */ | |
b6e68bc1 | 3520 | if (nr_online_nodes > 1 && cache_free_alien(cachep, objp)) |
729bd0b7 PE |
3521 | return; |
3522 | ||
3d880194 | 3523 | if (ac->avail < ac->limit) { |
1da177e4 | 3524 | STATS_INC_FREEHIT(cachep); |
1da177e4 LT |
3525 | } else { |
3526 | STATS_INC_FREEMISS(cachep); | |
3527 | cache_flusharray(cachep, ac); | |
1da177e4 | 3528 | } |
42c8c99c | 3529 | |
f68f8ddd JK |
3530 | if (sk_memalloc_socks()) { |
3531 | struct page *page = virt_to_head_page(objp); | |
3532 | ||
3533 | if (unlikely(PageSlabPfmemalloc(page))) { | |
3534 | cache_free_pfmemalloc(cachep, page, objp); | |
3535 | return; | |
3536 | } | |
3537 | } | |
3538 | ||
3539 | ac->entry[ac->avail++] = objp; | |
1da177e4 LT |
3540 | } |
3541 | ||
3542 | /** | |
3543 | * kmem_cache_alloc - Allocate an object | |
3544 | * @cachep: The cache to allocate from. | |
3545 | * @flags: See kmalloc(). | |
3546 | * | |
3547 | * Allocate an object from this cache. The flags are only relevant | |
3548 | * if the cache has no available objects. | |
3549 | */ | |
343e0d7a | 3550 | void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags) |
1da177e4 | 3551 | { |
48356303 | 3552 | void *ret = slab_alloc(cachep, flags, _RET_IP_); |
36555751 | 3553 | |
505f5dcb | 3554 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3555 | trace_kmem_cache_alloc(_RET_IP_, ret, |
8c138bc0 | 3556 | cachep->object_size, cachep->size, flags); |
36555751 EGM |
3557 | |
3558 | return ret; | |
1da177e4 LT |
3559 | } |
3560 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3561 | ||
7b0501dd JDB |
3562 | static __always_inline void |
3563 | cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags, | |
3564 | size_t size, void **p, unsigned long caller) | |
3565 | { | |
3566 | size_t i; | |
3567 | ||
3568 | for (i = 0; i < size; i++) | |
3569 | p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller); | |
3570 | } | |
3571 | ||
865762a8 | 3572 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2a777eac | 3573 | void **p) |
484748f0 | 3574 | { |
2a777eac JDB |
3575 | size_t i; |
3576 | ||
3577 | s = slab_pre_alloc_hook(s, flags); | |
3578 | if (!s) | |
3579 | return 0; | |
3580 | ||
3581 | cache_alloc_debugcheck_before(s, flags); | |
3582 | ||
3583 | local_irq_disable(); | |
3584 | for (i = 0; i < size; i++) { | |
3585 | void *objp = __do_cache_alloc(s, flags); | |
3586 | ||
2a777eac JDB |
3587 | if (unlikely(!objp)) |
3588 | goto error; | |
3589 | p[i] = objp; | |
3590 | } | |
3591 | local_irq_enable(); | |
3592 | ||
7b0501dd JDB |
3593 | cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_); |
3594 | ||
2a777eac JDB |
3595 | /* Clear memory outside IRQ disabled section */ |
3596 | if (unlikely(flags & __GFP_ZERO)) | |
3597 | for (i = 0; i < size; i++) | |
3598 | memset(p[i], 0, s->object_size); | |
3599 | ||
3600 | slab_post_alloc_hook(s, flags, size, p); | |
3601 | /* FIXME: Trace call missing. Christoph would like a bulk variant */ | |
3602 | return size; | |
3603 | error: | |
3604 | local_irq_enable(); | |
7b0501dd | 3605 | cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_); |
2a777eac JDB |
3606 | slab_post_alloc_hook(s, flags, i, p); |
3607 | __kmem_cache_free_bulk(s, i, p); | |
3608 | return 0; | |
484748f0 CL |
3609 | } |
3610 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3611 | ||
0f24f128 | 3612 | #ifdef CONFIG_TRACING |
85beb586 | 3613 | void * |
4052147c | 3614 | kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size) |
36555751 | 3615 | { |
85beb586 SR |
3616 | void *ret; |
3617 | ||
48356303 | 3618 | ret = slab_alloc(cachep, flags, _RET_IP_); |
85beb586 | 3619 | |
505f5dcb | 3620 | kasan_kmalloc(cachep, ret, size, flags); |
85beb586 | 3621 | trace_kmalloc(_RET_IP_, ret, |
ff4fcd01 | 3622 | size, cachep->size, flags); |
85beb586 | 3623 | return ret; |
36555751 | 3624 | } |
85beb586 | 3625 | EXPORT_SYMBOL(kmem_cache_alloc_trace); |
36555751 EGM |
3626 | #endif |
3627 | ||
1da177e4 | 3628 | #ifdef CONFIG_NUMA |
d0d04b78 ZL |
3629 | /** |
3630 | * kmem_cache_alloc_node - Allocate an object on the specified node | |
3631 | * @cachep: The cache to allocate from. | |
3632 | * @flags: See kmalloc(). | |
3633 | * @nodeid: node number of the target node. | |
3634 | * | |
3635 | * Identical to kmem_cache_alloc but it will allocate memory on the given | |
3636 | * node, which can improve the performance for cpu bound structures. | |
3637 | * | |
3638 | * Fallback to other node is possible if __GFP_THISNODE is not set. | |
3639 | */ | |
8b98c169 CH |
3640 | void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid) |
3641 | { | |
48356303 | 3642 | void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
36555751 | 3643 | |
505f5dcb | 3644 | kasan_slab_alloc(cachep, ret, flags); |
ca2b84cb | 3645 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
8c138bc0 | 3646 | cachep->object_size, cachep->size, |
ca2b84cb | 3647 | flags, nodeid); |
36555751 EGM |
3648 | |
3649 | return ret; | |
8b98c169 | 3650 | } |
1da177e4 LT |
3651 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
3652 | ||
0f24f128 | 3653 | #ifdef CONFIG_TRACING |
4052147c | 3654 | void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep, |
85beb586 | 3655 | gfp_t flags, |
4052147c EG |
3656 | int nodeid, |
3657 | size_t size) | |
36555751 | 3658 | { |
85beb586 SR |
3659 | void *ret; |
3660 | ||
592f4145 | 3661 | ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_); |
505f5dcb AP |
3662 | |
3663 | kasan_kmalloc(cachep, ret, size, flags); | |
85beb586 | 3664 | trace_kmalloc_node(_RET_IP_, ret, |
ff4fcd01 | 3665 | size, cachep->size, |
85beb586 SR |
3666 | flags, nodeid); |
3667 | return ret; | |
36555751 | 3668 | } |
85beb586 | 3669 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
36555751 EGM |
3670 | #endif |
3671 | ||
8b98c169 | 3672 | static __always_inline void * |
7c0cb9c6 | 3673 | __do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) |
97e2bde4 | 3674 | { |
343e0d7a | 3675 | struct kmem_cache *cachep; |
7ed2f9e6 | 3676 | void *ret; |
97e2bde4 | 3677 | |
61448479 DV |
3678 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3679 | return NULL; | |
2c59dd65 | 3680 | cachep = kmalloc_slab(size, flags); |
6cb8f913 CL |
3681 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3682 | return cachep; | |
7ed2f9e6 | 3683 | ret = kmem_cache_alloc_node_trace(cachep, flags, node, size); |
505f5dcb | 3684 | kasan_kmalloc(cachep, ret, size, flags); |
7ed2f9e6 AP |
3685 | |
3686 | return ret; | |
97e2bde4 | 3687 | } |
8b98c169 | 3688 | |
8b98c169 CH |
3689 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3690 | { | |
7c0cb9c6 | 3691 | return __do_kmalloc_node(size, flags, node, _RET_IP_); |
8b98c169 | 3692 | } |
dbe5e69d | 3693 | EXPORT_SYMBOL(__kmalloc_node); |
8b98c169 CH |
3694 | |
3695 | void *__kmalloc_node_track_caller(size_t size, gfp_t flags, | |
ce71e27c | 3696 | int node, unsigned long caller) |
8b98c169 | 3697 | { |
7c0cb9c6 | 3698 | return __do_kmalloc_node(size, flags, node, caller); |
8b98c169 CH |
3699 | } |
3700 | EXPORT_SYMBOL(__kmalloc_node_track_caller); | |
8b98c169 | 3701 | #endif /* CONFIG_NUMA */ |
1da177e4 LT |
3702 | |
3703 | /** | |
800590f5 | 3704 | * __do_kmalloc - allocate memory |
1da177e4 | 3705 | * @size: how many bytes of memory are required. |
800590f5 | 3706 | * @flags: the type of memory to allocate (see kmalloc). |
911851e6 | 3707 | * @caller: function caller for debug tracking of the caller |
1da177e4 | 3708 | */ |
7fd6b141 | 3709 | static __always_inline void *__do_kmalloc(size_t size, gfp_t flags, |
7c0cb9c6 | 3710 | unsigned long caller) |
1da177e4 | 3711 | { |
343e0d7a | 3712 | struct kmem_cache *cachep; |
36555751 | 3713 | void *ret; |
1da177e4 | 3714 | |
61448479 DV |
3715 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
3716 | return NULL; | |
2c59dd65 | 3717 | cachep = kmalloc_slab(size, flags); |
a5c96d8a LT |
3718 | if (unlikely(ZERO_OR_NULL_PTR(cachep))) |
3719 | return cachep; | |
48356303 | 3720 | ret = slab_alloc(cachep, flags, caller); |
36555751 | 3721 | |
505f5dcb | 3722 | kasan_kmalloc(cachep, ret, size, flags); |
7c0cb9c6 | 3723 | trace_kmalloc(caller, ret, |
3b0efdfa | 3724 | size, cachep->size, flags); |
36555751 EGM |
3725 | |
3726 | return ret; | |
7fd6b141 PE |
3727 | } |
3728 | ||
7fd6b141 PE |
3729 | void *__kmalloc(size_t size, gfp_t flags) |
3730 | { | |
7c0cb9c6 | 3731 | return __do_kmalloc(size, flags, _RET_IP_); |
1da177e4 LT |
3732 | } |
3733 | EXPORT_SYMBOL(__kmalloc); | |
3734 | ||
ce71e27c | 3735 | void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller) |
7fd6b141 | 3736 | { |
7c0cb9c6 | 3737 | return __do_kmalloc(size, flags, caller); |
7fd6b141 PE |
3738 | } |
3739 | EXPORT_SYMBOL(__kmalloc_track_caller); | |
1d2c8eea | 3740 | |
1da177e4 LT |
3741 | /** |
3742 | * kmem_cache_free - Deallocate an object | |
3743 | * @cachep: The cache the allocation was from. | |
3744 | * @objp: The previously allocated object. | |
3745 | * | |
3746 | * Free an object which was previously allocated from this | |
3747 | * cache. | |
3748 | */ | |
343e0d7a | 3749 | void kmem_cache_free(struct kmem_cache *cachep, void *objp) |
1da177e4 LT |
3750 | { |
3751 | unsigned long flags; | |
b9ce5ef4 GC |
3752 | cachep = cache_from_obj(cachep, objp); |
3753 | if (!cachep) | |
3754 | return; | |
1da177e4 LT |
3755 | |
3756 | local_irq_save(flags); | |
d97d476b | 3757 | debug_check_no_locks_freed(objp, cachep->object_size); |
3ac7fe5a | 3758 | if (!(cachep->flags & SLAB_DEBUG_OBJECTS)) |
8c138bc0 | 3759 | debug_check_no_obj_freed(objp, cachep->object_size); |
7c0cb9c6 | 3760 | __cache_free(cachep, objp, _RET_IP_); |
1da177e4 | 3761 | local_irq_restore(flags); |
36555751 | 3762 | |
ca2b84cb | 3763 | trace_kmem_cache_free(_RET_IP_, objp); |
1da177e4 LT |
3764 | } |
3765 | EXPORT_SYMBOL(kmem_cache_free); | |
3766 | ||
e6cdb58d JDB |
3767 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
3768 | { | |
3769 | struct kmem_cache *s; | |
3770 | size_t i; | |
3771 | ||
3772 | local_irq_disable(); | |
3773 | for (i = 0; i < size; i++) { | |
3774 | void *objp = p[i]; | |
3775 | ||
ca257195 JDB |
3776 | if (!orig_s) /* called via kfree_bulk */ |
3777 | s = virt_to_cache(objp); | |
3778 | else | |
3779 | s = cache_from_obj(orig_s, objp); | |
e6cdb58d JDB |
3780 | |
3781 | debug_check_no_locks_freed(objp, s->object_size); | |
3782 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
3783 | debug_check_no_obj_freed(objp, s->object_size); | |
3784 | ||
3785 | __cache_free(s, objp, _RET_IP_); | |
3786 | } | |
3787 | local_irq_enable(); | |
3788 | ||
3789 | /* FIXME: add tracing */ | |
3790 | } | |
3791 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3792 | ||
1da177e4 LT |
3793 | /** |
3794 | * kfree - free previously allocated memory | |
3795 | * @objp: pointer returned by kmalloc. | |
3796 | * | |
80e93eff PE |
3797 | * If @objp is NULL, no operation is performed. |
3798 | * | |
1da177e4 LT |
3799 | * Don't free memory not originally allocated by kmalloc() |
3800 | * or you will run into trouble. | |
3801 | */ | |
3802 | void kfree(const void *objp) | |
3803 | { | |
343e0d7a | 3804 | struct kmem_cache *c; |
1da177e4 LT |
3805 | unsigned long flags; |
3806 | ||
2121db74 PE |
3807 | trace_kfree(_RET_IP_, objp); |
3808 | ||
6cb8f913 | 3809 | if (unlikely(ZERO_OR_NULL_PTR(objp))) |
1da177e4 LT |
3810 | return; |
3811 | local_irq_save(flags); | |
3812 | kfree_debugcheck(objp); | |
6ed5eb22 | 3813 | c = virt_to_cache(objp); |
8c138bc0 CL |
3814 | debug_check_no_locks_freed(objp, c->object_size); |
3815 | ||
3816 | debug_check_no_obj_freed(objp, c->object_size); | |
7c0cb9c6 | 3817 | __cache_free(c, (void *)objp, _RET_IP_); |
1da177e4 LT |
3818 | local_irq_restore(flags); |
3819 | } | |
3820 | EXPORT_SYMBOL(kfree); | |
3821 | ||
e498be7d | 3822 | /* |
ce8eb6c4 | 3823 | * This initializes kmem_cache_node or resizes various caches for all nodes. |
e498be7d | 3824 | */ |
c3d332b6 | 3825 | static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp) |
e498be7d | 3826 | { |
c3d332b6 | 3827 | int ret; |
e498be7d | 3828 | int node; |
ce8eb6c4 | 3829 | struct kmem_cache_node *n; |
e498be7d | 3830 | |
9c09a95c | 3831 | for_each_online_node(node) { |
c3d332b6 JK |
3832 | ret = setup_kmem_cache_node(cachep, node, gfp, true); |
3833 | if (ret) | |
e498be7d CL |
3834 | goto fail; |
3835 | ||
e498be7d | 3836 | } |
c3d332b6 | 3837 | |
cafeb02e | 3838 | return 0; |
0718dc2a | 3839 | |
a737b3e2 | 3840 | fail: |
3b0efdfa | 3841 | if (!cachep->list.next) { |
0718dc2a CL |
3842 | /* Cache is not active yet. Roll back what we did */ |
3843 | node--; | |
3844 | while (node >= 0) { | |
18bf8541 CL |
3845 | n = get_node(cachep, node); |
3846 | if (n) { | |
ce8eb6c4 CL |
3847 | kfree(n->shared); |
3848 | free_alien_cache(n->alien); | |
3849 | kfree(n); | |
6a67368c | 3850 | cachep->node[node] = NULL; |
0718dc2a CL |
3851 | } |
3852 | node--; | |
3853 | } | |
3854 | } | |
cafeb02e | 3855 | return -ENOMEM; |
e498be7d CL |
3856 | } |
3857 | ||
18004c5d | 3858 | /* Always called with the slab_mutex held */ |
943a451a | 3859 | static int __do_tune_cpucache(struct kmem_cache *cachep, int limit, |
83b519e8 | 3860 | int batchcount, int shared, gfp_t gfp) |
1da177e4 | 3861 | { |
bf0dea23 JK |
3862 | struct array_cache __percpu *cpu_cache, *prev; |
3863 | int cpu; | |
1da177e4 | 3864 | |
bf0dea23 JK |
3865 | cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount); |
3866 | if (!cpu_cache) | |
d2e7b7d0 SS |
3867 | return -ENOMEM; |
3868 | ||
bf0dea23 JK |
3869 | prev = cachep->cpu_cache; |
3870 | cachep->cpu_cache = cpu_cache; | |
a87c75fb GT |
3871 | /* |
3872 | * Without a previous cpu_cache there's no need to synchronize remote | |
3873 | * cpus, so skip the IPIs. | |
3874 | */ | |
3875 | if (prev) | |
3876 | kick_all_cpus_sync(); | |
e498be7d | 3877 | |
1da177e4 | 3878 | check_irq_on(); |
1da177e4 LT |
3879 | cachep->batchcount = batchcount; |
3880 | cachep->limit = limit; | |
e498be7d | 3881 | cachep->shared = shared; |
1da177e4 | 3882 | |
bf0dea23 | 3883 | if (!prev) |
c3d332b6 | 3884 | goto setup_node; |
bf0dea23 JK |
3885 | |
3886 | for_each_online_cpu(cpu) { | |
97654dfa | 3887 | LIST_HEAD(list); |
18bf8541 CL |
3888 | int node; |
3889 | struct kmem_cache_node *n; | |
bf0dea23 | 3890 | struct array_cache *ac = per_cpu_ptr(prev, cpu); |
18bf8541 | 3891 | |
bf0dea23 | 3892 | node = cpu_to_mem(cpu); |
18bf8541 CL |
3893 | n = get_node(cachep, node); |
3894 | spin_lock_irq(&n->list_lock); | |
bf0dea23 | 3895 | free_block(cachep, ac->entry, ac->avail, node, &list); |
18bf8541 | 3896 | spin_unlock_irq(&n->list_lock); |
97654dfa | 3897 | slabs_destroy(cachep, &list); |
1da177e4 | 3898 | } |
bf0dea23 JK |
3899 | free_percpu(prev); |
3900 | ||
c3d332b6 JK |
3901 | setup_node: |
3902 | return setup_kmem_cache_nodes(cachep, gfp); | |
1da177e4 LT |
3903 | } |
3904 | ||
943a451a GC |
3905 | static int do_tune_cpucache(struct kmem_cache *cachep, int limit, |
3906 | int batchcount, int shared, gfp_t gfp) | |
3907 | { | |
3908 | int ret; | |
426589f5 | 3909 | struct kmem_cache *c; |
943a451a GC |
3910 | |
3911 | ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
3912 | ||
3913 | if (slab_state < FULL) | |
3914 | return ret; | |
3915 | ||
3916 | if ((ret < 0) || !is_root_cache(cachep)) | |
3917 | return ret; | |
3918 | ||
426589f5 VD |
3919 | lockdep_assert_held(&slab_mutex); |
3920 | for_each_memcg_cache(c, cachep) { | |
3921 | /* return value determined by the root cache only */ | |
3922 | __do_tune_cpucache(c, limit, batchcount, shared, gfp); | |
943a451a GC |
3923 | } |
3924 | ||
3925 | return ret; | |
3926 | } | |
3927 | ||
18004c5d | 3928 | /* Called with slab_mutex held always */ |
83b519e8 | 3929 | static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp) |
1da177e4 LT |
3930 | { |
3931 | int err; | |
943a451a GC |
3932 | int limit = 0; |
3933 | int shared = 0; | |
3934 | int batchcount = 0; | |
3935 | ||
7c00fce9 | 3936 | err = cache_random_seq_create(cachep, cachep->num, gfp); |
c7ce4f60 TG |
3937 | if (err) |
3938 | goto end; | |
3939 | ||
943a451a GC |
3940 | if (!is_root_cache(cachep)) { |
3941 | struct kmem_cache *root = memcg_root_cache(cachep); | |
3942 | limit = root->limit; | |
3943 | shared = root->shared; | |
3944 | batchcount = root->batchcount; | |
3945 | } | |
1da177e4 | 3946 | |
943a451a GC |
3947 | if (limit && shared && batchcount) |
3948 | goto skip_setup; | |
a737b3e2 AM |
3949 | /* |
3950 | * The head array serves three purposes: | |
1da177e4 LT |
3951 | * - create a LIFO ordering, i.e. return objects that are cache-warm |
3952 | * - reduce the number of spinlock operations. | |
a737b3e2 | 3953 | * - reduce the number of linked list operations on the slab and |
1da177e4 LT |
3954 | * bufctl chains: array operations are cheaper. |
3955 | * The numbers are guessed, we should auto-tune as described by | |
3956 | * Bonwick. | |
3957 | */ | |
3b0efdfa | 3958 | if (cachep->size > 131072) |
1da177e4 | 3959 | limit = 1; |
3b0efdfa | 3960 | else if (cachep->size > PAGE_SIZE) |
1da177e4 | 3961 | limit = 8; |
3b0efdfa | 3962 | else if (cachep->size > 1024) |
1da177e4 | 3963 | limit = 24; |
3b0efdfa | 3964 | else if (cachep->size > 256) |
1da177e4 LT |
3965 | limit = 54; |
3966 | else | |
3967 | limit = 120; | |
3968 | ||
a737b3e2 AM |
3969 | /* |
3970 | * CPU bound tasks (e.g. network routing) can exhibit cpu bound | |
1da177e4 LT |
3971 | * allocation behaviour: Most allocs on one cpu, most free operations |
3972 | * on another cpu. For these cases, an efficient object passing between | |
3973 | * cpus is necessary. This is provided by a shared array. The array | |
3974 | * replaces Bonwick's magazine layer. | |
3975 | * On uniprocessor, it's functionally equivalent (but less efficient) | |
3976 | * to a larger limit. Thus disabled by default. | |
3977 | */ | |
3978 | shared = 0; | |
3b0efdfa | 3979 | if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1) |
1da177e4 | 3980 | shared = 8; |
1da177e4 LT |
3981 | |
3982 | #if DEBUG | |
a737b3e2 AM |
3983 | /* |
3984 | * With debugging enabled, large batchcount lead to excessively long | |
3985 | * periods with disabled local interrupts. Limit the batchcount | |
1da177e4 LT |
3986 | */ |
3987 | if (limit > 32) | |
3988 | limit = 32; | |
3989 | #endif | |
943a451a GC |
3990 | batchcount = (limit + 1) / 2; |
3991 | skip_setup: | |
3992 | err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp); | |
c7ce4f60 | 3993 | end: |
1da177e4 | 3994 | if (err) |
1170532b | 3995 | pr_err("enable_cpucache failed for %s, error %d\n", |
b28a02de | 3996 | cachep->name, -err); |
2ed3a4ef | 3997 | return err; |
1da177e4 LT |
3998 | } |
3999 | ||
1b55253a | 4000 | /* |
ce8eb6c4 CL |
4001 | * Drain an array if it contains any elements taking the node lock only if |
4002 | * necessary. Note that the node listlock also protects the array_cache | |
b18e7e65 | 4003 | * if drain_array() is used on the shared array. |
1b55253a | 4004 | */ |
ce8eb6c4 | 4005 | static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n, |
18726ca8 | 4006 | struct array_cache *ac, int node) |
1da177e4 | 4007 | { |
97654dfa | 4008 | LIST_HEAD(list); |
18726ca8 JK |
4009 | |
4010 | /* ac from n->shared can be freed if we don't hold the slab_mutex. */ | |
4011 | check_mutex_acquired(); | |
1da177e4 | 4012 | |
1b55253a CL |
4013 | if (!ac || !ac->avail) |
4014 | return; | |
18726ca8 JK |
4015 | |
4016 | if (ac->touched) { | |
1da177e4 | 4017 | ac->touched = 0; |
18726ca8 | 4018 | return; |
1da177e4 | 4019 | } |
18726ca8 JK |
4020 | |
4021 | spin_lock_irq(&n->list_lock); | |
4022 | drain_array_locked(cachep, ac, node, false, &list); | |
4023 | spin_unlock_irq(&n->list_lock); | |
4024 | ||
4025 | slabs_destroy(cachep, &list); | |
1da177e4 LT |
4026 | } |
4027 | ||
4028 | /** | |
4029 | * cache_reap - Reclaim memory from caches. | |
05fb6bf0 | 4030 | * @w: work descriptor |
1da177e4 LT |
4031 | * |
4032 | * Called from workqueue/eventd every few seconds. | |
4033 | * Purpose: | |
4034 | * - clear the per-cpu caches for this CPU. | |
4035 | * - return freeable pages to the main free memory pool. | |
4036 | * | |
a737b3e2 AM |
4037 | * If we cannot acquire the cache chain mutex then just give up - we'll try |
4038 | * again on the next iteration. | |
1da177e4 | 4039 | */ |
7c5cae36 | 4040 | static void cache_reap(struct work_struct *w) |
1da177e4 | 4041 | { |
7a7c381d | 4042 | struct kmem_cache *searchp; |
ce8eb6c4 | 4043 | struct kmem_cache_node *n; |
7d6e6d09 | 4044 | int node = numa_mem_id(); |
bf6aede7 | 4045 | struct delayed_work *work = to_delayed_work(w); |
1da177e4 | 4046 | |
18004c5d | 4047 | if (!mutex_trylock(&slab_mutex)) |
1da177e4 | 4048 | /* Give up. Setup the next iteration. */ |
7c5cae36 | 4049 | goto out; |
1da177e4 | 4050 | |
18004c5d | 4051 | list_for_each_entry(searchp, &slab_caches, list) { |
1da177e4 LT |
4052 | check_irq_on(); |
4053 | ||
35386e3b | 4054 | /* |
ce8eb6c4 | 4055 | * We only take the node lock if absolutely necessary and we |
35386e3b CL |
4056 | * have established with reasonable certainty that |
4057 | * we can do some work if the lock was obtained. | |
4058 | */ | |
18bf8541 | 4059 | n = get_node(searchp, node); |
35386e3b | 4060 | |
ce8eb6c4 | 4061 | reap_alien(searchp, n); |
1da177e4 | 4062 | |
18726ca8 | 4063 | drain_array(searchp, n, cpu_cache_get(searchp), node); |
1da177e4 | 4064 | |
35386e3b CL |
4065 | /* |
4066 | * These are racy checks but it does not matter | |
4067 | * if we skip one check or scan twice. | |
4068 | */ | |
ce8eb6c4 | 4069 | if (time_after(n->next_reap, jiffies)) |
35386e3b | 4070 | goto next; |
1da177e4 | 4071 | |
5f0985bb | 4072 | n->next_reap = jiffies + REAPTIMEOUT_NODE; |
1da177e4 | 4073 | |
18726ca8 | 4074 | drain_array(searchp, n, n->shared, node); |
1da177e4 | 4075 | |
ce8eb6c4 CL |
4076 | if (n->free_touched) |
4077 | n->free_touched = 0; | |
ed11d9eb CL |
4078 | else { |
4079 | int freed; | |
1da177e4 | 4080 | |
ce8eb6c4 | 4081 | freed = drain_freelist(searchp, n, (n->free_limit + |
ed11d9eb CL |
4082 | 5 * searchp->num - 1) / (5 * searchp->num)); |
4083 | STATS_ADD_REAPED(searchp, freed); | |
4084 | } | |
35386e3b | 4085 | next: |
1da177e4 LT |
4086 | cond_resched(); |
4087 | } | |
4088 | check_irq_on(); | |
18004c5d | 4089 | mutex_unlock(&slab_mutex); |
8fce4d8e | 4090 | next_reap_node(); |
7c5cae36 | 4091 | out: |
a737b3e2 | 4092 | /* Set up the next iteration */ |
a9f2a846 VB |
4093 | schedule_delayed_work_on(smp_processor_id(), work, |
4094 | round_jiffies_relative(REAPTIMEOUT_AC)); | |
1da177e4 LT |
4095 | } |
4096 | ||
0d7561c6 | 4097 | void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo) |
1da177e4 | 4098 | { |
f728b0a5 | 4099 | unsigned long active_objs, num_objs, active_slabs; |
bf00bd34 DR |
4100 | unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0; |
4101 | unsigned long free_slabs = 0; | |
e498be7d | 4102 | int node; |
ce8eb6c4 | 4103 | struct kmem_cache_node *n; |
1da177e4 | 4104 | |
18bf8541 | 4105 | for_each_kmem_cache_node(cachep, node, n) { |
ca3b9b91 | 4106 | check_irq_on(); |
ce8eb6c4 | 4107 | spin_lock_irq(&n->list_lock); |
e498be7d | 4108 | |
bf00bd34 DR |
4109 | total_slabs += n->total_slabs; |
4110 | free_slabs += n->free_slabs; | |
f728b0a5 | 4111 | free_objs += n->free_objects; |
07a63c41 | 4112 | |
ce8eb6c4 CL |
4113 | if (n->shared) |
4114 | shared_avail += n->shared->avail; | |
e498be7d | 4115 | |
ce8eb6c4 | 4116 | spin_unlock_irq(&n->list_lock); |
1da177e4 | 4117 | } |
bf00bd34 DR |
4118 | num_objs = total_slabs * cachep->num; |
4119 | active_slabs = total_slabs - free_slabs; | |
f728b0a5 | 4120 | active_objs = num_objs - free_objs; |
1da177e4 | 4121 | |
0d7561c6 GC |
4122 | sinfo->active_objs = active_objs; |
4123 | sinfo->num_objs = num_objs; | |
4124 | sinfo->active_slabs = active_slabs; | |
bf00bd34 | 4125 | sinfo->num_slabs = total_slabs; |
0d7561c6 GC |
4126 | sinfo->shared_avail = shared_avail; |
4127 | sinfo->limit = cachep->limit; | |
4128 | sinfo->batchcount = cachep->batchcount; | |
4129 | sinfo->shared = cachep->shared; | |
4130 | sinfo->objects_per_slab = cachep->num; | |
4131 | sinfo->cache_order = cachep->gfporder; | |
4132 | } | |
4133 | ||
4134 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep) | |
4135 | { | |
1da177e4 | 4136 | #if STATS |
ce8eb6c4 | 4137 | { /* node stats */ |
1da177e4 LT |
4138 | unsigned long high = cachep->high_mark; |
4139 | unsigned long allocs = cachep->num_allocations; | |
4140 | unsigned long grown = cachep->grown; | |
4141 | unsigned long reaped = cachep->reaped; | |
4142 | unsigned long errors = cachep->errors; | |
4143 | unsigned long max_freeable = cachep->max_freeable; | |
1da177e4 | 4144 | unsigned long node_allocs = cachep->node_allocs; |
e498be7d | 4145 | unsigned long node_frees = cachep->node_frees; |
fb7faf33 | 4146 | unsigned long overflows = cachep->node_overflow; |
1da177e4 | 4147 | |
756a025f | 4148 | seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu", |
e92dd4fd JP |
4149 | allocs, high, grown, |
4150 | reaped, errors, max_freeable, node_allocs, | |
4151 | node_frees, overflows); | |
1da177e4 LT |
4152 | } |
4153 | /* cpu stats */ | |
4154 | { | |
4155 | unsigned long allochit = atomic_read(&cachep->allochit); | |
4156 | unsigned long allocmiss = atomic_read(&cachep->allocmiss); | |
4157 | unsigned long freehit = atomic_read(&cachep->freehit); | |
4158 | unsigned long freemiss = atomic_read(&cachep->freemiss); | |
4159 | ||
4160 | seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu", | |
b28a02de | 4161 | allochit, allocmiss, freehit, freemiss); |
1da177e4 LT |
4162 | } |
4163 | #endif | |
1da177e4 LT |
4164 | } |
4165 | ||
1da177e4 LT |
4166 | #define MAX_SLABINFO_WRITE 128 |
4167 | /** | |
4168 | * slabinfo_write - Tuning for the slab allocator | |
4169 | * @file: unused | |
4170 | * @buffer: user buffer | |
4171 | * @count: data length | |
4172 | * @ppos: unused | |
4173 | */ | |
b7454ad3 | 4174 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
b28a02de | 4175 | size_t count, loff_t *ppos) |
1da177e4 | 4176 | { |
b28a02de | 4177 | char kbuf[MAX_SLABINFO_WRITE + 1], *tmp; |
1da177e4 | 4178 | int limit, batchcount, shared, res; |
7a7c381d | 4179 | struct kmem_cache *cachep; |
b28a02de | 4180 | |
1da177e4 LT |
4181 | if (count > MAX_SLABINFO_WRITE) |
4182 | return -EINVAL; | |
4183 | if (copy_from_user(&kbuf, buffer, count)) | |
4184 | return -EFAULT; | |
b28a02de | 4185 | kbuf[MAX_SLABINFO_WRITE] = '\0'; |
1da177e4 LT |
4186 | |
4187 | tmp = strchr(kbuf, ' '); | |
4188 | if (!tmp) | |
4189 | return -EINVAL; | |
4190 | *tmp = '\0'; | |
4191 | tmp++; | |
4192 | if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3) | |
4193 | return -EINVAL; | |
4194 | ||
4195 | /* Find the cache in the chain of caches. */ | |
18004c5d | 4196 | mutex_lock(&slab_mutex); |
1da177e4 | 4197 | res = -EINVAL; |
18004c5d | 4198 | list_for_each_entry(cachep, &slab_caches, list) { |
1da177e4 | 4199 | if (!strcmp(cachep->name, kbuf)) { |
a737b3e2 AM |
4200 | if (limit < 1 || batchcount < 1 || |
4201 | batchcount > limit || shared < 0) { | |
e498be7d | 4202 | res = 0; |
1da177e4 | 4203 | } else { |
e498be7d | 4204 | res = do_tune_cpucache(cachep, limit, |
83b519e8 PE |
4205 | batchcount, shared, |
4206 | GFP_KERNEL); | |
1da177e4 LT |
4207 | } |
4208 | break; | |
4209 | } | |
4210 | } | |
18004c5d | 4211 | mutex_unlock(&slab_mutex); |
1da177e4 LT |
4212 | if (res >= 0) |
4213 | res = count; | |
4214 | return res; | |
4215 | } | |
871751e2 AV |
4216 | |
4217 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4218 | ||
871751e2 AV |
4219 | static inline int add_caller(unsigned long *n, unsigned long v) |
4220 | { | |
4221 | unsigned long *p; | |
4222 | int l; | |
4223 | if (!v) | |
4224 | return 1; | |
4225 | l = n[1]; | |
4226 | p = n + 2; | |
4227 | while (l) { | |
4228 | int i = l/2; | |
4229 | unsigned long *q = p + 2 * i; | |
4230 | if (*q == v) { | |
4231 | q[1]++; | |
4232 | return 1; | |
4233 | } | |
4234 | if (*q > v) { | |
4235 | l = i; | |
4236 | } else { | |
4237 | p = q + 2; | |
4238 | l -= i + 1; | |
4239 | } | |
4240 | } | |
4241 | if (++n[1] == n[0]) | |
4242 | return 0; | |
4243 | memmove(p + 2, p, n[1] * 2 * sizeof(unsigned long) - ((void *)p - (void *)n)); | |
4244 | p[0] = v; | |
4245 | p[1] = 1; | |
4246 | return 1; | |
4247 | } | |
4248 | ||
8456a648 JK |
4249 | static void handle_slab(unsigned long *n, struct kmem_cache *c, |
4250 | struct page *page) | |
871751e2 AV |
4251 | { |
4252 | void *p; | |
d31676df JK |
4253 | int i, j; |
4254 | unsigned long v; | |
b1cb0982 | 4255 | |
871751e2 AV |
4256 | if (n[0] == n[1]) |
4257 | return; | |
8456a648 | 4258 | for (i = 0, p = page->s_mem; i < c->num; i++, p += c->size) { |
d31676df JK |
4259 | bool active = true; |
4260 | ||
4261 | for (j = page->active; j < c->num; j++) { | |
4262 | if (get_free_obj(page, j) == i) { | |
4263 | active = false; | |
4264 | break; | |
4265 | } | |
4266 | } | |
4267 | ||
4268 | if (!active) | |
871751e2 | 4269 | continue; |
b1cb0982 | 4270 | |
d31676df JK |
4271 | /* |
4272 | * probe_kernel_read() is used for DEBUG_PAGEALLOC. page table | |
4273 | * mapping is established when actual object allocation and | |
4274 | * we could mistakenly access the unmapped object in the cpu | |
4275 | * cache. | |
4276 | */ | |
4277 | if (probe_kernel_read(&v, dbg_userword(c, p), sizeof(v))) | |
4278 | continue; | |
4279 | ||
4280 | if (!add_caller(n, v)) | |
871751e2 AV |
4281 | return; |
4282 | } | |
4283 | } | |
4284 | ||
4285 | static void show_symbol(struct seq_file *m, unsigned long address) | |
4286 | { | |
4287 | #ifdef CONFIG_KALLSYMS | |
871751e2 | 4288 | unsigned long offset, size; |
9281acea | 4289 | char modname[MODULE_NAME_LEN], name[KSYM_NAME_LEN]; |
871751e2 | 4290 | |
a5c43dae | 4291 | if (lookup_symbol_attrs(address, &size, &offset, modname, name) == 0) { |
871751e2 | 4292 | seq_printf(m, "%s+%#lx/%#lx", name, offset, size); |
a5c43dae | 4293 | if (modname[0]) |
871751e2 AV |
4294 | seq_printf(m, " [%s]", modname); |
4295 | return; | |
4296 | } | |
4297 | #endif | |
85c3e4a5 | 4298 | seq_printf(m, "%px", (void *)address); |
871751e2 AV |
4299 | } |
4300 | ||
4301 | static int leaks_show(struct seq_file *m, void *p) | |
4302 | { | |
0672aa7c | 4303 | struct kmem_cache *cachep = list_entry(p, struct kmem_cache, list); |
8456a648 | 4304 | struct page *page; |
ce8eb6c4 | 4305 | struct kmem_cache_node *n; |
871751e2 | 4306 | const char *name; |
db845067 | 4307 | unsigned long *x = m->private; |
871751e2 AV |
4308 | int node; |
4309 | int i; | |
4310 | ||
4311 | if (!(cachep->flags & SLAB_STORE_USER)) | |
4312 | return 0; | |
4313 | if (!(cachep->flags & SLAB_RED_ZONE)) | |
4314 | return 0; | |
4315 | ||
d31676df JK |
4316 | /* |
4317 | * Set store_user_clean and start to grab stored user information | |
4318 | * for all objects on this cache. If some alloc/free requests comes | |
4319 | * during the processing, information would be wrong so restart | |
4320 | * whole processing. | |
4321 | */ | |
4322 | do { | |
4323 | set_store_user_clean(cachep); | |
4324 | drain_cpu_caches(cachep); | |
4325 | ||
4326 | x[1] = 0; | |
871751e2 | 4327 | |
d31676df | 4328 | for_each_kmem_cache_node(cachep, node, n) { |
871751e2 | 4329 | |
d31676df JK |
4330 | check_irq_on(); |
4331 | spin_lock_irq(&n->list_lock); | |
871751e2 | 4332 | |
d31676df JK |
4333 | list_for_each_entry(page, &n->slabs_full, lru) |
4334 | handle_slab(x, cachep, page); | |
4335 | list_for_each_entry(page, &n->slabs_partial, lru) | |
4336 | handle_slab(x, cachep, page); | |
4337 | spin_unlock_irq(&n->list_lock); | |
4338 | } | |
4339 | } while (!is_store_user_clean(cachep)); | |
871751e2 | 4340 | |
871751e2 | 4341 | name = cachep->name; |
db845067 | 4342 | if (x[0] == x[1]) { |
871751e2 | 4343 | /* Increase the buffer size */ |
18004c5d | 4344 | mutex_unlock(&slab_mutex); |
6396bb22 KC |
4345 | m->private = kcalloc(x[0] * 4, sizeof(unsigned long), |
4346 | GFP_KERNEL); | |
871751e2 AV |
4347 | if (!m->private) { |
4348 | /* Too bad, we are really out */ | |
db845067 | 4349 | m->private = x; |
18004c5d | 4350 | mutex_lock(&slab_mutex); |
871751e2 AV |
4351 | return -ENOMEM; |
4352 | } | |
db845067 CL |
4353 | *(unsigned long *)m->private = x[0] * 2; |
4354 | kfree(x); | |
18004c5d | 4355 | mutex_lock(&slab_mutex); |
871751e2 AV |
4356 | /* Now make sure this entry will be retried */ |
4357 | m->count = m->size; | |
4358 | return 0; | |
4359 | } | |
db845067 CL |
4360 | for (i = 0; i < x[1]; i++) { |
4361 | seq_printf(m, "%s: %lu ", name, x[2*i+3]); | |
4362 | show_symbol(m, x[2*i+2]); | |
871751e2 AV |
4363 | seq_putc(m, '\n'); |
4364 | } | |
d2e7b7d0 | 4365 | |
871751e2 AV |
4366 | return 0; |
4367 | } | |
4368 | ||
a0ec95a8 | 4369 | static const struct seq_operations slabstats_op = { |
1df3b26f | 4370 | .start = slab_start, |
276a2439 WL |
4371 | .next = slab_next, |
4372 | .stop = slab_stop, | |
871751e2 AV |
4373 | .show = leaks_show, |
4374 | }; | |
a0ec95a8 AD |
4375 | |
4376 | static int slabstats_open(struct inode *inode, struct file *file) | |
4377 | { | |
b208ce32 RJ |
4378 | unsigned long *n; |
4379 | ||
4380 | n = __seq_open_private(file, &slabstats_op, PAGE_SIZE); | |
4381 | if (!n) | |
4382 | return -ENOMEM; | |
4383 | ||
4384 | *n = PAGE_SIZE / (2 * sizeof(unsigned long)); | |
4385 | ||
4386 | return 0; | |
a0ec95a8 AD |
4387 | } |
4388 | ||
4389 | static const struct file_operations proc_slabstats_operations = { | |
4390 | .open = slabstats_open, | |
4391 | .read = seq_read, | |
4392 | .llseek = seq_lseek, | |
4393 | .release = seq_release_private, | |
4394 | }; | |
4395 | #endif | |
4396 | ||
4397 | static int __init slab_proc_init(void) | |
4398 | { | |
4399 | #ifdef CONFIG_DEBUG_SLAB_LEAK | |
4400 | proc_create("slab_allocators", 0, NULL, &proc_slabstats_operations); | |
871751e2 | 4401 | #endif |
a0ec95a8 AD |
4402 | return 0; |
4403 | } | |
4404 | module_init(slab_proc_init); | |
1da177e4 | 4405 | |
04385fc5 KC |
4406 | #ifdef CONFIG_HARDENED_USERCOPY |
4407 | /* | |
afcc90f8 KC |
4408 | * Rejects incorrectly sized objects and objects that are to be copied |
4409 | * to/from userspace but do not fall entirely within the containing slab | |
4410 | * cache's usercopy region. | |
04385fc5 KC |
4411 | * |
4412 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4413 | * to indicate an error. | |
4414 | */ | |
f4e6e289 KC |
4415 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4416 | bool to_user) | |
04385fc5 KC |
4417 | { |
4418 | struct kmem_cache *cachep; | |
4419 | unsigned int objnr; | |
4420 | unsigned long offset; | |
4421 | ||
4422 | /* Find and validate object. */ | |
4423 | cachep = page->slab_cache; | |
4424 | objnr = obj_to_index(cachep, page, (void *)ptr); | |
4425 | BUG_ON(objnr >= cachep->num); | |
4426 | ||
4427 | /* Find offset within object. */ | |
4428 | offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep); | |
4429 | ||
afcc90f8 KC |
4430 | /* Allow address range falling entirely within usercopy region. */ |
4431 | if (offset >= cachep->useroffset && | |
4432 | offset - cachep->useroffset <= cachep->usersize && | |
4433 | n <= cachep->useroffset - offset + cachep->usersize) | |
f4e6e289 | 4434 | return; |
04385fc5 | 4435 | |
afcc90f8 KC |
4436 | /* |
4437 | * If the copy is still within the allocated object, produce | |
4438 | * a warning instead of rejecting the copy. This is intended | |
4439 | * to be a temporary method to find any missing usercopy | |
4440 | * whitelists. | |
4441 | */ | |
2d891fbc KC |
4442 | if (usercopy_fallback && |
4443 | offset <= cachep->object_size && | |
afcc90f8 KC |
4444 | n <= cachep->object_size - offset) { |
4445 | usercopy_warn("SLAB object", cachep->name, to_user, offset, n); | |
4446 | return; | |
4447 | } | |
04385fc5 | 4448 | |
f4e6e289 | 4449 | usercopy_abort("SLAB object", cachep->name, to_user, offset, n); |
04385fc5 KC |
4450 | } |
4451 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4452 | ||
00e145b6 MS |
4453 | /** |
4454 | * ksize - get the actual amount of memory allocated for a given object | |
4455 | * @objp: Pointer to the object | |
4456 | * | |
4457 | * kmalloc may internally round up allocations and return more memory | |
4458 | * than requested. ksize() can be used to determine the actual amount of | |
4459 | * memory allocated. The caller may use this additional memory, even though | |
4460 | * a smaller amount of memory was initially specified with the kmalloc call. | |
4461 | * The caller must guarantee that objp points to a valid object previously | |
4462 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
4463 | * must not be freed during the duration of the call. | |
4464 | */ | |
fd76bab2 | 4465 | size_t ksize(const void *objp) |
1da177e4 | 4466 | { |
7ed2f9e6 AP |
4467 | size_t size; |
4468 | ||
ef8b4520 CL |
4469 | BUG_ON(!objp); |
4470 | if (unlikely(objp == ZERO_SIZE_PTR)) | |
00e145b6 | 4471 | return 0; |
1da177e4 | 4472 | |
7ed2f9e6 AP |
4473 | size = virt_to_cache(objp)->object_size; |
4474 | /* We assume that ksize callers could use the whole allocated area, | |
4475 | * so we need to unpoison this area. | |
4476 | */ | |
4ebb31a4 | 4477 | kasan_unpoison_shadow(objp, size); |
7ed2f9e6 AP |
4478 | |
4479 | return size; | |
1da177e4 | 4480 | } |
b1aabecd | 4481 | EXPORT_SYMBOL(ksize); |